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 |
#ifdef ARCH_IS_PPC |
#ifdef ARCH_IS_PPC |
128 |
, { "ALTIVEC", XVID_CPU_ALTIVEC } |
{ "ALTIVEC", XVID_CPU_ALTIVEC }, |
129 |
#endif |
#endif |
130 |
#ifdef ARCH_IS_X86_64 |
#ifdef ARCH_IS_IA64 |
131 |
, { "X86_64", XVID_CPU_ASM} |
// { "IA64 ", XVID_CPU_IA64 }, |
132 |
#endif |
#endif |
133 |
//, { "IA64 ", XVID_CPU_IA64 } |
// { "TSC ", XVID_CPU_TSC }, |
134 |
//, { "TSC ", XVID_CPU_TSC } |
{ 0, 0 } |
135 |
, { 0, 0 } }; |
}; |
|
|
|
|
CPU cpu_short_list[] = |
|
|
{ { "PLAINC", 0 } |
|
|
#ifdef ARCH_IS_IA32 |
|
|
, { "MMX ", XVID_CPU_MMX } |
|
|
//, { "MMXEXT", XVID_CPU_MMXEXT | XVID_CPU_MMX } |
|
|
#endif |
|
|
//, { "IA64 ", XVID_CPU_IA64 } |
|
|
, { 0, 0 } }; |
|
|
|
|
|
CPU cpu_short_list2[] = |
|
|
{ { "PLAINC", 0 } |
|
|
#ifdef ARCH_IS_IA32 |
|
|
, { "MMX ", XVID_CPU_MMX } |
|
|
, { "SSE2 ", XVID_CPU_SSE2 | XVID_CPU_MMX } |
|
|
#endif |
|
|
, { 0, 0 } }; |
|
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 |
/********************************************************************* |
/********************************************************************* |
518 |
(iCrc!=8107)?"| ERROR": "" ); |
(iCrc!=8107)?"| ERROR": "" ); |
519 |
#endif |
#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 |
} |
} |
558 |
} |
} |
559 |
|
|
560 |
|
#undef ENTER |
561 |
|
#undef LEAVE |
562 |
|
#undef TEST_MB |
563 |
|
#undef TEST_MB2 |
564 |
|
|
565 |
/********************************************************************* |
/********************************************************************* |
566 |
* test transfer |
* test transfer |
567 |
*********************************************************************/ |
*********************************************************************/ |
591 |
} \ |
} \ |
592 |
emms(); \ |
emms(); \ |
593 |
t = (gettime_usec()-t -overhead) / nb_tests;\ |
t = (gettime_usec()-t -overhead) / nb_tests;\ |
594 |
s = calc_crc((uint8_t*)(DST), sizeof((DST)), CRC32_INITIAL) |
byte_swap((uint8_t*)(DST), 8*32*sizeof((DST)[0]), sizeof((DST)[0])); \ |
595 |
|
s = calc_crc((uint8_t*)(DST), 8*32*sizeof((DST)[0]), CRC32_INITIAL) |
596 |
|
|
597 |
#define TEST_TRANSFER(FUNC, DST, SRC) \ |
#define TEST_TRANSFER(FUNC, DST, SRC) \ |
598 |
TEST_TRANSFER_BEGIN(DST); \ |
TEST_TRANSFER_BEGIN(DST); \ |
618 |
} \ |
} \ |
619 |
emms(); \ |
emms(); \ |
620 |
t = (gettime_usec()-t -overhead) / nb_tests;\ |
t = (gettime_usec()-t -overhead) / nb_tests;\ |
621 |
s = calc_crc((uint8_t*)(DST), sizeof((DST)), CRC32_INITIAL) |
byte_swap((uint8_t*)(DST), 8*32*sizeof((DST)[0]), sizeof((DST)[0])); \ |
622 |
|
s = calc_crc((uint8_t*)(DST), 8*32*sizeof((DST)[0]), CRC32_INITIAL) |
623 |
|
|
624 |
#define TEST_TRANSFER2(FUNC, DST, SRC, R1) \ |
#define TEST_TRANSFER2(FUNC, DST, SRC, R1) \ |
625 |
TEST_TRANSFER2_BEGIN(DST,SRC); \ |
TEST_TRANSFER2_BEGIN(DST,SRC); \ |
636 |
const int nb_tests = 4000*speed_ref; |
const int nb_tests = 4000*speed_ref; |
637 |
int i; |
int i; |
638 |
CPU *cpu; |
CPU *cpu; |
639 |
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]; |
640 |
int16_t Src16[8*32], Dst16[8*32]; |
// int16_t Src16[8*32], Dst16[8*32]; |
641 |
|
DECLARE_ALIGNED_MATRIX(Src8, 8, 32, uint8_t, CACHE_LINE); |
642 |
|
DECLARE_ALIGNED_MATRIX(Dst8, 8, 32, uint8_t, CACHE_LINE); |
643 |
|
DECLARE_ALIGNED_MATRIX(Ref1, 8, 32, uint8_t, CACHE_LINE); |
644 |
|
DECLARE_ALIGNED_MATRIX(Ref2, 8, 32, uint8_t, CACHE_LINE); |
645 |
|
DECLARE_ALIGNED_MATRIX(Src16, 8, 32, uint16_t, CACHE_LINE); |
646 |
|
DECLARE_ALIGNED_MATRIX(Dst16, 8, 32, uint16_t, CACHE_LINE); |
647 |
|
|
648 |
printf( "\n === test transfer ===\n" ); |
printf( "\n === test transfer ===\n" ); |
649 |
|
|
665 |
cpu->name, t, s, |
cpu->name, t, s, |
666 |
(s!=0xee7ccbb4)?"| ERROR": ""); |
(s!=0xee7ccbb4)?"| ERROR": ""); |
667 |
|
|
668 |
|
/* New functions for field prediction by CK 1.10.2005 */ |
669 |
|
#pragma NEW8X4 |
670 |
|
TEST_TRANSFER(transfer8x4_copy, Dst8, Src8); |
671 |
|
printf("%s - 8to4 %.3f usec crc32=0x%08x %s\n", |
672 |
|
cpu->name, t, s, |
673 |
|
(s!=0xbb9c3db5)?"| ERROR": ""); |
674 |
|
/* End of new functions */ |
675 |
|
|
676 |
TEST_TRANSFER(transfer8x8_copy, Dst8, Src8); |
TEST_TRANSFER(transfer8x8_copy, Dst8, Src8); |
677 |
printf("%s - 8to8 %.3f usec crc32=0x%08x %s\n", |
printf("%s - 8to8 %.3f usec crc32=0x%08x %s\n", |
678 |
cpu->name, t, s, |
cpu->name, t, s, |
686 |
TEST_TRANSFER2(transfer_8to16sub, Dst16, Src8, Ref1); |
TEST_TRANSFER2(transfer_8to16sub, Dst16, Src8, Ref1); |
687 |
{ |
{ |
688 |
int s1, s2; |
int s1, s2; |
689 |
s1 = calc_crc((uint8_t*)Dst16, sizeof(Dst16), CRC32_INITIAL); |
s1 = calc_crc((uint8_t*)Dst16, 8*32*sizeof(Dst16[0]), CRC32_INITIAL); |
690 |
s2 = calc_crc((uint8_t*)Src8, sizeof(Src8), CRC32_INITIAL); |
s2 = calc_crc((uint8_t*)Src8, 8*32*sizeof(Src8[0]), CRC32_INITIAL); |
691 |
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", |
692 |
cpu->name, t, s1, s2, |
cpu->name, t, s1, s2, |
693 |
(s1!=0xa1e07163)?"| ERROR1": "", |
(s1!=0xa1e07163)?"| ERROR1": "", |
716 |
for(q=1; q<=max_Q; ++q) { \ |
for(q=1; q<=max_Q; ++q) { \ |
717 |
