maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/PowerPC/2008-07-15-Bswap.ll

382 lines
23 KiB
LLVM
Raw Normal View History

Adding -verify-machineinstrs option to PowerPC tests Currently we have a number of tests that fail with -verify-machineinstrs. To detect this cases earlier we add the option to the testcases with the exception of tests that will currently fail with this option. PR 27456 keeps track of this failures. No code review, as discussed with Hal Finkel. llvm-svn: 277624
2016-08-04 02:17:35 +08:00
; RUN: llc -verify-machineinstrs < %s
LegalizeTypes support for promotion of bswap. In LegalizeDAG the value is zero-extended to the new type before byte swapping. It doesn't matter how the extension is done since the new bits are shifted off anyway after the swap, so extend by any old rubbish bits. This results in the final assembler for the testcase being one line shorter. llvm-svn: 53604
2008-07-15 18:18:22 +08:00
target datalayout = "E-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f128:64:128"
target triple = "powerpc-apple-darwin9"
%struct.BiPartSrcDescriptor = type <{ %"struct.BiPartSrcDescriptor::$_105" }>
%"struct.BiPartSrcDescriptor::$_105" = type { %struct.BiPartSrcDescriptor_NO_VECTOR_ALIGNMENT_size_is_16 }
%struct.BiPartSrcDescriptor_NO_VECTOR_ALIGNMENT_size_is_16 = type { [2 x %struct.MotionVectors], [2 x i8], %struct.Map4x4ToPartIdx, [2 x i8], i8, i8 }
%struct.Condv = type opaque
%struct.DHBFLayerId = type { i8 }
%struct.DecodeComplexityInfo = type { i32, i32, i32, i32, %"struct.DecodeComplexityInfo::IntraStats", %"struct.DecodeComplexityInfo::InterStats" }
%"struct.DecodeComplexityInfo::InterStats" = type { i32, i32, i32, i32, [5 x i32], [3 x i32], [4 x [4 x i32]], [4 x i32], i32, %struct.MotionVectors, %struct.MotionVectors }
%"struct.DecodeComplexityInfo::IntraStats" = type { i32, i32, i32, [5 x i32], [3 x i32], [4 x i32], [3 x i32] }
%struct.DecodeComplexityOptions = type { i8, i8, i32, double, i8, float, i8, float, i8, i8, i8, i8, i8 }
%struct.DescriptorAllocator = type { %struct.Mutex*, %struct.Mutex*, i8**, i32, i32, i8**, i32, i32, i8**, i32, i32 }
%struct.DetailsFromSliceType = type <{ i8 }>
%struct.FlatnessAnalysis = type { i16, i16, i32, i32*, i8*, [512 x i32], [256 x i32] }
%struct.Frame = type <{ i8, i8, i8, i8, i8, [3 x i8], i32, i32, %struct.Mutex*, %struct.Condv*, [8 x i8], %struct.FramePixels, %struct.FrameMotionVectorCache, %struct.FrameIndex, i32, i8*, i8*, i8*, i8*, i16*, %struct.FlatnessAnalysis, %struct.NoiseAnalysis, %struct.VisualActivity, %struct.FrameMotionInfo, %struct.FrameMotionAnalysis, %struct.FrameDataRateParameters, %struct.FrameEncoderTags, %struct.DecodeComplexityInfo, %struct.DecodeComplexityOptions, %struct.MotionInfoFor16x16_FasterSP*, [1 x i32] }>
%struct.FrameDataRateParameters = type { i32, float, i8, i8 }
%struct.FrameEncoderTags = type { i8, i8, i32, i8, i8, float }
%struct.FrameIndex = type { i32, i32, i32, i32, i32, i32, i32, i32, i32, i8, i8, i8, i32, i32, %struct.Frame*, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, %struct.DHBFLayerId }
%struct.FrameMotionAnalysis = type { i32, i32, i32, %struct.MoEstMotion16x16*, %struct.MbAnalysis*, i32, i32, i16, i16, i32, i32, i32, i32, i8, i8 }
%struct.FrameMotionInfo = type { i32, i32, %struct.MoEstMbMotionInfo*, i32, i32, i32, i32, i32 }
%struct.FrameMotionVectorCache = type <{ %struct.ThreadAllocator**, i32, i32, i32, %struct.BiPartSrcDescriptor, %struct.BiPartSrcDescriptor, %struct.BiPartSrcDescriptor, [3 x %struct.BiPartSrcDescriptor*], %struct.BiPartSrcDescriptor** }>
%struct.FramePixels = type <{ i8, i8, i8, i8, i8, i8, i8, i8, i8*, i8*, i32, [4 x i8*], [4 x i8*], [2 x [4 x i32]], [2 x [4 x i32]], %struct.PixelData, %struct.InterpolationCache*, %struct.InterpolationCache*, %struct.InterpolationCache*, [16 x i16], [16 x i16], [12 x i8], %"struct.PortableSInt32Array<4>", %"struct.PortableSInt32Array<8>", %struct.ICOffsetArraysY, %struct.UVSrcOffsetEtcX_Struct*, i32*, i32*, [3 x i32] }>
