2002-11-02 13:28:08 +02:00
|
|
|
/*
|
|
|
|
* Copyright (c) 2002 Brian Foley
|
|
|
|
* Copyright (c) 2002 Dieter Shirley
|
2004-04-20 20:05:12 +03:00
|
|
|
* Copyright (c) 2003-2004 Romain Dolbeau <romain@dolbeau.org>
|
2002-11-02 13:28:08 +02:00
|
|
|
*
|
2011-03-18 19:35:10 +02:00
|
|
|
* This file is part of Libav.
|
2006-10-07 18:30:46 +03:00
|
|
|
*
|
2011-03-18 19:35:10 +02:00
|
|
|
* Libav is free software; you can redistribute it and/or
|
2002-11-02 13:28:08 +02:00
|
|
|
* 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.
|
2002-11-02 13:28:08 +02:00
|
|
|
*
|
2011-03-18 19:35:10 +02:00
|
|
|
* Libav is distributed in the hope that it will be useful,
|
2002-11-02 13:28:08 +02:00
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
* Lesser General Public License for more details.
|
|
|
|
*
|
|
|
|
* You should have received a copy of the GNU Lesser General Public
|
2011-03-18 19:35:10 +02:00
|
|
|
* License along with Libav; 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
|
2002-11-02 13:28:08 +02:00
|
|
|
*/
|
|
|
|
|
2012-08-15 23:27:52 +03:00
|
|
|
#include <string.h>
|
|
|
|
|
2010-09-08 18:07:14 +03:00
|
|
|
#include "libavutil/cpu.h"
|
2008-05-09 14:56:36 +03:00
|
|
|
#include "libavcodec/dsputil.h"
|
2002-09-02 11:48:12 +03:00
|
|
|
#include "dsputil_altivec.h"
|
|
|
|
|
2003-01-19 21:00:45 +02:00
|
|
|
/* ***** WARNING ***** WARNING ***** WARNING ***** */
|
|
|
|
/*
|
2008-07-20 21:58:30 +03:00
|
|
|
clear_blocks_dcbz32_ppc will not work properly on PowerPC processors with a
|
|
|
|
cache line size not equal to 32 bytes.
|
|
|
|
Fortunately all processor used by Apple up to at least the 7450 (aka second
|
|
|
|
generation G4) use 32 bytes cache line.
|
|
|
|
This is due to the use of the 'dcbz' instruction. It simply clear to zero a
|
|
|
|
single cache line, so you need to know the cache line size to use it !
|
|
|
|
It's absurd, but it's fast...
|
2003-06-29 03:39:57 +03:00
|
|
|
|
2008-07-20 21:58:30 +03:00
|
|
|
update 24/06/2003 : Apple released yesterday the G5, with a PPC970. cache line
|
|
|
|
size: 128 bytes. Oups.
|
|
|
|
The semantic of dcbz was changed, it always clear 32 bytes. so the function
|
|
|
|
below will work, but will be slow. So I fixed check_dcbz_effect to use dcbzl,
|
|
|
|
which is defined to clear a cache line (as dcbz before). So we still can
|
|
|
|
distinguish, and use dcbz (32 bytes) or dcbzl (one cache line) as required.
|
2003-06-29 03:39:57 +03:00
|
|
|
|
2008-07-20 21:58:30 +03:00
|
|
|
see <http://developer.apple.com/technotes/tn/tn2087.html>
|
|
|
|
and <http://developer.apple.com/technotes/tn/tn2086.html>
|
2003-01-19 21:00:45 +02:00
|
|
|
*/
|
2010-03-07 00:37:14 +02:00
|
|
|
static void clear_blocks_dcbz32_ppc(DCTELEM *blocks)
|
2003-01-19 21:00:45 +02:00
|
|
|
{
|
|
|
|
register int misal = ((unsigned long)blocks & 0x00000010);
|
|
|
|
register int i = 0;
|
|
|
|
if (misal) {
|
2008-07-20 21:58:30 +03:00
|
|
|
((unsigned long*)blocks)[0] = 0L;
|
|
|
|
((unsigned long*)blocks)[1] = 0L;
|
|
|
|
((unsigned long*)blocks)[2] = 0L;
|
|
|
|
((unsigned long*)blocks)[3] = 0L;
|
|
|
|
i += 16;
|
2003-01-19 21:00:45 +02:00
|
|
|
}
|
2004-10-12 01:47:09 +03:00
|
|
|
for ( ; i < sizeof(DCTELEM)*6*64-31 ; i += 32) {
|
2008-10-16 16:34:09 +03:00
|
|
|
__asm__ volatile("dcbz %0,%1" : : "b" (blocks), "r" (i) : "memory");
|
2003-01-19 21:00:45 +02:00
|
|
|
}
|
|
|
|
if (misal) {
|
2008-07-20 21:58:30 +03:00
|
|
|
((unsigned long*)blocks)[188] = 0L;
|
|
|
|
((unsigned long*)blocks)[189] = 0L;
|
|
|
|
((unsigned long*)blocks)[190] = 0L;
|
|
|
|
((unsigned long*)blocks)[191] = 0L;
|
|
|
|
i += 16;
|
2003-01-19 21:00:45 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2003-06-29 03:39:57 +03:00
|
|
|
/* same as above, when dcbzl clear a whole 128B cache line
|
|
|
|
i.e. the PPC970 aka G5 */
|
2009-01-14 01:44:16 +02:00
|
|
|
#if HAVE_DCBZL
|
2010-03-07 00:37:14 +02:00
|
|
|
static void clear_blocks_dcbz128_ppc(DCTELEM *blocks)
|
2003-06-29 03:39:57 +03:00
|
|
|
{
|
|
|
|
register int misal = ((unsigned long)blocks & 0x0000007f);
|
|
|
|
register int i = 0;
|
2008-07-20 21:58:30 +03:00
|
|
|
if (misal) {
|
|
|
|
// we could probably also optimize this case,
|
|
|
|
// but there's not much point as the machines
|
|
|
|
// aren't available yet (2003-06-26)
|
|
|
|
memset(blocks, 0, sizeof(DCTELEM)*6*64);
|
2003-06-29 03:39:57 +03:00
|
|
|
}
|
|
|
|
else
|
2008-07-20 21:58:30 +03:00
|
|
|
for ( ; i < sizeof(DCTELEM)*6*64 ; i += 128) {
|
2008-10-16 16:34:09 +03:00
|
|
|
__asm__ volatile("dcbzl %0,%1" : : "b" (blocks), "r" (i) : "memory");
|
2008-07-20 21:58:30 +03:00
|
|
|
}
|
2003-06-29 03:39:57 +03:00
|
|
|
}
|
|
|
|
#else
|
2010-03-07 00:37:14 +02:00
|
|
|
static void clear_blocks_dcbz128_ppc(DCTELEM *blocks)
|
2003-06-29 03:39:57 +03:00
|
|
|
{
|
2008-07-20 21:58:30 +03:00
|
|
|
memset(blocks, 0, sizeof(DCTELEM)*6*64);
|
2003-06-29 03:39:57 +03:00
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2009-01-14 01:44:16 +02:00
|
|
|
#if HAVE_DCBZL
|
2003-01-19 21:00:45 +02:00
|
|
|
/* check dcbz report how many bytes are set to 0 by dcbz */
|
2003-06-29 03:39:57 +03:00
|
|
|
/* update 24/06/2003 : replace dcbz by dcbzl to get
|
|
|
|
the intended effect (Apple "fixed" dcbz)
|
|
|
|
unfortunately this cannot be used unless the assembler
|
|
|
|
knows about dcbzl ... */
|
2010-03-07 00:37:14 +02:00
|
|
|
static long check_dcbzl_effect(void)
|
2003-01-19 21:00:45 +02:00
|
|
|
{
|
2008-07-20 21:58:30 +03:00
|
|
|
register char *fakedata = av_malloc(1024);
|
|
|
|
register char *fakedata_middle;
|
|
|
|
register long zero = 0;
|
|
|
|
register long i = 0;
|
|
|
|
long count = 0;
|
2003-01-19 21:00:45 +02:00
|
|
|
|
2008-07-20 21:58:30 +03:00
|
|
|
if (!fakedata) {
|
|
|
|
return 0L;
|
|
|
|
}
|
2003-01-19 21:00:45 +02:00
|
|
|
|
2008-07-20 21:58:30 +03:00
|
|
|
fakedata_middle = (fakedata + 512);
|
2003-01-19 21:00:45 +02:00
|
|
|
|
2008-07-20 21:58:30 +03:00
|
|
|
memset(fakedata, 0xFF, 1024);
|
2003-01-19 21:00:45 +02:00
|
|
|
|
2008-07-20 21:58:30 +03:00
|
|
|
/* below the constraint "b" seems to mean "Address base register"
|
|
|
|
in gcc-3.3 / RS/6000 speaks. seems to avoid using r0, so.... */
|
2008-10-16 16:34:09 +03:00
|
|
|
__asm__ volatile("dcbzl %0, %1" : : "b" (fakedata_middle), "r" (zero));
|
2003-01-19 21:00:45 +02:00
|
|
|
|
2008-07-20 21:58:30 +03:00
|
|
|
for (i = 0; i < 1024 ; i ++) {
|
|
|
