2001-08-14 00:38:25 +03:00
|
|
|
/*
|
2008-12-17 02:54:54 +02:00
|
|
|
* ARM optimized DSP utils
|
2009-01-19 17:46:40 +02:00
|
|
|
* Copyright (c) 2001 Lionel Ulmer
|
2001-08-14 00:38:25 +03:00
|
|
|
*
|
2006-10-07 18:30:46 +03:00
|
|
|
* This file is part of FFmpeg.
|
|
|
|
*
|
|
|
|
* FFmpeg is free software; you can redistribute it and/or
|
2002-05-26 01:45:33 +03:00
|
|
|
* modify it under the terms of the GNU Lesser General Public
|
|
|
|
* License as published by the Free Software Foundation; either
|
2006-10-07 18:30:46 +03:00
|
|
|
* version 2.1 of the License, or (at your option) any later version.
|
2001-08-14 00:38:25 +03:00
|
|
|
*
|
2006-10-07 18:30:46 +03:00
|
|
|
* FFmpeg is distributed in the hope that it will be useful,
|
2001-08-14 00:38:25 +03:00
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
2002-05-26 01:45:33 +03:00
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
* Lesser General Public License for more details.
|
2001-08-14 00:38:25 +03:00
|
|
|
*
|
2002-05-26 01:45:33 +03:00
|
|
|
* You should have received a copy of the GNU Lesser General Public
|
2006-10-07 18:30:46 +03:00
|
|
|
* License along with FFmpeg; if not, write to the Free Software
|
2006-01-13 00:43:26 +02:00
|
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
2001-08-14 00:38:25 +03:00
|
|
|
*/
|
|
|
|
|
2008-05-09 14:56:36 +03:00
|
|
|
#include "libavcodec/dsputil.h"
|
2009-01-14 01:44:16 +02:00
|
|
|
#if HAVE_IPP
|
2008-05-09 13:06:51 +03:00
|
|
|
#include <ipp.h>
|
2005-05-26 17:32:46 +03:00
|
|
|
#endif
|
|
|
|
|
2008-12-03 17:23:30 +02:00
|
|
|
void dsputil_init_iwmmxt(DSPContext* c, AVCodecContext *avctx);
|
|
|
|
void ff_float_init_arm_vfp(DSPContext* c, AVCodecContext *avctx);
|
2008-12-16 00:12:32 +02:00
|
|
|
void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx);
|
2001-08-14 00:38:25 +03:00
|
|
|
|
2008-12-03 17:23:30 +02:00
|
|
|
void j_rev_dct_ARM(DCTELEM *data);
|
|
|
|
void simple_idct_ARM(DCTELEM *data);
|
2001-08-14 00:38:25 +03:00
|
|
|
|
2008-12-03 17:23:30 +02:00
|
|
|
void simple_idct_armv5te(DCTELEM *data);
|
|
|
|
void simple_idct_put_armv5te(uint8_t *dest, int line_size, DCTELEM *data);
|
|
|
|
void simple_idct_add_armv5te(uint8_t *dest, int line_size, DCTELEM *data);
|
2006-09-26 23:26:22 +03:00
|
|
|
|
2008-12-03 17:23:30 +02:00
|
|
|
void ff_simple_idct_armv6(DCTELEM *data);
|
|
|
|
void ff_simple_idct_put_armv6(uint8_t *dest, int line_size, DCTELEM *data);
|
|
|
|
void ff_simple_idct_add_armv6(uint8_t *dest, int line_size, DCTELEM *data);
|
2007-01-28 23:32:08 +02:00
|
|
|
|
2008-12-16 00:12:37 +02:00
|
|
|
void ff_simple_idct_neon(DCTELEM *data);
|
|
|
|
void ff_simple_idct_put_neon(uint8_t *dest, int line_size, DCTELEM *data);
|
|
|
|
void ff_simple_idct_add_neon(uint8_t *dest, int line_size, DCTELEM *data);
|
|
|
|
|
2009-07-04 23:41:11 +03:00
|
|
|
void ff_vp3_idct_neon(DCTELEM *data);
|
|
|
|
void ff_vp3_idct_put_neon(uint8_t *dest, int line_size, DCTELEM *data);
|
|
|
|
void ff_vp3_idct_add_neon(uint8_t *dest, int line_size, DCTELEM *data);
|
|
|
|
|
2003-03-03 16:54:00 +02:00
|
|
|
/* XXX: local hack */
|
|
|
|
