1 |
/************************************************************************** |
/***************************************************************************** |
2 |
* |
* |
3 |
* XVID MPEG-4 VIDEO CODEC - Unit tests and benches |
* XVID MPEG-4 VIDEO CODEC |
4 |
|
* - Unit tests and benches - |
5 |
|
* |
6 |
|
* Copyright(C) 2002 Pascal Massimino <skal@planet-d.net> |
7 |
* |
* |
8 |
* This program is free software; you can redistribute it and/or modify |
* This program is free software; you can redistribute it and/or modify |
9 |
* it under the terms of the GNU General Public License as published by |
* it under the terms of the GNU General Public License as published by |
17 |
* |
* |
18 |
* You should have received a copy of the GNU General Public License |
* You should have received a copy of the GNU General Public License |
19 |
* along with this program; if not, write to the Free Software |
* along with this program; if not, write to the Free Software |
20 |
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
21 |
|
* |
22 |
|
* $Id$ |
23 |
* |
* |
24 |
*************************************************************************/ |
****************************************************************************/ |
25 |
|
|
26 |
/************************************************************************ |
/***************************************************************************** |
27 |
* |
* |
28 |
* 'Reference' output is at the end of file. |
* 'Reference' output is at the end of file. |
|
* Don't take the checksums and crc too seriouly, they aren't |
|
|
* bullet-proof... |
|
29 |
* |
* |
30 |
* compiles with something like: |
* compiles with something like: |
31 |
* gcc -o xvid_bench xvid_bench.c -I../src/ -lxvidcore -lm |
* gcc -o xvid_bench xvid_bench.c -I../src/ -lxvidcore -lm |
32 |
* |
* |
33 |
* History: |
****************************************************************************/ |
|
* |
|
|
* 06.06.2002 initial coding -Skal- |
|
|
* |
|
|
*************************************************************************/ |
|
34 |
|
|
35 |
#include <stdio.h> |
#include <stdio.h> |
36 |
#include <stdlib.h> |
#include <stdlib.h> |
37 |
#include <sys/time.h> // for gettimeofday |
#include <string.h> /* for memset */ |
|
#include <string.h> // for memset |
|
38 |
#include <assert.h> |
#include <assert.h> |
39 |
|
|
40 |
|
#ifndef WIN32 |
41 |
|
#include <sys/time.h> /* for gettimeofday */ |
42 |
|
#else |
43 |
|
#include <time.h> |
44 |
|
#endif |
45 |
|
|
46 |
|
|
47 |
#include "xvid.h" |
#include "xvid.h" |
48 |
|
|
49 |
// inner guts |
// inner guts |
52 |
#include "image/colorspace.h" |
#include "image/colorspace.h" |
53 |
#include "image/interpolate8x8.h" |
#include "image/interpolate8x8.h" |
54 |
#include "utils/mem_transfer.h" |
#include "utils/mem_transfer.h" |
55 |
#include "quant/quant_h263.h" |
#include "quant/quant.h" |
|
#include "quant/quant_mpeg4.h" |
|
56 |
#include "motion/sad.h" |
#include "motion/sad.h" |
57 |
#include "utils/emms.h" |
#include "utils/emms.h" |
58 |
#include "utils/timer.h" |
#include "utils/timer.h" |
59 |
#include "quant/quant_matrix.c" |
#include "quant/quant_matrix.c" |
60 |
#include "bitstream/cbp.h" |
#include "bitstream/cbp.h" |
61 |
|
|
62 |
const int speed_ref = 100; // on slow machines, decrease this value |
#include <math.h> |
63 |
|
|
64 |
|
#ifndef M_PI |
65 |
|
#define M_PI 3.14159265358979323846 |
66 |
|
#endif |
67 |
|
|
68 |
|
int speed_ref = 100; /* on slow machines, decrease this value */ |
69 |
|
int verbose = 0; |
70 |
|
unsigned int cpu_mask; |
71 |
|
|
72 |
/********************************************************************* |
/********************************************************************* |
73 |
* misc |
* misc |
76 |
/* returns time in micro-s*/ |
/* returns time in micro-s*/ |
77 |
double gettime_usec() |
double gettime_usec() |
78 |
{ |
{ |
79 |
|
#ifndef WIN32 |
80 |
struct timeval tv; |
struct timeval tv; |
81 |
gettimeofday(&tv, 0); |
gettimeofday(&tv, 0); |
82 |
return tv.tv_sec*1.0e6f + tv.tv_usec; |
return tv.tv_sec*1.0e6 + tv.tv_usec; |
83 |
|
#else |
84 |
|
clock_t clk; |
85 |
|
clk = clock(); |
86 |
|
return clk * 1000. / CLOCKS_PER_SEC; /* clock() returns time in Milliseconds */ |
87 |
|
#endif |
88 |
} |
} |
89 |
|
|
90 |
/* returns squared deviates (mean(v*v)-mean(v)^2) of a 8x8 block */ |
/* returns squared deviates (mean(v*v)-mean(v)^2) of a 8x8 block */ |
112 |
unsigned int cpu; |
unsigned int cpu; |
113 |
} CPU; |
} CPU; |
114 |
|
|
115 |
CPU cpu_list[] = |
CPU cpu_list[] = { |
116 |
{ { "PLAINC", 0 } |
{ "PLAINC ", 0 }, |
117 |
, { "MMX ", XVID_CPU_MMX } |
#ifdef ARCH_IS_IA32 |
118 |
, { "MMXEXT", XVID_CPU_MMXEXT | XVID_CPU_MMX } |
{ "MMX ", XVID_CPU_MMX }, |
119 |
, { "SSE2 ", XVID_CPU_SSE2 | XVID_CPU_MMX } |
{ "MMXEXT ", XVID_CPU_MMXEXT | XVID_CPU_MMX }, |
120 |
, { "3DNOW ", XVID_CPU_3DNOW } |
{ "SSE2 ", XVID_CPU_SSE2 | XVID_CPU_MMX }, |
121 |
, { "3DNOWE", XVID_CPU_3DNOWEXT } |
{ "3DNOW ", XVID_CPU_3DNOW }, |
122 |
//, { "TSC ", XVID_CPU_TSC } |
{ "3DNOWE ", XVID_CPU_3DNOW | XVID_CPU_3DNOWEXT }, |
123 |
, { 0, 0 } } |
#endif |
124 |
|
#ifdef ARCH_IS_PPC |
125 |
, cpu_short_list[] = |
{ "ALTIVEC", XVID_CPU_ALTIVEC }, |
126 |
{ { "PLAINC", 0 } |
#endif |
127 |
, { "MMX ", XVID_CPU_MMX } |
#ifdef ARCH_IS_X86_64 |
128 |
, { "MMXEXT", XVID_CPU_MMXEXT | XVID_CPU_MMX } |
{ "X86_64 ", XVID_CPU_ASM}, |
129 |
, { 0, 0 } } |
#endif |
130 |
|
#ifdef ARCH_IS_IA64 |
131 |
, cpu_short_list2[] = |
// { "IA64 ", XVID_CPU_IA64 }, |
132 |
{ { "PLAINC", 0 } |
#endif |
133 |
, { "MMX ", XVID_CPU_MMX } |
// { "TSC ", XVID_CPU_TSC }, |
134 |
, { "SSE2 ", XVID_CPU_SSE2 | XVID_CPU_MMX } |
{ 0, 0 } |
135 |
, { 0, 0 } }; |
}; |
136 |
|
|
137 |
|
|
138 |
int init_cpu(CPU *cpu) |
int init_cpu(CPU *cpu) |
139 |
{ |
{ |
140 |
int xerr, cpu_type; |
xvid_gbl_info_t xinfo; |
141 |
XVID_INIT_PARAM xinit; |
|
142 |
|
/* Get the available CPU flags */ |
143 |
|
memset(&xinfo, 0, sizeof(xinfo)); |
144 |
|
xinfo.version = XVID_VERSION; |
145 |
|
xvid_global(NULL, XVID_GBL_INFO, &xinfo, NULL); |
146 |
|
|
147 |
cpu_type = check_cpu_features() & cpu->cpu; |
/* Are we trying to test a subset of the host CPU features */ |
148 |
xinit.cpu_flags = cpu_type | XVID_CPU_FORCE; |
if ((xinfo.cpu_flags & cpu->cpu) == cpu->cpu) { |
149 |
// xinit.cpu_flags = XVID_CPU_MMX | XVID_CPU_FORCE; |
int xerr; |
150 |
xerr = xvid_init(NULL, 0, &xinit, NULL); |
xvid_gbl_init_t xinit; |
151 |
if (cpu->cpu>0 && (cpu_type==0 || xerr!=XVID_ERR_OK)) { |
memset(&xinit, 0, sizeof(xinit)); |
152 |
printf( "%s - skipped...\n", cpu->name ); |
xinit.cpu_flags = cpu->cpu | XVID_CPU_FORCE; |
153 |
|
xinit.version = XVID_VERSION; |
154 |
|
xerr = xvid_global(NULL, XVID_GBL_INIT, &xinit, NULL); |
155 |
|
if (xerr==XVID_ERR_FAIL) { |
156 |
|
/* libxvidcore failed to init */ |
157 |
return 0; |
return 0; |
158 |
} |
} |
159 |
|
} else { |
160 |
|
/* The host CPU doesn't support some required feature for this test */ |
161 |
|
return(0); |
162 |
|
} |
163 |
return 1; |
return 1; |
164 |
} |
} |
165 |
|
|
166 |
|
#define CRC32_REMAINDER 0xCBF43926 |
167 |
|
#define CRC32_INITIAL 0xffffffff |
168 |
|
|
169 |
|
#define DO1(c, crc) ((crc) = crc32tab[((unsigned int)((crc)>>24) ^ (*c++)) & 0xff] ^ ((crc) << 8)) |
170 |
|
#define DO2(c, crc) DO1(c, crc); DO1(c, crc); |
171 |
|
#define DO4(c, crc) DO2(c, crc); DO2(c, crc); |
172 |
|
#define DO8(c, crc) DO4(c, crc); DO4(c, crc); |
173 |
|
|
174 |
|
/****************************************************************************** |
175 |
|
* Precomputed AAL5 CRC32 lookup table |
176 |
|
******************************************************************************/ |
177 |
|
|
178 |
|
static unsigned long crc32tab[256] = { |
179 |
|
|
180 |
|
0x00000000L, 0x04C11DB7L, 0x09823B6EL, 0x0D4326D9L, |
181 |
|
0x130476DCL, 0x17C56B6BL, 0x1A864DB2L, 0x1E475005L, |
182 |
|
0x2608EDB8L, 0x22C9F00FL, 0x2F8AD6D6L, 0x2B4BCB61L, |
183 |
|
0x350C9B64L, 0x31CD86D3L, 0x3C8EA00AL, 0x384FBDBDL, |
184 |
|
0x4C11DB70L, 0x48D0C6C7L, 0x4593E01EL, 0x4152FDA9L, |
185 |
|
0x5F15ADACL, 0x5BD4B01BL, 0x569796C2L, 0x52568B75L, |
186 |
|
0x6A1936C8L, 0x6ED82B7FL, 0x639B0DA6L, 0x675A1011L, |
187 |
|
0x791D4014L, 0x7DDC5DA3L, 0x709F7B7AL, 0x745E66CDL, |
188 |
|
0x9823B6E0L, 0x9CE2AB57L, 0x91A18D8EL, 0x95609039L, |
189 |
|
0x8B27C03CL, 0x8FE6DD8BL, 0x82A5FB52L, 0x8664E6E5L, |
190 |
|
0xBE2B5B58L, 0xBAEA46EFL, 0xB7A96036L, 0xB3687D81L, |
191 |
|
0xAD2F2D84L, 0xA9EE3033L, 0xA4AD16EAL, 0xA06C0B5DL, |
192 |
|
0xD4326D90L, 0xD0F37027L, 0xDDB056FEL, 0xD9714B49L, |
193 |
|
0xC7361B4CL, 0xC3F706FBL, 0xCEB42022L, 0xCA753D95L, |
194 |
|
0xF23A8028L, 0xF6FB9D9FL, 0xFBB8BB46L, 0xFF79A6F1L, |
195 |
|
0xE13EF6F4L, 0xE5FFEB43L, 0xE8BCCD9AL, 0xEC7DD02DL, |
196 |
|
0x34867077L, 0x30476DC0L, 0x3D044B19L, 0x39C556AEL, |
197 |
|
0x278206ABL, 0x23431B1CL, 0x2E003DC5L, 0x2AC12072L, |
198 |
|
0x128E9DCFL, 0x164F8078L, 0x1B0CA6A1L, 0x1FCDBB16L, |
199 |
|
0x018AEB13L, 0x054BF6A4L, 0x0808D07DL, 0x0CC9CDCAL, |
200 |
|
0x7897AB07L, 0x7C56B6B0L, 0x71159069L, 0x75D48DDEL, |
201 |
|
0x6B93DDDBL, 0x6F52C06CL, 0x6211E6B5L, 0x66D0FB02L, |
202 |
|
0x5E9F46BFL, 0x5A5E5B08L, 0x571D7DD1L, 0x53DC6066L, |
203 |
|
0x4D9B3063L, 0x495A2DD4L, 0x44190B0DL, 0x40D816BAL, |
204 |
|
0xACA5C697L, 0xA864DB20L, 0xA527FDF9L, 0xA1E6E04EL, |
205 |
|
0xBFA1B04BL, 0xBB60ADFCL, 0xB6238B25L, 0xB2E29692L, |
206 |
|
0x8AAD2B2FL, 0x8E6C3698L, 0x832F1041L, 0x87EE0DF6L, |
207 |
|
0x99A95DF3L, 0x9D684044L, 0x902B669DL, 0x94EA7B2AL, |
208 |
|
0xE0B41DE7L, 0xE4750050L, 0xE9362689L, 0xEDF73B3EL, |
209 |
|
0xF3B06B3BL, 0xF771768CL, 0xFA325055L, 0xFEF34DE2L, |
210 |
|
0xC6BCF05FL, 0xC27DEDE8L, 0xCF3ECB31L, 0xCBFFD686L, |
211 |
|
0xD5B88683L, 0xD1799B34L, 0xDC3ABDEDL, 0xD8FBA05AL, |
212 |
|
0x690CE0EEL, 0x6DCDFD59L, 0x608EDB80L, 0x644FC637L, |
213 |
|
0x7A089632L, 0x7EC98B85L, 0x738AAD5CL, 0x774BB0EBL, |
214 |
|
0x4F040D56L, 0x4BC510E1L, 0x46863638L, 0x42472B8FL, |
215 |
|
0x5C007B8AL, 0x58C1663DL, 0x558240E4L, 0x51435D53L, |
216 |
|
0x251D3B9EL, 0x21DC2629L, 0x2C9F00F0L, 0x285E1D47L, |
217 |
|
0x36194D42L, 0x32D850F5L, 0x3F9B762CL, 0x3B5A6B9BL, |
218 |
|
0x0315D626L, 0x07D4CB91L, 0x0A97ED48L, 0x0E56F0FFL, |
219 |
|
0x1011A0FAL, 0x14D0BD4DL, 0x19939B94L, 0x1D528623L, |
220 |
|
0xF12F560EL, 0xF5EE4BB9L, 0xF8AD6D60L, 0xFC6C70D7L, |
221 |
|
0xE22B20D2L, 0xE6EA3D65L, 0xEBA91BBCL, 0xEF68060BL, |
222 |
|
0xD727BBB6L, 0xD3E6A601L, 0xDEA580D8L, 0xDA649D6FL, |
223 |
|
0xC423CD6AL, 0xC0E2D0DDL, 0xCDA1F604L, 0xC960EBB3L, |
224 |
|
0xBD3E8D7EL, 0xB9FF90C9L, 0xB4BCB610L, 0xB07DABA7L, |
225 |
|
0xAE3AFBA2L, 0xAAFBE615L, 0xA7B8C0CCL, 0xA379DD7BL, |
226 |
|
0x9B3660C6L, 0x9FF77D71L, 0x92B45BA8L, 0x9675461FL, |