%struct.ICOffsetArraysY = type { [21 x i32], [21 x i32], [4 x [21 x i32]] }
%struct.InterpolationCache = type opaque
%struct.LoopFilterInfo = type { %struct.BiPartSrcDescriptor**, i32, i32, i32, i32, i32*, i32, %"struct.LoopFilterInfo::SliceInfoStruct"*, i32, %struct.Mutex*, i16*, %struct.FramePixels*, i8*, i8*, i8*, i8*, i8*, %struct.PerMacroblockBoundaryStrengths*, %struct.Mutex*, i8*, i8*, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i8, i8, i8, i8*, i8*, i8, void (i8*, i8*, i32, i32, i32, i32, i32, i8*, i32)*, void (i8*, i8*, i32, i32, i32, i32, i32, i8*, i32, i8*)*, i32 }
%"struct.LoopFilterInfo::SliceInfoStruct" = type { %"struct.LoopFilterInfo::SliceInfoStruct::LFDisableStats", i8, i8, i8, i8, [17 x %struct.Frame*], [17 x %struct.Frame*] }
%"struct.LoopFilterInfo::SliceInfoStruct::LFDisableStats" = type { i32, i32 }
%struct.LoopFilterParam = type { i32, %struct.LoopFilterInfo*, %struct.FramePixels*, %struct.FrameMotionVectorCache* }
%struct.Map4x4ToPartIdx = type { i16 }
%struct.MbAnalysis = type { i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, %struct.RdCost, %struct.RdCost, i32 }
%struct.MoEstMbMotionInfo = type { i32, i32, i32, i32, [16 x %struct.MoEstPartMotionInfo] }
%struct.MoEstMotion16x16 = type { [2 x i8], [2 x %struct.MotionVectors], i8, [3 x %struct.MoEstPredCost] }
%struct.MoEstPartMotionInfo = type { i32, %struct.PartGeom, i32, i32, [2 x %struct.MotionVectors], [2 x i8], i16 }
%struct.MoEstPredCost = type { i32, i16, i16 }
%struct.MotionInfoFor16x16_FasterSP = type { [2 x %struct.MotionVectors], [2 x i8], i8, [2 x i32], i32, i32 }
%struct.MotionVectors = type { %"struct.MotionVectors::$_103" }
%"struct.MotionVectors::$_103" = type { i32 }
%struct.Mutex = type opaque
%struct.NoiseAnalysis = type { i16, i16, i32, i8*, i8*, i8*, [512 x i32] }
%struct.PartGeom = type { %struct.Map4x4ToPartIdx }
%struct.PerMacroblockBoundaryStrengths = type { [16 x i8], [16 x i8], [4 x i8], [4 x i8], [2 x i32] }
%struct.PixelData = type { i8*, i8*, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i8, i8 }
%"struct.PortableSInt32Array<4>" = type { [4 x i32] }
%"struct.PortableSInt32Array<8>" = type { [8 x i32] }
%struct.RdCost = type { i32, i32, i32, double }
%struct.ThreadAllocator = type { %struct.DescriptorAllocator*, %struct.BiPartSrcDescriptor*, [256 x %struct.BiPartSrcDescriptor*], i32, i32, i32 }
%struct.ThreadedBatch = type opaque
%struct.UVSrcOffsetEtcX_Struct = type <{ i16 }>
%struct.VisualActivity = type { i16, i16, i32, i32, i32*, i32*, i32, i32, i32*, i32, i32, i32, i32, i32, i8*, i32, [2 x i32], i32, i32, i32, i16*, i16, i16, i16, i16, float, i8*, i32*, i32, i32, i8 }
@_ZL33table_8_14_indexA_to_alpha_scalar = external constant [64 x i8] ; <[64 x i8]*> [#uses=0]
@_ZL32table_8_14_indexB_to_beta_scalar = external constant [64 x i8] ; <[64 x i8]*> [#uses=0]
@_ZL34table_8_15_indexA_bS_to_tc0_scalar = external constant [64 x [4 x i8]] ; <[64 x [4 x i8]]*> [#uses=0]
@gkDummy = external global i32 ; <i32*> [#uses=0]
@gkDetailsFromSliceTypeArray = external constant [10 x %struct.DetailsFromSliceType] ; <[10 x %struct.DetailsFromSliceType]*> [#uses=0]
declare i32 @_Z20LoopFilter_ConstructP14LoopFilterInfojj(%struct.LoopFilterInfo*, i32, i32)
declare i32 @_Z25LF_Threading2_assert_doneP14LoopFilterInfo(%struct.LoopFilterInfo*) nounwind
declare i32 @_Z54S_CalcIfLargeMVDeltaForBMbBothPredictionsFromSameFramePK19BiPartSrcDescriptorS1_ijj(%struct.BiPartSrcDescriptor*, %struct.BiPartSrcDescriptor*, i32, i32, i32) nounwind
declare void @_Z30LoopFilter_Internal_FilterLumaPhiiiiii(i8*, i32, i32, i32, i32, i32, i32) nounwind
declare void @_Z33LoopFilter_Internal_FilterChromaVPhiiiiiiiiii(i8*, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) nounwind
declare void @_Z33LoopFilter_Internal_FilterChromaHPhiiiiii(i8*, i32, i32, i32, i32, i32, i32) nounwind
declare void @_Z42LoopFilter_Internal_filter_macroblock_lumaPK14LoopFilterInfoPhS2_iiiPK30PerMacroblockBoundaryStrengthsjj(%struct.LoopFilterInfo*, i8*, i8*, i32, i32, i32, %struct.PerMacroblockBoundaryStrengths*, i32, i32) nounwind
