convert.dispatch.cpp
9.03 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html
#include "precomp.hpp"
#include "opencl_kernels_core.hpp"
#include "convert.simd.hpp"
#include "convert.simd_declarations.hpp" // defines CV_CPU_DISPATCH_MODES_ALL=AVX2,...,BASELINE based on CMakeLists.txt content
namespace cv {
namespace hal {
void cvt16f32f(const float16_t* src, float* dst, int len)
{
CV_INSTRUMENT_REGION();
CV_CPU_DISPATCH(cvt16f32f, (src, dst, len),
CV_CPU_DISPATCH_MODES_ALL);
}
void cvt32f16f(const float* src, float16_t* dst, int len)
{
CV_INSTRUMENT_REGION();
CV_CPU_DISPATCH(cvt32f16f, (src, dst, len),
CV_CPU_DISPATCH_MODES_ALL);
}
void addRNGBias32f(float* arr, const float* scaleBiasPairs, int len)
{
CV_INSTRUMENT_REGION();
CV_CPU_DISPATCH(addRNGBias32f, (arr, scaleBiasPairs, len),
CV_CPU_DISPATCH_MODES_ALL);
}
void addRNGBias64f(double* arr, const double* scaleBiasPairs, int len)
{
CV_INSTRUMENT_REGION();
CV_CPU_DISPATCH(addRNGBias64f, (arr, scaleBiasPairs, len),
CV_CPU_DISPATCH_MODES_ALL);
}
} // namespace
/* [TODO] Recover IPP calls
#if defined(HAVE_IPP)
#define DEF_CVT_FUNC_F(suffix, stype, dtype, ippFavor) \
static void cvt##suffix( const stype* src, size_t sstep, const uchar*, size_t, \
dtype* dst, size_t dstep, Size size, double*) \
{ \
CV_IPP_RUN(src && dst, CV_INSTRUMENT_FUN_IPP(ippiConvert_##ippFavor, src, (int)sstep, dst, (int)dstep, ippiSize(size.width, size.height)) >= 0) \
cvt_(src, sstep, dst, dstep, size); \
}
#define DEF_CVT_FUNC_F2(suffix, stype, dtype, ippFavor) \
static void cvt##suffix( const stype* src, size_t sstep, const uchar*, size_t, \
dtype* dst, size_t dstep, Size size, double*) \
{ \
CV_IPP_RUN(src && dst, CV_INSTRUMENT_FUN_IPP(ippiConvert_##ippFavor, src, (int)sstep, dst, (int)dstep, ippiSize(size.width, size.height), ippRndFinancial, 0) >= 0) \
cvt_(src, sstep, dst, dstep, size); \
}
#else
#define DEF_CVT_FUNC_F(suffix, stype, dtype, ippFavor) \
static void cvt##suffix( const stype* src, size_t sstep, const uchar*, size_t, \
dtype* dst, size_t dstep, Size size, double*) \
{ \
cvt_(src, sstep, dst, dstep, size); \
}
#define DEF_CVT_FUNC_F2 DEF_CVT_FUNC_F
#endif
#define DEF_CVT_FUNC(suffix, stype, dtype) \
static void cvt##suffix( const stype* src, size_t sstep, const uchar*, size_t, \
dtype* dst, size_t dstep, Size size, double*) \
{ \
cvt_(src, sstep, dst, dstep, size); \
}
#define DEF_CPY_FUNC(suffix, stype) \
static void cvt##suffix( const stype* src, size_t sstep, const uchar*, size_t, \
stype* dst, size_t dstep, Size size, double*) \
{ \
cpy_(src, sstep, dst, dstep, size); \
}
DEF_CPY_FUNC(8u, uchar)
DEF_CVT_FUNC_F(8s8u, schar, uchar, 8s8u_C1Rs)
DEF_CVT_FUNC_F(16u8u, ushort, uchar, 16u8u_C1R)
DEF_CVT_FUNC_F(16s8u, short, uchar, 16s8u_C1R)
DEF_CVT_FUNC_F(32s8u, int, uchar, 32s8u_C1R)
DEF_CVT_FUNC_F2(32f8u, float, uchar, 32f8u_C1RSfs)
DEF_CVT_FUNC(64f8u, double, uchar)
DEF_CVT_FUNC_F2(8u8s, uchar, schar, 8u8s_C1RSfs)
DEF_CVT_FUNC_F2(16u8s, ushort, schar, 16u8s_C1RSfs)
DEF_CVT_FUNC_F2(16s8s, short, schar, 16s8s_C1RSfs)
DEF_CVT_FUNC_F(32s8s, int, schar, 32s8s_C1R)
DEF_CVT_FUNC_F2(32f8s, float, schar, 32f8s_C1RSfs)
DEF_CVT_FUNC(64f8s, double, schar)
DEF_CVT_FUNC_F(8u16u, uchar, ushort, 8u16u_C1R)
DEF_CVT_FUNC_F(8s16u, schar, ushort, 8s16u_C1Rs)
DEF_CPY_FUNC(16u, ushort)
DEF_CVT_FUNC_F(16s16u, short, ushort, 16s16u_C1Rs)
DEF_CVT_FUNC_F2(32s16u, int, ushort, 32s16u_C1RSfs)
DEF_CVT_FUNC_F2(32f16u, float, ushort, 32f16u_C1RSfs)
DEF_CVT_FUNC(64f16u, double, ushort)
DEF_CVT_FUNC_F(8u16s, uchar, short, 8u16s_C1R)
DEF_CVT_FUNC_F(8s16s, schar, short, 8s16s_C1R)
DEF_CVT_FUNC_F2(16u16s, ushort, short, 16u16s_C1RSfs)
DEF_CVT_FUNC_F2(32s16s, int, short, 32s16s_C1RSfs)
DEF_CVT_FUNC(32f16s, float, short)
DEF_CVT_FUNC(64f16s, double, short)
DEF_CVT_FUNC_F(8u32s, uchar, int, 8u32s_C1R)
DEF_CVT_FUNC_F(8s32s, schar, int, 8s32s_C1R)
DEF_CVT_FUNC_F(16u32s, ushort, int, 16u32s_C1R)
DEF_CVT_FUNC_F(16s32s, short, int, 16s32s_C1R)
DEF_CPY_FUNC(32s, int)
DEF_CVT_FUNC_F2(32f32s, float, int, 32f32s_C1RSfs)
DEF_CVT_FUNC(64f32s, double, int)
DEF_CVT_FUNC_F(8u32f, uchar, float, 8u32f_C1R)
DEF_CVT_FUNC_F(8s32f, schar, float, 8s32f_C1R)
DEF_CVT_FUNC_F(16u32f, ushort, float, 16u32f_C1R)
DEF_CVT_FUNC_F(16s32f, short, float, 16s32f_C1R)
DEF_CVT_FUNC_F(32s32f, int, float, 32s32f_C1R)
DEF_CVT_FUNC(64f32f, double, float)
DEF_CVT_FUNC(8u64f, uchar, double)
DEF_CVT_FUNC(8s64f, schar, double)
DEF_CVT_FUNC(16u64f, ushort, double)
DEF_CVT_FUNC(16s64f, short, double)
DEF_CVT_FUNC(32s64f, int, double)
DEF_CVT_FUNC(32f64f, float, double)
DEF_CPY_FUNC(64s, int64)
*/
BinaryFunc getConvertFunc(int sdepth, int ddepth)
{
CV_INSTRUMENT_REGION();
CV_CPU_DISPATCH(getConvertFunc, (sdepth, ddepth),
CV_CPU_DISPATCH_MODES_ALL);
}
#ifdef HAVE_OPENCL
static bool ocl_convertFp16( InputArray _src, OutputArray _dst, int sdepth, int ddepth )
{
int type = _src.type(), cn = CV_MAT_CN(type);
_dst.createSameSize( _src, CV_MAKETYPE(ddepth, cn) );
int kercn = 1;
int rowsPerWI = 1;
String build_opt = format("-D HALF_SUPPORT -D srcT=%s -D dstT=%s -D rowsPerWI=%d%s",
sdepth == CV_32F ? "float" : "half",
sdepth == CV_32F ? "half" : "float",
rowsPerWI,
sdepth == CV_32F ? " -D FLOAT_TO_HALF " : "");
ocl::Kernel k(sdepth == CV_32F ? "convertFp16_FP32_to_FP16" : "convertFp16_FP16_to_FP32", ocl::core::halfconvert_oclsrc, build_opt);
if (k.empty())
return false;
UMat src = _src.getUMat();
UMat dst = _dst.getUMat();
ocl::KernelArg srcarg = ocl::KernelArg::ReadOnlyNoSize(src),
dstarg = ocl::KernelArg::WriteOnly(dst, cn, kercn);
k.args(srcarg, dstarg);
size_t globalsize[2] = { (size_t)src.cols * cn / kercn, ((size_t)src.rows + rowsPerWI - 1) / rowsPerWI };
return k.run(2, globalsize, NULL, false);
}
#endif
void Mat::convertTo(OutputArray _dst, int _type, double alpha, double beta) const
{
CV_INSTRUMENT_REGION();
if( empty() )
{
_dst.release();
return;
}
bool noScale = fabs(alpha-1) < DBL_EPSILON && fabs(beta) < DBL_EPSILON;
if( _type < 0 )
_type = _dst.fixedType() ? _dst.type() : type();
else
_type = CV_MAKETYPE(CV_MAT_DEPTH(_type), channels());
int sdepth = depth(), ddepth = CV_MAT_DEPTH(_type);
if( sdepth == ddepth && noScale )
{
copyTo(_dst);
return;
}
Mat src = *this;
if( dims <= 2 )
_dst.create( size(), _type );
else
_dst.create( dims, size, _type );
Mat dst = _dst.getMat();
BinaryFunc func = noScale ? getConvertFunc(sdepth, ddepth) : getConvertScaleFunc(sdepth, ddepth);
double scale[] = {alpha, beta};
int cn = channels();
CV_Assert( func != 0 );
if( dims <= 2 )
{
Size sz = getContinuousSize2D(src, dst, cn);
func( src.data, src.step, 0, 0, dst.data, dst.step, sz, scale );
}
else
{
const Mat* arrays[] = {&src, &dst, 0};
uchar* ptrs[2] = {};
NAryMatIterator it(arrays, ptrs);
Size sz((int)(it.size*cn), 1);
for( size_t i = 0; i < it.nplanes; i++, ++it )
func(ptrs[0], 1, 0, 0, ptrs[1], 1, sz, scale);
}
}
//==================================================================================================
void convertFp16(InputArray _src, OutputArray _dst)
{
CV_INSTRUMENT_REGION();
int sdepth = _src.depth(), ddepth = 0;
BinaryFunc func = 0;
switch( sdepth )
{
case CV_32F:
if(_dst.fixedType())
{
ddepth = _dst.depth();
CV_Assert(ddepth == CV_16S || ddepth == CV_16F);
CV_Assert(_dst.channels() == _src.channels());
}
else
ddepth = CV_16S;
func = getConvertFunc(CV_32F, CV_16F);
break;
case CV_16S:
case CV_16F:
ddepth = CV_32F;
func = getConvertFunc(CV_16F, CV_32F);
break;
default:
CV_Error(Error::StsUnsupportedFormat, "Unsupported input depth");
return;
}
CV_OCL_RUN(_src.dims() <= 2 && _dst.isUMat(),
ocl_convertFp16(_src, _dst, sdepth, ddepth))
Mat src = _src.getMat();
int type = CV_MAKETYPE(ddepth, src.channels());
_dst.create( src.dims, src.size, type );
Mat dst = _dst.getMat();
int cn = src.channels();
CV_Assert( func != 0 );
if( src.dims <= 2 )
{
Size sz = getContinuousSize2D(src, dst, cn);
func( src.data, src.step, 0, 0, dst.data, dst.step, sz, 0);
}
else
{
const Mat* arrays[] = {&src, &dst, 0};
uchar* ptrs[2] = {};
NAryMatIterator it(arrays, ptrs);
Size sz((int)(it.size*cn), 1);
for( size_t i = 0; i < it.nplanes; i++, ++it )
func(ptrs[0], 0, 0, 0, ptrs[1], 0, sz, 0);
}
}
} // namespace cv