cfRod commented on a change in pull request #20482:
URL: https://github.com/apache/incubator-mxnet/pull/20482#discussion_r702690341
##########
File path: src/operator/operator_common.h
##########
@@ -620,6 +620,10 @@ class OpSignature {
hash = hash * 2 + arr.dtype();
eles.push_back(arr.dtype());
AddSign(arr.shape());
+#if DNNL_AARCH64_USE_ACL == 1
+ auto ival = reinterpret_cast<const uint64_t>(arr.storage_handle().dptr);
+ AddSign(ival);
+#endif
Review comment:
Hi @szha,
As mentioned we have an issue here #2026, where re-using primitives causes
accuracy failures for an Arm build of MxNET (which is built with Compute
Library for Arm). This is a fix we have added to other frameworks like
TensorFlow too, issue here
https://github.com/tensorflow/tensorflow/issues/47415 and PR raised here
https://github.com/tensorflow/tensorflow/commit/06dd30ef987b5511351e979568d4ad06ccdc2844.
Please also refer to the discussion with oneDNN team
https://github.com/oneapi-src/oneDNN/issues/1113
for why this fix is necessary. As stated previously, this is a workaround we
have had to add at the framework level until this issue can be resolved in
future releases of Compute Library.
--
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]
