saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-06-03 23:49:44 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

[QNN EP] Disable early termination in GetCapability (#18140)

Browse source 
[QNN EP] Disable early termination in GetCapability if there are
multiple partition and context binary enabled

### Description
QNN EP context binary cache feature only support single partition for
now. We have early termination in GetCapability.
After the PR https://github.com/microsoft/onnxruntime/pull/17764.
There's no Level 1 optimization any more for the 1st GetCapability.
Graph transformer EnsureUniqueDQForNodeUnit is not applied. So if
there's initializer -> DQ -> shared by multiple node unit. The node is
not identified as node unit group. QNN EP report many not supported
nodes because of this in the 1st GetCapability call.
The 2nd GetCapability still works normally.
Disable the early termination in GetCapability, delay the decision to
Compile.
This commit is contained in:
Hector Li 2023-10-30 08:34:49 -07:00 committed by GitHub
parent 9bba990871
commit be2f72a315
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 2 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

8
onnxruntime/core/providers/qnn/qnn_execution_provider.cc
View file

@ -404,10 +404,6 @@ QNNExecutionProvider::GetCapability(const onnxruntime::GraphViewer& graph_viewer
}
}
if (num_of_partitions > 1) {
ORT_ENFORCE(!context_cache_enabled_, "Only support single partition for context cache feature.");
}
const auto summary_msg = MakeString("Number of partitions supported by QNN EP: ", num_of_partitions,
", number of nodes in the graph: ", num_nodes_in_graph,
", number of nodes supported by QNN: ", num_of_supported_nodes);
@ -485,7 +481,7 @@ Status QNNExecutionProvider::Compile(const std::vector<FusedNodeAndGraph>& fused
bool is_ctx_file_exist = qnn_cache_model_handler_->GetIsContextCacheFileExists();
if (is_qnn_ctx_model || (context_cache_enabled_ && is_ctx_file_exist)) {
ORT_ENFORCE(fused_nodes_and_graphs.size() == 1, "Only support single partition for context cache feature.");
ORT_RETURN_IF(fused_nodes_and_graphs.size() != 1, "Only support single partition for context cache feature.");
std::unique_ptr<qnn::QnnModel> qnn_model = std::make_unique<qnn::QnnModel>(logger, qnn_backend_manager_.get());
// Load and execute from cached context if exist
ORT_RETURN_IF_ERROR(qnn_cache_model_handler_->LoadQnnCtxFromOnnxModel(graph_viewer,
@ -509,7 +505,7 @@ Status QNNExecutionProvider::Compile(const std::vector<FusedNodeAndGraph>& fused
ORT_RETURN_IF_ERROR(CompileFromOrtGraph(fused_nodes_and_graphs, node_compute_funcs, logger));
if (context_cache_enabled_ && !is_qnn_ctx_model) {
ORT_ENFORCE(fused_nodes_and_graphs.size() == 1, "Only support single partition for context cache feature.");
ORT_RETURN_IF(fused_nodes_and_graphs.size() != 1, "Only support single partition for context cache feature.");
uint64_t buffer_size(0);
auto context_buffer = qnn_backend_manager_->GetContextBinaryBuffer(buffer_size);
ORT_RETURN_IF_ERROR(qnn_cache_model_handler_->GenerateCtxCacheOnnxModel(context_buffer.get(),