declare i32 @_Z40LoopFilter_Internal_FilterLumaPlaneMBAFFPK14LoopFilterInfojjj(%struct.LoopFilterInfo*, i32, i32, i32) nounwind
declare void @_Z18LoopFilter_DestroyP14LoopFilterInfo(%struct.LoopFilterInfo*)
declare void @MutexDispose(%struct.Mutex*)
declare void @_ZdaPv(i8*) nounwind
declare void @jvtDisposePTRVectorAligned(i8*)
declare void @jvtDisposePTR(i8*)
declare void @jvtDisposePTRMemAligned(i8*)
declare void @_Z31LoopFilter_Internal_ResetTablesP14LoopFilterInfo(%struct.LoopFilterInfo*) nounwind
define i32 @_Z60LoopFilter_Internal_CalculateBoundaryStrengths_MbaffFramePicPK14LoopFilterInfoP22FrameMotionVectorCachejj(%struct.LoopFilterInfo* %lfiPtr, %struct.FrameMotionVectorCache* %frameMotionVectorCachePtr, i32 %mbY_min, i32 %mbY_maxPlus1) nounwind {
entry:
icmp ult i32 %mbY_min, %mbY_maxPlus1 ; <i1>:0 [#uses=1]
br i1 %0, label %bb16, label %bb642
bb16: ; preds = %entry
bitcast %struct.PerMacroblockBoundaryStrengths* null to i32* ; <i32*>:1 [#uses=3]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr i32, i32* %1, i32 1 ; <i32*>:2 [#uses=0]
getelementptr i32, i32* %1, i32 2 ; <i32*>:3 [#uses=0]
getelementptr i32, i32* %1, i32 3 ; <i32*>:4 [#uses=0]
LegalizeTypes support for promotion of bswap. In LegalizeDAG the value is zero-extended to the new type before byte swapping. It doesn't matter how the extension is done since the new bits are shifted off anyway after the swap, so extend by any old rubbish bits. This results in the final assembler for the testcase being one line shorter. llvm-svn: 53604
2008-07-15 18:18:22 +08:00
bitcast [16 x i8]* null to i32* ; <i32*>:5 [#uses=3]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr i32, i32* %5, i32 1 ; <i32*>:6 [#uses=0]
getelementptr i32, i32* %5, i32 2 ; <i32*>:7 [#uses=0]
getelementptr i32, i32* %5, i32 3 ; <i32*>:8 [#uses=0]
LegalizeTypes support for promotion of bswap. In LegalizeDAG the value is zero-extended to the new type before byte swapping. It doesn't matter how the extension is done since the new bits are shifted off anyway after the swap, so extend by any old rubbish bits. This results in the final assembler for the testcase being one line shorter. llvm-svn: 53604
2008-07-15 18:18:22 +08:00
icmp eq i32 0, 0 ; <i1>:9 [#uses=0]
lshr i32 0, 30 ; <i32>:10 [#uses=0]
and i32 0, 268435455 ; <i32>:11 [#uses=0]
lshr i32 0, 28 ; <i32>:12 [#uses=1]
and i32 %12, 3 ; <i32>:13 [#uses=0]
and i32 0, 1 ; <i32>:14 [#uses=1]
icmp eq i32 %14, 0 ; <i1>:15 [#uses=0]
zext i8 0 to i32 ; <i32>:16 [#uses=1]
%.not656 = icmp ne i32 0, 0 ; <i1> [#uses=1]
icmp eq i8 0, 0 ; <i1>:17 [#uses=0]
trunc i32 0 to i8 ; <i8>:18 [#uses=2]
add i32 0, 1 ; <i32>:19 [#uses=1]
%.not658 = icmp ne i32 0, 0 ; <i1> [#uses=1]
and i32 0, 268369920 ; <i32>:20 [#uses=1]
icmp eq i32 %20, 268369920 ; <i1>:21 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr %struct.PerMacroblockBoundaryStrengths, %struct.PerMacroblockBoundaryStrengths* null, i32 0, i32 2 ; <[4 x i8]*>:22 [#uses=1]
getelementptr %struct.PerMacroblockBoundaryStrengths, %struct.PerMacroblockBoundaryStrengths* null, i32 0, i32 2, i32 0 ; <i8*>:23 [#uses=0]
LegalizeTypes support for promotion of bswap. In LegalizeDAG the value is zero-extended to the new type before byte swapping. It doesn't matter how the extension is done since the new bits are shifted off anyway after the swap, so extend by any old rubbish bits. This results in the final assembler for the testcase being one line shorter. llvm-svn: 53604
2008-07-15 18:18:22 +08:00
and i32 0, -2 ; <i32>:24 [#uses=1]
add i32 %24, -1 ; <i32>:25 [#uses=0]
bitcast [4 x i8]* %22 to i32* ; <i32*>:26 [#uses=3]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr i32, i32* %26, i32 1 ; <i32*>:27 [#uses=0]
getelementptr i32, i32* %26, i32 2 ; <i32*>:28 [#uses=0]
getelementptr i32, i32* %26, i32 3 ; <i32*>:29 [#uses=0]
