3 |
* XVID MPEG-4 VIDEO CODEC |
* XVID MPEG-4 VIDEO CODEC |
4 |
* - Native API implementation - |
* - Native API implementation - |
5 |
* |
* |
|
* This program is an implementation of a part of one or more MPEG-4 |
|
|
* Video tools as specified in ISO/IEC 14496-2 standard. Those intending |
|
|
* to use this software module in hardware or software products are |
|
|
* advised that its use may infringe existing patents or copyrights, and |
|
|
* any such use would be at such party's own risk. The original |
|
|
* developer of this software module and his/her company, and subsequent |
|
|
* editors and their companies, will have no liability for use of this |
|
|
* software or modifications or derivatives thereof. |
|
|
* |
|
6 |
* This program is free software ; you can redistribute it and/or modify |
* This program is free software ; you can redistribute it and/or modify |
7 |
* it under the terms of the GNU General Public License as published by |
* it under the terms of the GNU General Public License as published by |
8 |
* the Free Software Foundation ; either version 2 of the License, or |
* the Free Software Foundation ; either version 2 of the License, or |
17 |
* along with this program ; if not, write to the Free Software |
* along with this program ; if not, write to the Free Software |
18 |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
19 |
* |
* |
|
****************************************************************************/ |
|
|
|
|
|
/***************************************************************************** |
|
|
* |
|
|
* History |
|
|
* |
|
|
* - 23.06.2002 added XVID_CPU_CHKONLY |
|
|
* - 17.03.2002 Added interpolate8x8_halfpel_hv_xmm |
|
|
* - 22.12.2001 API change: added xvid_init() - Isibaar |
|
|
* - 16.12.2001 inital version; (c)2001 peter ross <pross@cs.rmit.edu.au> |
|
|
* |
|
20 |
* $Id$ |
* $Id$ |
21 |
* |
* |
22 |
****************************************************************************/ |
****************************************************************************/ |
45 |
#include "utils/timer.h" |
#include "utils/timer.h" |
46 |
#include "bitstream/mbcoding.h" |
#include "bitstream/mbcoding.h" |
47 |
|
|
48 |
#if defined(ARCH_X86) && defined(EXPERIMENTAL_SSE2_CODE) |
#if defined(ARCH_IS_IA32) |
49 |
|
|
50 |
#ifdef WIN32 |
#if defined(_MSC_VER) |
51 |
#include <windows.h> |
#include <windows.h> |
52 |
#else |
#else |
53 |
#include <signal.h> |
#include <signal.h> |
54 |
#include <setjmp.h> |
#include <setjmp.h> |
|
#endif |
|
|
|
|
|
|
|
|
#ifndef WIN32 |
|
55 |
|
|
56 |
static jmp_buf mark; |
static jmp_buf mark; |
57 |
|
|
64 |
|
|
65 |
|
|
66 |
/* |
/* |
67 |
calls the funcptr, and returns whether SIGILL (illegal instruction) was signalled |
* Calls the funcptr, and returns whether SIGILL (illegal instruction) was |
68 |
return values: |
* signalled |
69 |
-1 : could not determine |
* |
70 |
0 : SIGILL was *not* signalled |
* Return values: |
71 |
1 : SIGILL was signalled |
* -1 : could not determine |
72 |
|
* 0 : SIGILL was *not* signalled |
73 |
|
* 1 : SIGILL was signalled |
74 |
*/ |
*/ |
75 |
|
|
76 |
int |
int |
77 |
sigill_check(void (*func)()) |
sigill_check(void (*func)()) |
78 |
{ |
{ |
79 |
