This is an archive of the discontinued LLVM Phabricator instance.

Differential D74954

Add a basic tiling pass for parallel loops
ClosedPublic

Authored by bkramer on Feb 21 2020, 4:31 AM.

Details

Reviewers
herhut
nicolasvasilache
ftynse
Commits
rGbc1947a6f51f: Add a basic tiling pass for parallel loops
Summary

This exploits the fact that the iterations of parallel loops are
independent so tiling becomes just an index transformation. This pass
only tiles the innermost loop of a loop nest.

The ultimate goal is to allow vectorization of the tiled loops, but I
don't think we're there yet with the current rewriting, as the tiled
loops don't have a constant trip count.

Diff Detail

Event Timeline

bkramer created this revision.Feb 21 2020, 4:31 AM
Herald added a reviewer: nicolasvasilache. · View Herald TranscriptFeb 21 2020, 4:31 AM
Herald added a project: Restricted Project. · View Herald Transcript
Herald added subscribers: llvm-commits, Joonsoo, liufengdb and 12 others. · View Herald Transcript
bkramer updated this revision to Diff 245822.Feb 21 2020, 4:53 AM
  • use pass option instead of global option
  • fix the recursive walker to walk recursively
ftynse added a reviewer: ftynse.Feb 21 2020, 5:00 AM
herhut requested changes to this revision.Feb 21 2020, 5:01 AM
herhut added inline comments.
mlir/lib/Dialect/LoopOps/Transforms/ParallelLoopTiling.cpp
25

Please use pass options.

30

nit: Doc comments need to start with ///

Use parallel loop instead of ploop.

42

tileParallelLoop

44

Maybe newSteps.reserve?

68

I do not understand this part. We want the min of either the step size or dim - upperbound_of_outer.

77

you can do something like innerLoop.region().takeBody(op.region()) which simply transfers one region to the other. As the number of block arguments does not change, no need to fix uses or anything.

95

getBlock().getOps<loop::ParallelOp>() gives an (potentially empty) iterator over all parallel loops in the block.

This revision now requires changes to proceed.Feb 21 2020, 5:01 AM
Harbormaster completed remote builds in B46985: Diff 245816.Feb 21 2020, 5:02 AM
ftynse added a subscriber: poechsel.Feb 21 2020, 5:04 AM

We already several implementations of loop tiling (lib/Transforms/LoopTiling.cpp, lib/Dialect/Linalg/Transforms/Tiling.cpp, lib/Transforms/Utils/LoopUtils.cpp). Have you considered generalizing them instead of introducing yet another one? I suppose parallel loop nest as an operation removes the need for any supplementary preconditions, but the mechanics of the transformation should be very similar between parallel and sequential loops.

I agree we may want to peel off the last iterations to avoid complex conditions, but IMO it's better as an option to the transformation. @poechsel had a similar requested with loops produced by Linalg tiling, so it would make sense to try and reuse the infra here.

ftynse requested changes to this revision.Feb 21 2020, 5:07 AM
ftynse added inline comments.
mlir/test/Dialect/Loops/parallel-loop-tiling.mlir
20

Please don't pattern-match SSA names, and generally prefer avoiding non-essential parts of the IR from the test (e.g. repeated types in the pattern).

https://mlir.llvm.org/getting_started/TestingGuide/

Harbormaster completed remote builds in B46987: Diff 245822.Feb 21 2020, 5:20 AM
bkramer updated this revision to Diff 245832.Feb 21 2020, 6:00 AM
bkramer marked 9 inline comments as done.
  • One round of addressed review comments
herhut added a comment.Feb 21 2020, 6:18 AM
In D74954#1886199, @ftynse wrote:

We already several implementations of loop tiling (lib/Transforms/LoopTiling.cpp, lib/Dialect/Linalg/Transforms/Tiling.cpp, lib/Transforms/Utils/LoopUtils.cpp). Have you considered generalizing them instead of introducing yet another one? I suppose parallel loop nest as an operation removes the need for any supplementary preconditions, but the mechanics of the transformation should be very similar between parallel and sequential loops.

The complexity of the tiling passes is in the dependency analysis and not the rewriting itself. This one is purely structural. We would need to split the actual tiling rewrite form the analysis with interfaces for generating the tiled operation. I am not sure that is worth the code we would actually get to reuse.

mlir/lib/Dialect/LoopOps/Transforms/ParallelLoopTiling.cpp
53

The replication part could be a separate pattern/pass that rewrites a loop with dynamic upper bounds specified with a constant upper bound (so an affine.min with a constant) into something like

bound = affine.min(constant, other thing)
if (bound ==constant) {
  // loop with constant upper bound
} else {
  // loop with dynamic upper bound
}

Maybe this is good enough to make LLVM recognize the potential for vectorization. That rewrite could be applied independently.

97

Can we have an ArrayRef<unsigned> for tile sizes? So we can have different ones at different levels? I would suspect that we, for vectorization, always want to tile 1x...x1xN.

herhut added inline comments.Feb 21 2020, 6:19 AM
mlir/lib/Dialect/LoopOps/Transforms/ParallelLoopTiling.cpp
68

I always get confused by this myself. I think it needs to be upper_bound - current_index_of_outer_loop to get the number of remaining iterations. And then the min with the tiling size.

Harbormaster completed remote builds in B46991: Diff 245832.Feb 21 2020, 6:33 AM
bkramer updated this revision to Diff 245847.Feb 21 2020, 7:08 AM
bkramer marked an inline comment as done.
  • Make tileSizes a list
  • Fix min again
Harbormaster completed remote builds in B47001: Diff 245847.Feb 21 2020, 7:28 AM
dcaballe added a subscriber: dcaballe.Feb 21 2020, 11:30 AM

Just adding some minor nit comments!

We already several implementations of loop tiling (lib/Transforms/LoopTiling.cpp, lib/Dialect/Linalg/Transforms/Tiling.cpp, lib/Transforms/Utils/LoopUtils.cpp). Have you considered generalizing them instead of introducing yet another one? I suppose parallel loop nest as an operation removes the need for any supplementary preconditions, but the mechanics of the transformation should be very similar between parallel and sequential loops.

The complexity of the tiling passes is in the dependency analysis and not the rewriting itself. This one is purely structural. We would need to split the actual tiling rewrite form the analysis with interfaces for generating the tiled operation. I am not sure that is worth the code we would actually get to reuse.

I think it might be worth it since affine parallel for would be another candidate for tiling.

mlir/lib/Dialect/LoopOps/Transforms/ParallelLoopTiling.cpp
41

size_t i = 0; i != op.upperBound().size(); -> size_t i = 0, end = op.upperBound().size(); i != end; ?