for(tst=0; tst<nb_tests; ++tst) \ |
for(tst=0; tst<nb_tests; ++tst) \ |
718 |
(FUNC)((DST), (SRC), q, mpeg_quant_matrices); \ |
(FUNC)((DST), (SRC), q, mpeg_quant_matrices); \ |
719 |
s = calc_crc((uint8_t*)(DST), 64*sizeof(int16_t), s); \ |
byte_swap((uint8_t*)(DST), 64*sizeof((DST)[0]), sizeof((DST)[0])); \ |
720 |
|
s = calc_crc((uint8_t*)(DST), 64*sizeof((DST)[0]), s); \ |
721 |
} \ |
} \ |
722 |
emms(); \ |
emms(); \ |
723 |
} \ |
} \ |
730 |
set_intra_matrix( mpeg_quant_matrices, Quant ); \ |
set_intra_matrix( mpeg_quant_matrices, Quant ); \ |
731 |
emms(); \ |
emms(); \ |
732 |
for(q=1; q<=max_Q; ++q) { \ |
for(q=1; q<=max_Q; ++q) { \ |
733 |
|
init_intra_matrix( mpeg_quant_matrices, q ); \ |
734 |
for(tst=0; tst<nb_tests; ++tst) \ |
for(tst=0; tst<nb_tests; ++tst) \ |
735 |
(FUNC)((DST), (SRC), q, q, mpeg_quant_matrices); \ |
(FUNC)((DST), (SRC), q, q, mpeg_quant_matrices); \ |
736 |
s = calc_crc((uint8_t*)(DST), 64*sizeof(int16_t), s); \ |
byte_swap((uint8_t*)(DST), 64*sizeof((DST)[0]), sizeof((DST)[0])); \ |
737 |
|
s = calc_crc((uint8_t*)(DST), 64*sizeof((DST)[0]), s); \ |
738 |
} \ |
} \ |
739 |
emms(); \ |
emms(); \ |
740 |
} \ |
} \ |
741 |
t = (gettime_usec()-t-overhead)/nb_tests/qm |
t = (gettime_usec()-t-overhead)/nb_tests/qm |
742 |
|
|
743 |
#define TEST_INTRA(REFFUNC, NEWFUNC, RANGE) \ |
#define TEST_INTRA(REFFUNC, NEWFUNC, RANGE) \ |
744 |
{ int i,q,s;\ |
{ int32_t i,q,s;\ |
745 |
DECLARE_ALIGNED_MATRIX(Src, 8, 8, int16_t, 16); \ |
DECLARE_ALIGNED_MATRIX(Src, 8, 8, int16_t, 16); \ |
746 |
DECLARE_ALIGNED_MATRIX(Dst, 8, 8, int16_t, 16); \ |
DECLARE_ALIGNED_MATRIX(Dst, 8, 8, int16_t, 16); \ |
747 |
DECLARE_ALIGNED_MATRIX(Dst2,8, 8, int16_t, 16); \ |
DECLARE_ALIGNED_MATRIX(Dst2,8, 8, int16_t, 16); \ |
773 |
|
|
774 |
void test_quant() |
void test_quant() |
775 |
{ |
{ |
776 |
const int nb_tests = 1*speed_ref; |
const int32_t nb_tests = 1*speed_ref; |
777 |
const int max_Q = 31; |
const int32_t max_Q = 31; |
778 |
DECLARE_ALIGNED_MATRIX(mpeg_quant_matrices, 8, 64, uint16_t, 16); |
DECLARE_ALIGNED_MATRIX(mpeg_quant_matrices, 8, 64, uint16_t, 16); |
779 |
|
|
780 |
int i, qm; |
int32_t i, qm; |
781 |
CPU *cpu; |
CPU *cpu; |
782 |
DECLARE_ALIGNED_MATRIX(Src, 8, 8, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Src, 8, 8, int16_t, 16); |
783 |
DECLARE_ALIGNED_MATRIX(Dst, 8, 8, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Dst, 8, 8, int16_t, 16); |
796 |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
797 |
{ |
{ |
798 |
double t, overhead; |
double t, overhead; |
799 |
int tst, q; |
int32_t tst, q; |
800 |
uint32_t s; |
uint32_t s; |
801 |
|
|
802 |
if (!init_cpu(cpu)) |
if (!init_cpu(cpu)) |
820 |
TEST_QUANT2(quant_mpeg_intra, Dst, Src); |
TEST_QUANT2(quant_mpeg_intra, Dst, Src); |
821 |
printf("%s - quant_mpeg_intra %.3f usec crc32=0x%08x %s\n", |
printf("%s - quant_mpeg_intra %.3f usec crc32=0x%08x %s\n", |
822 |
cpu->name, t, s, |
cpu->name, t, s, |
823 |
(s!=0xfd6a21a4)? "| ERROR": ""); |
(s!=0x3b999af6)? "| ERROR": ""); |
824 |
|
|
825 |
TEST_QUANT(quant_mpeg_inter, Dst, Src); |
TEST_QUANT(quant_mpeg_inter, Dst, Src); |
826 |
printf("%s - quant_mpeg_inter %.3f usec crc32=0x%08x %s\n", |
printf("%s - quant_mpeg_inter %.3f usec crc32=0x%08x %s\n", |
954 |
* test non-zero AC counting |
* test non-zero AC counting |
955 |
*********************************************************************/ |
*********************************************************************/ |
956 |
|
|
957 |
#define TEST_CBP(FUNC, SRC) \ |
#define TEST_CBP(FUNC, SRC, NB) \ |
958 |
t = gettime_usec(); \ |
t = gettime_usec(); \ |
959 |
emms(); \ |
emms(); \ |
960 |
for(tst=0; tst<nb_tests; ++tst) { \ |
for(tst=0; tst<NB; ++tst) { \ |
961 |
cbp = (FUNC)((SRC)); \ |
cbp = (FUNC)((SRC)); \ |
962 |
} \ |
} \ |
963 |
emms(); \ |
emms(); \ |
966 |
void test_cbp() |
void test_cbp() |
967 |
{ |
{ |
968 |
const int nb_tests = 10000*speed_ref; |
const int nb_tests = 10000*speed_ref; |
969 |
int i; |
int i, n, m; |
970 |
CPU *cpu; |
CPU *cpu; |
971 |
DECLARE_ALIGNED_MATRIX(Src1, 6, 64, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Src1, 6, 64, int16_t, 16); |
972 |
DECLARE_ALIGNED_MATRIX(Src2, 6, 64, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Src2, 6, 64, int16_t, 16); |
973 |
DECLARE_ALIGNED_MATRIX(Src3, 6, 64, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Src3, 6, 64, int16_t, 16); |
974 |
DECLARE_ALIGNED_MATRIX(Src4, 6, 64, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Src4, 6, 64, int16_t, 16); |
975 |
|
DECLARE_ALIGNED_MATRIX(Src5, 6, 64, int16_t, 16); |
976 |
|
|
977 |
printf( "\n ===== test cbp =====\n" ); |
printf( "\n ===== test cbp =====\n" ); |
978 |
|
|
981 |
Src2[i] = (i<3*64); /* half-full */ |
Src2[i] = (i<3*64); /* half-full */ |
982 |
Src3[i] = ((i+32)>3*64); |
Src3[i] = ((i+32)>3*64); |
983 |
Src4[i] = (i==(3*64+2) || i==(5*64+9)); |
Src4[i] = (i==(3*64+2) || i==(5*64+9)); |
984 |
|
Src5[i] = ieee_rand(0,1) ? -1 : 1; /* +/- test */ |
985 |
} |
} |
986 |
|
|
987 |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
992 |
if (!init_cpu(cpu)) |
if (!init_cpu(cpu)) |
