3 |
* XVID MPEG-4 VIDEO CODEC |
* XVID MPEG-4 VIDEO CODEC |
4 |
* - Native API implementation - |
* - Native API implementation - |
5 |
* |
* |
|
* Copyright(C) 2001-2003 Peter Ross <pross@xvid.org> |
|
|
* |
|
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 |
37 |
#include "image/reduced.h" |
#include "image/reduced.h" |
38 |
#include "utils/mem_transfer.h" |
#include "utils/mem_transfer.h" |
39 |
#include "utils/mbfunctions.h" |
#include "utils/mbfunctions.h" |
40 |
#include "quant/quant.h" |
#include "quant/quant_h263.h" |
41 |
|
#include "quant/quant_mpeg4.h" |
42 |
#include "motion/motion.h" |
#include "motion/motion.h" |
43 |
#include "motion/sad.h" |
#include "motion/sad.h" |
44 |
#include "utils/emms.h" |
#include "utils/emms.h" |
45 |
#include "utils/timer.h" |
#include "utils/timer.h" |
46 |
#include "bitstream/mbcoding.h" |
#include "bitstream/mbcoding.h" |
|
#include "image/qpel.h" |
|
|
|
|
|
#if defined(_DEBUG) |
|
|
unsigned int xvid_debug = 0; /* xvid debug mask */ |
|
|
#endif |
|
47 |
|
|
48 |
#if defined(ARCH_IS_IA32) |
#if defined(ARCH_IS_IA32) |
49 |
|
|
50 |
#if defined(_MSC_VER) |
#if defined(_MSC_VER) |
51 |
# include <windows.h> |
# include <windows.h> |
52 |
#else |
#else |
153 |
|
|
154 |
|
|
155 |
static |
static |
156 |
int xvid_gbl_init(xvid_gbl_init_t * init) |
int xvid_init_init(XVID_INIT_PARAM * init_param) |
157 |
{ |
{ |
158 |
unsigned int cpu_flags; |
int cpu_flags; |
159 |
|
|
160 |
|
/* Inform the client the API version */ |
161 |
|
init_param->api_version = API_VERSION; |
162 |
|
|
163 |
|
/* Inform the client the core build - unused because we're still alpha */ |
164 |
|
init_param->core_build = 1000; |
165 |
|
|
166 |
if (XVID_VERSION_MAJOR(init->version) != 1) /* v1.x.x */ |
/* Do we have to force CPU features ? */ |
167 |
return XVID_ERR_VERSION; |
if ((init_param->cpu_flags & XVID_CPU_FORCE)) { |
168 |
|
|
169 |
|
cpu_flags = init_param->cpu_flags; |
170 |
|
|
171 |
|
} else { |
172 |
|
|
173 |
|
cpu_flags = detect_cpu_flags(); |
174 |
|
} |
175 |
|
|
176 |
|
if ((init_param->cpu_flags & XVID_CPU_CHKONLY)) |
177 |
|
{ |
178 |
|
init_param->cpu_flags = cpu_flags; |
179 |
|
return XVID_ERR_OK; |
180 |
|
} |
181 |
|
|
182 |
|
init_param->cpu_flags = cpu_flags; |
183 |
|
|
|
cpu_flags = (init->cpu_flags & XVID_CPU_FORCE) ? init->cpu_flags : detect_cpu_flags(); |
|
184 |
|
|
185 |
/* Initialize the function pointers */ |
/* Initialize the function pointers */ |
186 |
idct_int32_init(); |
idct_int32_init(); |
196 |
/* Restore FPU context : emms_c is a nop functions */ |
/* Restore FPU context : emms_c is a nop functions */ |
197 |
emms = emms_c; |
emms = emms_c; |
198 |
|
|
|
/* Qpel stuff */ |
|
|
xvid_QP_Funcs = &xvid_QP_Funcs_C; |
|
|
xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_C; |
|
|
xvid_Init_QP(); |
|
|
|
|
199 |
/* Quantization functions */ |
/* Quantization functions */ |
200 |
quant_h263_intra = quant_h263_intra_c; |
quant_intra = quant_intra_c; |
201 |
quant_h263_inter = quant_h263_inter_c; |
dequant_intra = dequant_intra_c; |
202 |
dequant_h263_intra = dequant_h263_intra_c; |
quant_inter = quant_inter_c; |
203 |
dequant_h263_inter = dequant_h263_inter_c; |
dequant_inter = dequant_inter_c; |
204 |
|
|
205 |
quant_mpeg_intra = quant_mpeg_intra_c; |
quant4_intra = quant4_intra_c; |
206 |
quant_mpeg_inter = quant_mpeg_inter_c; |
dequant4_intra = dequant4_intra_c; |
207 |
dequant_mpeg_intra = dequant_mpeg_intra_c; |
quant4_inter = quant4_inter_c; |
208 |
dequant_mpeg_inter = dequant_mpeg_inter_c; |
dequant4_inter = dequant4_inter_c; |
209 |
|
|
210 |
/* Block transfer related functions */ |