static void (*ff_put_pixels_clamped)(const DCTELEM *block, uint8_t *pixels, int line_size);
|
|
|
|
static void (*ff_add_pixels_clamped)(const DCTELEM *block, uint8_t *pixels, int line_size);
|
|
|
|
|
2005-05-26 17:32:46 +03:00
|
|
|
void put_pixels8_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
|
|
|
|
void put_pixels8_x2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
|
|
|
|
void put_pixels8_y2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
|
|
|
|
void put_pixels8_xy2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
|
|
|
|
|
|
|
|
void put_no_rnd_pixels8_x2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
|
|
|
|
void put_no_rnd_pixels8_y2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
|
|
|
|
void put_no_rnd_pixels8_xy2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
|
|
|
|
|
|
|
|
void put_pixels16_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
|
|
|
|
|
2008-12-03 17:23:30 +02:00
|
|
|
void ff_prefetch_arm(void *mem, int stride, int h);
|
2008-10-27 02:25:12 +02:00
|
|
|
|
2005-05-26 17:40:20 +03:00
|
|
|
CALL_2X_PIXELS(put_pixels16_x2_arm , put_pixels8_x2_arm , 8)
|
|
|
|
CALL_2X_PIXELS(put_pixels16_y2_arm , put_pixels8_y2_arm , 8)
|
|
|
|
CALL_2X_PIXELS(put_pixels16_xy2_arm, put_pixels8_xy2_arm, 8)
|
|
|
|
CALL_2X_PIXELS(put_no_rnd_pixels16_x2_arm , put_no_rnd_pixels8_x2_arm , 8)
|
|
|
|
CALL_2X_PIXELS(put_no_rnd_pixels16_y2_arm , put_no_rnd_pixels8_y2_arm , 8)
|
|
|
|
CALL_2X_PIXELS(put_no_rnd_pixels16_xy2_arm, put_no_rnd_pixels8_xy2_arm, 8)
|
2005-05-26 17:32:46 +03:00
|
|
|
|
2008-12-03 17:23:30 +02:00
|
|
|
void ff_add_pixels_clamped_ARM(short *block, unsigned char *dest,
|
2008-10-27 02:25:19 +02:00
|
|
|
int line_size);
|
2005-05-26 17:32:46 +03:00
|
|
|
|
2003-03-03 16:54:00 +02:00
|
|
|
/* XXX: those functions should be suppressed ASAP when all IDCTs are
|
|
|
|
converted */
|
2003-07-07 14:19:18 +03:00
|
|
|
static void j_rev_dct_ARM_put(uint8_t *dest, int line_size, DCTELEM *block)
|
2003-03-03 16:54:00 +02:00
|
|
|
{
|
|
|
|
j_rev_dct_ARM (block);
|
|
|
|
ff_put_pixels_clamped(block, dest, line_size);
|
|
|
|
}
|
2003-07-07 14:19:18 +03:00
|
|
|
static void j_rev_dct_ARM_add(uint8_t *dest, int line_size, DCTELEM *block)
|
2003-03-03 16:54:00 +02:00
|
|
|
{
|
|
|
|
j_rev_dct_ARM (block);
|
|
|
|
ff_add_pixels_clamped(block, dest, line_size);
|
|
|
|
}
|
2003-07-07 14:19:18 +03:00
|
|
|
static void simple_idct_ARM_put(uint8_t *dest, int line_size, DCTELEM *block)
|
|
|
|
{
|
|
|
|
simple_idct_ARM (block);
|
|
|
|
ff_put_pixels_clamped(block, dest, line_size);
|
|
|
|
}
|
|
|
|
static void simple_idct_ARM_add(uint8_t *dest, int line_size, DCTELEM *block)
|
|
|
|
{
|
|
|
|
simple_idct_ARM (block);
|
|
|
|
ff_add_pixels_clamped(block, dest, line_size);
|
|
|
|
}
|
2006-09-16 02:20:58 +03:00
|
|
|
|
2009-01-14 01:44:16 +02:00
|
|
|
#if HAVE_IPP
|
2005-05-26 17:32:46 +03:00
|
|
|
static void simple_idct_ipp(DCTELEM *block)
|
|
|
|
{
|
|
|
|
ippiDCT8x8Inv_Video_16s_C1I(block);
|
|
|
|
}
|
|
|
|
static void simple_idct_ipp_put(uint8_t *dest, int line_size, DCTELEM *block)
|
|
|
|
{
|
|
|
|