227 |
|
0x8832161AL, 0x8CF30BADL, 0x81B02D74L, 0x857130C3L, |
228 |
|
0x5D8A9099L, 0x594B8D2EL, 0x5408ABF7L, 0x50C9B640L, |
229 |
|
0x4E8EE645L, 0x4A4FFBF2L, 0x470CDD2BL, 0x43CDC09CL, |
230 |
|
0x7B827D21L, 0x7F436096L, 0x7200464FL, 0x76C15BF8L, |
231 |
|
0x68860BFDL, 0x6C47164AL, 0x61043093L, 0x65C52D24L, |
232 |
|
0x119B4BE9L, 0x155A565EL, 0x18197087L, 0x1CD86D30L, |
233 |
|
0x029F3D35L, 0x065E2082L, 0x0B1D065BL, 0x0FDC1BECL, |
234 |
|
0x3793A651L, 0x3352BBE6L, 0x3E119D3FL, 0x3AD08088L, |
235 |
|
0x2497D08DL, 0x2056CD3AL, 0x2D15EBE3L, 0x29D4F654L, |
236 |
|
0xC5A92679L, 0xC1683BCEL, 0xCC2B1D17L, 0xC8EA00A0L, |
237 |
|
0xD6AD50A5L, 0xD26C4D12L, 0xDF2F6BCBL, 0xDBEE767CL, |
238 |
|
0xE3A1CBC1L, 0xE760D676L, 0xEA23F0AFL, 0xEEE2ED18L, |
239 |
|
0xF0A5BD1DL, 0xF464A0AAL, 0xF9278673L, 0xFDE69BC4L, |
240 |
|
0x89B8FD09L, 0x8D79E0BEL, 0x803AC667L, 0x84FBDBD0L, |
241 |
|
0x9ABC8BD5L, 0x9E7D9662L, 0x933EB0BBL, 0x97FFAD0CL, |
242 |
|
0xAFB010B1L, 0xAB710D06L, 0xA6322BDFL, 0xA2F33668L, |
243 |
|
0xBCB4666DL, 0xB8757BDAL, 0xB5365D03L, 0xB1F740B4L |
244 |
|
|
245 |
|
}; |
246 |
|
|
247 |
|
uint32_t |
248 |
|
calc_crc(uint8_t *mem, int len, uint32_t crc) |
249 |
|
{ |
250 |
|
while( len >= 8) { |
251 |
|
DO8(mem, crc); |
252 |
|
len -= 8; |
253 |
|
} |
254 |
|
|
255 |
|
while( len ) { |
256 |
|
DO1(mem, crc); |
257 |
|
len--; |
258 |
|
} |
259 |
|
|
260 |
|
return crc; |
261 |
|
} |
262 |
|
|
263 |
|
void byte_swap(uint8_t *mem, int len, int element_size) { |
264 |
|
#ifdef ARCH_IS_BIG_ENDIAN |
265 |
|
int i; |
266 |
|
|
267 |
|
if(element_size == 1) { |
268 |
|
/* No need to swap */ |
269 |
|
} else if(element_size == 2) { |
270 |
|
uint8_t temp[2]; |
271 |
|
|
272 |
|
for(i=0; i < (len/2); i++ ) { |
273 |
|
temp[0] = mem[0]; |
274 |
|
temp[1] = mem[1]; |
275 |
|
mem[0] = temp[1]; |
276 |
|
mem[1] = temp[0]; |
277 |
|
|
278 |
|
mem += 2; |
279 |
|
} |
280 |
|
} else if(element_size == 4) { |
281 |
|
uint8_t temp[4]; |
282 |
|
|
283 |
|
for(i=0; i < (len/4); i++ ) { |
284 |
|
temp[0] = mem[0]; |
285 |
|
temp[1] = mem[1]; |
286 |
|
temp[2] = mem[2]; |
287 |
|
temp[3] = mem[3]; |
288 |
|
mem[0] = temp[3]; |
289 |
|
mem[1] = temp[2]; |
290 |
|
mem[2] = temp[1]; |
291 |
|
mem[3] = temp[0]; |
292 |
|
|
293 |
|
mem += 4; |
294 |
|
} |
295 |
|
} else { |
296 |
|
printf("ERROR: byte_swap unsupported element_size(%u)\n", element_size); |
297 |
|
} |
298 |
|
#endif |
299 |
|
} |
300 |
|
|
301 |
/********************************************************************* |
/********************************************************************* |
302 |
* test DCT |
* test DCT |
303 |
*********************************************************************/ |
*********************************************************************/ |
310 |
int tst; |
int tst; |
311 |
CPU *cpu; |
CPU *cpu; |
312 |
int i; |
int i; |
313 |
short iDst0[8*8], iDst[8*8], fDst[8*8]; |
DECLARE_ALIGNED_MATRIX(iDst0, 8, 8, short, 16); |
314 |
|
DECLARE_ALIGNED_MATRIX(iDst, 8, 8, short, 16); |
315 |
|
DECLARE_ALIGNED_MATRIX(fDst, 8, 8, short, 16); |
316 |
double overhead; |
double overhead; |
317 |
|
|
318 |
printf( "\n ===== test fdct/idct =====\n" ); |
printf( "\n ===== test fdct/idct =====\n" ); |
328 |
|
|
329 |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
330 |
{ |
{ |
331 |
double t; |
double t, PSNR, MSE; |
|
int iCrc, fCrc; |
|
332 |
|
|
333 |
if (!init_cpu(cpu)) |
if (!init_cpu(cpu)) |
334 |
continue; |
continue; |
344 |
} |
} |
345 |
emms(); |
emms(); |
346 |
t = (gettime_usec() - t - overhead) / nb_tests; |
t = (gettime_usec() - t - overhead) / nb_tests; |
347 |
iCrc=0; fCrc=0; |
MSE = 0.; |
348 |
for(i=0; i<8*8; ++i) { |
for(i=0; i<8*8; ++i) { |
349 |
iCrc += ABS(iDst[i] - iDst0[i]); |
double delta = 1.0*(iDst[i] - iDst0[i]); |
350 |
fCrc += fDst[i]^i; |
MSE += delta*delta; |
351 |
} |
} |
352 |
printf( "%s - %.3f usec iCrc=%d fCrc=%d\n", |
PSNR = (MSE==0.) ? 1.e6 : -4.3429448*log( MSE/64. ); |
353 |
cpu->name, t, iCrc, fCrc ); |
printf( "%s - %.3f usec PSNR=%.3f MSE=%.3f %s\n", |
354 |
// the norm tolerates ~1 bit of diff per coeff |
cpu->name, t, PSNR, MSE, |
355 |
if (ABS(iCrc)>=64) printf( "*** CRC ERROR! ***\n" ); |
(ABS(MSE)>=64)? "| ERROR" :""); |
356 |
} |
} |
357 |
} |
} |
358 |
|
|
366 |
int tst; |
int tst; |
367 |
CPU *cpu; |
CPU *cpu; |
368 |
int i; |
int i; |
369 |
uint8_t Cur[16*16], Ref1[16*16], Ref2[16*16]; |
DECLARE_ALIGNED_MATRIX(Cur, 16, 16, uint8_t, 16); |
370 |
|
DECLARE_ALIGNED_MATRIX(Ref1, 16, 16, uint8_t, 16); |
371 |
|
DECLARE_ALIGNED_MATRIX(Ref2, 16, 16, uint8_t, 16); |
372 |
|
|
373 |
printf( "\n ====== test SAD ======\n" ); |
printf( "\n ====== test SAD ======\n" ); |
374 |
for(i=0; i<16*16;++i) { |
for(i=0; i<16*16;++i) { |
389 |
for(tst=0; tst<nb_tests; ++tst) s = sad8(Cur, Ref1, 16); |
for(tst=0; tst<nb_tests; ++tst) s = sad8(Cur, Ref1, 16); |
390 |
emms(); |
emms(); |
391 |
t = (gettime_usec() - t) / nb_tests; |
t = (gettime_usec() - t) / nb_tests; |
392 |
printf( "%s - sad8 %.3f usec sad=%d\n", cpu->name, t, s ); |
printf("%s - sad8 %.3f usec sad=%d %s\n", |
393 |
if (s!=3776) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, s, |
394 |
|
(s!=3776)?"| ERROR": "" ); |
395 |
|
|
396 |
t = gettime_usec(); |
t = gettime_usec(); |
397 |
emms(); |
emms(); |
398 |
for(tst=0; tst<nb_tests; ++tst) s = sad16(Cur, Ref1, 16, -1); |
for(tst=0; tst<nb_tests; ++tst) s = sad16(Cur, Ref1, 16, -1); |
399 |
emms(); |
emms(); |
400 |
t = (gettime_usec() - t) / nb_tests; |
t = (gettime_usec() - t) / nb_tests; |
401 |
printf( "%s - sad16 %.3f usec sad=%d\n", cpu->name, t, s ); |
printf("%s - sad16 %.3f usec sad=%d %s\n", |
402 |
if (s!=27214) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, s, |
403 |
|
(s!=27214)?"| ERROR": "" ); |
404 |
|
|
405 |
t = gettime_usec(); |
t = gettime_usec(); |
406 |
emms(); |
emms(); |
407 |
for(tst=0; tst<nb_tests; ++tst) s = sad16bi(Cur, Ref1, Ref2, 16); |
for(tst=0; tst<nb_tests; ++tst) s = sad16bi(Cur, Ref1, Ref2, 16); |
408 |
emms(); |
emms(); |
409 |
t = (gettime_usec() - t) / nb_tests; |
t = (gettime_usec() - t) / nb_tests; |
410 |
printf( "%s - sad16bi %.3f usec sad=%d\n", cpu->name, t, s ); |
printf( "%s - sad16bi %.3f usec sad=%d %s\n", |
411 |
if (s!=26274) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, s, |
412 |
|
(s!=26274)?"| ERROR": "" ); |
413 |
|
|
414 |
t = gettime_usec(); |
t = gettime_usec(); |
415 |
emms(); |
emms(); |
416 |
for(tst=0; tst<nb_tests; ++tst) s = dev16(Cur, 16); |
for(tst=0; tst<nb_tests; ++tst) s = dev16(Cur, 16); |
417 |
emms(); |
emms(); |
418 |
t = (gettime_usec() - t) / nb_tests; |
t = (gettime_usec() - t) / nb_tests; |
419 |
printf( "%s - dev16 %.3f usec sad=%d\n", cpu->name, t, s ); |
printf( "%s - dev16 %.3f usec sad=%d %s\n", |
420 |
if (s!=3344) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, s, |
421 |
|
(s!=3344)?"| ERROR": "" ); |
422 |
|
|
423 |
printf( " --- \n" ); |
printf( " --- \n" ); |
424 |
} |
} |
436 |
#define LEAVE \ |
#define LEAVE \ |
437 |
emms(); \ |
emms(); \ |
438 |
t = (gettime_usec() - t) / nb_tests; \ |
t = (gettime_usec() - t) / nb_tests; \ |
439 |
iCrc = 0; \ |
iCrc = calc_crc((uint8_t*)Dst, sizeof(Dst), CRC32_INITIAL) |
|
for(i=0; i<16*8; ++i) { iCrc += Dst[i]^i; } |
|
440 |
|
|
441 |
#define TEST_MB(FUNC, R) \ |
#define TEST_MB(FUNC, R) \ |
442 |
ENTER \ |
ENTER \ |
454 |
const int nb_tests = 2000*speed_ref; |
const int nb_tests = 2000*speed_ref; |
455 |
CPU *cpu; |
CPU *cpu; |
456 |
const uint8_t Src0[16*9] = { |
const uint8_t Src0[16*9] = { |
457 |
// try to have every possible combinaison of rounding... |
/* try to have every possible combinaison of rounding... */ |
458 |
0, 0, 1, 0, 2, 0, 3, 0, 4 ,0,0,0, 0,0,0,0 |
0, 0, 1, 0, 2, 0, 3, 0, 4 ,0,0,0, 0,0,0,0, |
459 |
, 0, 1, 1, 1, 2, 1, 3, 1, 3 ,0,0,0, 0,0,0,0 |
0, 1, 1, 1, 2, 1, 3, 1, 3 ,0,0,0, 0,0,0,0, |
460 |
, 0, 2, 1, 2, 2, 2, 3, 2, 2 ,0,0,0, 0,0,0,0 |
0, 2, 1, 2, 2, 2, 3, 2, 2 ,0,0,0, 0,0,0,0, |
461 |
, 0, 3, 1, 3, 2, 3, 3, 3, 1 ,0,0,0, 0,0,0,0 |
0, 3, 1, 3, 2, 3, 3, 3, 1 ,0,0,0, 0,0,0,0, |
462 |
, 1, 3, 0, 2, 1, 0, 2, 3, 4 ,0,0,0, 0,0,0,0 |
1, 3, 0, 2, 1, 0, 2, 3, 4 ,0,0,0, 0,0,0,0, |
463 |
, 2, 2, 1, 2, 0, 1, 3, 5, 3 ,0,0,0, 0,0,0,0 |
2, 2, 1, 2, 0, 1, 3, 5, 3 ,0,0,0, 0,0,0,0, |
464 |
, 3, 1, 2, 3, 1, 2, 2, 6, 2 ,0,0,0, 0,0,0,0 |
3, 1, 2, 3, 1, 2, 2, 6, 2 ,0,0,0, 0,0,0,0, |
465 |
, 1, 0, 1, 3, 0, 3, 1, 6, 1 ,0,0,0, 0,0,0,0 |
1, 0, 1, 3, 0, 3, 1, 6, 1 ,0,0,0, 0,0,0,0, |
466 |
, 4, 3, 2, 1, 2, 3, 4, 0, 3 ,0,0,0, 0,0,0,0 |
4, 3, 2, 1, 2, 3, 4, 0, 3 ,0,0,0, 0,0,0,0 |
467 |
}; |
}; |
468 |
uint8_t Dst[16*8] = {0}; |
uint8_t Dst[16*8] = {0}; |
469 |
|
|
478 |
continue; |
continue; |
479 |
|
|
480 |
TEST_MB(interpolate8x8_halfpel_h, 0); |
TEST_MB(interpolate8x8_halfpel_h, 0); |
481 |
printf( "%s - interp- h-round0 %.3f usec iCrc=%d\n", cpu->name, t, iCrc ); |
printf("%s - interp- h-round0 %.3f usec crc32=0x%08x %s\n", |
482 |
if (iCrc!=8107) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, iCrc, |
483 |
|
(iCrc!=0x115381ba)?"| ERROR": "" ); |
484 |
|
|
485 |
TEST_MB(interpolate8x8_halfpel_h, 1); |
TEST_MB(interpolate8x8_halfpel_h, 1); |
486 |
printf( "%s - round1 %.3f usec iCrc=%d\n", cpu->name, t, iCrc ); |
printf("%s - round1 %.3f usec crc32=0x%08x %s\n", |
487 |
if (iCrc!=8100) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, iCrc, |
488 |
|
(iCrc!=0x2b1f528f)?"| ERROR": "" ); |
489 |
|
|
490 |
|
|
491 |
TEST_MB(interpolate8x8_halfpel_v, 0); |
TEST_MB(interpolate8x8_halfpel_v, 0); |
492 |
printf( "%s - interp- v-round0 %.3f usec iCrc=%d\n", cpu->name, t, iCrc ); |
printf("%s - interp- v-round0 %.3f usec crc32=0x%08x %s\n", |
493 |
if (iCrc!=8108) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, iCrc, |
494 |
|
(iCrc!=0x423cdcc7)?"| ERROR": "" ); |
495 |
|
|
496 |
TEST_MB(interpolate8x8_halfpel_v, 1); |
TEST_MB(interpolate8x8_halfpel_v, 1); |
497 |
printf( "%s - round1 %.3f usec iCrc=%d\n", cpu->name, t, iCrc ); |
printf("%s - round1 %.3f usec crc32=0x%08x %s\n", |
498 |
if (iCrc!=8105) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, iCrc, |
499 |
|
(iCrc!=0x42202efe)?"| ERROR": "" ); |
500 |
|
|
501 |
|
|
502 |
TEST_MB(interpolate8x8_halfpel_hv, 0); |
TEST_MB(interpolate8x8_halfpel_hv, 0); |
503 |
printf( "%s - interp-hv-round0 %.3f usec iCrc=%d\n", cpu->name, t, iCrc ); |
printf("%s - interp-hv-round0 %.3f usec crc32=0x%08x %s\n", |
504 |
if (iCrc!=8112) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, iCrc, |
505 |
|
(iCrc!=0xd198d387)?"| ERROR": "" ); |
506 |
|
|
507 |
TEST_MB(interpolate8x8_halfpel_hv, 1); |
TEST_MB(interpolate8x8_halfpel_hv, 1); |
508 |
printf( "%s - round1 %.3f usec iCrc=%d\n", cpu->name, t, iCrc ); |
printf("%s - round1 %.3f usec crc32=0x%08x %s\n", |
509 |
