Skip to content

Qualcomm AI Engine Direct - Fix mem_handel register twice issue #13410

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Aug 26, 2025
Merged

Qualcomm AI Engine Direct - Fix mem_handel register twice issue #13410

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion backends/qualcomm/runtime/QnnManager.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -291,7 +291,8 @@ Error QnnManager::RegisterCustomMem(
data_ptr,
unaligned_custom_mem_base,
total_custom_mem_size,
tensor_offset) == Error::Ok,
tensor_offset,
info) == Error::Ok,
Internal,
"Fail to register to shared memory.");

Expand Down
6 changes: 5 additions & 1 deletion backends/qualcomm/runtime/backends/QnnMemManager.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,13 +56,16 @@ Error QnnMemManager::RegisterIonMem(
return Error::Ok;
}

// TODO: Find a better way to unify RegisterCustomMem and
// PreRegisterCustomMemHandle
Error QnnMemManager::RegisterCustomMem(
const std::shared_ptr<TensorWrapper>& tensor_wrapper,
int32_t mem_fd,
void* mem_ptr,
void* unaligned_custom_mem_base,
size_t total_custom_mem_size,
size_t tensor_offset) {
size_t tensor_offset,
const CustomMemTensorInfo& info) {
const QnnInterface& qnn_interface = implementation_.GetQnnInterface();
Qnn_MemDescriptor_t descriptor = {
{tensor_wrapper->GetRank(), tensor_wrapper->GetDims(), nullptr},
Expand Down Expand Up @@ -94,6 +97,7 @@ Error QnnMemManager::RegisterCustomMem(
return Error::Internal;
}
tensor_wrapper->SetMemHandle(handle);
pre_registered_handles_.insert({info, handle});
registered_map_.insert({handle, mem_ptr});
if (log_level_ >= QnnExecuTorchLogLevel::kLogLevelInfo) {
QNN_EXECUTORCH_LOG_INFO(
Expand Down
5 changes: 4 additions & 1 deletion backends/qualcomm/runtime/backends/QnnMemManager.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,7 +41,8 @@ class QnnMemManager {
void* mem_ptr,
void* unaligned_custom_mem_base,
size_t total_custom_mem_size,
size_t tensor_offset);
size_t tensor_offset,
const CustomMemTensorInfo& info);

// Pre-register custom mem handle from SharedBuffer. Bring forward the
// memHandle creating time from execution to initialization.
Expand All @@ -67,7 +68,9 @@ class QnnMemManager {
const QnnImplementation& implementation_;
QnnContext* context_;
QnnExecuTorchLogLevel log_level_;
// Store the registered Qnn_MemHandle_t for de-registration
std::unordered_map<Qnn_MemHandle_t, void*> registered_map_;
// Store the pre-registered custom mem handles
std::unordered_map<CustomMemTensorInfo, void*> pre_registered_handles_;
std::unordered_map<executorch::aten::ScalarType, Qnn_DataType_t>
scalar_type_to_qnn_dtype_ = {
Expand Down
Loading