From 1412d3bc2f731f33279c27b229e84b804e247760 Mon Sep 17 00:00:00 2001
From: arlesniak <artur.lesniak@intel.com>
Date: Fri, 9 Jul 2021 11:09:32 +0200
Subject: [PATCH] Use CBLAS for SelectedRows elementwise add operation.
 (#34008)

* Use CBLAS for SelectedRows elementwise add operation. It's faster.

* template compilation fix

* reverted template compilation fix

* slimmed template compilation fix

Co-authored-by: Adam Osewski <adam.osewski@intel.com>
---
 .../operators/math/selected_rows_functor.cc   | 21 +++++++------------
 1 file changed, 7 insertions(+), 14 deletions(-)

diff --git a/paddle/fluid/operators/math/selected_rows_functor.cc b/paddle/fluid/operators/math/selected_rows_functor.cc
index a72bdec05d7..757cac4e4ff 100644
--- a/paddle/fluid/operators/math/selected_rows_functor.cc
+++ b/paddle/fluid/operators/math/selected_rows_functor.cc
@@ -300,33 +300,26 @@ template struct SelectedRowsAddToTensor<platform::CPUDeviceContext,
 // add or mul.
 namespace scatter {
 
-#ifdef PADDLE_WITH_MKLDNN
 template <typename T>
-typename std::enable_if<std::is_same<T, float>::value ||
-                        std::is_same<T, platform::bfloat16>::value>::type
+typename std::enable_if<std::is_same<T, platform::bfloat16>::value>::type
 elementwise_add_to(BlasT<platform::CPUDeviceContext, T>* blas, size_t data_len,
                    const T* in, T* out) {
+#ifdef PADDLE_WITH_MKLDNN
   onednn_handler_axpy(data_len, T(1.f), in, out);
-}
-
-template <typename T>
-typename std::enable_if<std::is_same<T, double>::value ||
-                        std::is_same<T, platform::complex<float>>::value ||
-                        std::is_same<T, platform::complex<double>>::value>::type
-elementwise_add_to(BlasT<platform::CPUDeviceContext, T>* blas, size_t data_len,
-                   const T* in, T* out) {
+#else
   blas->AXPY(data_len, T(1.f), in, out);
+#endif
 }
-#else
+
 template <typename T>
-typename std::enable_if<std::is_floating_point<T>::value ||
+typename std::enable_if<std::is_same<T, float>::value ||
+                        std::is_same<T, double>::value ||
                         std::is_same<T, platform::complex<float>>::value ||
                         std::is_same<T, platform::complex<double>>::value>::type
 elementwise_add_to(BlasT<platform::CPUDeviceContext, T>* blas, size_t data_len,
                    const T* in, T* out) {
   blas->AXPY(data_len, T(1.f), in, out);
 }
-#endif
 
 template <typename T>
 typename std::enable_if<std::is_integral<T>::value>::type elementwise_add_to(
-- 
GitLab