993 |
continue; |
continue; |
994 |
|
|
995 |
TEST_CBP(calc_cbp, Src1); |
TEST_CBP(calc_cbp, Src1, nb_tests); |
996 |
printf("%s - calc_cbp#1 %.3f usec cbp=0x%02x %s\n", |
printf("%s - calc_cbp#1 %.3f usec cbp=0x%02x %s\n", |
997 |
cpu->name, t, cbp, (cbp!=0x15)?"| ERROR": ""); |
cpu->name, t, cbp, (cbp!=0x15)?"| ERROR": ""); |
998 |
TEST_CBP(calc_cbp, Src2); |
TEST_CBP(calc_cbp, Src2, nb_tests); |
999 |
printf("%s - calc_cbp#2 %.3f usec cbp=0x%02x %s\n", |
printf("%s - calc_cbp#2 %.3f usec cbp=0x%02x %s\n", |
1000 |
cpu->name, t, cbp, (cbp!=0x38)?"| ERROR": ""); |
cpu->name, t, cbp, (cbp!=0x38)?"| ERROR": ""); |
1001 |
TEST_CBP(calc_cbp, Src3); |
TEST_CBP(calc_cbp, Src3, nb_tests); |
1002 |
printf("%s - calc_cbp#3 %.3f usec cbp=0x%02x %s\n", |
printf("%s - calc_cbp#3 %.3f usec cbp=0x%02x %s\n", |
1003 |
cpu->name, t, cbp, (cbp!=0x0f)?"| ERROR": "" ); |
cpu->name, t, cbp, (cbp!=0x0f)?"| ERROR": "" ); |
1004 |
TEST_CBP(calc_cbp, Src4); |
TEST_CBP(calc_cbp, Src4, nb_tests); |
1005 |
printf("%s - calc_cbp#4 %.3f usec cbp=0x%02x %s\n", |
printf("%s - calc_cbp#4 %.3f usec cbp=0x%02x %s\n", |
1006 |
cpu->name, t, cbp, (cbp!=0x05)?"| ERROR": "" ); |
cpu->name, t, cbp, (cbp!=0x05)?"| ERROR": "" ); |
1007 |
|
TEST_CBP(calc_cbp, Src5, nb_tests); |
1008 |
|
printf("%s - calc_cbp#4 %.3f usec cbp=0x%02x %s\n", |
1009 |
|
cpu->name, t, cbp, (cbp!=0x3f)?"| ERROR": "" ); |
1010 |
printf( " --- \n" ); |
printf( " --- \n" ); |
1011 |
} |
} |
1012 |
|
|
1013 |
|
for(cpu = cpu_list; cpu->name!=0; ++cpu) /* bench suggested by Carlo (carlo dot bramix at libero dot it) */ |
1014 |
|
{ |
1015 |
|
double t; |
1016 |
|
int tst, cbp, err; |
1017 |
|
|
1018 |
|
if (!init_cpu(cpu)) |
1019 |
|
continue; |
1020 |
|
|
1021 |
|
err = 0; |
1022 |
|
for(n=0; n<6; ++n) |
1023 |
|
{ |
1024 |
|
for(m=0; m<64; ++m) |
1025 |
|
{ |
1026 |
|
for(i=0; i<6*64; ++i) |
1027 |
|
Src1[i] = (i== (m + n*64)); |
1028 |
|
|
1029 |
|
TEST_CBP(calc_cbp, Src1, 1); |
1030 |
|
if (cbp!= (((m!=0)<<(5-n)))) |
1031 |
|
{ |
1032 |
|
printf( "%s - calc_cbp#5: ERROR at pos %d / %d!\n", cpu->name, n, m); |
1033 |
|
err = 1; |
1034 |
|
break; |
1035 |
|
} |
1036 |
|
} |
1037 |
|
} |
1038 |
|
if (!err) |
1039 |
|
printf( " %s - calc_cbp#5 : OK\n", cpu->name ); |
1040 |
|
|
1041 |
|
} |
1042 |
} |
} |
1043 |
|
|
1044 |
/********************************************************************* |
/********************************************************************* |
1432 |
* measure raw decoding speed |
* measure raw decoding speed |
1433 |
*********************************************************************/ |
*********************************************************************/ |
1434 |
|
|
1435 |
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) |
1436 |
{ |
{ |
1437 |
FILE *f = 0; |
FILE *f = 0; |
1438 |
void *dechandle = 0; |
void *dechandle = 0; |
1443 |
double t = 0.; |
double t = 0.; |
1444 |
int nb = 0; |
int nb = 0; |
1445 |
uint8_t *buf = 0; |
uint8_t *buf = 0; |
1446 |
uint8_t *rgb_out = 0; |
uint8_t *yuv_out = 0; |
1447 |
int buf_size, pos; |
int buf_size, pos; |
1448 |
uint32_t chksum = 0; |
uint32_t chksum = 0; |
1449 |
|
int bps = (width+31) & ~31; |
1450 |
|
|
1451 |
memset(&xinit, 0, sizeof(xinit)); |
memset(&xinit, 0, sizeof(xinit)); |
1452 |
xinit.cpu_flags = XVID_CPU_MMX | XVID_CPU_FORCE; |
xinit.cpu_flags = cpu_mask; |
1453 |
xinit.version = XVID_VERSION; |
xinit.version = XVID_VERSION; |
1454 |
xvid_global(NULL, 0, &xinit, NULL); |
xvid_global(NULL, 0, &xinit, NULL); |
1455 |
|
|
1459 |
xparam.version = XVID_VERSION; |
xparam.version = XVID_VERSION; |
1460 |
xerr = xvid_decore(NULL, XVID_DEC_CREATE, &xparam, NULL); |
xerr = xvid_decore(NULL, XVID_DEC_CREATE, &xparam, NULL); |
1461 |
if (xerr==XVID_ERR_FAIL) { |
if (xerr==XVID_ERR_FAIL) { |
1462 |
printf("can't init decoder (err=%d)\n", xerr); |
printf("ERROR: can't init decoder (err=%d)\n", xerr); |
1463 |
return; |
return; |
1464 |
} |
} |
1465 |
dechandle = xparam.handle; |
dechandle = xparam.handle; |
1467 |
|
|
1468 |
f = fopen(name, "rb"); |
f = fopen(name, "rb"); |
1469 |
if (f==0) { |
if (f==0) { |
1470 |
printf( "can't open file '%s'\n", name); |
printf( "ERROR: can't open file '%s'\n", name); |
1471 |
return; |
return; |
1472 |
} |
} |
1473 |
fseek(f, 0, SEEK_END); |
fseek(f, 0, SEEK_END); |
1474 |
buf_size = ftell(f); |
buf_size = ftell(f); |
1475 |
fseek(f, 0, SEEK_SET); |
fseek(f, 0, SEEK_SET); |
1476 |
if (buf_size<=0) { |
if (buf_size<=0) { |
1477 |
printf("error while stating file\n"); |
printf("ERROR: error while stating file\n"); |
1478 |
goto End; |
goto End; |
1479 |
} |
} |
|
else printf( "Input size: %d\n", buf_size); |
|
1480 |
|
|
1481 |
buf = malloc(buf_size); /* should be enuf' */ |
buf = malloc(buf_size); |
1482 |
rgb_out = calloc(4, width*height); /* <-room for _RGB24 */ |
yuv_out = calloc(1, bps*height*3/2 + 15); |
1483 |
if (buf==0 || rgb_out==0) { |
if (buf==0 || yuv_out==0) { |
1484 |
printf( "malloc failed!\n" ); |
printf( "ERROR: malloc failed!\n" ); |
1485 |
goto End; |
goto End; |
1486 |
} |
} |
1487 |
|
|
1488 |
if (fread(buf, buf_size, 1, f)!=1) { |
