[Mlir-commits] [mlir] [mlir][linalg] Introduce transpose semantic to 'linalg.matmul' ops. (PR #104783)

[llvmlistbot at llvm.org]

================
@@ -1161,9 +1202,11 @@ LogicalResult mlir::linalg::detail::verifyStructuredOpInterface(Operation *op) {
            << ") to be equal to the number of input/output operands ("
            << linalgOp->getNumOperands() << ")";
 
+  bool hasBroadcast = false;
+  MatmulOp matmulOp = dyn_cast<MatmulOp>(op);
----------------
MaheshRavishankar wrote:

I am not sure it is a good idea to do `MatmulOp` specific checks within something that is something for all structured ops. Can we refactor this to keep the matmul specific checks separate?

EDIT: Ok, looks like you left it here to have a ramp to enabling this for default. That would be fine, but some of the checks below dont seem general enough to be part of the ramp (see below).

https://github.com/llvm/llvm-project/pull/104783