[AMDGPU] Check for pointer operand while refining LDS align

[rampitec]

Summary

Also skips the propagation if alignment is 1.

[foad]

llvm/lib/Target/AMDGPU/AMDGPULowerModuleLDSPass.cpp

352–353

You look through bitcasts, so "Ptr" does not necessarily have pointer type, so you really do need this check.

[rampitec]

llvm/lib/Target/AMDGPU/AMDGPULowerModuleLDSPass.cpp

352–353

A bitcast cannot convert pointer to a non-pointer, and I am not handling ptrtoint.

[foad]

llvm/lib/Target/AMDGPU/AMDGPULowerModuleLDSPass.cpp

352–353

OK sorry, I missed that detail.

llvm/lib/Target/AMDGPU/AMDGPULowerModuleLDSPass.cpp

if (!F) {
         }
       }
     }
     return true;
   }
 
   void refineUsesAlignment(Value *Ptr, Align A, const DataLayout &DL,
                            unsigned MaxDepth = 5) {
-    if (!MaxDepth)
+    if (!MaxDepth || A == 1)
       return;
 
     for (User *U : Ptr->users()) {
       if (auto *LI = dyn_cast<LoadInst>(U)) {
         LI->setAlignment(std::max(A, LI->getAlign()));
         continue;
       }
       if (auto *SI = dyn_cast<StoreInst>(U)) {
+        if (SI->getPointerOperand() == Ptr)
           SI->setAlignment(std::max(A, SI->getAlign()));
         continue;
       }
       if (auto *AI = dyn_cast<AtomicRMWInst>(U)) {
+        // None of atomicrmw operations can work on pointers, but let's
+        // check it anyway in case it will or we will process ConstantExpr.
+        if (AI->getPointerOperand() == Ptr)
           AI->setAlignment(std::max(A, AI->getAlign()));
         continue;
       }
       if (auto *AI = dyn_cast<AtomicCmpXchgInst>(U)) {
+        if (AI->getPointerOperand() == Ptr)
           AI->setAlignment(std::max(A, AI->getAlign()));
         continue;
       }
       if (auto *GEP = dyn_cast<GetElementPtrInst>(U)) {
         unsigned BitWidth = DL.getIndexTypeSizeInBits(GEP->getType());
         APInt Off(BitWidth, 0);
         if (GEP->getPointerOperand() == Ptr &&
             GEP->accumulateConstantOffset(DL, Off)) {
           Align GA = commonAlignment(A, Off.getLimitedValue());

llvm/test/CodeGen/AMDGPU/lower-kernel-lds-super-align.ll

 ; RUN: opt -S -mtriple=amdgcn-- -amdgpu-lower-module-lds < %s | FileCheck --check-prefixes=CHECK,SUPER-ALIGN_ON %s
 ; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-lower-module-lds < %s | FileCheck --check-prefixes=CHECK,SUPER-ALIGN_ON %s
 ; RUN: opt -S -mtriple=amdgcn-- -amdgpu-lower-module-lds --amdgpu-super-align-lds-globals=false < %s | FileCheck --check-prefixes=CHECK,SUPER-ALIGN_OFF %s
 ; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-lower-module-lds --amdgpu-super-align-lds-globals=false < %s | FileCheck --check-prefixes=CHECK,SUPER-ALIGN_OFF %s
 
 ; CHECK: %llvm.amdgcn.kernel.k1.lds.t = type { [32 x i8] }
 ; CHECK: %llvm.amdgcn.kernel.k2.lds.t = type { i16, [2 x i8], i16 }
 ; CHECK: %llvm.amdgcn.kernel.k3.lds.t = type { [32 x i64], [32 x i32] }
+; CHECK: %llvm.amdgcn.kernel.k4.lds.t = type { [2 x i32 addrspace(3)*] }
 
 ; CHECK-NOT: @lds.1
 @lds.1 = internal unnamed_addr addrspace(3) global [32 x i8] undef, align 1
 
 ; SUPER-ALIGN_ON: @llvm.amdgcn.kernel.k1.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k1.lds.t undef, align 16
 ; SUPER-ALIGN_OFF: @llvm.amdgcn.kernel.k1.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k1.lds.t undef, align 1
 
