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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
21 |
* |
* |
22 |
* $Id: xvid_bench.c,v 1.11 2004-03-22 22:36:23 edgomez Exp $ |
* $Id: xvid_bench.c,v 1.38 2008-11-26 23:37:28 Isibaar Exp $ |
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 (should plug some .md5 here)... |
|
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 |
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 |
|
#include "bitstream/bitstream.h" |
62 |
|
|
63 |
#include <math.h> |
#include <math.h> |
64 |
|
|
66 |
#define M_PI 3.14159265358979323846 |
#define M_PI 3.14159265358979323846 |
67 |
#endif |
#endif |
68 |
|
|
69 |
const int speed_ref = 100; /* on slow machines, decrease this value */ |
int speed_ref = 100; /* on slow machines, decrease this value */ |
70 |
|
int verbose = 0; |
71 |
|
unsigned int cpu_mask; |
72 |
|
|
73 |
/********************************************************************* |
/********************************************************************* |
74 |
* misc |
* misc |
84 |
#else |
#else |
85 |
clock_t clk; |
clock_t clk; |
86 |
clk = clock(); |
clk = clock(); |
87 |
return clk * 1000000 / CLOCKS_PER_SEC; |
return clk * 1000. / CLOCKS_PER_SEC; /* clock() returns time in Milliseconds */ |
88 |
#endif |
#endif |
89 |
} |
} |
90 |
|
|
113 |
unsigned int cpu; |
unsigned int cpu; |
114 |
} CPU; |
} CPU; |
115 |
|
|
116 |
CPU cpu_list[] = |
CPU cpu_list[] = { |
117 |
{ { "PLAINC", 0 } |
{ "PLAINC ", 0 }, |
118 |
#ifdef ARCH_IS_IA32 |
#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) |
119 |
, { "MMX ", XVID_CPU_MMX } |
{ "MMX ", XVID_CPU_MMX }, |
120 |
, { "MMXEXT", XVID_CPU_MMXEXT | XVID_CPU_MMX } |
{ "MMXEXT ", XVID_CPU_MMXEXT | XVID_CPU_MMX }, |
121 |
, { "SSE2 ", XVID_CPU_SSE2 | XVID_CPU_MMX } |
{ "SSE2 ", XVID_CPU_SSE2 | XVID_CPU_MMX }, |
122 |
, { "3DNOW ", XVID_CPU_3DNOW } |
{ "SSE3 ", XVID_CPU_SSE3 | XVID_CPU_SSE2 | XVID_CPU_MMX }, |
123 |
, { "3DNOWE", XVID_CPU_3DNOW | XVID_CPU_3DNOWEXT } |
{ "SSE41 ", XVID_CPU_SSE41| XVID_CPU_SSE3 | XVID_CPU_SSE2 | XVID_CPU_MMX }, |
124 |
|
{ "3DNOW ", XVID_CPU_3DNOW }, |
125 |
|
{ "3DNOWE ", XVID_CPU_3DNOW | XVID_CPU_3DNOWEXT }, |
126 |
#endif |
#endif |
127 |
//, { "IA64 ", XVID_CPU_IA64 } |
#ifdef ARCH_IS_PPC |
128 |
//, { "TSC ", XVID_CPU_TSC } |
{ "ALTIVEC", XVID_CPU_ALTIVEC }, |
|
, { 0, 0 } }; |
|
|
|
|
|
CPU cpu_short_list[] = |
|
|
{ { "PLAINC", 0 } |
|
|
#ifdef ARCH_IS_IA32 |
|
|
, { "MMX ", XVID_CPU_MMX } |
|
|
//, { "MMXEXT", XVID_CPU_MMXEXT | XVID_CPU_MMX } |
|
129 |
#endif |
#endif |
130 |
//, { "IA64 ", XVID_CPU_IA64 } |
#ifdef ARCH_IS_IA64 |
131 |
, { 0, 0 } }; |
// { "IA64 ", XVID_CPU_IA64 }, |
|
|
|
|
CPU cpu_short_list2[] = |
|
|
{ { "PLAINC", 0 } |
|
|
#ifdef ARCH_IS_IA32 |
|
|
, { "MMX ", XVID_CPU_MMX } |
|
|
, { "SSE2 ", XVID_CPU_SSE2 | XVID_CPU_MMX } |
|
132 |
#endif |
#endif |
133 |
, { 0, 0 } }; |
// { "TSC ", XVID_CPU_TSC }, |
134 |
|
{ 0, 0 } |
135 |
|
}; |
136 |
|
|
137 |
|
|
138 |
int init_cpu(CPU *cpu) |
int init_cpu(CPU *cpu) |
245 |
}; |
}; |
246 |
|
|
247 |
uint32_t |
uint32_t |
248 |
calc_crc(uint8_t *mem, int len, uint32_t initial) |
calc_crc(uint8_t *mem, int len, uint32_t crc) |
249 |
{ |
{ |
|
|
|
|
register unsigned int crc; |
|
|
|
|
|
crc = initial; |
|
|
|
|
250 |
while( len >= 8) { |
while( len >= 8) { |
251 |
DO8(mem, crc); |
DO8(mem, crc); |
252 |
len -= 8; |
len -= 8; |
257 |
len--; |
len--; |
258 |
} |
} |
259 |
|
|
260 |
return(crc); |
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 |
/********************************************************************* |
/********************************************************************* |
413 |
|
|
414 |
t = gettime_usec(); |
t = gettime_usec(); |
415 |
emms(); |
emms(); |
416 |
|
for(tst=0; tst<nb_tests; ++tst) s = sad8bi(Cur, Ref1, Ref2, 8); |
417 |
|
emms(); |
418 |
|
t = (gettime_usec() - t) / nb_tests; |
419 |
|
printf( "%s - sad8bi %.3f usec sad=%d %s\n", |
420 |
|
cpu->name, t, s, |
421 |
|
(s!=4002)?"| ERROR": "" ); |
422 |
|
|
423 |
|
t = gettime_usec(); |
424 |
|
emms(); |
425 |
for(tst=0; tst<nb_tests; ++tst) s = dev16(Cur, 16); |
for(tst=0; tst<nb_tests; ++tst) s = dev16(Cur, 16); |
426 |
emms(); |
emms(); |
427 |
t = (gettime_usec() - t) / nb_tests; |
t = (gettime_usec() - t) / nb_tests; |
527 |
(iCrc!=8107)?"| ERROR": "" ); |
(iCrc!=8107)?"| ERROR": "" ); |
528 |
#endif |
#endif |
529 |
|
|
530 |
|
/* New functions for field prediction by CK 1.10.2005 */ |
531 |
|
#pragma NEW8X4 |
532 |
|
TEST_MB(interpolate8x4_halfpel_h, 0); |
533 |
|
printf("%s - interpfield-h -round0 %.3f usec crc32=0x%08x %s\n", |
534 |
|
cpu->name, t, iCrc, |
535 |
|
(iCrc!=0x9538d6df)?"| ERROR": "" ); |
536 |
|
|
537 |
|
TEST_MB(interpolate8x4_halfpel_h, 1); |
538 |
|
printf("%s - round1 %.3f usec crc32=0x%08x %s\n", |
539 |
|
cpu->name, t, iCrc, |
540 |
|
(iCrc!=0xde5f1db4)?"| ERROR": "" ); |
541 |
|
|
542 |
|
|
543 |
|
TEST_MB(interpolate8x4_halfpel_v, 0); |
544 |
|
printf("%s - interpfield- v-round0 %.3f usec crc32=0x%08x %s\n", |
545 |
|
cpu->name, t, iCrc, |
546 |
|
(iCrc!=0xea5a69ef)?"| ERROR": "" ); |
547 |
|
|
548 |
|
TEST_MB(interpolate8x4_halfpel_v, 1); |
549 |
|
printf("%s - round1 %.3f usec crc32=0x%08x %s\n", |
550 |
|
cpu->name, t, iCrc, |
551 |
|
(iCrc!=0x4f10ec0f)?"| ERROR": "" ); |
552 |
|
|
553 |
|
|
554 |
|
TEST_MB(interpolate8x4_halfpel_hv, 0); |
555 |
|
printf("%s - interpfield-hv-round0 %.3f usec crc32=0x%08x %s\n", |
556 |
|
cpu->name, t, iCrc, |
557 |
|
(iCrc!=0xf97ee367)?"| ERROR": "" ); |
558 |
|
|
559 |
|
TEST_MB(interpolate8x4_halfpel_hv, 1); |
560 |
|
printf("%s - round1 %.3f usec crc32=0x%08x %s\n", |
561 |
|
cpu->name, t, iCrc, |
562 |
|
(iCrc!=0xb6a9f581)?"| ERROR": "" ); |
563 |
|
/* End of 8x4 functions */ |
564 |
|
|
565 |
printf( " --- \n" ); |
printf( " --- \n" ); |
566 |
} |
} |
567 |
} |
} |
568 |
|
|
569 |
|
#undef ENTER |
570 |
|
#undef LEAVE |
571 |
|
#undef TEST_MB |
572 |
|
#undef TEST_MB2 |
573 |
|
|
574 |
/********************************************************************* |
/********************************************************************* |
575 |
* test transfer |
* test transfer |
576 |
*********************************************************************/ |
*********************************************************************/ |
600 |
} \ |
} \ |
601 |
emms(); \ |
emms(); \ |
602 |
t = (gettime_usec()-t -overhead) / nb_tests;\ |
t = (gettime_usec()-t -overhead) / nb_tests;\ |
603 |
s = calc_crc((uint8_t*)(DST), sizeof((DST)), CRC32_INITIAL) |
byte_swap((uint8_t*)(DST), 8*32*sizeof((DST)[0]), sizeof((DST)[0])); \ |
604 |
|
s = calc_crc((uint8_t*)(DST), 8*32*sizeof((DST)[0]), CRC32_INITIAL) |
605 |
|
|
606 |
#define TEST_TRANSFER(FUNC, DST, SRC) \ |
#define TEST_TRANSFER(FUNC, DST, SRC) \ |
607 |
TEST_TRANSFER_BEGIN(DST); \ |
TEST_TRANSFER_BEGIN(DST); \ |
627 |
} \ |
} \ |
628 |
emms(); \ |
emms(); \ |
629 |
t = (gettime_usec()-t -overhead) / nb_tests;\ |
t = (gettime_usec()-t -overhead) / nb_tests;\ |
630 |
s = calc_crc((uint8_t*)(DST), sizeof((DST)), CRC32_INITIAL) |
byte_swap((uint8_t*)(DST), 8*32*sizeof((DST)[0]), sizeof((DST)[0])); \ |
631 |
|
s = calc_crc((uint8_t*)(DST), 8*32*sizeof((DST)[0]), CRC32_INITIAL) |
632 |
|
|
633 |
#define TEST_TRANSFER2(FUNC, DST, SRC, R1) \ |
#define TEST_TRANSFER2(FUNC, DST, SRC, R1) \ |
634 |
TEST_TRANSFER2_BEGIN(DST,SRC); \ |
TEST_TRANSFER2_BEGIN(DST,SRC); \ |
645 |
const int nb_tests = 4000*speed_ref; |
const int nb_tests = 4000*speed_ref; |
646 |
int i; |
int i; |
647 |
CPU *cpu; |
CPU *cpu; |
648 |
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]; |