if (fread(buf, buf_size, 1, f)!=1) { |
1489 |
printf( "file-read failed\n" ); |
printf( "ERROR: file-read failed\n" ); |
1490 |
goto End; |
goto End; |
1491 |
} |
} |
1492 |
|
|
1494 |
pos = 0; |
pos = 0; |
1495 |
t = -gettime_usec(); |
t = -gettime_usec(); |
1496 |
while(1) { |
while(1) { |
1497 |
|
int y; |
1498 |
|
|
1499 |
memset(&xframe, 0, sizeof(xframe)); |
memset(&xframe, 0, sizeof(xframe)); |
1500 |
xframe.version = XVID_VERSION; |
xframe.version = XVID_VERSION; |
1501 |
xframe.bitstream = buf + pos; |
xframe.bitstream = buf + pos; |
1502 |
xframe.length = buf_size - pos; |
xframe.length = buf_size - pos; |
1503 |
xframe.output.plane[0] = rgb_out; |
xframe.output.plane[0] = (uint8_t*)(((size_t)yuv_out + 15) & ~15); |
1504 |
xframe.output.stride[0] = width; |
xframe.output.plane[1] = (uint8_t*)xframe.output.plane[0] + bps*height; |
1505 |
xframe.output.csp = XVID_CSP_BGR; |
xframe.output.plane[2] = (uint8_t*)xframe.output.plane[1] + bps/2; |
1506 |
|
xframe.output.stride[0] = bps; |
1507 |
|
xframe.output.stride[1] = bps; |
1508 |
|
xframe.output.stride[2] = bps; |
1509 |
|
xframe.output.csp = XVID_CSP_I420; |
1510 |
xerr = xvid_decore(dechandle, XVID_DEC_DECODE, &xframe, 0); |
xerr = xvid_decore(dechandle, XVID_DEC_DECODE, &xframe, 0); |
1511 |
|
if (xerr<0) { |
1512 |
|
printf("ERROR: decoding failed for frame #%d (err=%d)!\n", nb, xerr); |
1513 |
|
break; |
1514 |
|
} |
1515 |
|
else if (xerr==0) |
1516 |
|
break; |
1517 |
|
else if (verbose>0) printf("#%d %d\n", nb, xerr ); |
1518 |
|
|
1519 |
|
pos += xerr; |
1520 |
nb++; |
nb++; |
1521 |
pos += xframe.length; |
|
1522 |
if (with_chksum) { |
for(y=0; y<height/2; ++y) { |
1523 |
int k = width*height; |
chksum = calc_crc((uint8_t*)xframe.output.plane[0] + (2*y+0)*bps, width, chksum); |
1524 |
uint32_t *ptr = (uint32_t *)rgb_out; |
chksum = calc_crc((uint8_t*)xframe.output.plane[0] + (2*y+1)*bps, width, chksum); |
1525 |
while(k-->0) chksum += *ptr++; |
chksum = calc_crc((uint8_t*)xframe.output.plane[1] + y*bps, width/2, chksum); |
1526 |
|
chksum = calc_crc((uint8_t*)xframe.output.plane[2] + y*bps, width/2, chksum); |
1527 |
} |
} |
1528 |
if (pos==buf_size) |
if (pos==buf_size) |
1529 |
break; |
break; |
|
if (xerr==XVID_ERR_FAIL) { |
|
|
printf("decoding failed for frame #%d (err=%d)!\n", nb, xerr); |
|
|
break; |
|
|
} |
|
1530 |
} |
} |
1531 |
t += gettime_usec(); |
t += gettime_usec(); |
1532 |
|
if (ref_chksum==0) { |
1533 |
if (t>0.) |
if (t>0.) |
1534 |
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 ); |
1535 |
if (with_chksum) |
} |
1536 |
printf("checksum: 0x%.8x\n", chksum); |
else { |
1537 |
|
printf("FPS:%.1f Checksum: 0x%.8x Expected:0x%.8x | %s\n", |
1538 |
|
t>0. ? (float)(nb*1.e6f/t) : 0.f, chksum, ref_chksum, (chksum==ref_chksum) ? "OK" : "ERROR"); |
1539 |
|
} |
1540 |
|
|
1541 |
End: |
End: |
1542 |
if (rgb_out!=0) free(rgb_out); |
if (yuv_out!=0) free(yuv_out); |
1543 |
if (buf!=0) free(buf); |
if (buf!=0) free(buf); |
1544 |
if (dechandle!=0) { |
if (dechandle!=0) { |
1545 |
xerr= xvid_decore(dechandle, XVID_DEC_DESTROY, NULL, NULL); |
xerr= xvid_decore(dechandle, XVID_DEC_DESTROY, NULL, NULL); |
1546 |
if (xerr==XVID_ERR_FAIL) |
if (xerr==XVID_ERR_FAIL) |
1547 |
printf("destroy-decoder failed (err=%d)!\n", xerr); |
printf("ERROR: destroy-decoder failed (err=%d)!\n", xerr); |
1548 |
} |
} |
1549 |
if (f!=0) fclose(f); |
if (f!=0) fclose(f); |
1550 |
} |
} |
1708 |
} |
} |
1709 |
|
|
1710 |
/********************************************************************* |
/********************************************************************* |
1711 |
|
* test some YUV func |
1712 |
|
*********************************************************************/ |
1713 |
|
|
1714 |
|
#define ENTER \ |
1715 |
|
for(i=0; i<(int)sizeof(Dst0); ++i) Dst0[0][i] = 0; \ |
1716 |
|
t = gettime_usec(); \ |
1717 |
|
emms(); |
1718 |
|
|
1719 |
|
#define LEAVE \ |
1720 |
|
emms(); \ |
1721 |
|
t = (gettime_usec() - t) / nb_tests; \ |
1722 |
|
iCrc = calc_crc((uint8_t*)Dst0, sizeof(Dst0), CRC32_INITIAL) |
1723 |
|
|
1724 |
|
#define TEST_YUYV(FUNC, S, FLIP) \ |
1725 |
|
ENTER \ |
1726 |
|
for(tst=0; tst<nb_tests; ++tst) (FUNC)(Dst0[0], S*WIDTH, Src0[0], Src0[1], Src0[2], WIDTH, WIDTH/2, WIDTH, HEIGHT, (FLIP)); \ |
1727 |
|
LEAVE |
1728 |
|
|
1729 |
|
static const int yuv_CRCs[6][2] = { |
1730 |
|
{0x0f4fb96b,0x780b6a68} |
1731 |
|
, {0xa986b289,0x65e49b76} |
1732 |
|
, {0x7f19c152,0xd539b86e} |
1733 |
|
, {0x0f4fb96b,0x780b6a68} |
1734 |
|
, {0xa986b289,0x65e49b76} |
1735 |
|
, {0x36ab8b57,0x1cd92fee} |
1736 |
|
}; |
1737 |
|
|
1738 |
|
#define WIDTH 128 |
1739 |
|
#define HEIGHT 32 |
1740 |
|
void test_yuv() |
1741 |
|
{ |
1742 |
|
const int nb_tests = 200*speed_ref; |
1743 |
|
CPU *cpu; |
1744 |
|
uint8_t Src0[3][WIDTH*HEIGHT]; |
1745 |
|
uint8_t Dst0[4][WIDTH*HEIGHT]; |
1746 |
|
int i, j, with_flip; |
1747 |
|
double t; |
1748 |
|
int tst, iCrc; |
1749 |
|
|
1750 |
|
colorspace_init(); |
1751 |
|
ieee_reseed(1); |
1752 |
|
for(i=0; i<(int)sizeof(Src0); ++i) Src0[0][i] = ieee_rand(0,255); |
1753 |
|