116

please drop {} for single line for and if-else statements

mlir/test/Dialect/Loops/parallel-loop-tiling.mlir
38

Please add // ----- between tests for split-input-file to work

herhut added a comment.Feb 24 2020, 1:27 AM
In D74954#1887027, @dcaballe wrote:

Just adding some minor nit comments!

We already several implementations of loop tiling (lib/Transforms/LoopTiling.cpp, lib/Dialect/Linalg/Transforms/Tiling.cpp, lib/Transforms/Utils/LoopUtils.cpp). Have you considered generalizing them instead of introducing yet another one? I suppose parallel loop nest as an operation removes the need for any supplementary preconditions, but the mechanics of the transformation should be very similar between parallel and sequential loops.

The complexity of the tiling passes is in the dependency analysis and not the rewriting itself. This one is purely structural. We would need to split the actual tiling rewrite form the analysis with interfaces for generating the tiled operation. I am not sure that is worth the code we would actually get to reuse.

I think it might be worth it since affine parallel for would be another candidate for tiling.

I guess we would need to compute the new step sizes as affine expressions, as well, in such case. My point was how much code reuse we would get.

mlir/lib/Dialect/LoopOps/Transforms/ParallelLoopTiling.cpp
34

This comment does not state the actual rewriting, does it?

herhut accepted this revision.Feb 24 2020, 1:33 AM

Thanks!

LGTM with comments addressed.

mlir/lib/Dialect/LoopOps/Transforms/ParallelLoopTiling.cpp
40

mega-nit: tileSizeConstants.reserve

bkramer updated this revision to Diff 246166.Feb 24 2020, 2:42 AM
bkramer marked 4 inline comments as done.
  • Address moar comments
Herald added a reviewer: • espindola. · View Herald TranscriptFeb 24 2020, 2:42 AM
Herald added a reviewer: alexander-shaposhnikov. · View Herald Transcript
Herald added a reviewer: rupprecht. · View Herald Transcript
Herald added a reviewer: jhenderson. · View Herald Transcript
Herald added a reviewer: jdoerfert. · View Herald Transcript
Herald added a reviewer: sstefan1. · View Herald Transcript
Herald added a reviewer: mravishankar. · View Herald Transcript
Herald added a reviewer: antiagainst. · View Herald Transcript
Herald added a reviewer: rriddle. · View Herald Transcript
Herald added a reviewer: antiagainst. · View Herald Transcript
Herald added a reviewer: uenoku. · View Herald Transcript
Herald added projects: Restricted Project, Restricted Project, Restricted Project, Restricted Project, Restricted Project. · View Herald Transcript
Herald added subscribers: libc-commits, libcxx-commits, lldb-commits and 30 others. · View Herald Transcript
bkramer updated this revision to Diff 246167.Feb 24 2020, 2:43 AM
  • WTF phab?
Herald added a reviewer: mclow.lists. · View Herald TranscriptFeb 24 2020, 2:43 AM
Herald added a subscriber: • wuzish. · View Herald Transcript
teemperor removed subscribers: cfe-commits, lldb-commits, libcxx-commits, libc-commits.Feb 24 2020, 2:47 AM
This revision was not accepted when it landed; it landed in state Needs Review.Feb 24 2020, 2:48 AM
Closed by commit rGbc1947a6f51f: Add a basic tiling pass for parallel loops (authored by bkramer). · Explain Why
This revision was automatically updated to reflect the committed changes.
teemperor removed reviewers: • espindola, alexander-shaposhnikov, rupprecht, jhenderson, jdoerfert, sstefan1, mravishankar, antiagainst, rriddle, uenoku, mclow.lists.Feb 24 2020, 2:49 AM
teemperor removed projects: Restricted Project, Restricted Project, Restricted Project, Restricted Project, Restricted Project.
Harbormaster failed remote builds in B47116: Diff 246166!Feb 24 2020, 3:19 AM
Harbormaster completed remote builds in B47117: Diff 246167.Feb 24 2020, 3:29 AM
rriddle added inline comments.Feb 24 2020, 9:09 AM
mlir/lib/Dialect/LoopOps/Transforms/ParallelLoopTiling.cpp
113

nit: Please drop the trivial braces. Here and below.

Large Diff

This large diff affects 276 files. Files without inline comments have been collapsed. Expand All Files