649 |
int16_t Src16[8*32], Dst16[8*32]; |
// int16_t Src16[8*32], Dst16[8*32]; |
650 |
|
DECLARE_ALIGNED_MATRIX(Src8, 8, 32, uint8_t, CACHE_LINE); |
651 |
|
DECLARE_ALIGNED_MATRIX(Dst8, 8, 32, uint8_t, CACHE_LINE); |
652 |
|
DECLARE_ALIGNED_MATRIX(Ref1, 8, 32, uint8_t, CACHE_LINE); |
653 |
|
DECLARE_ALIGNED_MATRIX(Ref2, 8, 32, uint8_t, CACHE_LINE); |
654 |
|
DECLARE_ALIGNED_MATRIX(Src16, 8, 32, uint16_t, CACHE_LINE); |
655 |
|
DECLARE_ALIGNED_MATRIX(Dst16, 8, 32, uint16_t, CACHE_LINE); |
656 |
|
|
657 |
printf( "\n === test transfer ===\n" ); |
printf( "\n === test transfer ===\n" ); |
658 |
|
|
674 |
cpu->name, t, s, |
cpu->name, t, s, |
675 |
(s!=0xee7ccbb4)?"| ERROR": ""); |
(s!=0xee7ccbb4)?"| ERROR": ""); |
676 |
|
|
677 |
|
/* New functions for field prediction by CK 1.10.2005 */ |
678 |
|
#pragma NEW8X4 |
679 |
|
TEST_TRANSFER(transfer8x4_copy, Dst8, Src8); |
680 |
|
printf("%s - 8to4 %.3f usec crc32=0x%08x %s\n", |
681 |
|
cpu->name, t, s, |
682 |
|
(s!=0xbb9c3db5)?"| ERROR": ""); |
683 |
|
/* End of new functions */ |
684 |
|
|
685 |
TEST_TRANSFER(transfer8x8_copy, Dst8, Src8); |
TEST_TRANSFER(transfer8x8_copy, Dst8, Src8); |
686 |
printf("%s - 8to8 %.3f usec crc32=0x%08x %s\n", |
printf("%s - 8to8 %.3f usec crc32=0x%08x %s\n", |
687 |
cpu->name, t, s, |
cpu->name, t, s, |
695 |
TEST_TRANSFER2(transfer_8to16sub, Dst16, Src8, Ref1); |
TEST_TRANSFER2(transfer_8to16sub, Dst16, Src8, Ref1); |
696 |
{ |
{ |
697 |
int s1, s2; |
int s1, s2; |
698 |
s1 = calc_crc((uint8_t*)Dst16, sizeof(Dst16), CRC32_INITIAL); |
s1 = calc_crc((uint8_t*)Dst16, 8*32*sizeof(Dst16[0]), CRC32_INITIAL); |
699 |
s2 = calc_crc((uint8_t*)Src8, sizeof(Src8), CRC32_INITIAL); |
s2 = calc_crc((uint8_t*)Src8, 8*32*sizeof(Src8[0]), CRC32_INITIAL); |
700 |
printf("%s - 8to16sub %.3f usec crc32(1)=0x%08x crc32(2)=0x%08x %s %s\n", |
printf("%s - 8to16sub %.3f usec crc32(1)=0x%08x crc32(2)=0x%08x %s %s\n", |
701 |
cpu->name, t, s1, s2, |
cpu->name, t, s1, s2, |
702 |
(s1!=0xa1e07163)?"| ERROR1": "", |
(s1!=0xa1e07163)?"| ERROR1": "", |
725 |
for(q=1; q<=max_Q; ++q) { \ |
for(q=1; q<=max_Q; ++q) { \ |
726 |
for(tst=0; tst<nb_tests; ++tst) \ |
for(tst=0; tst<nb_tests; ++tst) \ |
727 |
(FUNC)((DST), (SRC), q, mpeg_quant_matrices); \ |
(FUNC)((DST), (SRC), q, mpeg_quant_matrices); \ |
728 |
s = calc_crc((uint8_t*)(DST), 64*sizeof(int16_t), s); \ |
byte_swap((uint8_t*)(DST), 64*sizeof((DST)[0]), sizeof((DST)[0])); \ |
729 |
|
s = calc_crc((uint8_t*)(DST), 64*sizeof((DST)[0]), s); \ |
730 |
} \ |
} \ |
731 |
emms(); \ |
emms(); \ |
732 |
} \ |
} \ |
739 |
set_intra_matrix( mpeg_quant_matrices, Quant ); \ |
set_intra_matrix( mpeg_quant_matrices, Quant ); \ |
740 |
emms(); \ |
emms(); \ |
741 |
for(q=1; q<=max_Q; ++q) { \ |
for(q=1; q<=max_Q; ++q) { \ |
742 |
|
init_intra_matrix( mpeg_quant_matrices, q ); \ |
743 |
for(tst=0; tst<nb_tests; ++tst) \ |
for(tst=0; tst<nb_tests; ++tst) \ |
744 |
(FUNC)((DST), (SRC), q, q, mpeg_quant_matrices); \ |
(FUNC)((DST), (SRC), q, q, mpeg_quant_matrices); \ |
745 |
s = calc_crc((uint8_t*)(DST), 64*sizeof(int16_t), s); \ |
byte_swap((uint8_t*)(DST), 64*sizeof((DST)[0]), sizeof((DST)[0])); \ |
746 |
|
s = calc_crc((uint8_t*)(DST), 64*sizeof((DST)[0]), s); \ |
747 |
} \ |
} \ |
748 |
emms(); \ |
emms(); \ |
749 |
} \ |
} \ |
750 |
t = (gettime_usec()-t-overhead)/nb_tests/qm |
t = (gettime_usec()-t-overhead)/nb_tests/qm |
751 |
|
|
752 |
|
#define TEST_INTRA(REFFUNC, NEWFUNC, RANGE) \ |
753 |
|
{ int32_t 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,q,mpeg_quant_matrices); \ |
761 |
|
(NEWFUNC)((Dst2),(Src),q,q,mpeg_quant_matrices); \ |
762 |
|
for(i=0;i<64;i++) \ |
763 |
|
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]); \ |
764 |
|
} \ |
765 |
|
} |
766 |
|
|
767 |
|
#define TEST_INTER(REFFUNC, NEWFUNC, RANGE) \ |
768 |
|
{ int i,q,s; \ |
769 |
|
DECLARE_ALIGNED_MATRIX(Src, 8, 8, int16_t, 16); \ |
770 |
|
DECLARE_ALIGNED_MATRIX(Dst, 8, 8, int16_t, 16); \ |
771 |
|
DECLARE_ALIGNED_MATRIX(Dst2,8, 8, int16_t, 16); \ |
772 |
|
for(q=1;q<=max_Q;q++) \ |
773 |
|
for(s=-RANGE;s<RANGE;s++) { \ |
774 |
|
for(i=0;i<64;i++) Src[i]=s; \ |
775 |
|
(REFFUNC)((Dst),(Src),q,mpeg_quant_matrices); \ |
776 |
|
(NEWFUNC)((Dst2),(Src),q,mpeg_quant_matrices); \ |
777 |
|
emms(); \ |
778 |
|
for(i=0;i<64;i++) \ |
779 |
|
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]); \ |
780 |
|
} \ |
781 |
|
} |
782 |
|
|
783 |
void test_quant() |
void test_quant() |
784 |
{ |
{ |
785 |
const int nb_tests = 1*speed_ref; |
const int32_t nb_tests = 1*speed_ref; |
786 |
const int max_Q = 31; |
const int32_t max_Q = 31; |
787 |
DECLARE_ALIGNED_MATRIX(mpeg_quant_matrices, 8, 64, uint16_t, 16); |
DECLARE_ALIGNED_MATRIX(mpeg_quant_matrices, 8, 64, uint16_t, 16); |
788 |
|
|
789 |
int i, qm; |
int32_t i, qm; |
790 |
CPU *cpu; |
CPU *cpu; |
791 |
DECLARE_ALIGNED_MATRIX(Src, 8, 8, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Src, 8, 8, int16_t, 16); |
792 |
DECLARE_ALIGNED_MATRIX(Dst, 8, 8, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Dst, 8, 8, int16_t, 16); |
793 |
|
DECLARE_ALIGNED_MATRIX(Dst2,8, 8, int16_t, 16); |
794 |
uint8_t Quant[8*8]; |
uint8_t Quant[8*8]; |
795 |
|
|
796 |
printf( "\n ===== test quant =====\n" ); |
printf( "\n ===== test quant =====\n" ); |
805 |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
806 |
{ |
{ |
807 |
double t, overhead; |
double t, overhead; |
808 |
int tst, q; |
int32_t tst, q; |
809 |
uint32_t s; |
uint32_t s; |
810 |
|
|
811 |
if (!init_cpu(cpu)) |
if (!init_cpu(cpu)) |
812 |
continue; |
continue; |
813 |
|
|
814 |
|
// exhaustive tests to compare against the (ref) C-version |
815 |
|
TEST_INTRA(quant_h263_intra_c, quant_h263_intra, 2048); |
816 |
|
TEST_INTRA(dequant_h263_intra_c, dequant_h263_intra , 512 ); |
817 |
|
TEST_INTER(quant_h263_inter_c, quant_h263_inter , 2048); |
818 |
|
TEST_INTER(dequant_h263_inter_c, dequant_h263_inter , 512 ); |
819 |
|
|
820 |
overhead = -gettime_usec(); |
overhead = -gettime_usec(); |
821 |
for(s=0,qm=1; qm<=255; ++qm) { |
for(s=0,qm=1; qm<=255; ++qm) { |
822 |
for(i=0; i<8*8; ++i) Quant[i] = qm; |
for(i=0; i<8*8; ++i) Quant[i] = qm; |
829 |
TEST_QUANT2(quant_mpeg_intra, Dst, Src); |
TEST_QUANT2(quant_mpeg_intra, Dst, Src); |
830 |
printf("%s - quant_mpeg_intra %.3f usec crc32=0x%08x %s\n", |
printf("%s - quant_mpeg_intra %.3f usec crc32=0x%08x %s\n", |
831 |
cpu->name, t, s, |
cpu->name, t, s, |
832 |
(s!=0xfd6a21a4)? "| ERROR": ""); |
(s!=0x3b999af6)? "| ERROR": ""); |
833 |
|
|
834 |
TEST_QUANT(quant_mpeg_inter, Dst, Src); |
TEST_QUANT(quant_mpeg_inter, Dst, Src); |
835 |
printf("%s - quant_mpeg_inter %.3f usec crc32=0x%08x %s\n", |
printf("%s - quant_mpeg_inter %.3f usec crc32=0x%08x %s\n", |
871 |
} |
} |
872 |
|
|
873 |
/********************************************************************* |
/********************************************************************* |
874 |
|
* test distortion operators |
875 |
|
*********************************************************************/ |
876 |
|
|
877 |
|
static void ieee_reseed(long s); |
878 |
|
static long ieee_rand(int Min, int Max); |
879 |
|
|
880 |
|
#define TEST_SSE(FUNCTION, SRC1, SRC2, STRIDE) \ |
881 |
|
do { \ |
882 |
|
t = gettime_usec(); \ |
883 |
|
tst = nb_tests; \ |
884 |
|
while((tst--)>0) sse = (FUNCTION)((SRC1), (SRC2), (STRIDE)); \ |
885 |
|
emms(); \ |
886 |
|
t = (gettime_usec() - t)/(double)nb_tests; \ |
887 |
|
} while(0) |
888 |
|
|
889 |
|
|
890 |
|
void test_sse() |
891 |
|
{ |
892 |
|
const int nb_tests = 100000*speed_ref; |
893 |
|
int i; |
894 |
|
