maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/PowerPC/2007-03-30-SpillerCrash.ll

1802 lines
138 KiB
LLVM
Raw Normal View History

PowerPC: Do not use llc -march in tests. `llc -march` is problematic because it only switches the target architecture, but leaves the operating system unchanged. This occasionally leads to indeterministic tests because the OS from LLVM_DEFAULT_TARGET_TRIPLE is used. However we can simply always use `llc -mtriple` instead. This changes all the tests to do this to avoid people using -march when they copy and paste parts of tests. This patch: - Removes -march if the .ll file already has a matching `target triple` directive or -mtriple argument. - In all other cases changes -march=ppc32/-march=ppc64 to -mtriple=ppc32--/-mtriple=ppc64-- See also the discussion in https://reviews.llvm.org/D35287 llvm-svn: 309754
2017-08-02 06:20:41 +08:00
; RUN: llc -verify-machineinstrs < %s -mtriple=ppc32-- -mcpu=g5
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
define void @test(<4 x float>*, { { i16, i16, i32 } }*) {
xOperationInitMasks.exit:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%.sub7896 = getelementptr [4 x <4 x i32>], [4 x <4 x i32>]* null, i32 0, i32 0 ; <<4 x i32>*> [#uses=24]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 175, i32 3 ; <<4 x float>*>:2 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 174, i32 2 ; <<4 x float>*>:3 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 174, i32 3 ; <<4 x float>*>:4 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 173, i32 1 ; <<4 x float>*>:5 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 173, i32 2 ; <<4 x float>*>:6 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 173, i32 3 ; <<4 x float>*>:7 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 172, i32 1 ; <<4 x float>*>:8 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 172, i32 2 ; <<4 x float>*>:9 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 172, i32 3 ; <<4 x float>*>:10 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 171, i32 1 ; <<4 x float>*>:11 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 171, i32 2 ; <<4 x float>*>:12 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 171, i32 3 ; <<4 x float>*>:13 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 170, i32 1 ; <<4 x float>*>:14 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 170, i32 2 ; <<4 x float>*>:15 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 170, i32 3 ; <<4 x float>*>:16 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 169, i32 1 ; <<4 x float>*>:17 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 169, i32 2 ; <<4 x float>*>:18 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 169, i32 3 ; <<4 x float>*>:19 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 168, i32 1 ; <<4 x float>*>:20 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 168, i32 2 ; <<4 x float>*>:21 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 168, i32 3 ; <<4 x float>*>:22 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 167, i32 1 ; <<4 x float>*>:23 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 167, i32 2 ; <<4 x float>*>:24 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 167, i32 3 ; <<4 x float>*>:25 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 166, i32 1 ; <<4 x float>*>:26 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 166, i32 2 ; <<4 x float>*>:27 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 166, i32 3 ; <<4 x float>*>:28 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 165, i32 1 ; <<4 x float>*>:29 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 165, i32 2 ; <<4 x float>*>:30 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 165, i32 3 ; <<4 x float>*>:31 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 164, i32 1 ; <<4 x float>*>:32 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 164, i32 2 ; <<4 x float>*>:33 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 164, i32 3 ; <<4 x float>*>:34 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 163, i32 1 ; <<4 x float>*>:35 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 163, i32 2 ; <<4 x float>*>:36 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 163, i32 3 ; <<4 x float>*>:37 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 162, i32 1 ; <<4 x float>*>:38 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 162, i32 2 ; <<4 x float>*>:39 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 162, i32 3 ; <<4 x float>*>:40 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 161, i32 1 ; <<4 x float>*>:41 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 161, i32 2 ; <<4 x float>*>:42 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 161, i32 3 ; <<4 x float>*>:43 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 160, i32 1 ; <<4 x float>*>:44 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 160, i32 2 ; <<4 x float>*>:45 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 160, i32 3 ; <<4 x float>*>:46 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 159, i32 1 ; <<4 x float>*>:47 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 159, i32 2 ; <<4 x float>*>:48 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 159, i32 3 ; <<4 x float>*>:49 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 158, i32 1 ; <<4 x float>*>:50 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 158, i32 2 ; <<4 x float>*>:51 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 158, i32 3 ; <<4 x float>*>:52 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 157, i32 1 ; <<4 x float>*>:53 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 157, i32 2 ; <<4 x float>*>:54 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 157, i32 3 ; <<4 x float>*>:55 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 156, i32 1 ; <<4 x float>*>:56 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 156, i32 2 ; <<4 x float>*>:57 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 156, i32 3 ; <<4 x float>*>:58 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 155, i32 1 ; <<4 x float>*>:59 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 155, i32 2 ; <<4 x float>*>:60 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 155, i32 3 ; <<4 x float>*>:61 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 154, i32 1 ; <<4 x float>*>:62 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 154, i32 2 ; <<4 x float>*>:63 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 154, i32 3 ; <<4 x float>*>:64 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 153, i32 1 ; <<4 x float>*>:65 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 153, i32 2 ; <<4 x float>*>:66 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 153, i32 3 ; <<4 x float>*>:67 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 152, i32 1 ; <<4 x float>*>:68 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 152, i32 2 ; <<4 x float>*>:69 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 152, i32 3 ; <<4 x float>*>:70 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 151, i32 1 ; <<4 x float>*>:71 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 151, i32 2 ; <<4 x float>*>:72 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 151, i32 3 ; <<4 x float>*>:73 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 150, i32 1 ; <<4 x float>*>:74 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 150, i32 2 ; <<4 x float>*>:75 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 150, i32 3 ; <<4 x float>*>:76 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 149, i32 1 ; <<4 x float>*>:77 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 149, i32 2 ; <<4 x float>*>:78 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 149, i32 3 ; <<4 x float>*>:79 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 148, i32 1 ; <<4 x float>*>:80 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 148, i32 2 ; <<4 x float>*>:81 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 148, i32 3 ; <<4 x float>*>:82 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 147, i32 1 ; <<4 x float>*>:83 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 147, i32 2 ; <<4 x float>*>:84 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 147, i32 3 ; <<4 x float>*>:85 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 146, i32 1 ; <<4 x float>*>:86 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 146, i32 2 ; <<4 x float>*>:87 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 146, i32 3 ; <<4 x float>*>:88 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 145, i32 1 ; <<4 x float>*>:89 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 145, i32 2 ; <<4 x float>*>:90 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 145, i32 3 ; <<4 x float>*>:91 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 144, i32 1 ; <<4 x float>*>:92 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 144, i32 2 ; <<4 x float>*>:93 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 144, i32 3 ; <<4 x float>*>:94 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 143, i32 1 ; <<4 x float>*>:95 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 143, i32 2 ; <<4 x float>*>:96 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 143, i32 3 ; <<4 x float>*>:97 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 142, i32 1 ; <<4 x float>*>:98 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 142, i32 2 ; <<4 x float>*>:99 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 142, i32 3 ; <<4 x float>*>:100 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 141, i32 1 ; <<4 x float>*>:101 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 141, i32 2 ; <<4 x float>*>:102 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 141, i32 3 ; <<4 x float>*>:103 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 140, i32 1 ; <<4 x float>*>:104 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 140, i32 2 ; <<4 x float>*>:105 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 140, i32 3 ; <<4 x float>*>:106 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 139, i32 1 ; <<4 x float>*>:107 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 139, i32 2 ; <<4 x float>*>:108 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 139, i32 3 ; <<4 x float>*>:109 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 138, i32 1 ; <<4 x float>*>:110 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 138, i32 2 ; <<4 x float>*>:111 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 138, i32 3 ; <<4 x float>*>:112 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 137, i32 1 ; <<4 x float>*>:113 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 137, i32 2 ; <<4 x float>*>:114 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 137, i32 3 ; <<4 x float>*>:115 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 136, i32 1 ; <<4 x float>*>:116 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 136, i32 2 ; <<4 x float>*>:117 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 136, i32 3 ; <<4 x float>*>:118 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 135, i32 1 ; <<4 x float>*>:119 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 135, i32 2 ; <<4 x float>*>:120 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 135, i32 3 ; <<4 x float>*>:121 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 134, i32 1 ; <<4 x float>*>:122 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 134, i32 2 ; <<4 x float>*>:123 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 134, i32 3 ; <<4 x float>*>:124 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 133, i32 1 ; <<4 x float>*>:125 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 133, i32 2 ; <<4 x float>*>:126 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 133, i32 3 ; <<4 x float>*>:127 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 132, i32 1 ; <<4 x float>*>:128 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 132, i32 2 ; <<4 x float>*>:129 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 132, i32 3 ; <<4 x float>*>:130 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 131, i32 1 ; <<4 x float>*>:131 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 131, i32 2 ; <<4 x float>*>:132 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 131, i32 3 ; <<4 x float>*>:133 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 130, i32 1 ; <<4 x float>*>:134 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 130, i32 2 ; <<4 x float>*>:135 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 130, i32 3 ; <<4 x float>*>:136 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 129, i32 1 ; <<4 x float>*>:137 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 129, i32 2 ; <<4 x float>*>:138 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 129, i32 3 ; <<4 x float>*>:139 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 128, i32 1 ; <<4 x float>*>:140 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 128, i32 2 ; <<4 x float>*>:141 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 128, i32 3 ; <<4 x float>*>:142 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 127, i32 1 ; <<4 x float>*>:143 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 127, i32 2 ; <<4 x float>*>:144 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 127, i32 3 ; <<4 x float>*>:145 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 126, i32 1 ; <<4 x float>*>:146 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 126, i32 2 ; <<4 x float>*>:147 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 126, i32 3 ; <<4 x float>*>:148 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 125, i32 1 ; <<4 x float>*>:149 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 125, i32 2 ; <<4 x float>*>:150 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 125, i32 3 ; <<4 x float>*>:151 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 124, i32 1 ; <<4 x float>*>:152 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 124, i32 2 ; <<4 x float>*>:153 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 