Skip to content

Commit

Permalink
Fix Memcpy transformer when dealing multiple EPs (microsoft#22413)
Browse files Browse the repository at this point in the history
### Description

Fix Memcpy transformer when dealing multiple EPs.

---------

Co-authored-by: Scott McKay <[email protected]>
Co-authored-by: Scott McKay <[email protected]>
  • Loading branch information
3 people authored Oct 16, 2024
1 parent f610605 commit c3a94c6
Showing 1 changed file with 0 additions and 7 deletions.
7 changes: 0 additions & 7 deletions onnxruntime/core/optimizer/transformer_memcpy.cc
Original file line number Diff line number Diff line change
Expand Up @@ -256,13 +256,6 @@ void TransformerMemcpyImpl::ProcessDefs(onnxruntime::Node& node, const KernelReg
}
} else if (node_provider_type != kCudaExecutionProvider && node_provider_type != kTensorrtExecutionProvider &&
node_provider_type != kRocmExecutionProvider && node_provider_type != kMIGraphXExecutionProvider) {
// TODO: copy between devices? i.e. multiple GPUs
if (node_provider_type != onnxruntime::kCpuExecutionProvider &&
node_provider_type != onnxruntime::kVitisAIExecutionProvider &&
!node_provider_type.empty()) {
ORT_THROW("Execution type '", node_provider_type, "' doesn't support memcpy ");
}

for (const auto* arg : node.InputDefs()) {
if (arg->Exists())
non_provider_input_defs_.insert(arg);
Expand Down

0 comments on commit c3a94c6

Please sign in to comment.