2002-05-29 19:16:22 +02:00
|
|
|
/*
|
|
|
|
|
* DSP utils mmx functions are compiled twice for rnd/no_rnd
|
|
|
|
|
* Copyright (c) 2000, 2001 Fabrice Bellard.
|
2004-01-10 17:04:55 +01:00
|
|
|
* Copyright (c) 2003-2004 Michael Niedermayer <michaelni@gmx.at>
|
2002-05-29 19:16:22 +02:00
|
|
|
*
|
2006-10-07 17:30:46 +02:00
|
|
|
* This file is part of FFmpeg.
|
|
|
|
|
*
|
|
|
|
|
* FFmpeg is free software; you can redistribute it and/or
|
2002-05-29 19:16:22 +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.
|
2002-05-29 19:16:22 +02:00
|
|
|
*
|
2006-10-07 17:30:46 +02:00
|
|
|
* FFmpeg is distributed in the hope that it will be useful,
|
2002-05-29 19:16:22 +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
|
2006-10-07 17:30:46 +02:00
|
|
|
* License along with FFmpeg; 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
|
2002-05-29 19:16:22 +02:00
|
|
|
*
|
|
|
|
|
* MMX optimization by Nick Kurshev <nickols_k@mail.ru>
|
|
|
|
|
* mostly rewritten by Michael Niedermayer <michaelni@gmx.at>
|
|
|
|
|
* and improved by Zdenek Kabelac <kabi@users.sf.net>
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
// put_pixels
|
2003-02-11 17:35:48 +01:00
|
|
|
static void DEF(put, pixels8_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
|
2002-05-29 19:16:22 +02:00
|
|
|
{
|
2002-05-30 17:14:56 +02:00
|
|
|
MOVQ_BFE(mm6);
|
2002-05-29 19:16:22 +02:00
|
|
|
__asm __volatile(
|
2005-12-22 02:10:11 +01:00
|
|
|
"lea (%3, %3), %%"REG_a" \n\t"
|
2006-08-12 18:37:31 +02:00
|
|
|
ASMALIGN(3)
|
2005-12-22 02:10:11 +01:00
|
|
|
"1: \n\t"
|
|
|
|
|
"movq (%1), %%mm0 \n\t"
|
|
|
|
|
"movq 1(%1), %%mm1 \n\t"
|
|
|
|
|
"movq (%1, %3), %%mm2 \n\t"
|
|
|
|
|
"movq 1(%1, %3), %%mm3 \n\t"
|
|
|
|
|
PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
|
|
|
|
|
"movq %%mm4, (%2) \n\t"
|
|
|
|
|
"movq %%mm5, (%2, %3) \n\t"
|
|
|
|
|
"add %%"REG_a", %1 \n\t"
|
|
|
|
|
"add %%"REG_a", %2 \n\t"
|
|
|
|
|
"movq (%1), %%mm0 \n\t"
|
|
|
|
|
"movq 1(%1), %%mm1 \n\t"
|
|
|
|
|
"movq (%1, %3), %%mm2 \n\t"
|
|
|
|
|
"movq 1(%1, %3), %%mm3 \n\t"
|
|
|
|
|
PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
|
|
|
|
|
"movq %%mm4, (%2) \n\t"
|
|
|
|
|
"movq %%mm5, (%2, %3) \n\t"
|
|
|
|
|
"add %%"REG_a", %1 \n\t"
|
|
|
|
|
"add %%"REG_a", %2 \n\t"
|
|
|
|
|
"subl $4, %0 \n\t"
|
|
|
|
|
"jnz 1b \n\t"
|
|
|
|
|
:"+g"(h), "+S"(pixels), "+D"(block)
|
|
|
|
|
:"r"((long)line_size)
|
|
|
|
|
:REG_a, "memory");
|
2002-05-29 19:16:22 +02:00
|
|
|
}
|
|
|
|
|
|
2005-09-06 23:25:35 +02:00
|
|
|
static void attribute_unused DEF(put, pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
|
2003-01-05 16:57:10 +01:00
|
|
|
{
|
|
|
|
|
MOVQ_BFE(mm6);
|
|
|
|
|
__asm __volatile(
|
2005-12-22 02:10:11 +01:00
|
|
|
"testl $1, %0 \n\t"
|
|
|
|
|
" jz 1f \n\t"
|
|
|
|
|
"movq (%1), %%mm0 \n\t"
|
|
|
|
|
"movq (%2), %%mm1 \n\t"
|
|
|
|
|
"add %4, %1 \n\t"
|
|
|
|
|
"add $8, %2 \n\t"
|
|
|
|
|
PAVGB(%%mm0, %%mm1, %%mm4, %%mm6)
|
|
|
|
|
"movq %%mm4, (%3) \n\t"
|
|
|
|
|
"add %5, %3 \n\t"
|
|
|
|
|
"decl %0 \n\t"
|
2006-08-12 18:37:31 +02:00
|
|
|
ASMALIGN(3)