LegalizeTypes support for promotion of bswap. In LegalizeDAG the value is zero-extended to the new type before byte swapping. It doesn't matter how the extension is done since the new bits are shifted off anyway after the swap, so extend by any old rubbish bits. This results in the final assembler for the testcase being one line shorter. llvm-svn: 53604
2008-07-15 18:18:22 +08:00
br label %bb144
bb144: ; preds = %bb395, %bb16
%idxEachField11.0773 = phi i32 [ 0, %bb16 ], [ %162, %bb395 ] ; <i32> [#uses=3]
%mbYLeft.2776 = phi i32 [ 0, %bb16 ], [ %mbYLeft.2776, %bb395 ] ; <i32> [#uses=3]
%mbXYLeft.2775 = phi i32 [ 0, %bb16 ], [ %mbXYLeft.2775, %bb395 ] ; <i32> [#uses=1]
%mixedModeLeftEdgeOfMbFlag.2774 = phi i32 [ 0, %bb16 ], [ 0, %bb395 ] ; <i32> [#uses=0]
%mbIndexLeft.2772 = phi i32 [ 0, %bb16 ], [ %mbIndexLeft.2772, %bb395 ] ; <i32> [#uses=2]
%boundaryStrengthsV.1771 = phi i8* [ null, %bb16 ], [ %158, %bb395 ] ; <i8*> [#uses=2]
%numEdgesToTest.1770 = phi i32 [ 4, %bb16 ], [ %numEdgesToTest.2, %bb395 ] ; <i32> [#uses=1]
icmp eq i32 %idxEachField11.0773, 0 ; <i1>:30 [#uses=0]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr %struct.BiPartSrcDescriptor*, %struct.BiPartSrcDescriptor** null, i32 %mbIndexLeft.2772 ; <%struct.BiPartSrcDescriptor**>:31 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load %struct.BiPartSrcDescriptor*, %struct.BiPartSrcDescriptor** %31, align 4 ; <%struct.BiPartSrcDescriptor*>:32 [#uses=0]
LegalizeTypes support for promotion of bswap. In LegalizeDAG the value is zero-extended to the new type before byte swapping. It doesn't matter how the extension is done since the new bits are shifted off anyway after the swap, so extend by any old rubbish bits. This results in the final assembler for the testcase being one line shorter. llvm-svn: 53604
2008-07-15 18:18:22 +08:00
%fMacroblockHasNonZeroBS.4 = select i1 %21, i32 1, i32 0 ; <i32> [#uses=1]
%numEdgesToTest.2 = select i1 %21, i32 1, i32 %numEdgesToTest.1770 ; <i32> [#uses=2]
store i8 32, i8* %boundaryStrengthsV.1771, align 1
br label %labelContinueEdgesLoopV
bb200: ; preds = %labelContinueEdgesLoopV
lshr i32 %159, 28 ; <i32>:33 [#uses=2]
and i32 %160, %16 ; <i32>:34 [#uses=1]
icmp eq i32 %34, 0 ; <i1>:35 [#uses=0]
icmp eq i32 %160, 0 ; <i1>:36 [#uses=3]
zext i1 %36 to i32 ; <i32>:37 [#uses=1]
or i32 %37, -1 ; <i32>:38 [#uses=1]
or i32 %38, %33 ; <i32>:39 [#uses=1]
icmp eq i32 %39, 0 ; <i1>:40 [#uses=1]
br i1 %40, label %bb205, label %bb206
bb205: ; preds = %bb200
store i8 32, i8* %158, align 1
br label %labelContinueEdgesLoopV
bb206: ; preds = %bb200
icmp eq i32 %33, 15 ; <i1>:41 [#uses=1]
br i1 %41, label %labelContinueEdgesLoopV, label %bb210.preheader
bb210.preheader: ; preds = %bb206
add i32 %160, 0 ; <i32>:42 [#uses=2]
%bothcond657 = and i1 %36, %.not656 ; <i1> [#uses=0]
shl i32 %idxEachField11.0773, 1 ; <i32>:43 [#uses=1]
add i32 %43, 0 ; <i32>:44 [#uses=0]
shl i32 %mbYLeft.2776, 2 ; <i32>:45 [#uses=0]
add i32 %42, -1 ; <i32>:46 [#uses=1]
icmp eq i32 0, 0 ; <i1>:47 [#uses=1]
%brmerge689.not = and i1 %47, false ; <i1> [#uses=0]
%bothcond659 = and i1 %36, %.not658 ; <i1> [#uses=0]
shl i32 %mbYLeft.2776, 1 ; <i32>:48 [#uses=1]
or i32 %48, 0 ; <i32>:49 [#uses=1]
shl i32 %49, 1 ; <i32>:50 [#uses=0]
add i32 0, 0 ; <i32>:51 [#uses=2]
mul i32 %51, 0 ; <i32>:52 [#uses=1]
add i32 %52, %42 ; <i32>:53 [#uses=1]
mul i32 %51, 0 ; <i32>:54 [#uses=1]
add i32 %46, %54 ; <i32>:55 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr %struct.BiPartSrcDescriptor*, %struct.BiPartSrcDescriptor** null, i32 %53 ; <%struct.BiPartSrcDescriptor**>:56 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load %struct.BiPartSrcDescriptor*, %struct.BiPartSrcDescriptor** %56, align 4 ; <%struct.BiPartSrcDescriptor*>:57 [#uses=7]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr %struct.BiPartSrcDescriptor*, %struct.BiPartSrcDescriptor** null, i32 %55 ; <%struct.BiPartSrcDescriptor**>:58 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load %struct.BiPartSrcDescriptor*, %struct.BiPartSrcDescriptor** %58, align 4 ; <%struct.BiPartSrcDescriptor*>:59 [#uses=5]