CPU *cpu; |
895 |
|
DECLARE_ALIGNED_MATRIX(Src1, 8, 8, int16_t, 16); |
896 |
|
DECLARE_ALIGNED_MATRIX(Src2, 8, 8, int16_t, 16); |
897 |
|
DECLARE_ALIGNED_MATRIX(Src3, 8, 8, int16_t, 16); |
898 |
|
DECLARE_ALIGNED_MATRIX(Src4, 8, 8, int16_t, 16); |
899 |
|
|
900 |
|
printf( "\n ===== test sse =====\n" ); |
901 |
|
|
902 |
|
ieee_reseed(1); |
903 |
|
for(i=0; i<64; ++i) { |
904 |
|
Src1[i] = ieee_rand(-2048, 2047); |
905 |
|
Src2[i] = ieee_rand(-2048, 2047); |
906 |
|
Src3[i] = ieee_rand(-2048, 2047); |
907 |
|
Src4[i] = ieee_rand(-2048, 2047); |
908 |
|
} |
909 |
|
|
910 |
|
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
911 |
|
{ |
912 |
|
double t; |
913 |
|
int tst, sse; |
914 |
|
|
915 |
|
if (!init_cpu(cpu)) |
916 |
|
continue; |
917 |
|
|
918 |
|
/* 16 bit element blocks */ |
919 |
|
TEST_SSE(sse8_16bit, Src1, Src2, 16); |
920 |
|
printf("%s - sse8_16bit#1 %.3f usec sse=%d %s\n", |
921 |
|
cpu->name, t, sse, (sse!=182013834)?"| ERROR": ""); |
922 |
|
TEST_SSE(sse8_16bit, Src1, Src3, 16); |
923 |
|
printf("%s - sse8_16bit#2 %.3f usec sse=%d %s\n", |
924 |
|
cpu->name, t, sse, (sse!=142545203)?"| ERROR": ""); |
925 |
|
TEST_SSE(sse8_16bit, Src1, Src4, 16); |
926 |
|
printf("%s - sse8_16bit#3 %.3f usec sse=%d %s\n", |
927 |
|
cpu->name, t, sse, (sse!=146340935)?"| ERROR": ""); |
928 |
|
TEST_SSE(sse8_16bit, Src2, Src3, 16); |
929 |
|
printf("%s - sse8_16bit#4 %.3f usec sse=%d %s\n", |
930 |
|
cpu->name, t, sse, (sse!=130136661)?"| ERROR": ""); |
931 |
|
TEST_SSE(sse8_16bit, Src2, Src4, 16); |
932 |
|
printf("%s - sse8_16bit#5 %.3f usec sse=%d %s\n", |
933 |
|
cpu->name, t, sse, (sse!=136870353)?"| ERROR": ""); |
934 |
|
TEST_SSE(sse8_16bit, Src3, Src4, 16); |
935 |
|
printf("%s - sse8_16bit#6 %.3f usec sse=%d %s\n", |
936 |
|
cpu->name, t, sse, (sse!=164107772)?"| ERROR": ""); |
937 |
|
|
938 |
|
/* 8 bit element blocks */ |
939 |
|
TEST_SSE(sse8_8bit, (int8_t*)Src1, (int8_t*)Src2, 8); |
940 |
|
printf("%s - sse8_8bit#1 %.3f usec sse=%d %s\n", |
941 |
|
cpu->name, t, sse, (sse!=1356423)?"| ERROR": ""); |
942 |
|
TEST_SSE(sse8_8bit, (int8_t*)Src1, (int8_t*)Src3, 8); |
943 |
|
printf("%s - sse8_8bit#2 %.3f usec sse=%d %s\n", |
944 |
|
cpu->name, t, sse, (sse!=1173074)?"| ERROR": ""); |
945 |
|
TEST_SSE(sse8_8bit, (int8_t*)Src1, (int8_t*)Src4, 8); |
946 |
|
printf("%s - sse8_8bit#3 %.3f usec sse=%d %s\n", |
947 |
|
cpu->name, t, sse, (sse!=1092357)?"| ERROR": ""); |
948 |
|
TEST_SSE(sse8_8bit, (int8_t*)Src2, (int8_t*)Src3, 8); |
949 |
|
printf("%s - sse8_8bit#4 %.3f usec sse=%d %s\n", |
950 |
|
cpu->name, t, sse, (sse!=1360239)?"| ERROR": ""); |
951 |
|
TEST_SSE(sse8_8bit, (int8_t*)Src2, (int8_t*)Src4, 8); |
952 |
|
printf("%s - sse8_8bit#5 %.3f usec sse=%d %s\n", |
953 |
|
cpu->name, t, sse, (sse!=1208414)?"| ERROR": ""); |
954 |
|
TEST_SSE(sse8_8bit, (int8_t*)Src3, (int8_t*)Src4, 8); |
955 |
|
printf("%s - sse8_8bit#6 %.3f usec sse=%d %s\n", |
956 |
|
cpu->name, t, sse, (sse!=1099285)?"| ERROR": ""); |
957 |
|
|
958 |
|
printf(" ---\n"); |
959 |
|
} |
960 |
|
} |
961 |
|
|
962 |
|
/********************************************************************* |
963 |
* test non-zero AC counting |
* test non-zero AC counting |
964 |
*********************************************************************/ |
*********************************************************************/ |
965 |
|
|
966 |
#define TEST_CBP(FUNC, SRC) \ |
#define TEST_CBP(FUNC, SRC, NB) \ |
967 |
t = gettime_usec(); \ |
t = gettime_usec(); \ |
968 |
emms(); \ |
emms(); \ |
969 |
for(tst=0; tst<nb_tests; ++tst) { \ |
for(tst=0; tst<NB; ++tst) { \ |
970 |
cbp = (FUNC)((SRC)); \ |
cbp = (FUNC)((SRC)); \ |
971 |
} \ |
} \ |
972 |
emms(); \ |
emms(); \ |
975 |
void test_cbp() |
void test_cbp() |
976 |
{ |
{ |
977 |
const int nb_tests = 10000*speed_ref; |
const int nb_tests = 10000*speed_ref; |
978 |
int i; |
int i, n, m; |
979 |
CPU *cpu; |
CPU *cpu; |
980 |
DECLARE_ALIGNED_MATRIX(Src1, 6, 64, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Src1, 6, 64, int16_t, 16); |
981 |
DECLARE_ALIGNED_MATRIX(Src2, 6, 64, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Src2, 6, 64, int16_t, 16); |
982 |
DECLARE_ALIGNED_MATRIX(Src3, 6, 64, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Src3, 6, 64, int16_t, 16); |
983 |
DECLARE_ALIGNED_MATRIX(Src4, 6, 64, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Src4, 6, 64, int16_t, 16); |
984 |
|
DECLARE_ALIGNED_MATRIX(Src5, 6, 64, int16_t, 16); |
985 |
|
|
986 |
printf( "\n ===== test cbp =====\n" ); |
printf( "\n ===== test cbp =====\n" ); |
987 |
|
|
990 |
Src2[i] = (i<3*64); /* half-full */ |
Src2[i] = (i<3*64); /* half-full */ |
991 |
Src3[i] = ((i+32)>3*64); |
Src3[i] = ((i+32)>3*64); |
992 |
Src4[i] = (i==(3*64+2) || i==(5*64+9)); |
Src4[i] = (i==(3*64+2) || i==(5*64+9)); |
993 |
|
Src5[i] = ieee_rand(0,1) ? -1 : 1; /* +/- test */ |
994 |
} |
} |
995 |
|
|
996 |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
1001 |
if (!init_cpu(cpu)) |
if (!init_cpu(cpu)) |
1002 |
continue; |
continue; |
1003 |
|
|
1004 |
TEST_CBP(calc_cbp, Src1); |
TEST_CBP(calc_cbp, Src1, nb_tests); |
1005 |
printf("%s - calc_cbp#1 %.3f usec cbp=0x%02x\n", |
printf("%s - calc_cbp#1 %.3f usec cbp=0x%02x %s\n", |
1006 |
cpu->name, t, cbp, (cbp!=0x15)?"| ERROR": ""); |
cpu->name, t, cbp, (cbp!=0x15)?"| ERROR": ""); |
1007 |
TEST_CBP(calc_cbp, Src2); |
TEST_CBP(calc_cbp, Src2, nb_tests); |
1008 |
printf("%s - calc_cbp#2 %.3f usec cbp=0x%02x\n", |
printf("%s - calc_cbp#2 %.3f usec cbp=0x%02x %s\n", |
1009 |
cpu->name, t, cbp, (cbp!=0x38)?"| ERROR": ""); |
cpu->name, t, cbp, (cbp!=0x38)?"| ERROR": ""); |
1010 |
TEST_CBP(calc_cbp, Src3); |
TEST_CBP(calc_cbp, Src3, nb_tests); |
1011 |
printf("%s - calc_cbp#3 %.3f usec cbp=0x%02x\n", |
printf("%s - calc_cbp#3 %.3f usec cbp=0x%02x %s\n", |
1012 |
cpu->name, t, cbp, (cbp!=0x0f)?"| ERROR": "" ); |
cpu->name, t, cbp, (cbp!=0x0f)?"| ERROR": "" ); |
1013 |
TEST_CBP(calc_cbp, Src4); |
TEST_CBP(calc_cbp, Src4, nb_tests); |
1014 |
printf("%s - calc_cbp#4 %.3f usec cbp=0x%02x\n", |
printf("%s - calc_cbp#4 %.3f usec cbp=0x%02x %s\n", |
1015 |
cpu->name, t, cbp, (cbp!=0x05)?"| ERROR": "" ); |
cpu->name, t, cbp, (cbp!=0x05)?"| ERROR": "" ); |
1016 |
|
TEST_CBP(calc_cbp, Src5, nb_tests); |
1017 |
|
printf("%s - calc_cbp#4 %.3f usec cbp=0x%02x %s\n", |
1018 |
|
cpu->name, t, cbp, (cbp!=0x3f)?"| ERROR": "" ); |
1019 |
printf( " --- \n" ); |
printf( " --- \n" ); |
1020 |
} |
} |
1021 |
|
|
1022 |
|
for(cpu = cpu_list; cpu->name!=0; ++cpu) /* bench suggested by Carlo (carlo dot bramix at libero dot it) */ |
1023 |
|
{ |
1024 |
|
double t; |
1025 |
|
int tst, cbp, err; |
1026 |
|
|
1027 |
|
if (!init_cpu(cpu)) |
1028 |
|
continue; |
1029 |
|
|
1030 |
|
err = 0; |
1031 |
|
for(n=0; n<6; ++n) |
1032 |
|
{ |
1033 |
|
for(m=0; m<64; ++m) |
1034 |
|
{ |
1035 |
|
for(i=0; i<6*64; ++i) |
1036 |
|
Src1[i] = (i== (m + n*64)); |
1037 |
|
|
1038 |
|
TEST_CBP(calc_cbp, Src1, 1); |
1039 |
|
if (cbp!= (((m!=0)<<(5-n)))) |
1040 |
|
{ |
1041 |
|
printf( "%s - calc_cbp#5: ERROR at pos %d / %d!\n", cpu->name, n, m); |
1042 |
|
err = 1; |
1043 |
|
break; |
1044 |
|
} |
1045 |
|
} |
1046 |
|
} |
1047 |
|
if (!err) |
1048 |
|
printf( " %s - calc_cbp#5 : OK\n", cpu->name ); |
1049 |
|
|
1050 |
|
} |
1051 |
} |
} |
1052 |
|
|
1053 |
/********************************************************************* |
/********************************************************************* |
1441 |
* measure raw decoding speed |
* measure raw decoding speed |
1442 |
*********************************************************************/ |
*********************************************************************/ |
1443 |
