Skip to content

O
opencv
Commit 1d01e435 authored by Tomoaki Teshima's avatar Tomoaki Teshima
Browse files
Options

use universal intrinsic in fast

parent bb5c40d9
Hide whitespace changes
Inline Side-by-side
Showing with 62 additions and 56 deletions
modules/features2d/src/fast.cpp
View file @ 1d01e435
......@@ -44,6 +44,7 @@ The references are:
#include "precomp.hpp"
#include "fast_score.hpp"
#include "opencl_kernels_features2d.hpp"
#include "opencv2/core/hal/intrin.hpp"
#include "opencv2/core/openvx/ovx_defs.hpp"
#if defined _MSC_VER
......@@ -58,9 +59,10 @@ void FAST_t(InputArray _img, std::vector<KeyPoint>& keypoints, int threshold, bo
{
Mat img = _img.getMat();
const int K = patternSize/2, N = patternSize + K + 1;
#if CV_SSE2
#if CV_SIMD128
const int quarterPatternSize = patternSize/4;
(void)quarterPatternSize;
v_uint8x16 delta = v_setall_u8(0x80), t = v_setall_u8((char)threshold), K16 = v_setall_u8((char)K);
bool hasSimd = hasSIMD128();
#endif
int i, j, k, pixel[25];
makeOffsets(pixel, (int)img.step, patternSize);
......@@ -69,12 +71,6 @@ void FAST_t(InputArray _img, std::vector<KeyPoint>& keypoints, int threshold, bo
threshold = std::min(std::max(threshold, 0), 255);
#if CV_SSE2
__m128i delta = _mm_set1_epi8(-128), t = _mm_set1_epi8((char)threshold), K16 = _mm_set1_epi8((char)K);
(void)K16;
(void)delta;
(void)t;
#endif
uchar threshold_tab[512];
for( i = -255; i <= 255; i++ )
threshold_tab[i+255] = (uchar)(i < -threshold ? 1 : i > threshold ? 2 : 0);
......@@ -99,66 +95,76 @@ void FAST_t(InputArray _img, std::vector<KeyPoint>& keypoints, int threshold, bo
if( i < img.rows - 3 )
{
j = 3;
#if CV_SSE2
if( patternSize == 16 )
#if CV_SIMD128
if( hasSimd )
{
for(; j < img.cols - 16 - 3; j += 16, ptr += 16)
if( patternSize == 16 )
{
__m128i m0, m1;
__m128i v0 = _mm_loadu_si128((const __m128i*)ptr);
__m128i v1 = _mm_xor_si128(_mm_subs_epu8(v0, t), delta);
v0 = _mm_xor_si128(_mm_adds_epu8(v0, t), delta);
__m128i x0 = _mm_sub_epi8(_mm_loadu_si128((const __m128i*)(ptr + pixel[0])), delta);
__m128i x1 = _mm_sub_epi8(_mm_loadu_si128((const __m128i*)(ptr + pixel[quarterPatternSize])), delta);
__m128i x2 = _mm_sub_epi8(_mm_loadu_si128((const __m128i*)(ptr + pixel[2*quarterPatternSize])), delta);
__m128i x3 = _mm_sub_epi8(_mm_loadu_si128((const __m128i*)(ptr + pixel[3*quarterPatternSize])), delta);
m0 = _mm_and_si128(_mm_cmpgt_epi8(x0, v0), _mm_cmpgt_epi8(x1, v0));
m1 = _mm_and_si128(_mm_cmpgt_epi8(v1, x0), _mm_cmpgt_epi8(v1, x1));
m0 = _mm_or_si128(m0, _mm_and_si128(_mm_cmpgt_epi8(x1, v0), _mm_cmpgt_epi8(x2, v0)));
m1 = _mm_or_si128(m1, _mm_and_si128(_mm_cmpgt_epi8(v1, x1), _mm_cmpgt_epi8(v1, x2)));
m0 = _mm_or_si128(m0, _mm_and_si128(_mm_cmpgt_epi8(x2, v0), _mm_cmpgt_epi8(x3, v0)));
