RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-26 12:50:30 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[AMDGPU] Preserve original symbol during attribute propagation

Browse Source 
AMDGPUPropagateAttributes can swap names while cloning a function.
Only do it if original symbol was not externally visible.

Differential Revision: https://reviews.llvm.org/D76789
This commit is contained in:
Stanislav Mekhanoshin 2020-03-25 11:06:34 -07:00
parent 931c0525ee
commit b2f5e79308
3 changed files with 30 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
lib/Target/AMDGPU/AMDGPUPropagateAttributes.cpp
View File

@ -269,17 +269,15 @@ AMDGPUPropagateAttributes::cloneWithFeatures(Function &F,
ValueToValueMapTy dummy; ValueToValueMapTy dummy;
Function *NewF = CloneFunction(&F, dummy); Function *NewF = CloneFunction(&F, dummy);
setFeatures(*NewF, NewFeatures); setFeatures(*NewF, NewFeatures);
NewF->setVisibility(GlobalValue::DefaultVisibility);
NewF->setLinkage(GlobalValue::InternalLinkage);
// Swap names. If that is the only clone it will retain the name of now // Swap names. If that is the only clone it will retain the name of now
// dead value. // dead value. Preserve original name for externally visible functions.
if (F.hasName()) { if (F.hasName() && F.hasLocalLinkage()) {
std::string NewName = std::string(NewF->getName()); std::string NewName = std::string(NewF->getName());
NewF->takeName(&F); NewF->takeName(&F);
F.setName(NewName); F.setName(NewName);
// Name has changed, it does not need an external symbol.
F.setVisibility(GlobalValue::DefaultVisibility);
F.setLinkage(GlobalValue::InternalLinkage);
} }
return NewF; return NewF;

6
lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
View File

@ -437,11 +437,11 @@ void AMDGPUTargetMachine::adjustPassManager(PassManagerBuilder &Builder) {
} }
PM.add(createAMDGPUUnifyMetadataPass()); PM.add(createAMDGPUUnifyMetadataPass());
PM.add(createAMDGPUPrintfRuntimeBinding()); PM.add(createAMDGPUPrintfRuntimeBinding());
PM.add(createAMDGPUPropagateAttributesLatePass(this)); if (Internalize)
if (Internalize) {
PM.add(createInternalizePass(mustPreserveGV)); PM.add(createInternalizePass(mustPreserveGV));
PM.add(createAMDGPUPropagateAttributesLatePass(this));
if (Internalize)
PM.add(createGlobalDCEPass()); PM.add(createGlobalDCEPass());
}
if (EarlyInline) if (EarlyInline)
PM.add(createAMDGPUAlwaysInlinePass(false)); PM.add(createAMDGPUAlwaysInlinePass(false));
}); });

29
test/CodeGen/AMDGPU/propagate-attributes-clone.ll
View File

@ -1,15 +1,23 @@
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -O1 < %s | FileCheck -check-prefix=OPT %s ; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -O1 < %s | FileCheck -check-prefixes=OPT,OPT-EXT %s
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -O1 --amdgpu-internalize-symbols < %s | FileCheck -check-prefixes=OPT,OPT-INT %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefix=LLC %s ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefix=LLC %s
; OPT: declare void @foo4() local_unnamed_addr #0 ; OPT: declare void @foo4() local_unnamed_addr #0
; OPT: define internal fastcc void @foo3.2() unnamed_addr #1 ; OPT: define internal fastcc void @0() unnamed_addr #1
; OPT: define void @foo2() local_unnamed_addr #1 ; OPT-EXT: define void @foo3() local_unnamed_addr #1
; OPT: define internal fastcc void @foo1.1() unnamed_addr #1 ; OPT-INT: define internal fastcc void @foo3.2() unnamed_addr #1
; OPT-EXT: define void @foo2() local_unnamed_addr #1
; OPT-INT: define internal fastcc void @foo2() unnamed_addr #1
; OPT-EXT: define void @foo1() local_unnamed_addr #1
; OPT-INT: define internal fastcc void @foo1.1() unnamed_addr #1
; OPT: define amdgpu_kernel void @kernel1() local_unnamed_addr #2 ; OPT: define amdgpu_kernel void @kernel1() local_unnamed_addr #2
; OPT: define amdgpu_kernel void @kernel2() local_unnamed_addr #3 ; OPT: define amdgpu_kernel void @kernel2() local_unnamed_addr #3
; OPT: define amdgpu_kernel void @kernel3() local_unnamed_addr #3 ; OPT: define amdgpu_kernel void @kernel3() local_unnamed_addr #3
; OPT: define void @foo1() local_unnamed_addr #4 ; OPT-EXT: define internal fastcc void @foo1.1() unnamed_addr #4
; OPT: define void @foo3() local_unnamed_addr #4 ; OPT-INT: define internal fastcc void @foo1() unnamed_addr #4
; OPT: define internal fastcc void @1() unnamed_addr #4
; OPT-EXT: define internal fastcc void @foo3.2() unnamed_addr #4
; OPT-INT: define internal fastcc void @foo3() unnamed_addr #4
; OPT: attributes #0 = { {{.*}} "target-features"="+wavefrontsize64" } ; OPT: attributes #0 = { {{.*}} "target-features"="+wavefrontsize64" }
; OPT: attributes #1 = { {{.*}} "target-features"="{{.*}},-wavefrontsize16,-wavefrontsize32,+wavefrontsize64{{.*}}" } ; OPT: attributes #1 = { {{.*}} "target-features"="{{.*}},-wavefrontsize16,-wavefrontsize32,+wavefrontsize64{{.*}}" }
; OPT: attributes #2 = { {{.*}} "target-features"="+wavefrontsize32" } ; OPT: attributes #2 = { {{.*}} "target-features"="+wavefrontsize32" }
@ -29,6 +37,8 @@
; LLC: foo2@gotpcrel32@hi+4 ; LLC: foo2@gotpcrel32@hi+4
; LLC: foo1@gotpcrel32@lo+4 ; LLC: foo1@gotpcrel32@lo+4
; LLC: foo1@gotpcrel32@hi+4 ; LLC: foo1@gotpcrel32@hi+4
; LLC: __unnamed_1@gotpcrel32@lo+4
; LLC: __unnamed_1@gotpcrel32@hi+4
; LLC: kernel1: ; LLC: kernel1:
; LLC: foo1@gotpcrel32@lo+4 ; LLC: foo1@gotpcrel32@lo+4
; LLC: foo1@gotpcrel32@hi+4 ; LLC: foo1@gotpcrel32@hi+4
@ -41,6 +51,12 @@
declare void @foo4() #1 declare void @foo4() #1
define void @0() #1 {
entry:
call void asm sideeffect "; sample asm", ""()
ret void
}
define void @foo3() #1 { define void @foo3() #1 {
entry: entry:
call void asm sideeffect "; sample asm", ""() call void asm sideeffect "; sample asm", ""()
@ -60,6 +76,7 @@ entry:
tail call void @foo2() tail call void @foo2()
tail call void @foo2() tail call void @foo2()
tail call void @foo1() tail call void @foo1()
tail call void @0()
ret void ret void
} }