[llvm-branch-commits] [llvm] 9abc457 - [NewPM][AMDGPU] Port amdgpu-simplifylib/amdgpu-usenative
Arthur Eubanks via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Mon Dec 28 10:43:41 PST 2020
Author: Arthur Eubanks
Date: 2020-12-28T10:38:51-08:00
New Revision: 9abc457724bd54014328a6f0b7ed230bacd9f610
URL: https://github.com/llvm/llvm-project/commit/9abc457724bd54014328a6f0b7ed230bacd9f610
DIFF: https://github.com/llvm/llvm-project/commit/9abc457724bd54014328a6f0b7ed230bacd9f610.diff
LOG: [NewPM][AMDGPU] Port amdgpu-simplifylib/amdgpu-usenative
And add them to the pipeline via
AMDGPUTargetMachine::registerPassBuilderCallbacks(), which mirrors
AMDGPUTargetMachine::adjustPassManager().
These passes can't be unconditionally added to PassRegistry.def since
they are only present when the AMDGPU backend is enabled. And there are
no target-specific headers in llvm/include, so parsing these pass names
must occur somewhere in the AMDGPU directory. I decided the best place
was inside the TargetMachine, since the PassBuilder invokes
TargetMachine::registerPassBuilderCallbacks() anyway. If we come up with
a cleaner solution for target-specific passes in the future that's fine,
but there aren't too many target-specific IR passes living in
target-specific directories so it shouldn't be too bad to change in the
future.
Reviewed By: ychen, arsenm
Differential Revision: https://reviews.llvm.org/D93863
Added:
Modified:
llvm/lib/Target/AMDGPU/AMDGPU.h
llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp
llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.h
llvm/lib/Target/AMDGPU/CMakeLists.txt
llvm/test/CodeGen/AMDGPU/simplify-libcalls.ll
llvm/test/CodeGen/AMDGPU/simplify-libcalls2.ll
llvm/utils/gn/secondary/llvm/lib/Target/AMDGPU/BUILD.gn
Removed:
################################################################################
diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.h b/llvm/lib/Target/AMDGPU/AMDGPU.h
index 132036fbcfd0..22d264e2880b 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPU.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPU.h
@@ -10,8 +10,9 @@
#ifndef LLVM_LIB_TARGET_AMDGPU_AMDGPU_H
#define LLVM_LIB_TARGET_AMDGPU_AMDGPU_H
-#include "llvm/IR/IntrinsicsR600.h" // TODO: Sink this.
#include "llvm/IR/IntrinsicsAMDGPU.h" // TODO: Sink this.
+#include "llvm/IR/IntrinsicsR600.h" // TODO: Sink this.
+#include "llvm/IR/PassManager.h"
#include "llvm/Support/CodeGen.h"
namespace llvm {
@@ -75,6 +76,14 @@ ModulePass *createAMDGPUPropagateAttributesLatePass(const TargetMachine *);
FunctionPass *createAMDGPURewriteOutArgumentsPass();
FunctionPass *createSIModeRegisterPass();
+struct AMDGPUSimplifyLibCallsPass : PassInfoMixin<AMDGPUSimplifyLibCallsPass> {
+ PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM);
+};
+
+struct AMDGPUUseNativeCallsPass : PassInfoMixin<AMDGPUUseNativeCallsPass> {
+ PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM);
+};
+
void initializeAMDGPUDAGToDAGISelPass(PassRegistry&);
void initializeAMDGPUMachineCFGStructurizerPass(PassRegistry&);
diff --git a/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp b/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp
index a72f16956e72..eedcb2e1a793 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp
@@ -26,6 +26,7 @@
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Module.h"
+#include "llvm/IR/PassManager.h"
#include "llvm/IR/ValueSymbolTable.h"
#include "llvm/InitializePasses.h"
#include "llvm/Support/Debug.h"
@@ -1750,6 +1751,40 @@ bool AMDGPUSimplifyLibCalls::runOnFunction(Function &F) {
return Changed;
}
+PreservedAnalyses AMDGPUSimplifyLibCallsPass::run(Function &F,
+ FunctionAnalysisManager &AM) {
+ AMDGPULibCalls Simplifier;
+ Simplifier.initNativeFuncs();
+
+ bool Changed = false;
+ auto AA = &AM.getResult<AAManager>(F);
+
+ LLVM_DEBUG(dbgs() << "AMDIC: process function ";
+ F.printAsOperand(dbgs(), false, F.getParent()); dbgs() << '\n';);
+
+ for (auto &BB : F) {
+ for (BasicBlock::iterator I = BB.begin(), E = BB.end(); I != E;) {
+ // Ignore non-calls.
