2001-07-22 16:18:56 +02:00
|
|
|
/*
|
|
|
|
* MMX optimized DSP utils
|
2009-01-19 16:46:40 +01:00
|
|
|
* Copyright (c) 2000, 2001 Fabrice Bellard
|
2004-01-10 17:04:55 +01:00
|
|
|
* Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
|
2001-07-22 16:18:56 +02:00
|
|
|
*
|
2012-03-06 13:00:42 +01:00
|
|
|
* MMX optimization by Nick Kurshev <nickols_k@mail.ru>
|
|
|
|
*
|
2011-03-18 18:35:10 +01:00
|
|
|
* This file is part of Libav.
|
2006-10-07 17:30:46 +02:00
|
|
|
*
|
2011-03-18 18:35:10 +01:00
|
|
|
* Libav is free software; you can redistribute it and/or
|
2002-05-26 00:45:33 +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 17:30:46 +02:00
|
|
|
* version 2.1 of the License, or (at your option) any later version.
|
2001-07-22 16:18:56 +02:00
|
|
|
*
|
2011-03-18 18:35:10 +01:00
|
|
|
* Libav is distributed in the hope that it will be useful,
|
2001-07-22 16:18:56 +02:00
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
2002-05-26 00:45:33 +02:00
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
* Lesser General Public License for more details.
|
2001-07-22 16:18:56 +02:00
|
|
|
*
|
2002-05-26 00:45:33 +02:00
|
|
|
* You should have received a copy of the GNU Lesser General Public
|
2011-03-18 18:35:10 +01:00
|
|
|
* License along with Libav; if not, write to the Free Software
|
2006-01-12 23:43:26 +01:00
|
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
2001-07-22 16:18:56 +02:00
|
|
|
*/
|
|
|
|
|
2013-05-08 01:45:39 +02:00
|
|
|
#include "config.h"
|
2010-09-04 11:59:08 +02:00
|
|
|
#include "libavutil/cpu.h"
|
2012-08-08 14:51:52 +02:00
|
|
|
#include "libavutil/x86/asm.h"
|
2013-05-08 01:52:57 +02:00
|
|
|
#include "dsputil_x86.h"
|
2014-01-27 15:06:54 +01:00
|
|
|
#include "inline_asm.h"
|
2013-01-27 04:45:43 +01:00
|
|
|
|
2012-07-22 23:14:20 +02:00
|
|
|
#if HAVE_INLINE_ASM
|
|
|
|
|
2013-01-20 01:02:29 +01:00
|
|
|
void ff_put_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels,
|
2012-03-06 13:00:42 +01:00
|
|
|
int line_size)
|
2001-07-22 16:18:56 +02:00
|
|
|
{
|
2013-01-20 01:02:29 +01:00
|
|
|
const int16_t *p;
|
2003-02-11 17:35:48 +01:00
|
|
|
uint8_t *pix;
|
2001-07-22 16:18:56 +02:00
|
|
|
|
|
|
|
/* read the pixels */
|
2012-03-06 13:00:42 +01:00
|
|
|
p = block;
|
2001-07-22 16:18:56 +02:00
|
|
|
pix = pixels;
|
2002-02-18 10:40:05 +01:00
|
|
|
/* unrolled loop */
|
2012-03-16 18:42:01 +01:00
|
|
|
__asm__ volatile (
|
dsputil_mmx: fix incorrect assembly code
In ff_put_pixels_clamped_mmx(), there are two assembly code blocks.
In the first block (in the unrolled loop), the instructions
"movq 8%3, %%mm1 \n\t", and so forth, have problems.
From above instruction, it is clear what the programmer wants: a load from
p + 8. But this assembly code doesn’t guarantee that. It only works if the
compiler puts p in a register to produce an instruction like this:
"movq 8(%edi), %mm1". During compiler optimization, it is possible that the
compiler will be able to constant propagate into p. Suppose p = &x[10000].
Then operand 3 can become 10000(%edi), where %edi holds &x. And the instruction
becomes "movq 810000(%edx)". That is, it will stride by 810000 instead of 8.
This will cause a segmentation fault.
This error was fixed in the second block of the assembly code, but not in
the unrolled loop.
