From e37444e768679c5e43f21c73f8587fcfd8d1500a Mon Sep 17 00:00:00 2001
From: Olivier BICHLER <olivier.bichler@cea.fr>
Date: Mon, 28 Apr 2025 15:06:20 +0200
Subject: [PATCH] Added _OPENMP guards
---
aidge_export_cpp/kernels/convolution.hpp | 4 ++++
aidge_export_cpp/kernels/fullyconnected.hpp | 5 +++++
aidge_export_cpp/kernels/leakyrelu.hpp | 2 ++
aidge_export_cpp/kernels/pooling.hpp | 2 ++
4 files changed, 13 insertions(+)
diff --git a/aidge_export_cpp/kernels/convolution.hpp b/aidge_export_cpp/kernels/convolution.hpp
index 40f22c6..7df55ab 100644
--- a/aidge_export_cpp/kernels/convolution.hpp
+++ b/aidge_export_cpp/kernels/convolution.hpp
@@ -48,7 +48,9 @@ void convolution_forward(
0, DILATED_KERNEL_HEIGHT);
const int iy = (oy * STRIDE_Y) - PADDING_Y;
+#ifdef _OPENMP
#pragma omp parallel for collapse(2)
+#endif
for (int ox = 0; ox < OUTPUTS_WIDTH; ++ox) {
for (int output = 0; output < NB_OUTPUTS; ++output) {
// moved to inner loop for collapsing -->
@@ -200,7 +202,9 @@ void convolution_depthwise_forward(
0, DILATED_KERNEL_HEIGHT);
const int iy = (oy * STRIDE_Y) - PADDING_Y;
+#ifdef _OPENMP
#pragma omp parallel for collapse(2)
+#endif
for (int ox = 0; ox < OUTPUTS_WIDTH; ++ox) {
for (int output = 0; output < NB_OUTPUTS; ++output) {
// moved to inner loop for collapsing -->
diff --git a/aidge_export_cpp/kernels/fullyconnected.hpp b/aidge_export_cpp/kernels/fullyconnected.hpp
index 2780de2..60805e7 100644
--- a/aidge_export_cpp/kernels/fullyconnected.hpp
+++ b/aidge_export_cpp/kernels/fullyconnected.hpp
@@ -28,6 +28,9 @@ void fullyconnected_forward (
// It is only an issue if the FC was after a flatten layer.
// Otherwise it is not an issue for the other FC because CHANNELS_WIDTH = CHANNELS_HEIGHT = 1
// Solution: Add a system to check dataformat
+#ifdef _OPENMP
+#pragma omp parallel for
+#endif
for (int och = 0; och < NB_OUTPUTS; och++) {
Bias_T weightedSum = (biases) ? biases[och] : Bias_T(0);
@@ -45,7 +48,9 @@ void fullyconnected_forward (
}
/*
Here the kernel to use with inputs in NHWC and weights in NHWC
+#ifdef _OPENMP
#pragma omp parallel for
+#endif
for (int och = 0; och < NB_OUTPUTS; och++) {
Bias_T weightedSum = (biases) ? biases[och] : Bias_T(0);
diff --git a/aidge_export_cpp/kernels/leakyrelu.hpp b/aidge_export_cpp/kernels/leakyrelu.hpp
index 07352cd..5e6598d 100644
--- a/aidge_export_cpp/kernels/leakyrelu.hpp
+++ b/aidge_export_cpp/kernels/leakyrelu.hpp
@@ -11,7 +11,9 @@ void leakyrelu_forward (
Output_T* __restrict outputs,
const float negative_slope)
{
+#ifdef _OPENMP
#pragma omp parallel for
+#endif
for (int i = 0; i < NB_DATA; ++i) {
if (inputs[i] >= 0) {
outputs[i] = inputs[i];
diff --git a/aidge_export_cpp/kernels/pooling.hpp b/aidge_export_cpp/kernels/pooling.hpp
index a86fd41..30fa766 100644
--- a/aidge_export_cpp/kernels/pooling.hpp
+++ b/aidge_export_cpp/kernels/pooling.hpp
@@ -36,7 +36,9 @@ void pooling_forward(
0, POOL_HEIGHT);
const int iy = (oy * STRIDE_Y) - PADDING_Y;
+#ifdef _OPENMP
#pragma omp parallel for collapse(2)
+#endif
for (int ox = 0; ox < OUTPUTS_WIDTH; ++ox) {
for (int output = 0; output < NB_OUTPUTS; ++output) {
// moved to inner loop for collapsing -->
--
GitLab