if (iCrc!=8103) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, iCrc, |
510 |
|
(iCrc!=0x9ecfd921)?"| ERROR": "" ); |
511 |
|
|
512 |
|
|
513 |
|
/* this is a new function, as of 06.06.2002 */ |
514 |
|
#if 0 |
515 |
|
TEST_MB2(interpolate8x8_avrg); |
516 |
|
printf("%s - interpolate8x8_c %.3f usec crc32=0x%08x %s\n", |
517 |
|
cpu->name, t, iCrc, |
518 |
|
(iCrc!=8107)?"| ERROR": "" ); |
519 |
|
#endif |
520 |
|
|
521 |
|
/* New functions for field prediction by CK 1.10.2005 */ |
522 |
|
#pragma NEW8X4 |
523 |
|
TEST_MB(interpolate8x4_halfpel_h, 0); |
524 |
|
printf("%s - interpfield-h -round0 %.3f usec crc32=0x%08x %s\n", |
525 |
|
cpu->name, t, iCrc, |
526 |
|
(iCrc!=0x9538d6df)?"| ERROR": "" ); |
527 |
|
|
528 |
|
TEST_MB(interpolate8x4_halfpel_h, 1); |
529 |
|
printf("%s - round1 %.3f usec crc32=0x%08x %s\n", |
530 |
|
cpu->name, t, iCrc, |
531 |
|
(iCrc!=0xde5f1db4)?"| ERROR": "" ); |
532 |
|
|
533 |
|
|
534 |
|
TEST_MB(interpolate8x4_halfpel_v, 0); |
535 |
|
printf("%s - interpfield- v-round0 %.3f usec crc32=0x%08x %s\n", |
536 |
|
cpu->name, t, iCrc, |
537 |
|
(iCrc!=0xea5a69ef)?"| ERROR": "" ); |
538 |
|
|
539 |
|
TEST_MB(interpolate8x4_halfpel_v, 1); |
540 |
|
printf("%s - round1 %.3f usec crc32=0x%08x %s\n", |
541 |
|
cpu->name, t, iCrc, |
542 |
|
(iCrc!=0x4f10ec0f)?"| ERROR": "" ); |
543 |
|
|
544 |
|
|
545 |
|
TEST_MB(interpolate8x4_halfpel_hv, 0); |
546 |
|
printf("%s - interpfield-hv-round0 %.3f usec crc32=0x%08x %s\n", |
547 |
|
cpu->name, t, iCrc, |
548 |
|
(iCrc!=0xf97ee367)?"| ERROR": "" ); |
549 |
|
|
550 |
|
TEST_MB(interpolate8x4_halfpel_hv, 1); |
551 |
|
printf("%s - round1 %.3f usec crc32=0x%08x %s\n", |
552 |
|
cpu->name, t, iCrc, |
553 |
|
(iCrc!=0xb6a9f581)?"| ERROR": "" ); |
554 |
|
/* End of 8x4 functions */ |
555 |
|
|
556 |
printf( " --- \n" ); |
printf( " --- \n" ); |
557 |
} |
} |
586 |
} \ |
} \ |
587 |
emms(); \ |
emms(); \ |
588 |
t = (gettime_usec()-t -overhead) / nb_tests;\ |
t = (gettime_usec()-t -overhead) / nb_tests;\ |
589 |
s = 0; for(i=0; i<8*32; ++i) { s += (DST)[i]^i; } |
byte_swap((uint8_t*)(DST), 8*32*sizeof((DST)[0]), sizeof((DST)[0])); \ |
590 |
|
s = calc_crc((uint8_t*)(DST), 8*32*sizeof((DST)[0]), CRC32_INITIAL) |
591 |
|
|
592 |
#define TEST_TRANSFER(FUNC, DST, SRC) \ |
#define TEST_TRANSFER(FUNC, DST, SRC) \ |
593 |
TEST_TRANSFER_BEGIN(DST); \ |
TEST_TRANSFER_BEGIN(DST); \ |
613 |
} \ |
} \ |
614 |
emms(); \ |
emms(); \ |
615 |
t = (gettime_usec()-t -overhead) / nb_tests;\ |
t = (gettime_usec()-t -overhead) / nb_tests;\ |
616 |
s = 0; for(i=0; i<8*32; ++i) { s += (DST)[i]; } |
byte_swap((uint8_t*)(DST), 8*32*sizeof((DST)[0]), sizeof((DST)[0])); \ |
617 |
|
s = calc_crc((uint8_t*)(DST), 8*32*sizeof((DST)[0]), CRC32_INITIAL) |
618 |
|
|
619 |
#define TEST_TRANSFER2(FUNC, DST, SRC, R1) \ |
#define TEST_TRANSFER2(FUNC, DST, SRC, R1) \ |
620 |
TEST_TRANSFER2_BEGIN(DST,SRC); \ |
TEST_TRANSFER2_BEGIN(DST,SRC); \ |
631 |
const int nb_tests = 4000*speed_ref; |
const int nb_tests = 4000*speed_ref; |
632 |
int i; |
int i; |
633 |
CPU *cpu; |
CPU *cpu; |
634 |
uint8_t Src8[8*32], Dst8[8*32], Ref1[8*32], Ref2[8*32]; |
// uint8_t Src8[8*32], Dst8[8*32], Ref1[8*32], Ref2[8*32]; |
635 |
int16_t Src16[8*32], Dst16[8*32]; |
// int16_t Src16[8*32], Dst16[8*32]; |
636 |
|
DECLARE_ALIGNED_MATRIX(Src8, 8, 32, uint8_t, CACHE_LINE); |
637 |
|
DECLARE_ALIGNED_MATRIX(Dst8, 8, 32, uint8_t, CACHE_LINE); |
638 |
|
DECLARE_ALIGNED_MATRIX(Ref1, 8, 32, uint8_t, CACHE_LINE); |
639 |
|
DECLARE_ALIGNED_MATRIX(Ref2, 8, 32, uint8_t, CACHE_LINE); |
640 |
|
DECLARE_ALIGNED_MATRIX(Src16, 8, 32, uint16_t, CACHE_LINE); |
641 |
|
DECLARE_ALIGNED_MATRIX(Dst16, 8, 32, uint16_t, CACHE_LINE); |
642 |
|
|
643 |
printf( "\n === test transfer ===\n" ); |
printf( "\n === test transfer ===\n" ); |
644 |
|
|
645 |
for(cpu = cpu_short_list; cpu->name!=0; ++cpu) |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
646 |
{ |
{ |
647 |
double t, overhead; |
double t, overhead; |
648 |
int tst, s; |
int tst, s; |
651 |
continue; |
continue; |
652 |
|
|
653 |
TEST_TRANSFER(transfer_8to16copy, Dst16, Src8); |
TEST_TRANSFER(transfer_8to16copy, Dst16, Src8); |
654 |
printf( "%s - 8to16 %.3f usec crc=%d\n", cpu->name, t, s ); |
printf("%s - 8to16 %.3f usec crc32=0x%08x %s\n", |
655 |
if (s!=28288) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, s, |
656 |
|
(s!=0x115814bb)?"| ERROR": ""); |
657 |
|
|
658 |
TEST_TRANSFER(transfer_16to8copy, Dst8, Src16); |
TEST_TRANSFER(transfer_16to8copy, Dst8, Src16); |
659 |
printf( "%s - 16to8 %.3f usec crc=%d\n", cpu->name, t, s ); |
printf( "%s - 16to8 %.3f usec crc32=0x%08x %s\n", |
660 |
if (s!=28288) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, s, |
661 |
|
(s!=0xee7ccbb4)?"| ERROR": ""); |
662 |
|
|
663 |
|
/* New functions for field prediction by CK 1.10.2005 */ |
664 |
|
#pragma NEW8X4 |
665 |
|
TEST_TRANSFER(transfer8x4_copy, Dst8, Src8); |
666 |
|
printf("%s - 8to4 %.3f usec crc32=0x%08x %s\n", |
667 |
|
cpu->name, t, s, |
668 |
|
(s!=0xbb9c3db5)?"| ERROR": ""); |
669 |
|
/* End of new functions */ |
670 |
|
|
671 |
TEST_TRANSFER(transfer8x8_copy, Dst8, Src8); |
TEST_TRANSFER(transfer8x8_copy, Dst8, Src8); |
672 |
printf( "%s - 8to8 %.3f usec crc=%d\n", cpu->name, t, s ); |
printf("%s - 8to8 %.3f usec crc32=0x%08x %s\n", |
673 |
if (s!=20352) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, s, |
674 |
|
(s!=0xd37b3295)?"| ERROR": ""); |
675 |
|
|
676 |
TEST_TRANSFER(transfer_16to8add, Dst8, Src16); |
TEST_TRANSFER(transfer_16to8add, Dst8, Src16); |
677 |
printf( "%s - 16to8add %.3f usec crc=%d\n", cpu->name, t, s ); |
printf("%s - 16to8add %.3f usec crc32=0x%08x %s\n", |
678 |
if (s!=25536) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, s, |
679 |
|
(s!=0xdd817bf4)?"| ERROR": "" ); |
680 |
|
|
681 |
TEST_TRANSFER2(transfer_8to16sub, Dst16, Src8, Ref1); |
TEST_TRANSFER2(transfer_8to16sub, Dst16, Src8, Ref1); |
682 |
printf( "%s - 8to16sub %.3f usec crc1=%d ", cpu->name, t, s ); |
{ |
683 |
if (s!=28064) printf( "*** CRC ERROR! ***\n" ); |
int s1, s2; |
684 |
s = 0; for(i=0; i<8*32; ++i) { s += (Src8[i]-Ref1[i])&i; } |
s1 = calc_crc((uint8_t*)Dst16, 8*32*sizeof(Dst16[0]), CRC32_INITIAL); |
685 |
printf( "crc2=%d\n", s); |
s2 = calc_crc((uint8_t*)Src8, 8*32*sizeof(Src8[0]), CRC32_INITIAL); |
686 |
if (s!=16256) printf( "*** CRC ERROR! ***\n" ); |
printf("%s - 8to16sub %.3f usec crc32(1)=0x%08x crc32(2)=0x%08x %s %s\n", |
687 |
|
cpu->name, t, s1, s2, |
688 |
|
(s1!=0xa1e07163)?"| ERROR1": "", |
689 |
|
(s2!=0xd86c5d23)?"| ERROR2": "" ); |
690 |
|
} |
691 |
|
|
692 |
TEST_TRANSFER3(transfer_8to16sub2, Dst16, Src8, Ref1, Ref2); |
TEST_TRANSFER3(transfer_8to16sub2, Dst16, Src8, Ref1, Ref2); |
693 |
printf( "%s - 8to16sub2 %.3f usec crc=%d\n", cpu->name, t, s ); |
printf("%s - 8to16sub2 %.3f usec crc32=0x%08x %s\n", |
694 |
if (s!=20384) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, s, |
695 |
|
(s!=0x99b6c4c7)?"| ERROR": "" ); |
696 |
|
|
697 |
printf( " --- \n" ); |
printf( " --- \n" ); |
698 |
} |
} |
704 |
|
|
705 |
#define TEST_QUANT(FUNC, DST, SRC) \ |
#define TEST_QUANT(FUNC, DST, SRC) \ |
706 |
t = gettime_usec(); \ |
t = gettime_usec(); \ |
707 |
|
for(s=CRC32_INITIAL,qm=1; qm<=255; ++qm) { \ |
708 |
|
for(i=0; i<8*8; ++i) Quant[i] = qm; \ |
709 |
|
set_inter_matrix( mpeg_quant_matrices, Quant ); \ |
710 |
emms(); \ |
emms(); \ |
711 |
|
for(q=1; q<=max_Q; ++q) { \ |
712 |
for(tst=0; tst<nb_tests; ++tst) \ |
for(tst=0; tst<nb_tests; ++tst) \ |
713 |
for(s=0, q=1; q<=max_Q; ++q) { \ |
(FUNC)((DST), (SRC), q, mpeg_quant_matrices); \ |
714 |
(FUNC)((DST), (SRC), q); \ |
byte_swap((uint8_t*)(DST), 64*sizeof((DST)[0]), sizeof((DST)[0])); \ |
715 |
for(i=0; i<64; ++i) s+=(DST)[i]^i; \ |
s = calc_crc((uint8_t*)(DST), 64*sizeof((DST)[0]), s); \ |
716 |
} \ |
} \ |
717 |
emms(); \ |
emms(); \ |
718 |
t = (gettime_usec()-t-overhead)/nb_tests; |
} \ |
719 |
|
t = (gettime_usec()-t-overhead)/nb_tests/qm |
720 |
|
|
721 |
#define TEST_QUANT2(FUNC, DST, SRC, MULT) \ |
#define TEST_QUANT2(FUNC, DST, SRC) \ |
722 |
t = gettime_usec(); \ |
t = gettime_usec(); \ |
723 |
|
for(s=CRC32_INITIAL,qm=1; qm<=255; ++qm) { \ |
724 |
|
for(i=0; i<8*8; ++i) Quant[i] = qm; \ |
725 |
|
set_intra_matrix( mpeg_quant_matrices, Quant ); \ |
726 |
emms(); \ |
emms(); \ |
727 |
|
for(q=1; q<=max_Q; ++q) { \ |
728 |
for(tst=0; tst<nb_tests; ++tst) \ |
for(tst=0; tst<nb_tests; ++tst) \ |
729 |
for(s=0, q=1; q<=max_Q; ++q) { \ |
(FUNC)((DST), (SRC), q, q, mpeg_quant_matrices); \ |
730 |
(FUNC)((DST), (SRC), q, MULT); \ |
byte_swap((uint8_t*)(DST), 64*sizeof((DST)[0]), sizeof((DST)[0])); \ |
731 |
for(i=0; i<64; ++i) s+=(DST)[i]^i; \ |
s = calc_crc((uint8_t*)(DST), 64*sizeof((DST)[0]), s); \ |
732 |
} \ |
} \ |
733 |
emms(); \ |
emms(); \ |
734 |
t = (gettime_usec()-t-overhead)/nb_tests; |
} \ |
735 |
|
t = (gettime_usec()-t-overhead)/nb_tests/qm |
736 |
|
|
737 |
|
#define TEST_INTRA(REFFUNC, NEWFUNC, RANGE) \ |
738 |
|
{ int i,q,s;\ |
739 |
|
DECLARE_ALIGNED_MATRIX(Src, 8, 8, int16_t, 16); \ |
740 |
|
DECLARE_ALIGNED_MATRIX(Dst, 8, 8, int16_t, 16); \ |
741 |
|
DECLARE_ALIGNED_MATRIX(Dst2,8, 8, int16_t, 16); \ |
742 |
|
for(q=1;q<=max_Q;q++) \ |
743 |
|
for(s=-RANGE;s<RANGE;s++) { \ |
744 |
|
for(i=0;i<64;i++) Src[i]=s; \ |
745 |
|
(REFFUNC)((Dst),(Src),q,q,mpeg_quant_matrices); \ |
746 |
|
(NEWFUNC)((Dst2),(Src),q,q,mpeg_quant_matrices); \ |
747 |
|
for(i=0;i<64;i++) \ |
748 |
|
if(Dst[i]!=Dst2[i]) printf("ERROR : " #NEWFUNC " i%d quant:%d input:%d C_result:%d ASM_result:%d\n",i,q,s,Dst[i],Dst2[i]); \ |
749 |
|
} \ |
750 |
|
} |
751 |
|
|
752 |
|
#define TEST_INTER(REFFUNC, NEWFUNC, RANGE) \ |
753 |
|
{ int i,q,s; \ |
754 |
|
DECLARE_ALIGNED_MATRIX(Src, 8, 8, int16_t, 16); \ |
755 |
|
DECLARE_ALIGNED_MATRIX(Dst, 8, 8, int16_t, 16); \ |
756 |
|
DECLARE_ALIGNED_MATRIX(Dst2,8, 8, int16_t, 16); \ |
757 |
|
for(q=1;q<=max_Q;q++) \ |
758 |
|
for(s=-RANGE;s<RANGE;s++) { \ |
759 |
|
for(i=0;i<64;i++) Src[i]=s; \ |
760 |
|
(REFFUNC)((Dst),(Src),q,mpeg_quant_matrices); \ |
761 |
|
(NEWFUNC)((Dst2),(Src),q,mpeg_quant_matrices); \ |
762 |
|
emms(); \ |
763 |
|
for(i=0;i<64;i++) \ |
764 |
|
if(Dst[i]!=Dst2[i]) printf("ERROR : " #NEWFUNC " i%d quant:%d input:%d C_result:%d ASM_result:%d\n",i,q,s,Dst[i],Dst2[i]); \ |
765 |
|
} \ |
766 |
|
} |
767 |
|
|
768 |
void test_quant() |
void test_quant() |
769 |
{ |
{ |
770 |
const int nb_tests = 150*speed_ref; |
const int nb_tests = 1*speed_ref; |
771 |
const int max_Q = 31; |
const int max_Q = 31; |
772 |
int i; |
DECLARE_ALIGNED_MATRIX(mpeg_quant_matrices, 8, 64, uint16_t, 16); |
773 |
|
|
774 |
|
int i, qm; |
775 |
CPU *cpu; |
CPU *cpu; |
776 |
int16_t Src[8*8], Dst[8*8]; |
