commit 48b3ef89de5f21a0169d8416e3d54081f82c7bf8
author Georgios Pinitas <georgios.pinitas@arm.com> Mon Oct 14 19:03:09 2019 +0100
committer Georgios Pinitas <georgios.pinitas@arm.com> Wed Oct 23 12:08:12 2019 +0000
tree f857d733ccf446c704823dc7ac796a96eb55095e
parent 1dce3101ef8d77c8cf0af7dfd4af6595a0136b91

COMPMID-2577: Fuse bias addition and activation in gemm assembly kernels

Change-Id: I7f52112d2d05b1ea3d3f3d4b19b8eafab05d6c44
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/2141
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Pablo Marquez <pablo.tello@arm.com>

src/core/NEON/kernels/assembly/Helpers.cpp

diff --git a/src/core/NEON/kernels/assembly/Helpers.cpp b/src/core/NEON/kernels/assembly/Helpers.cpp
index 3d8d66d..93ea6c8 100644
--- a/src/core/NEON/kernels/assembly/Helpers.cpp
+++ b/src/core/NEON/kernels/assembly/Helpers.cpp
@@ -24,16 +24,13 @@
 
 #include "arm_compute/core/NEON/kernels/assembly/Helpers.h"
 
-#include "arm_compute/core/NEON/kernels/assembly/arm_gemm.hpp"
-
 namespace arm_compute
 {
 arm_gemm::KernelDescription get_gemm_info(DataType                            input_type,
                                           const CPUInfo                      &ci,
                                           const unsigned int                  num_threads,
                                           const INEGEMMWrapperKernel::Params &p,
-                                          float                               alpha,
-                                          float                               beta,
+                                          arm_gemm::Activation                activation,
                                           bool                                pretranspose_hint)
 {
     switch(input_type)
@@ -42,25 +39,25 @@
         case DataType::QASYMM8:
         case DataType::U8:
         {
-            arm_gemm::GemmArgs<uint32_t> args(&ci, p.M, p.N, p.K, p.batches, p.multis, false, false, alpha, beta, num_threads, pretranspose_hint);
+            arm_gemm::GemmArgs args(&ci, p.M, p.N, p.K, p.batches, p.multis, false, false, activation, num_threads, pretranspose_hint);
             return arm_gemm::get_gemm_method<uint8_t, uint32_t>(args);
         }
         case DataType::S8:
         {
-            arm_gemm::GemmArgs<int32_t> args(&ci, p.M, p.N, p.K, p.batches, p.multis, false, false, alpha, beta, num_threads, pretranspose_hint);
+            arm_gemm::GemmArgs args(&ci, p.M, p.N, p.K, p.batches, p.multis, false, false, activation, num_threads, pretranspose_hint);
             return arm_gemm::get_gemm_method<int8_t, int32_t>(args);
         }
 #endif // __aarch64__
 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
         case DataType::F16:
         {
-            arm_gemm::GemmArgs<__fp16> args(&ci, p.M, p.N, p.K, p.batches, p.multis, false, false, alpha, beta, num_threads, pretranspose_hint);
+            arm_gemm::GemmArgs args(&ci, p.M, p.N, p.K, p.batches, p.multis, false, false, activation, num_threads, pretranspose_hint);
             return arm_gemm::get_gemm_method<__fp16, __fp16>(args);
         }
 #endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
         case DataType::F32:
         {
-            arm_gemm::GemmArgs<float> args(&ci, p.M, p.N, p.K, p.batches, p.multis, false, false, alpha, beta, num_threads, pretranspose_hint);
+            arm_gemm::GemmArgs args(&ci, p.M, p.N, p.K, p.batches, p.multis, false, false, activation, num_threads, pretranspose_hint);
             return arm_gemm::get_gemm_method<float, float>(args);
         }
         default: