[Mlir-commits] [mlir] [mlir][tensor] Improve `FoldTensorCastProducerOp` (dynamic shapes) (PR #114559)

[llvmlistbot at llvm.org]

================
@@ -4756,8 +4756,50 @@ struct FoldTensorCastProducerOp
         newResultTypes[dpsInitIdx++] = newOperands.back().getType();
     }
 
-    // Clone op.
-    Operation *newOp = clone(rewriter, op, newResultTypes, newOperands);
+    // For ops that have sizes-like attribute, update these accordingly.
+    // For now, only `tensor.pack` is supported.
+    // TODO: Generalize to make it work with other ops as well (e.g.
+    // `tensor.unpack`)
+    SmallVector<OpFoldResult> newMixedTileSizes;
----------------
MaheshRavishankar wrote:

I read your discourse thread on that. I agree that we need to do this for all ops, but the way the "size" operands map to result types for operands varies on an operand. I dont think we can "combine them all" into a single pattern. Lets just split the pattern. If you find a way to combine it then thats great, but I dont see a way.

https://github.com/llvm/llvm-project/pull/114559