|
|
1444 |
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) |
1445 |
{ |
{ |
1446 |
FILE *f = 0; |
FILE *f = 0; |
1447 |
void *dechandle = 0; |
void *dechandle = 0; |
1452 |
double t = 0.; |
double t = 0.; |
1453 |
int nb = 0; |
int nb = 0; |
1454 |
uint8_t *buf = 0; |
uint8_t *buf = 0; |
1455 |
uint8_t *rgb_out = 0; |
uint8_t *yuv_out = 0; |
1456 |
int buf_size, pos; |
int buf_size, pos; |
1457 |
uint32_t chksum = 0; |
uint32_t chksum = 0; |
1458 |
|
int bps = (width+31) & ~31; |
1459 |
|
|
1460 |
memset(&xinit, 0, sizeof(xinit)); |
memset(&xinit, 0, sizeof(xinit)); |
1461 |
xinit.cpu_flags = XVID_CPU_MMX | XVID_CPU_FORCE; |
xinit.cpu_flags = cpu_mask; |
1462 |
xinit.version = XVID_VERSION; |
xinit.version = XVID_VERSION; |
1463 |
xvid_global(NULL, 0, &xinit, NULL); |
xvid_global(NULL, 0, &xinit, NULL); |
1464 |
|
|
1468 |
xparam.version = XVID_VERSION; |
xparam.version = XVID_VERSION; |
1469 |
xerr = xvid_decore(NULL, XVID_DEC_CREATE, &xparam, NULL); |
xerr = xvid_decore(NULL, XVID_DEC_CREATE, &xparam, NULL); |
1470 |
if (xerr==XVID_ERR_FAIL) { |
if (xerr==XVID_ERR_FAIL) { |
1471 |
printf("can't init decoder (err=%d)\n", xerr); |
printf("ERROR: can't init decoder (err=%d)\n", xerr); |
1472 |
return; |
return; |
1473 |
} |
} |
1474 |
dechandle = xparam.handle; |
dechandle = xparam.handle; |
1476 |
|
|
1477 |
f = fopen(name, "rb"); |
f = fopen(name, "rb"); |
1478 |
if (f==0) { |
if (f==0) { |
1479 |
printf( "can't open file '%s'\n", name); |
printf( "ERROR: can't open file '%s'\n", name); |
1480 |
return; |
return; |
1481 |
} |
} |
1482 |
fseek(f, 0, SEEK_END); |
fseek(f, 0, SEEK_END); |
1483 |
buf_size = ftell(f); |
buf_size = ftell(f); |
1484 |
fseek(f, 0, SEEK_SET); |
fseek(f, 0, SEEK_SET); |
1485 |
if (buf_size<=0) { |
if (buf_size<=0) { |
1486 |
printf("error while stating file\n"); |
printf("ERROR: error while stating file\n"); |
1487 |
goto End; |
goto End; |
1488 |
} |
} |
|
else printf( "Input size: %d\n", buf_size); |
|
1489 |
|
|
1490 |
buf = malloc(buf_size); /* should be enuf' */ |
buf = malloc(buf_size); |
1491 |
rgb_out = calloc(4, width*height); /* <-room for _RGB24 */ |
yuv_out = calloc(1, bps*height*3/2 + 15); |
1492 |
if (buf==0 || rgb_out==0) { |
if (buf==0 || yuv_out==0) { |
1493 |
printf( "malloc failed!\n" ); |
printf( "ERROR: malloc failed!\n" ); |
1494 |
goto End; |
goto End; |
1495 |
} |
} |
1496 |
|
|
1497 |
if (fread(buf, buf_size, 1, f)!=1) { |
if (fread(buf, buf_size, 1, f)!=1) { |
1498 |
printf( "file-read failed\n" ); |
printf( "ERROR: file-read failed\n" ); |
1499 |
goto End; |
goto End; |
1500 |
} |
} |
1501 |
|
|
1503 |
pos = 0; |
pos = 0; |
1504 |
t = -gettime_usec(); |
t = -gettime_usec(); |
1505 |
while(1) { |
while(1) { |
1506 |
|
int y; |
1507 |
|
|
1508 |
memset(&xframe, 0, sizeof(xframe)); |
memset(&xframe, 0, sizeof(xframe)); |
1509 |
xframe.version = XVID_VERSION; |
xframe.version = XVID_VERSION; |
1510 |
xframe.bitstream = buf + pos; |
xframe.bitstream = buf + pos; |
1511 |
xframe.length = buf_size - pos; |
xframe.length = buf_size - pos; |
1512 |
xframe.output.plane[0] = rgb_out; |
xframe.output.plane[0] = (uint8_t*)(((size_t)yuv_out + 15) & ~15); |
1513 |
xframe.output.stride[0] = width; |
xframe.output.plane[1] = (uint8_t*)xframe.output.plane[0] + bps*height; |
1514 |
xframe.output.csp = XVID_CSP_BGR; |
xframe.output.plane[2] = (uint8_t*)xframe.output.plane[1] + bps/2; |
1515 |
|
xframe.output.stride[0] = bps; |
1516 |
|
xframe.output.stride[1] = bps; |
1517 |
|
xframe.output.stride[2] = bps; |
1518 |
|
xframe.output.csp = XVID_CSP_I420; |
1519 |
xerr = xvid_decore(dechandle, XVID_DEC_DECODE, &xframe, 0); |
xerr = xvid_decore(dechandle, XVID_DEC_DECODE, &xframe, 0); |
1520 |
|
if (xerr<0) { |
1521 |
|
printf("ERROR: decoding failed for frame #%d (err=%d)!\n", nb, xerr); |
1522 |
|
break; |
1523 |
|
} |
1524 |
|
else if (xerr==0) |
1525 |
|
break; |
1526 |
|
else if (verbose>0) printf("#%d %d\n", nb, xerr ); |
1527 |
|
|
1528 |
|
pos += xerr; |
1529 |
nb++; |
nb++; |
1530 |
pos += xframe.length; |
|
1531 |
if (with_chksum) { |
for(y=0; y<height/2; ++y) { |
1532 |
int k = width*height; |
chksum = calc_crc((uint8_t*)xframe.output.plane[0] + (2*y+0)*bps, width, chksum); |
1533 |
uint32_t *ptr = (uint32_t *)rgb_out; |
chksum = calc_crc((uint8_t*)xframe.output.plane[0] + (2*y+1)*bps, width, chksum); |
1534 |
while(k-->0) chksum += *ptr++; |
chksum = calc_crc((uint8_t*)xframe.output.plane[1] + y*bps, width/2, chksum); |
1535 |
|
chksum = calc_crc((uint8_t*)xframe.output.plane[2] + y*bps, width/2, chksum); |
1536 |
} |
} |
1537 |
if (pos==buf_size) |
if (pos==buf_size) |
1538 |
break; |
break; |
|
if (xerr==XVID_ERR_FAIL) { |
|
|
printf("decoding failed for frame #%d (err=%d)!\n", nb, xerr); |
|
|
break; |
|
|
} |
|
1539 |
} |
} |
1540 |
t += gettime_usec(); |
t += gettime_usec(); |
1541 |
|
if (ref_chksum==0) { |
1542 |
if (t>0.) |
if (t>0.) |
1543 |
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 ); |
1544 |
if (with_chksum) |
} |
1545 |
printf("checksum: 0x%.8x\n", chksum); |
else { |
1546 |
|
printf("FPS:%.1f Checksum: 0x%.8x Expected:0x%.8x | %s\n", |
1547 |
|
t>0. ? (float)(nb*1.e6f/t) : 0.f, chksum, ref_chksum, (chksum==ref_chksum) ? "OK" : "ERROR"); |
1548 |
|
} |
1549 |
|
|
1550 |
End: |
End: |
1551 |
if (rgb_out!=0) free(rgb_out); |
if (yuv_out!=0) free(yuv_out); |
1552 |
if (buf!=0) free(buf); |
if (buf!=0) free(buf); |
1553 |
if (dechandle!=0) { |
if (dechandle!=0) { |
1554 |
xerr= xvid_decore(dechandle, XVID_DEC_DESTROY, NULL, NULL); |
xerr= xvid_decore(dechandle, XVID_DEC_DESTROY, NULL, NULL); |
1555 |
if (xerr==XVID_ERR_FAIL) |
if (xerr==XVID_ERR_FAIL) |
1556 |
printf("destroy-decoder failed (err=%d)!\n", xerr); |
printf("ERROR: destroy-decoder failed (err=%d)!\n", xerr); |
1557 |
} |
} |
1558 |
if (f!=0) fclose(f); |
if (f!=0) fclose(f); |
1559 |
} |
} |
1717 |
} |
} |
1718 |
|
|
1719 |
/********************************************************************* |
/********************************************************************* |
1720 |
|
* test some YUV func |
1721 |
|
*********************************************************************/ |
1722 |
|
|
1723 |
|
#define ENTER \ |
1724 |
|
for(i=0; i<(int)sizeof(Dst0); ++i) Dst0[0][i] = 0; \ |
1725 |
|
t = gettime_usec(); \ |
1726 |
|
emms(); |
1727 |
|
|
1728 |
|
#define LEAVE \ |
1729 |
|
emms(); \ |
1730 |
|
t = (gettime_usec() - t) / nb_tests; \ |
1731 |
|
iCrc = calc_crc((uint8_t*)Dst0, sizeof(Dst0), CRC32_INITIAL) |
1732 |
|
|
1733 |
|
#define TEST_YUYV(FUNC, S, FLIP) \ |
1734 |
|
ENTER \ |
1735 |
|
for(tst=0; tst<nb_tests; ++tst) (FUNC)(Dst0[0], S*WIDTH, Src0[0], Src0[1], Src0[2], WIDTH, WIDTH/2, WIDTH, HEIGHT, (FLIP)); \ |
1736 |
|
LEAVE |
1737 |
|
|
1738 |
|
static const int yuv_CRCs[6][2] = { |
1739 |
|
{0x0f4fb96b,0x780b6a68} |
1740 |
|
, {0xa986b289,0x65e49b76} |
1741 |
|
, {0x7f19c152,0xd539b86e} |
1742 |
|
, {0x0f4fb96b,0x780b6a68} |
1743 |
|
, {0xa986b289,0x65e49b76} |
1744 |
|
, {0x36ab8b57,0x1cd92fee} |
1745 |
|
}; |
1746 |
|
|
1747 |
|
#define WIDTH 128 |
1748 |
|
#define HEIGHT 32 |
1749 |
|
void test_yuv() |
1750 |
|
{ |
1751 |
|
const int nb_tests = 200*speed_ref; |
1752 |
|
CPU *cpu; |
1753 |
|
uint8_t Src0[3][WIDTH*HEIGHT]; |
1754 |
|
uint8_t Dst0[4][WIDTH*HEIGHT]; |
1755 |
|
int i, j, with_flip; |
1756 |
|
double t; |
1757 |
|
int tst, iCrc; |
1758 |
|
|
1759 |
|
colorspace_init(); |
1760 |
|
ieee_reseed(1); |
1761 |
|
for(i=0; i<(int)sizeof(Src0); ++i) Src0[0][i] = ieee_rand(0,255); |
1762 |
|