#ifdef WIN32 |
#if defined(_MSC_VER) |
80 |
_try { |
_try { |
81 |
func(); |
func(); |
82 |
} |
} |
110 |
} |
} |
111 |
#endif |
#endif |
112 |
|
|
113 |
|
|
114 |
|
/* detect cpu flags */ |
115 |
|
static unsigned int |
116 |
|
detect_cpu_flags() |
117 |
|
{ |
118 |
|
/* enable native assembly optimizations by default */ |
119 |
|
unsigned int cpu_flags = XVID_CPU_ASM; |
120 |
|
|
121 |
|
#if defined(ARCH_IS_IA32) |
122 |
|
cpu_flags |= check_cpu_features(); |
123 |
|
if ((cpu_flags & XVID_CPU_SSE) && sigill_check(sse_os_trigger)) |
124 |
|
cpu_flags &= ~XVID_CPU_SSE; |
125 |
|
|
126 |
|
if ((cpu_flags & XVID_CPU_SSE2) && sigill_check(sse2_os_trigger)) |
127 |
|
cpu_flags &= ~XVID_CPU_SSE2; |
128 |
|
#endif |
129 |
|
|
130 |
|
#if defined(ARCH_IS_PPC) |
131 |
|
#if defined(ARCH_IS_PPC_ALTIVEC) |
132 |
|
cpu_flags |= XVID_CPU_ALTIVEC; |
133 |
|
#endif |
134 |
|
#endif |
135 |
|
|
136 |
|
return cpu_flags; |
137 |
|
} |
138 |
|
|
139 |
|
|
140 |
/***************************************************************************** |
/***************************************************************************** |
141 |
* XviD Init Entry point |
* XviD Init Entry point |
142 |
* |
* |
170 |
|
|
171 |
} else { |
} else { |
172 |
|
|
173 |
cpu_flags = check_cpu_features(); |
cpu_flags = detect_cpu_flags(); |
|
|
|
|
#if defined(ARCH_X86) && defined(EXPERIMENTAL_SSE2_CODE) |
|
|
if ((cpu_flags & XVID_CPU_SSE) && sigill_check(sse_os_trigger)) |
|
|
cpu_flags &= ~XVID_CPU_SSE; |
|
|
|
|
|
if ((cpu_flags & XVID_CPU_SSE2) && sigill_check(sse2_os_trigger)) |
|
|
cpu_flags &= ~XVID_CPU_SSE2; |
|
|
#endif |
|
174 |
} |
} |
175 |
|
|
176 |
if ((init_param->cpu_flags & XVID_CPU_CHKONLY)) |
if ((init_param->cpu_flags & XVID_CPU_CHKONLY)) |
211 |
transfer_8to16copy = transfer_8to16copy_c; |
transfer_8to16copy = transfer_8to16copy_c; |
212 |
transfer_16to8copy = transfer_16to8copy_c; |
transfer_16to8copy = transfer_16to8copy_c; |
213 |
transfer_8to16sub = transfer_8to16sub_c; |
transfer_8to16sub = transfer_8to16sub_c; |
214 |
|
transfer_8to16subro = transfer_8to16subro_c; |
215 |
transfer_8to16sub2 = transfer_8to16sub2_c; |
transfer_8to16sub2 = transfer_8to16sub2_c; |
216 |
transfer_16to8add = transfer_16to8add_c; |
transfer_16to8add = transfer_16to8add_c; |
217 |
transfer8x8_copy = transfer8x8_copy_c; |
transfer8x8_copy = transfer8x8_copy_c; |
239 |
interpolate8x8_avg4 = interpolate8x8_avg4_c; |
interpolate8x8_avg4 = interpolate8x8_avg4_c; |
240 |
|
|
241 |
/* reduced resoltuion */ |
/* reduced resoltuion */ |
|
|
|
242 |
copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_C; |
copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_C; |
243 |
add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_C; |
add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_C; |
|
#ifdef ARCH_X86 |
|
|
vfilter_31 = xvid_VFilter_31_x86; |
|
|
hfilter_31 = xvid_HFilter_31_x86; |
|
|
#else |
|
244 |
vfilter_31 = xvid_VFilter_31_C; |
vfilter_31 = xvid_VFilter_31_C; |
245 |
hfilter_31 = xvid_HFilter_31_C; |
hfilter_31 = xvid_HFilter_31_C; |
|
#endif |
|
246 |
filter_18x18_to_8x8 = xvid_Filter_18x18_To_8x8_C; |
filter_18x18_to_8x8 = xvid_Filter_18x18_To_8x8_C; |
247 |
filter_diff_18x18_to_8x8 = xvid_Filter_Diff_18x18_To_8x8_C; |
filter_diff_18x18_to_8x8 = xvid_Filter_Diff_18x18_To_8x8_C; |
248 |
|
|
298 |
dev16 = dev16_c; |
dev16 = dev16_c; |
299 |
sad16v = sad16v_c; |
sad16v = sad16v_c; |
300 |
|
|
301 |
// Halfpel8_Refine = Halfpel8_Refine_c; |
/* Halfpel8_Refine = Halfpel8_Refine_c; */ |
302 |
|
|
303 |
#ifdef ARCH_X86 |
#if defined(ARCH_IS_IA32) |
304 |
|
|
305 |
|
if ((cpu_flags & XVID_CPU_ASM)) |
306 |
|
{ |
307 |
|
vfilter_31 = xvid_VFilter_31_x86; |
308 |
|
hfilter_31 = xvid_HFilter_31_x86; |
309 |
|
} |
310 |
|
|
311 |
if ((cpu_flags & XVID_CPU_MMX) || (cpu_flags & XVID_CPU_MMXEXT) || |
if ((cpu_flags & XVID_CPU_MMX) || (cpu_flags & XVID_CPU_MMXEXT) || |
312 |
(cpu_flags & XVID_CPU_3DNOW) || (cpu_flags & XVID_CPU_3DNOWEXT) || |
(cpu_flags & XVID_CPU_3DNOW) || (cpu_flags & XVID_CPU_3DNOWEXT) || |
316 |
emms = emms_mmx; |
emms = emms_mmx; |
317 |
} |
} |
318 |
|
|
319 |
if ((cpu_flags & XVID_CPU_MMX) > 0) { |
if ((cpu_flags & XVID_CPU_MMX)) { |
320 |
|
|
321 |
/* Forward and Inverse Discrete Cosine Transformation functions */ |
/* Forward and Inverse Discrete Cosine Transformation functions */ |
322 |
fdct = fdct_mmx; |
fdct = fdct_mmx; |
337 |
transfer_8to16copy = transfer_8to16copy_mmx; |
transfer_8to16copy = transfer_8to16copy_mmx; |
338 |
transfer_16to8copy = transfer_16to8copy_mmx; |
transfer_16to8copy = transfer_16to8copy_mmx; |
339 |
transfer_8to16sub = transfer_8to16sub_mmx; |
transfer_8to16sub = transfer_8to16sub_mmx; |
340 |
|
transfer_8to16subro = transfer_8to16subro_mmx; |
341 |
transfer_8to16sub2 = transfer_8to16sub2_mmx; |
transfer_8to16sub2 = transfer_8to16sub2_mmx; |
342 |
transfer_16to8add = transfer_16to8add_mmx; |
transfer_16to8add = transfer_16to8add_mmx; |
343 |
transfer8x8_copy = transfer8x8_copy_mmx; |
transfer8x8_copy = transfer8x8_copy_mmx; |
387 |
sad8bi = sad8bi_mmx; |
sad8bi = sad8bi_mmx; |
388 |
dev16 = dev16_mmx; |
dev16 = dev16_mmx; |
389 |
sad16v = sad16v_mmx; |
sad16v = sad16v_mmx; |
|
|
|
390 |
} |
} |
391 |
|
|
392 |
/* these 3dnow functions are faster than mmx, but slower than xmm. */ |
/* these 3dnow functions are faster than mmx, but slower than xmm. */ |
393 |
if ((cpu_flags & XVID_CPU_3DNOW) > 0) { |
if ((cpu_flags & XVID_CPU_3DNOW)) { |
394 |
|
|
395 |
|
emms = emms_3dn; |
396 |
|
|
397 |
/* ME functions */ |
/* ME functions */ |
398 |
sad16bi = sad16bi_3dn; |
sad16bi = sad16bi_3dn; |
403 |
} |
} |
404 |
|
|
405 |
|
|
406 |
if ((cpu_flags & XVID_CPU_MMXEXT) > 0) { |
if ((cpu_flags & XVID_CPU_MMXEXT)) { |
407 |
|
|
408 |
/* Inverse DCT */ |
/* Inverse DCT */ |
409 |
idct = idct_xmm; |
idct = idct_xmm; |
441 |
sad16v = sad16v_xmm; |
sad16v = sad16v_xmm; |
442 |
} |
} |
443 |
|
|
444 |
if ((cpu_flags & XVID_CPU_3DNOW) > 0) { |
if ((cpu_flags & XVID_CPU_3DNOW)) { |
445 |
|
|
446 |
/* Interpolation */ |
/* Interpolation */ |
447 |
interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_3dn; |
interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_3dn; |
449 |
interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dn; |
interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dn; |
450 |
} |
} |
451 |
|
|
452 |
if ((cpu_flags & XVID_CPU_3DNOWEXT) > 0) { |
if ((cpu_flags & XVID_CPU_3DNOWEXT)) { |
453 |
