elvin-n commented on a change in pull request #8527:
URL: https://github.com/apache/tvm/pull/8527#discussion_r675118270
##########
File path: src/relay/qnn/op/batch_matmul.cc
##########
@@ -84,7 +84,9 @@ Expr MakeQuantizedBatchMatmul(Expr x, Expr y, Expr
x_zero_point, Expr y_zero_poi
Expr BatchMatmulFirstTerm(const Expr& quantized_x, const Expr& quantized_y,
const BatchMatmulAttrs* attrs) {
- return MakeBatchMatmul(quantized_x, quantized_y, attrs->out_dtype);
+ ICHECK(attrs->transpose_a == false && attrs->transpose_b == true)
+ << "Currently qnn.batch_matmul only support NT format.";
Review comment:
I believe for enabling of transposed tensors for qnn.batch_matmul, it is
required to set correct axes in `BatchMatmulSecondTerm` and
`BatchMatmulThirdTerm` depending on attrs->transpose_a and attrs->transpose_b
correspondingly
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
