[Mlir-commits] [mlir] [MLIR][GPU-LLVM] Add in-pass signature update option for opencl kernels (PR #105664)

[Petr Kurapov]

================
@@ -325,16 +359,45 @@ struct GPUToLLVMSPVConversionPass final
     LLVMTypeConverter converter(context, options);
     LLVMConversionTarget target(*context);
 
+    if (forceOpenclAddressSpaces) {
+      MemorySpaceToOpenCLMemorySpaceConverter converter;
+      AttrTypeReplacer replacer;
+      replacer.addReplacement([&converter](BaseMemRefType origType)
+                                  -> std::optional<BaseMemRefType> {
+        return converter.convertType<BaseMemRefType>(origType);
+      });
+
+      replacer.recursivelyReplaceElementsIn(getOperation(),
+                                            /*replaceAttrs=*/true,
+                                            /*replaceLocs=*/false,
+                                            /*replaceTypes=*/true);
+    }
+
     target.addIllegalOp<gpu::BarrierOp, gpu::BlockDimOp, gpu::BlockIdOp,
                         gpu::GPUFuncOp, gpu::GlobalIdOp, gpu::GridDimOp,
                         gpu::ReturnOp, gpu::ShuffleOp, gpu::ThreadIdOp>();
 
     populateGpuToLLVMSPVConversionPatterns(converter, patterns);
+    populateFuncToLLVMConversionPatterns(converter, patterns);
+    populateFinalizeMemRefToLLVMConversionPatterns(converter, patterns);
----------------
kurapov-peter wrote:

These are to handle the callees (which are `func.func` and hence aren't handled at all). I think the pass would be just incomplete without it since it leaves out memrefs between the caller and the callee, WDYT?

https://github.com/llvm/llvm-project/pull/105664