Revision Contents

PathSize
clang-tools-extra/
clangd/
7 lines
73 lines
7 lines
49 lines
76 lines
3 lines
86 lines
refactor/
3 lines
2 lines
6 lines
unittests/
4 lines
2 lines
71 lines
57 lines
2 lines
clang/
include/
clang/
Driver/
2 lines
11 lines
Parse/
3 lines
lib/
Basic/
Targets/
7 lines
2 lines
Driver/
ToolChains/
27 lines
2 lines
Frontend/
13 lines
Sema/
3 lines
test/
CodeGen/
10 lines
Driver/
5 lines
6 lines
14 lines
Preprocessor/
3 lines
701 lines
1334 lines
Sema/
2 lines
8 lines
SemaCXX/
5 lines
compiler-rt/
lib/
profile/
94 lines
libc/
1 line
cmake/
modules/
67 lines
docs/
15 lines
30 lines
fuzzing/
5 lines
string/
7 lines
38 lines
src/
signal/
linux/
2 lines
utils/
CPP/
10 lines
HdrGen/
5 lines
UnitTest/
23 lines
libcxx/
test/
support/
7 lines
7 lines
lld/
ELF/
2 lines
test/
ELF/
11 lines
10 lines
lldb/
packages/
Python/
lldbsuite/
test/
15 lines
source/
Plugins/
ExpressionParser/
Clang/
15 lines
260 lines
70 lines
179 lines
LanguageRuntime/
ObjC/
AppleObjCRuntime/
57 lines
ScriptInterpreter/
Python/
2 lines
SymbolFile/
DWARF/
9 lines
4 lines
12 lines
52 lines
2 lines
18 lines
test/
API/
functionalities/
data-formatter/
data-formatter-stl/
libcxx/
string/
4 lines
lang/
cpp/
operators/
2 lines
Shell/
SymbolFile/
Breakpad/
Inputs/
2 lines
DWARF/
18 lines
4 lines
1 line
llvm/
docs/
4 lines
130 lines
3 lines
include/
llvm/
ADT/
10 lines
Analysis/
5 lines
8 lines
CodeGen/
GlobalISel/
3 lines
5 lines
7 lines
42 lines
4 lines
ExecutionEngine/
JITLink/
17 lines
1 line
Orc/
38 lines
4 lines
IR/
8 lines
16 lines
MC/
7 lines
ObjectYAML/
2 lines
Target/
2 lines
lib/
Analysis/
36 lines
CodeGen/
32 lines
GlobalISel/
42 lines
11 lines
13 lines
SelectionDAG/
6 lines
130 lines
17 lines
4 lines
9 lines
27 lines
2 lines
19 lines
2 lines
ExecutionEngine/
JITLink/
30 lines
105 lines
114 lines
106 lines
Orc/
36 lines
20 lines
27 lines
30 lines
OrcError/
4 lines
IR/
6 lines
MC/
MCParser/
19 lines
ObjectYAML/
38 lines
2 lines
Passes/
16 lines
Target/
AMDGPU/
3 lines
17 lines
MCTargetDesc/
2 lines
ARM/
3 lines
Hexagon/
2 lines
2 lines
3 lines
PowerPC/
MCTargetDesc/
1 line
269 lines
SystemZ/
9 lines
143 lines
17 lines
VE/
MCTargetDesc/
2 lines
WebAssembly/
28 lines
X86/
MCTargetDesc/
2 lines
157 lines
105 lines
210 lines
95 lines
107 lines
57 lines
Transforms/
InstCombine/
7 lines
Instrumentation/
143 lines
Scalar/
16 lines
Utils/
4 lines
test/
CodeGen/
AArch64/
34 lines
AMDGPU/
3 lines
SystemZ/
20 lines
72 lines
54 lines
Thumb2/
LowOverheadLoops/
229 lines
X86/
312 lines
37 lines
501 lines
142 lines
29 lines
68 lines
80 lines
32 lines
1411 lines
143 lines
528 lines
2 lines
ExecutionEngine/
JITLink/
X86/
Inputs/
8 lines
31 lines
12 lines
OrcLazy/
28 lines
MC/
ELF/
14 lines
12 lines
12 lines
11 lines
Other/
33 lines
40 lines
Transforms/
Attributor/
1 line
CodeGenPrepare/
AArch64/
16 lines
X86/
31 lines
DeadStoreElimination/
MSSA/
1 line
1 line
1 line
1 line
3 lines
InstCombine/
16 lines
tools/
llvm-objcopy/
ELF/
6 lines
llvm-objdump/
AMDGPU/
1 line
Hexagon/
1 line
X86/
108 lines
1 line
1 line
llvm-readobj/
ELF/
8 lines
4 lines
obj2yaml/
1 line
yaml2obj/
ELF/
98 lines
tools/
lli/
98 lines
llvm-objdump/
78 lines
obj2yaml/
3 lines
unittests/
DebugInfo/
DWARF/
228 lines
Transforms/
Utils/
11 lines
utils/
TableGen/
4 lines
mlir/
docs/
103 lines
106 lines
40 lines
Tutorials/
Toy/
186 lines
32 lines
32 lines
14 lines
10 lines
68 lines
examples/
toy/
Ch2/
include/
toy/
39 lines
mlir/
75 lines
Ch3/
include/
toy/
47 lines
mlir/
75 lines
Ch4/
include/
toy/
47 lines
mlir/
75 lines
Ch5/
include/
toy/
47 lines
mlir/
75 lines
Ch6/
include/
toy/
45 lines
mlir/
75 lines
Ch7/
include/
toy/
64 lines
mlir/
75 lines
include/
mlir/
Dialect/
LLVMIR/
19 lines
31 lines
LoopOps/
5 lines
SPIRV/
20 lines
StandardOps/
IR/
49 lines
VectorOps/
9 lines
IR/
25 lines
11 lines
5 lines
1 line
Support/
20 lines
TableGen/
2 lines
18 lines
44 lines
lib/
Dialect/
LLVMIR/
IR/
70 lines
LoopOps/
Transforms/
1 line
133 lines
SPIRV/
30 lines
StandardOps/
IR/
298 lines
VectorOps/
52 lines
Parser/
9 lines
TableGen/
1 line
2 lines
48 lines
24 lines
test/
Dialect/
Loops/
80 lines
SPIRV/
6 lines
Examples/
Toy/
Ch2/
24 lines
8 lines
Ch3/
24 lines
8 lines
Ch4/
24 lines
8 lines
34 lines
Ch5/
14 lines
24 lines
8 lines
34 lines
Ch6/
14 lines
24 lines
10 lines
8 lines
34 lines
Ch7/
14 lines
24 lines
10 lines
8 lines
34 lines
30 lines
19 lines
IR/
8 lines
1 line
lib/
TestDialect/
13 lines
mlir-tblgen/
92 lines
19 lines
tools/
mlir-tblgen/
136 lines
913 lines
mlir-vulkan-runner/
3 lines