LegalizeTypes support for promotion of bswap. In LegalizeDAG the value is zero-extended to the new type before byte swapping. It doesn't matter how the extension is done since the new bits are shifted off anyway after the swap, so extend by any old rubbish bits. This results in the final assembler for the testcase being one line shorter. llvm-svn: 53604
2008-07-15 18:18:22 +08:00
icmp slt i32 %159, 0 ; <i1>:60 [#uses=0]
icmp eq %struct.BiPartSrcDescriptor* %57, %59 ; <i1>:61 [#uses=0]
bitcast %struct.BiPartSrcDescriptor* %57 to i16* ; <i16*>:62 [#uses=5]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load i16, i16* %62, align 2 ; <i16>:63 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr i16, i16* %62, i32 1 ; <i16*>:64 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load i16, i16* %64, align 2 ; <i16>:65 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr i16, i16* %62, i32 2 ; <i16*>:66 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load i16, i16* %66, align 2 ; <i16>:67 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr i16, i16* %62, i32 3 ; <i16*>:68 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load i16, i16* %68, align 2 ; <i16>:69 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr i16, i16* %62, i32 6 ; <i16*>:70 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load i16, i16* %70, align 2 ; <i16>:71 [#uses=2]
LegalizeTypes support for promotion of bswap. In LegalizeDAG the value is zero-extended to the new type before byte swapping. It doesn't matter how the extension is done since the new bits are shifted off anyway after the swap, so extend by any old rubbish bits. This results in the final assembler for the testcase being one line shorter. llvm-svn: 53604
2008-07-15 18:18:22 +08:00
bitcast %struct.BiPartSrcDescriptor* %59 to i16* ; <i16*>:72 [#uses=5]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load i16, i16* %72, align 2 ; <i16>:73 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr i16, i16* %72, i32 1 ; <i16*>:74 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load i16, i16* %74, align 2 ; <i16>:75 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr i16, i16* %72, i32 2 ; <i16*>:76 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load i16, i16* %76, align 2 ; <i16>:77 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr i16, i16* %72, i32 3 ; <i16*>:78 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load i16, i16* %78, align 2 ; <i16>:79 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr i16, i16* %72, i32 6 ; <i16*>:80 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load i16, i16* %80, align 2 ; <i16>:81 [#uses=2]
LegalizeTypes support for promotion of bswap. In LegalizeDAG the value is zero-extended to the new type before byte swapping. It doesn't matter how the extension is done since the new bits are shifted off anyway after the swap, so extend by any old rubbish bits. This results in the final assembler for the testcase being one line shorter. llvm-svn: 53604
2008-07-15 18:18:22 +08:00
sub i16 %63, %73 ; <i16>:82 [#uses=3]
sub i16 %65, %75 ; <i16>:83 [#uses=3]
sub i16 %67, %77 ; <i16>:84 [#uses=3]
sub i16 %69, %79 ; <i16>:85 [#uses=3]
sub i16 %71, %81 ; <i16>:86 [#uses=3]
sub i16 0, %82 ; <i16>:87 [#uses=1]
icmp slt i16 %82, 0 ; <i1>:88 [#uses=1]
%. = select i1 %88, i16 %87, i16 %82 ; <i16> [#uses=1]
sub i16 0, %83 ; <i16>:89 [#uses=1]
icmp slt i16 %83, 0 ; <i1>:90 [#uses=1]
%.660 = select i1 %90, i16 %89, i16 %83 ; <i16> [#uses=1]
sub i16 0, %84 ; <i16>:91 [#uses=1]
icmp slt i16 %84, 0 ; <i1>:92 [#uses=1]
%.661 = select i1 %92, i16 %91, i16 %84 ; <i16> [#uses=1]
sub i16 0, %85 ; <i16>:93 [#uses=1]
icmp slt i16 %85, 0 ; <i1>:94 [#uses=1]
%.662 = select i1 %94, i16 %93, i16 %85 ; <i16> [#uses=1]
sub i16 0, %86 ; <i16>:95 [#uses=1]
icmp slt i16 %86, 0 ; <i1>:96 [#uses=1]
