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 1.19 2005-05-23 09:29:43 Skal Exp $ |
* $Id: xvid_bench.c,v 1.23 2005-08-05 20:49:23 Skal Exp $ |
23 |
* |
* |
24 |
****************************************************************************/ |
****************************************************************************/ |
25 |
|
|
112 |
unsigned int cpu; |
unsigned int cpu; |
113 |
} CPU; |
} CPU; |
114 |
|
|
115 |
CPU cpu_list[] = |
CPU cpu_list[] = { |
116 |
{ { "PLAINC", 0 } |
{ "PLAINC ", 0 }, |
117 |
#ifdef ARCH_IS_IA32 |
#ifdef ARCH_IS_IA32 |
118 |
, { "MMX ", XVID_CPU_MMX } |
{ "MMX ", XVID_CPU_MMX }, |
119 |
, { "MMXEXT", XVID_CPU_MMXEXT | XVID_CPU_MMX } |
{ "MMXEXT ", XVID_CPU_MMXEXT | XVID_CPU_MMX }, |
120 |
, { "SSE2 ", XVID_CPU_SSE2 | XVID_CPU_MMX } |
{ "SSE2 ", XVID_CPU_SSE2 | XVID_CPU_MMX }, |
121 |
, { "3DNOW ", XVID_CPU_3DNOW } |
{ "3DNOW ", XVID_CPU_3DNOW }, |
122 |
, { "3DNOWE", XVID_CPU_3DNOW | XVID_CPU_3DNOWEXT } |
{ "3DNOWE ", XVID_CPU_3DNOW | XVID_CPU_3DNOWEXT }, |
123 |
#endif |
#endif |
124 |
#ifdef ARCH_IS_PPC |
#ifdef ARCH_IS_PPC |
125 |
, { "ALTIVEC", XVID_CPU_ALTIVEC } |
{ "ALTIVEC", XVID_CPU_ALTIVEC }, |
126 |
#endif |
#endif |
127 |
#ifdef ARCH_IS_X86_64 |
#ifdef ARCH_IS_X86_64 |
128 |
, { "X86_64", XVID_CPU_ASM} |
{ "X86_64 ", XVID_CPU_ASM}, |
129 |
#endif |
#endif |
130 |
//, { "IA64 ", XVID_CPU_IA64 } |
#ifdef ARCH_IS_IA64 |
131 |
//, { "TSC ", XVID_CPU_TSC } |
// { "IA64 ", XVID_CPU_IA64 }, |
|
, { 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 } |
|
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) |
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 |
/********************************************************************* |
/********************************************************************* |
302 |
* test DCT |
* test DCT |
303 |
*********************************************************************/ |
*********************************************************************/ |
551 |
} \ |
} \ |
552 |
emms(); \ |
emms(); \ |
553 |
t = (gettime_usec()-t -overhead) / nb_tests;\ |
t = (gettime_usec()-t -overhead) / nb_tests;\ |
554 |
s = calc_crc((uint8_t*)(DST), sizeof((DST)), CRC32_INITIAL) |
byte_swap((uint8_t*)(DST), 8*32*sizeof((DST)[0]), sizeof((DST)[0])); \ |
555 |
|
s = calc_crc((uint8_t*)(DST), 8*32*sizeof((DST)[0]), CRC32_INITIAL) |
556 |
|
|
557 |
#define TEST_TRANSFER(FUNC, DST, SRC) \ |
#define TEST_TRANSFER(FUNC, DST, SRC) \ |
558 |
TEST_TRANSFER_BEGIN(DST); \ |
TEST_TRANSFER_BEGIN(DST); \ |
578 |
} \ |
} \ |
579 |
emms(); \ |
