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 1.36 2008-11-14 15:43:27 Isibaar Exp $ |
* $Id: xvid_bench.c,v 1.38 2008-11-26 23:37:28 Isibaar Exp $ |
23 |
* |
* |
24 |
****************************************************************************/ |
****************************************************************************/ |
25 |
|
|
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 }, |
127 |
#ifdef ARCH_IS_PPC |
#ifdef ARCH_IS_PPC |
128 |
{ "ALTIVEC", XVID_CPU_ALTIVEC }, |
{ "ALTIVEC", XVID_CPU_ALTIVEC }, |
129 |
#endif |
#endif |
|
#ifdef ARCH_IS_X86_64 |
|
|
{ "X86_64 ", XVID_CPU_ASM}, |
|
|
#endif |
|
130 |
#ifdef ARCH_IS_IA64 |
#ifdef ARCH_IS_IA64 |
131 |
// { "IA64 ", XVID_CPU_IA64 }, |
// { "IA64 ", XVID_CPU_IA64 }, |
132 |
#endif |
#endif |
413 |
|
|
414 |
t = gettime_usec(); |
t = gettime_usec(); |
415 |
emms(); |
emms(); |
416 |
|
for(tst=0; tst<nb_tests; ++tst) s = sad8bi(Cur, Ref1, Ref2, 8); |
417 |
|
emms(); |
418 |
|
t = (gettime_usec() - t) / nb_tests; |
419 |
|
printf( "%s - sad8bi %.3f usec sad=%d %s\n", |
420 |
|
cpu->name, t, s, |
421 |
|
(s!=4002)?"| ERROR": "" ); |
422 |
|
|
423 |
|
t = gettime_usec(); |
424 |
|
emms(); |
425 |
for(tst=0; tst<nb_tests; ++tst) s = dev16(Cur, 16); |
for(tst=0; tst<nb_tests; ++tst) s = dev16(Cur, 16); |
426 |
emms(); |
emms(); |
427 |
t = (gettime_usec() - t) / nb_tests; |
t = (gettime_usec() - t) / nb_tests; |
739 |
set_intra_matrix( mpeg_quant_matrices, Quant ); \ |
set_intra_matrix( mpeg_quant_matrices, Quant ); \ |
740 |
emms(); \ |
emms(); \ |
741 |
for(q=1; q<=max_Q; ++q) { \ |
for(q=1; q<=max_Q; ++q) { \ |
742 |
|
init_intra_matrix( mpeg_quant_matrices, q ); \ |
743 |
for(tst=0; tst<nb_tests; ++tst) \ |
for(tst=0; tst<nb_tests; ++tst) \ |
744 |
(FUNC)((DST), (SRC), q, q, mpeg_quant_matrices); \ |
(FUNC)((DST), (SRC), q, q, mpeg_quant_matrices); \ |
745 |
byte_swap((uint8_t*)(DST), 64*sizeof((DST)[0]), sizeof((DST)[0])); \ |
byte_swap((uint8_t*)(DST), 64*sizeof((DST)[0]), sizeof((DST)[0])); \ |
750 |
t = (gettime_usec()-t-overhead)/nb_tests/qm |
t = (gettime_usec()-t-overhead)/nb_tests/qm |
751 |
|
|
752 |
#define TEST_INTRA(REFFUNC, NEWFUNC, RANGE) \ |
#define TEST_INTRA(REFFUNC, NEWFUNC, RANGE) \ |
753 |
{ int i,q,s;\ |
{ int32_t i,q,s;\ |
754 |
DECLARE_ALIGNED_MATRIX(Src, 8, 8, int16_t, 16); \ |
DECLARE_ALIGNED_MATRIX(Src, 8, 8, int16_t, 16); \ |
755 |
DECLARE_ALIGNED_MATRIX(Dst, 8, 8, int16_t, 16); \ |
DECLARE_ALIGNED_MATRIX(Dst, 8, 8, int16_t, 16); \ |
756 |
DECLARE_ALIGNED_MATRIX(Dst2,8, 8, int16_t, 16); \ |
DECLARE_ALIGNED_MATRIX(Dst2,8, 8, int16_t, 16); \ |
