[PATCH] D72808: [mlir] [VectorOps] Lowering of vector.extract_slices to LLVM IR

[River Riddle via Phabricator via llvm-commits]

rriddle added inline comments.


================
Comment at: mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp:902
 
+// Consume tuple elements extracted by tuple get operations.
+// Note that this is a conversion pattern, since we must also
----------------
//// -> /////


================
Comment at: mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp:927
+      rewriter.replaceOp(tupleGetOp,
+                         tupleOp.getOperands()[tupleGetOp.getIndex()]);
+      return matchSuccess();
----------------
getOperand(tupleGetOp.getIndex())


================
Comment at: mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp:934
+
+// The removal of a vector tuple will fail until all canonicalizations
+// occured that remove its uses (viz. all individual elements are
----------------
Same here.


================
Comment at: mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp:943
+//
+// TODO(ajcbik): avoid all this by simply relying on DCE
+//
----------------
Note, we don't run DCE during dialect conversion.


================
Comment at: mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp:1075
+    Value tuple = rewriter.create<vector::TupleOp>(loc, tupleType, tupleValues);
+    rewriter.replaceOp(op, tuple);
+    return matchSuccess();
----------------
replaceOpWithNewOp?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D72808/new/

https://reviews.llvm.org/D72808