emms(); \ |
580 |
t = (gettime_usec()-t -overhead) / nb_tests;\ |
t = (gettime_usec()-t -overhead) / nb_tests;\ |
581 |
s = calc_crc((uint8_t*)(DST), sizeof((DST)), CRC32_INITIAL) |
byte_swap((uint8_t*)(DST), 8*32*sizeof((DST)[0]), sizeof((DST)[0])); \ |
582 |
|
s = calc_crc((uint8_t*)(DST), 8*32*sizeof((DST)[0]), CRC32_INITIAL) |
583 |
|
|
584 |
#define TEST_TRANSFER2(FUNC, DST, SRC, R1) \ |
#define TEST_TRANSFER2(FUNC, DST, SRC, R1) \ |
585 |
TEST_TRANSFER2_BEGIN(DST,SRC); \ |
TEST_TRANSFER2_BEGIN(DST,SRC); \ |
596 |
const int nb_tests = 4000*speed_ref; |
const int nb_tests = 4000*speed_ref; |
597 |
int i; |
int i; |
598 |
CPU *cpu; |
CPU *cpu; |
599 |
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]; |
600 |
int16_t Src16[8*32], Dst16[8*32]; |
// int16_t Src16[8*32], Dst16[8*32]; |
601 |
|
DECLARE_ALIGNED_MATRIX(Src8, 8, 32, uint8_t, CACHE_LINE); |
602 |
|
DECLARE_ALIGNED_MATRIX(Dst8, 8, 32, uint8_t, CACHE_LINE); |
603 |
|
DECLARE_ALIGNED_MATRIX(Ref1, 8, 32, uint8_t, CACHE_LINE); |
604 |
|
DECLARE_ALIGNED_MATRIX(Ref2, 8, 32, uint8_t, CACHE_LINE); |
605 |
|
DECLARE_ALIGNED_MATRIX(Src16, 8, 32, uint16_t, CACHE_LINE); |
606 |
|
DECLARE_ALIGNED_MATRIX(Dst16, 8, 32, uint16_t, CACHE_LINE); |
607 |
|
|
608 |
printf( "\n === test transfer ===\n" ); |
printf( "\n === test transfer ===\n" ); |
609 |
|
|
638 |
TEST_TRANSFER2(transfer_8to16sub, Dst16, Src8, Ref1); |
TEST_TRANSFER2(transfer_8to16sub, Dst16, Src8, Ref1); |
639 |
{ |
{ |
640 |
int s1, s2; |
int s1, s2; |
641 |
s1 = calc_crc((uint8_t*)Dst16, sizeof(Dst16), CRC32_INITIAL); |
s1 = calc_crc((uint8_t*)Dst16, 8*32*sizeof(Dst16[0]), CRC32_INITIAL); |
642 |
s2 = calc_crc((uint8_t*)Src8, sizeof(Src8), CRC32_INITIAL); |
s2 = calc_crc((uint8_t*)Src8, 8*32*sizeof(Src8[0]), CRC32_INITIAL); |
643 |
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", |
644 |
cpu->name, t, s1, s2, |
cpu->name, t, s1, s2, |
645 |
(s1!=0xa1e07163)?"| ERROR1": "", |
(s1!=0xa1e07163)?"| ERROR1": "", |
668 |
for(q=1; q<=max_Q; ++q) { \ |
for(q=1; q<=max_Q; ++q) { \ |
669 |
for(tst=0; tst<nb_tests; ++tst) \ |
for(tst=0; tst<nb_tests; ++tst) \ |
670 |
(FUNC)((DST), (SRC), q, mpeg_quant_matrices); \ |
(FUNC)((DST), (SRC), q, mpeg_quant_matrices); \ |
671 |
s = calc_crc((uint8_t*)(DST), 64*sizeof(int16_t), s); \ |
byte_swap((uint8_t*)(DST), 64*sizeof((DST)[0]), sizeof((DST)[0])); \ |
672 |
|
s = calc_crc((uint8_t*)(DST), 64*sizeof((DST)[0]), s); \ |
673 |
} \ |
} \ |
674 |
emms(); \ |
emms(); \ |
675 |
} \ |
} \ |
684 |
for(q=1; q<=max_Q; ++q) { \ |
for(q=1; q<=max_Q; ++q) { \ |