for(i=0; i<(int)sizeof(Dst0); ++i) Dst0[0][i] = 0x5a; |
1763 |
|
|
1764 |
|
printf( "\n === test YUV ===\n" ); |
1765 |
|
|
1766 |
|
for(with_flip=0; with_flip<=1; ++with_flip) { |
1767 |
|
|
1768 |
|
init_cpu(&cpu_list[0]); |
1769 |
|
TEST_YUYV(yv12_to_yuyv_c, 4, with_flip); |
1770 |
|
printf(" yv12_to_yuyv_c %.3f usec crc32=0x%08x %s\n", |
1771 |
|
t, iCrc, (iCrc!=yuv_CRCs[0][with_flip])?"| ERROR": "" ); |
1772 |
|
TEST_YUYV(yv12_to_uyvy_c, 4, with_flip); |
1773 |
|
printf(" yv12_to_uyvy_c %.3f usec crc32=0x%08x %s\n", |
1774 |
|
t, iCrc, (iCrc!=yuv_CRCs[1][with_flip])?"| ERROR": "" ); |
1775 |
|
|
1776 |
|
TEST_YUYV(yv12_to_bgra_c, 4, with_flip); |
1777 |
|
printf(" yv12_to_bgra_c %.3f usec crc32=0x%08x %s\n", |
1778 |
|
t, iCrc, (iCrc!=yuv_CRCs[2][with_flip])?"| ERROR": "" ); |
1779 |
|
|
1780 |
|
#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) |
1781 |
|
init_cpu(&cpu_list[1]); |
1782 |
|
TEST_YUYV(yv12_to_yuyv_mmx, 4, with_flip); |
1783 |
|
printf(" yv12_to_yuyv_mmx %.3f usec crc32=0x%08x %s\n", |
1784 |
|
t, iCrc, (iCrc!=yuv_CRCs[3][with_flip])?"| ERROR": "" ); |
1785 |
|
|
1786 |
|
TEST_YUYV(yv12_to_uyvy_mmx, 4, with_flip); |
1787 |
|
printf(" yv12_to_uyvy_mmx %.3f usec crc32=0x%08x %s\n", |
1788 |
|
t, iCrc, (iCrc!=yuv_CRCs[4][with_flip])?"| ERROR": "" ); |
1789 |
|
|
1790 |
|
TEST_YUYV(yv12_to_bgra_mmx, 4, with_flip); |
1791 |
|
printf(" yv12_to_bgra_mmx %.3f usec crc32=0x%08x %s\n", |
1792 |
|
t, iCrc, (iCrc!=yuv_CRCs[5][with_flip])?"| ERROR": "" ); |
1793 |
|
|
1794 |
|
#endif |
1795 |
|
|
1796 |
|
#ifdef ARCH_IS_PPC |
1797 |
|
init_cpu(&cpu_list[1]); |
1798 |
|
TEST_YUYV(yv12_to_yuyv_altivec_c, 4, with_flip); |
1799 |
|
printf(" yv12_to_yuyv_altivec_c %.3f usec crc32=0x%08x %s\n", |
1800 |
|
t, iCrc, (iCrc!=yuv_CRCs[3][with_flip])?"| ERROR": "" ); |
1801 |
|
|
1802 |
|
TEST_YUYV(yv12_to_uyvy_altivec_c, 4, with_flip); |
1803 |
|
printf(" yv12_to_uyvy_altivec_c %.3f usec crc32=0x%08x %s\n", |
1804 |
|
t, iCrc, (iCrc!=yuv_CRCs[4][with_flip])?"| ERROR": "" ); |
1805 |
|
|
1806 |
|
TEST_YUYV(yv12_to_bgra_altivec_c, 4, with_flip); |
1807 |
|
printf(" yv12_to_bgra_altivec_c %.3f usec crc32=0x%08x %s\n", |
1808 |
|
t, iCrc, (iCrc!=yuv_CRCs[5][with_flip])?"| ERROR": "" ); |
1809 |
|
#endif |
1810 |
|
} |
1811 |
|
printf( " --- \n" ); |
1812 |
|
} |
1813 |
|
|
1814 |
|
#define TEST_YV2(FUNC, WITH_UV, WITH_FLIP) \ |
1815 |
|
ENTER \ |
1816 |
|
for(tst=0; tst<nb_tests; ++tst) (FUNC)(Dst0[0], Dst0[1], Dst0[2], WIDTH, WIDTH, \ |
1817 |
|
Src0[0], (WITH_UV) ? Src0[1] : 0, (WITH_UV) ? Src0[2] : 0, WIDTH, WIDTH, \ |
1818 |
|
WIDTH-2, HEIGHT-2, WITH_FLIP); \ |
1819 |
|
LEAVE |
1820 |
|
|
1821 |
|
#define PRINT_NxN(DATA,W,H,STR) { \ |
1822 |
|
int i,j; \ |
1823 |
|
for(j=0; j<(H); ++j) { \ |
1824 |
|
for(i=0; i<(W); ++i) printf( "0x%.2x ", (DATA)[i+j*(STR)] );\ |
1825 |
|
printf("\n"); \ |
1826 |
|
} \ |
1827 |
|
printf("---\n"); \ |
1828 |
|
} |
1829 |
|
|
1830 |
|
static const int yv12_CRCs[2][2] = { |
1831 |
|
{0x5cab7cf0,0xdab46541} |
1832 |
|
, {0xe8bae865,0x1faf77b7} |
1833 |
|
}; |
1834 |
|
|
1835 |
|
void test_yuv2() |
1836 |
|
{ |
1837 |
|
const int nb_tests = 800*speed_ref; |
1838 |
|
CPU *cpu; |
1839 |
|
uint8_t Src0[3][WIDTH*HEIGHT]; |
1840 |
|
uint8_t Dst0[3][WIDTH*HEIGHT]; |
1841 |
|
int with_uv, with_flip; |
1842 |
|
int i, j; |
1843 |
|
double t; |
1844 |
|
int tst, iCrc; |
1845 |
|
|
1846 |
|
colorspace_init(); |
1847 |
|
ieee_reseed(1); |
1848 |
|
for(i=0; i<(int)sizeof(Src0); ++i) Src0[0][i] = ieee_rand(0,255); |
1849 |
|
|
1850 |
|
printf( "\n === test YV2 ===\n" ); |
1851 |
|
for(with_flip=0; with_flip<=1; ++with_flip) { |
1852 |
|
for(with_uv=0; with_uv<=1; ++with_uv) { |
1853 |
|
init_cpu(&cpu_list[0]); |
1854 |
|
TEST_YV2(yv12_to_yv12_c, with_uv, with_flip); |
1855 |
|
printf(" yv12_to_yv12_c %.3f usec \tcrc32=0x%08x %s\n", |
1856 |
|
t, iCrc, (iCrc!=yv12_CRCs[with_flip][with_uv])?"| ERROR": "" ); |
1857 |
|
/* if (!with_uv) PRINT_NxN(Dst0[1], WIDTH/2, HEIGHT/2, WIDTH ); */ |
1858 |
|
|
1859 |
|
#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) |
1860 |
|
init_cpu(&cpu_list[1]); |
1861 |
|
TEST_YV2(yv12_to_yv12_mmx, with_uv, with_flip); |
1862 |
|
printf(" yv12_to_yv12_mmx %.3f usec \tcrc32=0x%08x %s\n", |
1863 |
|
t, iCrc, (iCrc!=yv12_CRCs[with_flip][with_uv])?"| ERROR": "" ); |
1864 |
|
/* if (!with_uv) PRINT_NxN(Dst0[1], WIDTH/2, HEIGHT/2, WIDTH ); */ |
1865 |
|
|
1866 |
|
TEST_YV2(yv12_to_yv12_xmm, with_uv, with_flip); |
1867 |
|
printf(" yv12_to_yv12_xmm %.3f usec \tcrc32=0x%08x %s\n", |
1868 |
|
t, iCrc, (iCrc!=yv12_CRCs[with_flip][with_uv])?"| ERROR": "" ); |
1869 |
|
#endif |
1870 |
|
} |
1871 |
|
|
1872 |
|
printf( " --- \n" ); |
1873 |
|
} |
1874 |
|
printf( " ===== \n" ); |
1875 |
|
} |
1876 |
|
|
1877 |
|
#undef WIDTH |
1878 |
|
#undef HEIGHT |
1879 |
|
#undef ENTER |
1880 |
|
#undef LEAVE |
1881 |
|
|
1882 |
|
/*********************************************************************/ |
1883 |
|
|
1884 |
|
static uint32_t __inline log2bin_v1(uint32_t value) |
1885 |
|
{ |
1886 |
|
int n = 0; |
1887 |
|
while (value) { |
1888 |
|
value >>= 1; |
1889 |
|
n++; |
1890 |
|
} |
1891 |
|
return n; |
1892 |
|
} |
1893 |
|
|
1894 |
|
static const uint8_t log2_tab_16[16] = { 0, 1, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4 }; |
1895 |
|
|
1896 |
|
static uint32_t __inline log2bin_v2(uint32_t value) |
1897 |
|
{ |
1898 |
|
int n = 0; |
1899 |
|
if (value & 0xffff0000) { |
1900 |
|
value >>= 16; |
1901 |
|
n += 16; |
1902 |
|
} |
1903 |
|
if (value & 0xff00) { |
1904 |
|
value >>= 8; |
1905 |
|
n += 8; |
1906 |
|
} |
1907 |
|
if (value & 0xf0) { |
1908 |
|
value >>= 4; |
1909 |
|
n += 4; |
1910 |
|
} |
1911 |
|
return n + log2_tab_16[value]; |
1912 |
|
} |
1913 |
|
|
1914 |
|
void test_log2bin() |
1915 |
|
{ |
1916 |
|
const int nb_tests = 3000*speed_ref; |
1917 |
|
int n, crc1=0, crc2=0; |
1918 |
|
uint32_t s, s0; |
1919 |
|
double t1, t2; |
1920 |
|
|
1921 |
|
t1 = gettime_usec(); |
1922 |
|
s0 = (int)(t1*31.241); |
1923 |
|
for(s=s0, n=0; n<nb_tests; ++n, s=(s*12363+31)&0x7fffffff) |
1924 |
|
crc1 += log2bin_v1(s); |
1925 |
|
t1 = (gettime_usec()-t1) / nb_tests; |
1926 |
|
|
1927 |
|
t2 = gettime_usec(); |
1928 |
|
for(s=s0, n=0; n<nb_tests; ++n, s=(s*12363+31)&0x7fffffff) |
1929 |
|
crc2 += log2bin_v2(s); |
1930 |
|
t2 = (gettime_usec() - t2) / nb_tests; |
1931 |
|
|
1932 |
|
printf( "log2bin_v1: %.3f sec crc=%d\n", t1, crc1 ); |
1933 |
|
printf( "log2bin_v2: %.3f sec crc=%d\n", t2, crc2 ); |
1934 |
|
if (crc1!=crc2) printf( " CRC ERROR !\n" ); |
1935 |
|
} |
1936 |
|
|
1937 |
|
/*********************************************************************/ |
1938 |
|
|
1939 |
|
static void __inline old_gcd(int *num, int *den) |
1940 |
|
{ |
1941 |
|
int i = *num; |
1942 |
|
while (i > 1) { |
1943 |
|
if (*num % i == 0 && *den % i == 0) { |
1944 |
|
*num /= i; |
1945 |
|
*den /= i; |
1946 |
|
i = *num; |
1947 |
|
continue; |
1948 |
|
} |
1949 |
|
i--; |
1950 |
|
} |
1951 |
|
} |
1952 |
|
|
1953 |
|
static uint32_t gcd(int num, int den) |
1954 |
|
{ |
1955 |
|
int tmp; |
1956 |
|
while( (tmp=num%den) ) { num = den; den = tmp; } |
1957 |
|
return den; |
1958 |
|
} |
1959 |
|
static void __inline new_gcd(int *num, int *den) |
1960 |
|
{ |
1961 |
|
const int div = gcd(*num, *den); |
1962 |
|
if (num) { |
1963 |
|
*num /= div; |
1964 |
|
*den /= div; |
1965 |
|
} |
1966 |
|
} |
1967 |
|
|
1968 |
|
void test_gcd() |
1969 |
|
{ |
1970 |
|
const int nb_tests = 10*speed_ref; |
1971 |
|
int i; |
1972 |
|
uint32_t crc1=0, crc2=0; |
1973 |
|
uint32_t n0, n, d0, d; |
1974 |
|
double t1, t2; |
1975 |
|
|
1976 |
