ashutosh-arm commented on code in PR #13407: URL: https://github.com/apache/tvm/pull/13407#discussion_r1024260209
########## src/runtime/contrib/arm_compute_lib/acl_utils.cc: ########## @@ -130,6 +130,8 @@ arm_compute::DataType MakeACLDataType(const DLDataType& data_type) { return arm_compute::DataType::F32; } else if (data_type.code == DLDataTypeCode::kDLUInt && data_type.bits == 8) { return arm_compute::DataType::QASYMM8; + } else if (data_type.code == DLDataTypeCode::kDLInt && data_type.bits == 8) { + return arm_compute::DataType::QASYMM8_SIGNED; Review Comment: After discussing this with @lhutton1, it looks like some of the buffers need to be symmetric as described on the https://www.tensorflow.org/lite/performance/quantization_spec. For Conv2D, Depthwise, Fully Connected layers, this could come handy. I wonder if the ACL errors out if the incorrect type is passed to the end APIs. -- 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-unsubscr...@tvm.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org