for(i=0; i<(int)sizeof(Dst0); ++i) Dst0[0][i] = 0x5a; |
1754 |
|
|
1755 |
|
printf( "\n === test YUV ===\n" ); |
1756 |
|
|
1757 |
|
for(with_flip=0; with_flip<=1; ++with_flip) { |
1758 |
|
|
1759 |
|
init_cpu(&cpu_list[0]); |
1760 |
|
TEST_YUYV(yv12_to_yuyv_c, 4, with_flip); |
1761 |
|
printf(" yv12_to_yuyv_c %.3f usec crc32=0x%08x %s\n", |
1762 |
|
t, iCrc, (iCrc!=yuv_CRCs[0][with_flip])?"| ERROR": "" ); |
1763 |
|
TEST_YUYV(yv12_to_uyvy_c, 4, with_flip); |
1764 |
|
printf(" yv12_to_uyvy_c %.3f usec crc32=0x%08x %s\n", |
1765 |
|
t, iCrc, (iCrc!=yuv_CRCs[1][with_flip])?"| ERROR": "" ); |
1766 |
|
|
1767 |
|
TEST_YUYV(yv12_to_bgra_c, 4, with_flip); |
1768 |
|
printf(" yv12_to_bgra_c %.3f usec crc32=0x%08x %s\n", |
1769 |
|
t, iCrc, (iCrc!=yuv_CRCs[2][with_flip])?"| ERROR": "" ); |
1770 |
|
|
1771 |
|
#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) |
1772 |
|
init_cpu(&cpu_list[1]); |
1773 |
|
TEST_YUYV(yv12_to_yuyv_mmx, 4, with_flip); |
1774 |
|
printf(" yv12_to_yuyv_mmx %.3f usec crc32=0x%08x %s\n", |
1775 |
|
t, iCrc, (iCrc!=yuv_CRCs[3][with_flip])?"| ERROR": "" ); |
1776 |
|
|
1777 |
|
TEST_YUYV(yv12_to_uyvy_mmx, 4, with_flip); |
1778 |
|
printf(" yv12_to_uyvy_mmx %.3f usec crc32=0x%08x %s\n", |
1779 |
|
t, iCrc, (iCrc!=yuv_CRCs[4][with_flip])?"| ERROR": "" ); |
1780 |
|
|
1781 |
|
TEST_YUYV(yv12_to_bgra_mmx, 4, with_flip); |
1782 |
|
printf(" yv12_to_bgra_mmx %.3f usec crc32=0x%08x %s\n", |
1783 |
|
t, iCrc, (iCrc!=yuv_CRCs[5][with_flip])?"| ERROR": "" ); |
1784 |
|
|
1785 |
|
#endif |
1786 |
|
|
1787 |
|
#ifdef ARCH_IS_PPC |
1788 |
|
init_cpu(&cpu_list[1]); |
1789 |
|
TEST_YUYV(yv12_to_yuyv_altivec_c, 4, with_flip); |
1790 |
|
printf(" yv12_to_yuyv_altivec_c %.3f usec crc32=0x%08x %s\n", |
1791 |
|
t, iCrc, (iCrc!=yuv_CRCs[3][with_flip])?"| ERROR": "" ); |
1792 |
|
|
1793 |
|
TEST_YUYV(yv12_to_uyvy_altivec_c, 4, with_flip); |
1794 |
|
printf(" yv12_to_uyvy_altivec_c %.3f usec crc32=0x%08x %s\n", |
1795 |
|
t, iCrc, (iCrc!=yuv_CRCs[4][with_flip])?"| ERROR": "" ); |
1796 |
|
|
1797 |
|
TEST_YUYV(yv12_to_bgra_altivec_c, 4, with_flip); |
1798 |
|
printf(" yv12_to_bgra_altivec_c %.3f usec crc32=0x%08x %s\n", |
1799 |
|
t, iCrc, (iCrc!=yuv_CRCs[5][with_flip])?"| ERROR": "" ); |
1800 |
|
#endif |
1801 |
|
} |
1802 |
|
printf( " --- \n" ); |
1803 |
|
} |
1804 |
|
|
1805 |
|
#define TEST_YV2(FUNC, WITH_UV, WITH_FLIP) \ |
1806 |
|
ENTER \ |
1807 |
|
for(tst=0; tst<nb_tests; ++tst) (FUNC)(Dst0[0], Dst0[1], Dst0[2], WIDTH, WIDTH, \ |
1808 |
|
Src0[0], (WITH_UV) ? Src0[1] : 0, (WITH_UV) ? Src0[2] : 0, WIDTH, WIDTH, \ |
1809 |
|
WIDTH-2, HEIGHT-2, WITH_FLIP); \ |
1810 |
|
LEAVE |
1811 |
|
|
1812 |
|
#define PRINT_NxN(DATA,W,H,STR) { \ |
1813 |
|
int i,j; \ |
1814 |
|
for(j=0; j<(H); ++j) { \ |
1815 |
|
for(i=0; i<(W); ++i) printf( "0x%.2x ", (DATA)[i+j*(STR)] );\ |
1816 |
|
printf("\n"); \ |
1817 |
|
} \ |
1818 |
|
printf("---\n"); \ |
1819 |
|
} |
1820 |
|
|
1821 |
|
static const int yv12_CRCs[2][2] = { |
1822 |
|
{0x5cab7cf0,0xdab46541} |
1823 |
|
, {0xe8bae865,0x1faf77b7} |
1824 |
|
}; |
1825 |
|
|
1826 |
|
void test_yuv2() |
1827 |
|
{ |
1828 |
|
const int nb_tests = 800*speed_ref; |
1829 |
|
CPU *cpu; |
1830 |
|
uint8_t Src0[3][WIDTH*HEIGHT]; |
1831 |
|
uint8_t Dst0[3][WIDTH*HEIGHT]; |
1832 |
|
int with_uv, with_flip; |
1833 |
|
int i, j; |
1834 |
|
double t; |
1835 |
|
int tst, iCrc; |
1836 |
|
|
1837 |
|
colorspace_init(); |
1838 |
|
ieee_reseed(1); |
1839 |
|
for(i=0; i<(int)sizeof(Src0); ++i) Src0[0][i] = ieee_rand(0,255); |
1840 |
|
|
1841 |
|
printf( "\n === test YV2 ===\n" ); |
1842 |
|
for(with_flip=0; with_flip<=1; ++with_flip) { |
1843 |
|
for(with_uv=0; with_uv<=1; ++with_uv) { |
1844 |
|
init_cpu(&cpu_list[0]); |
1845 |
|
TEST_YV2(yv12_to_yv12_c, with_uv, with_flip); |
1846 |
|
printf(" yv12_to_yv12_c %.3f usec \tcrc32=0x%08x %s\n", |
1847 |
|
t, iCrc, (iCrc!=yv12_CRCs[with_flip][with_uv])?"| ERROR": "" ); |
1848 |
|
/* if (!with_uv) PRINT_NxN(Dst0[1], WIDTH/2, HEIGHT/2, WIDTH ); */ |
1849 |
|
|
1850 |
|
#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) |
1851 |
|
init_cpu(&cpu_list[1]); |
1852 |
|
TEST_YV2(yv12_to_yv12_mmx, with_uv, with_flip); |
1853 |
|
printf(" yv12_to_yv12_mmx %.3f usec \tcrc32=0x%08x %s\n", |
1854 |
|
t, iCrc, (iCrc!=yv12_CRCs[with_flip][with_uv])?"| ERROR": "" ); |
1855 |
|
/* if (!with_uv) PRINT_NxN(Dst0[1], WIDTH/2, HEIGHT/2, WIDTH ); */ |
1856 |
|
|
1857 |
|
TEST_YV2(yv12_to_yv12_xmm, with_uv, with_flip); |
1858 |
|
printf(" yv12_to_yv12_xmm %.3f usec \tcrc32=0x%08x %s\n", |
1859 |
|
t, iCrc, (iCrc!=yv12_CRCs[with_flip][with_uv])?"| ERROR": "" ); |
1860 |
|
#endif |
1861 |
|
} |
1862 |
|
|
1863 |
|
printf( " --- \n" ); |
1864 |
|
} |
1865 |
|
printf( " ===== \n" ); |
1866 |
|
} |
1867 |
|
|
1868 |
|
#undef WIDTH |
1869 |
|
#undef HEIGHT |
1870 |
|
#undef ENTER |
1871 |
|
#undef LEAVE |
1872 |
|
|
1873 |
|
/*********************************************************************/ |
1874 |
|
|
1875 |
|
static uint32_t __inline log2bin_v1(uint32_t value) |
1876 |
|
{ |
1877 |
|
int n = 0; |
1878 |
|
while (value) { |
1879 |
|
value >>= 1; |
1880 |
|
n++; |
