From 7dfc8d27637beb5309c64a0e4599912f9d2c5307 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?xavier=20dupr=C3=A9?= Date: Fri, 25 Jun 2021 13:24:21 +0200 Subject: [PATCH] Use openmp to parallelize QLinearConv --- mlprodict/onnxrt/ops_cpu/op_qlinear_conv_.hpp | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/mlprodict/onnxrt/ops_cpu/op_qlinear_conv_.hpp b/mlprodict/onnxrt/ops_cpu/op_qlinear_conv_.hpp index a23a162e3..dd00bca66 100644 --- a/mlprodict/onnxrt/ops_cpu/op_qlinear_conv_.hpp +++ b/mlprodict/onnxrt/ops_cpu/op_qlinear_conv_.hpp @@ -344,7 +344,7 @@ template (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(complexity / thread_complexity) + 1; @@ -352,9 +352,9 @@ template output_image_size) thread_count = static_cast(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; @@ -391,9 +391,9 @@ template