|
|
454 |
/* Inverse DCT */ |
/* Inverse DCT */ |
455 |
idct = idct_3dne; |
idct = idct_3dne; |
458 |
transfer_8to16copy = transfer_8to16copy_3dne; |
transfer_8to16copy = transfer_8to16copy_3dne; |
459 |
transfer_16to8copy = transfer_16to8copy_3dne; |
transfer_16to8copy = transfer_16to8copy_3dne; |
460 |
transfer_8to16sub = transfer_8to16sub_3dne; |
transfer_8to16sub = transfer_8to16sub_3dne; |
461 |
|
transfer_8to16subro = transfer_8to16subro_3dne; |
462 |
transfer_8to16sub2 = transfer_8to16sub2_3dne; |
transfer_8to16sub2 = transfer_8to16sub2_3dne; |
463 |
transfer_16to8add = transfer_16to8add_3dne; |
transfer_16to8add = transfer_16to8add_3dne; |
464 |
transfer8x8_copy = transfer8x8_copy_3dne; |
transfer8x8_copy = transfer8x8_copy_3dne; |
486 |
} |
} |
487 |
|
|
488 |
|
|
489 |
if ((cpu_flags & XVID_CPU_SSE2) > 0) { |
if ((cpu_flags & XVID_CPU_SSE2)) { |
|
#ifdef EXPERIMENTAL_SSE2_CODE |
|
490 |
|
|
491 |
calc_cbp = calc_cbp_sse2; |
calc_cbp = calc_cbp_sse2; |
492 |
|
|
496 |
quant_inter = quant_inter_sse2; |
quant_inter = quant_inter_sse2; |
497 |
dequant_inter = dequant_inter_sse2; |
dequant_inter = dequant_inter_sse2; |
498 |
|
|
499 |
/* ME */ |
#if defined(EXPERIMENTAL_SSE2_CODE) |
500 |
|
/* ME; slower than xmm */ |
501 |
sad16 = sad16_sse2; |
sad16 = sad16_sse2; |
502 |
dev16 = dev16_sse2; |
dev16 = dev16_sse2; |
503 |
|
#endif |
504 |
/* Forward and Inverse DCT */ |
/* Forward and Inverse DCT */ |
505 |
idct = idct_sse2; |
idct = idct_sse2; |
506 |
fdct = fdct_sse2; |
fdct = fdct_sse2; |
|
#endif |
|
507 |
} |
} |
|
|
|
508 |
#endif |
#endif |
509 |
|
|
510 |
#ifdef ARCH_IA64 |
#if defined(ARCH_IS_IA64) |
511 |
if ((cpu_flags & XVID_CPU_IA64) > 0) { //use assembler routines? |
if ((cpu_flags & XVID_CPU_ASM)) { /* use assembler routines? */ |
512 |
idct_ia64_init(); |
idct_ia64_init(); |
513 |
fdct = fdct_ia64; |
fdct = fdct_ia64; |
514 |
idct = idct_ia64; //not yet working, crashes |
idct = idct_ia64; /*not yet working, crashes */ |
515 |
interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_ia64; |
interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_ia64; |
516 |
interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_ia64; |
interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_ia64; |
517 |
interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_ia64; |
interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_ia64; |
519 |
sad16bi = sad16bi_ia64; |
sad16bi = sad16bi_ia64; |
520 |
sad8 = sad8_ia64; |
sad8 = sad8_ia64; |
521 |
dev16 = dev16_ia64; |
dev16 = dev16_ia64; |
522 |
// Halfpel8_Refine = Halfpel8_Refine_ia64; |
/* Halfpel8_Refine = Halfpel8_Refine_ia64; */ |
523 |
quant_intra = quant_intra_ia64; |
quant_intra = quant_intra_ia64; |
524 |
dequant_intra = dequant_intra_ia64; |
dequant_intra = dequant_intra_ia64; |
525 |
quant_inter = quant_inter_ia64; |
quant_inter = quant_inter_ia64; |
530 |
transfer_8to16sub2 = transfer_8to16sub2_ia64; |
transfer_8to16sub2 = transfer_8to16sub2_ia64; |
531 |
transfer_16to8add = transfer_16to8add_ia64; |
transfer_16to8add = transfer_16to8add_ia64; |
532 |
transfer8x8_copy = transfer8x8_copy_ia64; |
transfer8x8_copy = transfer8x8_copy_ia64; |
533 |
DEBUG("Using IA-64 assembler routines.\n"); |
