[Mlir-commits] [mlir] [MLIR][GPUToLLVMSPV] Relax the width check in gpu.shuffle lowering (PR #183445)

[Jianhui Li]

================
@@ -345,7 +345,7 @@ struct GPUShuffleConversion final : ConvertOpToLLVMPattern<gpu::ShuffleOp> {
   LogicalResult
   matchAndRewrite(gpu::ShuffleOp op, OpAdaptor adaptor,
                   ConversionPatternRewriter &rewriter) const final {
-    if (!hasValidWidth(op))
+    if (getSubgroupSize(op) && !hasValidWidth(op))
----------------
Jianhui-Li wrote:

I change the code slightly to improve the readability. I moved the getSubgroupSize() out from the function to make it explicit that getSubgroupeSize() may not succeed, and in that case this pass don't treat it as critical compilation failure. 
Thanks for reviewing!

https://github.com/llvm/llvm-project/pull/183445