124, i32 3 ; <<4 x float>*>:154 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 123, i32 1 ; <<4 x float>*>:155 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 123, i32 2 ; <<4 x float>*>:156 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 123, i32 3 ; <<4 x float>*>:157 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 122, i32 1 ; <<4 x float>*>:158 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 122, i32 2 ; <<4 x float>*>:159 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 122, i32 3 ; <<4 x float>*>:160 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 121, i32 1 ; <<4 x float>*>:161 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 121, i32 2 ; <<4 x float>*>:162 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 121, i32 3 ; <<4 x float>*>:163 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 120, i32 1 ; <<4 x float>*>:164 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 120, i32 2 ; <<4 x float>*>:165 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 120, i32 3 ; <<4 x float>*>:166 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 119, i32 1 ; <<4 x float>*>:167 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 119, i32 2 ; <<4 x float>*>:168 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 119, i32 3 ; <<4 x float>*>:169 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 118, i32 1 ; <<4 x float>*>:170 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 118, i32 2 ; <<4 x float>*>:171 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 118, i32 3 ; <<4 x float>*>:172 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 117, i32 1 ; <<4 x float>*>:173 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 117, i32 2 ; <<4 x float>*>:174 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 117, i32 3 ; <<4 x float>*>:175 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 116, i32 1 ; <<4 x float>*>:176 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 116, i32 2 ; <<4 x float>*>:177 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 116, i32 3 ; <<4 x float>*>:178 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 115, i32 1 ; <<4 x float>*>:179 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 115, i32 2 ; <<4 x float>*>:180 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 115, i32 3 ; <<4 x float>*>:181 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 114, i32 1 ; <<4 x float>*>:182 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 114, i32 2 ; <<4 x float>*>:183 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 114, i32 3 ; <<4 x float>*>:184 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 113, i32 1 ; <<4 x float>*>:185 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 113, i32 2 ; <<4 x float>*>:186 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 113, i32 3 ; <<4 x float>*>:187 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 112, i32 1 ; <<4 x float>*>:188 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 112, i32 2 ; <<4 x float>*>:189 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 112, i32 3 ; <<4 x float>*>:190 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 111, i32 1 ; <<4 x float>*>:191 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 111, i32 2 ; <<4 x float>*>:192 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 111, i32 3 ; <<4 x float>*>:193 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 110, i32 1 ; <<4 x float>*>:194 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 110, i32 2 ; <<4 x float>*>:195 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 110, i32 3 ; <<4 x float>*>:196 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 109, i32 1 ; <<4 x float>*>:197 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 109, i32 2 ; <<4 x float>*>:198 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 109, i32 3 ; <<4 x float>*>:199 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 108, i32 1 ; <<4 x float>*>:200 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 108, i32 2 ; <<4 x float>*>:201 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 108, i32 3 ; <<4 x float>*>:202 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 107, i32 1 ; <<4 x float>*>:203 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 107, i32 2 ; <<4 x float>*>:204 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 107, i32 3 ; <<4 x float>*>:205 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 106, i32 1 ; <<4 x float>*>:206 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 106, i32 2 ; <<4 x float>*>:207 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 106, i32 3 ; <<4 x float>*>:208 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 105, i32 1 ; <<4 x float>*>:209 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 105, i32 2 ; <<4 x float>*>:210 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 105, i32 3 ; <<4 x float>*>:211 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 104, i32 1 ; <<4 x float>*>:212 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 104, i32 2 ; <<4 x float>*>:213 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 104, i32 3 ; <<4 x float>*>:214 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 103, i32 1 ; <<4 x float>*>:215 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 103, i32 2 ; <<4 x float>*>:216 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 103, i32 3 ; <<4 x float>*>:217 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 102, i32 1 ; <<4 x float>*>:218 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 102, i32 2 ; <<4 x float>*>:219 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 102, i32 3 ; <<4 x float>*>:220 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 101, i32 1 ; <<4 x float>*>:221 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 101, i32 2 ; <<4 x float>*>:222 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 101, i32 3 ; <<4 x float>*>:223 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 100, i32 1 ; <<4 x float>*>:224 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 100, i32 2 ; <<4 x float>*>:225 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 100, i32 3 ; <<4 x float>*>:226 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 99, i32 1 ; <<4 x float>*>:227 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 99, i32 2 ; <<4 x float>*>:228 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 99, i32 3 ; <<4 x float>*>:229 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 98, i32 1 ; <<4 x float>*>:230 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 98, i32 2 ; <<4 x float>*>:231 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 98, i32 3 ; <<4 x float>*>:232 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 97, i32 1 ; <<4 x float>*>:233 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 97, i32 2 ; <<4 x float>*>:234 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 97, i32 3 ; <<4 x float>*>:235 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 96, i32 1 ; <<4 x float>*>:236 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 96, i32 2 ; <<4 x float>*>:237 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 96, i32 3 ; <<4 x float>*>:238 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 95, i32 1 ; <<4 x float>*>:239 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 95, i32 2 ; <<4 x float>*>:240 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 95, i32 3 ; <<4 x float>*>:241 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 94, i32 1 ; <<4 x float>*>:242 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 94, i32 2 ; <<4 x float>*>:243 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 94, i32 3 ; <<4 x float>*>:244 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 93, i32 1 ; <<4 x float>*>:245 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 93, i32 2 ; <<4 x float>*>:246 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 93, i32 3 ; <<4 x float>*>:247 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 92, i32 1 ; <<4 x float>*>:248 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 92, i32 2 ; <<4 x float>*>:249 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 92, i32 3 ; <<4 x float>*>:250 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 91, i32 1 ; <<4 x float>*>:251 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 91, i32 2 ; <<4 x float>*>:252 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 91, i32 3 ; <<4 x float>*>:253 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 90, i32 1 ; <<4 x float>*>:254 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 90, i32 2 ; <<4 x float>*>:255 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 90, i32 3 ; <<4 x float>*>:256 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 89, i32 1 ; <<4 x float>*>:257 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 89, i32 2 ; <<4 x float>*>:258 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 89, i32 3 ; <<4 x float>*>:259 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 88, i32 1 ; <<4 x float>*>:260 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 88, i32 2 ; <<4 x float>*>:261 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 88, i32 3 ; <<4 x float>*>:262 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 87, i32 1 ; <<4 x float>*>:263 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 87, i32 2 ; <<4 x float>*>:264 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 87, i32 3 ; <<4 x float>*>:265 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 86, i32 1 ; <<4 x float>*>:266 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 86, i32 2 ; <<4 x float>*>:267 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 86, i32 3 ; <<4 x float>*>:268 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 85, i32 1 ; <<4 x float>*>:269 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 85, i32 2 ; <<4 x float>*>:270 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 85, i32 3 ; <<4 x float>*>:271 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 84, i32 1 ; <<4 x float>*>:272 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 84, i32 2 ; <<4 x float>*>:273 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 84, i32 3 ; <<4 x float>*>:274 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 83, i32 1 ; <<4 x float>*>:275 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 83, i32 2 ; <<4 x float>*>:276 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 83, i32 3 ; <<4 x float>*>:277 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 82, i32 1 ; <<4 x float>*>:278 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 82, i32 2 ; <<4 x float>*>:279 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 82, i32 3 ; <<4 x float>*>:280 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 81, i32 1 ; <<4 x float>*>:281 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 81, i32 2 ; <<4 x float>*>:282 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 81, i32 3 ; <<4 x float>*>:283 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 80, i32 1 ; <<4 x float>*>:284 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 80, i32 2 ; <<4 x float>*>:285 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 80, i32 3 ; <<4 x float>*>:286 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 79, i32 1 ; <<4 x float>*>:287 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 79, i32 2 ; <<4 x float>*>:288 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 79, i32 3 ; <<4 x float>*>:289 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 78, i32 1 ; <<4 x float>*>:290 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 78, i32 2 ; <<4 x float>*>:291 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 78, i32 3 ; <<4 x float>*>:292 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 77, i32 1 ; <<4 x float>*>:293 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 77, i32 2 ; <<4 x float>*>:294 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 77, i32 3 ; <<4 x float>*>:295 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 76, i32 1 ; <<4 x float>*>:296 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 76, i32 2 ; <<4 x float>*>:297 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 76, i32 3 ; <<4 x float>*>:298 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 75, i32 1 ; <<4 x float>*>:299 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 75, i32 2 ; <<4 x float>*>:300 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 75, i32 3 ; <<4 x float>*>:301 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 74, i32 1 ; <<4 x float>*>:302 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 74, i32 2 ; <<4 x float>*>:303 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 74, i32 3 ; <<4 x float>*>:304 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 73, i32 1 ; <<4 x float>*>:305 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 73, i32 2 ; <<4 x float>*>:306 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 73, i32 3 ; <<4 x float>*>:307 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 72, i32 1 ; <<4 x float>*>:308 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 72, i32 2 ; <<4 x float>*>:309 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 72, i32 3 ; <<4 x float>*>:310 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 71, i32 1 ; <<4 x float>*>:311 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 71, i32 2 ; <<4 x float>*>:312 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 71, i32 3 ; <<4 x float>*>:313 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 70, i32 1 ; <<4 x float>*>:314 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 70, i32 2 ; <<4 x float>*>:315 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 70, i32 3 ; <<4 x float>*>:316 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 69, i32 1 ; <<4 x float>*>:317 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 69, i32 2 ; <<4 x float>*>:318 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 69, i32 3 ; <<4 x float>*>:319 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 68, i32 1 ; <<4 x float>*>:320 