From b2c7ccced1c3965bd52692917a8fdd31eddd15d0 Mon Sep 17 00:00:00 2001 From: Jinzhe Zeng Date: Fri, 10 Mar 2023 15:17:32 -0500 Subject: [PATCH] enable openmp for gelu Fix #2373. --- source/lib/src/gelu.cc | 3 +++ 1 file changed, 3 insertions(+) diff --git a/source/lib/src/gelu.cc b/source/lib/src/gelu.cc index c9ff7c9f2b..f352fcd1b5 100644 --- a/source/lib/src/gelu.cc +++ b/source/lib/src/gelu.cc @@ -6,6 +6,7 @@ template void deepmd::gelu_cpu(FPTYPE* out, const FPTYPE* xx, const int_64 size) { +#pragma omp parallel for for (int ii = 0; ii < size; ii++) { out[ii] = xx[ii] * (FPTYPE)0.5 * ((FPTYPE)1.0 + @@ -19,6 +20,7 @@ void deepmd::gelu_grad_cpu(FPTYPE* out, const FPTYPE* xx, const FPTYPE* dy, const int_64 size) { +#pragma omp parallel for for (int ii = 0; ii < size; ii++) { const FPTYPE var = tanh((FPTYPE)SQRT_2_PI * @@ -36,6 +38,7 @@ void deepmd::gelu_grad_grad_cpu(FPTYPE* out, const FPTYPE* dy, const FPTYPE* dy_2, const int_64 size) { +#pragma omp parallel for for (int ii = 0; ii < size; ii++) { const FPTYPE var1 = tanh((FPTYPE)SQRT_2_PI *