How to reproduce:
This error is exposed when we build using Intel C++ Compiler, with
IPO+PGO optimization enabled. Crashed when decoding an MJPEG video.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>
2012-07-24 00:51:10 +02:00
|
|
|
"movq (%3), %%mm0 \n\t"
|
|
|
|
"movq 8(%3), %%mm1 \n\t"
|
|
|
|
"movq 16(%3), %%mm2 \n\t"
|
|
|
|
"movq 24(%3), %%mm3 \n\t"
|
|
|
|
"movq 32(%3), %%mm4 \n\t"
|
|
|
|
"movq 40(%3), %%mm5 \n\t"
|
|
|
|
"movq 48(%3), %%mm6 \n\t"
|
|
|
|
"movq 56(%3), %%mm7 \n\t"
|
2012-03-16 18:42:01 +01:00
|
|
|
"packuswb %%mm1, %%mm0 \n\t"
|
|
|
|
"packuswb %%mm3, %%mm2 \n\t"
|
|
|
|
"packuswb %%mm5, %%mm4 \n\t"
|
|
|
|
"packuswb %%mm7, %%mm6 \n\t"
|
|
|
|
"movq %%mm0, (%0) \n\t"
|
|
|
|
"movq %%mm2, (%0, %1) \n\t"
|
|
|
|
"movq %%mm4, (%0, %1, 2) \n\t"
|
|
|
|
"movq %%mm6, (%0, %2) \n\t"
|
2013-12-25 15:49:31 +01:00
|
|
|
:: "r" (pix), "r" ((x86_reg) line_size), "r" ((x86_reg) line_size * 3),
|
|
|
|
"r" (p)
|
2012-03-16 18:42:01 +01:00
|
|
|
: "memory");
|
2012-03-06 13:00:42 +01:00
|
|
|
pix += line_size * 4;
|
|
|
|
p += 32;
|
2002-02-18 10:40:05 +01:00
|
|
|
|
|
|
|
// if here would be an exact copy of the code above
|
|
|
|
// compiler would generate some very strange code
|
|
|
|
// thus using "r"
|
2012-03-16 18:42:01 +01:00
|
|
|
__asm__ volatile (
|
|
|
|
"movq (%3), %%mm0 \n\t"
|
|
|
|
"movq 8(%3), %%mm1 \n\t"
|
|
|
|
"movq 16(%3), %%mm2 \n\t"
|
|
|
|
"movq 24(%3), %%mm3 \n\t"
|
|
|
|
"movq 32(%3), %%mm4 \n\t"
|
|
|
|
"movq 40(%3), %%mm5 \n\t"
|
|
|
|
"movq 48(%3), %%mm6 \n\t"
|
|
|
|
"movq 56(%3), %%mm7 \n\t"
|
|
|
|
"packuswb %%mm1, %%mm0 \n\t"
|
|
|
|
"packuswb %%mm3, %%mm2 \n\t"
|
|
|
|
"packuswb %%mm5, %%mm4 \n\t"
|
|
|
|
"packuswb %%mm7, %%mm6 \n\t"
|
|
|
|
"movq %%mm0, (%0) \n\t"
|
|
|
|
"movq %%mm2, (%0, %1) \n\t"
|
|
|
|
"movq %%mm4, (%0, %1, 2) \n\t"
|
|
|
|
"movq %%mm6, (%0, %2) \n\t"
|
2013-12-25 15:49:31 +01:00
|
|
|
:: "r" (pix), "r" ((x86_reg) line_size), "r" ((x86_reg) line_size * 3),
|
|
|
|
"r" (p)
|
2012-03-16 18:42:01 +01:00
|
|
|
: "memory");
|
2001-07-22 16:18:56 +02:00
|
|
|
}
|
|
|
|
|
2012-03-16 18:42:01 +01:00
|
|
|
#define put_signed_pixels_clamped_mmx_half(off) \
|
|
|
|
"movq "#off"(%2), %%mm1 \n\t" \
|
|
|
|
"movq 16 + "#off"(%2), %%mm2 \n\t" \
|
|
|
|
"movq 32 + "#off"(%2), %%mm3 \n\t" \
|
|
|
|
"movq 48 + "#off"(%2), %%mm4 \n\t" \
|
|
|
|
"packsswb 8 + "#off"(%2), %%mm1 \n\t" \
|
|
|
|
"packsswb 24 + "#off"(%2), %%mm2 \n\t" \
|
|
|
|
"packsswb 40 + "#off"(%2), %%mm3 \n\t" \
|
|
|
|
"packsswb 56 + "#off"(%2), %%mm4 \n\t" \
|
|
|
|
"paddb %%mm0, %%mm1 \n\t" \
|
|
|
|
"paddb %%mm0, %%mm2 \n\t" \
|
|
|
|
"paddb %%mm0, %%mm3 \n\t" \
|
|
|
|
"paddb %%mm0, %%mm4 \n\t" \
|
|
|
|
"movq %%mm1, (%0) \n\t" \
|
|
|
|
"movq %%mm2, (%0, %3) \n\t" \
|
|
|
|
"movq %%mm3, (%0, %3, 2) \n\t" \
|
|
|
|
"movq %%mm4, (%0, %1) \n\t"
|
2009-04-02 23:02:42 +02:00
|
|
|
|
2013-01-20 01:02:29 +01:00
|
|
|
void ff_put_signed_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels,
|
2012-03-06 13:00:42 +01:00
|
|
|
int line_size)
|
2004-04-27 05:58:06 +02:00
|
|
|
{
|
2009-04-02 23:02:42 +02:00
|
|
|
x86_reg line_skip = line_size;
|
2009-04-03 16:03:49 +02:00
|
|
|
x86_reg line_skip3;
|
2009-04-02 23:02:42 +02:00
|
|
|
|
|
|
|
__asm__ volatile (
|
2012-03-16 18:42:01 +01:00
|
|
|
"movq "MANGLE(ff_pb_80)", %%mm0 \n\t"
|
|
|
|
"lea (%3, %3, 2), %1 \n\t"
|
|
|
|
put_signed_pixels_clamped_mmx_half(0)
|
|
|
|
"lea (%0, %3, 4), %0 \n\t"
|
|
|
|
put_signed_pixels_clamped_mmx_half(64)
|
2013-12-25 15:49:31 +01:00
|
|
|
: "+&r" (pixels), "=&r" (line_skip3)
|
|
|
|
: "r" (block), "r" (line_skip)
|
2012-03-16 18:42:01 +01:00
|
|
|
: "memory");
|
2004-04-27 05:58:06 +02:00
|
|
|
}
|
|
|
|
|
2013-01-20 01:02:29 +01:00
|
|
|
void ff_add_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels,
|
2012-03-06 13:00:42 +01:00
|
|
|
int line_size)
|
2001-07-22 16:18:56 +02:00
|
|
|
{
|
2013-01-20 01:02:29 +01:00
|
|
|
const int16_t *p;
|
2003-02-11 17:35:48 +01:00
|
|
|
uint8_t *pix;
|
2001-07-22 16:18:56 +02:00
|
|
|
int i;
|
|
|
|
|
|
|
|
/* read the pixels */
|
2012-03-06 13:00:42 +01:00
|
|
|
p = block;
|
2001-07-22 16:18:56 +02:00
|
|
|
pix = pixels;
|
2002-02-18 10:40:05 +01:00
|
|
|
MOVQ_ZERO(mm7);
|
|
|
|
i = 4;
|
2002-04-24 14:18:59 +02:00
|
|
|
do {
|
2012-03-16 18:42:01 +01:00
|
|
|
__asm__ volatile (
|
|
|
|
"movq (%2), %%mm0 \n\t"
|
|
|
|
"movq 8(%2), %%mm1 \n\t"
|
|
|
|
"movq 16(%2), %%mm2 \n\t"
|
|
|
|
"movq 24(%2), %%mm3 \n\t"
|
|
|
|
"movq %0, %%mm4 \n\t"
|
|
|
|
"movq %1, %%mm6 \n\t"
|
|
|
|
"movq %%mm4, %%mm5 \n\t"
|
|
|
|
"punpcklbw %%mm7, %%mm4 \n\t"
|
|
|
|
"punpckhbw %%mm7, %%mm5 \n\t"
|
|
|
|
"paddsw %%mm4, %%mm0 \n\t"
|
|
|
|
"paddsw %%mm5, %%mm1 \n\t"
|
|
|
|
"movq %%mm6, %%mm5 \n\t"
|
|
|
|
"punpcklbw %%mm7, %%mm6 \n\t"
|
|
|
|
"punpckhbw %%mm7, %%mm5 \n\t"
|
|
|
|
"paddsw %%mm6, %%mm2 \n\t"
|
|
|
|
"paddsw %%mm5, %%mm3 \n\t"
|
|
|
|
"packuswb %%mm1, %%mm0 \n\t"
|
|
|
|
"packuswb %%mm3, %%mm2 \n\t"
|
|
|
|
"movq %%mm0, %0 \n\t"
|
|
|
|
"movq %%mm2, %1 \n\t"
|
2013-12-25 15:49:31 +01:00
|
|
|
: "+m" (*pix), "+m" (*(pix + line_size))
|
|
|
|
: "r" (p)
|
2012-03-16 18:42:01 +01:00
|
|
|
: "memory");
|
2012-03-06 13:00:42 +01:00
|
|
|
pix += line_size * 2;
|
|
|
|
p += 16;
|
2002-04-24 14:18:59 +02:00
|
|
|
} while (--i);
|
2001-07-22 16:18:56 +02:00
|
|
|
}
|
|
|
|
|
2012-03-06 13:00:42 +01:00
|
|
|
/* Draw the edges of width 'w' of an image of size width, height
|
|
|
|
* this MMX version can only handle w == 8 || w == 16. */
|
2013-05-08 01:45:39 +02:00
|
|
|
void ff_draw_edges_mmx(uint8_t *buf, int wrap, int width, int height,
|
|
|
|
int w, int h, int sides)
|
2008-03-04 01:07:41 +01:00
|
|
|
{
|
|
|
|
uint8_t *ptr, *last_line;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
last_line = buf + (height - 1) * wrap;
|
|
|
|
/* left and right */
|
|
|
|
ptr = buf;
|
2012-03-06 13:00:42 +01:00
|
|
|
if (w == 8) {
|
2012-03-16 18:42:01 +01:00
|
|
|
__asm__ volatile (
|
|
|
|
"1: \n\t"
|
|
|
|
"movd (%0), %%mm0 \n\t"
|
|
|
|
"punpcklbw %%mm0, %%mm0 \n\t"
|
|
|
|
"punpcklwd %%mm0, %%mm0 \n\t"
|
|
|
|
"punpckldq %%mm0, %%mm0 \n\t"
|
|
|
|
"movq %%mm0, -8(%0) \n\t"
|
|
|
|
"movq -8(%0, %2), %%mm1 \n\t"
|
|
|
|
"punpckhbw %%mm1, %%mm1 \n\t"
|
|
|
|
"punpckhwd %%mm1, %%mm1 \n\t"
|
|
|
|
"punpckhdq %%mm1, %%mm1 \n\t"
|
|
|
|
"movq %%mm1, (%0, %2) \n\t"
|
|
|
|
"add %1, %0 \n\t"
|
|
|
|
"cmp %3, %0 \n\t"
|
|
|
|
"jb 1b \n\t"
|
2013-12-25 15:49:31 +01:00
|
|
|
: "+r" (ptr)
|
|
|
|
: "r" ((x86_reg) wrap), "r" ((x86_reg) width),
|
|
|
|
"r" (ptr + wrap * height));
|
2012-03-06 13:00:42 +01:00
|
|
|
} else {
|
2012-03-16 18:42:01 +01:00
|
|
|
__asm__ volatile (
|
|
|
|
"1: \n\t"
|
|
|
|
"movd (%0), %%mm0 \n\t"
|
|
|
|
"punpcklbw %%mm0, %%mm0 \n\t"
|
|
|
|
"punpcklwd %%mm0, %%mm0 \n\t"
|
|
|
|
"punpckldq %%mm0, %%mm0 \n\t"
|
|
|
|
"movq %%mm0, -8(%0) \n\t"
|
|
|
|
"movq %%mm0, -16(%0) \n\t"
|
|
|
|
"movq -8(%0, %2), %%mm1 \n\t"
|
|
|
|
"punpckhbw %%mm1, %%mm1 \n\t"
|
|
|
|
"punpckhwd %%mm1, %%mm1 \n\t"
|
|
|
|
"punpckhdq %%mm1, %%mm1 \n\t"