m1 = _mm_or_si128(m1, _mm_and_si128(_mm_cmpgt_epi8(v1, x2), _mm_cmpgt_epi8(v1, x3)));
m0 = _mm_or_si128(m0, _mm_and_si128(_mm_cmpgt_epi8(x3, v0), _mm_cmpgt_epi8(x0, v0)));
m1 = _mm_or_si128(m1, _mm_and_si128(_mm_cmpgt_epi8(v1, x3), _mm_cmpgt_epi8(v1, x0)));
m0 = _mm_or_si128(m0, m1);
int mask = _mm_movemask_epi8(m0);
if( mask == 0 )
continue;
if( (mask & 255) == 0 )
for(; j < img.cols - 16 - 3; j += 16, ptr += 16)
{
j -= 8;
ptr -= 8;
continue;
}
v_uint8x16 v = v_load(ptr);
v_int8x16 v0 = v_reinterpret_as_s8((v + t) ^ delta);
v_int8x16 v1 = v_reinterpret_as_s8((v - t) ^ delta);
v_int8x16 x0 = v_reinterpret_as_s8(v_sub_wrap(v_load(ptr + pixel[0]), delta));
v_int8x16 x1 = v_reinterpret_as_s8(v_sub_wrap(v_load(ptr + pixel[quarterPatternSize]), delta));
v_int8x16 x2 = v_reinterpret_as_s8(v_sub_wrap(v_load(ptr + pixel[2*quarterPatternSize]), delta));
v_int8x16 x3 = v_reinterpret_as_s8(v_sub_wrap(v_load(ptr + pixel[3*quarterPatternSize]), delta));
v_int8x16 m0, m1;
m0 = (v0 < x0) & (v0 < x1);
m1 = (x0 < v1) & (x1 < v1);
m0 = m0 | ((v0 < x1) & (v0 < x2));
m1 = m1 | ((x1 < v1) & (x2 < v1));
m0 = m0 | ((v0 < x2) & (v0 < x3));
m1 = m1 | ((x2 < v1) & (x3 < v1));
m0 = m0 | ((v0 < x3) & (v0 < x0));
m1 = m1 | ((x3 < v1) & (x0 < v1));
m0 = m0 | m1;
int mask = v_signmask(m0);
if( mask == 0 )
continue;
if( (mask & 255) == 0 )
{
j -= 8;
ptr -= 8;
continue;
}
__m128i c0 = _mm_setzero_si128(), c1 = c0, max0 = c0, max1 = c0;
for( k = 0; k < N; k++ )
{
__m128i x = _mm_xor_si128(_mm_loadu_si128((const __m128i*)(ptr + pixel[k])), delta);
m0 = _mm_cmpgt_epi8(x, v0);
m1 = _mm_cmpgt_epi8(v1, x);
v_int8x16 c0 = v_setzero_s8();
v_int8x16 c1 = v_setzero_s8();
v_uint8x16 max0 = v_setzero_u8();
v_uint8x16 max1 = v_setzero_u8();
for( k = 0; k < N; k++ )
{
v_int8x16 x = v_reinterpret_as_s8(v_load((ptr + pixel[k])) ^ delta);
m0 = v0 < x;
m1 = x < v1;
c0 = _mm_and_si128(_mm_sub_epi8(c0, m0), m0);
c1 = _mm_and_si128(_mm_sub_epi8(c1, m1), m1);
c0 = v_sub_wrap(c0, m0) & m0;
c1 = v_sub_wrap(c1, m1) & m1;
max0 = _mm_max_epu8(max0, c0);
max1 = _mm_max_epu8(max1, c1);
}
max0 = v_max(max0, v_reinterpret_as_u8(c0));
max1 = v_max(max1, v_reinterpret_as_u8(c1));
}
max0 = _mm_max_epu8(max0, max1);
int m = _mm_movemask_epi8(_mm_cmpgt_epi8(max0, K16));
max0 = v_max(max0, max1);
int m = v_signmask(K16 < max0);
for( k = 0; m > 0 && k < 16; k++, m >>= 1 )
if(m & 1)
for( k = 0; m > 0 && k < 16; k++, m >>= 1 )
{
cornerpos[ncorners++] = j+k;
if(nonmax_suppression)
curr[j+k] = (uchar)cornerScore<patternSize>(ptr+k, pixel, threshold);
if(m & 1)
{
cornerpos[ncorners++] = j+k;
if(nonmax_suppression)
curr[j+k] = (uchar)cornerScore<patternSize>(ptr+k, pixel, threshold);
}
}
}
}
}
#endif
#endif
for( ; j < img.cols - 3; j++, ptr++ )
{
int v = ptr[0];
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment