[Mlir-commits] [mlir] [mlir][bufferization] implement BufferizableOpInterface for concat op (PR #140171)

[Jeremy Kun]

j2kun wrote:

Turns out this was a bug in my code, and I added another test where the dimension not being concatenated along is dynamic to exercise the fix.

```patch
@@ -1110,7 +1110,7 @@ struct ConcatOpInterface
     for (const auto &[dimIdx, dimSize] :
          llvm::enumerate(tensorType.getShape())) {
       if (dimSize == ShapedType::kDynamic) {
-        auto dimOp = rewriter.create<memref::DimOp>(loc, dstBuffer, dimSize);
+        auto dimOp = rewriter.create<memref::DimOp>(loc, dstBuffer, dimIdx);
         sizes.push_back(dimOp.getResult());
         if (dimIdx == concatDim)
           dynamicConcatDim = true;
```

https://github.com/llvm/llvm-project/pull/140171