DECLARE_ALIGNED_MATRIX(Src, 8, 8, int16_t, 16); |
777 |
|
DECLARE_ALIGNED_MATRIX(Dst, 8, 8, int16_t, 16); |
778 |
|
DECLARE_ALIGNED_MATRIX(Dst2,8, 8, int16_t, 16); |
779 |
|
uint8_t Quant[8*8]; |
780 |
|
|
781 |
printf( "\n ===== test quant =====\n" ); |
printf( "\n ===== test quant =====\n" ); |
782 |
|
|
783 |
|
/* we deliberately enfringe the norm's specified range [-127,127], */ |
784 |
|
/* to test the robustness of the iquant module */ |
785 |
for(i=0; i<64; ++i) { |
for(i=0; i<64; ++i) { |
786 |
Src[i] = i-32; |
Src[i] = 1 + (i-32) * (i&6); |
787 |
Dst[i] = 0; |
Dst[i] = 0; |
788 |
} |
} |
789 |
|
|
790 |
|
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
|
for(cpu = cpu_short_list; cpu->name!=0; ++cpu) |
|
791 |
{ |
{ |
792 |
double t, overhead; |
double t, overhead; |
793 |
int tst, s, q; |
int tst, q; |
794 |
|
uint32_t s; |
795 |
|
|
796 |
if (!init_cpu(cpu)) |
if (!init_cpu(cpu)) |
797 |
continue; |
continue; |
798 |
|
|
799 |
set_inter_matrix( get_default_inter_matrix() ); |
// exhaustive tests to compare against the (ref) C-version |
800 |
set_intra_matrix( get_default_intra_matrix() ); |
TEST_INTRA(quant_h263_intra_c, quant_h263_intra, 2048); |
801 |
|
TEST_INTRA(dequant_h263_intra_c, dequant_h263_intra , 512 ); |
802 |
|
TEST_INTER(quant_h263_inter_c, quant_h263_inter , 2048); |
803 |
|
TEST_INTER(dequant_h263_inter_c, dequant_h263_inter , 512 ); |
804 |
|
|
805 |
overhead = -gettime_usec(); |
overhead = -gettime_usec(); |
806 |
for(tst=0; tst<nb_tests; ++tst) |
for(s=0,qm=1; qm<=255; ++qm) { |
807 |
for(s=0, q=1; q<=max_Q; ++q) |
for(i=0; i<8*8; ++i) Quant[i] = qm; |
808 |
for(i=0; i<64; ++i) s+=Dst[i]^i; |
set_inter_matrix(mpeg_quant_matrices, Quant ); |
809 |
|
for(q=1; q<=max_Q; ++q) |
810 |
|
for(i=0; i<64; ++i) s+=Dst[i]^i^qm; |
811 |
|
} |
812 |
overhead += gettime_usec(); |
overhead += gettime_usec(); |
813 |
|
|
814 |
TEST_QUANT2(quant4_intra, Dst, Src, 7); |
TEST_QUANT2(quant_mpeg_intra, Dst, Src); |
815 |
printf( "%s - quant4_intra %.3f usec crc=%d\n", cpu->name, t, s ); |
printf("%s - quant_mpeg_intra %.3f usec crc32=0x%08x %s\n", |
816 |
if (s!=55827) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, s, |
817 |
|
(s!=0xfd6a21a4)? "| ERROR": ""); |
818 |
TEST_QUANT(quant4_inter, Dst, Src); |
|
819 |
printf( "%s - quant4_inter %.3f usec crc=%d\n", cpu->name, t, s ); |
TEST_QUANT(quant_mpeg_inter, Dst, Src); |
820 |
if (s!=58201) printf( "*** CRC ERROR! ***\n" ); |
printf("%s - quant_mpeg_inter %.3f usec crc32=0x%08x %s\n", |
821 |
|
cpu->name, t, s, |
822 |
|
(s!=0xf6de7757)?"| ERROR": ""); |
823 |
TEST_QUANT2(dequant4_intra, Dst, Src, 7); |
|
824 |
printf( "%s - dequant4_intra %.3f usec crc=%d\n", cpu->name, t, s ); |
TEST_QUANT2(dequant_mpeg_intra, Dst, Src); |
825 |
if (s!=193340) printf( "*** CRC ERROR! ***\n" ); |
printf("%s - dequant_mpeg_intra %.3f usec crc32=0x%08x %s\n", |
826 |
|
cpu->name, t, s, |
827 |
TEST_QUANT(dequant4_inter, Dst, Src); |
(s!=0x2def7bc7)?"| ERROR": ""); |
828 |
printf( "%s - dequant4_inter %.3f usec crc=%d\n", cpu->name, t, s ); |
|
829 |
if (s!=116483) printf( "*** CRC ERROR! ***\n" ); |
TEST_QUANT(dequant_mpeg_inter, Dst, Src); |
830 |
|
printf("%s - dequant_mpeg_inter %.3f usec crc32=0x%08x %s\n", |
831 |
TEST_QUANT2(quant_intra, Dst, Src, 7); |
cpu->name, t, s, |
832 |
printf( "%s - quant_intra %.3f usec crc=%d\n", cpu->name, t, s ); |
(s!=0xd878c722)?"| ERROR": ""); |
833 |
if (s!=56885) printf( "*** CRC ERROR! ***\n" ); |
|
834 |
|
TEST_QUANT2(quant_h263_intra, Dst, Src); |
835 |
TEST_QUANT(quant_inter, Dst, Src); |
printf("%s - quant_h263_intra %.3f usec crc32=0x%08x %s\n", |
836 |
printf( "%s - quant_inter %.3f usec crc=%d\n", cpu->name, t, s ); |
cpu->name, t, s, |
837 |
if (s!=58056) printf( "*** CRC ERROR! ***\n" ); |
(s!=0x2eba9d43)?"| ERROR": ""); |
838 |
|
|
839 |
TEST_QUANT2(dequant_intra, Dst, Src, 7); |
TEST_QUANT(quant_h263_inter, Dst, Src); |
840 |
printf( "%s - dequant_intra %.3f usec crc=%d\n", cpu->name, t, s ); |
printf("%s - quant_h263_inter %.3f usec crc32=0x%08x %s\n", |
841 |
if (s!=-7936) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, s, |
842 |
|
(s!=0xbd315a7e)?"| ERROR": ""); |
843 |
TEST_QUANT(dequant_inter, Dst, Src); |
|
844 |
printf( "%s - dequant_inter %.3f usec crc=%d\n", cpu->name, t, s ); |
TEST_QUANT2(dequant_h263_intra, Dst, Src); |
845 |
// { int k,l; for(k=0; k<8; ++k) { for(l=0; l<8; ++l) printf( "[%.4d]", Dst[k*8+l]); printf("\n"); } } |
printf("%s - dequant_h263_intra %.3f usec crc32=0x%08x %s\n", |
846 |
if (s!=-33217) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, s, |
847 |
|
(s!=0x9841212a)?"| ERROR": ""); |
848 |
|
|
849 |
|
TEST_QUANT(dequant_h263_inter, Dst, Src); |
850 |
|
printf("%s - dequant_h263_inter %.3f usec crc32=0x%08x %s\n", |
851 |
|
cpu->name, t, s, |
852 |
|
(s!=0xe7df8fba)?"| ERROR": ""); |
853 |
|
|
854 |
|
printf( " --- \n" ); |
855 |
|
} |
856 |
|
} |
857 |
|
|
858 |
|
/********************************************************************* |
859 |
|
* test distortion operators |
860 |
|
*********************************************************************/ |
861 |
|
|
862 |
|
static void ieee_reseed(long s); |
863 |
|
static long ieee_rand(int Min, int Max); |
864 |
|
|
865 |
|
#define TEST_SSE(FUNCTION, SRC1, SRC2, STRIDE) \ |
866 |
|
do { \ |
867 |
|
t = gettime_usec(); \ |
868 |
|
tst = nb_tests; \ |
869 |
|
while((tst--)>0) sse = (FUNCTION)((SRC1), (SRC2), (STRIDE)); \ |
870 |
|
emms(); \ |
871 |
|
t = (gettime_usec() - t)/(double)nb_tests; \ |
872 |
|
} while(0) |
873 |
|
|
874 |
|
|
875 |
|
void test_sse() |
876 |
|
{ |
877 |
|
const int nb_tests = 100000*speed_ref; |
878 |
|
int i; |
879 |
|
CPU *cpu; |
880 |
|
DECLARE_ALIGNED_MATRIX(Src1, 8, 8, int16_t, 16); |
881 |
|
DECLARE_ALIGNED_MATRIX(Src2, 8, 8, int16_t, 16); |
882 |
|
DECLARE_ALIGNED_MATRIX(Src3, 8, 8, int16_t, 16); |
883 |
|
DECLARE_ALIGNED_MATRIX(Src4, 8, 8, int16_t, 16); |
884 |
|
|
885 |
|
printf( "\n ===== test sse =====\n" ); |
886 |
|
|
887 |
|
ieee_reseed(1); |
888 |
|
for(i=0; i<64; ++i) { |
889 |
|
Src1[i] = ieee_rand(-2048, 2047); |
890 |
|
Src2[i] = ieee_rand(-2048, 2047); |
891 |
|
Src3[i] = ieee_rand(-2048, 2047); |
892 |
|
Src4[i] = ieee_rand(-2048, 2047); |
893 |
|
} |
894 |
|
|
895 |
|
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
896 |
|
{ |
897 |
|
double t; |
898 |
|
int tst, sse; |
899 |
|
|
900 |
|
if (!init_cpu(cpu)) |
901 |
|
continue; |
902 |
|
|
903 |
|
/* 16 bit element blocks */ |
904 |
|
TEST_SSE(sse8_16bit, Src1, Src2, 16); |
905 |
|
printf("%s - sse8_16bit#1 %.3f usec sse=%d %s\n", |
906 |
|
cpu->name, t, sse, (sse!=182013834)?"| ERROR": ""); |
907 |
|
TEST_SSE(sse8_16bit, Src1, Src3, 16); |
908 |
|
printf("%s - sse8_16bit#2 %.3f usec sse=%d %s\n", |
909 |
|
cpu->name, t, sse, (sse!=142545203)?"| ERROR": ""); |
910 |
|
TEST_SSE(sse8_16bit, Src1, Src4, 16); |
911 |
|
printf("%s - sse8_16bit#3 %.3f usec sse=%d %s\n", |
912 |
|
cpu->name, t, sse, (sse!=146340935)?"| ERROR": ""); |
913 |
|
TEST_SSE(sse8_16bit, Src2, Src3, 16); |
914 |
|
printf("%s - sse8_16bit#4 %.3f usec sse=%d %s\n", |
915 |
|
cpu->name, t, sse, (sse!=130136661)?"| ERROR": ""); |
916 |
|
TEST_SSE(sse8_16bit, Src2, Src4, 16); |
917 |
|
printf("%s - sse8_16bit#5 %.3f usec sse=%d %s\n", |
918 |
|
cpu->name, t, sse, (sse!=136870353)?"| ERROR": ""); |
919 |
|
TEST_SSE(sse8_16bit, Src3, Src4, 16); |
920 |
|
printf("%s - sse8_16bit#6 %.3f usec sse=%d %s\n", |
921 |
|
cpu->name, t, sse, (sse!=164107772)?"| ERROR": ""); |
922 |
|
|
923 |
|
/* 8 bit element blocks */ |
924 |
|
TEST_SSE(sse8_8bit, (int8_t*)Src1, (int8_t*)Src2, 8); |
925 |
|
printf("%s - sse8_8bit#1 %.3f usec sse=%d %s\n", |
926 |
|
cpu->name, t, sse, (sse!=1356423)?"| ERROR": ""); |
927 |
|
TEST_SSE(sse8_8bit, (int8_t*)Src1, (int8_t*)Src3, 8); |
928 |
|
printf("%s - sse8_8bit#2 %.3f usec sse=%d %s\n", |
929 |
|
cpu->name, t, sse, (sse!=1173074)?"| ERROR": ""); |
930 |
|
TEST_SSE(sse8_8bit, (int8_t*)Src1, (int8_t*)Src4, 8); |
931 |
|
printf("%s - sse8_8bit#3 %.3f usec sse=%d %s\n", |
932 |
|
cpu->name, t, sse, (sse!=1092357)?"| ERROR": ""); |
933 |
|
TEST_SSE(sse8_8bit, (int8_t*)Src2, (int8_t*)Src3, 8); |
934 |
|
printf("%s - sse8_8bit#4 %.3f usec sse=%d %s\n", |
935 |
|
cpu->name, t, sse, (sse!=1360239)?"| ERROR": ""); |
936 |
|
TEST_SSE(sse8_8bit, (int8_t*)Src2, (int8_t*)Src4, 8); |
937 |
|
printf("%s - sse8_8bit#5 %.3f usec sse=%d %s\n", |
938 |
|
cpu->name, t, sse, (sse!=1208414)?"| ERROR": ""); |
939 |
|
TEST_SSE(sse8_8bit, (int8_t*)Src3, (int8_t*)Src4, 8); |
940 |
|
printf("%s - sse8_8bit#6 %.3f usec sse=%d %s\n", |
941 |
|
cpu->name, t, sse, (sse!=1099285)?"| ERROR": ""); |
942 |
|
|
943 |
printf( " --- \n" ); |
printf( " --- \n" ); |
944 |
} |
} |
948 |
* test non-zero AC counting |
* test non-zero AC counting |
949 |
*********************************************************************/ |
*********************************************************************/ |
950 |
|
|
951 |
#define TEST_CBP(FUNC, SRC) \ |
#define TEST_CBP(FUNC, SRC, NB) \ |
952 |
t = gettime_usec(); \ |
t = gettime_usec(); \ |
953 |
emms(); \ |
emms(); \ |
954 |
for(tst=0; tst<nb_tests; ++tst) { \ |
for(tst=0; tst<NB; ++tst) { \ |
955 |
cbp = (FUNC)((SRC)); \ |
cbp = (FUNC)((SRC)); \ |
956 |
} \ |
} \ |
957 |
emms(); \ |
emms(); \ |
960 |
void test_cbp() |
void test_cbp() |
961 |
{ |
{ |
962 |
const int nb_tests = 10000*speed_ref; |
const int nb_tests = 10000*speed_ref; |
963 |
int i; |
int i, n, m; |
964 |
CPU *cpu; |
CPU *cpu; |
965 |
int16_t Src1[6*64], Src2[6*64], Src3[6*64], Src4[6*64]; |
DECLARE_ALIGNED_MATRIX(Src1, 6, 64, int16_t, 16); |
966 |
|
DECLARE_ALIGNED_MATRIX(Src2, 6, 64, int16_t, 16); |
967 |
|
DECLARE_ALIGNED_MATRIX(Src3, 6, 64, int16_t, 16); |
968 |
|
DECLARE_ALIGNED_MATRIX(Src4, 6, 64, int16_t, 16); |
969 |
|
DECLARE_ALIGNED_MATRIX(Src5, 6, 64, int16_t, 16); |
970 |
|
|
971 |
printf( "\n ===== test cbp =====\n" ); |
printf( "\n ===== test cbp =====\n" ); |
972 |
|
|
973 |
for(i=0; i<6*64; ++i) { |
for(i=0; i<6*64; ++i) { |
974 |
Src1[i] = (i*i*3/8192)&(i/64)&1; // 'random' |
Src1[i] = (i*i*3/8192)&(i/64)&1; /* 'random' */ |
975 |
Src2[i] = (i<3*64); // half-full |
Src2[i] = (i<3*64); /* half-full */ |
976 |
Src3[i] = ((i+32)>3*64); |
Src3[i] = ((i+32)>3*64); |
977 |
Src4[i] = (i==(3*64+2) || i==(5*64+9)); |
Src4[i] = (i==(3*64+2) || i==(5*64+9)); |
978 |
|
Src5[i] = ieee_rand(0,1) ? -1 : 1; /* +/- test */ |
979 |
} |
} |
980 |
|
|
981 |
for(cpu = cpu_short_list2; cpu->name!=0; ++cpu) |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
982 |
{ |
{ |
983 |
double t; |
double t; |
984 |
int tst, cbp; |
int tst, cbp; |
986 |
if (!init_cpu(cpu)) |
if (!init_cpu(cpu)) |
987 |
continue; |
continue; |
988 |
|
|
989 |
TEST_CBP(calc_cbp, Src1); |
TEST_CBP(calc_cbp, Src1, nb_tests); |
990 |
printf( "%s - calc_cbp#1 %.3f usec cbp=0x%x\n", cpu->name, t, cbp ); |
printf("%s - calc_cbp#1 %.3f usec cbp=0x%02x %s\n", |
991 |
if (cbp!=0x15) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, cbp, (cbp!=0x15)?"| ERROR": ""); |
992 |
TEST_CBP(calc_cbp, Src2); |
TEST_CBP(calc_cbp, Src2, nb_tests); |
993 |
printf( "%s - calc_cbp#2 %.3f usec cbp=0x%x\n", cpu->name, t, cbp ); |
printf("%s - calc_cbp#2 %.3f usec cbp=0x%02x %s\n", |
994 |
if (cbp!=0x38) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, cbp, (cbp!=0x38)?"| ERROR": ""); |
995 |
TEST_CBP(calc_cbp, Src3); |
TEST_CBP(calc_cbp, Src3, nb_tests); |
996 |
printf( "%s - calc_cbp#3 %.3f usec cbp=0x%x\n", cpu->name, t, cbp ); |
printf("%s - calc_cbp#3 %.3f usec cbp=0x%02x %s\n", |
997 |
if (cbp!=0x0f) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, cbp, (cbp!=0x0f)?"| ERROR": "" ); |
998 |
TEST_CBP(calc_cbp, Src4); |
TEST_CBP(calc_cbp, Src4, nb_tests); |
999 |
printf( "%s - calc_cbp#4 %.3f usec cbp=0x%x\n", cpu->name, t, cbp ); |
printf("%s - calc_cbp#4 %.3f usec cbp=0x%02x %s\n", |
1000 |
if (cbp!=0x05) printf( "*** CRC ERROR! ***\n" ); |
cpu->name, t, cbp, (cbp!=0x05)?"| ERROR": "" ); |
1001 |
|
TEST_CBP(calc_cbp, Src5, nb_tests); |
1002 |
|
printf("%s - calc_cbp#4 %.3f usec cbp=0x%02x %s\n", |
1003 |
|
cpu->name, t, cbp, (cbp!=0x3f)?"| ERROR": "" ); |
1004 |
printf( " --- \n" ); |
printf( " --- \n" ); |
1005 |
} |
} |
1006 |
|
|
1007 |
|
for(cpu = cpu_list; cpu->name!=0; ++cpu) /* bench suggested by Carlo (carlo dot bramix at libero dot it) */ |
1008 |
|
{ |
1009 |
|
double t; |
1010 |
|
int tst, cbp, err; |
1011 |
|
|
1012 |
|
if (!init_cpu(cpu)) |
1013 |
|
continue; |
1014 |
|
|
1015 |
|
err = 0; |
1016 |
|
for(n=0; n<6; ++n) |
1017 |
|
{ |
1018 |
|
for(m=0; m<64; ++m) |
1019 |
|
{ |
1020 |
|
for(i=0; i<6*64; ++i) |
1021 |
|
Src1[i] = (i== (m + n*64)); |
1022 |
|
|
1023 |
|
TEST_CBP(calc_cbp, Src1, 1); |
1024 |
|
if (cbp!= (((m!=0)<<(5-n)))) |
1025 |
|
{ |
1026 |
|
printf( "%s - calc_cbp#5: ERROR at pos %d / %d!\n", cpu->name, n, m); |
1027 |
|
err = 1; |
1028 |
|
break; |
1029 |
|
} |
1030 |
|
} |
1031 |
|
} |
1032 |
|
if (!err) |
1033 |
|
printf( " %s - calc_cbp#5 : OK\n", cpu->name ); |
1034 |
|
|
1035 |
|
} |
1036 |
|
} |
1037 |
|
|
1038 |
|
/********************************************************************* |
1039 |
|
* fdct/idct IEEE1180 compliance |
1040 |
|
*********************************************************************/ |
1041 |
|
|
1042 |
|
typedef struct { |
1043 |
|
long Errors[64]; |
1044 |
|
long Sqr_Errors[64]; |
1045 |
|
long Max_Errors[64]; |
1046 |
|
long Nb; |
1047 |
|
} STATS_8x8; |
1048 |
|
|
1049 |
|
void init_stats(STATS_8x8 *S) |
1050 |
|
{ |
1051 |
|
int i; |
1052 |
|
for(i=0; i<64; ++i) { |
1053 |
|
S->Errors[i] = 0; |
1054 |
|
S->Sqr_Errors[i] = 0; |
1055 |
|
S->Max_Errors[i] = 0; |
1056 |
|
} |
1057 |
|
S->Nb = 0; |
1058 |
|
} |
1059 |
|
|
1060 |
|
void store_stats(STATS_8x8 *S, short Blk[64], short Ref[64]) |
1061 |
|
{ |
1062 |
|
int i; |
1063 |
|
for(i=0; i<64; ++i) |
1064 |
|
{ |
1065 |
|
short Err = Blk[i] - Ref[i]; |
1066 |
|
S->Errors[i] += Err; |
1067 |
|
S->Sqr_Errors[i] += Err * Err; |
1068 |
|
if (Err<0) Err = -Err; |
1069 |
|
if (S->Max_Errors[i]<Err) |
1070 |
|
S->Max_Errors[i] = Err; |
1071 |
|
} |
1072 |
|
S->Nb++; |
1073 |
|
} |
1074 |
|
|
1075 |
|
void print_stats(STATS_8x8 *S) |
1076 |
|
{ |
1077 |
|
int i; |
1078 |
|
double Norm; |
1079 |
|
|
1080 |
|
assert(S->Nb>0); |
1081 |
|
Norm = 1. / (double)S->Nb; |
1082 |
|
printf("\n== Max absolute values of errors ==\n"); |
1083 |
|
for(i=0; i<64; i++) { |
1084 |
|
printf(" %4ld", S->Max_Errors[i]); |
1085 |
|
if ((i&7)==7) printf("\n"); |
1086 |
|
} |
1087 |
|
|
1088 |
|
printf("\n== Mean square errors ==\n"); |
1089 |
|
for(i=0; i<64; i++) |
1090 |
|
{ |
1091 |
|
double Err = Norm * (double)S->Sqr_Errors[i]; |
1092 |
|
printf(" %.3f", Err); |
1093 |
|
if ((i&7)==7) printf("\n"); |
1094 |
|
} |
1095 |
|
|
1096 |
|
printf("\n== Mean errors ==\n"); |
1097 |
|
for(i=0; i<64; i++) |
1098 |
|
{ |
1099 |
|
double Err = Norm * (double)S->Errors[i]; |
1100 |
|
printf(" %.3f", Err); |
1101 |
|
if ((i&7)==7) printf("\n"); |
1102 |
|
} |
1103 |
|
printf("\n"); |
1104 |
|
} |
1105 |
|
|
1106 |
|
static const char *CHECK(double v, double l) { |
1107 |
|
if (fabs(v)<=l) return "ok"; |
1108 |
|
else return "FAIL!"; |
1109 |
|
} |
1110 |
|
|
1111 |
|
void report_stats(STATS_8x8 *S, const double *Limits) |
1112 |
|
{ |
1113 |
|
int i; |
1114 |
|
double Norm, PE, PMSE, OMSE, PME, OME; |
1115 |
|
|
1116 |
|
assert(S->Nb>0); |
1117 |
|
Norm = 1. / (double)S->Nb; |
1118 |
|
PE = 0.; |
1119 |
|
for(i=0; i<64; i++) { |
1120 |
|
if (PE<S->Max_Errors[i]) |
1121 |
|
PE = S->Max_Errors[i]; |
1122 |
|
} |
1123 |
|
|
1124 |
|
PMSE = 0.; |
1125 |
|
OMSE = 0.; |
1126 |
|
for(i=0; i<64; i++) |
1127 |
|
{ |
1128 |
|
double Err = Norm * (double)S->Sqr_Errors[i]; |
1129 |
|
OMSE += Err; |
1130 |
|
if (PMSE < Err) PMSE = Err; |
1131 |
|
} |
1132 |
|
OMSE /= 64.; |
1133 |
|
|
1134 |
|
PME = 0.; |
1135 |
|
OME = 0.; |
1136 |
|
for(i=0; i<64; i++) |
1137 |
|
{ |
1138 |
|
double Err = Norm * (double)S->Errors[i]; |
1139 |
|
OME += Err; |
1140 |
|
Err = fabs(Err); |
1141 |
|
if (PME < Err) PME = Err; |
1142 |
|
} |
1143 |
|
OME /= 64.; |
1144 |
|
|
1145 |
|
printf( "Peak error: %4.4f\n", PE ); |
1146 |
|
printf( "Peak MSE: %4.4f\n", PMSE ); |
1147 |
|
printf( "Overall MSE: %4.4f\n", OMSE ); |
1148 |
|
printf( "Peak ME: %4.4f\n", PME ); |
1149 |
|
printf( "Overall ME: %4.4f\n", OME ); |
1150 |
|
|
1151 |
|
if (Limits!=0) |
1152 |
|
{ |
1153 |
|
printf( "[PE<=%.4f %s] ", Limits[0], CHECK(PE, Limits[0]) ); |
1154 |
|
printf( "\n" ); |
1155 |
|
printf( "[PMSE<=%.4f %s]", Limits[1], CHECK(PMSE, Limits[1]) ); |
1156 |
|
printf( "[OMSE<=%.4f %s]", Limits[2], CHECK(OMSE, Limits[2]) ); |
1157 |
|
printf( "\n" ); |
1158 |
|
printf( "[PME<=%.4f %s] ", Limits[3], CHECK(PME , Limits[3]) ); |
1159 |
|
printf( "[OME<=%.4f %s] ", Limits[4], CHECK(OME , Limits[4]) ); |
1160 |
|
printf( "\n" ); |
1161 |
|
} |
1162 |
|
} |
1163 |
|
|
1164 |
|
///* ////////////////////////////////////////////////////// */ |
1165 |
|
/* Pseudo-random generator specified by IEEE 1180 */ |
1166 |
|
|
1167 |
|
static long ieee_seed = 1; |
1168 |
|
static void ieee_reseed(long s) { |
1169 |
|
ieee_seed = s; |
1170 |
|
} |
1171 |
|
static long ieee_rand(int Min, int Max) |
1172 |
|
{ |
1173 |
|
static double z = (double) 0x7fffffff; |
1174 |
|
|
1175 |
|
long i,j; |
1176 |
|
double x; |
1177 |
|
|
1178 |
|
ieee_seed = (ieee_seed * 1103515245) + 12345; |
1179 |
|
i = ieee_seed & 0x7ffffffe; |
1180 |
|
x = ((double) i) / z; |
1181 |
|
x *= (Max-Min+1); |
1182 |
|
j = (long)x; |
1183 |
|
j = j + Min; |
1184 |
|
assert(j>=Min && j<=Max); |
1185 |
|
return (short)j; |
1186 |
|
} |
1187 |
|
|
1188 |
|
#define CLAMP(x, M) (x) = ((x)<-(M)) ? (-(M)) : ((x)>=(M) ? ((M)-1) : (x)) |
1189 |
|
|
1190 |
|
static double Cos[8][8]; |
1191 |
|
static void init_ref_dct() |
1192 |
|
{ |
1193 |
|
int i, j; |
1194 |
|
for(i=0; i<8; i++) |
1195 |
|
{ |
1196 |
|
double scale = (i == 0) ? sqrt(0.125) : 0.5; |
1197 |
|
for (j=0; j<8; j++) |
1198 |
|
Cos[i][j] = scale*cos( (M_PI/8.0)*i*(j + 0.5) ); |
1199 |
|
} |
1200 |
|
} |
1201 |
|
|
1202 |
|
void ref_idct(short *M) |
1203 |
|
{ |
1204 |
|
int i, j, k; |
1205 |
|
double Tmp[8][8]; |
1206 |
|
|
1207 |
|
for(i=0; i<8; i++) { |
1208 |
|
for(j=0; j<8; j++) |
1209 |
|
{ |
1210 |
|
double Sum = 0.0; |
1211 |
|
for (k=0; k<8; k++) Sum += Cos[k][j]*M[8*i+k]; |
1212 |
|
Tmp[i][j] = Sum; |
1213 |
|
} |
1214 |
|
} |
1215 |
|
for(i=0; i<8; i++) { |
1216 |
|
for(j=0; j<8; j++) { |
1217 |
|
double Sum = 0.0; |
1218 |
|
for (k=0; k<8; k++) Sum += Cos[k][i]*Tmp[k][j]; |
1219 |
|
M[8*i+j] = (short)floor(Sum + .5); |
1220 |
|
} |
1221 |
|
} |
1222 |
|
} |
1223 |
|
|
1224 |
|
void ref_fdct(short *M) |
1225 |
|
{ |
1226 |
|
int i, j, k; |
1227 |
|
double Tmp[8][8]; |
1228 |
|
|
1229 |
|
for(i=0; i<8; i++) { |
1230 |
|
for(j=0; j<8; j++) |
1231 |
|
{ |
1232 |
|
double Sum = 0.0; |
1233 |
|
for (k=0; k<8; k++) Sum += Cos[j][k]*M[8*i+k]; |
1234 |
|
Tmp[i][j] = Sum; |
1235 |
|
} |
1236 |
|
} |
1237 |
|
for(i=0; i<8; i++) { |
1238 |
|
for(j=0; j<8; j++) { |
1239 |
|
double Sum = 0.0; |
1240 |
|
for (k=0; k<8; k++) Sum += Cos[i][k]*Tmp[k][j]; |
1241 |
|
M[8*i+j] = (short)floor(Sum + 0.5); |
1242 |
|
} |
1243 |
|
} |
1244 |
|
} |
1245 |
|
|
1246 |
|
void test_IEEE1180_compliance(int Min, int Max, int Sign) |
1247 |
|
{ |
1248 |
|
static const double ILimits[5] = { 1., 0.06, 0.02, 0.015, 0.0015 }; |
1249 |
|
int Loops = 10000; |
1250 |
|
int i, m, n; |
1251 |
|
DECLARE_ALIGNED_MATRIX(Blk0, 8, 8, short, 16); /* reference */ |
1252 |
|
DECLARE_ALIGNED_MATRIX(Blk, 8, 8, short, 16); |
1253 |
|
DECLARE_ALIGNED_MATRIX(iBlk, 8, 8, short, 16); |
1254 |
|
DECLARE_ALIGNED_MATRIX(Ref_FDCT, 8, 8, short, 16); |
1255 |
|
DECLARE_ALIGNED_MATRIX(Ref_IDCT, 8, 8, short, 16); |
1256 |
|
|
1257 |
|
STATS_8x8 FStats; /* forward dct stats */ |
1258 |
|
STATS_8x8 IStats; /* inverse dct stats */ |
1259 |
|
|
1260 |
|
CPU *cpu; |
1261 |
|
|
1262 |
|
init_ref_dct(); |
1263 |
|
|
1264 |
|
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
1265 |
|
{ |
1266 |
|
if (!init_cpu(cpu)) |
1267 |
|
continue; |
1268 |
|
|
1269 |
|
printf( "\n===== IEEE test for %s ==== (Min=%d Max=%d Sign=%d Loops=%d)\n", |
1270 |
|
cpu->name, Min, Max, Sign, Loops); |
1271 |
|
|
1272 |
|
init_stats(&IStats); |
1273 |
|
init_stats(&FStats); |
1274 |
|
|
1275 |
|
ieee_reseed(1); |
1276 |
|
for(n=0; n<Loops; ++n) |
1277 |
|
{ |
1278 |
|
for(i=0; i<64; ++i) |
1279 |
|
Blk0[i] = (short)ieee_rand(Min,Max) * Sign; |
1280 |
|
|
1281 |
|
/* hmm, I'm not quite sure this is exactly */ |
1282 |
|
/* the tests described in the norm. check... */ |
1283 |
|
|
1284 |
|
memcpy(Ref_FDCT, Blk0, 64*sizeof(short)); |
1285 |
|
ref_fdct(Ref_FDCT); |
1286 |
|
for(i=0; i<64; i++) CLAMP( Ref_FDCT[i], 2048 ); |
1287 |
|
|
1288 |
|
memcpy(Blk, Blk0, 64*sizeof(short)); |
1289 |
|
emms(); fdct(Blk); emms(); |
1290 |
|
for(i=0; i<64; i++) CLAMP( Blk[i], 2048 ); |
1291 |
|
|
1292 |
|
store_stats(&FStats, Blk, Ref_FDCT); |
1293 |
|
|
1294 |
|
|
1295 |
|
memcpy(Ref_IDCT, Ref_FDCT, 64*sizeof(short)); |
1296 |
|
ref_idct(Ref_IDCT); |
1297 |
|
for (i=0; i<64; i++) CLAMP( Ref_IDCT[i], 256 ); |
1298 |
|
|
1299 |
|
memcpy(iBlk, Ref_FDCT, 64*sizeof(short)); |
1300 |
|
emms(); idct(iBlk); emms(); |
1301 |
|
for(i=0; i<64; i++) CLAMP( iBlk[i], 256 ); |
1302 |
|
|
1303 |
|
store_stats(&IStats, iBlk, Ref_IDCT); |
1304 |
|
} |
1305 |
|
|
1306 |
|
|
1307 |
|
printf( "\n -- FDCT report --\n" ); |
1308 |
|
// print_stats(&FStats); |
1309 |
|
report_stats(&FStats, 0); /* so far I know, IEEE1180 says nothing for fdct */ |
1310 |
|
|
1311 |
|
for(i=0; i<64; i++) Blk[i] = 0; |
1312 |
|
emms(); fdct(Blk); emms(); |
1313 |
|
for(m=i=0; i<64; i++) if (Blk[i]!=0) m++; |
1314 |
|
printf( "FDCT(0) == 0 ? %s\n", (m!=0) ? "NOPE!" : "yup." ); |
1315 |
|
|
1316 |
|
printf( "\n -- IDCT report --\n" ); |
1317 |
|
// print_stats(&IStats); |
1318 |
|
report_stats(&IStats, ILimits); |
1319 |
|
|
1320 |
|
|
1321 |
|
for(i=0; i<64; i++) Blk[i] = 0; |
1322 |
|
emms(); idct(Blk); emms(); |
1323 |
|
for(m=i=0; i<64; i++) if (Blk[i]!=0) m++; |
1324 |
|
printf( "IDCT(0) == 0 ? %s\n", (m!=0) ? "NOPE!" : "yup." ); |
1325 |
|
} |
1326 |
|
} |
1327 |
|
|
1328 |
|
|
1329 |
|
void test_dct_saturation(int Min, int Max) |
1330 |
|
{ |
1331 |
|
/* test behaviour on input range fringe */ |
1332 |
|
|
1333 |
|
int i, n, p; |
1334 |
|
CPU *cpu; |
1335 |
|
// const short IDCT_MAX = 2047; /* 12bits input */ |
1336 |
|
// const short IDCT_MIN = -2048; |
1337 |
|
// const short IDCT_OUT = 256; /* 9bits ouput */ |
1338 |
|
const int Partitions = 4; |
1339 |
|
const int Loops = 10000 / Partitions; |
1340 |
|
|
1341 |
|
init_ref_dct(); |
1342 |
|
|
1343 |
|
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
1344 |
|
{ |
1345 |
|
short Blk0[64], Blk[64]; |
1346 |
|
STATS_8x8 Stats; |
1347 |
|
|
1348 |
|
if (!init_cpu(cpu)) |
1349 |
|
continue; |
1350 |
|
|
1351 |
|
printf( "\n===== IEEE test for %s Min=%d Max=%d =====\n", |
1352 |
|
cpu->name, Min, Max ); |
1353 |
|
|
1354 |
|
/* FDCT tests // */ |
1355 |
|
|
1356 |
|
init_stats(&Stats); |
1357 |
|
|
1358 |
|
/* test each computation channels separately */ |
1359 |
|
for(i=0; i<64; i++) Blk[i] = Blk0[i] = ((i/8)==(i%8)) ? Max : 0; |
1360 |
|
ref_fdct(Blk0); |
1361 |
|
emms(); fdct(Blk); emms(); |
1362 |
|
store_stats(&Stats, Blk, Blk0); |
1363 |
|
|
1364 |
|
for(i=0; i<64; i++) Blk[i] = Blk0[i] = ((i/8)==(i%8)) ? Min : 0; |
1365 |
|
ref_fdct(Blk0); |
1366 |
|
emms(); fdct(Blk); emms(); |
1367 |
|
store_stats(&Stats, Blk, Blk0); |
1368 |
|
|
1369 |
|
/* randomly saturated inputs */ |
1370 |
|
for(p=0; p<Partitions; ++p) |
1371 |
|
{ |
1372 |
|
for(n=0; n<Loops; ++n) |
1373 |
|
{ |
1374 |
|
for(i=0; i<64; ++i) |
1375 |
|
Blk0[i] = Blk[i] = (ieee_rand(0,Partitions)>=p)? Max : Min; |
1376 |
|
ref_fdct(Blk0); |
1377 |
|
emms(); fdct(Blk); emms(); |
1378 |
|
store_stats(&Stats, Blk, Blk0); |
1379 |
|
} |
1380 |
|
} |
1381 |
|
printf( "\n -- FDCT saturation report --\n" ); |
1382 |
|
report_stats(&Stats, 0); |
1383 |
|
|
1384 |
|
|
1385 |
|
/* IDCT tests // */ |
1386 |
|
#if 0 |
1387 |
|
/* no finished yet */ |
1388 |
|
|
1389 |
|
init_stats(&Stats); |
1390 |
|
|
1391 |
|
/* test each computation channel separately */ |
1392 |
|
for(i=0; i<64; i++) Blk[i] = Blk0[i] = ((i/8)==(i%8)) ? IDCT_MAX : 0; |
1393 |
|
ref_idct(Blk0); |
1394 |
|
emms(); idct(Blk); emms(); |
1395 |
|
for(i=0; i<64; i++) { CLAMP(Blk0[i], IDCT_OUT); CLAMP(Blk[i], IDCT_OUT); } |
1396 |
|
store_stats(&Stats, Blk, Blk0); |
1397 |
|
|
1398 |
|
for(i=0; i<64; i++) Blk[i] = Blk0[i] = ((i/8)==(i%8)) ? IDCT_MIN : 0; |
1399 |
|
ref_idct(Blk0); |
1400 |
|
emms(); idct(Blk); emms(); |
1401 |
|
for(i=0; i<64; i++) { CLAMP(Blk0[i], IDCT_OUT); CLAMP(Blk[i], IDCT_OUT); } |
1402 |
|
store_stats(&Stats, Blk, Blk0); |
1403 |
|
|
1404 |
|
/* randomly saturated inputs */ |
1405 |
|
for(p=0; p<Partitions; ++p) |
1406 |
|
{ |
1407 |
|
for(n=0; n<Loops; ++n) |
1408 |
|
{ |
1409 |
|
for(i=0; i<64; ++i) |
1410 |
|
Blk0[i] = Blk[i] = (ieee_rand(0,Partitions)>=p)? IDCT_MAX : IDCT_MIN; |
1411 |
|
ref_idct(Blk0); |
1412 |
|
emms(); idct(Blk); emms(); |
1413 |
|
for(i=0; i<64; i++) { CLAMP(Blk0[i],IDCT_OUT); CLAMP(Blk[i],IDCT_OUT); } |
1414 |
|
store_stats(&Stats, Blk, Blk0); |
1415 |
|
} |
1416 |
|
} |
1417 |
|
|
1418 |
|
printf( "\n -- IDCT saturation report --\n" ); |
1419 |
|
print_stats(&Stats); |
1420 |
|
report_stats(&Stats, 0); |
1421 |
|
#endif |
1422 |
|
} |
1423 |
} |
} |
1424 |
|
|
1425 |
/********************************************************************* |
/********************************************************************* |
1426 |
* measure raw decoding speed |
* measure raw decoding speed |
1427 |
*********************************************************************/ |
*********************************************************************/ |
1428 |
|
|
1429 |
void test_dec(const char *name, int width, int height, int with_chksum) |
void test_dec(const char *name, int width, int height, int ref_chksum) |
1430 |
{ |
{ |
1431 |
FILE *f = 0; |
FILE *f = 0; |
1432 |
void *dechandle = 0; |
void *dechandle = 0; |
1433 |
int xerr; |
int xerr; |
1434 |
XVID_INIT_PARAM xinit; |
xvid_gbl_init_t xinit; |
1435 |
XVID_DEC_PARAM xparam; |
xvid_dec_create_t xparam; |
1436 |
XVID_DEC_FRAME xframe; |
xvid_dec_frame_t xframe; |
1437 |
double t = 0.; |
double t = 0.; |
1438 |
int nb = 0; |
int nb = 0; |
1439 |
uint8_t *buf = 0; |
uint8_t *buf = 0; |
1440 |
uint8_t *rgb_out = 0; |
uint8_t *yuv_out = 0; |
1441 |
int buf_size, pos; |
int buf_size, pos; |
1442 |
uint32_t chksum = 0; |
uint32_t chksum = 0; |
1443 |
|
int bps = (width+31) & ~31; |
1444 |
|
|
1445 |
xinit.cpu_flags = 0; |
memset(&xinit, 0, sizeof(xinit)); |
1446 |
xvid_init(NULL, 0, &xinit, NULL); |
xinit.cpu_flags = cpu_mask; |
1447 |
printf( "API version: %d, core build:%d\n", xinit.api_version, xinit.core_build); |
xinit.version = XVID_VERSION; |
1448 |
|
xvid_global(NULL, 0, &xinit, NULL); |
1449 |
|
|
1450 |
|
memset(&xparam, 0, sizeof(xparam)); |
1451 |
xparam.width = width; |
xparam.width = width; |
1452 |
xparam.height = height; |
xparam.height = height; |
1453 |
|
xparam.version = XVID_VERSION; |
1454 |
xerr = xvid_decore(NULL, XVID_DEC_CREATE, &xparam, NULL); |
xerr = xvid_decore(NULL, XVID_DEC_CREATE, &xparam, NULL); |
1455 |
if (xerr!=XVID_ERR_OK) { |
if (xerr==XVID_ERR_FAIL) { |
1456 |
printf("can't init decoder (err=%d)\n", xerr); |
printf("ERROR: can't init decoder (err=%d)\n", xerr); |
1457 |
return; |
return; |
1458 |
} |
} |
1459 |
dechandle = xparam.handle; |
dechandle = xparam.handle; |
1461 |
|
|
1462 |
f = fopen(name, "rb"); |
f = fopen(name, "rb"); |
1463 |
if (f==0) { |
if (f==0) { |
1464 |
printf( "can't open file '%s'\n", name); |
printf( "ERROR: can't open file '%s'\n", name); |
1465 |
return; |
return; |
1466 |
} |
} |
1467 |
fseek(f, 0, SEEK_END); |
fseek(f, 0, SEEK_END); |
1468 |
buf_size = ftell(f); |
buf_size = ftell(f); |
1469 |
fseek(f, 0, SEEK_SET); |
fseek(f, 0, SEEK_SET); |
1470 |
if (buf_size<=0) { |
if (buf_size<=0) { |
1471 |
printf("error while stating file\n"); |
printf("ERROR: error while stating file\n"); |
1472 |
goto End; |
goto End; |
1473 |
} |
} |
|
else printf( "Input size: %d\n", buf_size); |
|
1474 |
|
|
1475 |
buf = malloc(buf_size); // should be enuf' |
buf = malloc(buf_size); |
1476 |
rgb_out = calloc(4, width*height); // <-room for _RGB24 |
yuv_out = calloc(1, bps*height*3/2 + 15); |
1477 |
if (buf==0 || rgb_out==0) { |
if (buf==0 || yuv_out==0) { |
1478 |
printf( "malloc failed!\n" ); |
printf( "ERROR: malloc failed!\n" ); |
1479 |
goto End; |
goto End; |
1480 |
} |
} |
1481 |
|
|
1482 |
if (fread(buf, buf_size, 1, f)!=1) { |
if (fread(buf, buf_size, 1, f)!=1) { |
1483 |
printf( "file-read failed\n" ); |
printf( "ERROR: file-read failed\n" ); |
1484 |
goto End; |
goto End; |
1485 |
} |
} |
1486 |
|
|
1488 |
pos = 0; |
pos = 0; |
1489 |
t = -gettime_usec(); |
t = -gettime_usec(); |
1490 |
while(1) { |
while(1) { |
1491 |
|
int y; |
1492 |
|
|
1493 |
|
memset(&xframe, 0, sizeof(xframe)); |
1494 |
|
xframe.version = XVID_VERSION; |
1495 |
xframe.bitstream = buf + pos; |
xframe.bitstream = buf + pos; |
1496 |
xframe.length = buf_size - pos; |
xframe.length = buf_size - pos; |
1497 |
xframe.image = rgb_out; |
xframe.output.plane[0] = (uint8_t*)(((size_t)yuv_out + 15) & ~15); |
1498 |
xframe.stride = width; |
xframe.output.plane[1] = (uint8_t*)xframe.output.plane[0] + bps*height; |
1499 |
xframe.colorspace = XVID_CSP_RGB24; |
xframe.output.plane[2] = (uint8_t*)xframe.output.plane[1] + bps/2; |
1500 |
|
xframe.output.stride[0] = bps; |
1501 |
|
xframe.output.stride[1] = bps; |
1502 |
|
xframe.output.stride[2] = bps; |
1503 |
|
xframe.output.csp = XVID_CSP_I420; |
1504 |
xerr = xvid_decore(dechandle, XVID_DEC_DECODE, &xframe, 0); |
xerr = xvid_decore(dechandle, XVID_DEC_DECODE, &xframe, 0); |
1505 |
|
if (xerr<0) { |
1506 |
|
printf("ERROR: decoding failed for frame #%d (err=%d)!\n", nb, xerr); |
1507 |
|
break; |
1508 |
|
} |
1509 |
|
else if (xerr==0) |
1510 |
|
break; |
1511 |
|
else if (verbose>0) printf("#%d %d\n", nb, xerr ); |
1512 |
|
|
1513 |
|
pos += xerr; |
1514 |
nb++; |
nb++; |
1515 |
pos += xframe.length; |
|
1516 |
if (with_chksum) { |
for(y=0; y<height/2; ++y) { |
1517 |
int k = width*height; |
chksum = calc_crc((uint8_t*)xframe.output.plane[0] + (2*y+0)*bps, width, chksum); |
1518 |
uint32_t *ptr = (uint32_t *)rgb_out; |
chksum = calc_crc((uint8_t*)xframe.output.plane[0] + (2*y+1)*bps, width, chksum); |
1519 |
while(k-->0) chksum += *ptr++; |
chksum = calc_crc((uint8_t*)xframe.output.plane[1] + y*bps, width/2, chksum); |
1520 |
|
chksum = calc_crc((uint8_t*)xframe.output.plane[2] + y*bps, width/2, chksum); |
1521 |
} |
} |
1522 |
if (pos==buf_size) |
if (pos==buf_size) |
1523 |
break; |
break; |
|
if (xerr!=XVID_ERR_OK) { |
|
|
printf("decoding failed for frame #%d (err=%d)!\n", nb, xerr); |
|
|
break; |
|
|
} |
|
1524 |
} |
} |
1525 |
t += gettime_usec(); |
t += gettime_usec(); |
1526 |
|
if (ref_chksum==0) { |
1527 |
if (t>0.) |
if (t>0.) |
1528 |
printf( "%d frames decoded in %.3f s -> %.1f FPS\n", nb, t*1.e-6f, (float)(nb*1.e6f/t) ); |
printf( "%d frames decoded in %.3f s -> %.1f FPS Checksum:0x%.8x\n", nb, t*1.e-6f, (float)(nb*1.e6f/t), chksum ); |
1529 |
if (with_chksum) |
} |
1530 |
printf("checksum: 0x%.8x\n", chksum); |
else { |
1531 |
|
printf("FPS:%.1f Checksum: 0x%.8x Expected:0x%.8x | %s\n", |
1532 |
|
t>0. ? (float)(nb*1.e6f/t) : 0.f, chksum, ref_chksum, (chksum==ref_chksum) ? "OK" : "ERROR"); |
1533 |
|
} |
1534 |
|
|
1535 |
End: |
End: |
1536 |
if (rgb_out!=0) free(rgb_out); |
if (yuv_out!=0) free(yuv_out); |
1537 |
if (buf!=0) free(buf); |
if (buf!=0) free(buf); |
1538 |
if (dechandle!=0) { |
if (dechandle!=0) { |
1539 |
xerr= xvid_decore(dechandle, XVID_DEC_DESTROY, NULL, NULL); |
xerr= xvid_decore(dechandle, XVID_DEC_DESTROY, NULL, NULL); |
1540 |
if (xerr!=XVID_ERR_OK) |
if (xerr==XVID_ERR_FAIL) |
1541 |
printf("destroy-decoder failed (err=%d)!\n", xerr); |
printf("ERROR: destroy-decoder failed (err=%d)!\n", xerr); |
1542 |
} |
} |
1543 |
if (f!=0) fclose(f); |
if (f!=0) fclose(f); |
1544 |
} |
} |
1550 |
void test_bugs1() |
void test_bugs1() |
1551 |
{ |
{ |
1552 |
CPU *cpu; |
CPU *cpu; |
1553 |
|
uint16_t mpeg_quant_matrices[64*8]; |
1554 |
|
|
1555 |
printf( "\n ===== (de)quant4_intra saturation bug? =====\n" ); |
printf( "\n ===== (de)quant4_intra saturation bug? =====\n" ); |
1556 |
|
|
1557 |
for(cpu = cpu_short_list; cpu->name!=0; ++cpu) |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
1558 |
{ |
{ |
1559 |
int i; |
int i; |
1560 |
int16_t Src[8*8], Dst[8*8]; |
int16_t Src[8*8], Dst[8*8]; |
1563 |
continue; |
continue; |
1564 |
|
|
1565 |
for(i=0; i<64; ++i) Src[i] = i-32; |
for(i=0; i<64; ++i) Src[i] = i-32; |
1566 |
set_intra_matrix( get_default_intra_matrix() ); |
set_intra_matrix( mpeg_quant_matrices, get_default_intra_matrix() ); |
1567 |
dequant4_intra(Dst, Src, 32, 5); |
dequant_mpeg_intra(Dst, Src, 31, 5, mpeg_quant_matrices); |
1568 |
printf( "dequant4_intra with CPU=%s: ", cpu->name); |
printf( "dequant_mpeg_intra with CPU=%s: ", cpu->name); |
1569 |
printf( " Out[]= " ); |
printf( " Out[]= " ); |
1570 |
for(i=0; i<64; ++i) printf( "[%d]", Dst[i]); |
for(i=0; i<64; ++i) printf( "[%d]", Dst[i]); |
1571 |
printf( "\n" ); |
printf( "\n" ); |
1573 |
|
|
1574 |
printf( "\n ===== (de)quant4_inter saturation bug? =====\n" ); |
printf( "\n ===== (de)quant4_inter saturation bug? =====\n" ); |
1575 |
|
|
1576 |
for(cpu = cpu_short_list; cpu->name!=0; ++cpu) |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
1577 |
{ |
{ |
1578 |
int i; |
int i; |
1579 |
int16_t Src[8*8], Dst[8*8]; |
int16_t Src[8*8], Dst[8*8]; |
1582 |
continue; |
continue; |
1583 |
|
|
1584 |
for(i=0; i<64; ++i) Src[i] = i-32; |
for(i=0; i<64; ++i) Src[i] = i-32; |
1585 |
set_inter_matrix( get_default_inter_matrix() ); |
set_inter_matrix( mpeg_quant_matrices, get_default_inter_matrix() ); |
1586 |
dequant4_inter(Dst, Src, 32); |
dequant_mpeg_inter(Dst, Src, 31, mpeg_quant_matrices); |
1587 |
printf( "dequant4_inter with CPU=%s: ", cpu->name); |
printf( "dequant_mpeg_inter with CPU=%s: ", cpu->name); |
1588 |
printf( " Out[]= " ); |
printf( " Out[]= " ); |
1589 |
for(i=0; i<64; ++i) printf( "[%d]", Dst[i]); |
for(i=0; i<64; ++i) printf( "[%d]", Dst[i]); |
1590 |
printf( "\n" ); |
printf( "\n" ); |
1594 |
void test_dct_precision_diffs() |
void test_dct_precision_diffs() |
1595 |
{ |
{ |
1596 |
CPU *cpu; |
CPU *cpu; |
1597 |
short Blk[8*8], Blk0[8*8]; |
DECLARE_ALIGNED_MATRIX(Blk, 8, 8, int16_t, 16); |
1598 |
|
DECLARE_ALIGNED_MATRIX(Blk0, 8, 8, int16_t, 16); |
1599 |
|
|
1600 |
printf( "\n ===== fdct/idct saturation diffs =====\n" ); |
printf( "\n ===== fdct/idct precision diffs =====\n" ); |
1601 |
|
|
1602 |
for(cpu = cpu_short_list; cpu->name!=0; ++cpu) |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
1603 |
{ |
{ |
1604 |
int i; |
int i; |
1605 |
|
|
1623 |
} |
} |
1624 |
} |
} |
1625 |
|
|
1626 |
|
void test_quant_bug() |
1627 |
|
{ |
1628 |
|
const int max_Q = 31; |
1629 |
|
int i, n, qm, q; |
1630 |
|
CPU *cpu; |
1631 |
|
DECLARE_ALIGNED_MATRIX(Src, 8, 8, int16_t, 16); |
1632 |
|
DECLARE_ALIGNED_MATRIX(Dst, 8, 8, int16_t, 16); |
1633 |
|
uint8_t Quant[8*8]; |
1634 |
|
CPU cpu_bug_list[] = { { "PLAINC", 0 }, { "MMX ", XVID_CPU_MMX }, {0,0} }; |
1635 |
|
uint16_t Crcs_Inter[2][32]; |
1636 |
|
uint16_t Crcs_Intra[2][32]; |
1637 |
|
DECLARE_ALIGNED_MATRIX(mpeg_quant_matrices, 8, 64, uint16_t, 16); |
1638 |
|
|
1639 |
|
printf( "\n ===== test MPEG4-quantize bug =====\n" ); |
1640 |
|
|
1641 |
|
for(i=0; i<64; ++i) Src[i] = 2048*(i-32)/32; |
1642 |
|
|
1643 |
|
#if 1 |
1644 |
|
for(qm=1; qm<=255; ++qm) |
1645 |
|
{ |
1646 |
|
for(i=0; i<8*8; ++i) Quant[i] = qm; |
1647 |
|
set_inter_matrix( mpeg_quant_matrices, Quant ); |
1648 |
|
|
1649 |
|
for(n=0, cpu = cpu_bug_list; cpu->name!=0; ++cpu, ++n) |
1650 |
|
{ |
1651 |
|
uint16_t s; |
1652 |
|
|
1653 |
|
if (!init_cpu(cpu)) |
1654 |
|
continue; |
1655 |
|
|
1656 |
|
for(q=1; q<=max_Q; ++q) { |
1657 |
|
emms(); |
1658 |
|
quant_mpeg_inter( Dst, Src, q, mpeg_quant_matrices ); |
1659 |
|
emms(); |
1660 |
|
for(s=0, i=0; i<64; ++i) s+=((uint16_t)Dst[i])^i; |
1661 |
|
Crcs_Inter[n][q] = s; |
1662 |
|
} |
1663 |
|
} |
1664 |
|
|
1665 |
|
for(q=1; q<=max_Q; ++q) |
1666 |
|
for(i=0; i<n-1; ++i) |
1667 |
|
if (Crcs_Inter[i][q]!=Crcs_Inter[i+1][q]) |
1668 |
|
printf( "Discrepancy Inter: qm=%d, q=%d -> %d/%d !\n", |
1669 |
|
qm, q, Crcs_Inter[i][q], Crcs_Inter[i+1][q]); |
1670 |
|
} |
1671 |
|
#endif |
1672 |
|
|
1673 |
|
#if 1 |
1674 |
|
for(qm=1; qm<=255; ++qm) |
1675 |
|
{ |
1676 |
|
for(i=0; i<8*8; ++i) Quant[i] = qm; |
1677 |
|
set_intra_matrix( mpeg_quant_matrices, Quant ); |
1678 |
|
|
1679 |
|
for(n=0, cpu = cpu_bug_list; cpu->name!=0; ++cpu, ++n) |
1680 |
|
{ |
1681 |
|
uint16_t s; |
1682 |
|
|
1683 |
|
if (!init_cpu(cpu)) |
1684 |
|
continue; |
1685 |
|
|
1686 |
|
for(q=1; q<=max_Q; ++q) { |
1687 |
|
emms(); |
1688 |
|
quant_mpeg_intra( Dst, Src, q, q, mpeg_quant_matrices); |
1689 |
|
emms(); |
1690 |
|
for(s=0, i=0; i<64; ++i) s+=((uint16_t)Dst[i])^i; |
1691 |
|
Crcs_Intra[n][q] = s; |
1692 |
|
} |
1693 |
|
} |
1694 |
|
|
1695 |
|
for(q=1; q<=max_Q; ++q) |
1696 |
|
for(i=0; i<n-1; ++i) |
1697 |
|
if (Crcs_Intra[i][q]!=Crcs_Intra[i+1][q]) |
1698 |
|
printf( "Discrepancy Intra: qm=%d, q=%d -> %d/%d!\n", |
1699 |
|
qm, q, Crcs_Inter[i][q], Crcs_Inter[i+1][q]); |
1700 |
|
} |
1701 |
|
#endif |
1702 |
|
} |
1703 |
|
/*********************************************************************/ |
1704 |
|
|
1705 |
|
static uint32_t __inline log2bin_v1(uint32_t value) |
1706 |
|
{ |
1707 |
|
int n = 0; |
1708 |
|
while (value) { |
1709 |
|
value >>= 1; |
1710 |
|
n++; |
1711 |
|
} |
1712 |
|
return n; |
1713 |
|
} |
1714 |
|
|
1715 |
|
static const uint8_t log2_tab_16[16] = { 0, 1, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4 }; |
1716 |
|
|
1717 |
|
static uint32_t __inline log2bin_v2(uint32_t value) |
1718 |
|
{ |
1719 |
|
int n = 0; |
1720 |
|
if (value & 0xffff0000) { |
1721 |
|
value >>= 16; |
1722 |
|
n += 16; |
1723 |
|
} |
1724 |
|
if (value & 0xff00) { |
1725 |
|
value >>= 8; |
1726 |
|
n += 8; |
1727 |
|
} |
1728 |
|
if (value & 0xf0) { |
1729 |
|
value >>= 4; |
1730 |
|
n += 4; |
1731 |
|
} |
1732 |
|
return n + log2_tab_16[value]; |
1733 |
|
} |
1734 |
|
|
1735 |
|
void test_log2bin() |
1736 |
|
{ |
1737 |
|
const int nb_tests = 3000*speed_ref; |
1738 |
|
int n, crc1=0, crc2=0; |
1739 |
|
uint32_t s, s0; |
1740 |
|
double t1, t2; |
1741 |
|
|
1742 |
|
t1 = gettime_usec(); |
1743 |
|
s0 = (int)(t1*31.241); |
1744 |
|
for(s=s0, n=0; n<nb_tests; ++n, s=(s*12363+31)&0x7fffffff) |
1745 |
|
crc1 += log2bin_v1(s); |
1746 |
|
t1 = (gettime_usec()-t1) / nb_tests; |
1747 |
|
|
1748 |
|
t2 = gettime_usec(); |
1749 |
|
for(s=s0, n=0; n<nb_tests; ++n, s=(s*12363+31)&0x7fffffff) |
1750 |
|
crc2 += log2bin_v2(s); |
1751 |
|
t2 = (gettime_usec() - t2) / nb_tests; |
1752 |
|
|
1753 |
|
printf( "log2bin_v1: %.3f sec crc=%d\n", t1, crc1 ); |
1754 |
|
printf( "log2bin_v2: %.3f sec crc=%d\n", t2, crc2 ); |
1755 |
|
if (crc1!=crc2) printf( " CRC ERROR !\n" ); |
1756 |
|
} |
1757 |
|
|
1758 |
|
/*********************************************************************/ |
1759 |
|
|
1760 |
|
static void __inline old_gcd(int *num, int *den) |
1761 |
|
{ |
1762 |
|
int i = *num; |
1763 |
|
while (i > 1) { |
1764 |
|
if (*num % i == 0 && *den % i == 0) { |
1765 |
|
*num /= i; |
1766 |
|
*den /= i; |
1767 |
|
i = *num; |
1768 |
|
continue; |
1769 |
|
} |
1770 |
|
i--; |
1771 |
|
} |
1772 |
|
} |
1773 |
|
|
1774 |
|
static uint32_t gcd(int num, int den) |
1775 |
|
{ |
1776 |
|
int tmp; |
1777 |
|
while( (tmp=num%den) ) { num = den; den = tmp; } |
1778 |
|
return den; |
1779 |
|
} |
1780 |
|
static void __inline new_gcd(int *num, int *den) |
1781 |
|
{ |
1782 |
|
const int div = gcd(*num, *den); |
1783 |
|
if (num) { |
1784 |
|
*num /= div; |
1785 |
|
*den /= div; |
1786 |
|
} |
1787 |
|
} |
1788 |
|
|
1789 |
|
void test_gcd() |
1790 |
|
{ |
1791 |
|
const int nb_tests = 10*speed_ref; |
1792 |
|
int i; |
1793 |
|
uint32_t crc1=0, crc2=0; |
1794 |
|
uint32_t n0, n, d0, d; |
1795 |
|
double t1, t2; |
1796 |
|
|
1797 |
|
t1 = gettime_usec(); |
1798 |
|
n0 = 0xfffff & (int)(t1*31.241); |
1799 |
|
d0 = 0xfffff & (int)( ((n0*4123)%17) | 1 ); |
1800 |
|
for(n=n0, d=d0, i=0; i<nb_tests; ++i) { |
1801 |
|
old_gcd(&n, &d); |
1802 |
|
crc1 = (((crc1>>4)^d) + ((crc1<<2)^n) ) & 0xffffff; |
1803 |
|
n = d; |
1804 |
|
d = (d*12363+31) & 0xffff; |
1805 |
|
d |= !d; |
1806 |
|
} |
1807 |
|
t1 = (gettime_usec()-t1) / nb_tests; |
1808 |
|
|
1809 |
|
t2 = gettime_usec(); |
1810 |
|
for(n=n0, d=d0, i=0; i<nb_tests; ++i) { |
1811 |
|
new_gcd(&n, &d); |
1812 |
|
crc2 = (((crc2>>4)^d) + ((crc2<<2)^n) ) & 0xffffff; |
1813 |
|
n = d; |
1814 |
|
d = (d*12363+31) & 0xffff; |
1815 |
|
d |= !d; |
1816 |
|
} |
1817 |
|
t2 = (gettime_usec() - t2) / nb_tests; |
1818 |
|
|
1819 |
|
printf( "old_gcd: %.3f sec crc=%d\n", t1, crc1 ); |
1820 |
|
printf( "new_gcd: %.3f sec crc=%d\n", t2, crc2 ); |
1821 |
|
if (crc1!=crc2) printf( " CRC ERROR !\n" ); |
1822 |
|
} |
1823 |
|
|
1824 |
|
/********************************************************************* |
1825 |
|
* test compiler |
1826 |
|
*********************************************************************/ |
1827 |
|
|
1828 |
|
void test_compiler() { |
1829 |
|
int nb_err = 0; |
1830 |
|
int32_t v; |
1831 |
|
if (sizeof(uint16_t)<2) { |
1832 |
|
printf( "ERROR: sizeof(uint16_t)<2 !!\n" ); |
1833 |
|
nb_err++; |
1834 |
|
} |
1835 |
|
if (sizeof(int16_t)<2) { |
1836 |
|
printf( "ERROR: sizeof(int16_t)<2 !!\n" ); |
1837 |
|
nb_err++; |
1838 |
|
} |
1839 |
|
if (sizeof(uint8_t)!=1) { |
1840 |
|
printf( "ERROR: sizeof(uint8_t)!=1 !!\n" ); |
1841 |
|
nb_err++; |
1842 |
|
} |
1843 |
|
if (sizeof(int8_t)!=1) { |
1844 |
|
printf( "ERROR: sizeof(int8_t)!=1 !!\n" ); |
1845 |
|
nb_err++; |
1846 |
|
} |
1847 |
|
if (sizeof(uint32_t)<4) { |
1848 |
|
printf( "ERROR: sizeof(uint32_t)<4 !!\n" ); |
1849 |
|
nb_err++; |
1850 |
|
} |
1851 |
|
if (sizeof(int32_t)<4) { |
1852 |
|
printf( "ERROR: sizeof(int32_t)<4 !!\n" ); |
1853 |
|
nb_err++; |
1854 |
|
} |
1855 |
|
/* yes, i know, this test is silly. But better be safe than sorry. :) */ |
1856 |
|
for(v=1000; v>=0; v--) { |
1857 |
|
if ( (v>>2) != v/4) |
1858 |
|
nb_err++; |
1859 |
|
} |
1860 |
|
for(v=-1000; v!=-1; v++) { |
1861 |
|
if ( (v>>2) != (v/4)-!!(v%4)) |
1862 |
|
nb_err++; |
1863 |
|
} |
1864 |
|
if (nb_err!=0) { |
1865 |
|
printf( "ERROR! please post your platform/compiler specs to xvid-devel@xvid.org !\n" ); |
1866 |
|
} |
1867 |
|
} |
1868 |
|
|
1869 |
/********************************************************************* |
/********************************************************************* |
1870 |
* main |
* main |
1871 |
*********************************************************************/ |
*********************************************************************/ |
1872 |
|
|
1873 |
int main(int argc, char *argv[]) |
static void arg_missing(const char *opt) |
1874 |
{ |
{ |
1875 |
int what = 0; |
printf( "missing argument after option '%s'\n", opt); |
1876 |
if (argc>1) what = atoi(argv[1]); |
exit(-1); |
1877 |
|
} |
1878 |
|
|
1879 |
|
int main(int argc, const char *argv[]) |
1880 |
|
{ |
1881 |
|
int c, what = 0; |
1882 |
|
int width, height; |
1883 |
|
uint32_t chksum = 0; |
1884 |
|
const char * test_bitstream = 0; |
1885 |
|
|
1886 |
|
cpu_mask = 0; // default => will use autodectect |
1887 |
|
for(c=1; c<argc; ++c) |
1888 |
|
{ |
1889 |
|
if (!strcmp(argv[c], "-v")) verbose++; |
1890 |
|
else if (!strcmp(argv[c], "-c")) cpu_mask = 0 /* PLAIN_C */ | XVID_CPU_FORCE; |
1891 |
|
else if (!strcmp(argv[c], "-mmx")) cpu_mask = XVID_CPU_MMX | XVID_CPU_FORCE; |
1892 |
|
else if (!strcmp(argv[c], "-mmxext")) cpu_mask = XVID_CPU_MMXEXT | XVID_CPU_MMX | XVID_CPU_FORCE; |
1893 |
|
else if (!strcmp(argv[c], "-sse2")) cpu_mask = XVID_CPU_SSE2 | XVID_CPU_MMXEXT | XVID_CPU_MMX | XVID_CPU_FORCE; |
1894 |
|
else if (!strcmp(argv[c], "-3dnow")) cpu_mask = XVID_CPU_3DNOW | XVID_CPU_FORCE; |
1895 |
|
else if (!strcmp(argv[c], "-3dnowe")) cpu_mask = XVID_CPU_3DNOW | XVID_CPU_3DNOWEXT | XVID_CPU_FORCE; |
1896 |
|
else if (!strcmp(argv[c], "-altivec")) cpu_mask = XVID_CPU_ALTIVEC | XVID_CPU_FORCE; |
1897 |
|
else if (!strcmp(argv[c], "-spd")) { |
1898 |
|
if (++c==argc) arg_missing( argv[argc-1] ); |
1899 |
|
speed_ref = atoi(argv[c]); |
1900 |
|
} |
1901 |
|
else if (argv[c][0]!='-') { |
1902 |
|
what = atoi(argv[c]); |
1903 |
|
if (what==9) { |
1904 |
|
if (c+4>argc) { |
1905 |
|
printf("usage: %s %d bitstream width height (checksum)\n", argv[0], what); |
1906 |
|
exit(-1); |
1907 |
|
} |
1908 |
|
test_bitstream = argv[++c]; |
1909 |
|
width = atoi(argv[++c]); |
1910 |
|
height = atoi(argv[++c]); |
1911 |
|
if (c+1<argc && argv[c+1][0]!='-') { |
1912 |
|
if (sscanf(argv[c+1], "0x%x", &chksum)!=1) { |
1913 |
|
printf( "can't read checksum value.\n" ); |
1914 |
|
exit(-1); |
1915 |
|
} |
1916 |
|
else c++; |
1917 |
|
} |
1918 |
|
// printf( "[%s] %dx%d (0x%.8x)\n", test_bitstream, width, height, chksum); |
1919 |
|
} |
1920 |
|
} |
1921 |
|
else { |
1922 |
|
printf( "unrecognized option '%s'\n", argv[c]); |
1923 |
|
exit(-1); |
1924 |
|
} |
1925 |
|
} |
1926 |
|
|
1927 |
|
|
1928 |
if (what==0 || what==1) test_dct(); |
if (what==0 || what==1) test_dct(); |
1929 |
if (what==0 || what==2) test_mb(); |
if (what==0 || what==2) test_mb(); |
1930 |
if (what==0 || what==3) test_sad(); |
if (what==0 || what==3) test_sad(); |
1931 |
if (what==0 || what==4) test_transfer(); |
if (what==0 || what==4) test_transfer(); |
1932 |
if (what==0 || what==5) test_quant(); |
if (what==0 || what==5) test_quant(); |
1933 |
if (what==0 || what==6) test_cbp(); |
if (what==0 || what==6) test_cbp(); |
1934 |
|
if (what==0 || what==10) test_sse(); |
1935 |
if (what==8) { |
if (what==0 || what==11) test_log2bin(); |
1936 |
int width, height; |
if (what==0 || what==12) test_gcd(); |
1937 |
if (argc<5) { |
if (what==0 || what==13) test_compiler(); |
1938 |
printf("usage: %s %d [bitstream] [width] [height]\n", argv[0], what); |
|
1939 |
return 1; |
|
1940 |
} |
if (what==7) { |
1941 |
width = atoi(argv[3]); |
test_IEEE1180_compliance(-256, 255, 1); |
1942 |
height = atoi(argv[4]); |
test_IEEE1180_compliance(-256, 255,-1); |
1943 |
test_dec(argv[2], width, height, (argc>5)); |
test_IEEE1180_compliance( -5, 5, 1); |
1944 |
|
test_IEEE1180_compliance( -5, 5,-1); |
1945 |
|
test_IEEE1180_compliance(-300, 300, 1); |
1946 |
|
test_IEEE1180_compliance(-300, 300,-1); |
1947 |
} |
} |
1948 |
|
if (what==8) test_dct_saturation(-256, 255); |
1949 |
|
|
1950 |
|
if (test_bitstream) |
1951 |
|
test_dec(test_bitstream, width, height, chksum); |
1952 |
if (what==-1) { |
if (what==-1) { |
|
test_bugs1(); |
|
1953 |
test_dct_precision_diffs(); |
test_dct_precision_diffs(); |
1954 |
|
test_bugs1(); |
1955 |
} |
} |
1956 |
return 0; |
if (what==-2) |
1957 |
|
test_quant_bug(); |
1958 |
|
|
1959 |
|
if ((what >= 0 && what <= 6) || what == 10) { |
1960 |
|
printf("\n\n" |
1961 |
|
"NB: If a function isn't optimised for a specific set of intructions,\n" |
1962 |
|
" a C function is used instead. So don't panic if some functions\n" |
1963 |
|
" may appear to be slow.\n"); |
1964 |
|
} |
1965 |
|
|
1966 |
|
#ifdef ARCH_IS_IA32 |
1967 |
|
if (what == 0 || what == 5) { |
1968 |
|
printf("\n" |
1969 |
|
"NB: MMX mpeg4 quantization is known to have very small errors (+/-1 magnitude)\n" |
1970 |
|
" for 1 or 2 coefficients a block. This is mainly caused by the fact the unit\n" |
1971 |
|
" test goes far behind the usual limits of real encoding. Please do not report\n" |
1972 |
|
" this error to the developers.\n"); |
1973 |
} |
} |
1974 |
|
#endif |
1975 |
|
|
1976 |
/********************************************************************* |
return 0; |
1977 |
* 'Reference' output (except for timing) on a PIII 1.13Ghz/linux |
} |
|
*********************************************************************/ |
|
|
/* |
|
1978 |
|
|
1979 |
===== test fdct/idct ===== |
/*********************************************************************/ |
|
PLAINC - 2.631 usec iCrc=3 fCrc=-85 |
|
|
MMX - 0.596 usec iCrc=3 fCrc=-67 |
|
|
MMXEXT - 0.608 usec iCrc=3 fCrc=-67 |
|
|
SSE2 - 0.605 usec iCrc=3 fCrc=-67 |
|
|
3DNOW - skipped... |
|
|
3DNOWE - skipped... |
|
|
|
|
|
=== test block motion === |
|
|
PLAINC - interp- h-round0 1.031 usec iCrc=8107 |
|
|
PLAINC - round1 1.022 usec iCrc=8100 |
|
|
PLAINC - interp- v-round0 1.002 usec iCrc=8108 |
|
|
PLAINC - round1 1.011 usec iCrc=8105 |
|
|
PLAINC - interp-hv-round0 1.623 usec iCrc=8112 |
|
|
PLAINC - round1 1.621 usec iCrc=8103 |
|
|
PLAINC - interpolate8x8_c 0.229 usec iCrc=8107 |
|
|
--- |
|
|
MMX - interp- h-round0 0.105 usec iCrc=8107 |
|
|
MMX - round1 0.105 usec iCrc=8100 |
|
|
MMX - interp- v-round0 0.106 usec iCrc=8108 |
|
|
MMX - round1 0.107 usec iCrc=8105 |
|
|
MMX - interp-hv-round0 0.145 usec iCrc=8112 |
|
|
MMX - round1 0.145 usec iCrc=8103 |
|
|
MMX - interpolate8x8_c 0.229 usec iCrc=8107 |
|
|
--- |
|
|
MMXEXT - interp- h-round0 0.027 usec iCrc=8107 |
|
|
MMXEXT - round1 0.041 usec iCrc=8100 |
|
|
MMXEXT - interp- v-round0 0.027 usec iCrc=8108 |
|
|
MMXEXT - round1 0.040 usec iCrc=8105 |
|
|
MMXEXT - interp-hv-round0 0.070 usec iCrc=8112 |
|
|
MMXEXT - round1 0.066 usec iCrc=8103 |
|
|
MMXEXT - interpolate8x8_c 0.027 usec iCrc=8107 |
|
|
--- |
|
|
SSE2 - interp- h-round0 0.106 usec iCrc=8107 |
|
|
SSE2 - round1 0.105 usec iCrc=8100 |
|
|
SSE2 - interp- v-round0 0.106 usec iCrc=8108 |
|
|
SSE2 - round1 0.106 usec iCrc=8105 |
|
|
SSE2 - interp-hv-round0 0.145 usec iCrc=8112 |
|
|
SSE2 - round1 0.145 usec iCrc=8103 |
|
|
SSE2 - interpolate8x8_c 0.237 usec iCrc=8107 |
|
|
--- |
|
|
3DNOW - skipped... |
|
|
3DNOWE - skipped... |
|
|
|
|
|
====== test SAD ====== |
|
|
PLAINC - sad8 0.296 usec sad=3776 |
|
|
PLAINC - sad16 1.599 usec sad=27214 |
|
|
PLAINC - sad16bi 2.350 usec sad=26274 |
|
|
PLAINC - dev16 1.610 usec sad=3344 |
|
|
--- |
|
|
MMX - sad8 0.057 usec sad=3776 |
|
|
MMX - sad16 0.178 usec sad=27214 |
|
|
MMX - sad16bi 2.381 usec sad=26274 |
|
|
MMX - dev16 0.312 usec sad=3344 |
|
|
--- |
|
|
MMXEXT - sad8 0.036 usec sad=3776 |
|
|
MMXEXT - sad16 0.106 usec sad=27214 |
|
|
MMXEXT - sad16bi 0.182 usec sad=26274 |
|
|
MMXEXT - dev16 0.193 usec sad=3344 |
|
|
--- |
|
|
SSE2 - sad8 0.057 usec sad=3776 |
|
|
SSE2 - sad16 0.178 usec sad=27214 |
|
|
SSE2 - sad16bi 2.427 usec sad=26274 |
|
|
SSE2 - dev16 0.313 usec sad=3344 |
|
|
--- |
|
|
3DNOW - skipped... |
|
|
3DNOWE - skipped... |
|
|
|
|
|
=== test transfer === |
|
|
PLAINC - 8to16 0.124 usec crc=28288 |
|
|
PLAINC - 16to8 0.753 usec crc=28288 |
|
|
PLAINC - 8to8 0.041 usec crc=20352 |
|
|
PLAINC - 16to8add 0.916 usec crc=25536 |
|
|
PLAINC - 8to16sub 0.812 usec crc1=28064 crc2=16256 |
|
|
PLAINC - 8to16sub2 0.954 usec crc=20384 |
|
|
--- |
|
|
MMX - 8to16 0.037 usec crc=28288 |
|
|
MMX - 16to8 0.016 usec crc=28288 |
|
|
MMX - 8to8 0.018 usec crc=20352 |
|
|
MMX - 16to8add 0.044 usec crc=25536 |
|
|
MMX - 8to16sub 0.065 usec crc1=28064 crc2=16256 |
|
|
MMX - 8to16sub2 0.110 usec crc=20384 |
|
|
--- |
|
|
MMXEXT - 8to16 0.032 usec crc=28288 |
|
|
MMXEXT - 16to8 0.023 usec crc=28288 |
|
|
MMXEXT - 8to8 0.018 usec crc=20352 |
|
|
MMXEXT - 16to8add 0.041 usec crc=25536 |
|
|
MMXEXT - 8to16sub 0.065 usec crc1=28064 crc2=16256 |
|
|
MMXEXT - 8to16sub2 0.069 usec crc=20384 |
|
|
--- |
|
|
|
|
|
===== test quant ===== |
|
|
PLAINC - quant4_intra 78.889 usec crc=55827 |
|
|
PLAINC - quant4_inter 71.957 usec crc=58201 |
|
|
PLAINC - dequant4_intra 34.968 usec crc=193340 |
|
|
PLAINC - dequant4_inter 40.792 usec crc=116483 |
|
|
PLAINC - quant_intra 30.845 usec crc=56885 |
|
|
PLAINC - quant_inter 34.842 usec crc=58056 |
|
|
PLAINC - dequant_intra 33.211 usec crc=-7936 |
|
|
PLAINC - dequant_inter 45.486 usec crc=-33217 |
|
|
--- |
|
|
MMX - quant4_intra 9.030 usec crc=55827 |
|
|
MMX - quant4_inter 8.234 usec crc=58201 |
|
|
MMX - dequant4_intra 18.330 usec crc=193340 |
|
|
MMX - dequant4_inter 19.181 usec crc=116483 |
|
|
MMX - quant_intra 7.124 usec crc=56885 |
|
|
MMX - quant_inter 6.861 usec crc=58056 |
|
|
MMX - dequant_intra 9.048 usec crc=-7936 |
|
|
MMX - dequant_inter 8.203 usec crc=-33217 |
|
|
--- |
|
|
MMXEXT - quant4_intra 9.045 usec crc=55827 |
|
|
MMXEXT - quant4_inter 8.232 usec crc=58201 |
|
|
MMXEXT - dequant4_intra 18.250 usec crc=193340 |
|
|
MMXEXT - dequant4_inter 19.256 usec crc=116483 |
|
|
MMXEXT - quant_intra 7.121 usec crc=56885 |
|
|
MMXEXT - quant_inter 6.855 usec crc=58056 |
|
|
MMXEXT - dequant_intra 9.034 usec crc=-7936 |
|
|
MMXEXT - dequant_inter 8.202 usec crc=-33217 |
|
|
--- |
|
|
|
|
|
===== test cbp ===== |
|
|
PLAINC - calc_cbp#1 0.545 usec cbp=0x15 |
|
|
PLAINC - calc_cbp#2 0.540 usec cbp=0x38 |
|
|
PLAINC - calc_cbp#3 0.477 usec cbp=0xf |
|
|
PLAINC - calc_cbp#4 0.739 usec cbp=0x5 |
|
|
--- |
|
|
MMX - calc_cbp#1 0.136 usec cbp=0x15 |
|
|
MMX - calc_cbp#2 0.131 usec cbp=0x38 |
|
|
MMX - calc_cbp#3 0.132 usec cbp=0xf |
|
|
MMX - calc_cbp#4 0.135 usec cbp=0x5 |
|
|
--- |
|
|
SSE2 - calc_cbp#1 0.135 usec cbp=0x15 |
|
|
SSE2 - calc_cbp#2 0.131 usec cbp=0x38 |
|
|
SSE2 - calc_cbp#3 0.134 usec cbp=0xf |
|
|
SSE2 - calc_cbp#4 0.136 usec cbp=0x5 |
|
|
--- |
|
|
*/ |
|