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/SystemZ/spill-01.ll

549 lines
16 KiB
LLVM
Raw Normal View History

[SystemZ] Automatically detect zEC12 and z196 hosts As on other hosts, the CPU identification instruction is priveleged, so we need to look through /proc/cpuinfo. I copied the PowerPC way of handling "generic". Several tests were implicitly assuming z10 and so failed on z196. llvm-svn: 193742
2013-10-31 20:14:17 +08:00
; Test spilling using MVC. The tests here assume z10 register pressure,
; without the high words being available.
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
;
[SystemZ] Automatically detect zEC12 and z196 hosts As on other hosts, the CPU identification instruction is priveleged, so we need to look through /proc/cpuinfo. I copied the PowerPC way of handling "generic". Several tests were implicitly assuming z10 and so failed on z196. llvm-svn: 193742
2013-10-31 20:14:17 +08:00
; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z10 | FileCheck %s
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
declare void @foo()
@g0 = global i32 0
@g1 = global i32 1
@g2 = global i32 2
@g3 = global i32 3
@g4 = global i32 4
@g5 = global i32 5
@g6 = global i32 6
@g7 = global i32 7
@g8 = global i32 8
@g9 = global i32 9
@h0 = global i64 0
@h1 = global i64 1
@h2 = global i64 2
@h3 = global i64 3
@h4 = global i64 4
@h5 = global i64 5
@h6 = global i64 6
@h7 = global i64 7
@h8 = global i64 8
@h9 = global i64 9
; This function shouldn't spill anything
define void @f1(i32 *%ptr0) {
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 14:24:09 +08:00
; CHECK-LABEL: f1:
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
; CHECK: stmg
; CHECK: aghi %r15, -160
; CHECK-NOT: %r15
; CHECK: brasl %r14, foo@PLT
; CHECK-NOT: %r15
; CHECK: lmg
; CHECK: br %r14
[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
%ptr1 = getelementptr i32, i32 *%ptr0, i32 2
%ptr2 = getelementptr i32, i32 *%ptr0, i32 4
%ptr3 = getelementptr i32, i32 *%ptr0, i32 6
%ptr4 = getelementptr i32, i32 *%ptr0, i32 8
%ptr5 = getelementptr i32, i32 *%ptr0, i32 10
%ptr6 = getelementptr i32, i32 *%ptr0, i32 12
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
[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
%val0 = load i32 , i32 *%ptr0
%val1 = load i32 , i32 *%ptr1
%val2 = load i32 , i32 *%ptr2
%val3 = load i32 , i32 *%ptr3
%val4 = load i32 , i32 *%ptr4
%val5 = load i32 , i32 *%ptr5
%val6 = load i32 , i32 *%ptr6
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
call void @foo()
store i32 %val0, i32 *%ptr0
store i32 %val1, i32 *%ptr1
store i32 %val2, i32 *%ptr2
store i32 %val3, i32 *%ptr3
store i32 %val4, i32 *%ptr4
store i32 %val5, i32 *%ptr5
store i32 %val6, i32 *%ptr6
ret void
}
; Test a case where at least one i32 load and at least one i32 store
; need spills.
define void @f2(i32 *%ptr0) {
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 14:24:09 +08:00
; CHECK-LABEL: f2:
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
; CHECK: mvc [[OFFSET1:16[04]]](4,%r15), [[OFFSET2:[0-9]+]]({{%r[0-9]+}})
; CHECK: brasl %r14, foo@PLT
; CHECK: mvc [[OFFSET2]](4,{{%r[0-9]+}}), [[OFFSET1]](%r15)
; CHECK: br %r14
[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
%ptr1 = getelementptr i32, i32 *%ptr0, i64 2
%ptr2 = getelementptr i32, i32 *%ptr0, i64 4
%ptr3 = getelementptr i32, i32 *%ptr0, i64 6
%ptr4 = getelementptr i32, i32 *%ptr0, i64 8
%ptr5 = getelementptr i32, i32 *%ptr0, i64 10
%ptr6 = getelementptr i32, i32 *%ptr0, i64 12
%ptr7 = getelementptr i32, i32 *%ptr0, i64 14
%ptr8 = getelementptr i32, i32 *%ptr0, i64 16
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
[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
%val0 = load i32 , i32 *%ptr0
%val1 = load i32 , i32 *%ptr1
%val2 = load i32 , i32 *%ptr2
%val3 = load i32 , i32 *%ptr3
%val4 = load i32 , i32 *%ptr4
%val5 = load i32 , i32 *%ptr5
%val6 = load i32 , i32 *%ptr6
%val7 = load i32 , i32 *%ptr7
%val8 = load i32 , i32 *%ptr8
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
call void @foo()
store i32 %val0, i32 *%ptr0
store i32 %val1, i32 *%ptr1
store i32 %val2, i32 *%ptr2
store i32 %val3, i32 *%ptr3
store i32 %val4, i32 *%ptr4
store i32 %val5, i32 *%ptr5
store i32 %val6, i32 *%ptr6
store i32 %val7, i32 *%ptr7
store i32 %val8, i32 *%ptr8
ret void
}
; Test a case where at least one i64 load and at least one i64 store
; need spills.