/* Block transfer related functions */ |
211 |
transfer_8to16copy = transfer_8to16copy_c; |
transfer_8to16copy = transfer_8to16copy_c; |
238 |
interpolate8x8_avg2 = interpolate8x8_avg2_c; |
interpolate8x8_avg2 = interpolate8x8_avg2_c; |
239 |
interpolate8x8_avg4 = interpolate8x8_avg4_c; |
interpolate8x8_avg4 = interpolate8x8_avg4_c; |
240 |
|
|
241 |
/* reduced resolution */ |
/* 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; |
244 |
vfilter_31 = xvid_VFilter_31_C; |
vfilter_31 = xvid_VFilter_31_C; |
302 |
|
|
303 |
#if defined(ARCH_IS_IA32) |
#if defined(ARCH_IS_IA32) |
304 |
|
|
305 |
if ((cpu_flags & XVID_CPU_ASM)) { |
if ((cpu_flags & XVID_CPU_ASM)) |
306 |
|
{ |
307 |
vfilter_31 = xvid_VFilter_31_x86; |
vfilter_31 = xvid_VFilter_31_x86; |
308 |
hfilter_31 = xvid_HFilter_31_x86; |
hfilter_31 = xvid_HFilter_31_x86; |
309 |
} |
} |
319 |
if ((cpu_flags & XVID_CPU_MMX)) { |
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_skal; |
fdct = fdct_mmx; |
323 |
idct = idct_mmx; |
idct = simple_idct_mmx; /* use simple idct by default */ |
|
|
|
|
/* Qpel stuff */ |
|
|
xvid_QP_Funcs = &xvid_QP_Funcs_mmx; |
|
|
xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_mmx; |
|
324 |
|
|
325 |
/* Quantization related functions */ |
/* Quantization related functions */ |
326 |
quant_h263_intra = quant_h263_intra_mmx; |
quant_intra = quant_intra_mmx; |
327 |
quant_h263_inter = quant_h263_inter_mmx; |
dequant_intra = dequant_intra_mmx; |
328 |
dequant_h263_intra = dequant_h263_intra_mmx; |
quant_inter = quant_inter_mmx; |
329 |
dequant_h263_inter = dequant_h263_inter_mmx; |
dequant_inter = dequant_inter_mmx; |
330 |
|
|
331 |
quant_mpeg_intra = quant_mpeg_intra_mmx; |
quant4_intra = quant4_intra_mmx; |
332 |
quant_mpeg_inter = quant_mpeg_inter_mmx; |
dequant4_intra = dequant4_intra_mmx; |
333 |
dequant_mpeg_intra = dequant_mpeg_intra_mmx; |
quant4_inter = quant4_inter_mmx; |
334 |
dequant_mpeg_inter = dequant_mpeg_inter_mmx; |
dequant4_inter = dequant4_inter_mmx; |
335 |
|
|
336 |
/* Block related functions */ |
/* Block related functions */ |
337 |
transfer_8to16copy = transfer_8to16copy_mmx; |
transfer_8to16copy = transfer_8to16copy_mmx; |
405 |
|
|
406 |
if ((cpu_flags & XVID_CPU_MMXEXT)) { |
if ((cpu_flags & XVID_CPU_MMXEXT)) { |
407 |
|
|
408 |
/* DCT */ |
/* Inverse DCT */ |
409 |
fdct = fdct_xmm_skal; |
/* idct = idct_xmm; Don't use Walken idct anymore! */ |
|
idct = idct_xmm; |
|
410 |
|
|
411 |
/* Interpolation */ |
/* Interpolation */ |
412 |
interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_xmm; |
interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_xmm; |
418 |
add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_xmm; |
add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_xmm; |
419 |
|
|
420 |
/* Quantization */ |
/* Quantization */ |
421 |
quant_mpeg_intra = quant_mpeg_intra_xmm; |
quant4_intra = quant4_intra_xmm; |
422 |
quant_mpeg_inter = quant_mpeg_inter_xmm; |
quant4_inter = quant4_inter_xmm; |
423 |
|
|
424 |
dequant_h263_intra = dequant_h263_intra_xmm; |
dequant_intra = dequant_intra_xmm; |
425 |
dequant_h263_inter = dequant_h263_inter_xmm; |
dequant_inter = dequant_inter_xmm; |
426 |
|
|
427 |
/* Buffer transfer */ |
/* Buffer transfer */ |
428 |
transfer_8to16sub2 = transfer_8to16sub2_xmm; |
transfer_8to16sub2 = transfer_8to16sub2_xmm; |
452 |
if ((cpu_flags & XVID_CPU_3DNOWEXT)) { |
if ((cpu_flags & XVID_CPU_3DNOWEXT)) { |
453 |
|
|
454 |
/* Inverse DCT */ |
/* Inverse DCT */ |
455 |
idct = idct_3dne; |
/* idct = idct_3dne; Don't use Walken idct anymore */ |
456 |
|
|
457 |
/* Buffer transfer */ |
/* Buffer transfer */ |
458 |