Diff 246166

clang-tools-extra/clangd/ClangdServer.h

clang-tools-extra/clangd/ClangdServer.cpp

clang-tools-extra/clangd/Hover.cpp

clang-tools-extra/clangd/Selection.h

clang-tools-extra/clangd/Selection.cpp

clang-tools-extra/clangd/SemanticSelection.cpp

clang-tools-extra/clangd/XRefs.cpp

clang-tools-extra/clangd/refactor/Rename.cpp

clang-tools-extra/clangd/refactor/Tweak.h

clang-tools-extra/clangd/refactor/Tweak.cpp

clang-tools-extra/clangd/unittests/FindTargetTests.cpp

clang-tools-extra/clangd/unittests/HoverTests.cpp

clang-tools-extra/clangd/unittests/SelectionTests.cpp

clang-tools-extra/clangd/unittests/TweakTesting.cpp

clang-tools-extra/clangd/unittests/TweakTests.cpp

clang/include/clang/Driver/CC1Options.td

clang/include/clang/Driver/Options.td

clang/include/clang/Parse/Parser.h

clang/lib/Basic/Targets/AArch64.cpp

clang/lib/Basic/Targets/X86.cpp

clang/lib/Driver/ToolChains/Clang.cpp

clang/lib/Driver/ToolChains/Darwin.cpp

clang/lib/Frontend/CompilerInvocation.cpp

clang/lib/Sema/SemaCast.cpp

clang/test/CodeGen/codemodels.c

clang/test/Driver/clang_f_opts.c

clang/test/Driver/code-model.c

clang/test/Driver/mbackchain.c

clang/test/Driver/mcmodel.c

clang/test/Preprocessor/aarch64-target-features.c

clang/test/Preprocessor/init-aarch64.c

clang/test/Preprocessor/init.c

clang/test/Sema/MicrosoftExtensions.c

clang/test/Sema/cast.c

clang/test/SemaCXX/cstyle-cast.cpp

compiler-rt/lib/profile/GCDAProfiling.c

libc/CMakeLists.txt

libc/cmake/modules/LLVMLibCRules.cmake

libc/docs/fuzzing.rst

libc/docs/source_layout.rst

libc/fuzzing/CMakeLists.txt

libc/fuzzing/string/CMakeLists.txt

libc/fuzzing/string/strcpy_fuzz.cpp

libc/src/signal/linux/raise.cpp

libc/utils/CPP/README.md

libc/utils/HdrGen/README.md

libc/utils/UnitTest/README.md

libcxx/test/support/count_new.h

libcxx/test/support/type_id.h

lld/ELF/Writer.cpp

lld/test/ELF/shuffle-sections-init-fini.s

lld/test/ELF/shuffle-sections.s

lldb/packages/Python/lldbsuite/test/lldbtest.py

lldb/source/Plugins/ExpressionParser/Clang/ClangASTSource.h

lldb/source/Plugins/ExpressionParser/Clang/ClangASTSource.cpp

lldb/source/Plugins/ExpressionParser/Clang/ClangExpressionDeclMap.h

lldb/source/Plugins/ExpressionParser/Clang/ClangExpressionDeclMap.cpp

lldb/source/Plugins/LanguageRuntime/ObjC/AppleObjCRuntime/AppleObjCDeclVendor.cpp

lldb/source/Plugins/ScriptInterpreter/Python/ScriptInterpreterPython.cpp

lldb/source/Plugins/SymbolFile/DWARF/DebugNamesDWARFIndex.h

lldb/source/Plugins/SymbolFile/DWARF/DebugNamesDWARFIndex.cpp

lldb/source/Plugins/SymbolFile/DWARF/ManualDWARFIndex.h

lldb/source/Plugins/SymbolFile/DWARF/ManualDWARFIndex.cpp

lldb/source/Plugins/SymbolFile/DWARF/SymbolFileDWARF.h

lldb/source/Plugins/SymbolFile/DWARF/SymbolFileDWARF.cpp

lldb/test/API/functionalities/data-formatter/data-formatter-stl/libcxx/string/TestDataFormatterLibcxxString.py