DPRINTF(DPRINTF_DEBUG, "Using IA-64 assembler routines."); |
534 |
} |
} |
535 |
#endif |
#endif |
536 |
|
|
537 |
#ifdef ARCH_PPC |
#if defined(ARCH_IS_PPC) |
538 |
#ifdef ARCH_PPC_ALTIVEC |
if ((cpu_flags & XVID_CPU_ASM)) |
539 |
|
{ |
540 |
|
calc_cbp = calc_cbp_ppc; |
541 |
|
} |
542 |
|
|
543 |
|
if ((cpu_flags & XVID_CPU_ALTIVEC)) |
544 |
|
{ |
545 |
calc_cbp = calc_cbp_altivec; |
calc_cbp = calc_cbp_altivec; |
546 |
fdct = fdct_altivec; |
fdct = fdct_altivec; |
547 |
idct = idct_altivec; |
idct = idct_altivec; |
549 |
sad16 = sad16_altivec; |
sad16 = sad16_altivec; |
550 |
sad8 = sad8_altivec; |
sad8 = sad8_altivec; |
551 |
dev16 = dev16_altivec; |
dev16 = dev16_altivec; |
552 |
#else |
} |
|
calc_cbp = calc_cbp_ppc; |
|
|
#endif |
|
553 |
#endif |
#endif |
554 |
|
|
555 |
return XVID_ERR_OK; |
return XVID_ERR_OK; |
560 |
static int |
static int |
561 |
xvid_init_convert(XVID_INIT_CONVERTINFO* convert) |
xvid_init_convert(XVID_INIT_CONVERTINFO* convert) |
562 |
{ |
{ |
563 |
// const int flip1 = (convert->input.colorspace & XVID_CSP_VFLIP) ^ (convert->output.colorspace & XVID_CSP_VFLIP); |
/* |
564 |
|
const int flip1 = |
565 |
|
(convert->input.colorspace & XVID_CSP_VFLIP) ^ |
566 |
|
(convert->output.colorspace & XVID_CSP_VFLIP); |
567 |
|
*/ |
568 |
const int width = convert->width; |
const int width = convert->width; |
569 |
const int height = convert->height; |
const int height = convert->height; |
570 |
const int width2 = convert->width/2; |
const int width2 = convert->width/2; |
650 |
#define TEST_FDCT (TEST_FORWARD) |
#define TEST_FDCT (TEST_FORWARD) |
651 |
#define TEST_IDCT (0) |
#define TEST_IDCT (0) |
652 |
|
|
653 |
int test_transform(void * funcA, void * funcB, const char * nameB, |
static int test_transform(void * funcA, void * funcB, const char * nameB, |
654 |
int test, int flags) |
int test, int flags) |
655 |
{ |
{ |
656 |
int i; |
int i; |
730 |
#define TEST_DEQUANT_INTRA (TEST_INTRA) |
#define TEST_DEQUANT_INTRA (TEST_INTRA) |
731 |
#define TEST_DEQUANT_INTER (0) |
#define TEST_DEQUANT_INTER (0) |
732 |
|
|
733 |
int test_quant(void * funcA, void * funcB, const char * nameB, |
static int test_quant(void * funcA, void * funcB, const char * nameB, |
734 |
int test, int flags) |
int test, int flags) |
735 |
{ |
{ |
736 |
int q,i; |
int q,i; |
737 |
int64_t timeSTART; |
int64_t timeSTART; |
738 |
int64_t timeA = 0; |
int64_t timeA = 0; |
739 |
int64_t timeB = 0; |
int64_t timeB = 0; |
740 |
int retA, retB; |
int retA = 0, retB = 0; |
741 |
DECLARE_ALIGNED_MATRIX(arrayX, 1, 64, int16_t, CACHE_LINE); |
DECLARE_ALIGNED_MATRIX(arrayX, 1, 64, int16_t, CACHE_LINE); |
742 |
DECLARE_ALIGNED_MATRIX(arrayA, 1, 64, int16_t, CACHE_LINE); |
DECLARE_ALIGNED_MATRIX(arrayA, 1, 64, int16_t, CACHE_LINE); |
743 |
DECLARE_ALIGNED_MATRIX(arrayB, 1, 64, int16_t, CACHE_LINE); |
DECLARE_ALIGNED_MATRIX(arrayB, 1, 64, int16_t, CACHE_LINE); |
817 |
|
|
818 |
printf("xvid_init_test\n"); |
printf("xvid_init_test\n"); |
819 |
|
|
820 |
#if defined(ARCH_X86) |
#if defined(ARCH_IS_IA32) |
821 |
cpu_flags = check_cpu_features(); |
cpu_flags = detect_cpu_flags(); |
822 |
idct_int32_init(); |
idct_int32_init(); |
823 |
emms_mmx(); |
emms_mmx(); |
824 |
|
|