[llvm-branch-commits] [llvm] AMDGPU: Fix inst-selection of large scratch offsets with sgpr base (PR #110256)

[via llvm-branch-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-amdgpu

Author: Petar Avramovic (petar-avramovic)

<details>
<summary>Changes</summary>

Use i32 for offset instead of i16, this way it does not get interpreted
as negative 16 bit offset.

---
Full diff: https://github.com/llvm/llvm-project/pull/110256.diff


2 Files Affected:

- (modified) llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp (+1-1) 
- (modified) llvm/test/CodeGen/AMDGPU/flat-scratch.ll (+2-2) 


``````````diff
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
index d3d5bc924525fc..48971a6840c779 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
@@ -1911,7 +1911,7 @@ bool AMDGPUDAGToDAGISel::SelectScratchSAddr(SDNode *Parent, SDValue Addr,
                     0);
   }
 
-  Offset = CurDAG->getTargetConstant(COffsetVal, DL, MVT::i16);
+  Offset = CurDAG->getTargetConstant(COffsetVal, DL, MVT::i32);
 
   return true;
 }
diff --git a/llvm/test/CodeGen/AMDGPU/flat-scratch.ll b/llvm/test/CodeGen/AMDGPU/flat-scratch.ll
index 667a8a38c62ecc..496ac80a3dfbcf 100644
--- a/llvm/test/CodeGen/AMDGPU/flat-scratch.ll
+++ b/llvm/test/CodeGen/AMDGPU/flat-scratch.ll
@@ -4926,7 +4926,7 @@ define amdgpu_gs void @sgpr_base_large_offset(ptr addrspace(1) %out, ptr addrspa
 ;
 ; GFX12-LABEL: sgpr_base_large_offset:
 ; GFX12:       ; %bb.0: ; %entry
-; GFX12-NEXT:    scratch_load_b32 v2, off, s0 offset:-24
+; GFX12-NEXT:    scratch_load_b32 v2, off, s0 offset:65512
 ; GFX12-NEXT:    s_wait_loadcnt 0x0
 ; GFX12-NEXT:    global_store_b32 v[0:1], v2, off
 ; GFX12-NEXT:    s_nop 0
@@ -4985,7 +4985,7 @@ define amdgpu_gs void @sgpr_base_large_offset(ptr addrspace(1) %out, ptr addrspa
 ;
 ; GFX12-PAL-LABEL: sgpr_base_large_offset:
 ; GFX12-PAL:       ; %bb.0: ; %entry
-; GFX12-PAL-NEXT:    scratch_load_b32 v2, off, s0 offset:-24
+; GFX12-PAL-NEXT:    scratch_load_b32 v2, off, s0 offset:65512
 ; GFX12-PAL-NEXT:    s_wait_loadcnt 0x0
 ; GFX12-PAL-NEXT:    global_store_b32 v[0:1], v2, off
 ; GFX12-PAL-NEXT:    s_nop 0

``````````

</details>


https://github.com/llvm/llvm-project/pull/110256