782 |
|
|
783 |
void test_quant() |
void test_quant() |
784 |
{ |
{ |
785 |
const int nb_tests = 1*speed_ref; |
const int32_t nb_tests = 1*speed_ref; |
786 |
const int max_Q = 31; |
const int32_t max_Q = 31; |
787 |
DECLARE_ALIGNED_MATRIX(mpeg_quant_matrices, 8, 64, uint16_t, 16); |
DECLARE_ALIGNED_MATRIX(mpeg_quant_matrices, 8, 64, uint16_t, 16); |
788 |
|
|
789 |
int i, qm; |
int32_t i, qm; |
790 |
CPU *cpu; |
CPU *cpu; |
791 |
DECLARE_ALIGNED_MATRIX(Src, 8, 8, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Src, 8, 8, int16_t, 16); |
792 |
DECLARE_ALIGNED_MATRIX(Dst, 8, 8, int16_t, 16); |
DECLARE_ALIGNED_MATRIX(Dst, 8, 8, int16_t, 16); |
805 |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
for(cpu = cpu_list; cpu->name!=0; ++cpu) |
806 |
{ |
{ |
807 |
double t, overhead; |
double t, overhead; |
808 |
int tst, q; |
int32_t tst, q; |
809 |
uint32_t s; |
uint32_t s; |
810 |
|
|
811 |
if (!init_cpu(cpu)) |
if (!init_cpu(cpu)) |
829 |
TEST_QUANT2(quant_mpeg_intra, Dst, Src); |
TEST_QUANT2(quant_mpeg_intra, Dst, Src); |
830 |
printf("%s - quant_mpeg_intra %.3f usec crc32=0x%08x %s\n", |
printf("%s - quant_mpeg_intra %.3f usec crc32=0x%08x %s\n", |
831 |
cpu->name, t, s, |
cpu->name, t, s, |
832 |
(s!=0xfd6a21a4)? "| ERROR": ""); |
(s!=0x3b999af6)? "| ERROR": ""); |
833 |
|
|
834 |
TEST_QUANT(quant_mpeg_inter, Dst, Src); |
TEST_QUANT(quant_mpeg_inter, Dst, Src); |
835 |
printf("%s - quant_mpeg_inter %.3f usec crc32=0x%08x %s\n", |
printf("%s - quant_mpeg_inter %.3f usec crc32=0x%08x %s\n", |
1730 |
t = (gettime_usec() - t) / nb_tests; \ |
t = (gettime_usec() - t) / nb_tests; \ |
1731 |
iCrc = calc_crc((uint8_t*)Dst0, sizeof(Dst0), CRC32_INITIAL) |
iCrc = calc_crc((uint8_t*)Dst0, sizeof(Dst0), CRC32_INITIAL) |
1732 |
|
|
1733 |
#define TEST_YUYV(FUNC, S) \ |
#define TEST_YUYV(FUNC, S, FLIP) \ |
1734 |
ENTER \ |
ENTER \ |
1735 |
for(tst=0; tst<nb_tests; ++tst) (FUNC)(Dst0[0], S*WIDTH, Src0[0], Src0[1], Src0[2], WIDTH, WIDTH/2, WIDTH, HEIGHT, 0); \ |
for(tst=0; tst<nb_tests; ++tst) (FUNC)(Dst0[0], S*WIDTH, Src0[0], Src0[1], Src0[2], WIDTH, WIDTH/2, WIDTH, HEIGHT, (FLIP)); \ |
1736 |
LEAVE |
LEAVE |
1737 |
|
|
1738 |
|
static const int yuv_CRCs[6][2] = { |
1739 |
|
{0x0f4fb96b,0x780b6a68} |
1740 |
|
, {0xa986b289,0x65e49b76} |
1741 |
|
, {0x7f19c152,0xd539b86e} |
1742 |
|
, {0x0f4fb96b,0x780b6a68} |
1743 |
|
, {0xa986b289,0x65e49b76} |
1744 |
|
, {0x36ab8b57,0x1cd92fee} |
1745 |
|
}; |
1746 |
|
|
1747 |
#define WIDTH 128 |
#define WIDTH 128 |
1748 |
#define HEIGHT 32 |
#define HEIGHT 32 |
1749 |
void test_yuv() |
void test_yuv() |
1752 |
CPU *cpu; |
CPU *cpu; |
1753 |
uint8_t Src0[3][WIDTH*HEIGHT]; |
uint8_t Src0[3][WIDTH*HEIGHT]; |
1754 |
uint8_t Dst0[4][WIDTH*HEIGHT]; |
uint8_t Dst0[4][WIDTH*HEIGHT]; |
1755 |
int i, j; |
int i, j, with_flip; |
1756 |
double t; |
double t; |
1757 |
int tst, iCrc; |
int tst, iCrc; |
1758 |
|
|
1763 |
|
|
1764 |
printf( "\n === test YUV ===\n" ); |
printf( "\n === test YUV ===\n" ); |
1765 |
|
|
1766 |
|
for(with_flip=0; with_flip<=1; ++with_flip) { |
1767 |
|
|
1768 |
init_cpu(&cpu_list[0]); |
init_cpu(&cpu_list[0]); |
1769 |
TEST_YUYV(yv12_to_yuyv_c, 4); |
TEST_YUYV(yv12_to_yuyv_c, 4, with_flip); |
1770 |
printf(" yv12_to_yuyv_c %.3f usec crc32=0x%08x %s\n", |
printf(" yv12_to_yuyv_c %.3f usec crc32=0x%08x %s\n", |
1771 |
t, iCrc, (iCrc!=0xeb1a0b0a)?"| ERROR": "" ); |
t, iCrc, (iCrc!=yuv_CRCs[0][with_flip])?"| ERROR": "" ); |
1772 |
TEST_YUYV(yv12_to_uyvy_c, 4); |
TEST_YUYV(yv12_to_uyvy_c, 4, with_flip); |
1773 |
printf(" yv12_to_uyvy_c %.3f usec crc32=0x%08x %s\n", |
printf(" yv12_to_uyvy_c %.3f usec crc32=0x%08x %s\n", |
1774 |
t, iCrc, (iCrc!=0x6e82f55b)?"| ERROR": "" ); |
t, iCrc, (iCrc!=yuv_CRCs[1][with_flip])?"| ERROR": "" ); |
1775 |
|
|
1776 |
|
TEST_YUYV(yv12_to_bgra_c, 4, with_flip); |
1777 |
|
printf(" yv12_to_bgra_c %.3f usec crc32=0x%08x %s\n", |
1778 |
|
t, iCrc, (iCrc!=yuv_CRCs[2][with_flip])?"| ERROR": "" ); |
1779 |
|
|
1780 |
#ifdef ARCH_IS_IA32 |
#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) |
1781 |
init_cpu(&cpu_list[1]); |
init_cpu(&cpu_list[1]); |
1782 |
TEST_YUYV(yv12_to_yuyv_mmx, 4); |
TEST_YUYV(yv12_to_yuyv_mmx, 4, with_flip); |
1783 |
printf(" yv12_to_yuyv_mmx %.3f usec crc32=0x%08x %s\n", |
printf(" yv12_to_yuyv_mmx %.3f usec crc32=0x%08x %s\n", |
1784 |
t, iCrc, (iCrc!=0xeb1a0b0a)?"| ERROR": "" ); |
t, iCrc, (iCrc!=yuv_CRCs[3][with_flip])?"| ERROR": "" ); |
1785 |
|
|
1786 |
TEST_YUYV(yv12_to_uyvy_mmx, 4); |
TEST_YUYV(yv12_to_uyvy_mmx, 4, with_flip); |
1787 |
printf(" yv12_to_uyvy_mmx %.3f usec crc32=0x%08x %s\n", |
printf(" yv12_to_uyvy_mmx %.3f usec crc32=0x%08x %s\n", |
1788 |
t, iCrc, (iCrc!=0x6e82f55b)?"| ERROR": "" ); |
t, iCrc, (iCrc!=yuv_CRCs[4][with_flip])?"| ERROR": "" ); |
1789 |
|
|
1790 |
|
TEST_YUYV(yv12_to_bgra_mmx, 4, with_flip); |
1791 |
|
printf(" yv12_to_bgra_mmx %.3f usec crc32=0x%08x %s\n", |
1792 |
|
t, iCrc, (iCrc!=yuv_CRCs[5][with_flip])?"| ERROR": "" ); |
1793 |
|
|
1794 |
#endif |
#endif |
1795 |
|
|
1796 |
#ifdef ARCH_IS_PPC |
#ifdef ARCH_IS_PPC |
1797 |
init_cpu(&cpu_list[1]); |
init_cpu(&cpu_list[1]); |
1798 |
