[llvm] [AMDGPU] Fix negative immediate offset for unbuffered smem loads (PR #89165)

[Matt Arsenault via llvm-commits]

================
@@ -1980,28 +1980,50 @@ bool AMDGPUDAGToDAGISel::SelectScratchSVAddr(SDNode *N, SDValue Addr,
   return true;
 }
 
+// For unbuffered smem loads, it is illegal for the Immediate Offset to be
+// negative if the resulting (Offset + (M0 or SOffset or zero) is negative.
+// Handle the case where the Immediate Offset + SOffset is negative.
+bool AMDGPUDAGToDAGISel::isSOffsetLegalWithImmOffset(SDValue *SOffset,
+                                                     bool Imm32Only,
+                                                     bool IsBuffer,
+                                                     int64_t ImmOffset) const {
+  if (AMDGPU::hasSMRDSignedImmOffset(*Subtarget) && !IsBuffer & !Imm32Only &&
+      ImmOffset < 0) {
----------------
arsenm wrote:

Move the AMDGPU::hasSMRDSignedImmOffset(*Subtarget) check last in this expression? 

https://github.com/llvm/llvm-project/pull/89165