1881 |
|
} |
1882 |
|
return n; |
1883 |
|
} |
1884 |
|
|
1885 |
|
static const uint8_t log2_tab_16[16] = { 0, 1, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4 }; |
1886 |
|
|
1887 |
|
static uint32_t __inline log2bin_v2(uint32_t value) |
1888 |
|
{ |
1889 |
|
int n = 0; |
1890 |
|
if (value & 0xffff0000) { |
1891 |
|
value >>= 16; |
1892 |
|
n += 16; |
1893 |
|
} |
1894 |
|
if (value & 0xff00) { |
1895 |
|
value >>= 8; |
1896 |
|
n += 8; |
1897 |
|
} |
1898 |
|
if (value & 0xf0) { |
1899 |
|
value >>= 4; |
1900 |
|
n += 4; |
1901 |
|
} |
1902 |
|
return n + log2_tab_16[value]; |
1903 |
|
} |
1904 |
|
|
1905 |
|
void test_log2bin() |
1906 |
|
{ |
1907 |
|
const int nb_tests = 3000*speed_ref; |
1908 |
|
int n, crc1=0, crc2=0; |
1909 |
|
uint32_t s, s0; |
1910 |
|
double t1, t2; |
1911 |
|
|
1912 |
|
t1 = gettime_usec(); |
1913 |
|
s0 = (int)(t1*31.241); |
1914 |
|
for(s=s0, n=0; n<nb_tests; ++n, s=(s*12363+31)&0x7fffffff) |
1915 |
|
crc1 += log2bin_v1(s); |
1916 |
|
t1 = (gettime_usec()-t1) / nb_tests; |
1917 |
|
|
1918 |
|
t2 = gettime_usec(); |
1919 |
|
for(s=s0, n=0; n<nb_tests; ++n, s=(s*12363+31)&0x7fffffff) |
1920 |
|
crc2 += log2bin_v2(s); |
1921 |
|
t2 = (gettime_usec() - t2) / nb_tests; |
1922 |
|
|
1923 |
|
printf( "log2bin_v1: %.3f sec crc=%d\n", t1, crc1 ); |
1924 |
|
printf( "log2bin_v2: %.3f sec crc=%d\n", t2, crc2 ); |
1925 |
|
if (crc1!=crc2) printf( " CRC ERROR !\n" ); |
1926 |
|
} |
1927 |
|
|
1928 |
|
/*********************************************************************/ |
1929 |
|
|
1930 |
|
static void __inline old_gcd(int *num, int *den) |
1931 |
|
{ |
1932 |
|
int i = *num; |
1933 |
|
while (i > 1) { |
1934 |
|
if (*num % i == 0 && *den % i == 0) { |
1935 |
|
*num /= i; |
1936 |
|
*den /= i; |
1937 |
|
i = *num; |
1938 |
|
continue; |
1939 |
|
} |
1940 |
|
i--; |
1941 |
|
} |
1942 |
|
} |
1943 |
|
|
1944 |
|
static uint32_t gcd(int num, int den) |
1945 |
|
{ |
1946 |
|
int tmp; |
1947 |
|
while( (tmp=num%den) ) { num = den; den = tmp; } |
1948 |
|
return den; |
1949 |
|
} |
1950 |
|
static void __inline new_gcd(int *num, int *den) |
1951 |
|
{ |
1952 |
|
const int div = gcd(*num, *den); |
1953 |
|
if (num) { |
1954 |
|
*num /= div; |
1955 |
|
*den /= div; |
1956 |
|
} |
1957 |
|
} |
1958 |
|
|
1959 |
|
void test_gcd() |
1960 |
|
{ |
1961 |
|
const int nb_tests = 10*speed_ref; |
1962 |
|
int i; |
1963 |
|
uint32_t crc1=0, crc2=0; |
1964 |
|
uint32_t n0, n, d0, d; |
1965 |
|
double t1, t2; |
1966 |
|
|
1967 |
|
t1 = gettime_usec(); |
1968 |
|
n0 = 0xfffff & (int)(t1*31.241); |
1969 |
|
d0 = 0xfffff & (int)( ((n0*4123)%17) | 1 ); |
1970 |
|
for(n=n0, d=d0, i=0; i<nb_tests; ++i) { |
1971 |
|
old_gcd(&n, &d); |
1972 |
|
crc1 = (((crc1>>4)^d) + ((crc1<<2)^n) ) & 0xffffff; |
1973 |
|
n = d; |
1974 |
|
d = (d*12363+31) & 0xffff; |
1975 |
|
d |= !d; |
1976 |
|
} |
1977 |
|
t1 = (gettime_usec()-t1) / nb_tests; |
1978 |
|
|
1979 |
|
t2 = gettime_usec(); |
1980 |
|
for(n=n0, d=d0, i=0; i<nb_tests; ++i) { |
1981 |
|
new_gcd(&n, &d); |
1982 |
|
crc2 = (((crc2>>4)^d) + ((crc2<<2)^n) ) & 0xffffff; |
1983 |
|
n = d; |
1984 |
|
d = (d*12363+31) & 0xffff; |
1985 |
|
d |= !d; |
1986 |
|
} |
1987 |
|
t2 = (gettime_usec() - t2) / nb_tests; |
1988 |
|
|
1989 |
|
printf( "old_gcd: %.3f sec crc=%d\n", t1, crc1 ); |
1990 |
|
printf( "new_gcd: %.3f sec crc=%d\n", t2, crc2 ); |
1991 |
|
if (crc1!=crc2) printf( " CRC ERROR !\n" ); |
1992 |
|
} |
1993 |
|
|
1994 |
|
/********************************************************************* |
1995 |
|
* test compiler |
1996 |
|
*********************************************************************/ |
1997 |
|
|
1998 |
|
void test_compiler() { |
1999 |
|
int nb_err = 0; |
2000 |
|
int32_t v; |
2001 |
|
if (sizeof(uint16_t)<2) { |
2002 |
|
printf( "ERROR: sizeof(uint16_t)<2 !!\n" ); |
2003 |
|
nb_err++; |
2004 |
|
} |
2005 |
|
if (sizeof(int16_t)<2) { |
2006 |
|
printf( "ERROR: sizeof(int16_t)<2 !!\n" ); |
2007 |
|
nb_err++; |
2008 |
|
} |
2009 |
|
if (sizeof(uint8_t)!=1) { |
2010 |
|
printf( "ERROR: sizeof(uint8_t)!=1 !!\n" ); |
2011 |
|
nb_err++; |
2012 |
|
} |
2013 |
|
if (sizeof(int8_t)!=1) { |
2014 |
|
printf( "ERROR: sizeof(int8_t)!=1 !!\n" ); |
2015 |
|
nb_err++; |
2016 |
|
} |
2017 |
|
if (sizeof(uint32_t)<4) { |
2018 |
|
printf( "ERROR: sizeof(uint32_t)<4 !!\n" ); |
2019 |
|
nb_err++; |
2020 |
|
} |
2021 |
|
if (sizeof(int32_t)<4) { |
2022 |
|
printf( "ERROR: sizeof(int32_t)<4 !!\n" ); |
2023 |
|
nb_err++; |
2024 |
|
} |
2025 |
|
/* yes, i know, this test is silly. But better be safe than sorry. :) */ |
2026 |
|
for(v=1000; v>=0; v--) { |
2027 |
|
if ( (v>>2) != v/4) |
2028 |
|
nb_err++; |
2029 |
|
} |
2030 |
|
for(v=-1000; v!=-1; v++) { |
2031 |
|
if ( (v>>2) != (v/4)-!!(v%4)) |
2032 |
|
nb_err++; |
2033 |
|
} |
2034 |
|
if (nb_err!=0) { |
2035 |
|
printf( "ERROR! please post your platform/compiler specs to xvid-devel@xvid.org !\n" ); |
2036 |
|
} |
2037 |
|
} |
2038 |
|
|
2039 |
|
/********************************************************************* |
2040 |
|
* test SSIM functions |
2041 |
|
*********************************************************************/ |
2042 |