|
t1 = gettime_usec(); |
1977 |
|
n0 = 0xfffff & (int)(t1*31.241); |
1978 |
|
d0 = 0xfffff & (int)( ((n0*4123)%17) | 1 ); |
1979 |
|
for(n=n0, d=d0, i=0; i<nb_tests; ++i) { |
1980 |
|
old_gcd(&n, &d); |
1981 |
|
crc1 = (((crc1>>4)^d) + ((crc1<<2)^n) ) & 0xffffff; |
1982 |
|
n = d; |
1983 |
|
d = (d*12363+31) & 0xffff; |
1984 |
|
d |= !d; |
1985 |
|
} |
1986 |
|
t1 = (gettime_usec()-t1) / nb_tests; |
1987 |
|
|
1988 |
|
t2 = gettime_usec(); |
1989 |
|
for(n=n0, d=d0, i=0; i<nb_tests; ++i) { |
1990 |
|
new_gcd(&n, &d); |
1991 |
|
crc2 = (((crc2>>4)^d) + ((crc2<<2)^n) ) & 0xffffff; |
1992 |
|
n = d; |
1993 |
|
d = (d*12363+31) & 0xffff; |
1994 |
|
d |= !d; |
1995 |
|
} |
1996 |
|
t2 = (gettime_usec() - t2) / nb_tests; |
1997 |
|
|
1998 |
|
printf( "old_gcd: %.3f sec crc=%d\n", t1, crc1 ); |
1999 |
|
printf( "new_gcd: %.3f sec crc=%d\n", t2, crc2 ); |
2000 |
|
if (crc1!=crc2) printf( " CRC ERROR !\n" ); |
2001 |
|
} |
2002 |
|
|
2003 |
|
/********************************************************************* |
2004 |
|
* test compiler |
2005 |
|
*********************************************************************/ |
2006 |
|
|
2007 |
|
void test_compiler() { |
2008 |
|
int nb_err = 0; |
2009 |
|
int32_t v; |
2010 |
|
if (sizeof(uint16_t)<2) { |
2011 |
|
printf( "ERROR: sizeof(uint16_t)<2 !!\n" ); |
2012 |
|
nb_err++; |
2013 |
|
} |
2014 |
|
if (sizeof(int16_t)<2) { |
2015 |
|
printf( "ERROR: sizeof(int16_t)<2 !!\n" ); |
2016 |
|
nb_err++; |
2017 |
|
} |
2018 |
|
if (sizeof(uint8_t)!=1) { |
2019 |
|
printf( "ERROR: sizeof(uint8_t)!=1 !!\n" ); |
2020 |
|
nb_err++; |
2021 |
|
} |
2022 |
|
if (sizeof(int8_t)!=1) { |
2023 |
|
printf( "ERROR: sizeof(int8_t)!=1 !!\n" ); |
2024 |
|
nb_err++; |
2025 |
|
} |
2026 |
|
if (sizeof(uint32_t)<4) { |
2027 |
|
printf( "ERROR: sizeof(uint32_t)<4 !!\n" ); |
2028 |
|
nb_err++; |
2029 |
|
} |
2030 |
|
if (sizeof(int32_t)<4) { |
2031 |
|
printf( "ERROR: sizeof(int32_t)<4 !!\n" ); |
2032 |
|
nb_err++; |
2033 |
|
} |
2034 |
|
/* yes, i know, this test is silly. But better be safe than sorry. :) */ |
2035 |
|
for(v=1000; v>=0; v--) { |
2036 |
|
if ( (v>>2) != v/4) |
2037 |
|
nb_err++; |
2038 |
|
} |
2039 |
|
for(v=-1000; v!=-1; v++) { |
2040 |
|
if ( (v>>2) != (v/4)-!!(v%4)) |
2041 |
|
nb_err++; |
2042 |
|
} |
2043 |
|
if (nb_err!=0) { |
2044 |
|
printf( "ERROR! please post your platform/compiler specs to xvid-devel@xvid.org !\n" ); |
2045 |
|
} |
2046 |
|
} |
2047 |
|
|
2048 |
|
/********************************************************************* |
2049 |
|
* test SSIM functions |
2050 |
|
*********************************************************************/ |
2051 |
|
|
2052 |
|
typedef int (*lumfunc)(uint8_t* ptr, int stride); |
2053 |
|
typedef void (*csfunc)(uint8_t* ptro, uint8_t* ptrc, int stride, int lumo, int lumc, int* pdevo, int* pdevc, int* pcorr); |
2054 |
|
|
2055 |
|
extern int lum_8x8_c(uint8_t* ptr, int stride); |
2056 |
|
extern int lum_8x8_mmx(uint8_t* ptr, int stride); |
2057 |
|
extern int lum_2x8_c(uint8_t* ptr, int stride); |
2058 |
|
extern void consim_c(uint8_t* ptro, uint8_t* ptrc, int stride, int lumo, int lumc, int* pdevo, int* pdevc, int* pcorr); |
2059 |
|
extern void consim_mmx(uint8_t* ptro, uint8_t* ptrc, int stride, int lumo, int lumc, int* pdevo, int* pdevc, int* pcorr); |
2060 |
|
extern void consim_sse2(uint8_t* ptro, uint8_t* ptrc, int stride, int lumo, int lumc, int* pdevo, int* pdevc, int* pcorr); |
2061 |
|
|
2062 |
|
void test_SSIM() |
2063 |
|
{ |
2064 |
|
const int nb_tests = 3000*speed_ref; |
2065 |
|
int tst; |
2066 |
|
CPU *cpu; |
2067 |
|
int i; |
2068 |
|
int devs[3]; |
2069 |
|
long lumo, lumc; |
2070 |
|
DECLARE_ALIGNED_MATRIX(Ref1, 16, 16, uint8_t, 16); |
2071 |
|
DECLARE_ALIGNED_MATRIX(Ref2, 16, 16, uint8_t, 16); |
2072 |
|
lumfunc lum8x8; |
2073 |
|
lumfunc lum2x8; |
2074 |
|
csfunc csim; |
2075 |
|
|
2076 |
|
ieee_reseed(1); |
2077 |
|
printf( "\n ====== test SSIM ======\n" ); |
2078 |
|
for(i=0; i<16*16;++i) { |
2079 |
|
long v1, v2; |
2080 |
|
v1 = ieee_rand(-256, 511); |
2081 |
|
v2 = ieee_rand(-256, 511); |
2082 |
|
Ref1[i] = (v1<0) ? 0 : (v1>255) ? 255 : v1; |
2083 |
|
Ref2[i] = (v2<0) ? 0 : (v2>255) ? 255 : v2; |
2084 |
|
} |
2085 |
|
lumc = ieee_rand(0, 255); |
2086 |
|
lumo = ieee_rand(0, 255); |
2087 |
|
|
2088 |
|
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
2089 |
|
{ |
2090 |
|
double t; |
2091 |
|
int m; |
2092 |
|
if (!init_cpu(cpu)) |
2093 |
|
continue; |
2094 |
|
lum8x8 = lum_8x8_c; |
2095 |
|
lum2x8 = lum_2x8_c; |
2096 |
|
csim = consim_c; |
2097 |
|
#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) |
2098 |
|
if (cpu->cpu & XVID_CPU_MMX){ |
2099 |
|
lum8x8 = lum_8x8_mmx; |
2100 |
|
csim = consim_mmx; |
2101 |
|
} |
2102 |
|
if (cpu->cpu & XVID_CPU_MMX){ |
2103 |
|
csim = consim_sse2; |
2104 |
|
} |
2105 |
|
#endif |
2106 |
|
t = gettime_usec(); |
2107 |
|
emms(); |
2108 |
|
for(tst=0; tst<nb_tests; ++tst) m = lum8x8(Ref1, 16); |
2109 |
|
emms(); |
2110 |
|
t = (gettime_usec() - t) / nb_tests; |
2111 |
|
printf("%s - ssim-lum8x8 %.3f usec m=%d %s\n", |
2112 |
|
cpu->name, t, m, |
2113 |
|
(m!=8230)?"| ERROR": "" ); |
2114 |
|
|
2115 |
|
t = gettime_usec(); |
2116 |
|
emms(); |
2117 |
|
for(tst=0; tst<nb_tests; ++tst) m = lum2x8(Ref1+8, 16); |
2118 |
|
emms(); |
2119 |
|
t = (gettime_usec() - t) / nb_tests; |
2120 |
|
printf("%s - ssim-lum2x8 %.3f usec m=%d %s\n", |
2121 |
|
cpu->name, t, m, |
2122 |
|
(m!=681)?"| ERROR": "" ); |
2123 |
|
|
2124 |
|
t = gettime_usec(); |
2125 |
|
emms(); |
2126 |
|
for(tst=0; tst<nb_tests; ++tst) csim(Ref1, Ref2, 16, lumo, lumc, devs+0, devs+1, devs+2); |
2127 |
|
emms(); |
2128 |
|
t = (gettime_usec() - t) / nb_tests; |
2129 |
|
printf("%s - ssim-consim %.3f usec devs=[0x%x 0x%x 0x%x] %s\n", |
2130 |
|
cpu->name, t, devs[0], devs[1], devs[2], |
2131 |
|
(devs[0]!=0x1bdf0f || devs[1]!=0x137258 || devs[2]!=0xcdb13)?"| ERROR": "" ); |
2132 |
|
printf( " --- \n" ); |
2133 |
|
} |
2134 |
|
} |
2135 |
|
|
2136 |
|
/********************************************************************* |
2137 |
|
* test bitstream functions |
2138 |
|
*********************************************************************/ |
2139 |
|
|
2140 |
|
#define BIT_BUF_SIZE 2000 |
2141 |
|
|
2142 |
|
static void test_bits() |
2143 |
|
{ |
2144 |
|
const int nb_tests = 50*speed_ref; |
2145 |
|
int tst; |
2146 |
|
uint32_t Crc; |
2147 |
|
uint8_t Buf[BIT_BUF_SIZE]; |
2148 |
|
uint32_t Extracted[BIT_BUF_SIZE*8]; /* worst case: bits read 1 by 1 */ |
2149 |
|
int Lens[BIT_BUF_SIZE*8]; |
2150 |
|
double t1; |
2151 |
|
|
2152 |
|
|
2153 |
|
printf( "\n === test bitstream ===\n" ); |
2154 |
|
ieee_reseed(1); |
2155 |
|
Crc = 0; |
2156 |
|
|
2157 |
|
t1 = gettime_usec(); |
2158 |
|
for(tst=0; tst<nb_tests; ++tst) { |
2159 |
|
Bitstream bs; |
2160 |
|
int m, m2, l, l2; |
2161 |
|
|
2162 |
|
for(l=0; l<BIT_BUF_SIZE; ++l) |
2163 |
|
Buf[l] = (uint8_t)ieee_rand(0,255); |
2164 |
|
|
2165 |
|
l = BIT_BUF_SIZE - ieee_rand(1,BIT_BUF_SIZE/10); |
2166 |
|
BitstreamInit(&bs, (void*)(Buf+BIT_BUF_SIZE-l), l); |
2167 |
|
|
2168 |
|
|
2169 |
|
BitstreamReset(&bs); |
2170 |
|
for(l2=l*8, m=0; l2>0; m++) { |
2171 |
|
const int b = ieee_rand(1,32); |
2172 |
|
Lens[m] = b; |
2173 |
|
l2 -= b; |
2174 |
|
if (l2<0) break; |
2175 |
|
Extracted[m] = BitstreamShowBits(&bs, b); |
2176 |
|
BitstreamSkip(&bs, b); |
2177 |
|
// printf( "<= %d: %d 0x%x\n", m, b, Extracted[m]); |
2178 |
|
} |
2179 |
|
|
2180 |
|
BitstreamReset(&bs); |
2181 |
|
for(m2=0; m2<m; ++m2) { |
2182 |
|
const int b = Lens[m2]; |
2183 |
|
const uint32_t v = BitstreamGetBits(&bs, b); |