transfer_8to16copy = transfer_8to16copy_3dne; |
transfer_8to16copy = transfer_8to16copy_3dne; |
464 |
transfer8x8_copy = transfer8x8_copy_3dne; |
transfer8x8_copy = transfer8x8_copy_3dne; |
465 |
|
|
466 |
/* Quantization */ |
/* Quantization */ |
467 |
quant_h263_intra = quant_h263_intra_3dne; |
dequant4_intra = dequant4_intra_3dne; |
468 |
quant_h263_inter = quant_h263_inter_3dne; |
dequant4_inter = dequant4_inter_3dne; |
469 |
dequant_mpeg_intra = dequant_mpeg_intra_3dne; |
quant_intra = quant_intra_3dne; |
470 |
dequant_mpeg_inter = dequant_mpeg_inter_3dne; |
quant_inter = quant_inter_3dne; |
471 |
dequant_h263_intra = dequant_h263_intra_3dne; |
dequant_intra = dequant_intra_3dne; |
472 |
dequant_h263_inter = dequant_h263_inter_3dne; |
dequant_inter = dequant_inter_3dne; |
473 |
|
|
474 |
/* ME functions */ |
/* ME functions */ |
475 |
calc_cbp = calc_cbp_3dne; |
calc_cbp = calc_cbp_3dne; |
485 |
interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dne; |
interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dne; |
486 |
} |
} |
487 |
|
|
488 |
#if defined(EXPERIMENTAL_SSE2_CODE) /* mark the whole SSE2 stuff as experimental. At least on |
|
|
my P4, it crashes... */ |
|
489 |
if ((cpu_flags & XVID_CPU_SSE2)) { |
if ((cpu_flags & XVID_CPU_SSE2)) { |
490 |
|
|
491 |
calc_cbp = calc_cbp_sse2; |
calc_cbp = calc_cbp_sse2; |
492 |
|
|
493 |
/* Quantization */ |
/* Quantization */ |
494 |
quant_h263_intra = quant_h263_intra_sse2; |
quant_intra = quant_intra_sse2; |
495 |
quant_h263_inter = quant_h263_inter_sse2; |
dequant_intra = dequant_intra_sse2; |
496 |
dequant_h263_intra = dequant_h263_intra_sse2; |
quant_inter = quant_inter_sse2; |
497 |
dequant_h263_inter = dequant_h263_inter_sse2; |
dequant_inter = dequant_inter_sse2; |
498 |
|
|
499 |
|
#if defined(EXPERIMENTAL_SSE2_CODE) |
500 |
/* ME; slower than xmm */ |
/* ME; slower than xmm */ |
501 |
sad16 = sad16_sse2; |
sad16 = sad16_sse2; |
502 |
dev16 = dev16_sse2; |
dev16 = dev16_sse2; |
|
} |
|
503 |
#endif |
#endif |
504 |
|
/* Forward and Inverse DCT */ |
505 |
|
/* idct = idct_sse2; |
506 |
|
/* fdct = fdct_sse2; Both are none to be unprecise - better deactivate for now */ |
507 |
|
} |
508 |
#endif |
#endif |
509 |
|
|
510 |
#if defined(ARCH_IS_IA64) |
#if defined(ARCH_IS_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_h263_intra = quant_h263_intra_ia64; |
quant_intra = quant_intra_ia64; |
524 |
quant_h263_inter = quant_h263_inter_ia64; |
dequant_intra = dequant_intra_ia64; |
525 |
dequant_h263_intra = dequant_h263_intra_ia64; |
quant_inter = quant_inter_ia64; |
526 |
dequant_h263_inter = dequant_h263_inter_ia64; |
dequant_inter = dequant_inter_ia64; |
527 |
transfer_8to16copy = transfer_8to16copy_ia64; |
transfer_8to16copy = transfer_8to16copy_ia64; |
528 |
transfer_16to8copy = transfer_16to8copy_ia64; |
transfer_16to8copy = transfer_16to8copy_ia64; |
529 |
transfer_8to16sub = transfer_8to16sub_ia64; |
transfer_8to16sub = transfer_8to16sub_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 |
|
DPRINTF(DPRINTF_DEBUG, "Using IA-64 assembler routines."); |
534 |
} |
} |
535 |
#endif |
#endif |
536 |
|
|
552 |
} |
} |
553 |
#endif |
#endif |
554 |
|
|
555 |
#if defined(_DEBUG) |
return XVID_ERR_OK; |
556 |
xvid_debug = init->debug; |
} |
557 |
#endif |
|
558 |
|
|
559 |
|
|
560 |
|
static int |
561 |
|
xvid_init_convert(XVID_INIT_CONVERTINFO* convert) |
562 |
|
{ |
563 |
|
/* |
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; |
569 |
|
const int height = convert->height; |
570 |
|
const int width2 = convert->width/2; |
571 |
|
const int height2 = convert->height/2; |
572 |
|
IMAGE img; |
573 |
|
|
574 |
|