685 |
for(tst=0; tst<nb_tests; ++tst) \ |
for(tst=0; tst<nb_tests; ++tst) \ |
686 |
(FUNC)((DST), (SRC), q, q, mpeg_quant_matrices); \ |
(FUNC)((DST), (SRC), q, q, mpeg_quant_matrices); \ |
687 |
s = calc_crc((uint8_t*)(DST), 64*sizeof(int16_t), s); \ |
byte_swap((uint8_t*)(DST), 64*sizeof((DST)[0]), sizeof((DST)[0])); \ |
688 |
|
s = calc_crc((uint8_t*)(DST), 64*sizeof((DST)[0]), s); \ |
689 |
} \ |
} \ |
690 |
emms(); \ |
emms(); \ |
691 |
} \ |
} \ |
905 |
* test non-zero AC counting |
* test non-zero AC counting |
906 |
*********************************************************************/ |
*********************************************************************/ |
907 |
|
|
908 |
#define TEST_CBP(FUNC, SRC) \ |
#define TEST_CBP(FUNC, SRC, NB) \ |
909 |
t = gettime_usec(); \ |
t = gettime_usec(); \ |
910 |
emms(); \ |
emms(); \ |
911 |
for(tst=0; tst<nb_tests; ++tst) { \ |
for(tst=0; tst<NB; ++tst) { \ |
912 |
cbp = (FUNC)((SRC)); \ |
cbp = (FUNC)((SRC)); \ |
913 |
} \ |
} \ |
914 |
emms(); \ |
emms(); \ |
917 |
void test_cbp() |
void test_cbp() |
918 |
{ |
{ |
919 |
const int nb_tests = 10000*speed_ref; |
const int nb_tests = 10000*speed_ref; |
920 |
int i; |
int i, n, m; |
921 |
CPU *cpu; |
CPU *cpu; |
922 |
DECLARE_ALIGNED_MATRIX(Src1, 6, 64, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Src1, 6, 64, int16_t, 16); |
923 |
DECLARE_ALIGNED_MATRIX(Src2, 6, 64, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Src2, 6, 64, int16_t, 16); |
941 |
if (!init_cpu(cpu)) |
if (!init_cpu(cpu)) |
942 |
continue; |
continue; |
943 |
|
|
944 |
TEST_CBP(calc_cbp, Src1); |
TEST_CBP(calc_cbp, Src1, nb_tests); |
945 |
printf("%s - calc_cbp#1 %.3f usec cbp=0x%02x %s\n", |
printf("%s - calc_cbp#1 %.3f usec cbp=0x%02x %s\n", |
946 |
cpu->name, t, cbp, (cbp!=0x15)?"| ERROR": ""); |
cpu->name, t, cbp, (cbp!=0x15)?"| ERROR": ""); |
947 |
TEST_CBP(calc_cbp, Src2); |
TEST_CBP(calc_cbp, Src2, nb_tests); |
948 |
printf("%s - calc_cbp#2 %.3f usec cbp=0x%02x %s\n", |
printf("%s - calc_cbp#2 %.3f usec cbp=0x%02x %s\n", |
949 |
cpu->name, t, cbp, (cbp!=0x38)?"| ERROR": ""); |
cpu->name, t, cbp, (cbp!=0x38)?"| ERROR": ""); |
950 |
TEST_CBP(calc_cbp, Src3); |
TEST_CBP(calc_cbp, Src3, nb_tests); |
951 |
printf("%s - calc_cbp#3 %.3f usec cbp=0x%02x %s\n", |
printf("%s - calc_cbp#3 %.3f usec cbp=0x%02x %s\n", |
952 |
cpu->name, t, cbp, (cbp!=0x0f)?"| ERROR": "" ); |
cpu->name, t, cbp, (cbp!=0x0f)?"| ERROR": "" ); |
953 |
TEST_CBP(calc_cbp, Src4); |
TEST_CBP(calc_cbp, Src4, nb_tests); |
954 |