%.663 = select i1 %96, i16 %95, i16 %86 ; <i16> [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr %struct.BiPartSrcDescriptor, %struct.BiPartSrcDescriptor* %57, i32 0, i32 0, i32 0, i32 1, i32 0 ; <i8*>:97 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load i8, i8* %97, align 1 ; <i8>:98 [#uses=1]
LegalizeTypes support for promotion of bswap. In LegalizeDAG the value is zero-extended to the new type before byte swapping. It doesn't matter how the extension is done since the new bits are shifted off anyway after the swap, so extend by any old rubbish bits. This results in the final assembler for the testcase being one line shorter. llvm-svn: 53604
2008-07-15 18:18:22 +08:00
zext i8 %98 to i32 ; <i32>:99 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr %struct.BiPartSrcDescriptor, %struct.BiPartSrcDescriptor* %57, i32 0, i32 0, i32 0, i32 1, i32 1 ; <i8*>:100 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load i8, i8* %100, align 1 ; <i8>:101 [#uses=1]
LegalizeTypes support for promotion of bswap. In LegalizeDAG the value is zero-extended to the new type before byte swapping. It doesn't matter how the extension is done since the new bits are shifted off anyway after the swap, so extend by any old rubbish bits. This results in the final assembler for the testcase being one line shorter. llvm-svn: 53604
2008-07-15 18:18:22 +08:00
zext i8 %101 to i32 ; <i32>:102 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr %struct.BiPartSrcDescriptor, %struct.BiPartSrcDescriptor* %57, i32 0, i32 0, i32 0, i32 3, i32 0 ; <i8*>:103 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load i8, i8* %103, align 1 ; <i8>:104 [#uses=2]
LegalizeTypes support for promotion of bswap. In LegalizeDAG the value is zero-extended to the new type before byte swapping. It doesn't matter how the extension is done since the new bits are shifted off anyway after the swap, so extend by any old rubbish bits. This results in the final assembler for the testcase being one line shorter. llvm-svn: 53604
2008-07-15 18:18:22 +08:00
zext i8 %104 to i32 ; <i32>:105 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr %struct.BiPartSrcDescriptor, %struct.BiPartSrcDescriptor* %59, i32 0, i32 0, i32 0, i32 3, i32 0 ; <i8*>:106 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load i8, i8* %106, align 1 ; <i8>:107 [#uses=2]
LegalizeTypes support for promotion of bswap. In LegalizeDAG the value is zero-extended to the new type before byte swapping. It doesn't matter how the extension is done since the new bits are shifted off anyway after the swap, so extend by any old rubbish bits. This results in the final assembler for the testcase being one line shorter. llvm-svn: 53604
2008-07-15 18:18:22 +08:00
zext i8 %107 to i32 ; <i32>:108 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr %struct.BiPartSrcDescriptor, %struct.BiPartSrcDescriptor* %57, i32 0, i32 0, i32 0, i32 3, i32 1 ; <i8*>:109 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load i8, i8* %109, align 1 ; <i8>:110 [#uses=1]
LegalizeTypes support for promotion of bswap. In LegalizeDAG the value is zero-extended to the new type before byte swapping. It doesn't matter how the extension is done since the new bits are shifted off anyway after the swap, so extend by any old rubbish bits. This results in the final assembler for the testcase being one line shorter. llvm-svn: 53604
2008-07-15 18:18:22 +08:00
zext i8 %110 to i32 ; <i32>:111 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr %struct.BiPartSrcDescriptor, %struct.BiPartSrcDescriptor* %59, i32 0, i32 0, i32 0, i32 3, i32 1 ; <i8*>:112 [#uses=1]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load i8, i8* %112, align 1 ; <i8>:113 [#uses=1]
LegalizeTypes support for promotion of bswap. In LegalizeDAG the value is zero-extended to the new type before byte swapping. It doesn't matter how the extension is done since the new bits are shifted off anyway after the swap, so extend by any old rubbish bits. This results in the final assembler for the testcase being one line shorter. llvm-svn: 53604
