Skip to content

Add 1MB temp allocator for executor #13533

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 4 commits into from
Aug 26, 2025
Merged

Add 1MB temp allocator for executor #13533

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
fea6359
migrate most aten ops tests to facto (#13483)
zonglinpeng Aug 20, 2025
f5f3524
fix MM nullptr from zero bias (#13523)
zonglinpeng Aug 20, 2025
3d6b898
Add 1MB temp allocator for executor (#13533)
zonglinpeng Aug 20, 2025
ef81355
Merge branch 'main' into export-D80578578
zonglinpeng Aug 26, 2025
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
7 changes: 5 additions & 2 deletions backends/cadence/fusion_g3/operators/op_native_layer_norm.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -221,12 +221,15 @@ std::tuple<Tensor&, Tensor&, Tensor&> native_layer_norm_out(
num_elm *= normalized_shape[i];
}

constexpr size_t kAlignment =
16; // 16-byte alignment for vectorized operations

float* weight_data;
if (weight.has_value()) {
weight_data = weight.value().mutable_data_ptr<float>();
} else {
executorch::runtime::Result<void*> temp_mem_weight =
ctx.allocate_temp(num_elm * sizeof(float));
ctx.allocate_temp(num_elm * sizeof(float), kAlignment);
weight_data = (float*)(temp_mem_weight.get());

for (int i = 0; i < num_elm; i++) {
Expand All @@ -238,7 +241,7 @@ std::tuple<Tensor&, Tensor&, Tensor&> native_layer_norm_out(
bias_data = bias.value().mutable_data_ptr<float>();
} else {
executorch::runtime::Result<void*> temp_mem_bias =
ctx.allocate_temp(num_elm * sizeof(float));
ctx.allocate_temp(num_elm * sizeof(float), kAlignment);
bias_data = (float*)(temp_mem_bias.get());

for (int i = 0; i < num_elm; i++) {
Expand Down
14 changes: 11 additions & 3 deletions backends/cadence/hifi/kernels/kernels.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,11 +21,19 @@ memcpy(void* dst, const void* src, size_t num_bytes) {
}

void* allocate_temp_memory(KernelRuntimeContext& ctx, size_t size) {
ET_LOG(Info, "Attempting to allocate %zu bytes of temp memory", size);
Result<void*> temp_mem_res = ctx.allocate_temp(size);
constexpr size_t kAlignment =
16; // 16-byte alignment for vectorized operations
ET_LOG(
Info,
"Attempting to allocate %zu bytes of temp memory (16-byte aligned)",
size);
Result<void*> temp_mem_res = ctx.allocate_temp(size, kAlignment);
if (temp_mem_res.ok()) {
void* ptr = temp_mem_res.get();
ET_LOG(Info, "Successfully allocated temp memory at %p", ptr);
ET_LOG(
Info,
"Successfully allocated temp memory at %p (16-byte aligned)",
ptr);
return ptr;
} else {
ET_LOG(
Expand Down
Loading