[llvm] [AMDGPU] Quit PromoteAllocaToVector if intrinsic is used (PR #68744)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-amdgpu

Author: Mariusz Sikora (mariusz-sikora-at-amd)

<details>
<summary>Changes</summary>

Attached test will cause crash without this change.

We should not remove isAssumeLikeIntrinsic instruction if it is used by other instruction.

---
Full diff: https://github.com/llvm/llvm-project/pull/68744.diff


2 Files Affected:

- (modified) llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp (+2) 
- (modified) llvm/test/CodeGen/AMDGPU/promote-alloca-mem-intrinsics.ll (+9) 


``````````diff
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp b/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
index 3707a960211eb49..6f5791fecdaf54c 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
@@ -772,6 +772,8 @@ bool AMDGPUPromoteAllocaImpl::tryPromoteAllocaToVector(AllocaInst &Alloca) {
 
     // Ignore assume-like intrinsics and comparisons used in assumes.
     if (isAssumeLikeIntrinsic(Inst)) {
+      if (!Inst->use_empty())
+        return RejectUser(Inst, "assume-line intrinsic cannot have any users");
       UsersToRemove.push_back(Inst);
       continue;
     }
diff --git a/llvm/test/CodeGen/AMDGPU/promote-alloca-mem-intrinsics.ll b/llvm/test/CodeGen/AMDGPU/promote-alloca-mem-intrinsics.ll
index 0bba1bdce95655b..5616bc0f5ef3c12 100644
--- a/llvm/test/CodeGen/AMDGPU/promote-alloca-mem-intrinsics.ll
+++ b/llvm/test/CodeGen/AMDGPU/promote-alloca-mem-intrinsics.ll
@@ -53,6 +53,15 @@ define amdgpu_kernel void @promote_with_objectsize(ptr addrspace(1) %out) #0 {
   ret void
 }
 
+; CHECK-LABEL: @promote_with_objectsize_8(
+; CHECK: [[PTR:%[0-9]+]] = getelementptr inbounds [64 x [8 x i32]], ptr addrspace(3) @promote_with_objectsize_8.alloca, i32 0, i32 %{{[0-9]+}}
+; CHECK: call i32 @llvm.objectsize.i32.p3(ptr addrspace(3) [[PTR]], i1 false, i1 false, i1 false)
+define amdgpu_kernel void @promote_with_objectsize_8(ptr addrspace(1) %out) #0 {
+  %alloca = alloca [8 x i32], align 4, addrspace(5)
+  %size = call i32 @llvm.objectsize.i32.p5(ptr addrspace(5) %alloca, i1 false, i1 false, i1 false)
+  store i32 %size, ptr addrspace(1) %out
+  ret void
+}
 ; CHECK-LABEL: @promote_alloca_used_twice_in_memcpy(
 ; CHECK: call void @llvm.memcpy.p3.p3.i64(ptr addrspace(3) align 8 dereferenceable(16) %arrayidx1, ptr addrspace(3) align 8 dereferenceable(16) %arrayidx2, i64 16, i1 false)
 define amdgpu_kernel void @promote_alloca_used_twice_in_memcpy(i32 %c) {

``````````

</details>


https://github.com/llvm/llvm-project/pull/68744