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 68, i32 2 ; <<4 x float>*>:321 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 68, i32 3 ; <<4 x float>*>:322 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 67, i32 1 ; <<4 x float>*>:323 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 67, i32 2 ; <<4 x float>*>:324 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 67, i32 3 ; <<4 x float>*>:325 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 66, i32 1 ; <<4 x float>*>:326 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 66, i32 2 ; <<4 x float>*>:327 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 66, i32 3 ; <<4 x float>*>:328 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 65, i32 1 ; <<4 x float>*>:329 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 65, i32 2 ; <<4 x float>*>:330 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 65, i32 3 ; <<4 x float>*>:331 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 64, i32 1 ; <<4 x float>*>:332 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 64, i32 2 ; <<4 x float>*>:333 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 64, i32 3 ; <<4 x float>*>:334 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 63, i32 1 ; <<4 x float>*>:335 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 63, i32 2 ; <<4 x float>*>:336 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 63, i32 3 ; <<4 x float>*>:337 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 62, i32 1 ; <<4 x float>*>:338 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 62, i32 2 ; <<4 x float>*>:339 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 62, i32 3 ; <<4 x float>*>:340 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 61, i32 1 ; <<4 x float>*>:341 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 61, i32 2 ; <<4 x float>*>:342 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 61, i32 3 ; <<4 x float>*>:343 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 60, i32 1 ; <<4 x float>*>:344 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 60, i32 2 ; <<4 x float>*>:345 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 60, i32 3 ; <<4 x float>*>:346 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 59, i32 1 ; <<4 x float>*>:347 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 59, i32 2 ; <<4 x float>*>:348 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 59, i32 3 ; <<4 x float>*>:349 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 58, i32 1 ; <<4 x float>*>:350 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 58, i32 2 ; <<4 x float>*>:351 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 58, i32 3 ; <<4 x float>*>:352 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 57, i32 1 ; <<4 x float>*>:353 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 57, i32 2 ; <<4 x float>*>:354 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 57, i32 3 ; <<4 x float>*>:355 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 56, i32 1 ; <<4 x float>*>:356 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 56, i32 2 ; <<4 x float>*>:357 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 56, i32 3 ; <<4 x float>*>:358 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 55, i32 1 ; <<4 x float>*>:359 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 55, i32 2 ; <<4 x float>*>:360 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 55, i32 3 ; <<4 x float>*>:361 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 54, i32 1 ; <<4 x float>*>:362 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 54, i32 2 ; <<4 x float>*>:363 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 54, i32 3 ; <<4 x float>*>:364 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 53, i32 1 ; <<4 x float>*>:365 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 53, i32 2 ; <<4 x float>*>:366 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 53, i32 3 ; <<4 x float>*>:367 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 52, i32 1 ; <<4 x float>*>:368 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 52, i32 2 ; <<4 x float>*>:369 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 52, i32 3 ; <<4 x float>*>:370 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 51, i32 1 ; <<4 x float>*>:371 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 51, i32 2 ; <<4 x float>*>:372 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 51, i32 3 ; <<4 x float>*>:373 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 50, i32 1 ; <<4 x float>*>:374 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 50, i32 2 ; <<4 x float>*>:375 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 50, i32 3 ; <<4 x float>*>:376 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 49, i32 1 ; <<4 x float>*>:377 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 49, i32 2 ; <<4 x float>*>:378 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 49, i32 3 ; <<4 x float>*>:379 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 48, i32 1 ; <<4 x float>*>:380 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 48, i32 2 ; <<4 x float>*>:381 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 48, i32 3 ; <<4 x float>*>:382 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 47, i32 1 ; <<4 x float>*>:383 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 47, i32 2 ; <<4 x float>*>:384 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 47, i32 3 ; <<4 x float>*>:385 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 46, i32 1 ; <<4 x float>*>:386 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 46, i32 2 ; <<4 x float>*>:387 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 46, i32 3 ; <<4 x float>*>:388 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 45, i32 1 ; <<4 x float>*>:389 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 45, i32 2 ; <<4 x float>*>:390 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 45, i32 3 ; <<4 x float>*>:391 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 44, i32 1 ; <<4 x float>*>:392 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 44, i32 2 ; <<4 x float>*>:393 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 44, i32 3 ; <<4 x float>*>:394 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 43, i32 1 ; <<4 x float>*>:395 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 43, i32 2 ; <<4 x float>*>:396 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 43, i32 3 ; <<4 x float>*>:397 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 42, i32 1 ; <<4 x float>*>:398 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 42, i32 2 ; <<4 x float>*>:399 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 42, i32 3 ; <<4 x float>*>:400 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 41, i32 1 ; <<4 x float>*>:401 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 41, i32 2 ; <<4 x float>*>:402 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 41, i32 3 ; <<4 x float>*>:403 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 40, i32 1 ; <<4 x float>*>:404 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 40, i32 2 ; <<4 x float>*>:405 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 40, i32 3 ; <<4 x float>*>:406 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 39, i32 1 ; <<4 x float>*>:407 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 39, i32 2 ; <<4 x float>*>:408 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 39, i32 3 ; <<4 x float>*>:409 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 38, i32 1 ; <<4 x float>*>:410 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 38, i32 2 ; <<4 x float>*>:411 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 38, i32 3 ; <<4 x float>*>:412 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 37, i32 1 ; <<4 x float>*>:413 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 37, i32 2 ; <<4 x float>*>:414 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 37, i32 3 ; <<4 x float>*>:415 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 36, i32 1 ; <<4 x float>*>:416 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 36, i32 2 ; <<4 x float>*>:417 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 36, i32 3 ; <<4 x float>*>:418 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 35, i32 1 ; <<4 x float>*>:419 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 35, i32 2 ; <<4 x float>*>:420 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 35, i32 3 ; <<4 x float>*>:421 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 34, i32 1 ; <<4 x float>*>:422 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 34, i32 2 ; <<4 x float>*>:423 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 34, i32 3 ; <<4 x float>*>:424 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 33, i32 1 ; <<4 x float>*>:425 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 33, i32 2 ; <<4 x float>*>:426 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 33, i32 3 ; <<4 x float>*>:427 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 32, i32 1 ; <<4 x float>*>:428 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 32, i32 2 ; <<4 x float>*>:429 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 32, i32 3 ; <<4 x float>*>:430 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 31, i32 1 ; <<4 x float>*>:431 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 31, i32 2 ; <<4 x float>*>:432 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 31, i32 3 ; <<4 x float>*>:433 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 30, i32 1 ; <<4 x float>*>:434 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 30, i32 2 ; <<4 x float>*>:435 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 30, i32 3 ; <<4 x float>*>:436 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 29, i32 1 ; <<4 x float>*>:437 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 29, i32 2 ; <<4 x float>*>:438 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 29, i32 3 ; <<4 x float>*>:439 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 28, i32 1 ; <<4 x float>*>:440 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 28, i32 2 ; <<4 x float>*>:441 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 28, i32 3 ; <<4 x float>*>:442 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 27, i32 1 ; <<4 x float>*>:443 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 27, i32 2 ; <<4 x float>*>:444 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 27, i32 3 ; <<4 x float>*>:445 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 26, i32 1 ; <<4 x float>*>:446 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 26, i32 2 ; <<4 x float>*>:447 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 26, i32 3 ; <<4 x float>*>:448 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 25, i32 1 ; <<4 x float>*>:449 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 25, i32 2 ; <<4 x float>*>:450 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 25, i32 3 ; <<4 x float>*>:451 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 24, i32 1 ; <<4 x float>*>:452 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 24, i32 2 ; <<4 x float>*>:453 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 24, i32 3 ; <<4 x float>*>:454 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 23, i32 1 ; <<4 x float>*>:455 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 23, i32 2 ; <<4 x float>*>:456 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 23, i32 3 ; <<4 x float>*>:457 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 22, i32 1 ; <<4 x float>*>:458 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 22, i32 2 ; <<4 x float>*>:459 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 22, i32 3 ; <<4 x float>*>:460 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 21, i32 1 ; <<4 x float>*>:461 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 21, i32 2 ; <<4 x float>*>:462 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 21, i32 3 ; <<4 x float>*>:463 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 20, i32 1 ; <<4 x float>*>:464 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 20, i32 2 ; <<4 x float>*>:465 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 20, i32 3 ; <<4 x float>*>:466 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 19, i32 1 ; <<4 x float>*>:467 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 19, i32 2 ; <<4 x float>*>:468 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 19, i32 3 ; <<4 x float>*>:469 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 18, i32 1 ; <<4 x float>*>:470 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 18, i32 2 ; <<4 x float>*>:471 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 18, i32 3 ; <<4 x float>*>:472 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 17, i32 1 ; <<4 x float>*>:473 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 17, i32 2 ; <<4 x float>*>:474 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 17, i32 3 ; <<4 x float>*>:475 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 16, i32 1 ; <<4 x float>*>:476 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 16, i32 2 ; <<4 x float>*>:477 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 16, i32 3 ; <<4 x float>*>:478 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 15, i32 1 ; <<4 x float>*>:479 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 15, i32 2 ; <<4 x float>*>:480 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 15, i32 3 ; <<4 x float>*>:481 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 14, i32 1 ; <<4 x float>*>:482 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 14, i32 2 ; <<4 x float>*>:483 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 14, i32 3 ; <<4 x float>*>:484 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 1 ; <<4 x float>*>:485 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 2 ; <<4 x float>*>:486 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 3 ; <<4 x float>*>:487 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 3, i32 1 ; <<4 x float>*>:488 