ippiDCT8x8Inv_Video_16s8u_C1R(block, dest, line_size);
|
|
|
|
}
|
|
|
|
|
|
|
|
void add_pixels_clamped_iwmmxt(const DCTELEM *block, uint8_t *pixels, int line_size);
|
|
|
|
|
|
|
|
static void simple_idct_ipp_add(uint8_t *dest, int line_size, DCTELEM *block)
|
|
|
|
{
|
|
|
|
ippiDCT8x8Inv_Video_16s_C1I(block);
|
2009-01-14 01:44:16 +02:00
|
|
|
#if HAVE_IWMMXT
|
2005-05-26 17:32:46 +03:00
|
|
|
add_pixels_clamped_iwmmxt(block, dest, line_size);
|
|
|
|
#else
|
2008-10-27 02:25:19 +02:00
|
|
|
ff_add_pixels_clamped_ARM(block, dest, line_size);
|
2005-05-26 17:32:46 +03:00
|
|
|
#endif
|
|
|
|
}
|
2006-09-16 02:20:58 +03:00
|
|
|
#endif
|
2003-03-03 16:54:00 +02:00
|
|
|
|
2008-06-30 10:31:41 +03:00
|
|
|
int mm_support(void)
|
|
|
|
{
|
2009-01-14 19:19:17 +02:00
|
|
|
return HAVE_IWMMXT * FF_MM_IWMMXT;
|
2008-06-30 10:31:41 +03:00
|
|
|
}
|
|
|
|
|
2008-12-17 02:54:54 +02:00
|
|
|
void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx)
|
2001-08-14 00:38:25 +03:00
|
|
|
{
|
2006-09-16 02:20:58 +03:00
|
|
|
int idct_algo= avctx->idct_algo;
|
2003-03-03 16:54:00 +02:00
|
|
|
|
|
|
|
ff_put_pixels_clamped = c->put_pixels_clamped;
|
|
|
|
ff_add_pixels_clamped = c->add_pixels_clamped;
|
|
|
|
|
2008-01-27 10:36:50 +02:00
|
|
|
if (avctx->lowres == 0) {
|
2008-01-27 10:37:28 +02:00
|
|
|
if(idct_algo == FF_IDCT_AUTO){
|
2009-01-14 01:44:16 +02:00
|
|
|
#if HAVE_IPP
|
2008-01-27 10:37:28 +02:00
|
|
|
idct_algo = FF_IDCT_IPP;
|
2009-01-14 01:44:16 +02:00
|
|
|
#elif HAVE_NEON
|
2008-12-16 00:12:37 +02:00
|
|
|
idct_algo = FF_IDCT_SIMPLENEON;
|
2009-01-14 01:44:16 +02:00
|
|
|
#elif HAVE_ARMV6
|
2008-01-27 10:37:28 +02:00
|
|
|
idct_algo = FF_IDCT_SIMPLEARMV6;
|
2009-01-14 01:44:16 +02:00
|
|
|
#elif HAVE_ARMV5TE
|
2008-01-27 10:37:28 +02:00
|
|
|
idct_algo = FF_IDCT_SIMPLEARMV5TE;
|
2005-05-26 17:32:46 +03:00
|
|
|
#else
|
2008-01-27 10:37:28 +02:00
|
|
|
idct_algo = FF_IDCT_ARM;
|
2005-05-26 17:32:46 +03:00
|
|
|
#endif
|
2008-01-27 10:37:28 +02:00
|
|
|
}
|
2006-09-16 02:20:58 +03:00
|
|
|
|
2008-01-27 10:37:28 +02:00
|
|
|
if(idct_algo==FF_IDCT_ARM){
|
|
|
|
c->idct_put= j_rev_dct_ARM_put;
|
|
|
|
c->idct_add= j_rev_dct_ARM_add;
|
|
|
|
c->idct = j_rev_dct_ARM;
|
2008-12-06 15:42:00 +02:00
|
|
|
c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
|
2008-01-27 10:37:28 +02:00
|
|
|
} else if (idct_algo==FF_IDCT_SIMPLEARM){
|
|
|
|
c->idct_put= simple_idct_ARM_put;
|
|
|
|
c->idct_add= simple_idct_ARM_add;
|
|
|
|
c->idct = simple_idct_ARM;
|
|
|
|
c->idct_permutation_type= FF_NO_IDCT_PERM;
|
2009-01-14 01:44:16 +02:00
|
|
|
#if HAVE_ARMV6
|
2008-01-27 10:37:28 +02:00
|
|
|
} else if (idct_algo==FF_IDCT_SIMPLEARMV6){
|
|
|
|
c->idct_put= ff_simple_idct_put_armv6;
|
|
|
|
c->idct_add= ff_simple_idct_add_armv6;
|
|
|
|
c->idct = ff_simple_idct_armv6;
|
|
|
|
c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
|
2007-01-28 23:32:08 +02:00
|
|
|
#endif
|
2009-01-14 01:44:16 +02:00
|
|
|
#if HAVE_ARMV5TE
|
2008-01-27 10:37:28 +02:00
|
|
|
} else if (idct_algo==FF_IDCT_SIMPLEARMV5TE){
|
|
|
|
c->idct_put= simple_idct_put_armv5te;
|
|
|
|
c->idct_add= simple_idct_add_armv5te;
|
|
