[GitHub] [tvm] sfvaroglu commented on a diff in pull request #10952: [QNN] Support input scale and zp of 1-element vector in qnn.conv2d_transpose

[GitBox <gi...@apache.org>]

sfvaroglu commented on code in PR #10952:
URL: https://github.com/apache/tvm/pull/10952#discussion_r852397609


##########
python/tvm/relay/qnn/op/legalizations.py:
##########
@@ -91,12 +91,30 @@ def qnn_conv2d_transpose_legalize(attrs, inputs, types):
     # Collect the input exprs.
     data, kernel, input_zero_point, kernel_zero_point, _, _ = inputs
 
-    shift_data = relay.subtract(

Review Comment:
   Without this change, the model I'm looking at fails with `Incompatible broadcast type TensorType([32, 32, 3, 3], int16) and TensorType([32], int16)`. Here is the op for more info:
   ```
     %2181 = qnn.conv2d_transpose(%2180, meta[relay.Constant][1003] /* ty=Tensor[(32, 32, 3, 3), int8] */, -26 /* ty=int32 */, meta[relay.Constant][1004] /* ty=Tensor[(32), int32] */, 0.196949f /* ty=float32 */, meta[relay.Constant][1005] /* ty=Tensor[(32), float32] */, channels=32, kernel_size=[3, 3], strides=[2, 2], padding=[0, 0, 1, 1], kernel_layout="IOHW", out_dtype="int32") /* ty=Tensor[(1, 32, 100, 100), int32] */;
   ```



-- 
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: commits-unsubscribe@tvm.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org