/
read_avg_prediction16x16_main.cpp
338 lines (319 loc) · 13.3 KB
/
read_avg_prediction16x16_main.cpp
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
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
#include <windows.h>
#include <stdio.h>
#include <stdlib.h>
#include <malloc.h>
#include <time.h>
#include <assert.h>
#include <conio.h>
#include <emmintrin.h>
#define MB_X_REF 68
#define MB_Y_REF 45
#define MB_SIZE 16
#define REF_STRIDE (MB_X_REF*MB_SIZE)
#ifdef _DEBUG
#define LOOP_TEST_COUNT_INTERP 1 //to have a larger runtime
#define LOOP_TEST_COUNT 1 //to have a larger runtime
#else
#define LOOP_TEST_COUNT_INTERP 50 //to have a larger runtime
#define LOOP_TEST_COUNT 100 //to have a larger runtime
#endif
extern "C" void read_luma_inter_pred_avg_16x16_SSE3( const BYTE *address1, const BYTE *address2, INT stride_src, BYTE *dst, INT stride_dst);
extern "C" void read_luma_inter_pred_avg_8x16_SSE3( const BYTE *address1, const BYTE *address2, INT stride_src, BYTE *dst, INT stride_dst);
extern "C" void read_luma_inter_pred_avg_16x8_SSE3( const BYTE *address1, const BYTE *address2, INT stride_src, BYTE *dst, INT stride_dst);
extern "C" void read_luma_inter_pred_avg_8x8_SSE3( const BYTE *address1, const BYTE *address2, INT stride_src, BYTE *dst, INT stride_dst);
#ifndef _WIN64
#ifdef _DEBUG
#pragma comment (lib, "libs/sad_lib_D_x32_vs10.lib")
#else
#pragma comment (lib, "libs/sad_lib_R_x32_vs10.lib")
#endif
#else
#ifdef _DEBUG
#pragma comment (lib, "libs/sad_lib_D_x64_vs10.lib")
#else
#pragma comment (lib, "libs/sad_lib_R_x64_vs10.lib")
#endif
#endif
void read_luma_inter_pred_avg_16x16_C( BYTE *address1, BYTE *address2, INT stride_src, BYTE *dst, INT stride_dst )
{
BYTE *h1=address1;
BYTE *h2=address2;
INT i, j;
for( i = 0; i < 16; i++ )
{
for( j = 0; j < 16; j++ )
{
dst[j]=(h1[j]+h2[j]+1 )>>1;
}
h1+=stride_src;
h2+=stride_src;
dst+=stride_dst;
}
}
void read_luma_inter_pred_avg_16x16_intrinsic( BYTE *address1, BYTE *address2, INT stride_src, BYTE *dst, INT stride_dst )
{
for(int i = 0; i < 16; i+=8)
{
__declspec(align(16)) __m128i r0, r1, r2, r3, r4, r5, r6, r7,
r0_x, r1_x, r2_x, r3_x, r4_x, r5_x, r6_x, r7_x;
int stride2 = (stride_src<<1);
int stride4 = (stride_src<<2);
int dst_stride2 = (stride_dst<<1);
int dst_stride4 = (stride_dst<<2);
r0 = _mm_loadu_si128((__m128i*)(address1));
r1 = _mm_loadu_si128((__m128i*)(address1+stride_dst));
r2 = _mm_loadu_si128((__m128i*)(address1+stride2));
r3 = _mm_loadu_si128((__m128i*)(address1+stride2+stride_src));
r4 = _mm_loadu_si128((__m128i*)(address1+stride4));
r5 = _mm_loadu_si128((__m128i*)(address1+stride4+stride_src));
r6 = _mm_loadu_si128((__m128i*)(address1+stride4+stride2));