switch (convert->input.colorspace & ~XVID_CSP_VFLIP) |
575 |
|
{ |
576 |
|
case XVID_CSP_YV12 : |
577 |
|
img.y = convert->input.y; |
578 |
|
img.v = (uint8_t*)convert->input.y + width*height; |
579 |
|
img.u = (uint8_t*)convert->input.y + width*height + width2*height2; |
580 |
|
image_output(&img, width, height, width, |
581 |
|
convert->output.y, convert->output.y_stride, |
582 |
|
convert->output.colorspace, convert->interlacing); |
583 |
|
break; |
584 |
|
|
585 |
|
default : |
586 |
|
return XVID_ERR_FORMAT; |
587 |
|
} |
588 |
|
|
589 |
|
|
590 |
|
emms(); |
591 |
|
return XVID_ERR_OK; |
592 |
|
} |
593 |
|
|
594 |
|
|
595 |
|
|
596 |
|
void fill8(uint8_t * block, int size, int value) |
597 |
|
{ |
598 |
|
int i; |
599 |
|
for (i = 0; i < size; i++) |
600 |
|
block[i] = value; |
601 |
|
} |
602 |
|
|
603 |
|
void fill16(int16_t * block, int size, int value) |
604 |
|
{ |
605 |
|
int i; |
606 |
|
for (i = 0; i < size; i++) |
607 |
|
block[i] = value; |
608 |
|
} |
609 |
|
|
610 |
|
#define RANDOM(min,max) min + (rand() % (max-min)) |
611 |
|
|
612 |
|
void random8(uint8_t * block, int size, int min, int max) |
613 |
|
{ |
614 |
|
int i; |
615 |
|
for (i = 0; i < size; i++) |
616 |
|
block[i] = RANDOM(min,max); |
617 |
|
} |
618 |
|
|
619 |
|
void random16(int16_t * block, int size, int min, int max) |
620 |
|
{ |
621 |
|
int i; |
622 |
|
for (i = 0; i < size; i++) |
623 |
|
block[i] = RANDOM(min,max); |
624 |
|
} |
625 |
|
|
626 |
|
int compare16(const int16_t * blockA, const int16_t * blockB, int size) |
627 |
|
{ |
628 |
|
int i; |
629 |
|
for (i = 0; i < size; i++) |
630 |
|
if (blockA[i] != blockB[i]) |
631 |
|
return 1; |
632 |
|
|
633 |
return 0; |
return 0; |
634 |
} |
} |
635 |
|
|
636 |
|
int diff16(const int16_t * blockA, const int16_t * blockB, int size) |
637 |
|
{ |
638 |
|
int i, diff = 0; |
639 |
|
for (i = 0; i < size; i++) |
640 |
|
diff += ABS(blockA[i]-blockB[i]); |
641 |
|
return diff; |
642 |
|
} |
643 |
|
|
644 |
static int |
|
645 |
xvid_gbl_info(xvid_gbl_info_t * info) |
#define XVID_TEST_RANDOM 0x00000001 /* random input data */ |
646 |
|
#define XVID_TEST_VERBOSE 0x00000002 /* verbose error output */ |
647 |
|
|
648 |
|
|
649 |
|
#define TEST_FORWARD 0x00000001 /* intra */ |
650 |
|
#define TEST_FDCT (TEST_FORWARD) |
651 |
|
#define TEST_IDCT (0) |
652 |
|
|
653 |
|
static int test_transform(void * funcA, void * funcB, const char * nameB, |
654 |
|
int test, int flags) |
655 |
{ |
{ |
656 |
if (XVID_VERSION_MAJOR(info->version) != 1) /* v1.x.x */ |
int i; |
657 |
return XVID_ERR_VERSION; |
int64_t timeSTART; |
658 |
|
int64_t timeA = 0; |
659 |
|
int64_t timeB = 0; |
660 |
|
DECLARE_ALIGNED_MATRIX(arrayA, 1, 64, int16_t, CACHE_LINE); |
661 |
|
DECLARE_ALIGNED_MATRIX(arrayB, 1, 64, int16_t, CACHE_LINE); |
662 |
|
int min, max; |
663 |
|
int count = 0; |
664 |
|
|
665 |
|
int tmp; |
666 |
|
int min_error = 0x10000*64; |
667 |
|
int max_error = 0; |
668 |
|
|
|
info->actual_version = XVID_VERSION; |
|
|
info->build = "dev-api-4"; |
|
|
info->cpu_flags = detect_cpu_flags(); |
|
669 |
|
|
670 |
#if defined(_SMP) && defined(WIN32) |
if ((test & TEST_FORWARD)) /* forward */ |
671 |
info->num_threads = pthread_num_processors_np();; |
{ |
672 |
#else |
min = -256; |
673 |
info->num_threads = 0; |
max = 255; |
674 |
#endif |
}else{ /* inverse */ |
675 |
|
min = -2048; |
676 |
|
max = 2047; |
677 |
|
} |
678 |
|
|
679 |
|
for (i = 0; i < 64*64; i++) |
680 |
|
{ |
681 |
|
if ((flags & XVID_TEST_RANDOM)) |
682 |
|
{ |
683 |
|
random16(arrayA, 64, min, max); |
684 |
|
}else{ |
685 |
|
fill16(arrayA, 64, i); |
686 |
|
} |
687 |
|
memcpy(arrayB, arrayA, 64*sizeof(int16_t)); |