2184 |
|
Crc |= (v!=Extracted[m2]); |
2185 |
|
// printf( "=> %d: %d 0x%x %c\n", m2, b, v, " *"[Crc]); |
2186 |
|
} |
2187 |
|
} |
2188 |
|
t1 = (gettime_usec() - t1) / nb_tests; |
2189 |
|
printf(" test_bits %.3f usec %s\n", t1, (Crc!=0)?"| ERROR": "" ); |
2190 |
|
} |
2191 |
|
|
2192 |
|
/********************************************************************* |
2193 |
* main |
* main |
2194 |
*********************************************************************/ |
*********************************************************************/ |
2195 |
|
|
2196 |
int main(int argc, char *argv[]) |
static void arg_missing(const char *opt) |
2197 |
|
{ |
2198 |
|
printf( "missing argument after option '%s'\n", opt); |
2199 |
|
exit(-1); |
2200 |
|
} |
2201 |
|
|
2202 |
|
int main(int argc, const char *argv[]) |
2203 |
{ |
{ |
2204 |
int what = 0; |
int c, what = 0; |
2205 |
if (argc>1) what = atoi(argv[1]); |
int width, height; |
2206 |
|
uint32_t chksum = 0; |
2207 |
|
const char * test_bitstream = 0; |
2208 |
|
|
2209 |
|
cpu_mask = 0; // default => will use autodectect |
2210 |
|
for(c=1; c<argc; ++c) |
2211 |
|
{ |
2212 |
|
if (!strcmp(argv[c], "-v")) verbose++; |
2213 |
|
else if (!strcmp(argv[c], "-c")) cpu_mask = 0 /* PLAIN_C */ | XVID_CPU_FORCE; |
2214 |
|
else if (!strcmp(argv[c], "-mmx")) cpu_mask = XVID_CPU_MMX | XVID_CPU_FORCE; |
2215 |
|
else if (!strcmp(argv[c], "-mmxext")) cpu_mask = XVID_CPU_MMXEXT | XVID_CPU_MMX | XVID_CPU_FORCE; |
2216 |
|
else if (!strcmp(argv[c], "-sse2")) cpu_mask = XVID_CPU_SSE2 | XVID_CPU_MMXEXT | XVID_CPU_MMX | XVID_CPU_FORCE; |
2217 |
|
else if (!strcmp(argv[c], "-sse3")) cpu_mask = XVID_CPU_SSE3 | XVID_CPU_SSE2 | XVID_CPU_MMXEXT | XVID_CPU_MMX | XVID_CPU_FORCE; |
2218 |
|
else if (!strcmp(argv[c], "-sse4")) cpu_mask = XVID_CPU_SSE41 | XVID_CPU_SSE3 | XVID_CPU_SSE2 | XVID_CPU_MMXEXT | XVID_CPU_MMX | XVID_CPU_FORCE; |
2219 |
|
else if (!strcmp(argv[c], "-3dnow")) cpu_mask = XVID_CPU_3DNOW | XVID_CPU_FORCE; |
2220 |
|
else if (!strcmp(argv[c], "-3dnowe")) cpu_mask = XVID_CPU_3DNOW | XVID_CPU_3DNOWEXT | XVID_CPU_FORCE; |
2221 |
|
else if (!strcmp(argv[c], "-altivec")) cpu_mask = XVID_CPU_ALTIVEC | XVID_CPU_FORCE; |
2222 |
|
else if (!strcmp(argv[c], "-spd")) { |
2223 |
|
if (++c==argc) arg_missing( argv[argc-1] ); |
2224 |
|
speed_ref = atoi(argv[c]); |
2225 |
|
} |
2226 |
|
else if (argv[c][0]!='-') { |
2227 |
|
what = atoi(argv[c]); |
2228 |
|
if (what==9) { |
2229 |
|
if (c+4>argc) { |
2230 |
|
printf("usage: %s %d bitstream width height (checksum)\n", argv[0], what); |
2231 |
|
exit(-1); |
2232 |
|
} |
2233 |
|
test_bitstream = argv[++c]; |
2234 |
|
width = atoi(argv[++c]); |
2235 |
|
height = atoi(argv[++c]); |
2236 |
|
if (c+1<argc && argv[c+1][0]!='-') { |
2237 |
|
if (sscanf(argv[c+1], "0x%x", &chksum)!=1) { |
2238 |
|
printf( "can't read checksum value.\n" ); |
2239 |
|
exit(-1); |
2240 |
|
} |
2241 |
|
else c++; |
2242 |
|
} |
2243 |
|
// printf( "[%s] %dx%d (0x%.8x)\n", test_bitstream, width, height, chksum); |
2244 |
|
} |
2245 |
|
} |
2246 |
|
else { |
2247 |
|
printf( "unrecognized option '%s'\n", argv[c]); |
2248 |
|
exit(-1); |
2249 |
|
} |
2250 |
|
} |
2251 |
|
|
2252 |
|
|
2253 |
if (what==0 || what==1) test_dct(); |
if (what==0 || what==1) test_dct(); |
2254 |
if (what==0 || what==2) test_mb(); |
if (what==0 || what==2) test_mb(); |
2255 |
if (what==0 || what==3) test_sad(); |
if (what==0 || what==3) test_sad(); |
2256 |
if (what==0 || what==4) test_transfer(); |
if (what==0 || what==4) test_transfer(); |
2257 |
if (what==0 || what==5) test_quant(); |
if (what==0 || what==5) test_quant(); |
2258 |
if (what==0 || what==6) test_cbp(); |
if (what==0 || what==6) test_cbp(); |
2259 |
|
if (what==0 || what==10) test_sse(); |
2260 |
|
if (what==0 || what==11) test_log2bin(); |
2261 |
|
if (what==0 || what==12) test_gcd(); |
2262 |
|
if (what==0 || what==13) test_compiler(); |
2263 |
|
if (what==0 || what==14) test_yuv(); |
2264 |
|
if (what==0 || what==15) test_SSIM(); |
2265 |
|
if (what==0 || what==16) test_yuv2(); |
2266 |
|
if (what==0 || what==17) test_bits(); |
2267 |
|
|
2268 |
if (what==7) { |
if (what==7) { |
2269 |
test_IEEE1180_compliance(-256, 255, 1); |
test_IEEE1180_compliance(-256, 255, 1); |
2275 |
} |
} |
2276 |
if (what==8) test_dct_saturation(-256, 255); |
if (what==8) test_dct_saturation(-256, 255); |
2277 |
|
|
2278 |
if (what==9) { |
if (test_bitstream) |
2279 |
int width, height; |
test_dec(test_bitstream, width, height, chksum); |
|
if (argc<5) { |
|
|
printf("usage: %s %d [bitstream] [width] [height]\n", argv[0], what); |
|
|
return 1; |
|
|
} |
|
|
width = atoi(argv[3]); |
|
|
height = atoi(argv[4]); |
|
|
test_dec(argv[2], width, height, (argc>5)); |
|
|
} |
|
|
|
|
2280 |
if (what==-1) { |
if (what==-1) { |
2281 |
test_dct_precision_diffs(); |
test_dct_precision_diffs(); |
2282 |
test_bugs1(); |
test_bugs1(); |
2284 |
if (what==-2) |
if (what==-2) |
2285 |
test_quant_bug(); |
test_quant_bug(); |
2286 |
|
|
2287 |
if (what >= 0 && what <= 6) { |
if ((what >= 0 && what <= 6) || what == 10) { |
2288 |
printf("\n\n" |
printf("\n\n" |
2289 |
"NB: If a function isn't optimised for a specific set of intructions,\n" |
"NB: If a function isn't optimised for a specific set of intructions,\n" |
2290 |
" a C function is used instead. So don't panic if some functions\n" |
" a C function is used instead. So don't panic if some functions\n" |
2291 |
" may appear to be slow.\n"); |
" may appear to be slow.\n"); |
2292 |
} |
} |
2293 |
|
|
2294 |
#ifdef ARCH_IS_IA32 |
#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) |
2295 |
if (what == 0 || what == 5) { |
if (what == 0 || what == 5) { |
2296 |
printf("\n" |
printf("\n" |
2297 |
"NB: MMX mpeg4 quantization is known to have very small errors (+/-1 magnitude)\n" |
"NB: MMX mpeg4 quantization is known to have very small errors (+/-1 magnitude)\n" |
2304 |
return 0; |
return 0; |
2305 |
} |
} |
2306 |
|
|
2307 |
/********************************************************************* |
/*********************************************************************/ |
|
* 'Reference' output (except for timing) on an Athlon XP 2200+ |
|
|
*********************************************************************/ |
|
|
|
|
|
/* as of 2002-01-07, there's a problem with MMX mpeg4-quantization */ |
|
|
/* as of 2003-11-30, the problem is still here */ |
|
|
|
|
|
/********************************************************************* |
|
|
|
|
|
|
|
|
===== test fdct/idct ===== |
|
|
PLAINC - 2.867 usec PSNR=13.291 MSE=3.000 |
|
|
MMX - -0.211 usec PSNR=9.611 MSE=7.000 |
|
|
MMXEXT - -0.256 usec PSNR=9.611 MSE=7.000 |
|
|
3DNOW - 2.855 usec PSNR=13.291 MSE=3.000 |
|
|
3DNOWE - 1.429 usec PSNR=13.291 MSE=3.000 |
|
|
|
|
|
=== test block motion === |
|
|
PLAINC - interp- h-round0 0.538 usec crc32=0x115381ba |
|
|
PLAINC - round1 0.527 usec crc32=0x2b1f528f |
|
|
PLAINC - interp- v-round0 0.554 usec crc32=0x423cdcc7 |
|
|
PLAINC - round1 0.551 usec crc32=0x42202efe |
|
|
PLAINC - interp-hv-round0 1.041 usec crc32=0xd198d387 |
|
|
PLAINC - round1 1.038 usec crc32=0x9ecfd921 |
|
|
--- |
|
|
MMX - interp- h-round0 0.051 usec crc32=0x115381ba |
|
|
MMX - round1 0.053 usec crc32=0x2b1f528f |
|
|
MMX - interp- v-round0 0.048 usec crc32=0x423cdcc7 |
|
|
MMX - round1 0.048 usec crc32=0x42202efe |
|
|
MMX - interp-hv-round0 0.074 usec crc32=0xd198d387 |
|
|
MMX - round1 0.073 usec crc32=0x9ecfd921 |
|
|
--- |
|
|
MMXEXT - interp- h-round0 0.020 usec crc32=0x115381ba |
|
|
MMXEXT - round1 0.025 usec crc32=0x2b1f528f |
|
|
