Automerge: [AMDGPULowerBufferFatPointers] Handle ptrtoaddr by extending the offset

Reviewed By: krzysz00

Pull Request: llvm/llvm-project#139413

Author: arichardson

llvm/lib/Target/AMDGPU/AMDGPULowerBufferFatPointers.cpp

@@ -1361,6 +1361,7 @@ class SplitPtrStructs : public InstVisitor<SplitPtrStructs, PtrParts> {
   PtrParts visitAtomicCmpXchgInst(AtomicCmpXchgInst &AI);
   PtrParts visitGetElementPtrInst(GetElementPtrInst &GEP);
 
+  PtrParts visitPtrToAddrInst(PtrToAddrInst &PA);
   PtrParts visitPtrToIntInst(PtrToIntInst &PI);
   PtrParts visitIntToPtrInst(IntToPtrInst &IP);
   PtrParts visitAddrSpaceCastInst(AddrSpaceCastInst &I);
@@ -1954,6 +1955,21 @@ PtrParts SplitPtrStructs::visitPtrToIntInst(PtrToIntInst &PI) {
   return {nullptr, nullptr};
 }
 
+PtrParts SplitPtrStructs::visitPtrToAddrInst(PtrToAddrInst &PA) {
+  Value *Ptr = PA.getPointerOperand();
+  if (!isSplitFatPtr(Ptr->getType()))
+    return {nullptr, nullptr};
+  IRB.SetInsertPoint(&PA);
+
+  auto [Rsrc, Off] = getPtrParts(Ptr);
+  Value *Res = IRB.CreateIntCast(Off, PA.getType(), /*isSigned=*/false);
+  copyMetadata(Res, &PA);
+  Res->takeName(&PA);
+  SplitUsers.insert(&PA);
+  PA.replaceAllUsesWith(Res);
+  return {nullptr, nullptr};
+}
+
 PtrParts SplitPtrStructs::visitIntToPtrInst(IntToPtrInst &IP) {
   if (!isSplitFatPtr(IP.getType()))
     return {nullptr, nullptr};

llvm/test/CodeGen/AMDGPU/lower-buffer-fat-pointers-pointer-ops.ll

@@ -1,7 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
-; FIXME: Currently generates invalid IR for ptrtoaddr
-; RUN: opt -S -mcpu=gfx900 -amdgpu-lower-buffer-fat-pointers -disable-verify < %s | FileCheck %s
-; RUN: opt -S -mcpu=gfx900 -passes=amdgpu-lower-buffer-fat-pointers -disable-verify < %s | FileCheck %s
+; RUN: opt -S -mcpu=gfx900 -amdgpu-lower-buffer-fat-pointers < %s | FileCheck %s
+; RUN: opt -S -mcpu=gfx900 -passes=amdgpu-lower-buffer-fat-pointers < %s | FileCheck %s
 
 target triple = "amdgcn--"
 
@@ -256,50 +255,50 @@ define i32 @ptrtoint_offset(ptr addrspace(7) %ptr) {
   ret i32 %ret
 }
 
-;; FIXME: this currently generates invalid IR
 define i32 @ptrtoaddr(ptr addrspace(7) %ptr) {
 ; CHECK-LABEL: define i32 @ptrtoaddr
 ; CHECK-SAME: ({ ptr addrspace(8), i32 } [[PTR:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:    [[RET:%.*]] = ptrtoaddr { ptr addrspace(8), i32 } [[PTR]] to i32
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[RET:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
 ; CHECK-NEXT:    ret i32 [[RET]]
 ;
   %ret = ptrtoaddr ptr addrspace(7) %ptr to i32
   ret i32 %ret
 }
 
-;; FIXME: this currently generates invalid IR
 define <2 x i32> @ptrtoaddr_vec(<2 x ptr addrspace(7)> %ptr) {
 ; CHECK-LABEL: define <2 x i32> @ptrtoaddr_vec
 ; CHECK-SAME: ({ <2 x ptr addrspace(8)>, <2 x i32> } [[PTR:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:    [[RET:%.*]] = ptrtoaddr { <2 x ptr addrspace(8)>, <2 x i32> } [[PTR]] to <2 x i32>
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { <2 x ptr addrspace(8)>, <2 x i32> } [[PTR]], 0
+; CHECK-NEXT:    [[RET:%.*]] = extractvalue { <2 x ptr addrspace(8)>, <2 x i32> } [[PTR]], 1
 ; CHECK-NEXT:    ret <2 x i32> [[RET]]
 ;
   %ret = ptrtoaddr <2 x ptr addrspace(7)> %ptr to <2 x i32>
   ret <2 x i32> %ret
 }
 
 ;; Check that we extend the offset to i160.
-;; FIXME: this currently generates invalid IR
 define i160 @ptrtoaddr_ext(ptr addrspace(7) %ptr) {
 ; CHECK-LABEL: define i160 @ptrtoaddr_ext
 ; CHECK-SAME: ({ ptr addrspace(8), i32 } [[PTR:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:    [[ADDR:%.*]] = ptrtoaddr { ptr addrspace(8), i32 } [[PTR]] to i32
-; CHECK-NEXT:    [[EXT:%.*]] = zext i32 [[ADDR]] to i160
-; CHECK-NEXT:    ret i160 [[EXT]]
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    [[RET:%.*]] = zext i32 [[PTR_OFF]] to i160
+; CHECK-NEXT:    ret i160 [[RET]]
 ;
   %addr = ptrtoaddr ptr addrspace(7) %ptr to i32
   %ext = zext i32 %addr to i160
   ret i160 %ext
 }
 
-;; FIXME: this currently generates invalid IR
 ;; Check that we truncate the offset to i16.
 define i16 @ptrtoaddr_trunc(ptr addrspace(7) %ptr) {
 ; CHECK-LABEL: define i16 @ptrtoaddr_trunc
 ; CHECK-SAME: ({ ptr addrspace(8), i32 } [[PTR:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:    [[ADDR:%.*]] = ptrtoaddr { ptr addrspace(8), i32 } [[PTR]] to i32
-; CHECK-NEXT:    [[TRUNC:%.*]] = trunc i32 [[ADDR]] to i16
-; CHECK-NEXT:    ret i16 [[TRUNC]]
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    [[RET:%.*]] = trunc i32 [[PTR_OFF]] to i16
+; CHECK-NEXT:    ret i16 [[RET]]
 ;
   %addr = ptrtoaddr ptr addrspace(7) %ptr to i32
   %trunc = trunc i32 %addr to i16