688 |
|
|
689 |
|
if ((test & TEST_FORWARD)) |
690 |
|
{ |
691 |
|
timeSTART = read_counter(); |
692 |
|
((fdctFunc*)funcA)(arrayA); |
693 |
|
timeA += read_counter() - timeSTART; |
694 |
|
|
695 |
|
timeSTART = read_counter(); |
696 |
|
((fdctFunc*)funcB)(arrayB); |
697 |
|
timeB += read_counter() - timeSTART; |
698 |
|
} |
699 |
|
else |
700 |
|
{ |
701 |
|
timeSTART = read_counter(); |
702 |
|
((idctFunc*)funcA)(arrayA); |
703 |
|
timeA += read_counter() - timeSTART; |
704 |
|
|
705 |
|
timeSTART = read_counter(); |
706 |
|
((idctFunc*)funcB)(arrayB); |
707 |
|
timeB += read_counter() - timeSTART; |
708 |
|
} |
709 |
|
|
710 |
|
tmp = diff16(arrayA, arrayB, 64) / 64; |
711 |
|
if (tmp > max_error) |
712 |
|
max_error = tmp; |
713 |
|
if (tmp < min_error) |
714 |
|
min_error = tmp; |
715 |
|
|
716 |
|
count++; |
717 |
|
} |
718 |
|
|
719 |
|
/* print the "average difference" of best/worst transforms */ |
720 |
|
printf("%s:\t%i\t(min_error:%i, max_error:%i)\n", nameB, (int)(timeB / count), min_error, max_error); |
721 |
|
|
722 |
return 0; |
return 0; |
723 |
} |
} |
724 |
|
|
725 |
|
|
726 |
static int |
#define TEST_QUANT 0x00000001 /* forward quantization */ |
727 |
xvid_gbl_convert(xvid_gbl_convert_t* convert) |
#define TEST_INTRA 0x00000002 /* intra */ |
728 |
|
#define TEST_QUANT_INTRA (TEST_QUANT|TEST_INTRA) |
729 |
|
#define TEST_QUANT_INTER (TEST_QUANT) |
730 |
|
#define TEST_DEQUANT_INTRA (TEST_INTRA) |
731 |
|
#define TEST_DEQUANT_INTER (0) |
732 |
|
|
733 |
|
static int test_quant(void * funcA, void * funcB, const char * nameB, |
734 |
|
int test, int flags) |
735 |
{ |
{ |
736 |
int width; |
int q,i; |
737 |
int height; |
int64_t timeSTART; |
738 |
int width2; |
int64_t timeA = 0; |
739 |
int height2; |
int64_t timeB = 0; |
740 |
IMAGE img; |
int retA = 0, retB = 0; |
741 |
|
DECLARE_ALIGNED_MATRIX(arrayX, 1, 64, int16_t, CACHE_LINE); |
742 |
|
DECLARE_ALIGNED_MATRIX(arrayA, 1, 64, int16_t, CACHE_LINE); |
743 |
|
DECLARE_ALIGNED_MATRIX(arrayB, 1, 64, int16_t, CACHE_LINE); |
744 |
|
int min, max; |
745 |
|
int count = 0; |
746 |
|
int errors = 0; |
747 |
|
|
748 |
if (XVID_VERSION_MAJOR(convert->version) != 1) /* v1.x.x */ |
if ((test & TEST_QUANT)) /* quant */ |
749 |
return XVID_ERR_VERSION; |
{ |
750 |
|
min = -2048; |
751 |
|
max = 2047; |
752 |
|
}else{ /* dequant */ |
753 |
|
min = -256; |
754 |
|
max = 255; |
755 |
|
} |
756 |
|
|
757 |
#if 0 |
for (q = 1; q <= 31; q++) /* quantizer */ |
758 |
const int flip1 = (convert->input.colorspace & XVID_CSP_VFLIP) ^ (convert->output.colorspace & XVID_CSP_VFLIP); |
{ |
759 |
#endif |
for (i = min; i < max; i++) /* input coeff */ |
760 |
width = convert->width; |
{ |
761 |
height = convert->height; |
if ((flags & XVID_TEST_RANDOM)) |
762 |
width2 = convert->width/2; |
{ |
763 |
height2 = convert->height/2; |
random16(arrayX, 64, min, max); |
764 |
|
}else{ |
765 |
|
fill16(arrayX, 64, i); |
766 |
|
} |
767 |
|
|
768 |
switch (convert->input.csp & ~XVID_CSP_VFLIP) |
if ((test & TEST_INTRA)) /* intra */ |
769 |
{ |
{ |
770 |
case XVID_CSP_YV12 : |
timeSTART = read_counter(); |
771 |
img.y = convert->input.plane[0]; |
((quanth263_intraFunc*)funcA)(arrayA, arrayX, q, q); |
772 |
img.v = (uint8_t*)convert->input.plane[0] + convert->input.stride[0]*height; |
timeA += read_counter() - timeSTART; |
773 |
img.u = (uint8_t*)convert->input.plane[0] + convert->input.stride[0]*height + (convert->input.stride[0]/2)*height2; |
|
774 |
image_output(&img, width, height, width, |
timeSTART = read_counter(); |
775 |
(uint8_t**)convert->output.plane, convert->output.stride, |
((quanth263_intraFunc*)funcB)(arrayB, arrayX, q, q); |
776 |
convert->output.csp, convert->interlacing); |
timeB += read_counter() - timeSTART; |
777 |
break; |
} |
778 |
|
else /* inter */ |
779 |
|
{ |
780 |
|
timeSTART = read_counter(); |
781 |
|
retA = ((quanth263_interFunc*)funcA)(arrayA, arrayX, q); |
782 |
|
timeA += read_counter() - timeSTART; |
783 |
|
|
784 |
|
timeSTART = read_counter(); |
785 |
|
retB = ((quanth263_interFunc*)funcB)(arrayB, arrayX, q); |
786 |
|
timeB += read_counter() - timeSTART; |
787 |
|
} |
788 |
|
|
789 |
default : |
/* compare return value from quant_inter, and compare (de)quantiz'd arrays */ |
790 |
return XVID_ERR_FORMAT; |
if ( ((test&TEST_QUANT) && !(test&TEST_INTRA) && retA != retB ) || |
791 |
|
compare16(arrayA, arrayB, 64)) |
792 |
|
{ |
793 |
|
errors++; |
794 |
|
if ((flags & XVID_TEST_VERBOSE)) |
795 |
|
printf("%s error: q=%i, i=%i\n", nameB, q, i); |
796 |
} |
} |
797 |
|
|
798 |
|
count++; |
799 |
|
} |
800 |
|
} |
801 |
|
|
802 |
|
printf("%s:\t%i", nameB, (int)(timeB / count)); |
803 |
|
if (errors>0) |
804 |
|
printf("\t(%i errors out of %i)", errors, count); |
805 |
|
printf("\n"); |
806 |
|
|
|
emms(); |
|
807 |
return 0; |
return 0; |
808 |
} |
} |
809 |
|
|
810 |
/***************************************************************************** |
|
811 |
* XviD Global Entry point |
|
812 |
* |
int xvid_init_test(int flags) |
813 |
* Well this function initialize all internal function pointers according |
{ |
814 |
* to the CPU features forced by the library client or autodetected (depending |
#if defined(ARCH_IS_IA32) |
815 |
* on the XVID_CPU_FORCE flag). It also initializes vlc coding tables and all |
int cpu_flags; |
816 |
* image colorspace transformation tables. |
#endif |
817 |
* |
|
818 |
****************************************************************************/ |
printf("XviD tests\n\n"); |
819 |
|
|
820 |
|
#if defined(ARCH_IS_IA32) |
821 |
|
cpu_flags = detect_cpu_flags(); |
822 |
|
#endif |
823 |
|
|
824 |
|
idct_int32_init(); |
825 |
|
emms(); |
826 |
|
|
827 |
|
srand(time(0)); |
828 |
|
|
829 |
|
/* fDCT test */ |
830 |
|
printf("--- fdct ---\n"); |
831 |
|
test_transform(fdct_int32, fdct_int32, "c", TEST_FDCT, flags); |
832 |
|
|
833 |
|
#if defined(ARCH_IS_IA32) |
834 |
|
if (cpu_flags & XVID_CPU_MMX) |
835 |
|
test_transform(fdct_int32, fdct_mmx, "mmx", TEST_FDCT, flags); |
836 |
|
if (cpu_flags & XVID_CPU_SSE2) |
837 |
|
test_transform(fdct_int32, fdct_sse2, "sse2", TEST_FDCT, flags); |
838 |
|
#endif |
839 |
|
|
840 |
|
/* iDCT test */ |
841 |
|
printf("\n--- idct ---\n"); |
842 |
|
test_transform(idct_int32, idct_int32, "c", TEST_IDCT, flags); |
843 |
|
|
844 |
|
#if defined(ARCH_IS_IA32) |
845 |
|
if (cpu_flags & XVID_CPU_MMX) |
846 |
|
test_transform(idct_int32, idct_mmx, "mmx", TEST_IDCT, flags); |
847 |
|
if (cpu_flags & XVID_CPU_MMXEXT) |
848 |
|
test_transform(idct_int32, idct_xmm, "xmm", TEST_IDCT, flags); |
849 |
|
if (cpu_flags & XVID_CPU_3DNOWEXT) |
850 |
|
test_transform(idct_int32, idct_3dne, "3dne", TEST_IDCT, flags); |
851 |
|
if (cpu_flags & XVID_CPU_SSE2) |
852 |
|
test_transform(idct_int32, idct_sse2, "sse2", TEST_IDCT, flags); |
853 |
|
#endif |
854 |
|
|
855 |
|
/* Intra quantization test */ |
856 |
|
printf("\n--- quant intra ---\n"); |
857 |
|
test_quant(quant_intra_c, quant_intra_c, "c", TEST_QUANT_INTRA, flags); |
858 |
|
|
859 |
|
#if defined(ARCH_IS_IA32) |
860 |
|
if (cpu_flags & XVID_CPU_MMX) |
861 |
|
test_quant(quant_intra_c, quant_intra_mmx, "mmx", TEST_QUANT_INTRA, flags); |
862 |
|
if (cpu_flags & XVID_CPU_3DNOWEXT) |
863 |
|
test_quant(quant_intra_c, quant_intra_3dne, "3dne", TEST_QUANT_INTRA, flags); |
