[PassBuilder] Add ThinOrFullLTOPhase to early simplication EP call backs
#114547
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
llvmbot
added
clang
Contributor
Author
This stack of pull requests is managed by Graphite. Learn more about stacking. |
Member
|
@llvm/pr-subscribers-clang-codegen @llvm/pr-subscribers-backend-amdgpu Author: Shilei Tian (shiltian) ChangesThe early simplication pipeline is used in non-LTO and (Thin/Full)LTO pre-link Full diff: https://github.com/llvm/llvm-project/pull/114547.diff 7 Files Affected:
diff --git a/clang/lib/CodeGen/BackendUtil.cpp b/clang/lib/CodeGen/BackendUtil.cpp
index ae33554a66b6b5..76cdf64b3c5a80 100644
--- a/clang/lib/CodeGen/BackendUtil.cpp
+++ b/clang/lib/CodeGen/BackendUtil.cpp
@@ -993,7 +993,7 @@ void EmitAssemblyHelper::RunOptimizationPipeline(
createModuleToFunctionPassAdaptor(ObjCARCExpandPass()));
});
PB.registerPipelineEarlySimplificationEPCallback(
- [](ModulePassManager &MPM, OptimizationLevel Level) {
+ [](ModulePassManager &MPM, OptimizationLevel Level, bool) {
if (Level != OptimizationLevel::O0)
MPM.addPass(ObjCARCAPElimPass());
});
diff --git a/llvm/include/llvm/Passes/PassBuilder.h b/llvm/include/llvm/Passes/PassBuilder.h
index e6ced0cccb9b3c..530db62bce0efa 100644
--- a/llvm/include/llvm/Passes/PassBuilder.h
+++ b/llvm/include/llvm/Passes/PassBuilder.h
@@ -478,7 +478,8 @@ class PassBuilder {
/// This extension point allows adding optimization right after passes that do
/// basic simplification of the input IR.
void registerPipelineEarlySimplificationEPCallback(
- const std::function<void(ModulePassManager &, OptimizationLevel)> &C) {
+ const std::function<void(ModulePassManager &, OptimizationLevel, bool)>
+ &C) {
PipelineEarlySimplificationEPCallbacks.push_back(C);
}
@@ -637,7 +638,8 @@ class PassBuilder {
void invokePipelineStartEPCallbacks(ModulePassManager &MPM,
OptimizationLevel Level);
void invokePipelineEarlySimplificationEPCallbacks(ModulePassManager &MPM,
- OptimizationLevel Level);
+ OptimizationLevel Level,
+ bool LTOPreLink = false);
static bool checkParametrizedPassName(StringRef Name, StringRef PassName) {
if (!Name.consume_front(PassName))
@@ -762,7 +764,8 @@ class PassBuilder {
FullLinkTimeOptimizationLastEPCallbacks;
SmallVector<std::function<void(ModulePassManager &, OptimizationLevel)>, 2>
PipelineStartEPCallbacks;
- SmallVector<std::function<void(ModulePassManager &, OptimizationLevel)>, 2>
+ SmallVector<std::function<void(ModulePassManager &, OptimizationLevel, bool)>,
+ 2>
PipelineEarlySimplificationEPCallbacks;
SmallVector<std::function<void(ModuleAnalysisManager &)>, 2>
diff --git a/llvm/lib/Passes/PassBuilderPipelines.cpp b/llvm/lib/Passes/PassBuilderPipelines.cpp
index 3f28dd39911f79..8c76e0c4955959 100644
--- a/llvm/lib/Passes/PassBuilderPipelines.cpp
+++ b/llvm/lib/Passes/PassBuilderPipelines.cpp
@@ -384,9 +384,9 @@ void PassBuilder::invokePipelineStartEPCallbacks(ModulePassManager &MPM,
C(MPM, Level);
}
void PassBuilder::invokePipelineEarlySimplificationEPCallbacks(
- ModulePassManager &MPM, OptimizationLevel Level) {
+ ModulePassManager &MPM, OptimizationLevel Level, bool LTOPreLink) {
for (auto &C : PipelineEarlySimplificationEPCallbacks)
- C(MPM, Level);
+ C(MPM, Level, LTOPreLink);
}
// Helper to add AnnotationRemarksPass.
@@ -1140,7 +1140,10 @@ PassBuilder::buildModuleSimplificationPipeline(OptimizationLevel Level,
MPM.addPass(LowerTypeTestsPass(nullptr, nullptr,
lowertypetests::DropTestKind::Assume));
- invokePipelineEarlySimplificationEPCallbacks(MPM, Level);
+ bool LTOPreLink = (Phase == ThinOrFullLTOPhase::FullLTOPreLink) ||
+ (Phase == ThinOrFullLTOPhase::ThinLTOPreLink);
+
+ invokePipelineEarlySimplificationEPCallbacks(MPM, Level, LTOPreLink);
// Interprocedural constant propagation now that basic cleanup has occurred
// and prior to optimizing globals.
@@ -2155,7 +2158,7 @@ ModulePassManager PassBuilder::buildO0DefaultPipeline(OptimizationLevel Level,
if (PGOOpt && PGOOpt->DebugInfoForProfiling)
MPM.addPass(createModuleToFunctionPassAdaptor(AddDiscriminatorsPass()));
- invokePipelineEarlySimplificationEPCallbacks(MPM, Level);
+ invokePipelineEarlySimplificationEPCallbacks(MPM, Level, LTOPreLink);
// Build a minimal pipeline based on the semantics required by LLVM,
// which is just that always inlining occurs. Further, disable generating
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index d93ec34a703d3d..1d295804fe3d2a 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -747,7 +747,7 @@ void AMDGPUTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
});
PB.registerPipelineEarlySimplificationEPCallback(
- [](ModulePassManager &PM, OptimizationLevel Level) {
+ [](ModulePassManager &PM, OptimizationLevel Level, bool LTOPreLink) {
PM.addPass(AMDGPUPrintfRuntimeBindingPass());
if (Level == OptimizationLevel::O0)
@@ -755,7 +755,8 @@ void AMDGPUTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
PM.addPass(AMDGPUUnifyMetadataPass());
- if (InternalizeSymbols) {
+ // We don't want to run internalization at per-module stage.
+ if (InternalizeSymbols && !LTOPreLink) {
PM.addPass(InternalizePass(mustPreserveGV));
PM.addPass(GlobalDCEPass());
}
@@ -823,8 +824,15 @@ void AMDGPUTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
PM.addPass(AMDGPUSwLowerLDSPass(*this));
if (EnableLowerModuleLDS)
PM.addPass(AMDGPULowerModuleLDSPass(*this));
- if (EnableAMDGPUAttributor && Level != OptimizationLevel::O0)
- PM.addPass(AMDGPUAttributorPass(*this));
+ if (Level != OptimizationLevel::O0) {
+ if (EnableAMDGPUAttributor)
+ PM.addPass(AMDGPUAttributorPass(*this));
+ // Do we really need internalization in LTO?
+ if (InternalizeSymbols) {
+ PM.addPass(InternalizePass(mustPreserveGV));
+ PM.addPass(GlobalDCEPass());
+ }
+ }
});
PB.registerRegClassFilterParsingCallback(
diff --git a/llvm/lib/Target/BPF/BPFTargetMachine.cpp b/llvm/lib/Target/BPF/BPFTargetMachine.cpp
index 7d91fa8bb824cf..6775930c8fb2dd 100644
--- a/llvm/lib/Target/BPF/BPFTargetMachine.cpp
+++ b/llvm/lib/Target/BPF/BPFTargetMachine.cpp
@@ -138,7 +138,7 @@ void BPFTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
FPM.addPass(BPFPreserveStaticOffsetPass(false));
});
PB.registerPipelineEarlySimplificationEPCallback(
- [=](ModulePassManager &MPM, OptimizationLevel) {
+ [=](ModulePassManager &MPM, OptimizationLevel, bool) {
MPM.addPass(BPFAdjustOptPass());
});
}
diff --git a/llvm/test/CodeGen/AMDGPU/print-pipeline-passes.ll b/llvm/test/CodeGen/AMDGPU/print-pipeline-passes.ll
index b9eda0c1cd3bb6..792e83095efaa7 100644
--- a/llvm/test/CodeGen/AMDGPU/print-pipeline-passes.ll
+++ b/llvm/test/CodeGen/AMDGPU/print-pipeline-passes.ll
@@ -3,9 +3,17 @@
; RUN: opt -mtriple=amdgcn--amdhsa -S -passes="lto<O2>" -print-pipeline-passes %s -o - | FileCheck %s
; RUN: opt -mtriple=amdgcn--amdhsa -S -passes="lto<O3>" -print-pipeline-passes %s -o - | FileCheck %s
+; RUN: opt -mtriple=amdgcn--amdhsa -S -passes="lto-pre-link<O0>" -print-pipeline-passes %s -o - | FileCheck --check-prefix=PRE %s
+; RUN: opt -mtriple=amdgcn--amdhsa -S -passes="lto-pre-link<O1>" -print-pipeline-passes %s -o - | FileCheck --check-prefix=PRE %s
+; RUN: opt -mtriple=amdgcn--amdhsa -S -passes="lto-pre-link<O2>" -print-pipeline-passes %s -o - | FileCheck --check-prefix=PRE %s
+; RUN: opt -mtriple=amdgcn--amdhsa -S -passes="lto-pre-link<O3>" -print-pipeline-passes %s -o - | FileCheck --check-prefix=PRE %s
+
+
; CHECK: amdgpu-attributor
; O0-NOT: amdgpu-attributor
+; PRE-NOT: internalize
+
define amdgpu_kernel void @kernel() {
entry:
ret void
diff --git a/llvm/tools/opt/NewPMDriver.cpp b/llvm/tools/opt/NewPMDriver.cpp
index 9a477193a29365..ccf271356b64c5 100644
--- a/llvm/tools/opt/NewPMDriver.cpp
+++ b/llvm/tools/opt/NewPMDriver.cpp
@@ -294,7 +294,7 @@ static void registerEPCallbacks(PassBuilder &PB) {
if (tryParsePipelineText<ModulePassManager>(
PB, PipelineEarlySimplificationEPPipeline))
PB.registerPipelineEarlySimplificationEPCallback(
- [&PB](ModulePassManager &PM, OptimizationLevel) {
+ [&PB](ModulePassManager &PM, OptimizationLevel, bool) {
ExitOnError Err("Unable to parse EarlySimplification pipeline: ");
Err(PB.parsePassPipeline(PM, PipelineEarlySimplificationEPPipeline));
});
|
nikic
reviewed
Nov 1, 2024
shiltian
changed the title
LTOPreLink to early simplication EP call backsThinOrFullLTOPhase to early simplication EP call backs
shiltian
changed the title
ThinOrFullLTOPhase to early simplication EP call backsThinOrFullLTOPhase to early simplication EP call backs
shiltian
force-pushed
the
users/shiltian/lto-pre-link-bool
branch
from
56ab85b to
c9be622
Compare
shiltian
force-pushed
the
users/shiltian/lto-pre-link-bool
branch
2 times, most recently
from
8ae74a4 to
e753c4f
Compare
shiltian
force-pushed
the
users/shiltian/bool-to-phase
branch
from
c271a32 to
8f36cd9
Compare
shiltian
force-pushed
the
users/shiltian/lto-pre-link-bool
branch
from
e753c4f to
912283a
Compare
shiltian
commented
Nov 1, 2024
shiltian
force-pushed
the
users/shiltian/lto-pre-link-bool
branch
from
912283a to
56d807b
Compare
The early simplication pipeline is used in non-LTO and (Thin/Full)LTO pre-link stage. There are some passes that we want them in non-LTO mode, but not at LTO pre-link stage. The control is missing currently. This PR adds the support. To demonstrate the use, we only enable the internalization pass in non-LTO mode for AMDGPU because having it run in pre-link stage causes some issues.
shiltian
force-pushed
the
users/shiltian/lto-pre-link-bool
branch
from
56d807b to
3da52a0
Compare
arsenm
approved these changes
Nov 4, 2024
PhilippRados
pushed a commit
to PhilippRados/llvm-project
that referenced
this pull request
Nov 6, 2024
…backs (llvm#114547) The early simplication pipeline is used in non-LTO and (Thin/Full)LTO pre-link stage. There are some passes that we want them in non-LTO mode, but not at LTO pre-link stage. The control is missing currently. This PR adds the support. To demonstrate the use, we only enable the internalization pass in non-LTO mode for AMDGPU because having it run in pre-link stage causes some issues.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The early simplication pipeline is used in non-LTO and (Thin/Full)LTO pre-link
stage. There are some passes that we want them in non-LTO mode, but not at LTO
pre-link stage. The control is missing currently. This PR adds the support. To
demonstrate the use, we only enable the internalization pass in non-LTO mode for
AMDGPU because having it run in pre-link stage causes some issues.