|
|
2043 |
|
typedef int (*lumfunc)(uint8_t* ptr, int stride); |
2044 |
|
typedef void (*csfunc)(uint8_t* ptro, uint8_t* ptrc, int stride, int lumo, int lumc, int* pdevo, int* pdevc, int* pcorr); |
2045 |
|
|
2046 |
|
extern int lum_8x8_c(uint8_t* ptr, int stride); |
2047 |
|
extern int lum_8x8_mmx(uint8_t* ptr, int stride); |
2048 |
|
extern int lum_2x8_c(uint8_t* ptr, int stride); |
2049 |
|
extern void consim_c(uint8_t* ptro, uint8_t* ptrc, int stride, int lumo, int lumc, int* pdevo, int* pdevc, int* pcorr); |
2050 |
|
extern void consim_mmx(uint8_t* ptro, uint8_t* ptrc, int stride, int lumo, int lumc, int* pdevo, int* pdevc, int* pcorr); |
2051 |
|
extern void consim_sse2(uint8_t* ptro, uint8_t* ptrc, int stride, int lumo, int lumc, int* pdevo, int* pdevc, int* pcorr); |
2052 |
|
|
2053 |
|
void test_SSIM() |
2054 |
|
{ |
2055 |
|
const int nb_tests = 3000*speed_ref; |
2056 |
|
int tst; |
2057 |
|
CPU *cpu; |
2058 |
|
int i; |
2059 |
|
int devs[3]; |
2060 |
|
long lumo, lumc; |
2061 |
|
DECLARE_ALIGNED_MATRIX(Ref1, 16, 16, uint8_t, 16); |
2062 |
|
DECLARE_ALIGNED_MATRIX(Ref2, 16, 16, uint8_t, 16); |
2063 |
|
lumfunc lum8x8; |
2064 |
|
lumfunc lum2x8; |
2065 |
|
csfunc csim; |
2066 |
|
|
2067 |
|
ieee_reseed(1); |
2068 |
|
printf( "\n ====== test SSIM ======\n" ); |
2069 |
|
for(i=0; i<16*16;++i) { |
2070 |
|
long v1, v2; |
2071 |
|
v1 = ieee_rand(-256, 511); |
2072 |
|
v2 = ieee_rand(-256, 511); |
2073 |
|
Ref1[i] = (v1<0) ? 0 : (v1>255) ? 255 : v1; |
2074 |
|
Ref2[i] = (v2<0) ? 0 : (v2>255) ? 255 : v2; |
2075 |
|
} |
2076 |
|
lumc = ieee_rand(0, 255); |
2077 |
|
lumo = ieee_rand(0, 255); |
2078 |
|
|
2079 |
|
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
2080 |
|
{ |
2081 |
|
double t; |
2082 |
|
int m; |
2083 |
|
if (!init_cpu(cpu)) |
2084 |
|
continue; |
2085 |
|
lum8x8 = lum_8x8_c; |
2086 |
|
lum2x8 = lum_2x8_c; |
2087 |
|
csim = consim_c; |
2088 |
|
#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) |
2089 |
|
if (cpu->cpu & XVID_CPU_MMX){ |
2090 |
|
lum8x8 = lum_8x8_mmx; |
2091 |
|
csim = consim_mmx; |
2092 |
|
} |
2093 |
|
if (cpu->cpu & XVID_CPU_MMX){ |
2094 |
|
csim = consim_sse2; |
2095 |
|
} |
2096 |
|
#endif |
2097 |
|
t = gettime_usec(); |
2098 |
|
emms(); |
2099 |
|
for(tst=0; tst<nb_tests; ++tst) m = lum8x8(Ref1, 16); |
2100 |
|
emms(); |
2101 |
|
t = (gettime_usec() - t) / nb_tests; |
2102 |
|
printf("%s - ssim-lum8x8 %.3f usec m=%d %s\n", |
2103 |
|
cpu->name, t, m, |
2104 |
|
(m!=8230)?"| ERROR": "" ); |
2105 |
|
|
2106 |
|
t = gettime_usec(); |
2107 |
|
emms(); |
2108 |
|
for(tst=0; tst<nb_tests; ++tst) m = lum2x8(Ref1+8, 16); |
2109 |
|
emms(); |
2110 |
|
t = (gettime_usec() - t) / nb_tests; |
2111 |
|
printf("%s - ssim-lum2x8 %.3f usec m=%d %s\n", |
2112 |
|
cpu->name, t, m, |
2113 |
|
(m!=681)?"| ERROR": "" ); |
2114 |
|
|
2115 |
|
t = gettime_usec(); |
2116 |
|
emms(); |
2117 |
|
for(tst=0; tst<nb_tests; ++tst) csim(Ref1, Ref2, 16, lumo, lumc, devs+0, devs+1, devs+2); |
2118 |
|
emms(); |
2119 |
|
t = (gettime_usec() - t) / nb_tests; |
2120 |
|
printf("%s - ssim-consim %.3f usec devs=[0x%x 0x%x 0x%x] %s\n", |
2121 |
|
cpu->name, t, devs[0], devs[1], devs[2], |
2122 |
|
(devs[0]!=0x1bdf0f || devs[1]!=0x137258 || devs[2]!=0xcdb13)?"| ERROR": "" ); |
2123 |
|
printf( " --- \n" ); |
2124 |
|
} |
2125 |
|
} |
2126 |
|
|
2127 |
|
/********************************************************************* |
2128 |
|
* test bitstream functions |
2129 |
|
*********************************************************************/ |
2130 |
|
|
2131 |
|
#define BIT_BUF_SIZE 2000 |
2132 |
|
|
2133 |
|
static void test_bits() |
2134 |
|
{ |
2135 |
|
const int nb_tests = 50*speed_ref; |
2136 |
|
int tst; |
2137 |
|
uint32_t Crc; |
2138 |
|
uint8_t Buf[BIT_BUF_SIZE]; |
2139 |
|
uint32_t Extracted[BIT_BUF_SIZE*8]; /* worst case: bits read 1 by 1 */ |
2140 |
|
int Lens[BIT_BUF_SIZE*8]; |
2141 |
|
double t1; |
2142 |
|
|
2143 |
|
|
2144 |
|
printf( "\n === test bitstream ===\n" ); |
2145 |
|
ieee_reseed(1); |
2146 |
|
Crc = 0; |
2147 |
|
|
2148 |
|
t1 = gettime_usec(); |
2149 |
|
for(tst=0; tst<nb_tests; ++tst) { |
2150 |
|
Bitstream bs; |
2151 |
|
int m, m2, l, l2; |
2152 |
|
|
2153 |
|
for(l=0; l<BIT_BUF_SIZE; ++l) |
2154 |
|
Buf[l] = (uint8_t)ieee_rand(0,255); |
2155 |
|
|
2156 |
|
l = BIT_BUF_SIZE - ieee_rand(1,BIT_BUF_SIZE/10); |
2157 |
|
BitstreamInit(&bs, (void*)(Buf+BIT_BUF_SIZE-l), l); |
2158 |
|
|
2159 |
|
|
2160 |
|
BitstreamReset(&bs); |
2161 |
|
for(l2=l*8, m=0; l2>0; m++) { |
2162 |
|
const int b = ieee_rand(1,32); |
2163 |
|
Lens[m] = b; |
2164 |
|
l2 -= b; |
2165 |
|
if (l2<0) break; |
2166 |
|
Extracted[m] = BitstreamShowBits(&bs, b); |
2167 |
|
BitstreamSkip(&bs, b); |
2168 |
|
// printf( "<= %d: %d 0x%x\n", m, b, Extracted[m]); |
2169 |
|
} |
2170 |
|
|
2171 |
|
BitstreamReset(&bs); |
2172 |
|
for(m2=0; m2<m; ++m2) { |
2173 |
|
const int b = Lens[m2]; |
2174 |
|
const uint32_t v = BitstreamGetBits(&bs, b); |
2175 |
|
Crc |= (v!=Extracted[m2]); |
2176 |
|
// printf( "=> %d: %d 0x%x %c\n", m2, b, v, " *"[Crc]); |
2177 |