printf("%s - calc_cbp#4 %.3f usec cbp=0x%02x %s\n", |
printf("%s - calc_cbp#4 %.3f usec cbp=0x%02x %s\n", |
955 |
cpu->name, t, cbp, (cbp!=0x05)?"| ERROR": "" ); |
cpu->name, t, cbp, (cbp!=0x05)?"| ERROR": "" ); |
956 |
printf( " --- \n" ); |
printf( " --- \n" ); |
957 |
} |
} |
958 |
|
|
959 |
|
for(cpu = cpu_list; cpu->name!=0; ++cpu) /* bench suggested by Carlo (carlo dot bramix at libero dot it) */ |
960 |
|
{ |
961 |
|
double t; |
962 |
|
int tst, cbp, err; |
963 |
|
|
964 |
|
if (!init_cpu(cpu)) |
965 |
|
continue; |
966 |
|
|
967 |
|
err = 0; |
968 |
|
for(n=0; n<6; ++n) |
969 |
|
{ |
970 |
|
for(m=0; m<64; ++m) |
971 |
|
{ |
972 |
|
for(i=0; i<6*64; ++i) |
973 |
|
Src1[i] = (i== (m + n*64)); |
974 |
|
|
975 |
|
TEST_CBP(calc_cbp, Src1, 1); |
976 |
|
if (cbp!= (((m!=0)<<(5-n)))) |
977 |
|
{ |
978 |
|
printf( "%s - calc_cbp#5: ERROR at pos %d / %d!\n", cpu->name, n, m); |
979 |
|
err = 1; |
980 |
|
break; |
981 |
|
} |
982 |
|
} |
983 |
|
} |
984 |
|
if (!err) |
985 |
|
printf( " %s - calc_cbp#5 : OK\n", cpu->name ); |
986 |
|
|
987 |
|
} |
988 |
} |
} |
989 |
|
|
990 |
/********************************************************************* |
/********************************************************************* |
1797 |
else if (!strcmp(argv[c], "-c")) cpu_mask = 0 /* PLAIN_C */ | XVID_CPU_FORCE; |
else if (!strcmp(argv[c], "-c")) cpu_mask = 0 /* PLAIN_C */ | XVID_CPU_FORCE; |
1798 |
else if (!strcmp(argv[c], "-mmx")) cpu_mask = XVID_CPU_MMX | XVID_CPU_FORCE; |
else if (!strcmp(argv[c], "-mmx")) cpu_mask = XVID_CPU_MMX | XVID_CPU_FORCE; |
1799 |
else if (!strcmp(argv[c], "-mmxext")) cpu_mask = XVID_CPU_MMXEXT | XVID_CPU_MMX | XVID_CPU_FORCE; |
else if (!strcmp(argv[c], "-mmxext")) cpu_mask = XVID_CPU_MMXEXT | XVID_CPU_MMX | XVID_CPU_FORCE; |
1800 |
else if (!strcmp(argv[c], "-sse2")) cpu_mask = XVID_CPU_SSE2 | XVID_CPU_MMX | XVID_CPU_FORCE; |
else if (!strcmp(argv[c], "-sse2")) cpu_mask = XVID_CPU_SSE2 | XVID_CPU_MMXEXT | XVID_CPU_MMX | XVID_CPU_FORCE; |
1801 |
else if (!strcmp(argv[c], "-3dnow")) cpu_mask = XVID_CPU_3DNOW | XVID_CPU_FORCE; |
else if (!strcmp(argv[c], "-3dnow")) cpu_mask = XVID_CPU_3DNOW | XVID_CPU_FORCE; |
1802 |
else if (!strcmp(argv[c], "-3dnowe")) cpu_mask = XVID_CPU_3DNOW | XVID_CPU_3DNOWEXT | XVID_CPU_FORCE; |
else if (!strcmp(argv[c], "-3dnowe")) cpu_mask = XVID_CPU_3DNOW | XVID_CPU_3DNOWEXT | XVID_CPU_FORCE; |
1803 |
else if (!strcmp(argv[c], "-altivec")) cpu_mask = XVID_CPU_ALTIVEC | XVID_CPU_FORCE; |
else if (!strcmp(argv[c], "-altivec")) cpu_mask = XVID_CPU_ALTIVEC | XVID_CPU_FORCE; |