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 3, i32 2 ; <<4 x float>*>:489 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 3, i32 3 ; <<4 x float>*>:490 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 180, i32 1 ; <<4 x float>*>:491 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 180, i32 2 ; <<4 x float>*>:492 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 180, i32 3 ; <<4 x float>*>:493 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 181, i32 1 ; <<4 x float>*>:494 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 181, i32 2 ; <<4 x float>*>:495 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 181, i32 3 ; <<4 x float>*>:496 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 182, i32 1 ; <<4 x float>*>:497 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 182, i32 2 ; <<4 x float>*>:498 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 182, i32 3 ; <<4 x float>*>:499 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 183, i32 1 ; <<4 x float>*>:500 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 183, i32 2 ; <<4 x float>*>:501 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 183, i32 3 ; <<4 x float>*>:502 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 184, i32 1 ; <<4 x float>*>:503 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 184, i32 2 ; <<4 x float>*>:504 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 184, i32 3 ; <<4 x float>*>:505 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 185, i32 1 ; <<4 x float>*>:506 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 185, i32 2 ; <<4 x float>*>:507 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 185, i32 3 ; <<4 x float>*>:508 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 190, i32 1 ; <<4 x float>*>:509 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 190, i32 2 ; <<4 x float>*>:510 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 190, i32 3 ; <<4 x float>*>:511 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 9, i32 1 ; <<4 x float>*>:512 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 9, i32 2 ; <<4 x float>*>:513 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 9, i32 3 ; <<4 x float>*>:514 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 10, i32 1 ; <<4 x float>*>:515 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 10, i32 2 ; <<4 x float>*>:516 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 10, i32 3 ; <<4 x float>*>:517 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 11, i32 1 ; <<4 x float>*>:518 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 11, i32 2 ; <<4 x float>*>:519 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 11, i32 3 ; <<4 x float>*>:520 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 12, i32 1 ; <<4 x float>*>:521 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 12, i32 2 ; <<4 x float>*>:522 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 12, i32 3 ; <<4 x float>*>:523 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 13, i32 1 ; <<4 x float>*>:524 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 13, i32 2 ; <<4 x float>*>:525 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 13, i32 3 ; <<4 x float>*>:526 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 5, i32 1 ; <<4 x float>*>:527 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 5, i32 2 ; <<4 x float>*>:528 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 5, i32 3 ; <<4 x float>*>:529 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 1 ; <<4 x float>*>:530 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 2 ; <<4 x float>*>:531 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 3 ; <<4 x float>*>:532 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 2, i32 1 ; <<4 x float>*>:533 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 2, i32 2 ; <<4 x float>*>:534 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 2, i32 3 ; <<4 x float>*>:535 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 6, i32 1 ; <<4 x float>*>:536 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 6, i32 2 ; <<4 x float>*>:537 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 6, i32 3 ; <<4 x float>*>:538 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 7, i32 1 ; <<4 x float>*>:539 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 7, i32 2 ; <<4 x float>*>:540 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 7, i32 3 ; <<4 x float>*>:541 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 186, i32 1 ; <<4 x float>*>:542 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 186, i32 2 ; <<4 x float>*>:543 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 186, i32 3 ; <<4 x float>*>:544 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 187, i32 1 ; <<4 x float>*>:545 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 187, i32 2 ; <<4 x float>*>:546 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 187, i32 3 ; <<4 x float>*>:547 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 8, i32 1 ; <<4 x float>*>:548 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 8, i32 2 ; <<4 x float>*>:549 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 8, i32 3 ; <<4 x float>*>:550 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:551 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 188, i32 1 ; <<4 x float>*>:552 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 188, i32 2 ; <<4 x float>*>:553 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %553 ; <<4 x float>>:554 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 188, i32 3 ; <<4 x float>*>:555 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> %554, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:556 [#uses=1]
call <4 x i32> @llvm.ppc.altivec.vcmpgtfp( <4 x float> zeroinitializer, <4 x float> %556 ) ; <<4 x i32>>:557 [#uses=0]
bitcast <4 x i32> zeroinitializer to <4 x float> ; <<4 x float>>:558 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 189, i32 0 ; <<4 x float>*>:559 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 189, i32 2 ; <<4 x float>*>:560 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> zeroinitializer, <4 x float>* %560
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 189, i32 3 ; <<4 x float>*>:561 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 189, i32 1 ; <<4 x float>*>:562 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 189, i32 2 ; <<4 x float>*>:563 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:564 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:565 [#uses=1]
store <4 x float> %565, <4 x float>* null
icmp eq i32 0, 0 ; <i1>:566 [#uses=1]
br i1 %566, label %.critedge, label %xPIF.exit
.critedge: ; preds = %xOperationInitMasks.exit
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [4 x <4 x i32>], [4 x <4 x i32>]* null, i32 0, i32 3 ; <<4 x i32>*>:567 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
and <4 x i32> zeroinitializer, zeroinitializer ; <<4 x i32>>:568 [#uses=0]
or <4 x i32> zeroinitializer, zeroinitializer ; <<4 x i32>>:569 [#uses=0]
icmp eq i32 0, 0 ; <i1>:570 [#uses=1]
br i1 %570, label %.critedge7898, label %xPBRK.exit
.critedge7898: ; preds = %.critedge
br label %xPIF.exit
xPIF.exit: ; preds = %.critedge7898, %xOperationInitMasks.exit
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 188, i32 1 ; <<4 x float>*>:571 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:572 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:573 [#uses=0]
icmp eq i32 0, 0 ; <i1>:574 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 3, i32 1 ; <<4 x float>*>:575 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %0 ; <<4 x float>>:576 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:577 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 2, i32 0 ; <<4 x float>*>:578 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 2, i32 1 ; <<4 x float>*>:579 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 2, i32 2 ; <<4 x float>*>:580 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 2, i32 3 ; <<4 x float>*>:581 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 3 ; <<4 x float>*>:582 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:583 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 1 ; <<4 x float>*>:584 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %584 ; <<4 x float>>:585 [#uses=1]
load <4 x float>, <4 x float>* null ; <<4 x float>>:586 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 3 ; <<4 x float>*>:587 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %587 ; <<4 x float>>:588 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> %583, <4 x float> undef, <4 x i32> < i32 3, i32 3, i32 3, i32 3 > ; <<4 x float>>:589 [#uses=1]
shufflevector <4 x float> %585, <4 x float> undef, <4 x i32> < i32 3, i32 3, i32 3, i32 3 > ; <<4 x float>>:590 [#uses=1]
shufflevector <4 x float> %588, <4 x float> undef, <4 x i32> < i32 3, i32 3, i32 3, i32 3 > ; <<4 x float>>:591 [#uses=1]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fmul <4 x float> zeroinitializer, %589 ; <<4 x float>>:592 [#uses=0]
fmul <4 x float> zeroinitializer, %590 ; <<4 x float>>:593 [#uses=0]
fmul <4 x float> zeroinitializer, zeroinitializer ; <<4 x float>>:594 [#uses=1]
fmul <4 x float> zeroinitializer, %591 ; <<4 x float>>:595 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 0 ; <<4 x float>*>:596 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %596 ; <<4 x float>>:597 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> zeroinitializer, <4 x float>* %596
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:598 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 2 ; <<4 x float>*>:599 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> %594, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 1, i32 2, i32 7 > ; <<4 x float>>:600 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 3 ; <<4 x float>*>:601 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %601 ; <<4 x float>>:602 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> zeroinitializer, <4 x float>* %601
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:603 [#uses=0]
load <4 x float>, <4 x float>* null ; <<4 x float>>:604 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 2 ; <<4 x float>*>:605 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %605 ; <<4 x float>>:606 [#uses=1]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fsub <4 x float> zeroinitializer, %604 ; <<4 x float>>:607 [#uses=2]
fsub <4 x float> zeroinitializer, %606 ; <<4 x float>>:608 [#uses=2]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:609 [#uses=0]
br i1 false, label %617, label %610
; <label>:610 ; preds = %xPIF.exit
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:611 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 1 ; <<4 x float>*>:612 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %612 ; <<4 x float>>:613 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> %607, <4 x float> %613, <4 x i32> < i32 0, i32 1, i32 2, i32 7 > ; <<4 x float>>:614 [#uses=1]
store <4 x float> %614, <4 x float>* %612
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 3 ; <<4 x float>*>:615 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %615 ; <<4 x float>>:616 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> zeroinitializer, <4 x float>* %615
br label %xST.exit400
; <label>:617 ; preds = %xPIF.exit
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:618 [#uses=0]
shufflevector <4 x i32> zeroinitializer, <4 x i32> undef, <4 x i32> < i32 1, i32 1, i32 1, i32 1 > ; <<4 x i32>>:619 [#uses=1]
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> %619, <4 x i32> zeroinitializer ) ; <i32>:620 [#uses=1]
icmp eq i32 %620, 0 ; <i1>:621 [#uses=1]
br i1 %621, label %625, label %622
; <label>:622 ; preds = %617
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 1 ; <<4 x float>*>:623 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> %607, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 1, i32 2, i32 7 > ; <<4 x float>>:624 [#uses=0]
br label %625
; <label>:625 ; preds = %622, %617
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:626 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:627 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 2 ; <<4 x float>*>:628 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %628 ; <<4 x float>>:629 [#uses=0]
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:630 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:631 [#uses=1]
icmp eq i32 %631, 0 ; <i1>:632 [#uses=1]
br i1 %632, label %xST.exit400, label %633
; <label>:633 ; preds = %625
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:634 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> %634, <4 x i32> < i32 0, i32 1, i32 2, i32 7 > ; <<4 x float>>:635 [#uses=1]
store <4 x float> %635, <4 x float>* null
br label %xST.exit400
xST.exit400: ; preds = %633, %625, %610
%.17218 = phi <4 x float> [ zeroinitializer, %610 ], [ %608, %633 ], [ %608, %625 ] ; <<4 x float>> [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 0 ; <<4 x float>*>:636 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %636 ; <<4 x float>>:637 [#uses=0]
load <4 x float>, <4 x float>* null ; <<4 x float>>:638 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 2 ; <<4 x float>*>:639 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:640 [#uses=2]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fmul <4 x float> %638, %638 ; <<4 x float>>:641 [#uses=1]
fmul <4 x float> zeroinitializer, zeroinitializer ; <<4 x float>>:642 [#uses=0]
fmul <4 x float> %640, %640 ; <<4 x float>>:643 [#uses=2]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> < i32 1, i32 1, i32 1, i32 1 > ; <<4 x float>>:644 [#uses=0]
shufflevector <4 x float> %643, <4 x float> undef, <4 x i32> < i32 1, i32 1, i32 1, i32 1 > ; <<4 x float>>:645 [#uses=1]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fadd <4 x float> %645, %643 ; <<4 x float>>:646 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> < i32 2, i32 2, i32 2, i32 2 > ; <<4 x float>>:647 [#uses=1]
shufflevector <4 x float> %641, <4 x float> undef, <4 x i32> < i32 2, i32 2, i32 2, i32 2 > ; <<4 x float>>:648 [#uses=1]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fadd <4 x float> zeroinitializer, %647 ; <<4 x float>>:649 [#uses=2]
fadd <4 x float> zeroinitializer, %648 ; <<4 x float>>:650 [#uses=0]
fadd <4 x float> zeroinitializer, zeroinitializer ; <<4 x float>>:651 [#uses=2]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
call <4 x float> @llvm.ppc.altivec.vrsqrtefp( <4 x float> %649 ) ; <<4 x float>>:652 [#uses=1]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fmul <4 x float> %652, %649 ; <<4 x float>>:653 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
call <4 x float> @llvm.ppc.altivec.vrsqrtefp( <4 x float> %651 ) ; <<4 x float>>:654 [#uses=1]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fmul <4 x float> %654, %651 ; <<4 x float>>:655 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
icmp eq i32 0, 0 ; <i1>:656 [#uses=1]
br i1 %656, label %665, label %657
; <label>:657 ; preds = %xST.exit400
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 0 ; <<4 x float>*>:658 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> %653, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:659 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 1 ; <<4 x float>*>:660 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %660 ; <<4 x float>>:661 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 2 ; <<4 x float>*>:662 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 3 ; <<4 x float>*>:663 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:664 [#uses=0]
br label %xST.exit402
; <label>:665 ; preds = %xST.exit400
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:666 [#uses=0]
br i1 false, label %669, label %667
; <label>:667 ; preds = %665
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:668 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br label %669
; <label>:669 ; preds = %667, %665
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:670 [#uses=0]
br label %xST.exit402
xST.exit402: ; preds = %669, %657
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 0 ; <<4 x float>*>:671 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:672 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 2 ; <<4 x float>*>:673 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 1 ; <<4 x float>*>:674 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %674 ; <<4 x float>>:675 [#uses=1]
load <4 x float>, <4 x float>* null ; <<4 x float>>:676 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:677 [#uses=1]
shufflevector <4 x float> %675, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:678 [#uses=1]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fmul <4 x float> zeroinitializer, %677 ; <<4 x float>>:679 [#uses=0]
fmul <4 x float> zeroinitializer, %678 ; <<4 x float>>:680 [#uses=0]
fmul <4 x float> zeroinitializer, zeroinitializer ; <<4 x float>>:681 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
icmp eq i32 0, 0 ; <i1>:682 [#uses=1]
br i1 %682, label %689, label %683
; <label>:683 ; preds = %xST.exit402
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 5, i32 1 ; <<4 x float>*>:684 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %684 ; <<4 x float>>:685 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 5, i32 2 ; <<4 x float>*>:686 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 5, i32 3 ; <<4 x float>*>:687 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> %681, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:688 [#uses=0]
br label %xST.exit405
; <label>:689 ; preds = %xST.exit402
shufflevector <4 x i32> zeroinitializer, <4 x i32> undef, <4 x i32> zeroinitializer ; <<4 x i32>>:690 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:691 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x i32> %691, <4 x i32> undef, <4 x i32> < i32 3, i32 3, i32 3, i32 3 > ; <<4 x i32>>:692 [#uses=1]
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> %692, <4 x i32> zeroinitializer ) ; <i32>:693 [#uses=1]
icmp eq i32 %693, 0 ; <i1>:694 [#uses=0]
br label %xST.exit405
xST.exit405: ; preds = %689, %683
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 3 ; <<4 x float>*>:695 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:696 [#uses=0]
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:697 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:698 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 5, i32 2 ; <<4 x float>*>:699 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:700 [#uses=1]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fadd <4 x float> zeroinitializer, %700 ; <<4 x float>>:701 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:702 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> %702, <4 x i32> zeroinitializer ) ; <i32>:703 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 1 ; <<4 x float>*>:704 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %704 ; <<4 x float>>:705 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> zeroinitializer, <4 x float>* %704
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:706 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> zeroinitializer, <4 x float>* null
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 3 ; <<4 x float>*>:707 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %707 ; <<4 x float>>:708 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> zeroinitializer, <4 x float>* %707
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:709 [#uses=0]
load <4 x float>, <4 x float>* null ; <<4 x float>>:710 [#uses=0]
load <4 x float>, <4 x float>* null ; <<4 x float>>:711 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> %711, <4 x float> undef, <4 x i32> < i32 2, i32 2, i32 2, i32 2 > ; <<4 x float>>:712 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 1 ; <<4 x float>*>:713 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 2 ; <<4 x float>*>:714 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %714 ; <<4 x float>>:715 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:716 [#uses=0]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fmul <4 x float> zeroinitializer, zeroinitializer ; <<4 x float>>:717 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:718 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 5, i32 0 ; <<4 x float>*>:719 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> zeroinitializer, <4 x float>* %719
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 5, i32 1 ; <<4 x float>*>:720 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> %717, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:721 [#uses=1]
store <4 x float> %721, <4 x float>* %720
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 5, i32 2 ; <<4 x float>*>:722 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %722 ; <<4 x float>>:723 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> %723, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:724 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 5, i32 3 ; <<4 x float>*>:725 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> zeroinitializer, <4 x float>* %725
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 5, i32 2 ; <<4 x float>*>:726 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %726 ; <<4 x float>>:727 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 5, i32 3 ; <<4 x float>*>:728 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %728 ; <<4 x float>>:729 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 0 ; <<4 x float>*>:730 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %730 ; <<4 x float>>:731 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 1 ; <<4 x float>*>:732 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %732 ; <<4 x float>>:733 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 3 ; <<4 x float>*>:734 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:735 [#uses=1]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fmul <4 x float> zeroinitializer, zeroinitializer ; <<4 x float>>:736 [#uses=1]
fmul <4 x float> zeroinitializer, zeroinitializer ; <<4 x float>>:737 [#uses=1]
fmul <4 x float> zeroinitializer, %735 ; <<4 x float>>:738 [#uses=1]
fmul <4 x float> zeroinitializer, zeroinitializer ; <<4 x float>>:739 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:740 [#uses=1]
icmp eq i32 %740, 0 ; <i1>:741 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 0 ; <<4 x float>*>:742 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %742 ; <<4 x float>>:743 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> %736, <4 x float> %743, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:744 [#uses=1]
store <4 x float> %744, <4 x float>* %742
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 1 ; <<4 x float>*>:745 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %745 ; <<4 x float>>:746 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> %737, <4 x float> %746, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:747 [#uses=0]
shufflevector <4 x float> %738, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:748 [#uses=1]
store <4 x float> %748, <4 x float>* null
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 3 ; <<4 x float>*>:749 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %749 ; <<4 x float>>:750 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> %739, <4 x float> %750, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:751 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 0 ; <<4 x float>*>:752 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 1 ; <<4 x float>*>:753 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %753 ; <<4 x float>>:754 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 2 ; <<4 x float>*>:755 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:756 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:757 [#uses=1]
shufflevector <4 x float> %756, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:758 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 2 ; <<4 x float>*>:759 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %759 ; <<4 x float>>:760 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 3 ; <<4 x float>*>:761 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:762 [#uses=0]
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:763 [#uses=1]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fadd <4 x float> %757, zeroinitializer ; <<4 x float>>:764 [#uses=0]
fadd <4 x float> %758, %763 ; <<4 x float>>:765 [#uses=0]
fmul <4 x float> zeroinitializer, zeroinitializer ; <<4 x float>>:766 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br i1 false, label %773, label %767
; <label>:767 ; preds = %xST.exit405
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 1 ; <<4 x float>*>:768 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:769 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> %769, <4 x i32> < i32 0, i32 1, i32 2, i32 7 > ; <<4 x float>>:770 [#uses=1]
store <4 x float> %770, <4 x float>* null
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 3 ; <<4 x float>*>:771 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %771 ; <<4 x float>>:772 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br label %xST.exit422
; <label>:773 ; preds = %xST.exit405
br label %xST.exit422
xST.exit422: ; preds = %773, %767
%.07267 = phi <4 x float> [ %766, %767 ], [ undef, %773 ] ; <<4 x float>> [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 3 ; <<4 x float>*>:774 [#uses=0]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fmul <4 x float> zeroinitializer, zeroinitializer ; <<4 x float>>:775 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
icmp eq i32 0, 0 ; <i1>:776 [#uses=1]
br i1 %776, label %780, label %777
; <label>:777 ; preds = %xST.exit422
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 2 ; <<4 x float>*>:778 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 3 ; <<4 x float>*>:779 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br label %xST.exit431
; <label>:780 ; preds = %xST.exit422
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:781 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 2 ; <<4 x float>*>:782 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %782 ; <<4 x float>>:783 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> zeroinitializer, <4 x float>* %782
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:784 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x i32> %784, <4 x i32> undef, <4 x i32> < i32 3, i32 3, i32 3, i32 3 > ; <<4 x i32>>:785 [#uses=0]
icmp eq i32 0, 0 ; <i1>:786 [#uses=0]
br label %xST.exit431
xST.exit431: ; preds = %780, %777
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 2 ; <<4 x float>*>:787 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:788 [#uses=0]
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:789 [#uses=2]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> %789, <4 x i32> zeroinitializer ) ; <i32>:790 [#uses=1]
icmp eq i32 %790, 0 ; <i1>:791 [#uses=0]
shufflevector <4 x i32> %789, <4 x i32> undef, <4 x i32> zeroinitializer ; <<4 x i32>>:792 [#uses=1]
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> %792, <4 x i32> zeroinitializer ) ; <i32>:793 [#uses=1]
icmp eq i32 %793, 0 ; <i1>:794 [#uses=1]
br i1 %794, label %797, label %795
; <label>:795 ; preds = %xST.exit431
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:796 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> zeroinitializer, <4 x float>* null
br label %797
; <label>:797 ; preds = %795, %xST.exit431
%.07332 = phi <4 x float> [ zeroinitializer, %795 ], [ undef, %xST.exit431 ] ; <<4 x float>> [#uses=0]
shufflevector <4 x i32> zeroinitializer, <4 x i32> undef, <4 x i32> < i32 1, i32 1, i32 1, i32 1 > ; <<4 x i32>>:798 [#uses=0]
br i1 false, label %xST.exit434, label %799
; <label>:799 ; preds = %797
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:800 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> zeroinitializer, <4 x float>* null
br label %xST.exit434
xST.exit434: ; preds = %799, %797
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:801 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x i32> %801, <4 x i32> undef, <4 x i32> < i32 2, i32 2, i32 2, i32 2 > ; <<4 x i32>>:802 [#uses=0]
shufflevector <4 x i32> zeroinitializer, <4 x i32> undef, <4 x i32> < i32 3, i32 3, i32 3, i32 3 > ; <<4 x i32>>:803 [#uses=0]
icmp eq i32 0, 0 ; <i1>:804 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 0 ; <<4 x float>*>:805 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %805 ; <<4 x float>>:806 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 1 ; <<4 x float>*>:807 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %807 ; <<4 x float>>:808 [#uses=0]
load <4 x float>, <4 x float>* null ; <<4 x float>>:809 [#uses=0]
load <4 x float>, <4 x float>* null ; <<4 x float>>:810 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 0 ; <<4 x float>*>:811 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 2 ; <<4 x float>*>:812 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %812 ; <<4 x float>>:813 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 3 ; <<4 x float>*>:814 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %814 ; <<4 x float>>:815 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:816 [#uses=0]
unreachable
xPBRK.exit: ; preds = %.critedge
store <4 x i32> < i32 -1, i32 -1, i32 -1, i32 -1 >, <4 x i32>* %.sub7896
store <4 x i32> zeroinitializer, <4 x i32>* null
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 186, i32 1 ; <<4 x float>*>:817 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %817 ; <<4 x float>>:818 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 186, i32 2 ; <<4 x float>*>:819 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %819 ; <<4 x float>>:820 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 186, i32 3 ; <<4 x float>*>:821 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %821 ; <<4 x float>>:822 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:823 [#uses=1]
shufflevector <4 x float> %818, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:824 [#uses=1]
shufflevector <4 x float> %820, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:825 [#uses=1]
shufflevector <4 x float> %822, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:826 [#uses=1]
shufflevector <4 x float> %823, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:827 [#uses=0]
shufflevector <4 x float> %824, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:828 [#uses=1]
store <4 x float> %828, <4 x float>* null
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:829 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> %825, <4 x float> %829, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:830 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 186, i32 3 ; <<4 x float>*>:831 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %831 ; <<4 x float>>:832 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> %826, <4 x float> %832, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:833 [#uses=1]
store <4 x float> %833, <4 x float>* %831
br label %xLS.exit449
xLS.exit449: ; preds = %1215, %xPBRK.exit
%.27464 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.17463, %1215 ] ; <<4 x float>> [#uses=2]
%.27469 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.17468, %1215 ] ; <<4 x float>> [#uses=2]
%.27474 = phi <4 x float> [ undef, %xPBRK.exit ], [ zeroinitializer, %1215 ] ; <<4 x float>> [#uses=1]
%.17482 = phi <4 x float> [ undef, %xPBRK.exit ], [ zeroinitializer, %1215 ] ; <<4 x float>> [#uses=0]
%.17486 = phi <4 x float> [ undef, %xPBRK.exit ], [ zeroinitializer, %1215 ] ; <<4 x float>> [#uses=0]
%.17490 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.07489, %1215 ] ; <<4 x float>> [#uses=2]
%.17494 = phi <4 x float> [ undef, %xPBRK.exit ], [ zeroinitializer, %1215 ] ; <<4 x float>> [#uses=0]
%.27504 = phi <4 x float> [ undef, %xPBRK.exit ], [ zeroinitializer, %1215 ] ; <<4 x float>> [#uses=0]
%.17513 = phi <4 x float> [ undef, %xPBRK.exit ], [ zeroinitializer, %1215 ] ; <<4 x float>> [#uses=0]
%.17517 = phi <4 x float> [ undef, %xPBRK.exit ], [ zeroinitializer, %1215 ] ; <<4 x float>> [#uses=0]
%.17552 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.07551, %1215 ] ; <<4 x float>> [#uses=2]
%.17556 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.07555, %1215 ] ; <<4 x float>> [#uses=2]
%.17560 = phi <4 x float> [ undef, %xPBRK.exit ], [ zeroinitializer, %1215 ] ; <<4 x float>> [#uses=0]
%.17583 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.07582, %1215 ] ; <<4 x float>> [#uses=2]
%.17591 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.07590, %1215 ] ; <<4 x float>> [#uses=2]
%.17599 = phi <4 x float> [ undef, %xPBRK.exit ], [ zeroinitializer, %1215 ] ; <<4 x float>> [#uses=0]
%.17618 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.07617, %1215 ] ; <<4 x float>> [#uses=2]
%.17622 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.07621, %1215 ] ; <<4 x float>> [#uses=2]
%.17626 = phi <4 x float> [ undef, %xPBRK.exit ], [ zeroinitializer, %1215 ] ; <<4 x float>> [#uses=0]
%.17653 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.07652, %1215 ] ; <<4 x float>> [#uses=2]
%.17657 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.07656, %1215 ] ; <<4 x float>> [#uses=2]
%.17661 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.07660, %1215 ] ; <<4 x float>> [#uses=2]
%.17665 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.07664, %1215 ] ; <<4 x float>> [#uses=2]
%.17723 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.07722, %1215 ] ; <<4 x float>> [#uses=2]
%.17727 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.07726, %1215 ] ; <<4 x float>> [#uses=2]
%.17731 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.07730, %1215 ] ; <<4 x float>> [#uses=2]
%.17735 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.07734, %1215 ] ; <<4 x float>> [#uses=2]
%.17770 = phi <4 x float> [ undef, %xPBRK.exit ], [ %.07769, %1215 ] ; <<4 x float>> [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 186, i32 0 ; <<4 x float>*>:834 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:835 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 186, i32 2 ; <<4 x float>*>:836 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 186, i32 3 ; <<4 x float>*>:837 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:838 [#uses=0]
shufflevector <4 x float> %835, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:839 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr <4 x float>, <4 x float>* null, i32 878 ; <<4 x float>*>:840 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %840 ; <<4 x float>>:841 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
call <4 x float> @llvm.ppc.altivec.vcfsx( <4 x i32> zeroinitializer, i32 0 ) ; <<4 x float>>:842 [#uses=1]
shufflevector <4 x float> %842, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:843 [#uses=2]
call <4 x i32> @llvm.ppc.altivec.vcmpgtfp( <4 x float> %843, <4 x float> %839 ) ; <<4 x i32>>:844 [#uses=1]
bitcast <4 x i32> %844 to <4 x float> ; <<4 x float>>:845 [#uses=1]
call <4 x i32> @llvm.ppc.altivec.vcmpgtfp( <4 x float> %843, <4 x float> zeroinitializer ) ; <<4 x i32>>:846 [#uses=0]
bitcast <4 x i32> zeroinitializer to <4 x float> ; <<4 x float>>:847 [#uses=1]
icmp eq i32 0, 0 ; <i1>:848 [#uses=1]
br i1 %848, label %854, label %849
; <label>:849 ; preds = %xLS.exit449
shufflevector <4 x float> zeroinitializer, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:850 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 189, i32 1 ; <<4 x float>*>:851 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> zeroinitializer, <4 x float>* %851
shufflevector <4 x float> zeroinitializer, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:852 [#uses=1]
store <4 x float> %852, <4 x float>* null
shufflevector <4 x float> %847, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:853 [#uses=0]
br label %xST.exit451
; <label>:854 ; preds = %xLS.exit449
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:855 [#uses=0]
br i1 false, label %859, label %856
; <label>:856 ; preds = %854
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 189, i32 0 ; <<4 x float>*>:857 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %857 ; <<4 x float>>:858 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> zeroinitializer, <4 x float>* %857
br label %859
; <label>:859 ; preds = %856, %854
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:860 [#uses=0]
br i1 false, label %864, label %861
; <label>:861 ; preds = %859
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 189, i32 1 ; <<4 x float>*>:862 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> %845, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:863 [#uses=1]
store <4 x float> %863, <4 x float>* %862
br label %864
; <label>:864 ; preds = %861, %859
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:865 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x i32> %865, <4 x i32> undef, <4 x i32> < i32 2, i32 2, i32 2, i32 2 > ; <<4 x i32>>:866 [#uses=0]
br i1 false, label %868, label %867
; <label>:867 ; preds = %864
store <4 x float> zeroinitializer, <4 x float>* null
br label %868
; <label>:868 ; preds = %867, %864
shufflevector <4 x i32> zeroinitializer, <4 x i32> undef, <4 x i32> < i32 3, i32 3, i32 3, i32 3 > ; <<4 x i32>>:869 [#uses=0]
br label %xST.exit451
xST.exit451: ; preds = %868, %849
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 189, i32 0 ; <<4 x float>*>:870 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 189, i32 1 ; <<4 x float>*>:871 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:872 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:873 [#uses=1]
bitcast <4 x float> zeroinitializer to <4 x i32> ; <<4 x i32>>:874 [#uses=1]
xor <4 x i32> %874, < i32 -1, i32 -1, i32 -1, i32 -1 > ; <<4 x i32>>:875 [#uses=0]
bitcast <4 x float> %873 to <4 x i32> ; <<4 x i32>>:876 [#uses=1]
xor <4 x i32> %876, < i32 -1, i32 -1, i32 -1, i32 -1 > ; <<4 x i32>>:877 [#uses=0]
bitcast <4 x float> zeroinitializer to <4 x i32> ; <<4 x i32>>:878 [#uses=1]
xor <4 x i32> %878, < i32 -1, i32 -1, i32 -1, i32 -1 > ; <<4 x i32>>:879 [#uses=1]
bitcast <4 x i32> %879 to <4 x float> ; <<4 x float>>:880 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:881 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
icmp eq i32 0, 0 ; <i1>:882 [#uses=1]
br i1 %882, label %888, label %883
; <label>:883 ; preds = %xST.exit451
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 189, i32 0 ; <<4 x float>*>:884 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> zeroinitializer, <4 x float>* %884
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 189, i32 1 ; <<4 x float>*>:885 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:886 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 189, i32 3 ; <<4 x float>*>:887 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br label %xST.exit453
; <label>:888 ; preds = %xST.exit451
shufflevector <4 x i32> %881, <4 x i32> undef, <4 x i32> zeroinitializer ; <<4 x i32>>:889 [#uses=0]
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:890 [#uses=0]
br i1 false, label %894, label %891
; <label>:891 ; preds = %888
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 189, i32 1 ; <<4 x float>*>:892 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:893 [#uses=1]
store <4 x float> %893, <4 x float>* %892
br label %894
; <label>:894 ; preds = %891, %888
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:895 [#uses=1]
icmp eq i32 %895, 0 ; <i1>:896 [#uses=1]
br i1 %896, label %898, label %897
; <label>:897 ; preds = %894
br label %898
; <label>:898 ; preds = %897, %894
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:899 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br i1 false, label %xST.exit453, label %900
; <label>:900 ; preds = %898
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 189, i32 3 ; <<4 x float>*>:901 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %901 ; <<4 x float>>:902 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> %902, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:903 [#uses=0]
br label %xST.exit453
xST.exit453: ; preds = %900, %898, %883
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 189, i32 1 ; <<4 x float>*>:904 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:905 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 189, i32 3 ; <<4 x float>*>:906 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:907 [#uses=1]
shufflevector <4 x float> %905, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:908 [#uses=1]
bitcast <4 x float> zeroinitializer to <4 x i32> ; <<4 x i32>>:909 [#uses=0]
bitcast <4 x float> %908 to <4 x i32> ; <<4 x i32>>:910 [#uses=0]
bitcast <4 x float> %907 to <4 x i32> ; <<4 x i32>>:911 [#uses=0]
bitcast <4 x float> zeroinitializer to <4 x i32> ; <<4 x i32>>:912 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:913 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 2, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:914 [#uses=0]
br i1 false, label %915, label %xPIF.exit455
; <label>:915 ; preds = %xST.exit453
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:916 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [4 x <4 x i32>], [4 x <4 x i32>]* null, i32 0, i32 3 ; <<4 x i32>*>:917 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x i32> zeroinitializer, <4 x i32>* %917
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:918 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
and <4 x i32> %918, zeroinitializer ; <<4 x i32>>:919 [#uses=0]
br label %.critedge7899
.critedge7899: ; preds = %.critedge7899, %915
or <4 x i32> zeroinitializer, zeroinitializer ; <<4 x i32>>:920 [#uses=1]
br i1 false, label %.critedge7899, label %xPBRK.exit456
xPBRK.exit456: ; preds = %.critedge7899
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 2, <4 x i32> %920, <4 x i32> zeroinitializer ) ; <i32>:921 [#uses=0]
unreachable
xPIF.exit455: ; preds = %xST.exit453
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 186, i32 0 ; <<4 x float>*>:922 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %922 ; <<4 x float>>:923 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 186, i32 1 ; <<4 x float>*>:924 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %924 ; <<4 x float>>:925 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 186, i32 2 ; <<4 x float>*>:926 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 186, i32 3 ; <<4 x float>*>:927 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:928 [#uses=0]