|
|
c->idct = simple_idct_armv5te;
|
|
|
|
c->idct_permutation_type = FF_NO_IDCT_PERM;
|
2006-09-26 23:26:22 +03:00
|
|
|
#endif
|
2009-01-14 01:44:16 +02:00
|
|
|
#if HAVE_IPP
|
2008-01-27 10:37:28 +02:00
|
|
|
} else if (idct_algo==FF_IDCT_IPP){
|
|
|
|
c->idct_put= simple_idct_ipp_put;
|
|
|
|
c->idct_add= simple_idct_ipp_add;
|
|
|
|
c->idct = simple_idct_ipp;
|
|
|
|
c->idct_permutation_type= FF_NO_IDCT_PERM;
|
2008-12-16 00:12:37 +02:00
|
|
|
#endif
|
2009-01-14 01:44:16 +02:00
|
|
|
#if HAVE_NEON
|
2008-12-16 00:12:37 +02:00
|
|
|
} else if (idct_algo==FF_IDCT_SIMPLENEON){
|
|
|
|
c->idct_put= ff_simple_idct_put_neon;
|
|
|
|
c->idct_add= ff_simple_idct_add_neon;
|
|
|
|
c->idct = ff_simple_idct_neon;
|
|
|
|
c->idct_permutation_type = FF_PARTTRANS_IDCT_PERM;
|
2009-07-04 23:41:11 +03:00
|
|
|
} else if ((CONFIG_VP3_DECODER || CONFIG_VP5_DECODER || CONFIG_VP6_DECODER || CONFIG_THEORA_DECODER) &&
|
|
|
|
idct_algo==FF_IDCT_VP3){
|
|
|
|
c->idct_put= ff_vp3_idct_put_neon;
|
|
|
|
c->idct_add= ff_vp3_idct_add_neon;
|
|
|
|
c->idct = ff_vp3_idct_neon;
|
|
|
|
c->idct_permutation_type = FF_TRANSPOSE_IDCT_PERM;
|
2006-09-16 02:20:58 +03:00
|
|
|
#endif
|
2008-01-27 10:37:28 +02:00
|
|
|
}
|
2008-01-27 10:36:50 +02:00
|
|
|
}
|
2005-05-26 17:32:46 +03:00
|
|
|
|
2007-09-05 14:05:33 +03:00
|
|
|
c->put_pixels_tab[0][0] = put_pixels16_arm;
|
2008-12-06 15:42:00 +02:00
|
|
|
c->put_pixels_tab[0][1] = put_pixels16_x2_arm;
|
|
|
|
c->put_pixels_tab[0][2] = put_pixels16_y2_arm;
|
2007-09-05 14:05:33 +03:00
|
|
|
c->put_pixels_tab[0][3] = put_pixels16_xy2_arm;
|
|
|
|
c->put_no_rnd_pixels_tab[0][0] = put_pixels16_arm;
|
2008-12-06 15:42:00 +02:00
|
|
|
c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_arm;
|
|
|
|
c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_arm;
|
2007-09-05 14:05:33 +03:00
|
|
|
c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_arm;
|
2008-12-06 15:42:00 +02:00
|
|
|
c->put_pixels_tab[1][0] = put_pixels8_arm;
|
|
|
|
c->put_pixels_tab[1][1] = put_pixels8_x2_arm;
|
2007-09-05 14:05:33 +03:00
|
|
|
c->put_pixels_tab[1][2] = put_pixels8_y2_arm;
|
|
|
|
c->put_pixels_tab[1][3] = put_pixels8_xy2_arm;
|
2008-12-06 15:42:00 +02:00
|
|
|
c->put_no_rnd_pixels_tab[1][0] = put_pixels8_arm;
|
|
|
|
c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_arm;
|
|
|
|
c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_arm;
|
2007-09-05 14:05:33 +03:00
|
|
|
c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_arm;
|
2005-05-26 17:32:46 +03:00
|
|
|
|
2009-01-14 01:44:16 +02:00
|
|
|
#if HAVE_ARMV5TE
|
2008-10-27 02:25:12 +02:00
|
|
|
c->prefetch = ff_prefetch_arm;
|
2008-08-25 05:29:04 +03:00
|
|
|
#endif
|
|
|
|
|
2009-01-14 01:44:16 +02:00
|
|
|
#if HAVE_IWMMXT
|
2005-05-26 17:32:46 +03:00
|
|
|
dsputil_init_iwmmxt(c, avctx);
|
|
|
|
#endif
|
2009-01-14 01:44:16 +02:00
|
|
|
#if HAVE_ARMVFP
|
2008-05-11 23:06:03 +03:00
|
|
|
ff_float_init_arm_vfp(c, avctx);
|
|
|
|
#endif
|
2009-01-14 01:44:16 +02:00
|
|
|
#if HAVE_NEON
|
2008-12-16 00:12:32 +02:00
|
|
|
ff_dsputil_init_neon(c, avctx);
|
|
|
|
#endif
|
2001-08-14 00:38:25 +03:00
|
|
|
}
|