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/X86/psubus.ll

351 lines
12 KiB
LLVM
Raw Normal View History

[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
; RUN: llc -mcpu=core2 < %s | FileCheck %s -check-prefix=SSSE3
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
; RUN: llc -mcpu=corei7-avx < %s | FileCheck %s -check-prefix=AVX1
; RUN: llc -mcpu=core-avx2 < %s | FileCheck %s -check-prefix=AVX2
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
target triple = "x86_64-apple-macosx10.8.0"
define void @test1(i16* nocapture %head) nounwind {
vector.ph:
[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
%0 = getelementptr inbounds i16, i16* %head, i64 0
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%1 = bitcast i16* %0 to <8 x i16>*
[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
%2 = load <8 x i16>, <8 x i16>* %1, align 2
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%3 = icmp slt <8 x i16> %2, zeroinitializer
%4 = xor <8 x i16> %2, <i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768>
%5 = select <8 x i1> %3, <8 x i16> %4, <8 x i16> zeroinitializer
store <8 x i16> %5, <8 x i16>* %1, align 2
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
ret void
; SSSE3: @test1
; SSSE3: # BB#0:
; SSSE3-NEXT: movdqu (%rdi), %xmm0
; SSSE3-NEXT: psubusw LCPI0_0(%rip), %xmm0
; SSSE3-NEXT: movdqu %xmm0, (%rdi)
; SSSE3-NEXT: retq
; AVX1: @test1
; AVX1: # BB#0:
; AVX1-NEXT: vmovdqu (%rdi), %xmm0
; AVX1-NEXT: vpsubusw LCPI0_0(%rip), %xmm0, %xmm0
; AVX1-NEXT: vmovdqu %xmm0, (%rdi)
; AVX1-NEXT: retq
; AVX2: @test1
; AVX2: # BB#0:
; AVX2-NEXT: vmovdqu (%rdi), %xmm0
; AVX2-NEXT: vpsubusw LCPI0_0(%rip), %xmm0, %xmm0
; AVX2-NEXT: vmovdqu %xmm0, (%rdi)
; AVX2-NEXT: retq
}
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
define void @test2(i16* nocapture %head) nounwind {
vector.ph:
[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
%0 = getelementptr inbounds i16, i16* %head, i64 0
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%1 = bitcast i16* %0 to <8 x i16>*
[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
%2 = load <8 x i16>, <8 x i16>* %1, align 2
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%3 = icmp ugt <8 x i16> %2, <i16 32766, i16 32766, i16 32766, i16 32766, i16 32766, i16 32766, i16 32766, i16 32766>
%4 = add <8 x i16> %2, <i16 -32767, i16 -32767, i16 -32767, i16 -32767, i16 -32767, i16 -32767, i16 -32767, i16 -32767>
%5 = select <8 x i1> %3, <8 x i16> %4, <8 x i16> zeroinitializer
store <8 x i16> %5, <8 x i16>* %1, align 2
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
ret void
; SSSE3: @test2
; SSSE3: # BB#0:
; SSSE3-NEXT: movdqu (%rdi), %xmm0
; SSSE3-NEXT: psubusw LCPI1_0(%rip), %xmm0
; SSSE3-NEXT: movdqu %xmm0, (%rdi)
; SSSE3-NEXT: retq
; AVX1: @test2
; AVX1: # BB#0:
; AVX1-NEXT: vmovdqu (%rdi), %xmm0
; AVX1-NEXT: vpsubusw LCPI1_0(%rip), %xmm0, %xmm0
; AVX1-NEXT: vmovdqu %xmm0, (%rdi)
; AVX1-NEXT: retq
; AVX2: @test2
; AVX2: # BB#0:
; AVX2-NEXT: vmovdqu (%rdi), %xmm0
; AVX2-NEXT: vpsubusw LCPI1_0(%rip), %xmm0, %xmm0
; AVX2-NEXT: vmovdqu %xmm0, (%rdi)
; AVX2-NEXT: retq
}
define void @test3(i16* nocapture %head, i16 zeroext %w) nounwind {
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
vector.ph:
%0 = insertelement <8 x i16> undef, i16 %w, i32 0
%broadcast15 = shufflevector <8 x i16> %0, <8 x i16> undef, <8 x i32> zeroinitializer
[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
%1 = getelementptr inbounds i16, i16* %head, i64 0
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%2 = bitcast i16* %1 to <8 x i16>*
[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
%3 = load <8 x i16>, <8 x i16>* %2, align 2
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%4 = icmp ult <8 x i16> %3, %broadcast15
%5 = sub <8 x i16> %3, %broadcast15
%6 = select <8 x i1> %4, <8 x i16> zeroinitializer, <8 x i16> %5
store <8 x i16> %6, <8 x i16>* %2, align 2
ret void
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
; SSSE3: @test3
; SSSE3: # BB#0:
; SSSE3-NEXT: movd %esi, %xmm0
; SSSE3-NEXT: pshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; SSSE3-NEXT: movdqu (%rdi), %xmm1
; SSSE3-NEXT: psubusw %xmm0, %xmm1
; SSSE3-NEXT: movdqu %xmm1, (%rdi)
; SSSE3-NEXT: retq
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
; AVX1: @test3
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
; AVX1: # BB#0:
; AVX1-NEXT: vmovd %esi, %xmm0
; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; AVX1-NEXT: vmovdqu (%rdi), %xmm1
; AVX1-NEXT: vpsubusw %xmm0, %xmm1, %xmm0
; AVX1-NEXT: vmovdqu %xmm0, (%rdi)
; AVX1-NEXT: retq
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
; AVX2: @test3
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
; AVX2: # BB#0:
; AVX2-NEXT: vmovd %esi, %xmm0
; AVX2-NEXT: vpbroadcastw %xmm0, %xmm0
; AVX2-NEXT: vmovdqu (%rdi), %xmm1
; AVX2-NEXT: vpsubusw %xmm0, %xmm1, %xmm0
; AVX2-NEXT: vmovdqu %xmm0, (%rdi)
; AVX2-NEXT: retq
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
}
define void @test4(i8* nocapture %head) nounwind {
vector.ph:
[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
%0 = getelementptr inbounds i8, i8* %head, i64 0
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%1 = bitcast i8* %0 to <16 x i8>*
[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
%2 = load <16 x i8>, <16 x i8>* %1, align 1
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%3 = icmp slt <16 x i8> %2, zeroinitializer
%4 = xor <16 x i8> %2, <i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128>
%5 = select <16 x i1> %3, <16 x i8> %4, <16 x i8> zeroinitializer
store <16 x i8> %5, <16 x i8>* %1, align 1
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
ret void
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
; SSSE3: @test4
; SSSE3: # BB#0:
; SSSE3-NEXT: movdqu (%rdi), %xmm0
; SSSE3-NEXT: psubusb LCPI3_0(%rip), %xmm0
; SSSE3-NEXT: movdqu %xmm0, (%rdi)
; SSSE3-NEXT: retq
; AVX1: @test4
; AVX1: # BB#0:
; AVX1-NEXT: vmovdqu (%rdi), %xmm0
; AVX1-NEXT: vpsubusb LCPI3_0(%rip), %xmm0, %xmm0
; AVX1-NEXT: vmovdqu %xmm0, (%rdi)
; AVX1-NEXT: retq
; AVX2: @test4
; AVX2: # BB#0:
; AVX2-NEXT: vmovdqu (%rdi), %xmm0
; AVX2-NEXT: vpsubusb LCPI3_0(%rip), %xmm0, %xmm0
; AVX2-NEXT: vmovdqu %xmm0, (%rdi)
; AVX2-NEXT: retq
}
define void @test5(i8* nocapture %head) nounwind {
vector.ph:
[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
%0 = getelementptr inbounds i8, i8* %head, i64 0
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%1 = bitcast i8* %0 to <16 x i8>*
[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
%2 = load <16 x i8>, <16 x i8>* %1, align 1
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%3 = icmp ugt <16 x i8> %2, <i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126>
%4 = add <16 x i8> %2, <i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127>
%5 = select <16 x i1> %3, <16 x i8> %4, <16 x i8> zeroinitializer
store <16 x i8> %5, <16 x i8>* %1, align 1
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
ret void
; SSSE3: @test5
; SSSE3: # BB#0:
; SSSE3-NEXT: movdqu (%rdi), %xmm0
; SSSE3-NEXT: psubusb LCPI4_0(%rip), %xmm0
; SSSE3-NEXT: movdqu %xmm0, (%rdi)
; SSSE3-NEXT: retq
; AVX1: @test5
; AVX1: # BB#0:
; AVX1-NEXT: vmovdqu (%rdi), %xmm0
; AVX1-NEXT: vpsubusb LCPI4_0(%rip), %xmm0
; AVX1-NEXT: vmovdqu %xmm0, (%rdi)
; AVX1-NEXT: retq
; AVX2: @test5
; AVX2: # BB#0:
; AVX2-NEXT: vmovdqu (%rdi), %xmm0
; AVX2-NEXT: vpsubusb LCPI4_0(%rip), %xmm0
; AVX2-NEXT: vmovdqu %xmm0, (%rdi)
; AVX2-NEXT: retq
}
define void @test6(i8* nocapture %head, i8 zeroext %w) nounwind {
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
vector.ph:
%0 = insertelement <16 x i8> undef, i8 %w, i32 0
%broadcast15 = shufflevector <16 x i8> %0, <16 x i8> undef, <16 x i32> zeroinitializer
[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
%1 = getelementptr inbounds i8, i8* %head, i64 0
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%2 = bitcast i8* %1 to <16 x i8>*
[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
%3 = load <16 x i8>, <16 x i8>* %2, align 1
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%4 = icmp ult <16 x i8> %3, %broadcast15
%5 = sub <16 x i8> %3, %broadcast15
%6 = select <16 x i1> %4, <16 x i8> zeroinitializer, <16 x i8> %5
store <16 x i8> %6, <16 x i8>* %2, align 1
ret void
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
; SSSE3: @test6
; SSSE3: # BB#0:
; SSSE3-NEXT: movd %esi, %xmm0
; SSSE3-NEXT: pxor %xmm1, %xmm1
; SSSE3-NEXT: pshufb %xmm1, %xmm0
; SSSE3-NEXT: movdqu (%rdi), %xmm1
; SSSE3-NEXT: psubusb %xmm0, %xmm1
; SSSE3-NEXT: movdqu %xmm1, (%rdi)
; SSSE3-NEXT: retq
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
; AVX1: @test6
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
; AVX1: # BB#0:
; AVX1-NEXT: vmovd %esi, %xmm0
; AVX1-NEXT: vpxor %xmm1, %xmm1
; AVX1-NEXT: vpshufb %xmm1, %xmm0
; AVX1-NEXT: vmovdqu (%rdi), %xmm1
; AVX1-NEXT: vpsubusb %xmm0, %xmm1, %xmm0
; AVX1-NEXT: vmovdqu %xmm0, (%rdi)
; AVX1-NEXT: retq
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
; AVX2: @test6
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
; AVX2: # BB#0:
; AVX2-NEXT: vmovd %esi, %xmm0
; AVX2-NEXT: vpbroadcastb %xmm0, %xmm0
; AVX2-NEXT: vmovdqu (%rdi), %xmm1
; AVX2-NEXT: vpsubusb %xmm0, %xmm1, %xmm0
; AVX2-NEXT: vmovdqu %xmm0, (%rdi)
; AVX2-NEXT: retq
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
}
define void @test7(i16* nocapture %head) nounwind {
vector.ph:
[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
%0 = getelementptr inbounds i16, i16* %head, i64 0
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%1 = bitcast i16* %0 to <16 x i16>*
[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
%2 = load <16 x i16>, <16 x i16>* %1, align 2
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%3 = icmp slt <16 x i16> %2, zeroinitializer
%4 = xor <16 x i16> %2, <i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768>
%5 = select <16 x i1> %3, <16 x i16> %4, <16 x i16> zeroinitializer
store <16 x i16> %5, <16 x i16>* %1, align 2
ret void
; AVX2: @test7
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
; AVX2: # BB#0:
; AVX2-NEXT: vmovdqu (%rdi), %ymm0
; AVX2-NEXT: vpsubusw LCPI6_0(%rip), %ymm0, %ymm0
; AVX2-NEXT: vmovdqu %ymm0, (%rdi)
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
}
define void @test8(i16* nocapture %head) nounwind {
vector.ph:
[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
%0 = getelementptr inbounds i16, i16* %head, i64 0
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%1 = bitcast i16* %0 to <16 x i16>*
[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
%2 = load <16 x i16>, <16 x i16>* %1, align 2
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%3 = icmp ugt <16 x i16> %2, <i16 32766, i16 32766, i16 32766, i16 32766, i16 32766, i16 32766, i16 32766, i16 32766, i16 32766, i16 32766, i16 32766, i16 32766, i16 32766, i16 32766, i16 32766, i16 32766>
%4 = add <16 x i16> %2, <i16 -32767, i16 -32767, i16 -32767, i16 -32767, i16 -32767, i16 -32767, i16 -32767, i16 -32767, i16 -32767, i16 -32767, i16 -32767, i16 -32767, i16 -32767, i16 -32767, i16 -32767, i16 -32767>
%5 = select <16 x i1> %3, <16 x i16> %4, <16 x i16> zeroinitializer
store <16 x i16> %5, <16 x i16>* %1, align 2
ret void
; AVX2: @test8
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
; AVX2: # BB#0:
; AVX2-NEXT: vmovdqu (%rdi), %ymm0
; AVX2-NEXT: vpsubusw LCPI7_0(%rip), %ymm0, %ymm0
; AVX2-NEXT: vmovdqu %ymm0, (%rdi)
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
}
define void @test9(i16* nocapture %head, i16 zeroext %w) nounwind {
vector.ph:
%0 = insertelement <16 x i16> undef, i16 %w, i32 0
%broadcast15 = shufflevector <16 x i16> %0, <16 x i16> undef, <16 x i32> zeroinitializer
[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
%1 = getelementptr inbounds i16, i16* %head, i64 0
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%2 = bitcast i16* %1 to <16 x i16>*
[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
%3 = load <16 x i16>, <16 x i16>* %2, align 2
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%4 = icmp ult <16 x i16> %3, %broadcast15
%5 = sub <16 x i16> %3, %broadcast15
%6 = select <16 x i1> %4, <16 x i16> zeroinitializer, <16 x i16> %5
store <16 x i16> %6, <16 x i16>* %2, align 2
ret void
; AVX2: @test9
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
; AVX2: # BB#0:
; AVX2-NEXT: vmovd %esi, %xmm0
; AVX2-NEXT: vpbroadcastw %xmm0, %ymm0
; AVX2-NEXT: vmovdqu (%rdi), %ymm1
; AVX2-NEXT: vpsubusw %ymm0, %ymm1, %ymm0
; AVX2-NEXT: vmovdqu %ymm0, (%rdi)
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
}
define void @test10(i8* nocapture %head) nounwind {
vector.ph:
[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
%0 = getelementptr inbounds i8, i8* %head, i64 0
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%1 = bitcast i8* %0 to <32 x i8>*
[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
%2 = load <32 x i8>, <32 x i8>* %1, align 1
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%3 = icmp slt <32 x i8> %2, zeroinitializer
%4 = xor <32 x i8> %2, <i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128, i8 -128>
%5 = select <32 x i1> %3, <32 x i8> %4, <32 x i8> zeroinitializer
store <32 x i8> %5, <32 x i8>* %1, align 1
ret void
; AVX2: @test10
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
; AVX2: # BB#0:
; AVX2-NEXT: vmovdqu (%rdi), %ymm0
; AVX2-NEXT: vpsubusb LCPI9_0(%rip), %ymm0, %ymm0
; AVX2-NEXT: vmovdqu %ymm0, (%rdi)
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
}
define void @test11(i8* nocapture %head) nounwind {
vector.ph:
[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
%0 = getelementptr inbounds i8, i8* %head, i64 0
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%1 = bitcast i8* %0 to <32 x i8>*
[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
%2 = load <32 x i8>, <32 x i8>* %1, align 1
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%3 = icmp ugt <32 x i8> %2, <i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126, i8 126>
%4 = add <32 x i8> %2, <i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127, i8 -127>
%5 = select <32 x i1> %3, <32 x i8> %4, <32 x i8> zeroinitializer
store <32 x i8> %5, <32 x i8>* %1, align 1
ret void
; AVX2: @test11
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
; AVX2: # BB#0:
; AVX2-NEXT: vmovdqu (%rdi), %ymm0
; AVX2-NEXT: vpsubusb LCPI10_0(%rip), %ymm0, %ymm0
; AVX2-NEXT: vmovdqu %ymm0, (%rdi)
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
}
define void @test12(i8* nocapture %head, i8 zeroext %w) nounwind {
vector.ph:
%0 = insertelement <32 x i8> undef, i8 %w, i32 0
%broadcast15 = shufflevector <32 x i8> %0, <32 x i8> undef, <32 x i32> zeroinitializer
[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
%1 = getelementptr inbounds i8, i8* %head, i64 0
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%2 = bitcast i8* %1 to <32 x i8>*
[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
%3 = load <32 x i8>, <32 x i8>* %2, align 1
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
%4 = icmp ult <32 x i8> %3, %broadcast15
%5 = sub <32 x i8> %3, %broadcast15
%6 = select <32 x i1> %4, <32 x i8> zeroinitializer, <32 x i8> %5
store <32 x i8> %6, <32 x i8>* %2, align 1
ret void
; AVX2: @test12
[X86][SSE] Simplified PSUBUS tests Removed loops from PSUBUS tests - ensures folding is tested. Also renamed SSE2 tests SSSE3 to match cpu. This is a follow up commit agreed in http://reviews.llvm.org/D7094 llvm-svn: 226871
2015-01-23 06:19:58 +08:00
; AVX2: # BB#0:
; AVX2-NEXT: vmovd %esi, %xmm0
; AVX2-NEXT: vpbroadcastb %xmm0, %ymm0
; AVX2-NEXT: vmovdqu (%rdi), %ymm1
; AVX2-NEXT: vpsubusb %ymm0, %ymm1, %ymm0
; AVX2-NEXT: vmovdqu %ymm0, (%rdi)
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible. We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases if y is a constant. DAGCombiner canonicalizes those so we first have to undo the canonicalization for those cases. The pattern occurs in gzip when the loop vectorizer is enabled. Part of PR14613. llvm-svn: 170273
2012-12-16 00:47:44 +08:00
}