|
|
|
|
"movq %%mm1, (%0, %2) \n\t"
|
|
|
|
"movq %%mm1, 8(%0, %2) \n\t"
|
|
|
|
"add %1, %0 \n\t"
|
|
|
|
"cmp %3, %0 \n\t"
|
|
|
|
"jb 1b \n\t"
|
2013-12-25 15:49:31 +01:00
|
|
|
: "+r" (ptr)
|
|
|
|
: "r" ((x86_reg) wrap), "r" ((x86_reg) width),
|
|
|
|
"r" (ptr + wrap * height));
|
2008-03-04 01:07:41 +01:00
|
|
|
}
|
|
|
|
|
2011-03-26 22:31:13 +01:00
|
|
|
/* top and bottom (and hopefully also the corners) */
|
2012-03-06 13:00:42 +01:00
|
|
|
if (sides & EDGE_TOP) {
|
|
|
|
for (i = 0; i < h; i += 4) {
|
|
|
|
ptr = buf - (i + 1) * wrap - w;
|
2012-03-16 18:42:01 +01:00
|
|
|
__asm__ volatile (
|
|
|
|
"1: \n\t"
|
|
|
|
"movq (%1, %0), %%mm0 \n\t"
|
|
|
|
"movq %%mm0, (%0) \n\t"
|
|
|
|
"movq %%mm0, (%0, %2) \n\t"
|
|
|
|
"movq %%mm0, (%0, %2, 2) \n\t"
|
|
|
|
"movq %%mm0, (%0, %3) \n\t"
|
|
|
|
"add $8, %0 \n\t"
|
|
|
|
"cmp %4, %0 \n\t"
|
|
|
|
"jb 1b \n\t"
|
2013-12-25 15:49:31 +01:00
|
|
|
: "+r" (ptr)
|
|
|
|
: "r" ((x86_reg) buf - (x86_reg) ptr - w),
|
|
|
|
"r" ((x86_reg) - wrap), "r" ((x86_reg) - wrap * 3),
|
|
|
|
"r" (ptr + width + 2 * w));
|
2011-03-26 22:31:13 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-03-06 13:00:42 +01:00
|
|
|
if (sides & EDGE_BOTTOM) {
|
|
|
|
for (i = 0; i < h; i += 4) {
|
|
|
|
ptr = last_line + (i + 1) * wrap - w;
|
2012-03-16 18:42:01 +01:00
|
|
|
__asm__ volatile (
|
|
|
|
"1: \n\t"
|
|
|
|
"movq (%1, %0), %%mm0 \n\t"
|
|
|
|
"movq %%mm0, (%0) \n\t"
|
|
|
|
"movq %%mm0, (%0, %2) \n\t"
|
|
|
|
"movq %%mm0, (%0, %2, 2) \n\t"
|
|
|
|
"movq %%mm0, (%0, %3) \n\t"
|
|
|
|
"add $8, %0 \n\t"
|
|
|
|
"cmp %4, %0 \n\t"
|
|
|
|
"jb 1b \n\t"
|
2013-12-25 15:49:31 +01:00
|
|
|
: "+r" (ptr)
|
|
|
|
: "r" ((x86_reg) last_line - (x86_reg) ptr - w),
|
|
|
|
"r" ((x86_reg) wrap), "r" ((x86_reg) wrap * 3),
|
|
|
|
"r" (ptr + width + 2 * w));
|
2011-03-26 22:31:13 +01:00
|
|
|
}
|
2008-03-04 01:07:41 +01:00
|
|
|
}
|
|
|
|
}
|
2003-01-05 16:57:10 +01:00
|
|
|
|
2013-05-08 01:45:39 +02:00
|
|
|
void ff_gmc_mmx(uint8_t *dst, uint8_t *src,
|
|
|
|
int stride, int h, int ox, int oy,
|
|
|
|
int dxx, int dxy, int dyx, int dyy,
|
|
|
|
int shift, int r, int width, int height)
|
2011-02-01 02:55:56 +01:00
|
|
|
{
|
2012-03-06 13:00:42 +01:00
|
|
|
const int w = 8;
|
|
|
|
const int ix = ox >> (16 + shift);
|
|
|
|
const int iy = oy >> (16 + shift);
|
|
|
|
const int oxs = ox >> 4;
|
|
|
|
const int oys = oy >> 4;
|
|
|
|
const int dxxs = dxx >> 4;
|
|
|
|
const int dxys = dxy >> 4;
|
|
|
|
const int dyxs = dyx >> 4;
|
|
|
|
const int dyys = dyy >> 4;
|
|
|
|
const uint16_t r4[4] = { r, r, r, r };
|
|
|
|
const uint16_t dxy4[4] = { dxys, dxys, dxys, dxys };
|
|
|
|
const uint16_t dyy4[4] = { dyys, dyys, dyys, dyys };
|
2013-12-25 15:49:31 +01:00
|
|
|
const uint64_t shift2 = 2 * shift;
|
2006-04-04 11:23:45 +02:00
|
|
|
int x, y;
|
|
|
|
|
2012-03-06 13:00:42 +01:00
|
|
|
const int dxw = (dxx - (1 << (16 + shift))) * (w - 1);
|
|
|
|
const int dyh = (dyy - (1 << (16 + shift))) * (h - 1);
|
|
|
|
const int dxh = dxy * (h - 1);
|
|
|
|
const int dyw = dyx * (w - 1);
|
2013-12-25 15:49:31 +01:00
|
|
|
|
2012-03-06 13:00:42 +01:00
|
|
|
if ( // non-constant fullpel offset (3% of blocks)
|
|
|
|
((ox ^ (ox + dxw)) | (ox ^ (ox + dxh)) | (ox ^ (ox + dxw + dxh)) |
|
2013-12-25 15:49:31 +01:00
|
|
|
(oy ^ (oy + dyw)) | (oy ^ (oy + dyh)) | (oy ^ (oy + dyw + dyh))) >> (16 + shift) ||
|
2006-04-04 11:23:45 +02:00
|
|
|
// uses more than 16 bits of subpel mv (only at huge resolution)
|
2013-12-25 15:49:31 +01:00
|
|
|
(dxx | dxy | dyx | dyy) & 15 ||
|
|
|
|
(unsigned) ix >= width - w ||
|
|
|
|
(unsigned) iy >= height - h) {
|
2012-03-16 18:42:01 +01:00
|
|
|
// FIXME could still use mmx for some of the rows
|
2012-03-06 13:00:42 +01:00
|
|
|
ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy,
|
|
|
|
shift, r, width, height);
|
2006-04-04 11:23:45 +02:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2012-03-06 13:00:42 +01:00
|
|
|
src += ix + iy * stride;
|
2006-04-05 06:13:41 +02:00
|
|
|
|
2012-03-16 18:42:01 +01:00
|
|
|
__asm__ volatile (
|
|
|
|
"movd %0, %%mm6 \n\t"
|
|
|
|
"pxor %%mm7, %%mm7 \n\t"
|
|
|
|
"punpcklwd %%mm6, %%mm6 \n\t"
|
|
|
|
"punpcklwd %%mm6, %%mm6 \n\t"
|
2013-12-25 15:49:31 +01:00
|
|
|
:: "r" (1 << shift));
|
2006-04-04 11:23:45 +02:00
|
|
|
|
2012-03-06 13:00:42 +01:00
|
|
|
for (x = 0; x < w; x += 4) {
|
|
|
|
uint16_t dx4[4] = { oxs - dxys + dxxs * (x + 0),
|
|
|
|
oxs - dxys + dxxs * (x + 1),
|
|
|
|
oxs - dxys + dxxs * (x + 2),
|
|
|
|
oxs - dxys + dxxs * (x + 3) };
|
|
|
|
uint16_t dy4[4] = { oys - dyys + dyxs * (x + 0),
|
|
|
|
oys - dyys + dyxs * (x + 1),
|
|
|
|
oys - dyys + dyxs * (x + 2),
|
|
|
|