define void @f3(i64 *%ptr0) {
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 14:24:09 +08:00
; CHECK-LABEL: f3:
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
; CHECK: mvc 160(8,%r15), [[OFFSET:[0-9]+]]({{%r[0-9]+}})
; CHECK: brasl %r14, foo@PLT
; CHECK: mvc [[OFFSET]](8,{{%r[0-9]+}}), 160(%r15)
; CHECK: br %r14
[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
%ptr1 = getelementptr i64, i64 *%ptr0, i64 2
%ptr2 = getelementptr i64, i64 *%ptr0, i64 4
%ptr3 = getelementptr i64, i64 *%ptr0, i64 6
%ptr4 = getelementptr i64, i64 *%ptr0, i64 8
%ptr5 = getelementptr i64, i64 *%ptr0, i64 10
%ptr6 = getelementptr i64, i64 *%ptr0, i64 12
%ptr7 = getelementptr i64, i64 *%ptr0, i64 14
%ptr8 = getelementptr i64, i64 *%ptr0, i64 16
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
[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
%val0 = load i64 , i64 *%ptr0
%val1 = load i64 , i64 *%ptr1
%val2 = load i64 , i64 *%ptr2
%val3 = load i64 , i64 *%ptr3
%val4 = load i64 , i64 *%ptr4
%val5 = load i64 , i64 *%ptr5
%val6 = load i64 , i64 *%ptr6
%val7 = load i64 , i64 *%ptr7
%val8 = load i64 , i64 *%ptr8
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
call void @foo()
store i64 %val0, i64 *%ptr0
store i64 %val1, i64 *%ptr1
store i64 %val2, i64 *%ptr2
store i64 %val3, i64 *%ptr3
store i64 %val4, i64 *%ptr4
store i64 %val5, i64 *%ptr5
store i64 %val6, i64 *%ptr6
store i64 %val7, i64 *%ptr7
store i64 %val8, i64 *%ptr8
ret void
}
; Test a case where at least at least one f32 load and at least one f32 store
; need spills. The 8 call-saved FPRs could be used for 8 of the %vals
; (and are at the time of writing), but it would really be better to use
; MVC for all 10.
define void @f4(float *%ptr0) {
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 14:24:09 +08:00
; CHECK-LABEL: f4:
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
; CHECK: mvc [[OFFSET1:16[04]]](4,%r15), [[OFFSET2:[0-9]+]]({{%r[0-9]+}})
; CHECK: brasl %r14, foo@PLT
; CHECK: mvc [[OFFSET2]](4,{{%r[0-9]+}}), [[OFFSET1]](%r15)
; CHECK: br %r14
[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
%ptr1 = getelementptr float, float *%ptr0, i64 2
%ptr2 = getelementptr float, float *%ptr0, i64 4
%ptr3 = getelementptr float, float *%ptr0, i64 6
%ptr4 = getelementptr float, float *%ptr0, i64 8
%ptr5 = getelementptr float, float *%ptr0, i64 10
%ptr6 = getelementptr float, float *%ptr0, i64 12
%ptr7 = getelementptr float, float *%ptr0, i64 14
%ptr8 = getelementptr float, float *%ptr0, i64 16
%ptr9 = getelementptr float, float *%ptr0, i64 18
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
[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
%val0 = load float , float *%ptr0
%val1 = load float , float *%ptr1
%val2 = load float , float *%ptr2
%val3 = load float , float *%ptr3
%val4 = load float , float *%ptr4
%val5 = load float , float *%ptr5
%val6 = load float , float *%ptr6
%val7 = load float , float *%ptr7
%val8 = load float , float *%ptr8
%val9 = load float , float *%ptr9
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
call void @foo()
store float %val0, float *%ptr0
store float %val1, float *%ptr1
store float %val2, float *%ptr2
store float %val3, float *%ptr3
store float %val4, float *%ptr4
store float %val5, float *%ptr5
store float %val6, float *%ptr6
store float %val7, float *%ptr7
store float %val8, float *%ptr8
store float %val9, float *%ptr9
ret void
}
; Similarly for f64.
define void @f5(double *%ptr0) {
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 14:24:09 +08:00
; CHECK-LABEL: f5:
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
; CHECK: mvc 160(8,%r15), [[OFFSET:[0-9]+]]({{%r[0-9]+}})
; CHECK: brasl %r14, foo@PLT
; CHECK: mvc [[OFFSET]](8,{{%r[0-9]+}}), 160(%r15)
; CHECK: br %r14
[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
%ptr1 = getelementptr double, double *%ptr0, i64 2
%ptr2 = getelementptr double, double *%ptr0, i64 4
%ptr3 = getelementptr double, double *%ptr0, i64 6
%ptr4 = getelementptr double, double *%ptr0, i64 8
%ptr5 = getelementptr double, double *%ptr0, i64 10
%ptr6 = getelementptr double, double *%ptr0, i64 12
%ptr7 = getelementptr double, double *%ptr0, i64 14
%ptr8 = getelementptr double, double *%ptr0, i64 16
%ptr9 = getelementptr double, double *%ptr0, i64 18
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
[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
%val0 = load double , double *%ptr0
%val1 = load double , double *%ptr1
%val2 = load double , double *%ptr2
%val3 = load double , double *%ptr3
%val4 = load double , double *%ptr4
%val5 = load double , double *%ptr5
%val6 = load double , double *%ptr6
%val7 = load double , double *%ptr7
%val8 = load double , double *%ptr8
%val9 = load double , double *%ptr9
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
call void @foo()
store double %val0, double *%ptr0
store double %val1, double *%ptr1
store double %val2, double *%ptr2
store double %val3, double *%ptr3
store double %val4, double *%ptr4
store double %val5, double *%ptr5
store double %val6, double *%ptr6
store double %val7, double *%ptr7
store double %val8, double *%ptr8
store double %val9, double *%ptr9
ret void
}
; Repeat f2 with atomic accesses. We shouldn't use MVC here.
define void @f6(i32 *%ptr0) {
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 14:24:09 +08:00
; CHECK-LABEL: f6:
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
; CHECK-NOT: mvc
; CHECK: br %r14
[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
%ptr1 = getelementptr i32, i32 *%ptr0, i64 2
%ptr2 = getelementptr i32, i32 *%ptr0, i64 4
%ptr3 = getelementptr i32, i32 *%ptr0, i64 6
%ptr4 = getelementptr i32, i32 *%ptr0, i64 8
%ptr5 = getelementptr i32, i32 *%ptr0, i64 10
%ptr6 = getelementptr i32, i32 *%ptr0, i64 12
%ptr7 = getelementptr i32, i32 *%ptr0, i64 14
%ptr8 = getelementptr i32, i32 *%ptr0, i64 16
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
[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
%val0 = load atomic i32 , i32 *%ptr0 unordered, align 4
%val1 = load atomic i32 , i32 *%ptr1 unordered, align 4
%val2 = load atomic i32 , i32 *%ptr2 unordered, align 4
%val3 = load atomic i32 , i32 *%ptr3 unordered, align 4
%val4 = load atomic i32 , i32 *%ptr4 unordered, align 4
%val5 = load atomic i32 , i32 *%ptr5 unordered, align 4
%val6 = load atomic i32 , i32 *%ptr6 unordered, align 4
%val7 = load atomic i32 , i32 *%ptr7 unordered, align 4
%val8 = load atomic i32 , i32 *%ptr8 unordered, align 4
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
call void @foo()
store atomic i32 %val0, i32 *%ptr0 unordered, align 4
store atomic i32 %val1, i32 *%ptr1 unordered, align 4
store atomic i32 %val2, i32 *%ptr2 unordered, align 4
store atomic i32 %val3, i32 *%ptr3 unordered, align 4
store atomic i32 %val4, i32 *%ptr4 unordered, align 4
store atomic i32 %val5, i32 *%ptr5 unordered, align 4
store atomic i32 %val6, i32 *%ptr6 unordered, align 4
store atomic i32 %val7, i32 *%ptr7 unordered, align 4
store atomic i32 %val8, i32 *%ptr8 unordered, align 4
ret void
}
; ...likewise volatile accesses.
define void @f7(i32 *%ptr0) {
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 14:24:09 +08:00
; CHECK-LABEL: f7:
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
; CHECK-NOT: mvc
; CHECK: br %r14
[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
%ptr1 = getelementptr i32, i32 *%ptr0, i64 2
%ptr2 = getelementptr i32, i32 *%ptr0, i64 4
%ptr3 = getelementptr i32, i32 *%ptr0, i64 6
%ptr4 = getelementptr i32, i32 *%ptr0, i64 8
%ptr5 = getelementptr i32, i32 *%ptr0, i64 10
%ptr6 = getelementptr i32, i32 *%ptr0, i64 12
%ptr7 = getelementptr i32, i32 *%ptr0, i64 14
%ptr8 = getelementptr i32, i32 *%ptr0, i64 16
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
[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
%val0 = load volatile i32 , i32 *%ptr0
%val1 = load volatile i32 , i32 *%ptr1
%val2 = load volatile i32 , i32 *%ptr2
%val3 = load volatile i32 , i32 *%ptr3
%val4 = load volatile i32 , i32 *%ptr4
%val5 = load volatile i32 , i32 *%ptr5
%val6 = load volatile i32 , i32 *%ptr6
%val7 = load volatile i32 , i32 *%ptr7
%val8 = load volatile i32 , i32 *%ptr8
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
call void @foo()
store volatile i32 %val0, i32 *%ptr0
store volatile i32 %val1, i32 *%ptr1
store volatile i32 %val2, i32 *%ptr2
store volatile i32 %val3, i32 *%ptr3
store volatile i32 %val4, i32 *%ptr4
store volatile i32 %val5, i32 *%ptr5
store volatile i32 %val6, i32 *%ptr6
store volatile i32 %val7, i32 *%ptr7
store volatile i32 %val8, i32 *%ptr8
ret void
}
; Check that LRL and STRL are not converted.
define void @f8() {
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 14:24:09 +08:00
; CHECK-LABEL: f8:
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
; CHECK-NOT: mvc
; CHECK: br %r14
[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
%val0 = load i32 , i32 *@g0
%val1 = load i32 , i32 *@g1
%val2 = load i32 , i32 *@g2
%val3 = load i32 , i32 *@g3
%val4 = load i32 , i32 *@g4
%val5 = load i32 , i32 *@g5
%val6 = load i32 , i32 *@g6
%val7 = load i32 , i32 *@g7
%val8 = load i32 , i32 *@g8
%val9 = load i32 , i32 *@g9
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
call void @foo()
store i32 %val0, i32 *@g0
store i32 %val1, i32 *@g1
store i32 %val2, i32 *@g2
store i32 %val3, i32 *@g3
store i32 %val4, i32 *@g4
store i32 %val5, i32 *@g5
store i32 %val6, i32 *@g6
store i32 %val7, i32 *@g7
store i32 %val8, i32 *@g8
store i32 %val9, i32 *@g9
ret void
}
; Likewise LGRL and STGRL.