bitcast { { i16, i16, i32 } }* %1 to <4 x float>* ; <<4 x float>*>:929 [#uses=0]
bitcast <4 x float> zeroinitializer to <4 x i32> ; <<4 x i32>>:930 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:931 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
icmp eq i32 0, 0 ; <i1>:932 [#uses=1]
br i1 %932, label %934, label %933
; <label>:933 ; preds = %xPIF.exit455
store <4 x float> zeroinitializer, <4 x float>* null
br label %934
; <label>:934 ; preds = %933, %xPIF.exit455
shufflevector <4 x i32> zeroinitializer, <4 x i32> undef, <4 x i32> < i32 1, i32 1, i32 1, i32 1 > ; <<4 x i32>>:935 [#uses=0]
icmp eq i32 0, 0 ; <i1>:936 [#uses=1]
br i1 %936, label %xST.exit459, label %937
; <label>:937 ; preds = %934
br label %xST.exit459
xST.exit459: ; preds = %937, %934
shufflevector <4 x i32> zeroinitializer, <4 x i32> undef, <4 x i32> < i32 2, i32 2, i32 2, i32 2 > ; <<4 x i32>>:938 [#uses=1]
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> %938, <4 x i32> zeroinitializer ) ; <i32>:939 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 188, i32 2 ; <<4 x float>*>:940 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> zeroinitializer, <4 x float>* %940
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:941 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> %941, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:942 [#uses=1]
store <4 x float> %942, <4 x float>* null
shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:943 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:944 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:945 [#uses=0]
br i1 false, label %947, label %946
; <label>:946 ; preds = %xST.exit459
br label %947
; <label>:947 ; preds = %946, %xST.exit459
shufflevector <4 x i32> zeroinitializer, <4 x i32> undef, <4 x i32> < i32 1, i32 1, i32 1, i32 1 > ; <<4 x i32>>:948 [#uses=0]
icmp eq i32 0, 0 ; <i1>:949 [#uses=1]
br i1 %949, label %952, label %950
; <label>:950 ; preds = %947
shufflevector <4 x i32> zeroinitializer, <4 x i32> undef, <4 x i32> zeroinitializer ; <<4 x i32>>:951 [#uses=1]
call void @llvm.ppc.altivec.stvewx( <4 x i32> %951, i8* null )
br label %952
; <label>:952 ; preds = %950, %947
br i1 false, label %955, label %953
; <label>:953 ; preds = %952
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [4 x <4 x i32>], [4 x <4 x i32>]* null, i32 0, i32 2 ; <<4 x i32>*>:954 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br label %955
; <label>:955 ; preds = %953, %952
shufflevector <4 x i32> zeroinitializer, <4 x i32> undef, <4 x i32> < i32 3, i32 3, i32 3, i32 3 > ; <<4 x i32>>:956 [#uses=0]
icmp eq i32 0, 0 ; <i1>:957 [#uses=1]
br i1 %957, label %xStoreDestAddressWithMask.exit461, label %958
; <label>:958 ; preds = %955
shufflevector <4 x i32> zeroinitializer, <4 x i32> undef, <4 x i32> zeroinitializer ; <<4 x i32>>:959 [#uses=1]
call void @llvm.ppc.altivec.stvewx( <4 x i32> %959, i8* null )
br label %xStoreDestAddressWithMask.exit461
xStoreDestAddressWithMask.exit461: ; preds = %958, %955
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %0 ; <<4 x float>>:960 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:961 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 3, i32 0 ; <<4 x float>*>:962 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br i1 false, label %968, label %xST.exit463
xST.exit463: ; preds = %xStoreDestAddressWithMask.exit461
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 3, i32 1 ; <<4 x float>*>:963 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 3, i32 2 ; <<4 x float>*>:964 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 3, i32 3 ; <<4 x float>*>:965 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %0 ; <<4 x float>>:966 [#uses=3]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:967 [#uses=0]
br i1 false, label %972, label %969
; <label>:968 ; preds = %xStoreDestAddressWithMask.exit461
unreachable
; <label>:969 ; preds = %xST.exit463
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 2, i32 1 ; <<4 x float>*>:970 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 2, i32 2 ; <<4 x float>*>:971 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> %966, <4 x float>* %971
store <4 x float> %966, <4 x float>* null
br label %xST.exit465
; <label>:972 ; preds = %xST.exit463
call <4 x i32> @llvm.ppc.altivec.vsel( <4 x i32> zeroinitializer, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <<4 x i32>>:973 [#uses=0]
store <4 x float> zeroinitializer, <4 x float>* null
store <4 x float> zeroinitializer, <4 x float>* null
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:974 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
bitcast <4 x float> %966 to <4 x i32> ; <<4 x i32>>:975 [#uses=1]
call <4 x i32> @llvm.ppc.altivec.vsel( <4 x i32> zeroinitializer, <4 x i32> %975, <4 x i32> zeroinitializer ) ; <<4 x i32>>:976 [#uses=1]
bitcast <4 x i32> %976 to <4 x float> ; <<4 x float>>:977 [#uses=1]
store <4 x float> %977, <4 x float>* null
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 2, i32 3 ; <<4 x float>*>:978 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
bitcast <4 x float> zeroinitializer to <4 x i32> ; <<4 x i32>>:979 [#uses=1]
call <4 x i32> @llvm.ppc.altivec.vsel( <4 x i32> %979, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <<4 x i32>>:980 [#uses=1]
bitcast <4 x i32> %980 to <4 x float> ; <<4 x float>>:981 [#uses=0]
br label %xST.exit465
xST.exit465: ; preds = %972, %969
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %0 ; <<4 x float>>:982 [#uses=3]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
icmp eq i32 0, 0 ; <i1>:983 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 0 ; <<4 x float>*>:984 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br i1 %983, label %989, label %985
; <label>:985 ; preds = %xST.exit465
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 1 ; <<4 x float>*>:986 [#uses=0]
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 2 ; <<4 x float>*>:987 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> %982, <4 x float>* %987
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 3 ; <<4 x float>*>:988 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br label %xST.exit467
; <label>:989 ; preds = %xST.exit465
bitcast <4 x float> %982 to <4 x i32> ; <<4 x i32>>:990 [#uses=0]
shufflevector <4 x i32> zeroinitializer, <4 x i32> undef, <4 x i32> zeroinitializer ; <<4 x i32>>:991 [#uses=0]
store <4 x float> zeroinitializer, <4 x float>* %984
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 1 ; <<4 x float>*>:992 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:993 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 2 ; <<4 x float>*>:994 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
bitcast <4 x i32> zeroinitializer to <4 x float> ; <<4 x float>>:995 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 1, i32 3 ; <<4 x float>*>:996 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
bitcast <4 x float> zeroinitializer to <4 x i32> ; <<4 x i32>>:997 [#uses=1]
bitcast <4 x float> %982 to <4 x i32> ; <<4 x i32>>:998 [#uses=1]
shufflevector <4 x i32> zeroinitializer, <4 x i32> undef, <4 x i32> < i32 3, i32 3, i32 3, i32 3 > ; <<4 x i32>>:999 [#uses=1]
call <4 x i32> @llvm.ppc.altivec.vsel( <4 x i32> %997, <4 x i32> %998, <4 x i32> %999 ) ; <<4 x i32>>:1000 [#uses=1]
bitcast <4 x i32> %1000 to <4 x float> ; <<4 x float>>:1001 [#uses=0]
br label %xST.exit467
xST.exit467: ; preds = %989, %985
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %0 ; <<4 x float>>:1002 [#uses=5]
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:1003 [#uses=2]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> %1003, <4 x i32> zeroinitializer ) ; <i32>:1004 [#uses=0]
br i1 false, label %1011, label %1005
; <label>:1005 ; preds = %xST.exit467
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:1006 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 1 ; <<4 x float>*>:1007 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %1007 ; <<4 x float>>:1008 [#uses=0]
load <4 x float>, <4 x float>* null ; <<4 x float>>:1009 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 3 ; <<4 x float>*>:1010 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br label %xST.exit469
; <label>:1011 ; preds = %xST.exit467
shufflevector <4 x i32> %1003, <4 x i32> undef, <4 x i32> zeroinitializer ; <<4 x i32>>:1012 [#uses=0]
icmp eq i32 0, 0 ; <i1>:1013 [#uses=1]
br i1 %1013, label %1015, label %1014
; <label>:1014 ; preds = %1011
br label %1015
; <label>:1015 ; preds = %1014, %1011
%.07472 = phi <4 x float> [ %1002, %1014 ], [ %.27474, %1011 ] ; <<4 x float>> [#uses=0]
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:1016 [#uses=1]
icmp eq i32 %1016, 0 ; <i1>:1017 [#uses=1]
br i1 %1017, label %1021, label %1018
; <label>:1018 ; preds = %1015
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 1 ; <<4 x float>*>:1019 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> %1002, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 1, i32 2, i32 7 > ; <<4 x float>>:1020 [#uses=0]
br label %1021
; <label>:1021 ; preds = %1018, %1015
%.07467 = phi <4 x float> [ %1002, %1018 ], [ %.27469, %1015 ] ; <<4 x float>> [#uses=2]
icmp eq i32 0, 0 ; <i1>:1022 [#uses=1]
br i1 %1022, label %1025, label %1023
; <label>:1023 ; preds = %1021
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 2 ; <<4 x float>*>:1024 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
store <4 x float> zeroinitializer, <4 x float>* %1024
br label %1025
; <label>:1025 ; preds = %1023, %1021
%.07462 = phi <4 x float> [ %1002, %1023 ], [ %.27464, %1021 ] ; <<4 x float>> [#uses=2]
icmp eq i32 0, 0 ; <i1>:1026 [#uses=1]
br i1 %1026, label %xST.exit469, label %1027
; <label>:1027 ; preds = %1025
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 3 ; <<4 x float>*>:1028 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br label %xST.exit469
xST.exit469: ; preds = %1027, %1025, %1005
%.17463 = phi <4 x float> [ %.27464, %1005 ], [ %.07462, %1027 ], [ %.07462, %1025 ] ; <<4 x float>> [#uses=1]
%.17468 = phi <4 x float> [ %.27469, %1005 ], [ %.07467, %1027 ], [ %.07467, %1025 ] ; <<4 x float>> [#uses=1]
%.07489 = phi <4 x float> [ %1002, %1005 ], [ %.17490, %1027 ], [ %.17490, %1025 ] ; <<4 x float>> [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:1029 [#uses=0]
load <4 x float>, <4 x float>* null ; <<4 x float>>:1030 [#uses=0]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fsub <4 x float> zeroinitializer, zeroinitializer ; <<4 x float>>:1031 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br i1 false, label %1037, label %1032
; <label>:1032 ; preds = %xST.exit469
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:1033 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 2 ; <<4 x float>*>:1034 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %1034 ; <<4 x float>>:1035 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 3 ; <<4 x float>*>:1036 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br label %xST.exit472
; <label>:1037 ; preds = %xST.exit469
icmp eq i32 0, 0 ; <i1>:1038 [#uses=1]
br i1 %1038, label %1040, label %1039
; <label>:1039 ; preds = %1037
br label %1040
; <label>:1040 ; preds = %1039, %1037
%.07507 = phi <4 x float> [ zeroinitializer, %1039 ], [ zeroinitializer, %1037 ] ; <<4 x float>> [#uses=0]
icmp eq i32 0, 0 ; <i1>:1041 [#uses=1]
br i1 %1041, label %1045, label %1042
; <label>:1042 ; preds = %1040
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 4, i32 1 ; <<4 x float>*>:1043 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %1043 ; <<4 x float>>:1044 [#uses=0]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br label %1045
; <label>:1045 ; preds = %1042, %1040
br i1 false, label %1048, label %1046
; <label>:1046 ; preds = %1045
shufflevector <4 x float> %1031, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 1, i32 2, i32 7 > ; <<4 x float>>:1047 [#uses=0]
br label %1048
; <label>:1048 ; preds = %1046, %1045
icmp eq i32 0, 0 ; <i1>:1049 [#uses=1]
br i1 %1049, label %xST.exit472, label %1050
; <label>:1050 ; preds = %1048
br label %xST.exit472
xST.exit472: ; preds = %1050, %1048, %1032
br i1 false, label %1052, label %1051
; <label>:1051 ; preds = %xST.exit472
br label %xST.exit474
; <label>:1052 ; preds = %xST.exit472
br i1 false, label %1054, label %1053
; <label>:1053 ; preds = %1052
br label %1054
; <label>:1054 ; preds = %1053, %1052
br i1 false, label %1056, label %1055
; <label>:1055 ; preds = %1054
br label %1056
; <label>:1056 ; preds = %1055, %1054
br i1 false, label %1058, label %1057
; <label>:1057 ; preds = %1056
br label %1058
; <label>:1058 ; preds = %1057, %1056
br i1 false, label %xST.exit474, label %1059
; <label>:1059 ; preds = %1058
br label %xST.exit474
xST.exit474: ; preds = %1059, %1058, %1051
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:1060 [#uses=1]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fmul <4 x float> zeroinitializer, zeroinitializer ; <<4 x float>>:1061 [#uses=1]
fmul <4 x float> %1060, zeroinitializer ; <<4 x float>>:1062 [#uses=2]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br i1 false, label %1065, label %1063
; <label>:1063 ; preds = %xST.exit474
shufflevector <4 x float> %1062, <4 x float> zeroinitializer, <4 x i32> < i32 0, i32 1, i32 2, i32 7 > ; <<4 x float>>:1064 [#uses=1]
store <4 x float> %1064, <4 x float>* null
br label %xST.exit476
; <label>:1065 ; preds = %xST.exit474
br i1 false, label %1067, label %1066
; <label>:1066 ; preds = %1065
br label %1067
; <label>:1067 ; preds = %1066, %1065
shufflevector <4 x i32> zeroinitializer, <4 x i32> undef, <4 x i32> < i32 1, i32 1, i32 1, i32 1 > ; <<4 x i32>>:1068 [#uses=0]
br i1 false, label %1070, label %1069
; <label>:1069 ; preds = %1067
br label %1070
; <label>:1070 ; preds = %1069, %1067
br i1 false, label %1072, label %1071
; <label>:1071 ; preds = %1070
br label %1072
; <label>:1072 ; preds = %1071, %1070
br i1 false, label %xST.exit476, label %1073
; <label>:1073 ; preds = %1072
br label %xST.exit476
xST.exit476: ; preds = %1073, %1072, %1063
%.07551 = phi <4 x float> [ %1062, %1063 ], [ %.17552, %1073 ], [ %.17552, %1072 ] ; <<4 x float>> [#uses=1]
