Skip to content

O
opencv
Commit ea222877 authored by Alexander Alekhin's avatar Alexander Alekhin
Browse files
Options

Merge pull request #2881 from mlyashko:moments_opt

parents ea1b14ee 44ffa420
Show whitespace changes
Inline Side-by-side
Showing with 108 additions and 62 deletions
modules/imgproc/src/moments.cpp
View file @ ea222877
...@@ -369,11 +369,16 @@ Moments::Moments( double _m00, double _m10, double _m01, double _m20, double _m1 ...@@ -369,11 +369,16 @@ Moments::Moments( double _m00, double _m10, double _m01, double _m20, double _m1
#ifdef HAVE_OPENCL #ifdef HAVE_OPENCL
static bool ocl_moments( InputArray _src, Moments& m) static bool ocl_moments( InputArray _src, Moments& m, bool binary)
{ {
const int TILE_SIZE = 32; const int TILE_SIZE = 32;
const int K = 10; const int K = 10;
ocl::Kernel k("moments", ocl::imgproc::moments_oclsrc, format("-D TILE_SIZE=%d", TILE_SIZE));
ocl::Kernel k = ocl::Kernel("moments", ocl::imgproc::moments_oclsrc,
format("-D TILE_SIZE=%d%s",
TILE_SIZE,
binary ? " -D OP_MOMENTS_BINARY" : ""));
if( k.empty() ) if( k.empty() )
return false; return false;
...@@ -451,8 +456,8 @@ cv::Moments cv::moments( InputArray _src, bool binary ) ...@@ -451,8 +456,8 @@ cv::Moments cv::moments( InputArray _src, bool binary )
return m; return m;
#ifdef HAVE_OPENCL #ifdef HAVE_OPENCL
if( !(ocl::useOpenCL() && type == CV_8UC1 && !binary && if( !(ocl::useOpenCL() && type == CV_8UC1 &&
_src.isUMat() && ocl_moments(_src, m)) ) _src.isUMat() && ocl_moments(_src, m, binary)) )
#endif #endif
{ {
Mat mat = _src.getMat(); Mat mat = _src.getMat();
......
modules/imgproc/src/opencl/moments.cl
View file @ ea222877
...@@ -4,6 +4,7 @@ ...@@ -4,6 +4,7 @@
#error "TILE SIZE should be 32" #error "TILE SIZE should be 32"
#endif #endif
__kernel void moments(__global const uchar* src, int src_step, int src_offset, __kernel void moments(__global const uchar* src, int src_step, int src_offset,
int src_rows, int src_cols, __global int* mom0, int xtiles) int src_rows, int src_cols, __global int* mom0, int xtiles)
{ {
...@@ -14,78 +15,127 @@ __kernel void moments(__global const uchar* src, int src_step, int src_offset, ...@@ -14,78 +15,127 @@ __kernel void moments(__global const uchar* src, int src_step, int src_offset,
int ypix = y0*TILE_SIZE + y; int ypix = y0*TILE_SIZE + y;
__local int mom[TILE_SIZE][10]; __local int mom[TILE_SIZE][10];
if( x_min < src_cols && y0*TILE_SIZE < src_rows ) if (x_min < src_cols && y0*TILE_SIZE < src_rows)
{ {
if( ypix < src_rows ) if (ypix < src_rows)
{ {
int x_max = min(src_cols - x_min, TILE_SIZE); int x_max = min(src_cols - x_min, TILE_SIZE);
__global const uchar* ptr = src + src_offset + ypix*src_step + x_min; __global const uchar* ptr = src + src_offset + ypix*src_step + x_min;
int4 S = (int4)(0,0,0,0), p; int4 S = (int4)(0, 0, 0, 0), p;
#define SUM_ELEM(elem, ofs) \ #define SUM_ELEM(elem, ofs) \
(int4)(1, (ofs), (ofs)*(ofs), (ofs)*(ofs)*(ofs))*elem (int4)(1, (ofs), (ofs)*(ofs), (ofs)*(ofs)*(ofs))*elem
x = x_max & -4; x = x_max & -4;
if( x_max >= 4 ) if (x_max >= 4)
{ {
p = convert_int4(vload4(0, ptr)); p = convert_int4(vload4(0, ptr));
S += SUM_ELEM(p.s0, 0) + SUM_ELEM(p.s1, 1) + SUM_ELEM(p.s2, 2) + SUM_ELEM(p.s3, 3); #ifdef OP_MOMENTS_BINARY
p = min(p, 1);
#endif
S += (int4)(p.s0, 0, 0, 0) + (int4)(p.s1, p.s1, p.s1, p.s1) +
(int4)(p.s2, p.s2 * 2, p.s2 * 4, p.s2 * 8) + (int4)(p.s3, p.s3 * 3, p.s3 * 9, p.s3 * 27);
//SUM_ELEM(p.s0, 0) + SUM_ELEM(p.s1, 1) + SUM_ELEM(p.s2, 2) + SUM_ELEM(p.s3, 3);
if( x_max >= 8 ) if (x_max >= 8)
{ {
p = convert_int4(vload4(0, ptr+4)); p = convert_int4(vload4(0, ptr + 4));
S += SUM_ELEM(p.s0, 4) + SUM_ELEM(p.s1, 5) + SUM_ELEM(p.s2, 6) + SUM_ELEM(p.s3, 7); #ifdef OP_MOMENTS_BINARY
p = min(p, 1);
#endif
S += (int4)(p.s0, p.s0 * 4, p.s0 * 16, p.s0 * 64) + (int4)(p.s1, p.s1 * 5, p.s1 * 25, p.s1 * 125) +
(int4)(p.s2, p.s2 * 6, p.s2 * 36, p.s2 * 216) + (int4)(p.s3, p.s3 * 7, p.s3 * 49, p.s3 * 343);
//SUM_ELEM(p.s0, 4) + SUM_ELEM(p.s1, 5) + SUM_ELEM(p.s2, 6) + SUM_ELEM(p.s3, 7);
if( x_max >= 12 ) if (x_max >= 12)
{ {
p = convert_int4(vload4(0, ptr+8)); p = convert_int4(vload4(0, ptr + 8));
S += SUM_ELEM(p.s0, 8) + SUM_ELEM(p.s1, 9) + SUM_ELEM(p.s2, 10) + SUM_ELEM(p.s3, 11); #ifdef OP_MOMENTS_BINARY
p = min(p, 1);
#endif
S += (int4)(p.s0, p.s0 * 8, p.s0 * 64, p.s0 * 512) + (int4)(p.s1, p.s1 * 9, p.s1 * 81, p.s1 * 729) +
(int4)(p.s2, p.s2 * 10, p.s2 * 100, p.s2 * 1000) + (int4)(p.s3, p.s3 * 11, p.s3 * 121, p.s3 * 1331);
//SUM_ELEM(p.s0, 8) + SUM_ELEM(p.s1, 9) + SUM_ELEM(p.s2, 10) + SUM_ELEM(p.s3, 11);
if( x_max >= 16 ) if (x_max >= 16)
{ {
p = convert_int4(vload4(0, ptr+12)); p = convert_int4(vload4(0, ptr + 12));
S += SUM_ELEM(p.s0, 12) + SUM_ELEM(p.s1, 13) + SUM_ELEM(p.s2, 14) + SUM_ELEM(p.s3, 15); #ifdef OP_MOMENTS_BINARY
p = min(p, 1);
#endif
S += (int4)(p.s0, p.s0 * 12, p.s0 * 144, p.s0 * 1728) + (int4)(p.s1, p.s1 * 13, p.s1 * 169, p.s1 * 2197) +