define void @f9() {
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 14:24:09 +08:00
; CHECK-LABEL: f9:
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
; CHECK-NOT: mvc
; CHECK: br %r14
[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
%val0 = load i64 , i64 *@h0
%val1 = load i64 , i64 *@h1
%val2 = load i64 , i64 *@h2
%val3 = load i64 , i64 *@h3
%val4 = load i64 , i64 *@h4
%val5 = load i64 , i64 *@h5
%val6 = load i64 , i64 *@h6
%val7 = load i64 , i64 *@h7
%val8 = load i64 , i64 *@h8
%val9 = load i64 , i64 *@h9
[SystemZ] Use MVC to spill loads and stores Try to use MVC when spilling the destination of a simple load or the source of a simple store. As explained in the comment, this doesn't yet handle the case where the load or store location is also a frame index, since that could lead to two simultaneous scavenger spills, something the backend can't handle yet. spill-02.py tests that this restriction kicks in, but unfortunately I've not yet found a case that would fail without it. The volatile trick I used for other scavenger tests doesn't work here because we can't use MVC for volatile accesses anyway. I'm planning on relaxing the restriction later, hopefully with a test that does trigger the problem... Tests @f8 and @f9 also showed that L(G)RL and ST(G)RL were wrongly classified as SimpleBDX{Load,Store}. It wouldn't be easy to test for that bug separately, which is why I didn't split out the fix as a separate patch. llvm-svn: 185434
2013-07-02 23:28:56 +08:00
call void @foo()
store i64 %val0, i64 *@h0
store i64 %val1, i64 *@h1
store i64 %val2, i64 *@h2
store i64 %val3, i64 *@h3
store i64 %val4, i64 *@h4
store i64 %val5, i64 *@h5
store i64 %val6, i64 *@h6
store i64 %val7, i64 *@h7
store i64 %val8, i64 *@h8
store i64 %val9, i64 *@h9
ret void
}
Fix double renaming bug in stack coloring pass The stack coloring pass renumbered frame indexes with a loop of the form: for each frame index FI for each instruction I that uses FI for each use of FI in I rename FI to FI' This caused problems if an instruction used two frame indexes F0 and F1 and if F0 was renamed to F1 and F1 to F2. The first time we visited the instruction we changed F0 to F1, then we changed both F1s to F2. In other words, the problem was that SSRefs recorded which instructions used an FI, but not which MachineOperands and MachineMemOperands within that instruction used it. This is easily fixed for MachineOperands by walking the instructions once and processing each operand in turn. There's already a loop to do that for dead store elimination, so it seemed more efficient to fuse the two at the block level. MachineMemOperands are more tricky because they can be shared between instructions. The patch handles them by making SSRefs an array of MachineMemOperands rather than an array of MachineInstrs. We might end up processing the same MachineMemOperand twice, but that's OK because we always know from the SSRefs index what the original frame index was. llvm-svn: 185703
2013-07-05 22:24:47 +08:00
; This showed a problem with the way stack coloring updated instructions.
; The copy from %val9 to %newval8 can be done using an MVC, which then
; has two frame index operands. Stack coloring chose a valid renumbering
; [FI0, FI1] -> [FI1, FI2], but applied it in the form FI0 -> FI1 -> FI2,
; so that both operands ended up being the same.
define void @f10() {
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 14:24:09 +08:00
; CHECK-LABEL: f10:
Fix double renaming bug in stack coloring pass The stack coloring pass renumbered frame indexes with a loop of the form: for each frame index FI for each instruction I that uses FI for each use of FI in I rename FI to FI' This caused problems if an instruction used two frame indexes F0 and F1 and if F0 was renamed to F1 and F1 to F2. The first time we visited the instruction we changed F0 to F1, then we changed both F1s to F2. In other words, the problem was that SSRefs recorded which instructions used an FI, but not which MachineOperands and MachineMemOperands within that instruction used it. This is easily fixed for MachineOperands by walking the instructions once and processing each operand in turn. There's already a loop to do that for dead store elimination, so it seemed more efficient to fuse the two at the block level. MachineMemOperands are more tricky because they can be shared between instructions. The patch handles them by making SSRefs an array of MachineMemOperands rather than an array of MachineInstrs. We might end up processing the same MachineMemOperand twice, but that's OK because we always know from the SSRefs index what the original frame index was. llvm-svn: 185703
2013-07-05 22:24:47 +08:00
; CHECK: lgrl [[REG:%r[0-9]+]], h9
; CHECK: stg [[REG]], [[VAL9:[0-9]+]](%r15)
; CHECK: brasl %r14, foo@PLT