|
if (fakedata[i] == (char)0)
|
|
|
|
count++;
|
|
|
|
}
|
2003-01-19 21:00:45 +02:00
|
|
|
|
2008-07-20 21:58:30 +03:00
|
|
|
av_free(fakedata);
|
2005-12-17 20:14:38 +02:00
|
|
|
|
2008-07-20 21:58:30 +03:00
|
|
|
return count;
|
2003-01-19 21:00:45 +02:00
|
|
|
}
|
2003-06-29 03:39:57 +03:00
|
|
|
#else
|
2010-03-07 00:37:14 +02:00
|
|
|
static long check_dcbzl_effect(void)
|
2003-06-29 03:39:57 +03:00
|
|
|
{
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
#endif
|
2003-01-19 21:00:45 +02:00
|
|
|
|
2006-10-12 13:26:07 +03:00
|
|
|
static void prefetch_ppc(void *mem, int stride, int h)
|
|
|
|
{
|
|
|
|
register const uint8_t *p = mem;
|
|
|
|
do {
|
2008-10-16 16:34:09 +03:00
|
|
|
__asm__ volatile ("dcbt 0,%0" : : "r" (p));
|
2006-10-12 13:26:07 +03:00
|
|
|
p+= stride;
|
|
|
|
} while(--h);
|
|
|
|
}
|
|
|
|
|
2012-02-15 13:06:44 +03:00
|
|
|
void ff_dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx)
|
2002-09-02 11:48:12 +03:00
|
|
|
{
|
2011-07-21 12:05:15 +03:00
|
|
|
const int high_bit_depth = avctx->bits_per_raw_sample > 8;
|
2011-03-29 18:48:59 +03:00
|
|
|
|
2007-10-02 13:34:57 +03:00
|
|
|
// Common optimizations whether AltiVec is available or not
|
2006-10-12 13:26:07 +03:00
|
|
|
c->prefetch = prefetch_ppc;
|
2011-03-29 18:48:59 +03:00
|
|
|
if (!high_bit_depth) {
|
2006-08-03 19:13:50 +03:00
|
|
|
switch (check_dcbzl_effect()) {
|
|
|
|
case 32:
|
|
|
|
c->clear_blocks = clear_blocks_dcbz32_ppc;
|
|
|
|
break;
|
|
|
|
case 128:
|
|
|
|
c->clear_blocks = clear_blocks_dcbz128_ppc;
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
2011-03-29 18:48:59 +03:00
|
|
|
}
|
2004-09-16 22:05:22 +03:00
|
|
|
|
2009-01-14 01:44:16 +02:00
|
|
|
#if HAVE_ALTIVEC
|
2012-02-15 13:06:44 +03:00
|
|
|
if(CONFIG_H264_DECODER) ff_dsputil_h264_init_ppc(c, avctx);
|
2005-12-17 20:14:38 +02:00
|
|
|
|
2010-09-08 18:07:14 +03:00
|
|
|
if (av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) {
|
2012-02-15 13:06:44 +03:00
|
|
|
ff_dsputil_init_altivec(c, avctx);
|
2012-02-15 15:42:56 +03:00
|
|
|
ff_float_init_altivec(c, avctx);
|
|
|
|
ff_int_init_altivec(c, avctx);
|
|
|
|
c->gmc1 = ff_gmc1_altivec;
|
2003-03-03 16:54:00 +02:00
|
|
|
|
2009-01-14 01:44:16 +02:00
|
|
|
#if CONFIG_ENCODERS
|
2011-07-20 22:01:56 +03:00
|
|
|
if (avctx->bits_per_raw_sample <= 8 &&
|
|
|
|
(avctx->dct_algo == FF_DCT_AUTO ||
|
|
|
|
avctx->dct_algo == FF_DCT_ALTIVEC)) {
|
2012-02-15 15:42:56 +03:00
|
|
|
c->fdct = ff_fdct_altivec;
|
2005-12-22 03:10:11 +02:00
|
|
|
}
|
2003-10-26 12:14:05 +02:00
|
|
|
#endif //CONFIG_ENCODERS
|
|
|
|
|
2012-04-12 15:55:49 +03:00
|
|
|
if (avctx->bits_per_raw_sample <= 8) {
|
2008-07-20 21:58:30 +03:00
|
|
|
if ((avctx->idct_algo == FF_IDCT_AUTO) ||
|
|
|
|
(avctx->idct_algo == FF_IDCT_ALTIVEC)) {
|
2012-02-15 15:42:56 +03:00
|
|
|
c->idct_put = ff_idct_put_altivec;
|
|
|
|
c->idct_add = ff_idct_add_altivec;
|
2008-07-20 21:58:30 +03:00
|
|
|
c->idct_permutation_type = FF_TRANSPOSE_IDCT_PERM;
|
|
|
|
}
|
2003-03-03 16:54:00 +02:00
|
|
|
}
|
2005-12-17 20:14:38 +02:00
|
|
|
|
2002-09-02 11:48:12 +03:00
|
|
|
}
|
2006-10-08 16:55:02 +03:00
|
|
|
#endif /* HAVE_ALTIVEC */
|
2002-09-02 11:48:12 +03:00
|
|
|
}
|