r7 = _mm_loadu_si128((__m128i*)(address1+stride4+stride2+stride_src));
r0_x = _mm_loadu_si128((__m128i*)(address2));
r1_x = _mm_loadu_si128((__m128i*)(address2+stride_src));
r2_x = _mm_loadu_si128((__m128i*)(address2+stride2));
r3_x = _mm_loadu_si128((__m128i*)(address2+stride2+stride_src));
r4_x = _mm_loadu_si128((__m128i*)(address2+stride4));
r5_x = _mm_loadu_si128((__m128i*)(address2+stride4+stride_dst));
r6_x = _mm_loadu_si128((__m128i*)(address2+stride4+stride2));
r7_x = _mm_loadu_si128((__m128i*)(address2+stride4+stride2+stride_dst));
r0 = _mm_avg_epu8(r0, r0_x);
r1 = _mm_avg_epu8(r1, r1_x);
r2 = _mm_avg_epu8(r2, r2_x);
r3 = _mm_avg_epu8(r3, r3_x);
r4 = _mm_avg_epu8(r4, r4_x);
r5 = _mm_avg_epu8(r5, r5_x);
r6 = _mm_avg_epu8(r6, r6_x);
r7 = _mm_avg_epu8(r7, r7_x);
_mm_storeu_si128((__m128i*)(dst), r0);
_mm_storeu_si128((__m128i*)(dst+stride_dst), r1);
_mm_storeu_si128((__m128i*)(dst+dst_stride2), r2);
_mm_storeu_si128((__m128i*)(dst+dst_stride2+stride_dst), r3);
_mm_storeu_si128((__m128i*)(dst+dst_stride4), r4);
_mm_storeu_si128((__m128i*)(dst+dst_stride4+stride_dst), r5);
_mm_storeu_si128((__m128i*)(dst+dst_stride4+dst_stride2), r6);
_mm_storeu_si128((__m128i*)(dst+dst_stride4+dst_stride2+stride_dst), r7);
address1 += (stride4<<1);
address2 += (stride4<<1);
dst += (dst_stride4<<1);
}
}
void read_luma_inter_pred_avg_8x16_intrinsic( BYTE *address1, BYTE *address2, INT stride_src, BYTE *dst, INT stride_dst )
{
int i;
int src_stride = stride_src;
int dst_stride = stride_dst;
const unsigned char* src1 = address1;
const unsigned char* src2 = address2;
for( i = 0; i < 16; i+=8)
{
__declspec(align(16)) __m128i r0, r1, r2, r3, r4, r5, r6, r7,
r0_x, r1_x, r2_x, r3_x, r4_x, r5_x, r6_x, r7_x;
int stride2 = (src_stride<<1);
int stride4 = (src_stride<<2);
int dst_stride2 = (dst_stride<<1);
int dst_stride4 = (dst_stride<<2);
r0 = _mm_loadl_epi64((__m128i*)(src1));
r1 = _mm_loadl_epi64((__m128i*)(src1+src_stride));
r2 = _mm_loadl_epi64((__m128i*)(src1+stride2));
r3 = _mm_loadl_epi64((__m128i*)(src1+stride2+src_stride));
r4 = _mm_loadl_epi64((__m128i*)(src1+stride4));
r5 = _mm_loadl_epi64((__m128i*)(src1+stride4+src_stride));
r6 = _mm_loadl_epi64((__m128i*)(src1+stride4+stride2));
r7 = _mm_loadl_epi64((__m128i*)(src1+stride4+stride2+src_stride));
r0_x = _mm_loadl_epi64((__m128i*)(src2));
r1_x = _mm_loadl_epi64((__m128i*)(src2+src_stride));
r2_x = _mm_loadl_epi64((__m128i*)(src2+stride2));
r3_x = _mm_loadl_epi64((__m128i*)(src2+stride2+src_stride));
r4_x = _mm_loadl_epi64((__m128i*)(src2+stride4));
r5_x = _mm_loadl_epi64((__m128i*)(src2+stride4+src_stride));
r6_x = _mm_loadl_epi64((__m128i*)(src2+stride4+stride2));
r7_x = _mm_loadl_epi64((__m128i*)(src2+stride4+stride2+src_stride));
r0 = _mm_avg_epu8(r0, r0_x);
r1 = _mm_avg_epu8(r1, r1_x);
r2 = _mm_avg_epu8(r2, r2_x);
r3 = _mm_avg_epu8(r3, r3_x);
r4 = _mm_avg_epu8(r4, r4_x);
r5 = _mm_avg_epu8(r5, r5_x);
r6 = _mm_avg_epu8(r6, r6_x);
r7 = _mm_avg_epu8(r7, r7_x);
_mm_storel_epi64((__m128i*)(dst), r0);
_mm_storel_epi64((__m128i*)(dst+dst_stride), r1);
_mm_storel_epi64((__m128i*)(dst+dst_stride2), r2);
_mm_storel_epi64((__m128i*)(dst+dst_stride2+dst_stride), r3);
_mm_storel_epi64((__m128i*)(dst+dst_stride4), r4);
_mm_storel_epi64((__m128i*)(dst+dst_stride4+dst_stride), r5);
_mm_storel_epi64((__m128i*)(dst+dst_stride4+dst_stride2), r6);
_mm_storel_epi64((__m128i*)(dst+dst_stride4+dst_stride2+dst_stride), r7);
src1 += (stride4<<1);
src2 += (stride4<<1);
dst += (dst_stride4<<1);
}
}
void read_luma_inter_pred_avg_16x8_intrinsic( BYTE *address1, BYTE *address2, INT stride_src, BYTE *dst, INT stride_dst )
{
int src_stride = stride_src;
int dst_stride = stride_dst;
const unsigned char* src1 = address1;
const unsigned char* src2 = address2;
__declspec(align(16)) __m128i r0, r1, r2, r3, r4, r5, r6, r7,
r0_x, r1_x, r2_x, r3_x, r4_x, r5_x, r6_x, r7_x;
int stride2 = (src_stride<<1);
int stride4 = (src_stride<<2);
int dst_stride2 = (dst_stride<<1);
int dst_stride4 = (dst_stride<<2);
r0 = _mm_loadu_si128((__m128i*)(src1));
r1 = _mm_loadu_si128((__m128i*)(src1+src_stride));
r2 = _mm_loadu_si128((__m128i*)(src1+stride2));
r3 = _mm_loadu_si128((__m128i*)(src1+stride2+src_stride));
r4 = _mm_loadu_si128((__m128i*)(src1+stride4));
r5 = _mm_loadu_si128((__m128i*)(src1+stride4+src_stride));
r6 = _mm_loadu_si128((__m128i*)(src1+stride4+stride2));
r7 = _mm_loadu_si128((__m128i*)(src1+stride4+stride2+src_stride));
r0_x = _mm_loadu_si128((__m128i*)(src2));
r1_x = _mm_loadu_si128((__m128i*)(src2+src_stride));
r2_x = _mm_loadu_si128((__m128i*)(src2+stride2));
r3_x = _mm_loadu_si128((__m128i*)(src2+stride2+src_stride));
r4_x = _mm_loadu_si128((__m128i*)(src2+stride4));
r5_x = _mm_loadu_si128((__m128i*)(src2+stride4+src_stride));
r6_x = _mm_loadu_si128((__m128i*)(src2+stride4+stride2));
r7_x = _mm_loadu_si128((__m128i*)(src2+stride4+stride2+src_stride));
r0 = _mm_avg_epu8(r0, r0_x);
r1 = _mm_avg_epu8(r1, r1_x);
r2 = _mm_avg_epu8(r2, r2_x);
r3 = _mm_avg_epu8(r3, r3_x);
r4 = _mm_avg_epu8(r4, r4_x);
r5 = _mm_avg_epu8(r5, r5_x);
r6 = _mm_avg_epu8(r6, r6_x);
r7 = _mm_avg_epu8(r7, r7_x);
_mm_storeu_si128((__m128i*)(dst), r0);
_mm_storeu_si128((__m128i*)(dst+dst_stride), r1);
_mm_storeu_si128((__m128i*)(dst+dst_stride2), r2);
_mm_storeu_si128((__m128i*)(dst+dst_stride2+dst_stride), r3);
_mm_storeu_si128((__m128i*)(dst+dst_stride4), r4);
_mm_storeu_si128((__m128i*)(dst+dst_stride4+dst_stride), r5);
_mm_storeu_si128((__m128i*)(dst+dst_stride4+dst_stride2), r6);
_mm_storeu_si128((__m128i*)(dst+dst_stride4+dst_stride2+dst_stride), r7);
src1 += (stride4<<1);
src2 += (stride4<<1);
dst += (dst_stride4<<1);
}
void read_luma_inter_pred_avg_8x8_intrinsic( BYTE *address1, BYTE *address2, INT stride_src, BYTE *dst, INT stride_dst )
{
int src_stride = stride_src;
int dst_stride = stride_dst;
const unsigned char* src1 = address1;
const unsigned char* src2 = address2;
__declspec(align(16)) __m128i r0, r1, r2, r3, r4, r5, r6, r7,
r0_x, r1_x, r2_x, r3_x, r4_x, r5_x, r6_x, r7_x;
int stride2 = (src_stride<<1);
int stride4 = (src_stride<<2);
int dst_stride2 = (dst_stride<<1);
int dst_stride4 = (dst_stride<<2);
r0 = _mm_loadl_epi64((__m128i*)(src1));
r1 = _mm_loadl_epi64((__m128i*)(src1+src_stride));
r2 = _mm_loadl_epi64((__m128i*)(src1+stride2));
r3 = _mm_loadl_epi64((__m128i*)(src1+stride2+src_stride));
r4 = _mm_loadl_epi64((__m128i*)(src1+stride4));
r5 = _mm_loadl_epi64((__m128i*)(src1+stride4+src_stride));
r6 = _mm_loadl_epi64((__m128i*)(src1+stride4+stride2));
r7 = _mm_loadl_epi64((__m128i*)(src1+stride4+stride2+src_stride));
r0_x = _mm_loadl_epi64((__m128i*)(src2));
r1_x = _mm_loadl_epi64((__m128i*)(src2+src_stride));
r2_x = _mm_loadl_epi64((__m128i*)(src2+stride2));
r3_x = _mm_loadl_epi64((__m128i*)(src2+stride2+src_stride));
r4_x = _mm_loadl_epi64((__m128i*)(src2+stride4));
r5_x = _mm_loadl_epi64((__m128i*)(src2+stride4+src_stride));
r6_x = _mm_loadl_epi64((__m128i*)(src2+stride4+stride2));
r7_x = _mm_loadl_epi64((__m128i*)(src2+stride4+stride2+src_stride));
r0 = _mm_avg_epu8(r0, r0_x);
r1 = _mm_avg_epu8(r1, r1_x);
r2 = _mm_avg_epu8(r2, r2_x);
r3 = _mm_avg_epu8(r3, r3_x);
r4 = _mm_avg_epu8(r4, r4_x);
r5 = _mm_avg_epu8(r5, r5_x);
r6 = _mm_avg_epu8(r6, r6_x);
r7 = _mm_avg_epu8(r7, r7_x);
_mm_storel_epi64((__m128i*)(dst), r0);
_mm_storel_epi64((__m128i*)(dst+dst_stride), r1);
_mm_storel_epi64((__m128i*)(dst+dst_stride2), r2);
_mm_storel_epi64((__m128i*)(dst+dst_stride2+dst_stride), r3);
_mm_storel_epi64((__m128i*)(dst+dst_stride4), r4);
_mm_storel_epi64((__m128i*)(dst+dst_stride4+dst_stride), r5);
_mm_storel_epi64((__m128i*)(dst+dst_stride4+dst_stride2), r6);
_mm_storel_epi64((__m128i*)(dst+dst_stride4+dst_stride2+dst_stride), r7);
src1 += (stride4<<1);
src2 += (stride4<<1);
dst += (dst_stride4<<1);
}
void DoSpeedTest(unsigned char* address1,int stride1, unsigned char* address2,int stride2,unsigned int* orio,unsigned int* newo)
{
UINT so,eo,sn,en;
UINT diffo,diffn,diffnc;
diffo = 0;
diffn = 0;
diffnc = 0;
for( int loop=0;loop<LOOP_TEST_COUNT;loop++)
{
so = GetTickCount();
//do a full search motion estimation. This does not require to be correct. It is a speed test
for( int mby = 0; mby < MB_Y_REF; mby++ )
for( int mbx = 0; mbx < MB_X_REF; mbx++ )
{
int index = ( mby * MB_SIZE ) * REF_STRIDE + mbx * MB_SIZE;
BYTE* refa = address1 + index;
BYTE* cura = address2 + index;
BYTE *out = (BYTE*)orio + index;
for( int loop=0;loop<LOOP_TEST_COUNT;loop++)
// read_luma_inter_pred_avg_16x16_C( refa, cura, stride1, (BYTE*)out, stride2 );
// read_luma_inter_pred_avg_16x16_intrinsic( refa, cura, stride1, (BYTE*)out, stride2 );
// read_luma_inter_pred_avg_8x16_intrinsic( refa, cura, stride1, (BYTE*)out, stride2 );
// read_luma_inter_pred_avg_16x8_intrinsic( refa, cura, stride1, (BYTE*)out, stride2 );
read_luma_inter_pred_avg_8x8_intrinsic( refa, cura, stride1, (BYTE*)out, stride2 );
}
eo = GetTickCount();
diffo += eo - so;
sn = GetTickCount();
for( int mby = 0; mby < MB_Y_REF; mby++ )
for( int mbx = 0; mbx < MB_X_REF; mbx++ )
{
int index = ( mby * MB_SIZE ) * REF_STRIDE + mbx * MB_SIZE;
BYTE* refa = address1 + index;
BYTE* cura = address2 + index;
BYTE *out = (BYTE*)newo + index;
for( int loop=0;loop<LOOP_TEST_COUNT;loop++)
// read_luma_inter_pred_avg_16x16_SSE3( refa, cura, stride1, (BYTE*)out, stride2 );
// read_luma_inter_pred_avg_8x16_SSE3( refa, cura, stride1, (BYTE*)out, stride2 );
// read_luma_inter_pred_avg_16x8_SSE3( refa, cura, stride1, (BYTE*)out, stride2 );
read_luma_inter_pred_avg_8x8_SSE3( refa, cura, stride1, (BYTE*)out, stride2 );
}
en = GetTickCount();
diffn += en - sn;
/**/
}
printf("MS for old algorithm : %u\n",diffo);
printf("for new algorithm : %u\n",diffn);
// printf("speed diff PCT : %u\n",(diffn)*100/(diffo));
printf("Results match : %s \n", memcmp( newo,orio, MB_X_REF * MB_SIZE * MB_Y_REF * MB_SIZE * sizeof( UINT ) ) == 0 ? "true":"false");
}
int main()
{
volatile int stride = MB_X_REF * MB_SIZE;
BYTE* address1 = (BYTE*)malloc( MB_X_REF * MB_SIZE * MB_Y_REF * MB_SIZE * sizeof( BYTE ) );
BYTE* address2 = (BYTE*)malloc( MB_X_REF * MB_SIZE * MB_Y_REF * MB_SIZE * sizeof( BYTE ) );
UINT* orio = (UINT*)malloc( MB_X_REF * MB_SIZE * MB_Y_REF * MB_SIZE * sizeof( UINT ) );
UINT* newo = (UINT*)malloc( MB_X_REF * MB_SIZE * MB_Y_REF * MB_SIZE * sizeof( UINT ) );
memset( orio, 0, MB_X_REF * MB_SIZE * MB_Y_REF * MB_SIZE * sizeof( UINT ) );
memset( newo, 0, MB_X_REF * MB_SIZE * MB_Y_REF * MB_SIZE * sizeof( UINT ) );
for(int i=0;i<MB_X_REF * MB_SIZE * MB_Y_REF * MB_SIZE;i++)
{
address1[i] = i+1;
address2[i] = i;
}
DoSpeedTest( address1, stride, address2, stride, orio, newo );
getch();
free( address1 );
free( address2 );
free( orio );
free( newo );
return 0;
}