 ; CHECK: @llvm.amdgcn.kernel.k2.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k2.lds.t undef, align 4
 ; SUPER-ALIGN_ON:  @llvm.amdgcn.kernel.k3.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k3.lds.t undef, align 16
 ; SUPER-ALIGN_OFF: @llvm.amdgcn.kernel.k3.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k3.lds.t undef, align 8
 
+; SUPER-ALIGN_ON:  @llvm.amdgcn.kernel.k4.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k4.lds.t undef, align 16
+; SUPER-ALIGN_OFF: @llvm.amdgcn.kernel.k4.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k4.lds.t undef, align 4
+
 ; CHECK-LABEL: @k1
 ; CHECK:  %1 = getelementptr inbounds [32 x i8], [32 x i8] addrspace(3)* getelementptr inbounds (%llvm.amdgcn.kernel.k1.lds.t, %llvm.amdgcn.kernel.k1.lds.t addrspace(3)* @llvm.amdgcn.kernel.k1.lds, i32 0, i32 0), i32 0, i32 0
 ; CHECK:  %2 = addrspacecast i8 addrspace(3)* %1 to i8*
 ; CHECK:  %ptr = getelementptr inbounds i8, i8* %2, i64 %x
 ; CHECK:  store i8 1, i8* %ptr, align 1
 define amdgpu_kernel void @k1(i64 %x) {
   %ptr = getelementptr inbounds i8, i8* addrspacecast ([32 x i8] addrspace(3)* @lds.1 to i8*), i64 %x
   store i8 1, i8 addrspace(0)* %ptr, align 1
define amdgpu_kernel void @k3(i64 %x) {
 
   store i32 21, i32* %ptr1.ac, align 4
   store i32 22, i32* %ptr2.ac, align 4
   store i32 23, i32* %ptr3.ac, align 4
   store i32 24, i32* %ptr4.ac, align 4
 
   ret void
 }
+
+@lds.6 = internal unnamed_addr addrspace(3) global [2 x i32 addrspace(3)*] undef, align 4
+
+; Check that aligment is not propagated if use is not a pointer operand.
+
+; CHECK-LABEL: @k4
+; SUPER-ALIGN_ON:  store i32 undef, i32 addrspace(3)* %ptr, align 8
+; SUPER-ALIGN_OFF: store i32 undef, i32 addrspace(3)* %ptr, align 4
+; CHECK:           store i32 addrspace(3)* %ptr, i32 addrspace(3)** undef, align 4
+; SUPER-ALIGN_ON:  %val1 = cmpxchg volatile i32 addrspace(3)* %ptr, i32 1, i32 2 monotonic monotonic, align 8
+; SUPER-ALIGN_OFF: %val1 = cmpxchg volatile i32 addrspace(3)* %ptr, i32 1, i32 2 monotonic monotonic, align 4
+; CHECK:           %val2 = cmpxchg volatile i32 addrspace(3)** undef, i32 addrspace(3)* %ptr, i32 addrspace(3)* undef monotonic monotonic, align 4
+define amdgpu_kernel void @k4() {
+  %gep = getelementptr inbounds i32 addrspace(3)*, i32 addrspace(3)* addrspace(3)* bitcast ([2 x i32 addrspace(3)*] addrspace(3)* @lds.6 to i32 addrspace(3)* addrspace(3)*), i64 1
+  %ptr = bitcast i32 addrspace(3)* addrspace(3)* %gep to i32 addrspace(3)*
+  store i32 undef, i32 addrspace(3)* %ptr, align 4
+  store i32 addrspace(3)* %ptr, i32 addrspace(3)** undef, align 4
+  %val1 = cmpxchg volatile i32 addrspace(3)* %ptr, i32 1, i32 2 monotonic monotonic, align 4
+  %val2 = cmpxchg volatile i32 addrspace(3)** undef, i32 addrspace(3)* %ptr, i32 addrspace(3)* undef monotonic monotonic, align 4
+  ret void
+}