potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-30 15:53:03 +04:00
Code Packages Releases Wiki Activity

Provide a few AVX512 optimized functions for the DNN module

Browse Source 
This patch adds AVX512 optimized fastConv as well as the hookups
needed to get these called in the convolution_layer.

AVX512 fastConv is code-identical on a C level to the AVX2 one,
but is measurably faster due to AVX512 having more registers available
to cache results in.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
This commit is contained in:
Arjan van de Ven
2017-12-25 21:08:22 +00:00
parent fc8e848a54
commit 2938860b3f
4 changed files with 101 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/dnn/src/layers/fully_connected_layer.cpp
+8 -1
View File
@@ -139,7 +139,7 @@ public:
class FullyConnected : public ParallelLoopBody
{
public:
FullyConnected() : srcMat(0), weights(0), biasMat(0), activ(0), dstMat(0), nstripes(0), useAVX(false), useAVX2(false) {}
FullyConnected() : srcMat(0), weights(0), biasMat(0), activ(0), dstMat(0), nstripes(0), useAVX(false), useAVX2(false), useAVX512(false) {}
static void run(const Mat& srcMat, const Mat& weights, const Mat& biasMat,
Mat& dstMat, const ActivationLayer* activ, int nstripes)
@@ -161,6 +161,7 @@ public:
p.activ = activ;
p.useAVX = checkHardwareSupport(CPU_AVX);
p.useAVX2 = checkHardwareSupport(CPU_AVX2);
p.useAVX512 = checkHardwareSupport(CPU_AVX_512DQ);
parallel_for_(Range(0, nstripes), p, nstripes);
}
@@ -195,6 +196,11 @@ public:
memcpy(sptr, sptr_, vecsize*sizeof(sptr[0]));
#if CV_TRY_AVX512
if( useAVX512 )
opt_AVX512::fastGEMM1T( sptr, wptr, wstep, biasptr, dptr, nw, vecsize);
else
#endif
#if CV_TRY_AVX2
if( useAVX2 )
opt_AVX2::fastGEMM1T( sptr, wptr, wstep, biasptr, dptr, nw, vecsize);
@@ -255,6 +261,7 @@ public:
int nstripes;
bool useAVX;
bool useAVX2;
bool useAVX512;
};
#ifdef HAVE_OPENCL