2008-07-15 18:18:22 +08:00
zext i8 %113 to i32 ; <i32>:114 [#uses=1]
lshr i32 %99, 4 ; <i32>:115 [#uses=1]
and i32 %115, 2 ; <i32>:116 [#uses=1]
lshr i32 %102, 5 ; <i32>:117 [#uses=1]
or i32 %116, %117 ; <i32>:118 [#uses=3]
icmp eq i32 %118, 0 ; <i1>:119 [#uses=0]
icmp eq i32 %118, 1 ; <i1>:120 [#uses=1]
br i1 %120, label %bb297, label %bb298
bb297: ; preds = %bb210.preheader
br label %bb298
bb298: ; preds = %bb297, %bb210.preheader
%vu8Mask_0.1 = phi i8 [ -1, %bb297 ], [ 0, %bb210.preheader ] ; <i8> [#uses=1]
%vu8Mask_1.1 = phi i8 [ -1, %bb297 ], [ 0, %bb210.preheader ] ; <i8> [#uses=1]
%vu8Mask_2.1 = phi i8 [ -1, %bb297 ], [ 0, %bb210.preheader ] ; <i8> [#uses=0]
%vu8Mask_3.1 = phi i8 [ -1, %bb297 ], [ 0, %bb210.preheader ] ; <i8> [#uses=1]
%vu8Mask_4.1 = phi i8 [ 0, %bb297 ], [ 0, %bb210.preheader ] ; <i8> [#uses=0]
%vu8Mask_5.1 = phi i8 [ 0, %bb297 ], [ 0, %bb210.preheader ] ; <i8> [#uses=1]
%vu8Mask_6.1 = phi i8 [ 0, %bb297 ], [ 0, %bb210.preheader ] ; <i8> [#uses=0]
%vu8Mask_7.1 = phi i8 [ 0, %bb297 ], [ 0, %bb210.preheader ] ; <i8> [#uses=1]
%vu8Mask_12.1 = phi i8 [ -1, %bb297 ], [ 0, %bb210.preheader ] ; <i8> [#uses=0]
%vu8Mask_13.1 = phi i8 [ -1, %bb297 ], [ 0, %bb210.preheader ] ; <i8> [#uses=0]
icmp eq i32 %118, 2 ; <i1>:121 [#uses=0]
and i8 %vu8Mask_1.1, 3 ; <i8>:122 [#uses=0]
and i8 %vu8Mask_5.1, 3 ; <i8>:123 [#uses=0]
and i8 %vu8Mask_3.1, %18 ; <i8>:124 [#uses=0]
and i8 %vu8Mask_7.1, %18 ; <i8>:125 [#uses=0]
icmp eq i8 %104, %107 ; <i1>:126 [#uses=1]
br i1 %126, label %bb328, label %bb303
bb303: ; preds = %bb298
call i16 @llvm.bswap.i16( i16 %81 ) ; <i16>:127 [#uses=1]
sub i16 %63, %77 ; <i16>:128 [#uses=3]
sub i16 %65, %79 ; <i16>:129 [#uses=3]
sub i16 %67, %73 ; <i16>:130 [#uses=3]
sub i16 %69, %75 ; <i16>:131 [#uses=3]
sub i16 %71, %127 ; <i16>:132 [#uses=3]
sub i16 0, %128 ; <i16>:133 [#uses=1]
icmp slt i16 %128, 0 ; <i1>:134 [#uses=1]
%.673 = select i1 %134, i16 %133, i16 %128 ; <i16> [#uses=1]
sub i16 0, %129 ; <i16>:135 [#uses=1]
icmp slt i16 %129, 0 ; <i1>:136 [#uses=1]
%.674 = select i1 %136, i16 %135, i16 %129 ; <i16> [#uses=1]
sub i16 0, %130 ; <i16>:137 [#uses=1]
icmp slt i16 %130, 0 ; <i1>:138 [#uses=1]
%.675 = select i1 %138, i16 %137, i16 %130 ; <i16> [#uses=1]
sub i16 0, %131 ; <i16>:139 [#uses=1]
icmp slt i16 %131, 0 ; <i1>:140 [#uses=1]
%.676 = select i1 %140, i16 %139, i16 %131 ; <i16> [#uses=1]
sub i16 0, %132 ; <i16>:141 [#uses=1]
icmp slt i16 %132, 0 ; <i1>:142 [#uses=1]
%.677 = select i1 %142, i16 %141, i16 %132 ; <i16> [#uses=1]
br label %bb328
bb328: ; preds = %bb303, %bb298
%vu16Delta_0.0 = phi i16 [ %.673, %bb303 ], [ %., %bb298 ] ; <i16> [#uses=1]
%vu16Delta_1.0 = phi i16 [ %.674, %bb303 ], [ %.660, %bb298 ] ; <i16> [#uses=0]
%vu16Delta_2.0 = phi i16 [ %.675, %bb303 ], [ %.661, %bb298 ] ; <i16> [#uses=0]
%vu16Delta_3.0 = phi i16 [ %.676, %bb303 ], [ %.662, %bb298 ] ; <i16> [#uses=0]
%vu16Delta_6.0 = phi i16 [ %.677, %bb303 ], [ %.663, %bb298 ] ; <i16> [#uses=0]
lshr i16 %vu16Delta_0.0, 8 ; <i16>:143 [#uses=1]
trunc i16 %143 to i8 ; <i8>:144 [#uses=1]
and i8 %144, %vu8Mask_0.1 ; <i8>:145 [#uses=1]
icmp eq i8 %145, 0 ; <i1>:146 [#uses=0]
sub i32 %105, %114 ; <i32>:147 [#uses=1]
sub i32 %111, %108 ; <i32>:148 [#uses=1]
or i32 %147, %148 ; <i32>:149 [#uses=1]
icmp eq i32 %149, 0 ; <i1>:150 [#uses=0]
call i32 @_Z54S_CalcIfLargeMVDeltaForBMbBothPredictionsFromSameFramePK19BiPartSrcDescriptorS1_ijj( %struct.BiPartSrcDescriptor* %57, %struct.BiPartSrcDescriptor* %59, i32 %19, i32 0, i32 0 ) nounwind ; <i32>:151 [#uses=0]
unreachable
labelContinueEdgesLoopV: ; preds = %bb206, %bb205, %bb144
%fEdgeHasNonZeroBS.0 = phi i32 [ 0, %bb205 ], [ 0, %bb144 ], [ 1, %bb206 ] ; <i32> [#uses=2]
%fMacroblockHasNonZeroBS.6 = phi i32 [ %152, %bb205 ], [ %fMacroblockHasNonZeroBS.4, %bb144 ], [ %152, %bb206 ] ; <i32> [#uses=1]
%ixEdge.1 = phi i32 [ %160, %bb205 ], [ 0, %bb144 ], [ %160, %bb206 ] ; <i32> [#uses=1]
%bfNZ12.2 = phi i32 [ %159, %bb205 ], [ 0, %bb144 ], [ %159, %bb206 ] ; <i32> [#uses=1]
%boundaryStrengthsV.3 = phi i8* [ %158, %bb205 ], [ %boundaryStrengthsV.1771, %bb144 ], [ %158, %bb206 ] ; <i8*> [#uses=3]
or i32 %fMacroblockHasNonZeroBS.6, %fEdgeHasNonZeroBS.0 ; <i32>:152 [#uses=2]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
load i8, i8* %boundaryStrengthsV.3, align 1 ; <i8>:153 [#uses=1]
LegalizeTypes support for promotion of bswap. In LegalizeDAG the value is zero-extended to the new type before byte swapping. It doesn't matter how the extension is done since the new bits are shifted off anyway after the swap, so extend by any old rubbish bits. This results in the final assembler for the testcase being one line shorter. llvm-svn: 53604
2008-07-15 18:18:22 +08:00
trunc i32 %fEdgeHasNonZeroBS.0 to i8 ; <i8>:154 [#uses=1]
shl i8 %154, 5 ; <i8>:155 [#uses=1]
xor i8 %155, 32 ; <i8>:156 [#uses=1]
or i8 %153, %156 ; <i8>:157 [#uses=1]
store i8 %157, i8* %boundaryStrengthsV.3, align 1
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
getelementptr i8, i8* %boundaryStrengthsV.3, i32 4 ; <i8*>:158 [#uses=4]
LegalizeTypes support for promotion of bswap. In LegalizeDAG the value is zero-extended to the new type before byte swapping. It doesn't matter how the extension is done since the new bits are shifted off anyway after the swap, so extend by any old rubbish bits. This results in the final assembler for the testcase being one line shorter. llvm-svn: 53604
2008-07-15 18:18:22 +08:00
shl i32 %bfNZ12.2, 4 ; <i32>:159 [#uses=4]
add i32 %ixEdge.1, 1 ; <i32>:160 [#uses=6]
icmp ult i32 %160, %numEdgesToTest.2 ; <i1>:161 [#uses=1]
br i1 %161, label %bb200, label %bb395
bb395: ; preds = %labelContinueEdgesLoopV
add i32 %idxEachField11.0773, 1 ; <i32>:162 [#uses=2]
icmp ugt i32 %162, 0 ; <i1>:163 [#uses=1]
br i1 %163, label %bb398, label %bb144
bb398: ; preds = %bb395
call void asm sideeffect "dcbt $0, $1", "b%,r,~{memory}"( i32 19, i32* null ) nounwind
unreachable
bb642: ; preds = %entry
ret i32 0
}
declare i16 @llvm.bswap.i16(i16) nounwind readnone
declare i8* @jvtNewPtrVectorAligned(i32)
declare i8* @jvtNewPtr(i32)
declare i8* @jvtNewPtrMemAligned(i32)
declare %struct.Mutex* @MutexNew()
declare i8* @_Znam(i32)
declare i32 @_Z24LoopFilter_FilterMbGroupP14LoopFilterInfoP11FramePixelsP22FrameMotionVectorCacheP19ThreadedBatchStructjjij(%struct.LoopFilterInfo*, %struct.FramePixels*, %struct.FrameMotionVectorCache*, %struct.ThreadedBatch*, i32, i32, i32, i32)
declare void @MutexLock(%struct.Mutex*)
declare void @MutexUnlock(%struct.Mutex*)
declare i32 @_Z35LoopFilter_Internal_FilterLumaPlanePK14LoopFilterInfojjjjj(%struct.LoopFilterInfo*, i32, i32, i32, i32, i32)
declare i32 @_Z37LoopFilter_Internal_FilterChromaPlanePK14LoopFilterInfojjjjj(%struct.LoopFilterInfo*, i32, i32, i32, i32, i32)
declare void @_Z44LoopFilter_Internal_filter_macroblock_chromaPK14LoopFilterInfoPhS2_iiiPK30PerMacroblockBoundaryStrengthsjj(%struct.LoopFilterInfo*, i8*, i8*, i32, i32, i32, %struct.PerMacroblockBoundaryStrengths*, i32, i32) nounwind
declare i32 @_Z42LoopFilter_Internal_FilterChromaPlaneMBAFFPK14LoopFilterInfojjj(%struct.LoopFilterInfo*, i32, i32, i32) nounwind
declare i32 @_Z26LF_Threading2_ProcessTasksP14LoopFilterInfoP11FramePixelsP22FrameMotionVectorCacheij(%struct.LoopFilterInfo*, %struct.FramePixels*, %struct.FrameMotionVectorCache*, i32, i32)
declare i32 @_Z46LoopFilter_Internal_CalculateBoundaryStrengthsPK14LoopFilterInfoP22FrameMotionVectorCachejj(%struct.LoopFilterInfo*, %struct.FrameMotionVectorCache*, i32, i32)
declare i32 @_Z44LoopFilter_Internal_FilterLumaChromaPlane_PPP14LoopFilterInfojjjjj(%struct.LoopFilterInfo*, i32, i32, i32, i32, i32)
declare i32 @_Z22LoopFilter_FilterFrameP14LoopFilterInfoP11FramePixelsP22FrameMotionVectorCacheP19ThreadedBatchStructij(%struct.LoopFilterInfo*, %struct.FramePixels*, %struct.FrameMotionVectorCache*, %struct.ThreadedBatch*, i32, i32)
declare void @_Z34LF_Threading2_ProcessTasks_WrapperPv(i8*)