This is an archive of the discontinued LLVM Phabricator instance.

Differential D16777

AMDGPU: Skip promote alloca with no optimizations
ClosedPublic

Authored by arsenm on Feb 1 2016, 9:56 AM.

Details

Reviewers
• tstellarAMD

Diff Detail

Event Timeline

arsenm updated this revision to Diff 46554.Feb 1 2016, 9:56 AM
arsenm retitled this revision from to AMDGPU: Skip promote alloca with no optimizations.
arsenm updated this object.
arsenm added a reviewer: • tstellarAMD.
arsenm added a subscriber: llvm-commits.
Herald added a subscriber: arsenm. · View Herald TranscriptFeb 1 2016, 9:56 AM
• tstellarAMD accepted this revision.Feb 2 2016, 8:42 AM
• tstellarAMD edited edge metadata.

LGTM.

This revision is now accepted and ready to land.Feb 2 2016, 8:42 AM
arsenm closed this revision.Feb 2 2016, 11:39 AM

r259551

Revision Contents

PathSize
lib/
Target/
AMDGPU/
2 lines
2 lines
test/
CodeGen/
AMDGPU/
38 lines

Diff 46554

lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp

Show First 20 Lines • Show All 92 Lines • ▼ Show 20 Linesbool AMDGPUPromoteAlloca::doInitialization(Module &M) {
IsAMDGCN = TT.getArch() == Triple::amdgcn; IsAMDGCN = TT.getArch() == Triple::amdgcn;
IsAMDHSA = TT.getOS() == Triple::AMDHSA; IsAMDHSA = TT.getOS() == Triple::AMDHSA;
return false; return false;
} }
bool AMDGPUPromoteAlloca::runOnFunction(Function &F) { bool AMDGPUPromoteAlloca::runOnFunction(Function &F) {
if (!TM) if (!TM || F.hasFnAttribute(Attribute::OptimizeNone))
return false; return false;
FunctionType *FTy = F.getFunctionType(); FunctionType *FTy = F.getFunctionType();
// If the function has any arguments in the local address space, then it's // If the function has any arguments in the local address space, then it's
// possible these arguments require the entire local memory space, so // possible these arguments require the entire local memory space, so
// we cannot use local memory in the pass. // we cannot use local memory in the pass.
for (Type *ParamTy : FTy->params()) { for (Type *ParamTy : FTy->params()) {
▲ Show 20 Lines • Show All 466 Lines • Show Last 20 Lines

lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

Show First 20 Lines • Show All 223 Lines • ▼ Show 20 Linesvoid AMDGPUPassConfig::addIRPasses() {
addPass(createAMDGPUOpenCLImageTypeLoweringPass()); addPass(createAMDGPUOpenCLImageTypeLoweringPass());
TargetPassConfig::addIRPasses(); TargetPassConfig::addIRPasses();
} }
void AMDGPUPassConfig::addCodeGenPrepare() { void AMDGPUPassConfig::addCodeGenPrepare() {
const AMDGPUTargetMachine &TM = getAMDGPUTargetMachine(); const AMDGPUTargetMachine &TM = getAMDGPUTargetMachine();
const AMDGPUSubtarget &ST = *TM.getSubtargetImpl(); const AMDGPUSubtarget &ST = *TM.getSubtargetImpl();
if (ST.isPromoteAllocaEnabled()) { if (TM.getOptLevel() > CodeGenOpt::None && ST.isPromoteAllocaEnabled()) {
addPass(createAMDGPUPromoteAlloca(&TM)); addPass(createAMDGPUPromoteAlloca(&TM));
addPass(createSROAPass()); addPass(createSROAPass());
} }
TargetPassConfig::addCodeGenPrepare(); TargetPassConfig::addCodeGenPrepare();
} }
bool bool
AMDGPUPassConfig::addPreISel() { AMDGPUPassConfig::addPreISel() {
▲ Show 20 Lines • Show All 129 Lines • Show Last 20 Lines

test/CodeGen/AMDGPU/promote-alloca-no-opts.ll

  • This file was added.
; RUN: llc -O0 -march=amdgcn -mtriple=amdgcn-unknown-amdhsa -mattr=+promote-alloca < %s | FileCheck -check-prefix=NOOPTS -check-prefix=ALL %s
; RUN: llc -O1 -march=amdgcn -mtriple=amdgcn-unknown-amdhsa -mattr=+promote-alloca < %s | FileCheck -check-prefix=OPTS -check-prefix=ALL %s
; ALL-LABEL: {{^}}promote_alloca_i32_array_array:
; NOOPTS: workgroup_group_segment_byte_size = 0{{$}}
; NOOPTS-NOT ds_write
; OPTS: ds_write
define void @promote_alloca_i32_array_array(i32 addrspace(1)* %out, i32 %index) #0 {
entry:
%alloca = alloca [2 x [2 x i32]]
%gep0 = getelementptr inbounds [2 x [2 x i32]], [2 x [2 x i32]]* %alloca, i32 0, i32 0, i32 0
%gep1 = getelementptr inbounds [2 x [2 x i32]], [2 x [2 x i32]]* %alloca, i32 0, i32 0, i32 1
store i32 0, i32* %gep0
store i32 1, i32* %gep1
%gep2 = getelementptr inbounds [2 x [2 x i32]], [2 x [2 x i32]]* %alloca, i32 0, i32 0, i32 %index
%load = load i32, i32* %gep2
store i32 %load, i32 addrspace(1)* %out
ret void
}
; ALL-LABEL: {{^}}optnone_promote_alloca_i32_array_array:
; ALL: workgroup_group_segment_byte_size = 0{{$}}
; ALL-NOT ds_write
define void @optnone_promote_alloca_i32_array_array(i32 addrspace(1)* %out, i32 %index) #1 {
entry:
%alloca = alloca [2 x [2 x i32]]
%gep0 = getelementptr inbounds [2 x [2 x i32]], [2 x [2 x i32]]* %alloca, i32 0, i32 0, i32 0
%gep1 = getelementptr inbounds [2 x [2 x i32]], [2 x [2 x i32]]* %alloca, i32 0, i32 0, i32 1
store i32 0, i32* %gep0
store i32 1, i32* %gep1
%gep2 = getelementptr inbounds [2 x [2 x i32]], [2 x [2 x i32]]* %alloca, i32 0, i32 0, i32 %index
%load = load i32, i32* %gep2
store i32 %load, i32 addrspace(1)* %out
ret void
}
attributes #0 = { nounwind }
attributes #1 = { nounwind optnone noinline }