; CHECK: brasl %r14, foo@PLT
; CHECK: mvc [[NEWVAL8:[0-9]+]](8,%r15), [[VAL9]](%r15)
; CHECK: brasl %r14, foo@PLT
; CHECK: lg [[REG:%r[0-9]+]], [[NEWVAL8]](%r15)
; CHECK: stgrl [[REG]], h8
; CHECK: br %r14
entry:
[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
%val8 = load volatile i64 , i64 *@h8
%val0 = load volatile i64 , i64 *@h0
%val1 = load volatile i64 , i64 *@h1
%val2 = load volatile i64 , i64 *@h2
%val3 = load volatile i64 , i64 *@h3
%val4 = load volatile i64 , i64 *@h4
%val5 = load volatile i64 , i64 *@h5
%val6 = load volatile i64 , i64 *@h6
%val7 = load volatile i64 , i64 *@h7
%val9 = load volatile i64 , i64 *@h9
Fix double renaming bug in stack coloring pass The stack coloring pass renumbered frame indexes with a loop of the form: for each frame index FI for each instruction I that uses FI for each use of FI in I rename FI to FI' This caused problems if an instruction used two frame indexes F0 and F1 and if F0 was renamed to F1 and F1 to F2. The first time we visited the instruction we changed F0 to F1, then we changed both F1s to F2. In other words, the problem was that SSRefs recorded which instructions used an FI, but not which MachineOperands and MachineMemOperands within that instruction used it. This is easily fixed for MachineOperands by walking the instructions once and processing each operand in turn. There's already a loop to do that for dead store elimination, so it seemed more efficient to fuse the two at the block level. MachineMemOperands are more tricky because they can be shared between instructions. The patch handles them by making SSRefs an array of MachineMemOperands rather than an array of MachineInstrs. We might end up processing the same MachineMemOperand twice, but that's OK because we always know from the SSRefs index what the original frame index was. llvm-svn: 185703
2013-07-05 22:24:47 +08:00
call void @foo()
store volatile i64 %val0, i64 *@h0
store volatile i64 %val1, i64 *@h1
store volatile i64 %val2, i64 *@h2
store volatile i64 %val3, i64 *@h3
store volatile i64 %val4, i64 *@h4
store volatile i64 %val5, i64 *@h5
store volatile i64 %val6, i64 *@h6
store volatile i64 %val7, i64 *@h7
[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
%check = load volatile i64 , i64 *@h0
Fix double renaming bug in stack coloring pass The stack coloring pass renumbered frame indexes with a loop of the form: for each frame index FI for each instruction I that uses FI for each use of FI in I rename FI to FI' This caused problems if an instruction used two frame indexes F0 and F1 and if F0 was renamed to F1 and F1 to F2. The first time we visited the instruction we changed F0 to F1, then we changed both F1s to F2. In other words, the problem was that SSRefs recorded which instructions used an FI, but not which MachineOperands and MachineMemOperands within that instruction used it. This is easily fixed for MachineOperands by walking the instructions once and processing each operand in turn. There's already a loop to do that for dead store elimination, so it seemed more efficient to fuse the two at the block level. MachineMemOperands are more tricky because they can be shared between instructions. The patch handles them by making SSRefs an array of MachineMemOperands rather than an array of MachineInstrs. We might end up processing the same MachineMemOperand twice, but that's OK because we always know from the SSRefs index what the original frame index was. llvm-svn: 185703
2013-07-05 22:24:47 +08:00
%cond = icmp eq i64 %check, 0
br i1 %cond, label %skip, label %fallthru
fallthru:
call void @foo()
store volatile i64 %val0, i64 *@h0
store volatile i64 %val1, i64 *@h1
store volatile i64 %val2, i64 *@h2
store volatile i64 %val3, i64 *@h3
store volatile i64 %val4, i64 *@h4
store volatile i64 %val5, i64 *@h5
store volatile i64 %val6, i64 *@h6
store volatile i64 %val7, i64 *@h7
store volatile i64 %val8, i64 *@h8
br label %skip
skip:
%newval8 = phi i64 [ %val8, %entry ], [ %val9, %fallthru ]
call void @foo()
store volatile i64 %val0, i64 *@h0
store volatile i64 %val1, i64 *@h1
store volatile i64 %val2, i64 *@h2
store volatile i64 %val3, i64 *@h3
store volatile i64 %val4, i64 *@h4
store volatile i64 %val5, i64 *@h5
store volatile i64 %val6, i64 *@h6
store volatile i64 %val7, i64 *@h7
store volatile i64 %newval8, i64 *@h8
store volatile i64 %val9, i64 *@h9
ret void
}
[SystemZ] Remove no-op MVCs The stack coloring pass has code to delete stores and loads that become trivially dead after coloring. Extend it to cope with single instructions that copy from one frame index to another. The testcase happens to show an example of this kicking in at the moment. It did occur in Real Code too though. llvm-svn: 185705
2013-07-05 22:38:48 +08:00
; This used to generate a no-op MVC. It is very sensitive to spill heuristics.
