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 1.18 2005-05-17 21:03:32 Skal Exp $ |
* $Id: xvid_bench.c,v 1.25 2005-09-23 12:53:35 suxen_drol 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); |
924 |
DECLARE_ALIGNED_MATRIX(Src3, 6, 64, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Src3, 6, 64, int16_t, 16); |
925 |
DECLARE_ALIGNED_MATRIX(Src4, 6, 64, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Src4, 6, 64, int16_t, 16); |
926 |
|
DECLARE_ALIGNED_MATRIX(Src5, 6, 64, int16_t, 16); |
927 |
|
|
928 |
printf( "\n ===== test cbp =====\n" ); |
printf( "\n ===== test cbp =====\n" ); |
929 |
|
|
932 |
Src2[i] = (i<3*64); /* half-full */ |
Src2[i] = (i<3*64); /* half-full */ |
933 |
Src3[i] = ((i+32)>3*64); |
Src3[i] = ((i+32)>3*64); |
934 |
Src4[i] = (i==(3*64+2) || i==(5*64+9)); |
Src4[i] = (i==(3*64+2) || i==(5*64+9)); |
935 |
|
Src5[i] = ieee_rand(0,1) ? -1 : 1; /* +/- test */ |
936 |
} |
} |
937 |
|
|
938 |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
943 |
if (!init_cpu(cpu)) |
if (!init_cpu(cpu)) |
944 |
continue; |
continue; |
945 |
|
|
946 |
TEST_CBP(calc_cbp, Src1); |
TEST_CBP(calc_cbp, Src1, nb_tests); |
947 |
printf("%s - calc_cbp#1 %.3f usec cbp=0x%02x %s\n", |
printf("%s - calc_cbp#1 %.3f usec cbp=0x%02x %s\n", |
948 |
cpu->name, t, cbp, (cbp!=0x15)?"| ERROR": ""); |
cpu->name, t, cbp, (cbp!=0x15)?"| ERROR": ""); |
949 |
TEST_CBP(calc_cbp, Src2); |
TEST_CBP(calc_cbp, Src2, nb_tests); |
950 |
printf("%s - calc_cbp#2 %.3f usec cbp=0x%02x %s\n", |
printf("%s - calc_cbp#2 %.3f usec cbp=0x%02x %s\n", |
951 |
cpu->name, t, cbp, (cbp!=0x38)?"| ERROR": ""); |
cpu->name, t, cbp, (cbp!=0x38)?"| ERROR": ""); |
952 |
TEST_CBP(calc_cbp, Src3); |
TEST_CBP(calc_cbp, Src3, nb_tests); |
953 |
printf("%s - calc_cbp#3 %.3f usec cbp=0x%02x %s\n", |
printf("%s - calc_cbp#3 %.3f usec cbp=0x%02x %s\n", |
954 |
cpu->name, t, cbp, (cbp!=0x0f)?"| ERROR": "" ); |
cpu->name, t, cbp, (cbp!=0x0f)?"| ERROR": "" ); |
955 |
TEST_CBP(calc_cbp, Src4); |
TEST_CBP(calc_cbp, Src4, nb_tests); |
956 |
printf("%s - calc_cbp#4 %.3f usec cbp=0x%02x %s\n", |
printf("%s - calc_cbp#4 %.3f usec cbp=0x%02x %s\n", |
957 |
cpu->name, t, cbp, (cbp!=0x05)?"| ERROR": "" ); |
cpu->name, t, cbp, (cbp!=0x05)?"| ERROR": "" ); |
958 |
|
TEST_CBP(calc_cbp, Src5, nb_tests); |
959 |
|
printf("%s - calc_cbp#4 %.3f usec cbp=0x%02x %s\n", |
960 |
|
cpu->name, t, cbp, (cbp!=0x3f)?"| ERROR": "" ); |
961 |
printf( " --- \n" ); |
printf( " --- \n" ); |
962 |
} |
} |
963 |
|
|
964 |
|
for(cpu = cpu_list; cpu->name!=0; ++cpu) /* bench suggested by Carlo (carlo dot bramix at libero dot it) */ |
965 |
|
{ |
966 |
|
double t; |
967 |
|
int tst, cbp, err; |
968 |
|
|
969 |
|
if (!init_cpu(cpu)) |
970 |
|
continue; |
971 |
|
|
972 |
|
err = 0; |
973 |
|
for(n=0; n<6; ++n) |
974 |
|
{ |
975 |
|
for(m=0; m<64; ++m) |
976 |
|
{ |
977 |
|
for(i=0; i<6*64; ++i) |
978 |
|
Src1[i] = (i== (m + n*64)); |
979 |
|
|
980 |
|
TEST_CBP(calc_cbp, Src1, 1); |
981 |
|
if (cbp!= (((m!=0)<<(5-n)))) |
982 |
|
{ |
983 |
|
printf( "%s - calc_cbp#5: ERROR at pos %d / %d!\n", cpu->name, n, m); |
984 |
|
err = 1; |
985 |
|
break; |
986 |
|
} |
987 |
|
} |
988 |
|
} |
989 |
|
if (!err) |
990 |
|
printf( " %s - calc_cbp#5 : OK\n", cpu->name ); |
991 |
|
|
992 |
|
} |
993 |
} |
} |
994 |
|
|
995 |
/********************************************************************* |
/********************************************************************* |
1452 |
xframe.bitstream = buf + pos; |
xframe.bitstream = buf + pos; |
1453 |
xframe.length = buf_size - pos; |
xframe.length = buf_size - pos; |
1454 |
xframe.output.plane[0] = (uint8_t*)(((size_t)yuv_out + 15) & ~15); |
xframe.output.plane[0] = (uint8_t*)(((size_t)yuv_out + 15) & ~15); |
1455 |
xframe.output.plane[1] = xframe.output.plane[0] + bps*height; |
xframe.output.plane[1] = (uint8_t*)xframe.output.plane[0] + bps*height; |
1456 |
xframe.output.plane[2] = xframe.output.plane[1] + bps/2; |
xframe.output.plane[2] = (uint8_t*)xframe.output.plane[1] + bps/2; |
1457 |
xframe.output.stride[0] = bps; |
xframe.output.stride[0] = bps; |
1458 |
xframe.output.stride[1] = bps; |
xframe.output.stride[1] = bps; |
1459 |
xframe.output.stride[2] = bps; |
xframe.output.stride[2] = bps; |
1471 |
nb++; |
nb++; |
1472 |
|
|
1473 |
for(y=0; y<height/2; ++y) { |
for(y=0; y<height/2; ++y) { |
1474 |
chksum = calc_crc(xframe.output.plane[0] + (2*y+0)*bps, width, chksum); |
chksum = calc_crc((uint8_t*)xframe.output.plane[0] + (2*y+0)*bps, width, chksum); |
1475 |
chksum = calc_crc(xframe.output.plane[0] + (2*y+1)*bps, width, chksum); |
chksum = calc_crc((uint8_t*)xframe.output.plane[0] + (2*y+1)*bps, width, chksum); |
1476 |
chksum = calc_crc(xframe.output.plane[1] + y*bps, width/2, chksum); |
chksum = calc_crc((uint8_t*)xframe.output.plane[1] + y*bps, width/2, chksum); |
1477 |
chksum = calc_crc(xframe.output.plane[2] + y*bps, width/2, chksum); |
chksum = calc_crc((uint8_t*)xframe.output.plane[2] + y*bps, width/2, chksum); |
1478 |
} |
} |
1479 |
if (pos==buf_size) |
if (pos==buf_size) |
1480 |
break; |
break; |
1669 |
return n; |
return n; |
1670 |
} |
} |
1671 |
|
|
1672 |
static const uint8_t log2_tab_16[256] = { 0, 1, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4 }; |
static const uint8_t log2_tab_16[16] = { 0, 1, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4 }; |
1673 |
|
|
1674 |
static uint32_t __inline log2bin_v2(uint32_t value) |
static uint32_t __inline log2bin_v2(uint32_t value) |
1675 |
{ |
{ |
1706 |
for(s=s0, n=0; n<nb_tests; ++n, s=(s*12363+31)&0x7fffffff) |
for(s=s0, n=0; n<nb_tests; ++n, s=(s*12363+31)&0x7fffffff) |
1707 |
crc2 += log2bin_v2(s); |
crc2 += log2bin_v2(s); |
1708 |
t2 = (gettime_usec() - t2) / nb_tests; |
t2 = (gettime_usec() - t2) / nb_tests; |
1709 |
|
|
1710 |
printf( "log2bin_v1: %.3f sec crc=%d\n", t1, crc1 ); |
printf( "log2bin_v1: %.3f sec crc=%d\n", t1, crc1 ); |
1711 |
printf( "log2bin_v2: %.3f sec crc=%d\n", t2, crc2 ); |
printf( "log2bin_v2: %.3f sec crc=%d\n", t2, crc2 ); |
1712 |
if (crc1!=crc2) printf( " CRC ERROR !\n" ); |
if (crc1!=crc2) printf( " CRC ERROR !\n" ); |
1713 |
} |
} |
1714 |
|
|
1715 |
|
/*********************************************************************/ |
1716 |
|
|
1717 |
|
static void __inline old_gcd(int *num, int *den) |
1718 |
|
{ |
1719 |
|
int i = *num; |
1720 |
|
while (i > 1) { |
1721 |
|
if (*num % i == 0 && *den % i == 0) { |
1722 |
|
