[Mlir-commits] [mlir] [mlir][tosa] Add support for matmul_t (PR #203894)

[Luke Hutton]

================
@@ -248,6 +248,14 @@ LogicalResult ProfileInfoDepot::populateProfileInfo(tosa::MatMulOp op) {
   return success();
 }
 
+template <>
+LogicalResult ProfileInfoDepot::populateProfileInfo(tosa::MatMulTOp op) {
+  addValue(op.getA());
+  addValue(op.getB());
+  addValue(op.getOutput());
----------------
lhutton1 wrote:

Apologies, yes, this is redundant as the types are checked in the verifier: https://github.com/llvm/llvm-project/blob/main/mlir/lib/Dialect/Tosa/IR/TosaOps.cpp#L2116.

That said, it does seem as though the script that generates TosaComplianceData.h.inc as part of the spec is now producing entries for zero points for matmul_t similarly to matmul (it's not immediately clear to me when that started). Therefore, to make copy and pasting the entries easier, I think we should add the zero points here like you suggested. It's also possible we may want to relax the verifier in the future.

https://github.com/llvm/llvm-project/pull/203894