AMDGPU: Fix emitting multiple stack loads for stack passed workitems

The same stack is loaded for each workitem ID, and each use. Nothing
prevents you from creating multiple fixed stack objects with the same
offsets, so this was creating a load for each unique frame index,
despite them being the same offset. Re-use the same frame index so the
loads are CSEable.

llvm-svn: 371148

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
index a9596b9..04f1da2 100644 (file)
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -4162,14 +4162,28 @@ SDValue AMDGPUTargetLowering::CreateLiveInRegister(SelectionDAG &DAG,
   return DAG.getCopyFromReg(DAG.getEntryNode(), SL, VReg, VT);
 }
 
+// This may be called multiple times, and nothing prevents creating multiple
+// objects at the same offset. See if we already defined this object.
+static int getOrCreateFixedStackObject(MachineFrameInfo &MFI, unsigned Size,
+                                       int64_t Offset) {
+  for (int I = MFI.getObjectIndexBegin(); I < 0; ++I) {
+    if (MFI.getObjectOffset(I) == Offset) {
+      assert(MFI.getObjectSize(I) == Size);
+      return I;
+    }
+  }
+
+  return MFI.CreateFixedObject(Size, Offset, true);
+}
+
 SDValue AMDGPUTargetLowering::loadStackInputValue(SelectionDAG &DAG,
                                                   EVT VT,
                                                   const SDLoc &SL,
                                                   int64_t Offset) const {
   MachineFunction &MF = DAG.getMachineFunction();
   MachineFrameInfo &MFI = MF.getFrameInfo();
+  int FI = getOrCreateFixedStackObject(MFI, VT.getStoreSize(), Offset);
 
-  int FI = MFI.CreateFixedObject(VT.getStoreSize(), Offset, true);
   auto SrcPtrInfo = MachinePointerInfo::getStack(MF, Offset);
   SDValue Ptr = DAG.getFrameIndex(FI, MVT::i32);
 

diff --git a/llvm/test/CodeGen/AMDGPU/callee-special-input-vgprs.ll b/llvm/test/CodeGen/AMDGPU/callee-special-input-vgprs.ll
index 2df3ad9..d5211ce 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/callee-special-input-vgprs.ll
+++ b/llvm/test/CodeGen/AMDGPU/callee-special-input-vgprs.ll
@@ -543,19 +543,25 @@ define void @func_call_too_many_args_use_workitem_id_x_byval() #1 {
   ret void
 }
 
+; Only one stack load should be emitted for all 3 values.
 ; GCN-LABEL: {{^}}too_many_args_use_workitem_id_xyz:
-; GCN: buffer_store_dword v32, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
-; GCN: buffer_load_dword v32, off, s[0:3], s32{{$}}
-; GCN: v_and_b32_e32 v32, 0x3ff, v32
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+]}}, v32
-; GCN: buffer_load_dword v32, off, s[0:3], s32{{$}}
-; GCN: v_bfe_u32 v32, v32, 10, 10
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+]}}, v32
+; GCN: buffer_store_dword v32, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; GCN: buffer_store_dword v33, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GCN-NOT: buffer_load_dword v{{[0-9]+}}, off, s[0:3], s32{{$}}
 ; GCN: buffer_load_dword v32, off, s[0:3], s32{{$}}
-; GCN: v_bfe_u32 v32, v32, 20, 10
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+]}}, v32
-
-; GCN: buffer_load_dword v32, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GCN-NOT: buffer_load_dword
+
+; GCN: v_and_b32_e32 [[AND_X:v[0-9]+]], 0x3ff, v32
+; GCN-NOT: buffer_load_dword
+; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+]}}, [[AND_X]]
+; GCN-NOT: buffer_load_dword
+; GCN: v_bfe_u32 [[BFE_Y:v[0-9]+]], v32, 10, 10
+; GCN-NEXT: v_bfe_u32 [[BFE_Z:v[0-9]+]], v32, 20, 10
+; GCN-NEXT: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+]}}, [[BFE_Y]]
+; GCN-NEXT: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+]}}, [[BFE_Z]]
+
+; GCN: buffer_load_dword v33, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GCN: buffer_load_dword v32, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
 ; GCN-NEXT: s_waitcnt
 ; GCN-NEXT: s_setpc_b64
 define void @too_many_args_use_workitem_id_xyz(