moments work now and work more or less fast

This commit is contained in:
Vadim Pisarevsky 2013-12-26 02:57:08 +04:00
parent 6035925f41
commit 83f749afd2
4 changed files with 88 additions and 118 deletions

View File

@ -2261,6 +2261,12 @@ void _OutputArray::release() const
((Mat*)obj)->release();
return;
}
if( k == UMAT )
{
((UMat*)obj)->release();
return;
}
if( k == GPU_MAT )
{

View File

@ -363,36 +363,31 @@ Moments::Moments( double _m00, double _m10, double _m01, double _m20, double _m1
nu30 = mu30*s3; nu21 = mu21*s3; nu12 = mu12*s3; nu03 = mu03*s3;
}
static const int OCL_TILE_SIZE = 32;
static bool ocl_moments( InputArray _src, Moments& m, bool binary )
static bool ocl_moments( InputArray _src, Moments& m)
{
printf("!!!!!!!!!!!!!!!!!! ocl moments !!!!!!!!!!!!!!!!!!!\n");
const int TILE_SIZE = 16;
const int K = 10;
ocl::Kernel k("moments", ocl::imgproc::moments_oclsrc, binary ? "-D BINARY_MOMENTS" : "");
ocl::Kernel k("moments", ocl::imgproc::moments_oclsrc, format("-D TILE_SIZE=%d", TILE_SIZE));
if( k.empty() )
return false;
UMat src = _src.getUMat();
Size sz = src.size();
int xtiles = (sz.width + OCL_TILE_SIZE-1)/OCL_TILE_SIZE;
int ytiles = (sz.height + OCL_TILE_SIZE-1)/OCL_TILE_SIZE;
int xtiles = (sz.width + TILE_SIZE-1)/TILE_SIZE;
int ytiles = (sz.height + TILE_SIZE-1)/TILE_SIZE;
int ntiles = xtiles*ytiles;
UMat umbuf(1, ntiles*K, CV_32S);
umbuf.setTo(Scalar::all(0));
size_t globalsize[] = {xtiles, ytiles};
size_t localsize[] = {1, 1};
bool ok = k.args(ocl::KernelArg::ReadOnly(src),
ocl::KernelArg::PtrWriteOnly(umbuf),
OCL_TILE_SIZE, xtiles, ytiles).run(2, globalsize, localsize, false);
xtiles).run(2, globalsize, 0, true);
if(!ok)
return false;
Mat mbuf;
umbuf.copyTo(mbuf);
Mat mbuf = umbuf.getMat(ACCESS_READ);
for( int i = 0; i < ntiles; i++ )
{
double x = (i % xtiles)*OCL_TILE_SIZE, y = (i / xtiles)*OCL_TILE_SIZE;
double x = (i % xtiles)*TILE_SIZE, y = (i / xtiles)*TILE_SIZE;
const int* mom = mbuf.ptr<int>() + i*K;
double xm = x * mom[0], ym = y * mom[0];
@ -452,10 +447,8 @@ cv::Moments cv::moments( InputArray _src, bool binary )
if( size.width <= 0 || size.height <= 0 )
return m;
if( ocl::useOpenCL() && depth == CV_8U &&
size.width >= OCL_TILE_SIZE &&
size.height >= OCL_TILE_SIZE &&
/*_src.isUMat() &&*/ ocl_moments(_src, m, binary) )
if( ocl::useOpenCL() && depth == CV_8U && !binary &&
_src.isUMat() && ocl_moments(_src, m) )
;
else
{

View File

@ -1,110 +1,70 @@
/* See LICENSE file in the root OpenCV directory */
#ifdef BINARY_MOMENTS
#define READ_PIX(ref) (ref != 0)
#else
#define READ_PIX(ref) ref
#endif
__kernel void moments(__global const uchar* src, int src_step, int src_offset,
int src_rows, int src_cols, __global int* mom0,
int tile_size, int xtiles, int ytiles)
int src_rows, int src_cols, __global int* mom0, int xtiles)
{
int x = get_global_id(0);
int y = get_global_id(1);
int x_min = x*tile_size;
int y_min = y*tile_size;
int x_min = x*TILE_SIZE;
int y_min = y*TILE_SIZE;
if( x_min < src_cols && y_min < src_rows )
{
int x_max = src_cols - x_min;
int y_max = src_rows - y_min;
int m[10]={0,0,0,0,0,0,0,0,0,0};
__global const uchar* ptr = (src + src_offset);// + y_min*src_step + x_min;
int x_max = min(src_cols - x_min, TILE_SIZE);
int y_max = min(src_rows - y_min, TILE_SIZE);
int m00=0, m10=0, m01=0, m20=0, m11=0, m02=0, m30=0, m21=0, m12=0, m03=0;
__global const uchar* ptr = src + src_offset + y_min*src_step + x_min;
__global int* mom = mom0 + (xtiles*y + x)*10;
x_max = x_max < tile_size ? x_max : tile_size;
y_max = y_max < tile_size ? y_max : tile_size;
for( y = 0; y < y_max; y++ )
for( y = 0; y < y_max; y++, ptr += src_step )
{
int x00, x10, x20, x30;
int sx, sy, p;
x00 = x10 = x20 = x30 = 0;
sy = y*y;
int4 S = (int4)(0,0,0,0);
for( x = 0; x < x_max; x++ )
for( x = 0; x <= x_max - 4; x += 4 )
{
p = ptr[0];//READ_PIX(ptr[x]);
sx = x*x;
x00 += p;
x10 += x*p;
x20 += sx*p;
x30 += x*sx*p;
int4 p = convert_int4(vload4(0, ptr + x));
#define SUM_ELEM(elem, ofs) \
(int4)(elem, (x+ofs)*elem, (x+ofs)*(x+ofs)*elem, (x+ofs)*(x+ofs)*(x+ofs)*elem)
S += SUM_ELEM(p.s0, 0) + SUM_ELEM(p.s1, 1) + SUM_ELEM(p.s2, 2) + SUM_ELEM(p.s3, 3);
}
m[0] += x00;
m[1] += x10;
m[2] += y*x00;
m[3] += x20;
m[4] += y*x10;
m[5] += sy*x00;
m[6] += x30;
m[7] += y*x20;
m[8] += sy*x10;
m[9] += y*sy*x00;
//ptr += src_step;
if( x < x_max )
{
int ps = ptr[x];
S += SUM_ELEM(ps, 0);
if( x+1 < x_max )
{
ps = ptr[x+1];
S += SUM_ELEM(ps, 1);
if( x+2 < x_max )
{
ps = ptr[x+2];
S += SUM_ELEM(ps, 2);
}
}
}
int sy = y*y;
m00 += S.s0;
m10 += S.s1;
m01 += y*S.s0;
m20 += S.s2;
m11 += y*S.s1;
m02 += sy*S.s0;
m30 += S.s3;
m21 += y*S.s2;
m12 += sy*S.s1;
m03 += y*sy*S.s0;
}
mom[0] = m[0];
mom[1] = m[1];
mom[2] = m[2];
mom[3] = m[3];
mom[4] = m[4];
mom[5] = m[5];
mom[6] = m[6];
mom[7] = m[7];
mom[8] = m[8];
mom[9] = m[9];
mom[0] = m00;
mom[1] = m10;
mom[2] = m01;
mom[3] = m20;
mom[4] = m11;
mom[5] = m02;
mom[6] = m30;
mom[7] = m21;
mom[8] = m12;
mom[9] = m03;
}
}
/*__kernel void moments(__global const uchar* src, int src_step, int src_offset,
int src_rows, int src_cols, __global float* mom0,
int tile_size, int xtiles, int ytiles)
{
int x = get_global_id(0);
int y = get_global_id(1);
if( x < xtiles && y < ytiles )
{
//int x_min = x*tile_size;
//int y_min = y*tile_size;
//int x_max = src_cols - x_min;
//int y_max = src_rows - y_min;
__global const uchar* ptr = src + src_offset;// + src_step*y_min + x_min;
__global float* mom = mom0;// + (y*xtiles + x)*16;
//int x00, x10, x20, x30, m00=0;
//x_max = min(x_max, tile_size);
//y_max = min(y_max, tile_size);
//int m00 = 0;
//for( y = 0; y < y_max; y++, ptr += src_step )
//{
//int x00 = 0, x10 = 0, x20 = 0, x30 = 0;
//for( x = 0; x < x_max; x++ )
//{
int p = ptr[x];
//m00 = p;
//x10 += x*p;
/*x20 += x*x*p;
x30 += x*x*x*p;
//}
//m00 = m00 + x00;
//}
mom[0] = p;
}
}*/

View File

@ -60,6 +60,7 @@ protected:
void run_func();
int coi;
bool is_binary;
bool try_umat;
};
@ -102,20 +103,25 @@ void CV_MomentsTest::get_test_array_types_and_sizes( int test_case_idx,
{
RNG& rng = ts->get_rng();
cvtest::ArrayTest::get_test_array_types_and_sizes( test_case_idx, sizes, types );
int cn = cvtest::randInt(rng) % 4 + 1;
int cn = (cvtest::randInt(rng) % 4) + 1;
int depth = cvtest::randInt(rng) % 4;
depth = depth == 0 ? CV_8U : depth == 1 ? CV_16U : depth == 2 ? CV_16S : CV_32F;
if( cn == 2 )
is_binary = cvtest::randInt(rng) % 2 != 0;
if( depth == 0 && !is_binary )
try_umat = cvtest::randInt(rng) % 5 != 0;
else
try_umat = cvtest::randInt(rng) % 2 != 0;
if( cn == 2 || try_umat )
cn = 1;
sizes[INPUT][0].height = sizes[INPUT][0].width;
types[INPUT][0] = CV_MAKETYPE(depth, cn);
types[OUTPUT][0] = types[REF_OUTPUT][0] = CV_64FC1;
sizes[OUTPUT][0] = sizes[REF_OUTPUT][0] = cvSize(MOMENT_COUNT,1);
if(CV_MAT_DEPTH(types[INPUT][0])>=CV_32S)
sizes[INPUT][0].width = MAX(sizes[INPUT][0].width, 3);
is_binary = cvtest::randInt(rng) % 2 != 0;
coi = 0;
cvmat_allowed = true;
if( cn > 1 )
@ -150,7 +156,16 @@ void CV_MomentsTest::run_func()
{
CvMoments* m = (CvMoments*)test_mat[OUTPUT][0].ptr<double>();
double* others = (double*)(m + 1);
cvMoments( test_array[INPUT][0], m, is_binary );
if( try_umat )
{
UMat u;
test_mat[INPUT][0].clone().copyTo(u);
Moments new_m = moments(u, is_binary != 0);
*m = new_m;
}
else
cvMoments( test_array[INPUT][0], m, is_binary );
others[0] = cvGetNormalizedCentralMoment( m, 2, 0 );
others[1] = cvGetNormalizedCentralMoment( m, 1, 1 );
others[2] = cvGetNormalizedCentralMoment( m, 0, 2 );
@ -275,10 +290,6 @@ void CV_MomentsTest::prepare_to_validation( int /*test_case_idx*/ )
mdata[6] = m.mu03 * s3;
}
test_mat[REF_OUTPUT][0].copyTo(test_mat[OUTPUT][0]);
cout << "ref moments: " << test_mat[REF_OUTPUT][0] << "\n";
cout << "fun moments: " << test_mat[OUTPUT][0] << "\n";
double* a = test_mat[REF_OUTPUT][0].ptr<double>();
double* b = test_mat[OUTPUT][0].ptr<double>();
for( i = 0; i < MOMENT_COUNT; i++ )