oys - dyys + dyxs * (x + 3) };
|
|
|
|
|
|
|
|
for (y = 0; y < h; y++) {
|
2012-03-16 18:42:01 +01:00
|
|
|
__asm__ volatile (
|
|
|
|
"movq %0, %%mm4 \n\t"
|
|
|
|
"movq %1, %%mm5 \n\t"
|
|
|
|
"paddw %2, %%mm4 \n\t"
|
|
|
|
"paddw %3, %%mm5 \n\t"
|
|
|
|
"movq %%mm4, %0 \n\t"
|
|
|
|
"movq %%mm5, %1 \n\t"
|
|
|
|
"psrlw $12, %%mm4 \n\t"
|
|
|
|
"psrlw $12, %%mm5 \n\t"
|
2013-12-25 15:49:31 +01:00
|
|
|
: "+m" (*dx4), "+m" (*dy4)
|
|
|
|
: "m" (*dxy4), "m" (*dyy4));
|
2006-04-04 11:23:45 +02:00
|
|
|
|
2012-03-16 18:42:01 +01:00
|
|
|
__asm__ volatile (
|
|
|
|
"movq %%mm6, %%mm2 \n\t"
|
|
|
|
"movq %%mm6, %%mm1 \n\t"
|
|
|
|
"psubw %%mm4, %%mm2 \n\t"
|
|
|
|
"psubw %%mm5, %%mm1 \n\t"
|
|
|
|
"movq %%mm2, %%mm0 \n\t"
|
|
|
|
"movq %%mm4, %%mm3 \n\t"
|
|
|
|
"pmullw %%mm1, %%mm0 \n\t" // (s - dx) * (s - dy)
|
|
|
|
"pmullw %%mm5, %%mm3 \n\t" // dx * dy
|
|
|
|
"pmullw %%mm5, %%mm2 \n\t" // (s - dx) * dy
|
|
|
|
"pmullw %%mm4, %%mm1 \n\t" // dx * (s - dy)
|
|
|
|
|
|
|
|
"movd %4, %%mm5 \n\t"
|
|
|
|
"movd %3, %%mm4 \n\t"
|
2006-04-04 11:23:45 +02:00
|
|
|
"punpcklbw %%mm7, %%mm5 \n\t"
|
|
|
|
"punpcklbw %%mm7, %%mm4 \n\t"
|
2012-03-16 18:42:01 +01:00
|
|
|
"pmullw %%mm5, %%mm3 \n\t" // src[1, 1] * dx * dy
|
|
|
|
"pmullw %%mm4, %%mm2 \n\t" // src[0, 1] * (s - dx) * dy
|
2006-04-04 11:23:45 +02:00
|
|
|
|
2012-03-16 18:42:01 +01:00
|
|
|
"movd %2, %%mm5 \n\t"
|
|
|
|
"movd %1, %%mm4 \n\t"
|
2006-04-04 11:23:45 +02:00
|
|
|
"punpcklbw %%mm7, %%mm5 \n\t"
|
|
|
|
"punpcklbw %%mm7, %%mm4 \n\t"
|
2012-03-16 18:42:01 +01:00
|
|
|
"pmullw %%mm5, %%mm1 \n\t" // src[1, 0] * dx * (s - dy)
|
|
|
|
"pmullw %%mm4, %%mm0 \n\t" // src[0, 0] * (s - dx) * (s - dy)
|
|
|
|
"paddw %5, %%mm1 \n\t"
|
|
|
|
"paddw %%mm3, %%mm2 \n\t"
|
|
|
|
"paddw %%mm1, %%mm0 \n\t"
|
|
|
|
"paddw %%mm2, %%mm0 \n\t"
|
|
|
|
|
|
|
|
"psrlw %6, %%mm0 \n\t"
|
|
|
|
"packuswb %%mm0, %%mm0 \n\t"
|
|
|
|
"movd %%mm0, %0 \n\t"
|
|
|
|
|
2013-12-25 15:49:31 +01:00
|
|
|
: "=m" (dst[x + y * stride])
|
|
|
|
: "m" (src[0]), "m" (src[1]),
|
|
|
|
"m" (src[stride]), "m" (src[stride + 1]),
|
|
|
|
"m" (*r4), "m" (shift2));
|
2006-04-04 11:23:45 +02:00
|
|
|
src += stride;
|
|
|
|
}
|
2012-03-06 13:00:42 +01:00
|
|
|
src += 4 - h * stride;
|
2006-04-04 11:23:45 +02:00
|
|
|
}
|
|
|
|
}
|
2012-07-22 23:14:20 +02:00
|
|
|
|
|
|
|
#endif /* HAVE_INLINE_ASM */
|