nuoya
/
mindspore2022
forked from huawei/mindspore2022
0
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

!25255 fix the weight update bug in the shared weight scene 

Merge pull request !25255 from limingqi107/r1.5_lmq
This commit is contained in:
i-robot 2021-10-21 12:10:37 +00:00 committed by Gitee
parent 09f5aa0af1 aba9ba4ef0
commit b98b46cddc
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
mindspore/ccsrc/runtime/framework/actor/data_prepare_actor.cc
View File

@ -421,7 +421,6 @@ void DataPrepareActor::PrepareDataForWeightNode(const AnfNodePtr &backend_node,
UpdateRefCount(host_tensor_address.get(), true);
}
MS_EXCEPTION_IF_NULL(host_tensor_address);
DeviceTensorStore::GetInstance().Insert(front_node.get(), host_tensor_address);
if (host_tensor_address->DeviceType() == device_tensor->DeviceType()) {
AnfAlgo::SetOutputAddr(host_tensor_address, 0, backend_node.get());
} else {
@ -429,6 +428,9 @@ void DataPrepareActor::PrepareDataForWeightNode(const AnfNodePtr &backend_node,
<< ", device tensor type:" << device_tensor->DeviceType();
}
}
// Maybe the same host_tensor_address corresponds to the different front_node in shared weight scene,
// so need update the device tensor store always.
DeviceTensorStore::GetInstance().Insert(front_node.get(), host_tensor_address);
// If the ptr of device tensor is not nullptr, it indicates that the device data has been prepared.
MS_EXCEPTION_IF_NULL(host_tensor_address);