AMDGPU: Don't try to use LDS/vector for private if pointer value stored

If the pointer is the store's value operand, this would produce
a broken module. Make sure the use is actually for the pointer operand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243462 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp b/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
index cd968031a6101157cda990e50eecf14e644e3f5a..57b7a73bf56c24333b4b60f8ab88721dd18879b4 100644 (file)
--- a/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
+++ b/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
@@ -134,13 +134,17 @@ static Value* GEPToVectorIndex(GetElementPtrInst *GEP) {
 //
 // TODO: Check isTriviallyVectorizable for calls and handle other
 // instructions.
-static bool canVectorizeInst(Instruction *Inst) {
+static bool canVectorizeInst(Instruction *Inst, User *User) {
   switch (Inst->getOpcode()) {
   case Instruction::Load:
-  case Instruction::Store:
   case Instruction::BitCast:
   case Instruction::AddrSpaceCast:
     return true;
+  case Instruction::Store: {
+    // Must be the stored pointer operand, not a stored value.
+    StoreInst *SI = cast<StoreInst>(Inst);
+    return SI->getPointerOperand() == User;
+  }
   default:
     return false;
   }
@@ -166,7 +170,7 @@ static bool tryPromoteAllocaToVector(AllocaInst *Alloca) {
   for (User *AllocaUser : Alloca->users()) {
     GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(AllocaUser);
     if (!GEP) {
-      if (!canVectorizeInst(cast<Instruction>(AllocaUser)))
+      if (!canVectorizeInst(cast<Instruction>(AllocaUser), Alloca))
         return false;
 
       WorkList.push_back(AllocaUser);
@@ -184,7 +188,7 @@ static bool tryPromoteAllocaToVector(AllocaInst *Alloca) {
 
     GEPVectorIdx[GEP] = Index;
     for (User *GEPUser : AllocaUser->users()) {
-      if (!canVectorizeInst(cast<Instruction>(GEPUser)))
+      if (!canVectorizeInst(cast<Instruction>(GEPUser), AllocaUser))
         return false;
 
       WorkList.push_back(GEPUser);
@@ -255,6 +259,12 @@ static bool collectUsesWithPtrTypes(Value *Val, std::vector<Value*> &WorkList) {
     if (UseInst && UseInst->getOpcode() == Instruction::PtrToInt)
       return false;
 
+    if (StoreInst *SI = dyn_cast_or_null<StoreInst>(UseInst)) {
+      // Reject if the stored value is not the pointer operand.
+      if (SI->getPointerOperand() != Val)
+        return false;
+    }
+
     if (!User->getType()->isPointerTy())
       continue;
 

diff --git a/test/CodeGen/AMDGPU/promote-alloca-stored-pointer-value.ll b/test/CodeGen/AMDGPU/promote-alloca-stored-pointer-value.ll
new file mode 100644 (file)
index 0000000..2ee98cc
--- /dev/null
+++ b/test/CodeGen/AMDGPU/promote-alloca-stored-pointer-value.ll
@@ -0,0 +1,52 @@
+; RUN: llc -march=amdgcn < %s | FileCheck -check-prefix=GCN %s
+
+; Pointer value is stored in a candidate for LDS usage.
+
+; GCN-LABEL: {{^}}stored_lds_pointer_value:
+; GCN: buffer_store_dword v
+define void @stored_lds_pointer_value(float* addrspace(1)* %ptr) #0 {
+  %tmp = alloca float
+  store float 0.0, float *%tmp
+  store float* %tmp, float* addrspace(1)* %ptr
+  ret void
+}
+
+; GCN-LABEL: {{^}}stored_lds_pointer_value_gep:
+; GCN-DAG: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD0
+; GCN-DAG: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD1
+; GCN: buffer_store_dword v
+; GCN: buffer_store_dword v
+define void @stored_lds_pointer_value_gep(float* addrspace(1)* %ptr, i32 %idx) #0 {
+bb:
+  %tmp = alloca float, i32 16
+  store float 0.0, float* %tmp
+  %tmp2 = getelementptr inbounds float, float* %tmp, i32 %idx
+  store float* %tmp2, float* addrspace(1)* %ptr
+  ret void
+}
+
+; Pointer value is stored in a candidate for vector usage
+; GCN-LABEL: {{^}}stored_vector_pointer_value:
+; GCN-DAG: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD0
+; GCN-DAG: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD1
+; GCN: buffer_store_dword
+; GCN: buffer_store_dword
+; GCN: buffer_store_dword
+; GCN: buffer_store_dword
+define void @stored_vector_pointer_value(i32* addrspace(1)* %out, i32 %index) {
+entry:
+  %tmp0 = alloca [4 x i32]
+  %x = getelementptr [4 x i32], [4 x i32]* %tmp0, i32 0, i32 0
+  %y = getelementptr [4 x i32], [4 x i32]* %tmp0, i32 0, i32 1
+  %z = getelementptr [4 x i32], [4 x i32]* %tmp0, i32 0, i32 2
+  %w = getelementptr [4 x i32], [4 x i32]* %tmp0, i32 0, i32 3
+  store i32 0, i32* %x
+  store i32 1, i32* %y
+  store i32 2, i32* %z
+  store i32 3, i32* %w
+  %tmp1 = getelementptr [4 x i32], [4 x i32]* %tmp0, i32 0, i32 %index
+  store i32* %tmp1, i32* addrspace(1)* %out
+  ret void
+}
+
+attributes #0 = { nounwind }