TEST_YUYV(yv12_to_yuyv_altivec_c, 4); |
TEST_YUYV(yv12_to_yuyv_altivec_c, 4, with_flip); |
1799 |
printf(" yv12_to_yuyv_altivec_c %.3f usec crc32=0x%08x %s\n", |
printf(" yv12_to_yuyv_altivec_c %.3f usec crc32=0x%08x %s\n", |
1800 |
t, iCrc, (iCrc!=0xeb1a0b0a)?"| ERROR": "" ); |
t, iCrc, (iCrc!=yuv_CRCs[3][with_flip])?"| ERROR": "" ); |
1801 |
|
|
1802 |
TEST_YUYV(yv12_to_uyvy_altivec_c, 4); |
TEST_YUYV(yv12_to_uyvy_altivec_c, 4, with_flip); |
1803 |
printf(" yv12_to_uyvy_altivec_c %.3f usec crc32=0x%08x %s\n", |
printf(" yv12_to_uyvy_altivec_c %.3f usec crc32=0x%08x %s\n", |
1804 |
t, iCrc, (iCrc!=0x6e82f55b)?"| ERROR": "" ); |
t, iCrc, (iCrc!=yuv_CRCs[4][with_flip])?"| ERROR": "" ); |
1805 |
|
|
1806 |
|
TEST_YUYV(yv12_to_bgra_altivec_c, 4, with_flip); |
1807 |
|
printf(" yv12_to_bgra_altivec_c %.3f usec crc32=0x%08x %s\n", |
1808 |
|
t, iCrc, (iCrc!=yuv_CRCs[5][with_flip])?"| ERROR": "" ); |
1809 |
#endif |
#endif |
1810 |
|
} |
1811 |
printf( " --- \n" ); |
printf( " --- \n" ); |
1812 |
} |
} |
1813 |
|
|
1856 |
t, iCrc, (iCrc!=yv12_CRCs[with_flip][with_uv])?"| ERROR": "" ); |
t, iCrc, (iCrc!=yv12_CRCs[with_flip][with_uv])?"| ERROR": "" ); |
1857 |
/* if (!with_uv) PRINT_NxN(Dst0[1], WIDTH/2, HEIGHT/2, WIDTH ); */ |
/* if (!with_uv) PRINT_NxN(Dst0[1], WIDTH/2, HEIGHT/2, WIDTH ); */ |
1858 |
|
|
1859 |
#ifdef ARCH_IS_IA32 |
#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) |
1860 |
init_cpu(&cpu_list[1]); |
init_cpu(&cpu_list[1]); |
1861 |
TEST_YV2(yv12_to_yv12_mmx, with_uv, with_flip); |
TEST_YV2(yv12_to_yv12_mmx, with_uv, with_flip); |
1862 |
printf(" yv12_to_yv12_mmx %.3f usec \tcrc32=0x%08x %s\n", |
printf(" yv12_to_yv12_mmx %.3f usec \tcrc32=0x%08x %s\n", |
2094 |
lum8x8 = lum_8x8_c; |
lum8x8 = lum_8x8_c; |
2095 |
lum2x8 = lum_2x8_c; |
lum2x8 = lum_2x8_c; |
2096 |
csim = consim_c; |
csim = consim_c; |
2097 |
#ifdef ARCH_IS_IA32 |
#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) |
2098 |
if (cpu->cpu & XVID_CPU_MMX){ |
if (cpu->cpu & XVID_CPU_MMX){ |
2099 |
lum8x8 = lum_8x8_mmx; |
lum8x8 = lum_8x8_mmx; |
2100 |
csim = consim_mmx; |
csim = consim_mmx; |
2291 |
" may appear to be slow.\n"); |
" may appear to be slow.\n"); |
2292 |
} |
} |
2293 |
|
|
2294 |
#ifdef ARCH_IS_IA32 |
#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) |
2295 |
if (what == 0 || what == 5) { |
if (what == 0 || what == 5) { |
2296 |
printf("\n" |
printf("\n" |
2297 |
"NB: MMX mpeg4 quantization is known to have very small errors (+/-1 magnitude)\n" |
"NB: MMX mpeg4 quantization is known to have very small errors (+/-1 magnitude)\n" |