From 51477ab274bbb40be259844bdaab9685f693b028 Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Mon, 24 Jul 2017 23:40:23 +0000
Subject: [PATCH] Increased 2.3 times performance of Yolo on CPU by using OpenMP in both darknet_no_gpu and yolo_cpp_dll_no_gpu
---
build/darknet/yolo_cpp_dll_no_gpu.vcxproj | 1 +
src/gemm.c | 21 +++++++++++++--------
build/darknet/darknet_no_gpu.vcxproj | 2 ++
3 files changed, 16 insertions(+), 8 deletions(-)
diff --git a/build/darknet/darknet_no_gpu.vcxproj b/build/darknet/darknet_no_gpu.vcxproj
index 3f4681d..6da02a2 100644
--- a/build/darknet/darknet_no_gpu.vcxproj
+++ b/build/darknet/darknet_no_gpu.vcxproj
@@ -89,6 +89,7 @@
<AdditionalIncludeDirectories>C:\opencv_2.4.9\opencv\build\include;..\..\3rdparty\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
<PreprocessorDefinitions>_MBCS;OPENCV;_TIMESPEC_DEFINED;_CRT_SECURE_NO_WARNINGS;WIN32;NDEBUG;_CONSOLE;_LIB;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<UndefinePreprocessorDefinitions>CUDNN</UndefinePreprocessorDefinitions>
+ <OpenMPSupport>true</OpenMPSupport>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
@@ -133,6 +134,7 @@
<PrecompiledHeaderCompileAs>CompileAsCpp</PrecompiledHeaderCompileAs>
<CompileAs>Default</CompileAs>
<UndefinePreprocessorDefinitions>CUDNN</UndefinePreprocessorDefinitions>
+ <OpenMPSupport>true</OpenMPSupport>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
diff --git a/build/darknet/yolo_cpp_dll_no_gpu.vcxproj b/build/darknet/yolo_cpp_dll_no_gpu.vcxproj
index 96afb0d..2b746f5 100644
--- a/build/darknet/yolo_cpp_dll_no_gpu.vcxproj
+++ b/build/darknet/yolo_cpp_dll_no_gpu.vcxproj
@@ -139,6 +139,7 @@
<UndefinePreprocessorDefinitions>
</UndefinePreprocessorDefinitions>
<MultiProcessorCompilation>true</MultiProcessorCompilation>
+ <OpenMPSupport>true</OpenMPSupport>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
diff --git a/src/gemm.c b/src/gemm.c
index 3003be0..a4db8a4 100644
--- a/src/gemm.c
+++ b/src/gemm.c
@@ -151,14 +151,19 @@
C[i*ldc + j] *= BETA;
}
}
- if(!TA && !TB)
- gemm_nn(M, N, K, ALPHA,A,lda, B, ldb,C,ldc);
- else if(TA && !TB)
- gemm_tn(M, N, K, ALPHA,A,lda, B, ldb,C,ldc);
- else if(!TA && TB)
- gemm_nt(M, N, K, ALPHA,A,lda, B, ldb,C,ldc);
- else
- gemm_tt(M, N, K, ALPHA,A,lda, B, ldb,C,ldc);
+
+ int t;
+ #pragma omp parallel for
+ for (t = 0; t < M; ++t) {
+ if (!TA && !TB)
+ gemm_nn(1, N, K, ALPHA, A + t*lda, lda, B, ldb, C + t*ldc, ldc);
+ else if (TA && !TB)
+ gemm_tn(1, N, K, ALPHA, A + t, lda, B, ldb, C + t*ldc, ldc);
+ else if (!TA && TB)
+ gemm_nt(1, N, K, ALPHA, A + t*lda, lda, B, ldb, C + t*ldc, ldc);
+ else
+ gemm_tt(1, N, K, ALPHA, A + t, lda, B, ldb, C + t*ldc, ldc);
+ }
}
#ifdef GPU
--
Gitblit v1.10.0