[AMDGPU] Extend WorkGroupID* codegen for compute shaders

[cdevadas]

Summary

Currently, the codegen support for llvm.amdgcn.workgroup.id*
intrinsics are enabled only for compute kernels. In addition,
this patch enables their selection for compute shaders on
subtargets that have architected SGPRs.

[cdevadas]

Renamed the testcase.

[cdevadas]

Extended only for amdgpu_cs and when the subtarget has architected SGPRs enabled.

[cdevadas]

Ping

[foad]

Looks OK to me, thanks.

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

2428–2436

[cdevadas]

Split the assertion into smaller chunks.

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

SDValue SITargetLowering::LowerFormalArguments(
   CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), ArgLocs,
                  *DAG.getContext());
 
   bool IsGraphics = AMDGPU::isGraphics(CallConv);
   bool IsKernel = AMDGPU::isKernel(CallConv);
   bool IsEntryFunc = AMDGPU::isEntryFunctionCC(CallConv);
 
   if (IsGraphics) {
     assert(!Info->hasDispatchPtr() && !Info->hasKernargSegmentPtr() &&
            (!Info->hasFlatScratchInit() || Subtarget->enableFlatScratch()) &&
-           !Info->hasWorkGroupIDX() && !Info->hasWorkGroupIDY() &&
-           !Info->hasWorkGroupIDZ() && !Info->hasWorkGroupInfo() &&
-           !Info->hasLDSKernelId() && !Info->hasWorkItemIDX() &&
-           !Info->hasWorkItemIDY() && !Info->hasWorkItemIDZ());
+           ((!Info->hasWorkGroupIDX() && !Info->hasWorkGroupIDY() &&
+             !Info->hasWorkGroupIDZ()) ||
+            Subtarget->hasArchitectedSGPRs()) &&
+           !Info->hasWorkGroupInfo() && !Info->hasLDSKernelId() &&
+           !Info->hasWorkItemIDX() && !Info->hasWorkItemIDY() &&
+           !Info->hasWorkItemIDZ());
   }
 
   if (CallConv == CallingConv::AMDGPU_PS) {
     processPSInputArgs(Splits, CallConv, Ins, Skipped, FType, Info);
 
     // At least one interpolation mode must be enabled or else the GPU will
     // hang.
     //
     // Check PSInputAddr instead of PSInputEnable. The idea is that if the user

llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp

SIMachineFunctionInfo::SIMachineFunctionInfo(const Function &F,
   }
 
   bool isAmdHsaOrMesa = ST.isAmdHsaOrMesa(F);
   if (isAmdHsaOrMesa && !ST.enableFlatScratch())
     PrivateSegmentBuffer = true;
   else if (ST.isMesaGfxShader(F))
     ImplicitBufferPtr = true;
 
-  if (!AMDGPU::isGraphics(CC)) {
+  if (!AMDGPU::isGraphics(CC) || ST.hasArchitectedSGPRs()) {
     if (IsKernel || !F.hasFnAttribute("amdgpu-no-workgroup-id-x"))
       WorkGroupIDX = true;
 
     if (!F.hasFnAttribute("amdgpu-no-workgroup-id-y"))
       WorkGroupIDY = true;
 
     if (!F.hasFnAttribute("amdgpu-no-workgroup-id-z"))
       WorkGroupIDZ = true;
+  }
 
+  if (!AMDGPU::isGraphics(CC)) {
     if (IsKernel || !F.hasFnAttribute("amdgpu-no-workitem-id-x"))
       WorkItemIDX = true;
 
     if (!F.hasFnAttribute("amdgpu-no-workitem-id-y") &&
         ST.getMaxWorkitemID(F, 1) != 0)
       WorkItemIDY = true;
 
     if (!F.hasFnAttribute("amdgpu-no-workitem-id-z") &&

llvm/test/CodeGen/AMDGPU/lower-work-group-ids-intrinsics.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 -mattr=+architected-sgprs --verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9,GFX9-SDAG %s
+; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 -mattr=+architected-sgprs -global-isel --verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s
+define amdgpu_cs void @_amdgpu_cs_main() {
+; GFX9-SDAG-LABEL: _amdgpu_cs_main:
+; GFX9-SDAG:       ; %bb.0: ; %.entry
+; GFX9-SDAG-NEXT:    s_lshr_b32 s2, ttmp7, 16
+; GFX9-SDAG-NEXT:    s_and_b32 s1, ttmp7, 0xffff
+; GFX9-SDAG-NEXT:    v_mov_b32_e32 v0, ttmp9
+; GFX9-SDAG-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-SDAG-NEXT:    v_mov_b32_e32 v2, s2
+; GFX9-SDAG-NEXT:    buffer_store_dwordx3 v[0:2], off, s[0:3], 0
+; GFX9-SDAG-NEXT:    s_endpgm
+;
+; GFX9-GISEL-LABEL: _amdgpu_cs_main:
+; GFX9-GISEL:       ; %bb.0: ; %.entry
+; GFX9-GISEL-NEXT:    s_mov_b32 s0, ttmp9
+; GFX9-GISEL-NEXT:    s_and_b32 s1, ttmp7, 0xffff
+; GFX9-GISEL-NEXT:    s_lshr_b32 s2, ttmp7, 16
+; GFX9-GISEL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-GISEL-NEXT:    v_mov_b32_e32 v2, s2
+; GFX9-GISEL-NEXT:    buffer_store_dwordx3 v[0:2], off, s[0:3], 0
+; GFX9-GISEL-NEXT:    s_endpgm
+.entry:
+  %idx = call i32 @llvm.amdgcn.workgroup.id.x()
+  %idy = call i32 @llvm.amdgcn.workgroup.id.y()
+  %idz = call i32 @llvm.amdgcn.workgroup.id.z()
+  %ielemx = insertelement <3 x i32> undef, i32 %idx, i64 0
+  %ielemy = insertelement <3 x i32> %ielemx, i32 %idy, i64 1
+  %ielemz = insertelement <3 x i32> %ielemy, i32 %idz, i64 2
+  call void @llvm.amdgcn.raw.buffer.store.v3i32(<3 x i32> %ielemz, <4 x i32> undef, i32 0, i32 0, i32 0)
+  ret void
+}
+
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare i32 @llvm.amdgcn.workgroup.id.x()
+
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare i32 @llvm.amdgcn.workgroup.id.y()
+
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare i32 @llvm.amdgcn.workgroup.id.z()
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(write)
+declare void @llvm.amdgcn.raw.buffer.store.v3i32(<3 x i32>, <4 x i32>, i32, i32, i32 immarg)
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; GFX9: {{.*}}