864 |
|
if (cpu_flags & XVID_CPU_SSE2) |
865 |
|
test_quant(quant_intra_c, quant_intra_sse2, "sse2", TEST_QUANT_INTRA, flags); |
866 |
|
#endif |
867 |
|
|
868 |
|
/* Inter quantization test */ |
869 |
|
printf("\n--- quant inter ---\n"); |
870 |
|
test_quant(quant_inter_c, quant_inter_c, "c", TEST_QUANT_INTER, flags); |
871 |
|
|
872 |
|
#if defined(ARCH_IS_IA32) |
873 |
|
if (cpu_flags & XVID_CPU_MMX) |
874 |
|
test_quant(quant_inter_c, quant_inter_mmx, "mmx", TEST_QUANT_INTER, flags); |
875 |
|
if (cpu_flags & XVID_CPU_3DNOWEXT) |
876 |
|
test_quant(quant_inter_c, quant_inter_3dne, "3dne", TEST_QUANT_INTER, flags); |
877 |
|
if (cpu_flags & XVID_CPU_SSE2) |
878 |
|
test_quant(quant_inter_c, quant_inter_sse2, "sse2", TEST_QUANT_INTER, flags); |
879 |
|
#endif |
880 |
|
|
881 |
|
/* Intra dequantization test */ |
882 |
|
printf("\n--- dequant intra ---\n"); |
883 |
|
test_quant(dequant_intra_c, dequant_intra_c, "c", TEST_DEQUANT_INTRA, flags); |
884 |
|
|
885 |
|
#if defined(ARCH_IS_IA32) |
886 |
|
if (cpu_flags & XVID_CPU_MMX) |
887 |
|
test_quant(dequant_intra_c, dequant_intra_mmx, "mmx", TEST_DEQUANT_INTRA, flags); |
888 |
|
if (cpu_flags & XVID_CPU_MMXEXT) |
889 |
|
test_quant(dequant_intra_c, dequant_intra_xmm, "xmm", TEST_DEQUANT_INTRA, flags); |
890 |
|
if (cpu_flags & XVID_CPU_3DNOWEXT) |
891 |
|
test_quant(dequant_intra_c, dequant_intra_3dne, "3dne", TEST_DEQUANT_INTRA, flags); |
892 |
|
if (cpu_flags & XVID_CPU_SSE2) |
893 |
|
test_quant(dequant_intra_c, dequant_intra_sse2, "sse2", TEST_DEQUANT_INTRA, flags); |
894 |
|
#endif |
895 |
|
|
896 |
|
/* Inter dequantization test */ |
897 |
|
printf("\n--- dequant inter ---\n"); |
898 |
|
test_quant(dequant_inter_c, dequant_inter_c, "c", TEST_DEQUANT_INTER, flags); |
899 |
|
|
900 |
|
#if defined(ARCH_IS_IA32) |
901 |
|
if (cpu_flags & XVID_CPU_MMX) |
902 |
|
test_quant(dequant_inter_c, dequant_inter_mmx, "mmx", TEST_DEQUANT_INTER, flags); |
903 |
|
if (cpu_flags & XVID_CPU_MMXEXT) |
904 |
|
test_quant(dequant_inter_c, dequant_inter_xmm, "xmm", TEST_DEQUANT_INTER, flags); |
905 |
|
if (cpu_flags & XVID_CPU_3DNOWEXT) |
906 |
|
test_quant(dequant_inter_c, dequant_inter_3dne, "3dne", TEST_DEQUANT_INTER, flags); |
907 |
|
if (cpu_flags & XVID_CPU_SSE2) |
908 |
|
test_quant(dequant_inter_c, dequant_inter_sse2, "sse2", TEST_DEQUANT_INTER, flags); |
909 |
|
#endif |
910 |
|
|
911 |
|
/* Intra quantization test */ |
912 |
|
printf("\n--- quant4 intra ---\n"); |
913 |
|
test_quant(quant4_intra_c, quant4_intra_c, "c", TEST_QUANT_INTRA, flags); |
914 |
|
|
915 |
|
#if defined(ARCH_IS_IA32) |
916 |
|
if (cpu_flags & XVID_CPU_MMX) |
917 |
|
test_quant(quant4_intra_c, quant4_intra_mmx, "mmx", TEST_QUANT_INTRA, flags); |
918 |
|
if (cpu_flags & XVID_CPU_MMXEXT) |
919 |
|
test_quant(quant4_intra_c, quant4_intra_xmm, "xmm", TEST_QUANT_INTRA, flags); |
920 |
|
#endif |
921 |
|
|
922 |
|
/* Inter quantization test */ |
923 |
|
printf("\n--- quant4 inter ---\n"); |
924 |
|
test_quant(quant4_inter_c, quant4_inter_c, "c", TEST_QUANT_INTER, flags); |
925 |
|
|
926 |
|
#if defined(ARCH_IS_IA32) |
927 |
|
if (cpu_flags & XVID_CPU_MMX) |
928 |
|
test_quant(quant4_inter_c, quant4_inter_mmx, "mmx", TEST_QUANT_INTER, flags); |
929 |
|
if (cpu_flags & XVID_CPU_MMXEXT) |
930 |
|
test_quant(quant4_inter_c, quant4_inter_xmm, "xmm", TEST_QUANT_INTER, flags); |
931 |
|
#endif |
932 |
|
|
933 |
|
/* Intra dequantization test */ |
934 |
|
printf("\n--- dequant4 intra ---\n"); |
935 |
|
test_quant(dequant4_intra_c, dequant4_intra_c, "c", TEST_DEQUANT_INTRA, flags); |
936 |
|
|
937 |
|
#if defined(ARCH_IS_IA32) |
938 |
|
if (cpu_flags & XVID_CPU_MMX) |
