AMDGPU: Preserve alignments on new created globals

[arsenm]

Summary

Also switch to internal linkage, and include the name of the function in the name.

[• tstellarAMD]

LGTM. Is the alignment on global variables actually used for things? I thought it was the alignment on loads/stores that mattered.

[arsenm]

In D16860#343776, @tstellarAMD wrote:

LGTM. Is the alignment on global variables actually used for things? I thought it was the alignment on loads/stores that mattered.

The alignment of the load/store must be >= the alignment of the actual address of the indexed value

[arsenm]

r259911

lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp

void AMDGPUPromoteAlloca::visitAlloca(AllocaInst &I) {
   if (!collectUsesWithPtrTypes(&I, WorkList)) {
     DEBUG(dbgs() << " Do not know how to convert all uses\n");
     return;
   }
 
   DEBUG(dbgs() << "Promoting alloca to local memory\n");
   LocalMemAvailable -= AllocaSize;
 
+  Function *F = I.getParent()->getParent();
+
   Type *GVTy = ArrayType::get(I.getAllocatedType(), 256);
   GlobalVariable *GV = new GlobalVariable(
-      *Mod, GVTy, false, GlobalValue::ExternalLinkage, 0, I.getName(), 0,
-      GlobalVariable::NotThreadLocal, AMDGPUAS::LOCAL_ADDRESS);
+      *Mod, GVTy, false, GlobalValue::InternalLinkage,
+      UndefValue::get(GVTy),
+      Twine(F->getName()) + Twine('.') + I.getName(),
+      nullptr,
+      GlobalVariable::NotThreadLocal,
+      AMDGPUAS::LOCAL_ADDRESS);
+  GV->setUnnamedAddr(true);
+  GV->setAlignment(I.getAlignment());
 
   Value *TCntY, *TCntZ;
 
   std::tie(TCntY, TCntZ) = getLocalSizeYZ(Builder);
   Value *TIdX = getWorkitemID(Builder, 0);
   Value *TIdY = getWorkitemID(Builder, 1);
   Value *TIdZ = getWorkitemID(Builder, 2);
 

test/CodeGen/AMDGPU/private-memory.ll

 ; RUN: llc -show-mc-encoding -mattr=+promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=SI < %s | FileCheck %s -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC
 ; RUN: llc -show-mc-encoding -mattr=+promote-alloca -verify-machineinstrs -mtriple=amdgcn--amdhsa -mcpu=kaveri < %s | FileCheck %s -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC -check-prefix=HSA-PROMOTE
 ; RUN: llc -show-mc-encoding -mattr=-promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=SI < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC
 ; RUN: llc -show-mc-encoding -mattr=-promote-alloca -verify-machineinstrs -mtriple=amdgcn-amdhsa -mcpu=kaveri < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC -check-prefix=HSA-ALLOCA
 ; RUN: llc -show-mc-encoding -mattr=+promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=tonga < %s | FileCheck %s -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC
 ; RUN: llc -show-mc-encoding -mattr=-promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=tonga < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC
 
 ; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri -amdgpu-promote-alloca < %s | FileCheck -check-prefix=HSAOPT -check-prefix=OPT %s
 ; RUN: opt -S -mtriple=amdgcn-unknown-unknown -mcpu=kaveri -amdgpu-promote-alloca < %s | FileCheck -check-prefix=NOHSAOPT -check-prefix=OPT %s
 
 declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
 
+; HSAOPT: @mova_same_clause.stack = internal unnamed_addr addrspace(3) global [256 x [5 x i32]] undef, align 4
+; HSAOPT: @high_alignment.stack = internal unnamed_addr addrspace(3) global [256 x [8 x i32]] undef, align 16
+
+
 ; FUNC-LABEL: {{^}}mova_same_clause:
 ; OPT-LABEL: @mova_same_clause(
 
 ; R600: LDS_WRITE
 ; R600: LDS_WRITE
 ; R600: LDS_READ
 ; R600: LDS_READ
 

 ; HSAOPT: [[WORKITEM_ID_Z:%[0-9]+]] = call i32 @llvm.amdgcn.workitem.id.z(), !range !1
 
 ; HSAOPT: [[Y_SIZE_X_Z_SIZE:%[0-9]+]] = mul nuw nsw i32 [[EXTRACTY]], [[LDZU]]
 ; HSAOPT: [[YZ_X_XID:%[0-9]+]] = mul i32 [[Y_SIZE_X_Z_SIZE]], [[WORKITEM_ID_X]]
 ; HSAOPT: [[Y_X_Z_SIZE:%[0-9]+]] = mul nuw nsw i32 [[WORKITEM_ID_Y]], [[LDZU]]
 ; HSAOPT: [[ADD_YZ_X_X_YZ_SIZE:%[0-9]+]] = add i32 [[YZ_X_XID]], [[Y_X_Z_SIZE]]
 ; HSAOPT: [[ADD_ZID:%[0-9]+]] = add i32 [[ADD_YZ_X_X_YZ_SIZE]], [[WORKITEM_ID_Z]]
 
-; HSAOPT: [[LOCAL_GEP:%[0-9]+]] = getelementptr inbounds [256 x [5 x i32]], [256 x [5 x i32]] addrspace(3)* @stack, i32 0, i32 [[ADD_ZID]]
+; HSAOPT: [[LOCAL_GEP:%[0-9]+]] = getelementptr inbounds [256 x [5 x i32]], [256 x [5 x i32]] addrspace(3)* @mova_same_clause.stack, i32 0, i32 [[ADD_ZID]]
 ; HSAOPT: %arrayidx1 = getelementptr inbounds [5 x i32], [5 x i32] addrspace(3)* [[LOCAL_GEP]], i32 0, i32 {{%[0-9]+}}
 ; HSAOPT: %arrayidx3 = getelementptr inbounds [5 x i32], [5 x i32] addrspace(3)* [[LOCAL_GEP]], i32 0, i32 {{%[0-9]+}}
 ; HSAOPT: %arrayidx10 = getelementptr inbounds [5 x i32], [5 x i32] addrspace(3)* [[LOCAL_GEP]], i32 0, i32 0
 ; HSAOPT: %arrayidx12 = getelementptr inbounds [5 x i32], [5 x i32] addrspace(3)* [[LOCAL_GEP]], i32 0, i32 1
 
 
 ; NOHSAOPT: call i32 @llvm.r600.read.local.size.y(), !range !0
 ; NOHSAOPT: call i32 @llvm.r600.read.local.size.z(), !range !0
entry:
   store i32 %2, i32 addrspace(1)* %out, align 4
   %arrayidx12 = getelementptr inbounds [5 x i32], [5 x i32]* %stack, i32 0, i32 1
   %3 = load i32, i32* %arrayidx12
   %arrayidx13 = getelementptr inbounds i32, i32 addrspace(1)* %out, i32 1
   store i32 %3, i32 addrspace(1)* %arrayidx13
   ret void
 }
 
+; OPT-LABEL: @high_alignment(
+; OPT: getelementptr inbounds [256 x [8 x i32]], [256 x [8 x i32]] addrspace(3)* @high_alignment.stack, i32 0, i32 %{{[0-9]+}}
+define void @high_alignment(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) {
+entry:
+  %stack = alloca [8 x i32], align 16
+  %0 = load i32, i32 addrspace(1)* %in, align 4
+  %arrayidx1 = getelementptr inbounds [8 x i32], [8 x i32]* %stack, i32 0, i32 %0
+  store i32 4, i32* %arrayidx1, align 4
+  %arrayidx2 = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 1
+  %1 = load i32, i32 addrspace(1)* %arrayidx2, align 4
+  %arrayidx3 = getelementptr inbounds [8 x i32], [8 x i32]* %stack, i32 0, i32 %1
+  store i32 5, i32* %arrayidx3, align 4
+  %arrayidx10 = getelementptr inbounds [8 x i32], [8 x i32]* %stack, i32 0, i32 0
+  %2 = load i32, i32* %arrayidx10, align 4
+  store i32 %2, i32 addrspace(1)* %out, align 4
+  %arrayidx12 = getelementptr inbounds [8 x i32], [8 x i32]* %stack, i32 0, i32 1
+  %3 = load i32, i32* %arrayidx12
+  %arrayidx13 = getelementptr inbounds i32, i32 addrspace(1)* %out, i32 1
+  store i32 %3, i32 addrspace(1)* %arrayidx13
+  ret void
+}
+
 ; FUNC-LABEL: {{^}}no_replace_inbounds_gep:
 ; OPT-LABEL: @no_replace_inbounds_gep(
 ; OPT: alloca [5 x i32]
 
 ; SI-NOT: ds_write
 define void @no_replace_inbounds_gep(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) {
 entry:
   %stack = alloca [5 x i32], align 4

test/CodeGen/AMDGPU/promote-alloca-mem-intrinsics.ll

 ; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -amdgpu-promote-alloca < %s | FileCheck %s
 
 declare void @llvm.memcpy.p0i8.p1i8.i32(i8* nocapture, i8 addrspace(1)* nocapture, i32, i32, i1) #0
 declare void @llvm.memcpy.p1i8.p0i8.i32(i8 addrspace(1)* nocapture, i8* nocapture, i32, i32, i1) #0
 
 declare void @llvm.memmove.p0i8.p1i8.i32(i8* nocapture, i8 addrspace(1)* nocapture, i32, i32, i1) #0
 declare void @llvm.memmove.p1i8.p0i8.i32(i8 addrspace(1)* nocapture, i8* nocapture, i32, i32, i1) #0
 
 declare void @llvm.memset.p0i8.i32(i8* nocapture, i8, i32, i32, i1) #0
 
 declare i32 @llvm.objectsize.i32.p0i8(i8*, i1) #1
 
 ; CHECK-LABEL: @promote_with_memcpy(
-; CHECK: getelementptr inbounds [256 x [17 x i32]], [256 x [17 x i32]] addrspace(3)* @alloca, i32 0, i32 %{{[0-9]+}}
+; CHECK: getelementptr inbounds [256 x [17 x i32]], [256 x [17 x i32]] addrspace(3)* @promote_with_memcpy.alloca, i32 0, i32 %{{[0-9]+}}
 ; CHECK: call void @llvm.memcpy.p3i8.p1i8.i32(i8 addrspace(3)* %alloca.bc, i8 addrspace(1)* %in.bc, i32 68, i32 4, i1 false)
 ; CHECK: call void @llvm.memcpy.p1i8.p3i8.i32(i8 addrspace(1)* %out.bc, i8 addrspace(3)* %alloca.bc, i32 68, i32 4, i1 false)
 define void @promote_with_memcpy(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
   %alloca = alloca [17 x i32], align 16
   %alloca.bc = bitcast [17 x i32]* %alloca to i8*
   %in.bc = bitcast i32 addrspace(1)* %in to i8 addrspace(1)*
   %out.bc = bitcast i32 addrspace(1)* %out to i8 addrspace(1)*
   call void @llvm.memcpy.p0i8.p1i8.i32(i8* %alloca.bc, i8 addrspace(1)* %in.bc, i32 68, i32 4, i1 false)
   call void @llvm.memcpy.p1i8.p0i8.i32(i8 addrspace(1)* %out.bc, i8* %alloca.bc, i32 68, i32 4, i1 false)
   ret void
 }
 
 ; CHECK-LABEL: @promote_with_memmove(
-; CHECK: getelementptr inbounds [256 x [17 x i32]], [256 x [17 x i32]] addrspace(3)* @alloca.1, i32 0, i32 %{{[0-9]+}}
+; CHECK: getelementptr inbounds [256 x [17 x i32]], [256 x [17 x i32]] addrspace(3)* @promote_with_memmove.alloca, i32 0, i32 %{{[0-9]+}}
 ; CHECK: call void @llvm.memmove.p3i8.p1i8.i32(i8 addrspace(3)* %alloca.bc, i8 addrspace(1)* %in.bc, i32 68, i32 4, i1 false)
 ; CHECK: call void @llvm.memmove.p1i8.p3i8.i32(i8 addrspace(1)* %out.bc, i8 addrspace(3)* %alloca.bc, i32 68, i32 4, i1 false)
 define void @promote_with_memmove(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
   %alloca = alloca [17 x i32], align 16
   %alloca.bc = bitcast [17 x i32]* %alloca to i8*
   %in.bc = bitcast i32 addrspace(1)* %in to i8 addrspace(1)*
   %out.bc = bitcast i32 addrspace(1)* %out to i8 addrspace(1)*
   call void @llvm.memmove.p0i8.p1i8.i32(i8* %alloca.bc, i8 addrspace(1)* %in.bc, i32 68, i32 4, i1 false)
   call void @llvm.memmove.p1i8.p0i8.i32(i8 addrspace(1)* %out.bc, i8* %alloca.bc, i32 68, i32 4, i1 false)
   ret void
 }
 
 ; CHECK-LABEL: @promote_with_memset(
-; CHECK: getelementptr inbounds [256 x [17 x i32]], [256 x [17 x i32]] addrspace(3)* @alloca.2, i32 0, i32 %{{[0-9]+}}
+; CHECK: getelementptr inbounds [256 x [17 x i32]], [256 x [17 x i32]] addrspace(3)* @promote_with_memset.alloca, i32 0, i32 %{{[0-9]+}}
 ; CHECK: call void @llvm.memset.p3i8.i32(i8 addrspace(3)* %alloca.bc, i8 7, i32 68, i32 4, i1 false)
 define void @promote_with_memset(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
   %alloca = alloca [17 x i32], align 16
   %alloca.bc = bitcast [17 x i32]* %alloca to i8*
   %in.bc = bitcast i32 addrspace(1)* %in to i8 addrspace(1)*
   %out.bc = bitcast i32 addrspace(1)* %out to i8 addrspace(1)*
   call void @llvm.memset.p0i8.i32(i8* %alloca.bc, i8 7, i32 68, i32 4, i1 false)
   ret void
 }
 
 ; CHECK-LABEL: @promote_with_objectsize(
-; CHECK: [[PTR:%[0-9]+]] = getelementptr inbounds [256 x [17 x i32]], [256 x [17 x i32]] addrspace(3)* @alloca.3, i32 0, i32 %{{[0-9]+}}
+; CHECK: [[PTR:%[0-9]+]] = getelementptr inbounds [256 x [17 x i32]], [256 x [17 x i32]] addrspace(3)* @promote_with_objectsize.alloca, i32 0, i32 %{{[0-9]+}}
 ; CHECK: call i32 @llvm.objectsize.i32.p3i8(i8 addrspace(3)* %alloca.bc, i1 false)
 define void @promote_with_objectsize(i32 addrspace(1)* %out) #0 {
   %alloca = alloca [17 x i32], align 16
   %alloca.bc = bitcast [17 x i32]* %alloca to i8*
   %size = call i32 @llvm.objectsize.i32.p0i8(i8* %alloca.bc, i1 false)
   store i32 %size, i32 addrspace(1)* %out
   ret void
 }
 
 attributes #0 = { nounwind }
 attributes #1 = { nounwind readnone }