define void @f11() {
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 14:24:09 +08:00
; CHECK-LABEL: f11:
[SystemZ] Remove no-op MVCs The stack coloring pass has code to delete stores and loads that become trivially dead after coloring. Extend it to cope with single instructions that copy from one frame index to another. The testcase happens to show an example of this kicking in at the moment. It did occur in Real Code too though. llvm-svn: 185705
2013-07-05 22:38:48 +08:00
; CHECK-NOT: mvc [[OFFSET:[0-9]+]](8,%r15), [[OFFSET]](%r15)
; CHECK: br %r14
entry:
[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
%val0 = load volatile i64 , i64 *@h0
%val1 = load volatile i64 , i64 *@h1
%val2 = load volatile i64 , i64 *@h2
%val3 = load volatile i64 , i64 *@h3
%val4 = load volatile i64 , i64 *@h4
%val5 = load volatile i64 , i64 *@h5
%val6 = load volatile i64 , i64 *@h6
%val7 = load volatile i64 , i64 *@h7
%altval0 = load volatile i64 , i64 *@h0
%altval1 = load volatile i64 , i64 *@h1
[SystemZ] Remove no-op MVCs The stack coloring pass has code to delete stores and loads that become trivially dead after coloring. Extend it to cope with single instructions that copy from one frame index to another. The testcase happens to show an example of this kicking in at the moment. It did occur in Real Code too though. llvm-svn: 185705
2013-07-05 22:38:48 +08:00
call void @foo()
store volatile i64 %val0, i64 *@h0
store volatile i64 %val1, i64 *@h1
store volatile i64 %val2, i64 *@h2
store volatile i64 %val3, i64 *@h3
store volatile i64 %val4, i64 *@h4
store volatile i64 %val5, i64 *@h5
store volatile i64 %val6, i64 *@h6
store volatile i64 %val7, i64 *@h7
[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
%check = load volatile i64 , i64 *@h0
[SystemZ] Remove no-op MVCs The stack coloring pass has code to delete stores and loads that become trivially dead after coloring. Extend it to cope with single instructions that copy from one frame index to another. The testcase happens to show an example of this kicking in at the moment. It did occur in Real Code too though. llvm-svn: 185705
2013-07-05 22:38:48 +08:00
%cond = icmp eq i64 %check, 0
br i1 %cond, label %a1, label %b1
a1:
call void @foo()
br label %join1
b1:
call void @foo()
br label %join1
join1:
%newval0 = phi i64 [ %val0, %a1 ], [ %altval0, %b1 ]
call void @foo()
store volatile i64 %val1, i64 *@h1
store volatile i64 %val2, i64 *@h2
store volatile i64 %val3, i64 *@h3
store volatile i64 %val4, i64 *@h4
store volatile i64 %val5, i64 *@h5
store volatile i64 %val6, i64 *@h6
store volatile i64 %val7, i64 *@h7
br i1 %cond, label %a2, label %b2
a2:
call void @foo()
br label %join2
b2:
call void @foo()
br label %join2
join2:
%newval1 = phi i64 [ %val1, %a2 ], [ %altval1, %b2 ]
call void @foo()
store volatile i64 %val2, i64 *@h2
store volatile i64 %val3, i64 *@h3
store volatile i64 %val4, i64 *@h4
store volatile i64 %val5, i64 *@h5
store volatile i64 %val6, i64 *@h6
store volatile i64 %val7, i64 *@h7
call void @foo()
store volatile i64 %newval0, i64 *@h0
store volatile i64 %newval1, i64 *@h1
store volatile i64 %val2, i64 *@h2
store volatile i64 %val3, i64 *@h3
store volatile i64 %val4, i64 *@h4
store volatile i64 %val5, i64 *@h5
store volatile i64 %val6, i64 *@h6
store volatile i64 %val7, i64 *@h7
ret void
}