|
} |
2178 |
|
} |
2179 |
|
t1 = (gettime_usec() - t1) / nb_tests; |
2180 |
|
printf(" test_bits %.3f usec %s\n", t1, (Crc!=0)?"| ERROR": "" ); |
2181 |
|
} |
2182 |
|
|
2183 |
|
/********************************************************************* |
2184 |
* main |
* main |
2185 |
*********************************************************************/ |
*********************************************************************/ |
2186 |
|
|
2187 |
int main(int argc, char *argv[]) |
static void arg_missing(const char *opt) |
2188 |
|
{ |
2189 |
|
printf( "missing argument after option '%s'\n", opt); |
2190 |
|
exit(-1); |
2191 |
|
} |
2192 |
|
|
2193 |
|
int main(int argc, const char *argv[]) |
2194 |
|
{ |
2195 |
|
int c, what = 0; |
2196 |
|
int width, height; |
2197 |
|
uint32_t chksum = 0; |
2198 |
|
const char * test_bitstream = 0; |
2199 |
|
|
2200 |
|
cpu_mask = 0; // default => will use autodectect |
2201 |
|
for(c=1; c<argc; ++c) |
2202 |
{ |
{ |
2203 |
int what = 0; |
if (!strcmp(argv[c], "-v")) verbose++; |
2204 |
if (argc>1) what = atoi(argv[1]); |
else if (!strcmp(argv[c], "-c")) cpu_mask = 0 /* PLAIN_C */ | XVID_CPU_FORCE; |
2205 |
|
else if (!strcmp(argv[c], "-mmx")) cpu_mask = XVID_CPU_MMX | XVID_CPU_FORCE; |
2206 |
|
else if (!strcmp(argv[c], "-mmxext")) cpu_mask = XVID_CPU_MMXEXT | XVID_CPU_MMX | XVID_CPU_FORCE; |
2207 |
|
else if (!strcmp(argv[c], "-sse2")) cpu_mask = XVID_CPU_SSE2 | XVID_CPU_MMXEXT | XVID_CPU_MMX | XVID_CPU_FORCE; |
2208 |
|
else if (!strcmp(argv[c], "-sse3")) cpu_mask = XVID_CPU_SSE3 | XVID_CPU_SSE2 | XVID_CPU_MMXEXT | XVID_CPU_MMX | XVID_CPU_FORCE; |
2209 |
|
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; |
2210 |
|
else if (!strcmp(argv[c], "-3dnow")) cpu_mask = XVID_CPU_3DNOW | XVID_CPU_FORCE; |
2211 |
|
else if (!strcmp(argv[c], "-3dnowe")) cpu_mask = XVID_CPU_3DNOW | XVID_CPU_3DNOWEXT | XVID_CPU_FORCE; |
2212 |
|
else if (!strcmp(argv[c], "-altivec")) cpu_mask = XVID_CPU_ALTIVEC | XVID_CPU_FORCE; |
2213 |
|
else if (!strcmp(argv[c], "-spd")) { |
2214 |
|
if (++c==argc) arg_missing( argv[argc-1] ); |
2215 |
|
speed_ref = atoi(argv[c]); |
2216 |
|
} |
2217 |
|
else if (argv[c][0]!='-') { |
2218 |
|
what = atoi(argv[c]); |
2219 |
|
if (what==9) { |
2220 |
|
if (c+4>argc) { |
2221 |
|
printf("usage: %s %d bitstream width height (checksum)\n", argv[0], what); |
2222 |
|
exit(-1); |
2223 |
|
} |
2224 |
|
test_bitstream = argv[++c]; |
2225 |
|
width = atoi(argv[++c]); |
2226 |
|
height = atoi(argv[++c]); |
2227 |
|
if (c+1<argc && argv[c+1][0]!='-') { |
2228 |
|
if (sscanf(argv[c+1], "0x%x", &chksum)!=1) { |
2229 |
|
printf( "can't read checksum value.\n" ); |
2230 |
|
exit(-1); |
2231 |
|
} |
2232 |
|
else c++; |
2233 |
|
} |
2234 |
|
// printf( "[%s] %dx%d (0x%.8x)\n", test_bitstream, width, height, chksum); |
2235 |
|
} |
2236 |
|
} |
2237 |
|
else { |
2238 |
|
printf( "unrecognized option '%s'\n", argv[c]); |
2239 |
|
exit(-1); |
2240 |
|
} |
2241 |
|
} |
2242 |
|
|
2243 |
|
|
2244 |
if (what==0 || what==1) test_dct(); |
if (what==0 || what==1) test_dct(); |
2245 |
if (what==0 || what==2) test_mb(); |
if (what==0 || what==2) test_mb(); |
2246 |
if (what==0 || what==3) test_sad(); |
if (what==0 || what==3) test_sad(); |
2248 |
if (what==0 || what==5) test_quant(); |
if (what==0 || what==5) test_quant(); |
2249 |
if (what==0 || what==6) test_cbp(); |
if (what==0 || what==6) test_cbp(); |
2250 |
if (what==0 || what==10) test_sse(); |
if (what==0 || what==10) test_sse(); |
2251 |
|
if (what==0 || what==11) test_log2bin(); |
2252 |
|
if (what==0 || what==12) test_gcd(); |
2253 |
|
if (what==0 || what==13) test_compiler(); |
2254 |
|
if (what==0 || what==14) test_yuv(); |
2255 |
|
if (what==0 || what==15) test_SSIM(); |
2256 |
|
if (what==0 || what==16) test_yuv2(); |
2257 |
|
if (what==0 || what==17) test_bits(); |
2258 |
|
|
2259 |
if (what==7) { |
if (what==7) { |
2260 |
test_IEEE1180_compliance(-256, 255, 1); |
test_IEEE1180_compliance(-256, 255, 1); |
2266 |
} |
} |
2267 |
if (what==8) test_dct_saturation(-256, 255); |
if (what==8) test_dct_saturation(-256, 255); |
2268 |
|
|
2269 |
if (what==9) { |
if (test_bitstream) |
2270 |
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)); |
|
|
} |
|
|
|
|
2271 |
if (what==-1) { |
if (what==-1) { |
2272 |
test_dct_precision_diffs(); |
test_dct_precision_diffs(); |
2273 |
test_bugs1(); |
test_bugs1(); |
2282 |
" may appear to be slow.\n"); |
" may appear to be slow.\n"); |
2283 |
} |
} |
2284 |
|
|
2285 |
#ifdef ARCH_IS_IA32 |
#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) |
2286 |
if (what == 0 || what == 5) { |
if (what == 0 || what == 5) { |
2287 |
printf("\n" |
printf("\n" |
2288 |
"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" |
2295 |
return 0; |
return 0; |
2296 |
} |
} |
2297 |
|
|
2298 |
/********************************************************************* |
/*********************************************************************/ |
|
* '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 |
|
|
|
|
|
*********************************************************************/ |
|