AMDGPU/SI: Emit constant variables in the .hsatext section when targeting HSA

[arsenm]

lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp

128–135

I would split the HSA vs. Non HSA into separate ifs

[• tstellarAMD]

Address review comment.

[arsenm]

LGTM

lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp

static bool isModuleLinkage(const GlobalValue *GV) {
    return true;
   case GlobalValue::ExternalLinkage:
    return false;
   default: llvm_unreachable("unknown linkage type");
   }
 }
 
 void AMDGPUAsmPrinter::EmitGlobalVariable(const GlobalVariable *GV) {
 
-  if (TM.getTargetTriple().getOS() != Triple::AMDHSA ||
-      GV->isDeclaration()) {
+  if (TM.getTargetTriple().getOS() != Triple::AMDHSA) {
+    AsmPrinter::EmitGlobalVariable(GV);
+    return;
+  }
+
+  if (GV->isDeclaration() || GV->getLinkage() == GlobalValue::PrivateLinkage) {
     AsmPrinter::EmitGlobalVariable(GV);
     return;
   }
 
   // Group segment variables aren't emitted in HSA.
   if (AMDGPU::isGroupSegment(GV))
     return;
 
   AMDGPUTargetStreamer *TS =

lib/Target/AMDGPU/AMDGPUTargetObjectFile.cpp

MCSection *AMDGPUHSATargetObjectFile::SelectSectionForGlobal(
   if (AMDGPU::isGlobalSegment(GV)) {
     if (isAgentAllocation(GV))
       return DataGlobalAgentSection;
 
     if (isProgramAllocation(GV))
       return DataGlobalProgramSection;
   }
 
-  if (Kind.isReadOnly() && AMDGPU::isReadOnlySegment(GV))
-    return RodataReadonlyAgentSection;
-
-  return TargetLoweringObjectFileELF::SelectSectionForGlobal(GV, Kind, Mang, TM);
+  return AMDGPUTargetObjectFile::SelectSectionForGlobal(GV, Kind, Mang, TM);
 }

test/CodeGen/AMDGPU/global-constant.ll

-; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=NOHSA %s
+; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=kaveri -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=HSA %s
 
 @readonly = private unnamed_addr addrspace(2) constant [4 x float] [float 0.0, float 1.0, float 2.0, float 3.0]
 @readonly2 = private unnamed_addr addrspace(2) constant [4 x float] [float 4.0, float 5.0, float 6.0, float 7.0]
 
 ; GCN-LABEL: {{^}}main:
 ; GCN: s_getpc_b64 s{{\[}}[[PC0_LO:[0-9]+]]:[[PC0_HI:[0-9]+]]{{\]}}
 ; GCN-NEXT: s_add_u32 s{{[0-9]+}}, s[[PC0_LO]], readonly
 ; GCN: s_addc_u32 s{{[0-9]+}}, s[[PC0_HI]], 0
 ; GCN: s_getpc_b64 s{{\[}}[[PC1_LO:[0-9]+]]:[[PC1_HI:[0-9]+]]{{\]}}
 ; GCN-NEXT: s_add_u32 s{{[0-9]+}}, s[[PC1_LO]], readonly
 ; GCN: s_addc_u32 s{{[0-9]+}}, s[[PC1_HI]], 0
-; GCN: .text
+; NOHSA: .text
+; HSA: .hsatext
 ; GCN: readonly:
 ; GCN: readonly2:
 define void @main(i32 %index, float addrspace(1)* %out) {
   %ptr = getelementptr [4 x float], [4 x float] addrspace(2) * @readonly, i32 0, i32 %index
   %val = load float, float addrspace(2)* %ptr
   store float %val, float addrspace(1)* %out
   %ptr2 = getelementptr [4 x float], [4 x float] addrspace(2) * @readonly2, i32 0, i32 %index
   %val2 = load float, float addrspace(2)* %ptr2
   store float %val2, float addrspace(1)* %out
   ret void
 }