Skip to content
This repository was archived by the owner on Jan 13, 2024. It is now read-only.

Use openmp to parallelize QLinearConv #276

Merged
sdpython merged 2 commits into from
Jun 25, 2021
Merged

Use openmp to parallelize QLinearConv #276

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7dfc8d2
Use openmp to parallelize QLinearConv
sdpython Jun 25, 2021
eb1c9f7
Merge branch 'master' of https://github.com/sdpython/mlprodict into omp
sdpython Jun 25, 2021
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
12 changes: 6 additions & 6 deletions mlprodict/onnxrt/ops_cpu/op_qlinear_conv_.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -344,17 +344,17 @@ template <typename T1, typename T2, typename T3 = T1, typename T4 = int32_t,
static_cast<double>(kernel_dim);

// OMP
#if false //USE_OPENMP
#if USE_OPENMP
int32_t thread_count = maximum_thread_count;
if (complexity < thread_complexity * maximum_thread_count)
thread_count = static_cast<int32_t>(complexity / thread_complexity) + 1;
// Ensure that every thread produces at least one output.
if (thread_count > output_image_size)
thread_count = static_cast<int32_t>(output_image_size);
thread_count = std::min(thread_count, ::omp_get_max_threads());
#else
#else
int32_t thread_count = 1;
#endif
#endif

for (int64_t image_id = 0; image_id < N; ++image_id) {
const T1* input_data = Xdata;
Expand Down Expand Up @@ -391,9 +391,9 @@ template <typename T1, typename T2, typename T3 = T1, typename T4 = int32_t,
}
}

#if false // USE_OPENMP
#pragma omp parallel for
#endif
#if USE_OPENMP
#pragma omp parallel for
#endif
for (int32_t batch_idx = 0; batch_idx < thread_count; ++batch_idx) {
int64_t output_start, output_end;
std::ptrdiff_t work_per_batch = output_image_size / thread_count;
Expand Down