lldb/test/API/lang/cpp/operators/main.cpp

lldb/test/Shell/SymbolFile/Breakpad/Inputs/basic-elf.yaml

lldb/test/Shell/SymbolFile/DWARF/dwp-debug-types.s

lldb/test/Shell/SymbolFile/DWARF/dwp.s

lldb/test/Shell/lit-lldb-init.in

llvm/docs/Extensions.rst

llvm/docs/LangRef.rst

llvm/docs/LoopTerminology.rst

llvm/include/llvm/ADT/STLExtras.h

llvm/include/llvm/Analysis/CFGPrinter.h

llvm/include/llvm/Analysis/LoopInfo.h

llvm/include/llvm/CodeGen/GlobalISel/GISelKnownBits.h

llvm/include/llvm/CodeGen/ISDOpcodes.h

llvm/include/llvm/CodeGen/ScoreboardHazardRecognizer.h

llvm/include/llvm/CodeGen/SelectionDAG.h

llvm/include/llvm/CodeGen/TargetLowering.h

llvm/include/llvm/ExecutionEngine/JITLink/JITLink.h

llvm/include/llvm/ExecutionEngine/JITLink/MachO_x86_64.h

llvm/include/llvm/ExecutionEngine/Orc/Core.h

llvm/include/llvm/ExecutionEngine/Orc/OrcError.h

llvm/include/llvm/IR/Intrinsics.td

llvm/include/llvm/IR/PatternMatch.h

llvm/include/llvm/MC/MCInstrItineraries.h

llvm/include/llvm/ObjectYAML/ELFYAML.h

llvm/include/llvm/Target/TargetSelectionDAG.td

llvm/lib/Analysis/CFGPrinter.cpp

llvm/lib/CodeGen/CodeGenPrepare.cpp

llvm/lib/CodeGen/GlobalISel/GISelKnownBits.cpp

llvm/lib/CodeGen/MachinePipeliner.cpp

llvm/lib/CodeGen/ScoreboardHazardRecognizer.cpp

llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp

llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp

llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

llvm/lib/CodeGen/TargetLoweringBase.cpp

llvm/lib/ExecutionEngine/JITLink/BasicGOTAndStubsBuilder.h

llvm/lib/ExecutionEngine/JITLink/JITLinkGeneric.h

llvm/lib/ExecutionEngine/JITLink/JITLinkGeneric.cpp

llvm/lib/ExecutionEngine/JITLink/MachO_x86_64.cpp

llvm/lib/ExecutionEngine/Orc/CompileOnDemandLayer.cpp

llvm/lib/ExecutionEngine/Orc/Core.cpp

llvm/lib/ExecutionEngine/Orc/LLJIT.cpp

llvm/lib/ExecutionEngine/Orc/ObjectLinkingLayer.cpp

llvm/lib/ExecutionEngine/OrcError/OrcError.cpp

llvm/lib/IR/Verifier.cpp

llvm/lib/MC/MCParser/ELFAsmParser.cpp

llvm/lib/ObjectYAML/ELFEmitter.cpp

llvm/lib/ObjectYAML/ELFYAML.cpp

llvm/lib/Passes/StandardInstrumentations.cpp

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

llvm/lib/Target/AMDGPU/MCTargetDesc/AMDGPUTargetStreamer.cpp

llvm/lib/Target/ARM/ARMISelLowering.cpp

llvm/lib/Target/Hexagon/HexagonInstrInfo.h

llvm/lib/Target/Hexagon/HexagonInstrInfo.cpp

llvm/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp

llvm/lib/Target/PowerPC/MCTargetDesc/PPCAsmBackend.cpp

llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp

llvm/lib/Target/SystemZ/SystemZFrameLowering.h

llvm/lib/Target/SystemZ/SystemZFrameLowering.cpp

llvm/lib/Target/SystemZ/SystemZISelLowering.cpp

llvm/lib/Target/VE/MCTargetDesc/VEMCExpr.cpp

llvm/lib/Target/WebAssembly/WebAssemblyFixIrreducibleControlFlow.cpp

llvm/lib/Target/X86/MCTargetDesc/LLVMBuild.txt

llvm/lib/Target/X86/X86ISelDAGToDAG.cpp

llvm/lib/Target/X86/X86ISelLowering.h

llvm/lib/Target/X86/X86ISelLowering.cpp

llvm/lib/Target/X86/X86InstrAVX512.td

llvm/lib/Target/X86/X86InstrFragmentsSIMD.td

llvm/lib/Target/X86/X86InstrSSE.td

llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp

llvm/lib/Transforms/Instrumentation/GCOVProfiling.cpp

llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp

llvm/lib/Transforms/Utils/KnowledgeRetention.cpp

llvm/test/CodeGen/AArch64/sat-add.ll

llvm/test/CodeGen/AMDGPU/hsa.ll

llvm/test/CodeGen/SystemZ/frame-23.ll

llvm/test/CodeGen/SystemZ/frame-24.ll

llvm/test/CodeGen/SystemZ/frameaddr-02.ll

llvm/test/CodeGen/Thumb2/LowOverheadLoops/extending-loads.ll

llvm/test/CodeGen/X86/atomic-fp.ll

llvm/test/CodeGen/X86/atomic-load-store-wide.ll

llvm/test/CodeGen/X86/atomic-mi.ll

llvm/test/CodeGen/X86/atomic-non-integer.ll

llvm/test/CodeGen/X86/merge-consecutive-loads-128.ll

llvm/test/CodeGen/X86/recip-fastmath.ll

llvm/test/CodeGen/X86/recip-fastmath2.ll

llvm/test/CodeGen/X86/sat-add.ll

llvm/test/CodeGen/X86/sdiv_fix_sat.ll

llvm/test/CodeGen/X86/sse-intrinsics-fast-isel.ll

llvm/test/CodeGen/X86/udiv_fix_sat.ll

llvm/test/CodeGen/X86/vec_fneg.ll

llvm/test/ExecutionEngine/JITLink/X86/Inputs/MachO_GOTAndStubsOptimizationHelper.s