(int4)(p.s2, p.s2 * 14, p.s2 * 196, p.s2 * 2744) + (int4)(p.s3, p.s3 * 15, p.s3 * 225, p.s3 * 3375);
//SUM_ELEM(p.s0, 12) + SUM_ELEM(p.s1, 13) + SUM_ELEM(p.s2, 14) + SUM_ELEM(p.s3, 15);
} }
} }
} }
} }
if( x_max >= 20 ) if (x_max >= 20)
{ {
p = convert_int4(vload4(0, ptr+16)); p = convert_int4(vload4(0, ptr + 16));
S += SUM_ELEM(p.s0, 16) + SUM_ELEM(p.s1, 17) + SUM_ELEM(p.s2, 18) + SUM_ELEM(p.s3, 19); #ifdef OP_MOMENTS_BINARY
p = min(p, 1);
#endif
S += (int4)(p.s0, p.s0 * 16, p.s0 * 256, p.s0 * 4096) + (int4)(p.s1, p.s1 * 17, p.s1 * 289, p.s1 * 4913) +
(int4)(p.s2, p.s2 * 18, p.s2 * 324, p.s2 * 5832) + (int4)(p.s3, p.s3 * 19, p.s3 * 361, p.s3 * 6859);
//SUM_ELEM(p.s0, 16) + SUM_ELEM(p.s1, 17) + SUM_ELEM(p.s2, 18) + SUM_ELEM(p.s3, 19);
if( x_max >= 24 ) if (x_max >= 24)
{ {
p = convert_int4(vload4(0, ptr+20)); p = convert_int4(vload4(0, ptr + 20));
S += SUM_ELEM(p.s0, 20) + SUM_ELEM(p.s1, 21) + SUM_ELEM(p.s2, 22) + SUM_ELEM(p.s3, 23); #ifdef OP_MOMENTS_BINARY
p = min(p, 1);
#endif
S += (int4)(p.s0, p.s0 * 20, p.s0 * 400, p.s0 * 8000) + (int4)(p.s1, p.s1 * 21, p.s1 * 441, p.s1 * 9261) +
(int4)(p.s2, p.s2 * 22, p.s2 * 484, p.s2 * 10648) + (int4)(p.s3, p.s3 * 23, p.s3 * 529, p.s3 * 12167);
//SUM_ELEM(p.s0, 20) + SUM_ELEM(p.s1, 21) + SUM_ELEM(p.s2, 22) + SUM_ELEM(p.s3, 23);
if( x_max >= 28 ) if (x_max >= 28)
{ {
p = convert_int4(vload4(0, ptr+24)); p = convert_int4(vload4(0, ptr + 24));
S += SUM_ELEM(p.s0, 24) + SUM_ELEM(p.s1, 25) + SUM_ELEM(p.s2, 26) + SUM_ELEM(p.s3, 27); #ifdef OP_MOMENTS_BINARY
p = min(p, 1);
#endif
S += (int4)(p.s0, p.s0 * 24, p.s0 * 576, p.s0 * 13824) + (int4)(p.s1, p.s1 * 25, p.s1 * 625, p.s1 * 15625) +
(int4)(p.s2, p.s2 * 26, p.s2 * 676, p.s2 * 17576) + (int4)(p.s3, p.s3 * 27, p.s3 * 729, p.s3 * 19683);
//SUM_ELEM(p.s0, 24) + SUM_ELEM(p.s1, 25) + SUM_ELEM(p.s2, 26) + SUM_ELEM(p.s3, 27);
if( x_max >= 32 ) if (x_max >= 32)
{ {
p = convert_int4(vload4(0, ptr+28)); p = convert_int4(vload4(0, ptr + 28));
S += SUM_ELEM(p.s0, 28) + SUM_ELEM(p.s1, 29) + SUM_ELEM(p.s2, 30) + SUM_ELEM(p.s3, 31); #ifdef OP_MOMENTS_BINARY
p = min(p, 1);
#endif
S += (int4)(p.s0, p.s0 * 28, p.s0 * 784, p.s0 * 21952) + (int4)(p.s1, p.s1 * 29, p.s1 * 841, p.s1 * 24389) +
(int4)(p.s2, p.s2 * 30, p.s2 * 900, p.s2 * 27000) + (int4)(p.s3, p.s3 * 31, p.s3 * 961, p.s3 * 29791);
//SUM_ELEM(p.s0, 28) + SUM_ELEM(p.s1, 29) + SUM_ELEM(p.s2, 30) + SUM_ELEM(p.s3, 31);
} }
} }
} }
} }
if( x < x_max ) if (x < x_max)