+ CallInst *CI = dyn_cast<CallInst>(I);
+ ++I;
+ // Ignore intrinsics that do not become real instructions.
+ if (!CI || isa<DbgInfoIntrinsic>(CI) || CI->isLifetimeStartOrEnd())
+ continue;
+
+ // Ignore indirect calls.
+ Function *Callee = CI->getCalledFunction();
+ if (Callee == 0)
+ continue;
+
+ LLVM_DEBUG(dbgs() << "AMDIC: try folding " << *CI << "\n";
+ dbgs().flush());
+ if (Simplifier.fold(CI, AA))
+ Changed = true;
+ }
+ }
+ return Changed ? PreservedAnalyses::none() : PreservedAnalyses::all();
+}
+
bool AMDGPUUseNativeCalls::runOnFunction(Function &F) {
if (skipFunction(F) || UseNative.empty())
return false;
@@ -1772,3 +1807,32 @@ bool AMDGPUUseNativeCalls::runOnFunction(Function &F) {
}
return Changed;
}
+
+PreservedAnalyses AMDGPUUseNativeCallsPass::run(Function &F,
+ FunctionAnalysisManager &AM) {
+ if (UseNative.empty())
+ return PreservedAnalyses::all();
+
+ AMDGPULibCalls Simplifier;
+ Simplifier.initNativeFuncs();
+
+ bool Changed = false;
+ for (auto &BB : F) {
+ for (BasicBlock::iterator I = BB.begin(), E = BB.end(); I != E;) {
+ // Ignore non-calls.
+ CallInst *CI = dyn_cast<CallInst>(I);
+ ++I;
+ if (!CI)
+ continue;
+
+ // Ignore indirect calls.
+ Function *Callee = CI->getCalledFunction();
+ if (Callee == 0)
+ continue;
+
+ if (Simplifier.useNative(CI))
+ Changed = true;
+ }
+ }
+ return Changed ? PreservedAnalyses::none() : PreservedAnalyses::all();
+}
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index ccfd62ea3e09..6ea99bdf9206 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -40,8 +40,10 @@
#include "llvm/IR/Attributes.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/LegacyPassManager.h"
+#include "llvm/IR/PassManager.h"
#include "llvm/InitializePasses.h"
#include "llvm/Pass.h"
+#include "llvm/Passes/PassBuilder.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/TargetRegistry.h"
@@ -52,6 +54,7 @@
#include "llvm/Transforms/Scalar.h"
#include "llvm/Transforms/Scalar/GVN.h"
#include "llvm/Transforms/Utils.h"
+#include "llvm/Transforms/Utils/SimplifyLibCalls.h"
#include "llvm/Transforms/Vectorize.h"
#include <memory>
@@ -482,6 +485,33 @@ void AMDGPUTargetMachine::adjustPassManager(PassManagerBuilder &Builder) {
});
}
+void AMDGPUTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB,
+ bool DebugPassManager) {
+ PB.registerPipelineParsingCallback(
+ [](StringRef PassName, FunctionPassManager &PM,
+ ArrayRef<PassBuilder::PipelineElement>) {
+ if (PassName == "amdgpu-simplifylib") {
+ PM.addPass(AMDGPUSimplifyLibCallsPass());
+ return true;
+ }
+ if (PassName == "amdgpu-usenative") {
+ PM.addPass(AMDGPUUseNativeCallsPass());
+ return true;
+ }
+ return false;
+ });
+
+ PB.registerPipelineStartEPCallback([DebugPassManager](
+ ModulePassManager &PM,
+ PassBuilder::OptimizationLevel Level) {
+ FunctionPassManager FPM(DebugPassManager);
+ FPM.addPass(AMDGPUUseNativeCallsPass());
+ if (EnableLibCallSimplify && Level != PassBuilder::OptimizationLevel::O0)
+ FPM.addPass(AMDGPUSimplifyLibCallsPass());
+ PM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM)));
+ });
+}
+
//===----------------------------------------------------------------------===//
// R600 Target Machine (R600 -> Cayman)
//===----------------------------------------------------------------------===//
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.h b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.h
index 56d3237832be..d5fd769912d0 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.h
@@ -56,6 +56,9 @@ class AMDGPUTargetMachine : public LLVMTargetMachine {
void adjustPassManager(PassManagerBuilder &) override;
+ void registerPassBuilderCallbacks(PassBuilder &PB,
+ bool DebugPassManager) override;
+
/// Get the integer value of a null pointer in the given address space.
static int64_t getNullPointerValue(unsigned AddrSpace) {
return (AddrSpace == AMDGPUAS::LOCAL_ADDRESS ||
diff --git a/llvm/lib/Target/AMDGPU/CMakeLists.txt b/llvm/lib/Target/AMDGPU/CMakeLists.txt
index 48846c82e7f7..9b02fb44b0eb 100644
--- a/llvm/lib/Target/AMDGPU/CMakeLists.txt
+++ b/llvm/lib/Target/AMDGPU/CMakeLists.txt
@@ -152,6 +152,7 @@ add_llvm_target(AMDGPUCodeGen
Core
IPO
MC
+ Passes
AMDGPUDesc
AMDGPUInfo
AMDGPUUtils
diff --git a/llvm/test/CodeGen/AMDGPU/simplify-libcalls.ll b/llvm/test/CodeGen/AMDGPU/simplify-libcalls.ll
index 40e7f6e7f8f6..2ae332ffa437 100644
--- a/llvm/test/CodeGen/AMDGPU/simplify-libcalls.ll
+++ b/llvm/test/CodeGen/AMDGPU/simplify-libcalls.ll
@@ -1,6 +1,9 @@
; RUN: opt -S -O1 -mtriple=amdgcn-- -amdgpu-simplify-libcall < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=GCN-POSTLINK %s
; RUN: opt -S -O1 -mtriple=amdgcn-- -amdgpu-simplify-libcall -amdgpu-prelink <%s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=GCN-PRELINK %s
; RUN: opt -S -O1 -mtriple=amdgcn-- -amdgpu-use-native -amdgpu-prelink < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=GCN-NATIVE %s
+; RUN: opt -S -passes='default<O1>' -mtriple=amdgcn-- -amdgpu-simplify-libcall < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=GCN-POSTLINK %s
+; RUN: opt -S -passes='default<O1>' -mtriple=amdgcn-- -amdgpu-simplify-libcall -amdgpu-prelink <%s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=GCN-PRELINK %s
+; RUN: opt -S -passes='default<O1>' -mtriple=amdgcn-- -amdgpu-use-native -amdgpu-prelink < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=GCN-NATIVE %s
; GCN-LABEL: {{^}}define amdgpu_kernel void @test_sincos
; GCN-POSTLINK: call fast float @_Z3sinf(
diff --git a/llvm/test/CodeGen/AMDGPU/simplify-libcalls2.ll b/llvm/test/CodeGen/AMDGPU/simplify-libcalls2.ll
index 16150d0dd7c3..50a34fac05a6 100644
--- a/llvm/test/CodeGen/AMDGPU/simplify-libcalls2.ll
+++ b/llvm/test/CodeGen/AMDGPU/simplify-libcalls2.ll
@@ -1,5 +1,6 @@
; REQUIRES: asserts
; RUN: opt -S -amdgpu-simplifylib -debug-only=amdgpu-simplifylib -mtriple=amdgcn-unknown-amdhsa -disable-output < %s 2>&1 | FileCheck %s
+; RUN: opt -S -passes=amdgpu-simplifylib -debug-only=amdgpu-simplifylib -mtriple=amdgcn-unknown-amdhsa -disable-output < %s 2>&1 | FileCheck %s
; CHECK-NOT: AMDIC: try folding call void @llvm.lifetime.start.p0i8
; CHECK-NOT: AMDIC: try folding call void @llvm.lifetime.end.p0i8
diff --git a/llvm/utils/gn/secondary/llvm/lib/Target/AMDGPU/BUILD.gn b/llvm/utils/gn/secondary/llvm/lib/Target/AMDGPU/BUILD.gn
index b9154a449557..fdc0751231b5 100644
--- a/llvm/utils/gn/secondary/llvm/lib/Target/AMDGPU/BUILD.gn
+++ b/llvm/utils/gn/secondary/llvm/lib/Target/AMDGPU/BUILD.gn
@@ -112,6 +112,7 @@ static_library("LLVMAMDGPUCodeGen") {
"//llvm/lib/CodeGen/SelectionDAG",
"//llvm/lib/IR",
"//llvm/lib/MC",
+ "//llvm/lib/Passes",
"//llvm/lib/Support",
"//llvm/lib/Target",
"//llvm/lib/Transforms/IPO",
More information about the llvm-branch-commits
mailing list