MMXEXT - interp- v-round0 0.016 usec crc32=0x423cdcc7 |
|
|
MMXEXT - round1 0.024 usec crc32=0x42202efe |
|
|
MMXEXT - interp-hv-round0 0.037 usec crc32=0xd198d387 |
|
|
MMXEXT - round1 0.037 usec crc32=0x9ecfd921 |
|
|
--- |
|
|
3DNOW - interp- h-round0 0.020 usec crc32=0x115381ba |
|
|
3DNOW - round1 0.029 usec crc32=0x2b1f528f |
|
|
3DNOW - interp- v-round0 0.016 usec crc32=0x423cdcc7 |
|
|
3DNOW - round1 0.024 usec crc32=0x42202efe |
|
|
3DNOW - interp-hv-round0 0.038 usec crc32=0xd198d387 |
|
|
3DNOW - round1 0.039 usec crc32=0x9ecfd921 |
|
|
--- |
|
|
3DNOWE - interp- h-round0 0.020 usec crc32=0x115381ba |
|
|
3DNOWE - round1 0.024 usec crc32=0x2b1f528f |
|
|
3DNOWE - interp- v-round0 0.016 usec crc32=0x423cdcc7 |
|
|
3DNOWE - round1 0.021 usec crc32=0x42202efe |
|
|
3DNOWE - interp-hv-round0 0.037 usec crc32=0xd198d387 |
|
|
3DNOWE - round1 0.036 usec crc32=0x9ecfd921 |
|
|
--- |
|
|
|
|
|
====== test SAD ====== |
|
|
PLAINC - sad8 0.505 usec sad=3776 |
|
|
PLAINC - sad16 1.941 usec sad=27214 |
|
|
PLAINC - sad16bi 4.925 usec sad=26274 |
|
|
PLAINC - dev16 4.254 usec sad=3344 |
|
|
--- |
|
|
MMX - sad8 0.036 usec sad=3776 |
|
|
MMX - sad16 0.107 usec sad=27214 |
|
|
MMX - sad16bi 0.259 usec sad=26274 |
|
|
MMX - dev16 0.187 usec sad=3344 |
|
|
--- |
|
|
MMXEXT - sad8 0.016 usec sad=3776 |
|
|
MMXEXT - sad16 0.050 usec sad=27214 |
|
|
MMXEXT - sad16bi 0.060 usec sad=26274 |
|
|
MMXEXT - dev16 0.086 usec sad=3344 |
|
|
--- |
|
|
3DNOW - sad8 0.506 usec sad=3776 |
|
|
3DNOW - sad16 1.954 usec sad=27214 |
|
|
3DNOW - sad16bi 0.119 usec sad=26274 |
|
|
3DNOW - dev16 4.252 usec sad=3344 |
|
|
--- |
|
|
3DNOWE - sad8 0.017 usec sad=3776 |
|
|
3DNOWE - sad16 0.038 usec sad=27214 |
|
|
3DNOWE - sad16bi 0.052 usec sad=26274 |
|
|
3DNOWE - dev16 0.067 usec sad=3344 |
|
|
--- |
|
|
|
|
|
=== test transfer === |
|
|
PLAINC - 8to16 0.603 usec crc32=0x115814bb |
|
|
PLAINC - 16to8 1.077 usec crc32=0xee7ccbb4 |
|
|
PLAINC - 8to8 0.679 usec crc32=0xd37b3295 |
|
|
PLAINC - 16to8add 1.341 usec crc32=0xdd817bf4 |
|
|
PLAINC - 8to16sub 1.566 usec crc32(1)=0xa1e07163 crc32(2)=0xd86c5d23 |
|
|
PLAINC - 8to16sub2 2.206 usec crc32=0x99b6c4c7 |
|
|
--- |
|
|
MMX - 8to16 -0.025 usec crc32=0x115814bb |
|
|
MMX - 16to8 -0.049 usec crc32=0xee7ccbb4 |
|
|
MMX - 8to8 0.014 usec crc32=0xd37b3295 |
|
|
MMX - 16to8add 0.011 usec crc32=0xdd817bf4 |
|
|
MMX - 8to16sub 0.108 usec crc32(1)=0xa1e07163 crc32(2)=0xd86c5d23 |
|
|
MMX - 8to16sub2 0.164 usec crc32=0x99b6c4c7 |
|
|
--- |
|
|
MMXEXT - 8to16 -0.054 usec crc32=0x115814bb |
|
|
MMXEXT - 16to8 0.010 usec crc32=0xee7ccbb4 |
|
|
MMXEXT - 8to8 0.015 usec crc32=0xd37b3295 |
|
|
MMXEXT - 16to8add 0.008 usec crc32=0xdd817bf4 |
|
|
MMXEXT - 8to16sub 0.263 usec crc32(1)=0xa1e07163 crc32(2)=0xd86c5d23 |
|
|
MMXEXT - 8to16sub2 0.178 usec crc32=0x99b6c4c7 |
|
|
--- |
|
|
3DNOW - 8to16 0.666 usec crc32=0x115814bb |
|
|
3DNOW - 16to8 1.078 usec crc32=0xee7ccbb4 |
|
|
3DNOW - 8to8 0.665 usec crc32=0xd37b3295 |
|
|
3DNOW - 16to8add 1.365 usec crc32=0xdd817bf4 |
|
|
3DNOW - 8to16sub 1.356 usec crc32(1)=0xa1e07163 crc32(2)=0xd86c5d23 |
|
|
3DNOW - 8to16sub2 2.098 usec crc32=0x99b6c4c7 |
|
|
--- |
|
|
3DNOWE - 8to16 -0.024 usec crc32=0x115814bb |
|
|
3DNOWE - 16to8 0.010 usec crc32=0xee7ccbb4 |
|
|
3DNOWE - 8to8 0.014 usec crc32=0xd37b3295 |
|
|
3DNOWE - 16to8add 0.016 usec crc32=0xdd817bf4 |
|
|
3DNOWE - 8to16sub -0.000 usec crc32(1)=0xa1e07163 crc32(2)=0xd86c5d23 |
|
|
3DNOWE - 8to16sub2 -0.031 usec crc32=0x99b6c4c7 |
|
|
--- |
|
|
|
|
|
===== test quant ===== |
|
|
PLAINC - quant_mpeg_intra 98.631 usec crc32=0xfd6a21a4 |
|
|
PLAINC - quant_mpeg_inter 104.876 usec crc32=0xf6de7757 |
|
|
PLAINC - dequant_mpeg_intra 50.285 usec crc32=0x2def7bc7 |
|
|
PLAINC - dequant_mpeg_inter 58.316 usec crc32=0xd878c722 |
|
|
PLAINC - quant_h263_intra 33.803 usec crc32=0x2eba9d43 |
|
|
PLAINC - quant_h263_inter 45.411 usec crc32=0xbd315a7e |
|
|
PLAINC - dequant_h263_intra 39.302 usec crc32=0x9841212a |
|
|
PLAINC - dequant_h263_inter 44.124 usec crc32=0xe7df8fba |
|
|
--- |
|
|
MMX - quant_mpeg_intra 4.273 usec crc32=0xdacabdb6 | ERROR |
|
|
MMX - quant_mpeg_inter 3.576 usec crc32=0x72883ab6 | ERROR |
|
|
MMX - dequant_mpeg_intra 3.793 usec crc32=0x2def7bc7 |
|
|
MMX - dequant_mpeg_inter 4.808 usec crc32=0xd878c722 |
|
|
MMX - quant_h263_intra 2.881 usec crc32=0x2eba9d43 |
|
|
MMX - quant_h263_inter 2.550 usec crc32=0xbd315a7e |
|
|
MMX - dequant_h263_intra 2.974 usec crc32=0x9841212a |
|
|
MMX - dequant_h263_inter 2.906 usec crc32=0xe7df8fba |
|
|
--- |
|
|
MMXEXT - quant_mpeg_intra 4.221 usec crc32=0xfd6a21a4 |
|
|
MMXEXT - quant_mpeg_inter 4.339 usec crc32=0xf6de7757 |
|
|
MMXEXT - dequant_mpeg_intra 3.802 usec crc32=0x2def7bc7 |
|
|
MMXEXT - dequant_mpeg_inter 4.821 usec crc32=0xd878c722 |
|
|
MMXEXT - quant_h263_intra 2.884 usec crc32=0x2eba9d43 |
|
|
MMXEXT - quant_h263_inter 2.554 usec crc32=0xbd315a7e |
|
|
MMXEXT - dequant_h263_intra 2.728 usec crc32=0x9841212a |
|
|
MMXEXT - dequant_h263_inter 2.611 usec crc32=0xe7df8fba |
|
|
--- |
|
|
3DNOW - quant_mpeg_intra 98.512 usec crc32=0xfd6a21a4 |
|
|
3DNOW - quant_mpeg_inter 104.873 usec crc32=0xf6de7757 |
|
|
3DNOW - dequant_mpeg_intra 50.219 usec crc32=0x2def7bc7 |
|
|
3DNOW - dequant_mpeg_inter 58.254 usec crc32=0xd878c722 |
|
|
3DNOW - quant_h263_intra 33.778 usec crc32=0x2eba9d43 |
|
|
3DNOW - quant_h263_inter 41.998 usec crc32=0xbd315a7e |
|
|
3DNOW - dequant_h263_intra 39.344 usec crc32=0x9841212a |
|
|
3DNOW - dequant_h263_inter 43.607 usec crc32=0xe7df8fba |
|
|
--- |
|
|
3DNOWE - quant_mpeg_intra 98.490 usec crc32=0xfd6a21a4 |
|
|
3DNOWE - quant_mpeg_inter 104.889 usec crc32=0xf6de7757 |
|
|
3DNOWE - dequant_mpeg_intra 3.277 usec crc32=0x2def7bc7 |
|
|
3DNOWE - dequant_mpeg_inter 4.485 usec crc32=0xd878c722 |
|
|
3DNOWE - quant_h263_intra 1.882 usec crc32=0x2eba9d43 |
|
|
3DNOWE - quant_h263_inter 2.246 usec crc32=0xbd315a7e |
|
|
3DNOWE - dequant_h263_intra 3.457 usec crc32=0x9841212a |
|
|
3DNOWE - dequant_h263_inter 3.275 usec crc32=0xe7df8fba |
|
|
--- |
|
|
|
|
|
===== test cbp ===== |
|
|
PLAINC - calc_cbp#1 0.168 usec cbp=0x15 |
|
|
PLAINC - calc_cbp#2 0.168 usec cbp=0x38 |
|
|
PLAINC - calc_cbp#3 0.157 usec cbp=0x0f |
|
|
PLAINC - calc_cbp#4 0.235 usec cbp=0x05 |
|
|
--- |
|
|
MMX - calc_cbp#1 0.070 usec cbp=0x15 |
|
|
MMX - calc_cbp#2 0.062 usec cbp=0x38 |
|
|
MMX - calc_cbp#3 0.062 usec cbp=0x0f |
|
|
MMX - calc_cbp#4 0.061 usec cbp=0x05 |
|
|
--- |
|
|
MMXEXT - calc_cbp#1 0.062 usec cbp=0x15 |
|
|
MMXEXT - calc_cbp#2 0.061 usec cbp=0x38 |
|
|
MMXEXT - calc_cbp#3 0.061 usec cbp=0x0f |
|
|
MMXEXT - calc_cbp#4 0.061 usec cbp=0x05 |
|
|
--- |
|
|
3DNOW - calc_cbp#1 0.168 usec cbp=0x15 |
|
|
3DNOW - calc_cbp#2 0.168 usec cbp=0x38 |
|
|
3DNOW - calc_cbp#3 0.157 usec cbp=0x0f |
|
|
3DNOW - calc_cbp#4 0.238 usec cbp=0x05 |
|
|
--- |
|
|
3DNOWE - calc_cbp#1 0.049 usec cbp=0x15 |
|
|
3DNOWE - calc_cbp#2 0.049 usec cbp=0x38 |
|
|
3DNOWE - calc_cbp#3 0.049 usec cbp=0x0f |
|
|
3DNOWE - calc_cbp#4 0.049 usec cbp=0x05 |
|
|
--- |
|
|
|
|
|
|
|
|
NB: If a function isn't optimised for a specific set of intructions, |
|
|
a C function is used instead. So don't panic if some functions |
|
|
may appear to be slow. |
|
|
|
|
|
NB: MMX mpeg4 quantization is known to have very small errors (+/-1 magnitude) |
|
|
for 1 or 2 coefficients a block. This is mainly caused by the fact the unit |
|
|
test goes far behind the usual limits of real encoding. Please do not report |
|
|
this error to the developers |
|
|
|
|
|
*********************************************************************/ |
|