%.07555 = phi <4 x float> [ %1061, %1063 ], [ %.17556, %1073 ], [ %.17556, %1072 ] ; <<4 x float>> [#uses=1]
br i1 false, label %1075, label %1074
; <label>:1074 ; preds = %xST.exit476
br label %xST.exit479
; <label>:1075 ; preds = %xST.exit476
br i1 false, label %1077, label %1076
; <label>:1076 ; preds = %1075
br label %1077
; <label>:1077 ; preds = %1076, %1075
br i1 false, label %1079, label %1078
; <label>:1078 ; preds = %1077
br label %1079
; <label>:1079 ; preds = %1078, %1077
br i1 false, label %1081, label %1080
; <label>:1080 ; preds = %1079
br label %1081
; <label>:1081 ; preds = %1080, %1079
br i1 false, label %xST.exit479, label %1082
; <label>:1082 ; preds = %1081
br label %xST.exit479
xST.exit479: ; preds = %1082, %1081, %1074
br i1 false, label %1084, label %1083
; <label>:1083 ; preds = %xST.exit479
br label %xST.exit482
; <label>:1084 ; preds = %xST.exit479
br i1 false, label %1086, label %1085
; <label>:1085 ; preds = %1084
br label %1086
; <label>:1086 ; preds = %1085, %1084
br i1 false, label %1088, label %1087
; <label>:1087 ; preds = %1086
br label %1088
; <label>:1088 ; preds = %1087, %1086
br i1 false, label %1090, label %1089
; <label>:1089 ; preds = %1088
br label %1090
; <label>:1090 ; preds = %1089, %1088
br i1 false, label %xST.exit482, label %1091
; <label>:1091 ; preds = %1090
br label %xST.exit482
xST.exit482: ; preds = %1091, %1090, %1083
br i1 false, label %1093, label %1092
; <label>:1092 ; preds = %xST.exit482
br label %xST.exit486
; <label>:1093 ; preds = %xST.exit482
br i1 false, label %1095, label %1094
; <label>:1094 ; preds = %1093
br label %1095
; <label>:1095 ; preds = %1094, %1093
br i1 false, label %1097, label %1096
; <label>:1096 ; preds = %1095
br label %1097
; <label>:1097 ; preds = %1096, %1095
br i1 false, label %1099, label %1098
; <label>:1098 ; preds = %1097
br label %1099
; <label>:1099 ; preds = %1098, %1097
br i1 false, label %xST.exit486, label %1100
; <label>:1100 ; preds = %1099
br label %xST.exit486
xST.exit486: ; preds = %1100, %1099, %1092
br i1 false, label %1102, label %1101
; <label>:1101 ; preds = %xST.exit486
br label %xST.exit489
; <label>:1102 ; preds = %xST.exit486
br i1 false, label %1104, label %1103
; <label>:1103 ; preds = %1102
br label %1104
; <label>:1104 ; preds = %1103, %1102
br i1 false, label %1106, label %1105
; <label>:1105 ; preds = %1104
br label %1106
; <label>:1106 ; preds = %1105, %1104
br i1 false, label %1108, label %1107
; <label>:1107 ; preds = %1106
br label %1108
; <label>:1108 ; preds = %1107, %1106
br i1 false, label %xST.exit489, label %1109
; <label>:1109 ; preds = %1108
br label %xST.exit489
xST.exit489: ; preds = %1109, %1108, %1101
br i1 false, label %1111, label %1110
; <label>:1110 ; preds = %xST.exit489
br label %xST.exit492
; <label>:1111 ; preds = %xST.exit489
br i1 false, label %1113, label %1112
; <label>:1112 ; preds = %1111
br label %1113
; <label>:1113 ; preds = %1112, %1111
br i1 false, label %1115, label %1114
; <label>:1114 ; preds = %1113
br label %1115
; <label>:1115 ; preds = %1114, %1113
br i1 false, label %1117, label %1116
; <label>:1116 ; preds = %1115
br label %1117
; <label>:1117 ; preds = %1116, %1115
br i1 false, label %xST.exit492, label %1118
; <label>:1118 ; preds = %1117
br label %xST.exit492
xST.exit492: ; preds = %1118, %1117, %1110
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:1119 [#uses=1]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fmul <4 x float> %1119, zeroinitializer ; <<4 x float>>:1120 [#uses=1]
fmul <4 x float> zeroinitializer, zeroinitializer ; <<4 x float>>:1121 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br i1 false, label %1123, label %1122
; <label>:1122 ; preds = %xST.exit492
br label %xST.exit495
; <label>:1123 ; preds = %xST.exit492
br i1 false, label %1125, label %1124
; <label>:1124 ; preds = %1123
br label %1125
; <label>:1125 ; preds = %1124, %1123
br i1 false, label %1127, label %1126
; <label>:1126 ; preds = %1125
br label %1127
; <label>:1127 ; preds = %1126, %1125
br i1 false, label %1129, label %1128
; <label>:1128 ; preds = %1127
br label %1129
; <label>:1129 ; preds = %1128, %1127
br i1 false, label %xST.exit495, label %1130
; <label>:1130 ; preds = %1129
br label %xST.exit495
xST.exit495: ; preds = %1130, %1129, %1122
%.07582 = phi <4 x float> [ %1121, %1122 ], [ %.17583, %1130 ], [ %.17583, %1129 ] ; <<4 x float>> [#uses=1]
%.07590 = phi <4 x float> [ %1120, %1122 ], [ %.17591, %1130 ], [ %.17591, %1129 ] ; <<4 x float>> [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:1131 [#uses=1]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fadd <4 x float> %1131, zeroinitializer ; <<4 x float>>:1132 [#uses=1]
fadd <4 x float> zeroinitializer, zeroinitializer ; <<4 x float>>:1133 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br i1 false, label %1135, label %1134
; <label>:1134 ; preds = %xST.exit495
br label %xST.exit498
; <label>:1135 ; preds = %xST.exit495
br i1 false, label %1137, label %1136
; <label>:1136 ; preds = %1135
br label %1137
; <label>:1137 ; preds = %1136, %1135
br i1 false, label %1139, label %1138
; <label>:1138 ; preds = %1137
br label %1139
; <label>:1139 ; preds = %1138, %1137
br i1 false, label %1141, label %1140
; <label>:1140 ; preds = %1139
br label %1141
; <label>:1141 ; preds = %1140, %1139
br i1 false, label %xST.exit498, label %1142
; <label>:1142 ; preds = %1141
br label %xST.exit498
xST.exit498: ; preds = %1142, %1141, %1134
%.07617 = phi <4 x float> [ %1133, %1134 ], [ %.17618, %1142 ], [ %.17618, %1141 ] ; <<4 x float>> [#uses=1]
%.07621 = phi <4 x float> [ %1132, %1134 ], [ %.17622, %1142 ], [ %.17622, %1141 ] ; <<4 x float>> [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:1143 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 2 ; <<4 x float>*>:1144 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %1144 ; <<4 x float>>:1145 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 3 ; <<4 x float>*>:1146 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %1146 ; <<4 x float>>:1147 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> %1143, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:1148 [#uses=1]
shufflevector <4 x float> %1145, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:1149 [#uses=1]
shufflevector <4 x float> %1147, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>>:1150 [#uses=1]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fmul <4 x float> zeroinitializer, zeroinitializer ; <<4 x float>>:1151 [#uses=1]
fmul <4 x float> zeroinitializer, %1148 ; <<4 x float>>:1152 [#uses=1]
fmul <4 x float> zeroinitializer, %1149 ; <<4 x float>>:1153 [#uses=1]
fmul <4 x float> zeroinitializer, %1150 ; <<4 x float>>:1154 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br i1 false, label %1156, label %1155
; <label>:1155 ; preds = %xST.exit498
br label %xST.exit501
; <label>:1156 ; preds = %xST.exit498
br i1 false, label %1158, label %1157
; <label>:1157 ; preds = %1156
br label %1158
; <label>:1158 ; preds = %1157, %1156
br i1 false, label %1160, label %1159
; <label>:1159 ; preds = %1158
br label %1160
; <label>:1160 ; preds = %1159, %1158
br i1 false, label %1162, label %1161
; <label>:1161 ; preds = %1160
br label %1162
; <label>:1162 ; preds = %1161, %1160
br i1 false, label %xST.exit501, label %1163
; <label>:1163 ; preds = %1162
br label %xST.exit501
xST.exit501: ; preds = %1163, %1162, %1155
%.07652 = phi <4 x float> [ %1154, %1155 ], [ %.17653, %1163 ], [ %.17653, %1162 ] ; <<4 x float>> [#uses=1]
%.07656 = phi <4 x float> [ %1153, %1155 ], [ %.17657, %1163 ], [ %.17657, %1162 ] ; <<4 x float>> [#uses=1]
%.07660 = phi <4 x float> [ %1152, %1155 ], [ %.17661, %1163 ], [ %.17661, %1162 ] ; <<4 x float>> [#uses=1]
%.07664 = phi <4 x float> [ %1151, %1155 ], [ %.17665, %1163 ], [ %.17665, %1162 ] ; <<4 x float>> [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:1164 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 2 ; <<4 x float>*>:1165 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %1165 ; <<4 x float>>:1166 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr [193 x [4 x <4 x float>]], [193 x [4 x <4 x float>]]* null, i32 0, i32 0, i32 3 ; <<4 x float>*>:1167 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* %1167 ; <<4 x float>>:1168 [#uses=1]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fadd <4 x float> zeroinitializer, zeroinitializer ; <<4 x float>>:1169 [#uses=1]
fadd <4 x float> zeroinitializer, %1164 ; <<4 x float>>:1170 [#uses=1]
fadd <4 x float> zeroinitializer, %1166 ; <<4 x float>>:1171 [#uses=1]
fadd <4 x float> zeroinitializer, %1168 ; <<4 x float>>:1172 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br i1 false, label %1174, label %1173
; <label>:1173 ; preds = %xST.exit501
br label %xST.exit504
; <label>:1174 ; preds = %xST.exit501
br i1 false, label %1176, label %1175
; <label>:1175 ; preds = %1174
br label %1176
; <label>:1176 ; preds = %1175, %1174
br i1 false, label %1178, label %1177
; <label>:1177 ; preds = %1176
br label %1178
; <label>:1178 ; preds = %1177, %1176
br i1 false, label %1180, label %1179
; <label>:1179 ; preds = %1178
br label %1180
; <label>:1180 ; preds = %1179, %1178
br i1 false, label %xST.exit504, label %1181
; <label>:1181 ; preds = %1180
br label %xST.exit504
xST.exit504: ; preds = %1181, %1180, %1173
%.07722 = phi <4 x float> [ %1172, %1173 ], [ %.17723, %1181 ], [ %.17723, %1180 ] ; <<4 x float>> [#uses=1]
%.07726 = phi <4 x float> [ %1171, %1173 ], [ %.17727, %1181 ], [ %.17727, %1180 ] ; <<4 x float>> [#uses=1]
%.07730 = phi <4 x float> [ %1170, %1173 ], [ %.17731, %1181 ], [ %.17731, %1180 ] ; <<4 x float>> [#uses=1]
%.07734 = phi <4 x float> [ %1169, %1173 ], [ %.17735, %1181 ], [ %.17735, %1180 ] ; <<4 x float>> [#uses=1]
Split the Add, Sub, and Mul instruction opcodes into separate integer and floating-point opcodes, introducing FAdd, FSub, and FMul. For now, the AsmParser, BitcodeReader, and IRBuilder all preserve backwards compatability, and the Core LLVM APIs preserve backwards compatibility for IR producers. Most front-ends won't need to change immediately. This implements the first step of the plan outlined here: http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt llvm-svn: 72897
2009-06-05 06:49:04 +08:00
fadd <4 x float> zeroinitializer, zeroinitializer ; <<4 x float>>:1182 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
br i1 false, label %1184, label %1183
; <label>:1183 ; preds = %xST.exit504
br label %xST.exit507
; <label>:1184 ; preds = %xST.exit504
br i1 false, label %1186, label %1185
; <label>:1185 ; preds = %1184
br label %1186
; <label>:1186 ; preds = %1185, %1184
br i1 false, label %1188, label %1187
; <label>:1187 ; preds = %1186
store <4 x float> zeroinitializer, <4 x float>* null
br label %1188
; <label>:1188 ; preds = %1187, %1186
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:1189 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x i32> %1189, <4 x i32> undef, <4 x i32> < i32 2, i32 2, i32 2, i32 2 > ; <<4 x i32>>:1190 [#uses=1]
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> %1190, <4 x i32> zeroinitializer ) ; <i32>:1191 [#uses=1]
icmp eq i32 %1191, 0 ; <i1>:1192 [#uses=1]
br i1 %1192, label %1196, label %1193
; <label>:1193 ; preds = %1188
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:1194 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> %1194, <4 x i32> < i32 0, i32 1, i32 2, i32 7 > ; <<4 x float>>:1195 [#uses=1]
store <4 x float> %1195, <4 x float>* null
br label %1196
; <label>:1196 ; preds = %1193, %1188
%.07742 = phi <4 x float> [ zeroinitializer, %1193 ], [ zeroinitializer, %1188 ] ; <<4 x float>> [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:1197 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x i32> %1197, <4 x i32> undef, <4 x i32> < i32 3, i32 3, i32 3, i32 3 > ; <<4 x i32>>:1198 [#uses=1]
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> %1198, <4 x i32> zeroinitializer ) ; <i32>:1199 [#uses=1]
icmp eq i32 %1199, 0 ; <i1>:1200 [#uses=1]
br i1 %1200, label %xST.exit507, label %1201
; <label>:1201 ; preds = %1196
store <4 x float> zeroinitializer, <4 x float>* null
br label %xST.exit507
xST.exit507: ; preds = %1201, %1196, %1183
%.07769 = phi <4 x float> [ %1182, %1183 ], [ %.17770, %1201 ], [ %.17770, %1196 ] ; <<4 x float>> [#uses=1]
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> zeroinitializer, <4 x i32> zeroinitializer ) ; <i32>:1202 [#uses=1]
icmp eq i32 %1202, 0 ; <i1>:1203 [#uses=1]
br i1 %1203, label %1207, label %1204
; <label>:1204 ; preds = %xST.exit507
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:1205 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> %1205, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:1206 [#uses=1]
store <4 x float> %1206, <4 x float>* null
br label %1207
; <label>:1207 ; preds = %1204, %xST.exit507
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x i32>, <4 x i32>* %.sub7896 ; <<4 x i32>>:1208 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x i32> %1208, <4 x i32> undef, <4 x i32> < i32 1, i32 1, i32 1, i32 1 > ; <<4 x i32>>:1209 [#uses=1]
call i32 @llvm.ppc.altivec.vcmpequw.p( i32 0, <4 x i32> %1209, <4 x i32> zeroinitializer ) ; <i32>:1210 [#uses=1]
icmp eq i32 %1210, 0 ; <i1>:1211 [#uses=1]
br i1 %1211, label %1215, label %1212
; <label>:1212 ; preds = %1207
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load <4 x float>, <4 x float>* null ; <<4 x float>>:1213 [#uses=1]
New test case. llvm-svn: 35515
2007-03-31 04:41:18 +08:00
shufflevector <4 x float> zeroinitializer, <4 x float> %1213, <4 x i32> < i32 0, i32 5, i32 6, i32 7 > ; <<4 x float>>:1214 [#uses=1]
store <4 x float> %1214, <4 x float>* null
br label %1215
; <label>:1215 ; preds = %1212, %1207
store <4 x float> zeroinitializer, <4 x float>* null
br label %xLS.exit449
}
declare <4 x i32> @llvm.ppc.altivec.vsel(<4 x i32>, <4 x i32>, <4 x i32>)
declare void @llvm.ppc.altivec.stvewx(<4 x i32>, i8*)
declare <4 x float> @llvm.ppc.altivec.vrsqrtefp(<4 x float>)
declare <4 x float> @llvm.ppc.altivec.vcfsx(<4 x i32>, i32)
declare i32 @llvm.ppc.altivec.vcmpequw.p(i32, <4 x i32>, <4 x i32>)
declare <4 x i32> @llvm.ppc.altivec.vcmpgtfp(<4 x float>, <4 x float>)