llvm/test/ExecutionEngine/JITLink/X86/MachO_GOTAndStubsOptimization.s

llvm/test/ExecutionEngine/JITLink/X86/MachO_x86-64_relocations.s

llvm/test/ExecutionEngine/OrcLazy/static-initializers-in-objectfiles.ll

llvm/test/MC/ELF/exclude-debug-dwo.s

llvm/test/MC/ELF/section-entsize-changed.s

llvm/test/MC/ELF/section-flags-changed.s

llvm/test/MC/ELF/section-type-changed.s

llvm/test/Other/cfg_deopt_unreach.ll

llvm/test/Other/module-pass-printer.ll

llvm/test/Transforms/Attributor/liveness.ll

llvm/test/Transforms/CodeGenPrepare/AArch64/overflow-intrinsics.ll

llvm/test/Transforms/CodeGenPrepare/X86/overflow-intrinsics.ll

llvm/test/Transforms/DeadStoreElimination/MSSA/OverwriteStoreBegin.ll

llvm/test/Transforms/DeadStoreElimination/MSSA/OverwriteStoreEnd.ll

llvm/test/Transforms/DeadStoreElimination/MSSA/memset-missing-debugloc.ll

llvm/test/Transforms/DeadStoreElimination/MSSA/multiblock-loops.ll

llvm/test/Transforms/DeadStoreElimination/MSSA/multiblock-memintrinsics.ll

llvm/test/Transforms/InstCombine/fneg.ll

llvm/test/tools/llvm-objcopy/ELF/only-keep-debug.test

llvm/test/tools/llvm-objdump/AMDGPU/source-lines.ll

llvm/test/tools/llvm-objdump/Hexagon/source-interleave-hexagon.ll

llvm/test/tools/llvm-objdump/X86/source-interleave-function-from-debug.test

llvm/test/tools/llvm-objdump/X86/source-interleave-x86_64.test

llvm/test/tools/llvm-objdump/embedded-source.test

llvm/test/tools/llvm-readobj/ELF/verneed-invalid.test

llvm/test/tools/llvm-readobj/ELF/versioninfo.test

llvm/test/tools/obj2yaml/elf-gnu-hash-section.yaml

llvm/test/tools/yaml2obj/ELF/section-address-assign.yaml

llvm/tools/lli/lli.cpp

llvm/tools/llvm-objdump/llvm-objdump.cpp

llvm/tools/obj2yaml/elf2yaml.cpp

llvm/unittests/DebugInfo/DWARF/DWARFDebugLineTest.cpp

llvm/unittests/Transforms/Utils/KnowledgeRetentionTest.cpp

llvm/utils/TableGen/SubtargetEmitter.cpp

mlir/docs/OpDefinitions.md

mlir/docs/Quantization.md

mlir/docs/ShapeInference.md

mlir/docs/Tutorials/Toy/Ch-2.md

mlir/docs/Tutorials/Toy/Ch-3.md

mlir/docs/Tutorials/Toy/Ch-4.md

mlir/docs/Tutorials/Toy/Ch-5.md

mlir/docs/Tutorials/Toy/Ch-6.md

mlir/docs/Tutorials/Toy/Ch-7.md

mlir/examples/toy/Ch2/include/toy/Ops.td

mlir/examples/toy/Ch2/mlir/Dialect.cpp

mlir/examples/toy/Ch3/include/toy/Ops.td

mlir/examples/toy/Ch3/mlir/Dialect.cpp

mlir/examples/toy/Ch4/include/toy/Ops.td

mlir/examples/toy/Ch4/mlir/Dialect.cpp

mlir/examples/toy/Ch5/include/toy/Ops.td

mlir/examples/toy/Ch5/mlir/Dialect.cpp

mlir/examples/toy/Ch6/include/toy/Ops.td

mlir/examples/toy/Ch6/mlir/Dialect.cpp

mlir/examples/toy/Ch7/include/toy/Ops.td

mlir/examples/toy/Ch7/mlir/Dialect.cpp

mlir/include/mlir/Dialect/LLVMIR/LLVMOpBase.td

mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td

mlir/include/mlir/Dialect/LoopOps/Passes.h

mlir/include/mlir/Dialect/SPIRV/SPIRVControlFlowOps.td

mlir/include/mlir/Dialect/StandardOps/IR/Ops.td

mlir/include/mlir/Dialect/VectorOps/VectorOps.td

mlir/include/mlir/IR/OpBase.td

mlir/include/mlir/IR/OpImplementation.h

mlir/include/mlir/IR/OperationSupport.h

mlir/include/mlir/InitAllPasses.h

mlir/include/mlir/Support/STLExtras.h

mlir/include/mlir/TableGen/Constraint.h

mlir/include/mlir/TableGen/Operator.h

mlir/include/mlir/TableGen/Successor.h

mlir/lib/Dialect/LLVMIR/IR/LLVMDialect.cpp

mlir/lib/Dialect/LoopOps/Transforms/CMakeLists.txt

mlir/lib/Dialect/LoopOps/Transforms/ParallelLoopTiling.cpp

  • This file was added.
//===- ParallelLoopTiling.cpp - Tiles loop.parallel ---------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file implements loop tiling on parallel loops.
//
//===----------------------------------------------------------------------===//
#include "mlir/Dialect/AffineOps/AffineOps.h"
#include "mlir/Dialect/LoopOps/LoopOps.h"
#include "mlir/Dialect/LoopOps/Passes.h"
#include "mlir/Dialect/StandardOps/IR/Ops.h"
#include "mlir/Pass/Pass.h"
#include "mlir/Transforms/RegionUtils.h"
#include "llvm/Support/CommandLine.h"
using namespace mlir;
using loop::ParallelOp;
/// Tile a parallel loop of the form
/// loop.parallel (%i0, %i1) = (%arg0, %arg1) to (%arg2, %arg3)
herhutUnsubmitted
Done
ReplyInline Actions

Please use pass options.

herhut: Please use pass options.
/// step (%arg4, %arg5)
///
/// into
/// loop.parallel (%i0, %i1) = (%arg0, %arg1) to (%arg2, %arg3)
/// step (%arg4*tileSize[0],
herhutUnsubmitted
Done
ReplyInline Actions

nit: Doc comments need to start with ///

Use parallel loop instead of ploop.

herhut: nit: Doc comments need to start with `///` Use parallel loop instead of ploop.
/// %arg5*tileSize[1])
/// loop.parallel (%j0, %j1) = (0, 0) to (min(tileSize[0], %arg2-%j0)
/// min(tileSize[1], %arg3-%j1))
/// step (%arg4, %arg5)
herhutUnsubmitted
Done
ReplyInline Actions

This comment does not state the actual rewriting, does it?

herhut: This comment does not state the actual rewriting, does it?
/// The old loop is replaced with the new one.
static void tileParallelLoop(ParallelOp op, ArrayRef<int64_t> tileSizes) {
OpBuilder b(op);
auto zero = b.create<ConstantIndexOp>(op.getLoc(), 0);
SmallVector<Value, 2> tileSizeConstants;
tileSizeConstants.reserve(op.upperBound().size());
herhutUnsubmitted
Done
ReplyInline Actions

mega-nit: tileSizeConstants.reserve

herhut: mega-nit: `tileSizeConstants.reserve`
for (size_t i = 0, end = op.upperBound().size(); i != end; ++i) {
dcaballeUnsubmitted
Done
ReplyInline Actions

size_t i = 0; i != op.upperBound().size(); -> size_t i = 0, end = op.upperBound().size(); i != end; ?

dcaballe: `size_t i = 0; i != op.upperBound().size();` -> `size_t i = 0, end = op.upperBound().size(); i !
if (i < tileSizes.size())
herhutUnsubmitted
Done
ReplyInline Actions

tileParallelLoop

herhut: tileParallelLoop
tileSizeConstants.push_back(
b.create<ConstantIndexOp>(op.getLoc(), tileSizes[i]));
herhutUnsubmitted
Done
ReplyInline Actions

Maybe newSteps.reserve?

herhut: Maybe `newSteps.reserve`?
else
// Just pick 1 for the remaining dimensions.
tileSizeConstants.push_back(b.create<ConstantIndexOp>(op.getLoc(), 1));
}
// Create the outer loop with adjusted steps.
SmallVector<Value, 2> newSteps;
newSteps.reserve(op.step().size());
for (auto step : llvm::zip(op.step(), tileSizeConstants)) {
herhutUnsubmitted
Not Done
ReplyInline Actions

The replication part could be a separate pattern/pass that rewrites a loop with dynamic upper bounds specified with a constant upper bound (so an affine.min with a constant) into something like

bound = affine.min(constant, other thing)
if (bound ==constant) {
  // loop with constant upper bound
} else {
  // loop with dynamic upper bound
}

Maybe this is good enough to make LLVM recognize the potential for vectorization. That rewrite could be applied independently.

herhut: The replication part could be a separate pattern/pass that rewrites a loop with dynamic upper…
newSteps.push_back(
b.create<MulIOp>(op.getLoc(), std::get<0>(step), std::get<1>(step)));
}
auto outerLoop = b.create<ParallelOp>(op.getLoc(), op.lowerBound(),
op.upperBound(), newSteps);
b.setInsertionPointToStart(outerLoop.getBody());
// Compute min(size, dim - offset) to avoid out-of-bounds accesses.
// FIXME: Instead of using min, we want to replicate the tail. This would give
// the inner loop constant bounds for easy vectorization.
auto minMap = AffineMap::get(
/*dimCount=*/3, /*symbolCount=*/0,
{getAffineDimExpr(/*position=*/0, b.getContext()),
getAffineDimExpr(/*position=*/1, b.getContext()) -
getAffineDimExpr(/*position=*/2, b.getContext())});
herhutUnsubmitted
Done
ReplyInline Actions

I do not understand this part. We want the min of either the step size or dim - upperbound_of_outer.

herhut: I do not understand this part. We want the min of either the step size or dim…
herhutUnsubmitted
Not Done
ReplyInline Actions

I always get confused by this myself. I think it needs to be upper_bound - current_index_of_outer_loop to get the number of remaining iterations. And then the min with the tiling size.

herhut: I always get confused by this myself. I think it needs to be `upper_bound…
// Create the inner loop with adjusted bounds.
SmallVector<Value, 2> newBounds;
newBounds.reserve(op.upperBound().size());
for (auto bounds : llvm::zip(tileSizeConstants, outerLoop.upperBound(),
outerLoop.getInductionVars())) {
newBounds.push_back(b.create<AffineMinOp>(
op.getLoc(), b.getIndexType(), minMap,
ValueRange{std::get<0>(bounds), std::get<1>(bounds),
herhutUnsubmitted
Done
ReplyInline Actions

you can do something like innerLoop.region().takeBody(op.region()) which simply transfers one region to the other. As the number of block arguments does not change, no need to fix uses or anything.

herhut: you can do something like innerLoop.region().takeBody(op.region()) which simply transfers one…
std::get<2>(bounds)}));
}
auto innerLoop = b.create<ParallelOp>(
op.getLoc(), SmallVector<Value, 2>(newBounds.size(), zero), newBounds,
op.step());
// Steal the body of the old parallel loop and erase it.
innerLoop.region().takeBody(op.region());
op.erase();
}
/// Get a list of most nested parallel loops. Assumes that ParallelOps are only
/// directly nested.
static bool getInnermostNestedLoops(Block *block,
SmallVectorImpl<ParallelOp> &loops) {
bool hasInnerLoop = false;
for (auto parallelOp : block->getOps<ParallelOp>()) {
hasInnerLoop = true;
herhutUnsubmitted
Done
ReplyInline Actions

getBlock().getOps<loop::ParallelOp>() gives an (potentially empty) iterator over all parallel loops in the block.

herhut: `getBlock().getOps<loop::ParallelOp>()` gives an (potentially empty) iterator over all parallel…
if (!getInnermostNestedLoops(parallelOp.getBody(), loops))
loops.push_back(parallelOp);
herhutUnsubmitted
Not Done
ReplyInline Actions

Can we have an ArrayRef<unsigned> for tile sizes? So we can have different ones at different levels? I would suspect that we, for vectorization, always want to tile 1x...x1xN.

herhut: Can we have an ArrayRef<unsigned> for tile sizes? So we can have different ones at different…
}
return hasInnerLoop;
}
namespace {
struct ParallelLoopTiling : public FunctionPass<ParallelLoopTiling> {
ParallelLoopTiling() = default;
ParallelLoopTiling(const ParallelLoopTiling &) {} // tileSize is non-copyable.
explicit ParallelLoopTiling(ArrayRef<int64_t> tileSizes) {
this->tileSizes = tileSizes;
}
void runOnFunction() override {
SmallVector<ParallelOp, 2> mostNestedParallelOps;
for (Block &block : getFunction()) {
getInnermostNestedLoops(&block, mostNestedParallelOps);
rriddleUnsubmitted
Not Done
ReplyInline Actions

nit: Please drop the trivial braces. Here and below.

rriddle: nit: Please drop the trivial braces. Here and below.
}
for (ParallelOp pLoop : mostNestedParallelOps) {
tileParallelLoop(pLoop, tileSizes);
dcaballeUnsubmitted
Done
ReplyInline Actions

please drop {} for single line for and if-else statements

dcaballe: please drop {} for single line for and if-else statements
}
}
ListOption<int64_t> tileSizes{
*this, "parallel-loop-tile-sizes",
llvm::cl::desc("factors to tile parallel loops by"), llvm::cl::ZeroOrMore,
llvm::cl::MiscFlags::CommaSeparated};
};
} // namespace
std::unique_ptr<Pass>
mlir::createParallelLoopTilingPass(ArrayRef<int64_t> tileSizes) {
return std::make_unique<ParallelLoopTiling>(tileSizes);
}
static PassRegistration<ParallelLoopTiling> pass("parallel-loop-tiling",
"Tile parallel loops.");

mlir/lib/Dialect/SPIRV/SPIRVOps.cpp

mlir/lib/Dialect/StandardOps/IR/Ops.cpp

mlir/lib/Dialect/VectorOps/VectorOps.cpp

mlir/lib/Parser/Parser.cpp

mlir/lib/TableGen/CMakeLists.txt

mlir/lib/TableGen/Constraint.cpp

mlir/lib/TableGen/Operator.cpp

mlir/lib/TableGen/Successor.cpp

mlir/test/Dialect/Loops/parallel-loop-tiling.mlir

  • This file was added.
// RUN: mlir-opt %s -pass-pipeline='func(parallel-loop-tiling{parallel-loop-tile-sizes=1,4})' -split-input-file | FileCheck %s --dump-input-on-failure
func @parallel_loop(%arg0 : index, %arg1 : index, %arg2 : index,
%arg3 : index, %arg4 : index, %arg5 : index,
%A: memref<?x?xf32>, %B: memref<?x?xf32>,
%C: memref<?x?xf32>, %result: memref<?x?xf32>) {
loop.parallel (%i0, %i1) = (%arg0, %arg1) to (%arg2, %arg3) step (%arg4, %arg5) {
%B_elem = load %B[%i0, %i1] : memref<?x?xf32>
%C_elem = load %C[%i0, %i1] : memref<?x?xf32>
%sum_elem = addf %B_elem, %C_elem : f32
store %sum_elem, %result[%i0, %i1] : memref<?x?xf32>
}
return
}
// CHECK: #map0 = affine_map<(d0, d1, d2) -> (d0, d1 - d2)>
// CHECK-LABEL: func @parallel_loop(
// CHECK-SAME: [[VAL_0:%.*]]: index, [[VAL_1:%.*]]: index, [[VAL_2:%.*]]: index, [[VAL_3:%.*]]: index, [[VAL_4:%.*]]: index, [[VAL_5:%.*]]: index, [[VAL_6:%.*]]: memref<?x?xf32>, [[VAL_7:%.*]]: memref<?x?xf32>, [[VAL_8:%.*]]: memref<?x?xf32>, [[VAL_9:%.*]]: memref<?x?xf32>) {
// CHECK: [[VAL_10:%.*]] = constant 0 : index
// CHECK: [[VAL_11:%.*]] = constant 1 : index
ftynseUnsubmitted
Done
ReplyInline Actions

Please don't pattern-match SSA names, and generally prefer avoiding non-essential parts of the IR from the test (e.g. repeated types in the pattern).

https://mlir.llvm.org/getting_started/TestingGuide/

ftynse: Please don't pattern-match SSA names, and generally prefer avoiding non-essential parts of the…
// CHECK: [[VAL_12:%.*]] = constant 4 : index
// CHECK: [[VAL_13:%.*]] = muli [[VAL_4]], [[VAL_11]] : index
// CHECK: [[VAL_14:%.*]] = muli [[VAL_5]], [[VAL_12]] : index
// CHECK: loop.parallel ([[VAL_15:%.*]], [[VAL_16:%.*]]) = ([[VAL_0]], [[VAL_1]]) to ([[VAL_2]], [[VAL_3]]) step ([[VAL_13]], [[VAL_14]]) {
// CHECK: [[VAL_17:%.*]] = affine.min #map0([[VAL_11]], [[VAL_2]], [[VAL_15]])
// CHECK: [[VAL_18:%.*]] = affine.min #map0([[VAL_12]], [[VAL_3]], [[VAL_16]])
// CHECK: loop.parallel ([[VAL_19:%.*]], [[VAL_20:%.*]]) = ([[VAL_10]], [[VAL_10]]) to ([[VAL_17]], [[VAL_18]]) step ([[VAL_4]], [[VAL_5]]) {
// CHECK: [[VAL_21:%.*]] = load [[VAL_7]]{{\[}}[[VAL_19]], [[VAL_20]]] : memref<?x?xf32>
// CHECK: [[VAL_22:%.*]] = load [[VAL_8]]{{\[}}[[VAL_19]], [[VAL_20]]] : memref<?x?xf32>
// CHECK: [[VAL_23:%.*]] = addf [[VAL_21]], [[VAL_22]] : f32
// CHECK: store [[VAL_23]], [[VAL_9]]{{\[}}[[VAL_19]], [[VAL_20]]] : memref<?x?xf32>
// CHECK: }
// CHECK: }
// CHECK: return
// -----
func @tile_nested_innermost() {
dcaballeUnsubmitted
Not Done
ReplyInline Actions

Please add // ----- between tests for split-input-file to work

dcaballe: Please add `// -----` between tests for split-input-file to work
%c2 = constant 2 : index
%c0 = constant 0 : index
%c1 = constant 1 : index
loop.parallel (%i, %j) = (%c0, %c0) to (%c2, %c2) step (%c1, %c1) {
loop.parallel (%k, %l) = (%c0, %c0) to (%c2, %c2) step (%c1, %c1) {
}
}
loop.parallel (%i, %j) = (%c0, %c0) to (%c2, %c2) step (%c1, %c1) {
}
return
}
// CHECK-LABEL: func @tile_nested_innermost() {
// CHECK: [[VAL_24:%.*]] = constant 2 : index
// CHECK: [[VAL_25:%.*]] = constant 0 : index
// CHECK: [[VAL_26:%.*]] = constant 1 : index
// CHECK: loop.parallel ([[VAL_27:%.*]], [[VAL_28:%.*]]) = ([[VAL_25]], [[VAL_25]]) to ([[VAL_24]], [[VAL_24]]) step ([[VAL_26]], [[VAL_26]]) {
// CHECK: [[VAL_29:%.*]] = constant 0 : index
// CHECK: [[VAL_30:%.*]] = constant 1 : index
// CHECK: [[VAL_31:%.*]] = constant 4 : index
// CHECK: [[VAL_32:%.*]] = muli [[VAL_26]], [[VAL_30]] : index
// CHECK: [[VAL_33:%.*]] = muli [[VAL_26]], [[VAL_31]] : index
// CHECK: loop.parallel ([[VAL_34:%.*]], [[VAL_35:%.*]]) = ([[VAL_25]], [[VAL_25]]) to ([[VAL_24]], [[VAL_24]]) step ([[VAL_32]], [[VAL_33]]) {
// CHECK: [[VAL_36:%.*]] = affine.min #map0([[VAL_30]], [[VAL_24]], [[VAL_34]])
// CHECK: [[VAL_37:%.*]] = affine.min #map0([[VAL_31]], [[VAL_24]], [[VAL_35]])
// CHECK: loop.parallel ([[VAL_38:%.*]], [[VAL_39:%.*]]) = ([[VAL_29]], [[VAL_29]]) to ([[VAL_36]], [[VAL_37]]) step ([[VAL_26]], [[VAL_26]]) {
// CHECK: }
// CHECK: }
// CHECK: }
// CHECK: [[VAL_40:%.*]] = constant 0 : index
// CHECK: [[VAL_41:%.*]] = constant 1 : index
// CHECK: [[VAL_42:%.*]] = constant 4 : index
// CHECK: [[VAL_43:%.*]] = muli [[VAL_26]], [[VAL_41]] : index
// CHECK: [[VAL_44:%.*]] = muli [[VAL_26]], [[VAL_42]] : index
// CHECK: loop.parallel ([[VAL_45:%.*]], [[VAL_46:%.*]]) = ([[VAL_25]], [[VAL_25]]) to ([[VAL_24]], [[VAL_24]]) step ([[VAL_43]], [[VAL_44]]) {
// CHECK: [[VAL_47:%.*]] = affine.min #map0([[VAL_41]], [[VAL_24]], [[VAL_45]])
// CHECK: [[VAL_48:%.*]] = affine.min #map0([[VAL_42]], [[VAL_24]], [[VAL_46]])
// CHECK: loop.parallel ([[VAL_49:%.*]], [[VAL_50:%.*]]) = ([[VAL_40]], [[VAL_40]]) to ([[VAL_47]], [[VAL_48]]) step ([[VAL_26]], [[VAL_26]]) {
// CHECK: }
// CHECK: }
// CHECK: return
// CHECK: }

mlir/test/Dialect/SPIRV/control-flow-ops.mlir

mlir/test/Examples/Toy/Ch2/codegen.toy

mlir/test/Examples/Toy/Ch2/scalar.toy

mlir/test/Examples/Toy/Ch3/codegen.toy

mlir/test/Examples/Toy/Ch3/scalar.toy

mlir/test/Examples/Toy/Ch4/codegen.toy

mlir/test/Examples/Toy/Ch4/scalar.toy

mlir/test/Examples/Toy/Ch4/shape_inference.mlir

mlir/test/Examples/Toy/Ch5/affine-lowering.mlir

mlir/test/Examples/Toy/Ch5/codegen.toy

mlir/test/Examples/Toy/Ch5/scalar.toy

mlir/test/Examples/Toy/Ch5/shape_inference.mlir

mlir/test/Examples/Toy/Ch6/affine-lowering.mlir

mlir/test/Examples/Toy/Ch6/codegen.toy

mlir/test/Examples/Toy/Ch6/llvm-lowering.mlir

mlir/test/Examples/Toy/Ch6/scalar.toy

mlir/test/Examples/Toy/Ch6/shape_inference.mlir

mlir/test/Examples/Toy/Ch7/affine-lowering.mlir

mlir/test/Examples/Toy/Ch7/codegen.toy

mlir/test/Examples/Toy/Ch7/llvm-lowering.mlir

mlir/test/Examples/Toy/Ch7/scalar.toy

mlir/test/Examples/Toy/Ch7/shape_inference.mlir

mlir/test/Examples/Toy/Ch7/struct-codegen.toy

mlir/test/Examples/Toy/Ch7/struct-opt.mlir

mlir/test/IR/invalid-ops.mlir

mlir/test/IR/invalid.mlir

mlir/test/lib/TestDialect/TestOps.td

mlir/test/mlir-tblgen/op-format-spec.td

mlir/test/mlir-tblgen/op-format.mlir

mlir/tools/mlir-tblgen/OpDefinitionsGen.cpp

mlir/tools/mlir-tblgen/OpFormatGen.cpp

mlir/tools/mlir-vulkan-runner/VulkanRuntime.cpp