*num /= i; |
1723 |
|
*den /= i; |
1724 |
|
i = *num; |
1725 |
|
continue; |
1726 |
|
} |
1727 |
|
i--; |
1728 |
|
} |
1729 |
|
} |
1730 |
|
|
1731 |
|
static uint32_t gcd(int num, int den) |
1732 |
|
{ |
1733 |
|
int tmp; |
1734 |
|
while( (tmp=num%den) ) { num = den; den = tmp; } |
1735 |
|
return den; |
1736 |
|
} |
1737 |
|
static void __inline new_gcd(int *num, int *den) |
1738 |
|
{ |
1739 |
|
const int div = gcd(*num, *den); |
1740 |
|
if (num) { |
1741 |
|
*num /= div; |
1742 |
|
*den /= div; |
1743 |
|
} |
1744 |
|
} |
1745 |
|
|
1746 |
|
void test_gcd() |
1747 |
|
{ |
1748 |
|
const int nb_tests = 10*speed_ref; |
1749 |
|
int i; |
1750 |
|
uint32_t crc1=0, crc2=0; |
1751 |
|
uint32_t n0, n, d0, d; |
1752 |
|
double t1, t2; |
1753 |
|
|
1754 |
|
t1 = gettime_usec(); |
1755 |
|
n0 = 0xfffff & (int)(t1*31.241); |
1756 |
|
d0 = 0xfffff & (int)( ((n0*4123)%17) | 1 ); |
1757 |
|
for(n=n0, d=d0, i=0; i<nb_tests; ++i) { |
1758 |
|
old_gcd(&n, &d); |
1759 |
|
crc1 = (((crc1>>4)^d) + ((crc1<<2)^n) ) & 0xffffff; |
1760 |
|
n = d; |
1761 |
|
d = (d*12363+31) & 0xffff; |
1762 |
|
d |= !d; |
1763 |
|
} |
1764 |
|
t1 = (gettime_usec()-t1) / nb_tests; |
1765 |
|
|
1766 |
|
t2 = gettime_usec(); |
1767 |
|
for(n=n0, d=d0, i=0; i<nb_tests; ++i) { |
1768 |
|
new_gcd(&n, &d); |
1769 |
|
crc2 = (((crc2>>4)^d) + ((crc2<<2)^n) ) & 0xffffff; |
1770 |
|
n = d; |
1771 |
|
d = (d*12363+31) & 0xffff; |
1772 |
|
d |= !d; |
1773 |
|
} |
1774 |
|
t2 = (gettime_usec() - t2) / nb_tests; |
1775 |
|
|
1776 |
|
printf( "old_gcd: %.3f sec crc=%d\n", t1, crc1 ); |
1777 |
|
printf( "new_gcd: %.3f sec crc=%d\n", t2, crc2 ); |
1778 |
|
if (crc1!=crc2) printf( " CRC ERROR !\n" ); |
1779 |
|
} |
1780 |
|
|
1781 |
/********************************************************************* |
/********************************************************************* |
1782 |
* main |
* main |
1783 |
*********************************************************************/ |
*********************************************************************/ |
1802 |
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; |
1803 |
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; |
1804 |
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; |
1805 |
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; |
1806 |
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; |
1807 |
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; |
1808 |
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; |
1845 |
if (what==0 || what==6) test_cbp(); |
if (what==0 || what==6) test_cbp(); |
1846 |
if (what==0 || what==10) test_sse(); |
if (what==0 || what==10) test_sse(); |
1847 |
if (what==0 || what==11) test_log2bin(); |
if (what==0 || what==11) test_log2bin(); |
1848 |
|
if (what==0 || what==12) test_gcd(); |
1849 |
|
|
1850 |
|
|
1851 |
if (what==7) { |
if (what==7) { |
1887 |
return 0; |
return 0; |
1888 |
} |
} |
1889 |
|
|
1890 |
/********************************************************************* |
/*********************************************************************/ |
|
* '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 |
|
|
|
|
|
*********************************************************************/ |
|