[Mlir-commits] [mlir] [mlir] Fix bug in UnPackOp tiling implementation causing infinite loop (PR #113571)

[llvmlistbot at llvm.org]

================
@@ -554,13 +554,14 @@ struct UnPackOpTiling
     sliceSrcIndices.append(numInnerTiles, zeroAttr);
     sliceSrcSizes.append(unpackOp.getMixedTiles());
     sliceSrcStrides.append(numInnerTiles, oneAttr);
-    Value sliceSource =
+    SmallVector<Operation *> generatedSlices;
+    ExtractSliceOp sliceSource =
         b.create<ExtractSliceOp>(loc, unpackOp.getSource(), sliceSrcIndices,
                                  sliceSrcSizes, sliceSrcStrides);
+    generatedSlices.push_back(sliceSource);
----------------
Max191 wrote:

I looked at a few other implementations and they seemed to only add the init argument slices, so I thought there was some context I was missing. I checked some other implementations and they add the input slices as well, though, so I think the ops I looked at happened to be special cases. I think it is good as I had it.

https://github.com/llvm/llvm-project/pull/113571