Chiebot-Mirror
/
opencv
mirror of https://github.com/opencv/opencv.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

replace v_add with +

Browse Source
pull/22440/head
Zihao Mu 2 years ago
parent 2619099fe5
commit 2cd7e17b65
2 changed files with 35 additions and 34 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 13
      modules/dnn/src/layers/fast_convolution/fast_convolution.cpp
  2. 56
      modules/dnn/src/layers/fast_convolution/fast_convolution.simd.hpp

13
modules/dnn/src/layers/fast_convolution/fast_convolution.cpp
Unescape View File

@ -502,10 +502,11 @@ void runFastConv2d(InputArray _input, OutputArray _output, const Ptr<FastConv2d>
{
for (; j + 7 < out_width; j += 8)
{
v_float32x4 v0 = v_add(v_load(cptr + j), vbias);
v_float32x4 v1 = v_add(v_load(cptr + j + 4), vbias);
v0 = v_add(v0, v_load(pbptr + j));
v1 = v_add(v1, v_load(pbptr + j + 4));
v_float32x4 v0 = v_load(cptr + j) + vbias;
v_float32x4 v1 = v_load(cptr + j + 4) + vbias;
v0 += v_load(pbptr + j);
v1 += v_load(pbptr + j + 4);
if (ifMinMaxAct)
{
@ -521,8 +522,8 @@ void runFastConv2d(InputArray _input, OutputArray _output, const Ptr<FastConv2d>
{
for (; j + 7 < out_width; j += 8)
{
v_float32x4 v0 = v_add(v_load(cptr + j), vbias);
v_float32x4 v1 = v_add(v_load(cptr + j + 4), vbias);
v_float32x4 v0 = v_load(cptr + j) + vbias;
v_float32x4 v1 = v_load(cptr + j + 4) + vbias;
if (ifMinMaxAct)
{

56
modules/dnn/src/layers/fast_convolution/fast_convolution.simd.hpp
Unescape View File

@ -13,7 +13,7 @@ namespace dnn {
void convBlock(int np, const float* a, const float* b, float* c, int ldc, bool init_c)
{
#if 0 // CV_SIMD128 && CONV_MR == 4 && CONV_NR == 24
#if CV_SIMD128 && CONV_MR == 4 && CONV_NR == 24
v_float32x4 c0 = v_setzero_f32(), c1 = c0, c2 = c0, c3 = c0, c4 = c0, c5 = c0;
v_float32x4 c6 = v_setzero_f32(), c7 = c6, c8 = c6, c9 = c6, c10 = c6, c11 = c6;
v_float32x4 c12 = v_setzero_f32(), c13 = c12, c14 = c12, c15 = c12, c16 = c12, c17 = c12;
@ -59,33 +59,33 @@ void convBlock(int np, const float* a, const float* b, float* c, int ldc, bool i
if (!init_c)
{
c0 = v_add(c0, v_load(c));
c1 = v_add(c1, v_load(c + 4));
c2 = v_add(c2, v_load(c + 8));
c3 = v_add(c3, v_load(c + 12));
c4 = v_add(c4, v_load(c + 16));
c5 = v_add(c5, v_load(c + 20));
c6 = v_add(c6 , v_load(c + ldc));
c7 = v_add(c7 , v_load(c + ldc + 4));
c8 = v_add(c8 , v_load(c + ldc + 8));
c9 = v_add(c9 , v_load(c + ldc + 12));
c10 = v_add(c10, v_load(c + ldc + 16));
c11 = v_add(c11, v_load(c + ldc + 20));
c12 = v_add(c12, v_load(c + ldc*2));
c13 = v_add(c13, v_load(c + ldc*2 + 4));
c14 = v_add(c14, v_load(c + ldc*2 + 8));
c15 = v_add(c15, v_load(c + ldc*2 + 12));
c16 = v_add(c16, v_load(c + ldc*2 + 16));
c17 = v_add(c17, v_load(c + ldc*2 + 20));
c18 = v_add(c18, v_load(c + ldc*3));
c19 = v_add(c19, v_load(c + ldc*3 + 4));
c20 = v_add(c20, v_load(c + ldc*3 + 8));
c21 = v_add(c21, v_load(c + ldc*3 + 12));
c22 = v_add(c22, v_load(c + ldc*3 + 16));
c23 = v_add(c23, v_load(c + ldc*3 + 20));
c0 += v_load(c);
c1 += v_load(c + 4);
c2 += v_load(c + 8);
c3 += v_load(c + 12);
c4 += v_load(c + 16);
c5 += v_load(c + 20);
c6 += v_load(c + ldc);
c7 += v_load(c + ldc + 4);
c8 += v_load(c + ldc + 8);
c9 += v_load(c + ldc + 12);
c10 += v_load(c + ldc + 16);
c11 += v_load(c + ldc + 20);
c12 += v_load(c + ldc*2);
c13 += v_load(c + ldc*2 + 4);
c14 += v_load(c + ldc*2 + 8);
c15 += v_load(c + ldc*2 + 12);
c16 += v_load(c + ldc*2 + 16);
c17 += v_load(c + ldc*2 + 20);
c18 += v_load(c + ldc*3);
c19 += v_load(c + ldc*3 + 4);
c20 += v_load(c + ldc*3 + 8);
c21 += v_load(c + ldc*3 + 12);
c22 += v_load(c + ldc*3 + 16);
c23 += v_load(c + ldc*3 + 20);
}
v_store(c, c0);

Write Preview
Loading…
Cancel
Save