{ {
int ps = ptr[x]; int ps = ptr[x];
#ifdef OP_MOMENTS_BINARY
ps = min(ps, 1);
#endif
S += SUM_ELEM(ps, x); S += SUM_ELEM(ps, x);
if( x+1 < x_max ) if (x + 1 < x_max)
{ {
ps = ptr[x+1]; ps = ptr[x + 1];
S += SUM_ELEM(ps, x+1); #ifdef OP_MOMENTS_BINARY
if( x+2 < x_max ) ps = min(ps, 1);
#endif
S += SUM_ELEM(ps, x + 1);
if (x + 2 < x_max)
{ {
ps = ptr[x+2]; ps = ptr[x + 2];
S += SUM_ELEM(ps, x+2); #ifdef OP_MOMENTS_BINARY
ps = min(ps, 1);
#endif
S += SUM_ELEM(ps, x + 2);
} }
} }
} }
...@@ -108,19 +158,19 @@ __kernel void moments(__global const uchar* src, int src_step, int src_offset, ...@@ -108,19 +158,19 @@ __kernel void moments(__global const uchar* src, int src_step, int src_offset,
mom[y][5] = mom[y][6] = mom[y][7] = mom[y][8] = mom[y][9] = 0; mom[y][5] = mom[y][6] = mom[y][7] = mom[y][8] = mom[y][9] = 0;
barrier(CLK_LOCAL_MEM_FENCE); barrier(CLK_LOCAL_MEM_FENCE);
#define REDUCE(d) \ #define REDUCE(d) \
if( y < d ) \ if (y < d) \
{ \ { \
mom[y][0] += mom[y+d][0]; \ mom[y][0] += mom[y + d][0]; \
mom[y][1] += mom[y+d][1]; \ mom[y][1] += mom[y + d][1]; \
mom[y][2] += mom[y+d][2]; \ mom[y][2] += mom[y + d][2]; \
mom[y][3] += mom[y+d][3]; \ mom[y][3] += mom[y + d][3]; \
mom[y][4] += mom[y+d][4]; \ mom[y][4] += mom[y + d][4]; \
mom[y][5] += mom[y+d][5]; \ mom[y][5] += mom[y + d][5]; \
mom[y][6] += mom[y+d][6]; \ mom[y][6] += mom[y + d][6]; \
mom[y][7] += mom[y+d][7]; \ mom[y][7] += mom[y + d][7]; \
mom[y][8] += mom[y+d][8]; \ mom[y][8] += mom[y + d][8]; \
mom[y][9] += mom[y+d][9]; \ mom[y][9] += mom[y + d][9]; \
} \ } \
barrier(CLK_LOCAL_MEM_FENCE) barrier(CLK_LOCAL_MEM_FENCE)
...@@ -129,19 +179,10 @@ __kernel void moments(__global const uchar* src, int src_step, int src_offset, ...@@ -129,19 +179,10 @@ __kernel void moments(__global const uchar* src, int src_step, int src_offset,
REDUCE(4); REDUCE(4);
REDUCE(2); REDUCE(2);
if( y == 0 ) if (y < 10)
{ {
__global int* momout = mom0 + (y0*xtiles + x0)*10; __global int* momout = mom0 + (y0*xtiles + x0) * 10;
momout[0] = mom[0][0] + mom[1][0]; momout[y] = mom[0][y] + mom[1][y];
momout[1] = mom[0][1] + mom[1][1];
momout[2] = mom[0][2] + mom[1][2];
momout[3] = mom[0][3] + mom[1][3];
momout[4] = mom[0][4] + mom[1][4];
momout[5] = mom[0][5] + mom[1][5];
momout[6] = mom[0][6] + mom[1][6];
momout[7] = mom[0][7] + mom[1][7];
momout[8] = mom[0][8] + mom[1][8];
momout[9] = mom[0][9] + mom[1][9];
} }
} }
} }
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