939 |
|
test_quant(dequant4_intra_c, dequant4_intra_mmx, "mmx", TEST_DEQUANT_INTRA, flags); |
940 |
|
if (cpu_flags & XVID_CPU_3DNOWEXT) |
941 |
|
test_quant(dequant4_intra_c, dequant4_intra_3dne, "3dne", TEST_DEQUANT_INTRA, flags); |
942 |
|
#endif |
943 |
|
|
944 |
|
/* Inter dequantization test */ |
945 |
|
printf("\n--- dequant4 inter ---\n"); |
946 |
|
test_quant(dequant4_inter_c, dequant4_inter_c, "c", TEST_DEQUANT_INTER, flags); |
947 |
|
|
948 |
|
#if defined(ARCH_IS_IA32) |
949 |
|
if (cpu_flags & XVID_CPU_MMX) |
950 |
|
test_quant(dequant4_inter_c, dequant4_inter_mmx, "mmx", TEST_DEQUANT_INTER, flags); |
951 |
|
if (cpu_flags & XVID_CPU_3DNOWEXT) |
952 |
|
test_quant(dequant4_inter_c, dequant4_inter_3dne, "3dne", TEST_DEQUANT_INTER, flags); |
953 |
|
#endif |
954 |
|
|
955 |
|
emms(); |
956 |
|
|
957 |
|
return XVID_ERR_OK; |
958 |
|
} |
959 |
|
|
960 |
|
|
961 |
int |
int |
962 |
xvid_global(void *handle, |
xvid_init(void *handle, |
963 |
int opt, |
int opt, |
964 |
void *param1, |
void *param1, |
965 |
void *param2) |
void *param2) |
966 |
{ |
{ |
967 |
switch(opt) |
switch(opt) |
968 |
{ |
{ |
969 |
case XVID_GBL_INIT : |
case XVID_INIT_INIT : |
970 |
return xvid_gbl_init((xvid_gbl_init_t*)param1); |
return xvid_init_init((XVID_INIT_PARAM*)param1); |
|
|
|
|
case XVID_GBL_INFO : |
|
|
return xvid_gbl_info((xvid_gbl_info_t*)param1); |
|
971 |
|
|
972 |
case XVID_GBL_CONVERT : |
case XVID_INIT_CONVERT : |
973 |
return xvid_gbl_convert((xvid_gbl_convert_t*)param1); |
return xvid_init_convert((XVID_INIT_CONVERTINFO*)param1); |
974 |
|
|
975 |
|
case XVID_INIT_TEST : |
976 |
|
{ |
977 |
|
ptr_t flags = (ptr_t)param1; |
978 |
|
return xvid_init_test((int)flags); |
979 |
|
} |
980 |
default : |
default : |
981 |
return XVID_ERR_FAIL; |
return XVID_ERR_FAIL; |
982 |
} |
} |
999 |
void *param2) |
void *param2) |
1000 |
{ |
{ |
1001 |
switch (opt) { |
switch (opt) { |
1002 |
|
case XVID_DEC_DECODE: |
1003 |
|
return decoder_decode((DECODER *) handle, (XVID_DEC_FRAME *) param1, (XVID_DEC_STATS*) param2); |
1004 |
|
|
1005 |
case XVID_DEC_CREATE: |
case XVID_DEC_CREATE: |
1006 |
return decoder_create((xvid_dec_create_t *) param1); |
return decoder_create((XVID_DEC_PARAM *) param1); |
1007 |
|
|
1008 |
case XVID_DEC_DESTROY: |
case XVID_DEC_DESTROY: |
1009 |
return decoder_destroy((DECODER *) handle); |
return decoder_destroy((DECODER *) handle); |
1010 |
|
|
|
case XVID_DEC_DECODE: |
|
|
return decoder_decode((DECODER *) handle, (xvid_dec_frame_t *) param1, (xvid_dec_stats_t*) param2); |
|
|
|
|
1011 |
default: |
default: |
1012 |
return XVID_ERR_FAIL; |
return XVID_ERR_FAIL; |
1013 |
} |
} |
1033 |
switch (opt) { |
switch (opt) { |
1034 |
case XVID_ENC_ENCODE: |
case XVID_ENC_ENCODE: |
1035 |
|
|
1036 |
return enc_encode((Encoder *) handle, |
if (((Encoder *) handle)->mbParam.max_bframes >= 0) |
1037 |
(xvid_enc_frame_t *) param1, |
return encoder_encode_bframes((Encoder *) handle, |
1038 |
(xvid_enc_stats_t *) param2); |
(XVID_ENC_FRAME *) param1, |
1039 |
|
(XVID_ENC_STATS *) param2); |
1040 |
|
else |
1041 |
|
return encoder_encode((Encoder *) handle, |
1042 |
|
(XVID_ENC_FRAME *) param1, |
1043 |
|
(XVID_ENC_STATS *) param2); |
1044 |
|
|
1045 |
case XVID_ENC_CREATE: |
case XVID_ENC_CREATE: |
1046 |
return enc_create((xvid_enc_create_t *) param1); |
return encoder_create((XVID_ENC_PARAM *) param1); |
1047 |
|
|
1048 |
case XVID_ENC_DESTROY: |
case XVID_ENC_DESTROY: |
1049 |
return enc_destroy((Encoder *) handle); |
return encoder_destroy((Encoder *) handle); |
1050 |
|
|
1051 |
default: |
default: |
1052 |
return XVID_ERR_FAIL; |
return XVID_ERR_FAIL; |