|
2005-12-22 02:10:11 +01:00
|
|
|
"1: \n\t"
|
|
|
|
|
"movq (%1), %%mm0 \n\t"
|
|
|
|
|
"movq (%2), %%mm1 \n\t"
|
|
|
|
|
"add %4, %1 \n\t"
|
|
|
|
|
"movq (%1), %%mm2 \n\t"
|
|
|
|
|
"movq 8(%2), %%mm3 \n\t"
|
|
|
|
|
"add %4, %1 \n\t"
|
|
|
|
|
PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
|
|
|
|
|
"movq %%mm4, (%3) \n\t"
|
|
|
|
|
"add %5, %3 \n\t"
|
|
|
|
|
"movq %%mm5, (%3) \n\t"
|
|
|
|
|
"add %5, %3 \n\t"
|
|
|
|
|
"movq (%1), %%mm0 \n\t"
|
|
|
|
|
"movq 16(%2), %%mm1 \n\t"
|
|
|
|
|
"add %4, %1 \n\t"
|
|
|
|
|
"movq (%1), %%mm2 \n\t"
|
|
|
|
|
"movq 24(%2), %%mm3 \n\t"
|
|
|
|
|
"add %4, %1 \n\t"
|
|
|
|
|
"add $32, %2 \n\t"
|
|
|
|
|
PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
|
|
|
|
|
"movq %%mm4, (%3) \n\t"
|
|
|
|
|
"add %5, %3 \n\t"
|
|
|
|
|
"movq %%mm5, (%3) \n\t"
|
|
|
|
|
"add %5, %3 \n\t"
|
|
|
|
|
"subl $4, %0 \n\t"
|
|
|
|
|
"jnz 1b \n\t"
|
2003-01-07 14:00:38 +01:00
|
|
|
#ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cant be used
|
|
|
|
|
:"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
|
|
|
|
|
#else
|
|
|
|
|
:"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
|
|
|
|
|
#endif
|
2005-12-22 02:10:11 +01:00
|
|
|
:"S"((long)src1Stride), "D"((long)dstStride)
|
|
|
|
|
:"memory");
|
2003-01-05 16:57:10 +01:00
|
|
|
}
|
|
|
|
|
|
2003-02-11 17:35:48 +01:00
|
|
|
static void DEF(put, pixels16_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
|
2002-09-11 14:39:53 +02:00
|
|
|
{
|
|
|
|
|
MOVQ_BFE(mm6);
|
|
|
|
|
__asm __volatile(
|
2005-12-22 02:10:11 +01:00
|
|
|
"lea (%3, %3), %%"REG_a" \n\t"
|
2006-08-12 18:37:31 +02:00
|
|
|
ASMALIGN(3)
|
2005-12-22 02:10:11 +01:00
|
|
|
"1: \n\t"
|
|
|
|
|
"movq (%1), %%mm0 \n\t"
|
|
|
|
|
"movq 1(%1), %%mm1 \n\t"
|
|
|
|
|
"movq (%1, %3), %%mm2 \n\t"
|
|
|
|
|
"movq 1(%1, %3), %%mm3 \n\t"
|
|
|
|
|
PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
|
|
|
|
|
"movq %%mm4, (%2) \n\t"
|
|
|
|
|
"movq %%mm5, (%2, %3) \n\t"
|
|
|
|
|
"movq 8(%1), %%mm0 \n\t"
|
|
|
|
|
"movq 9(%1), %%mm1 \n\t"
|
|
|
|
|
"movq 8(%1, %3), %%mm2 \n\t"
|
|
|
|
|
"movq 9(%1, %3), %%mm3 \n\t"
|
|
|
|
|
PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
|
|
|
|
|
"movq %%mm4, 8(%2) \n\t"
|
|
|
|
|
"movq %%mm5, 8(%2, %3) \n\t"
|
|
|
|
|
"add %%"REG_a", %1 \n\t"
|
|
|
|
|
"add %%"REG_a", %2 \n\t"
|
|
|
|
|
"movq (%1), %%mm0 \n\t"
|
|
|
|
|
"movq 1(%1), %%mm1 \n\t"
|
|
|
|
|
"movq (%1, %3), %%mm2 \n\t"
|
|
|
|
|
"movq 1(%1, %3), %%mm3 \n\t"
|
|
|
|
|
PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
|
|
|
|
|
"movq %%mm4, (%2) \n\t"
|
|
|
|
|
"movq %%mm5, (%2, %3) \n\t"
|
|
|
|
|
"movq 8(%1), %%mm0 \n\t"
|
|
|
|
|
"movq 9(%1), %%mm1 \n\t"
|
|
|
|
|
"movq 8(%1, %3), %%mm2 \n\t"
|
|
|
|
|
"movq 9(%1, %3), %%mm3 \n\t"
|
|
|
|
|
PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
|
|
|
|
|
"movq %%mm4, 8(%2) \n\t"
|
|
|
|
|
"movq %%mm5, 8(%2, %3) \n\t"
|
|
|
|
|
"add %%"REG_a", %1 \n\t"
|
|
|
|
|
"add %%"REG_a", %2 \n\t"
|
|
|
|
|
"subl $4, %0 \n\t"
|
|
|
|
|
"jnz 1b \n\t"
|
|
|
|
|
:"+g"(h), "+S"(pixels), "+D"(block)
|
|
|
|
|
:"r"((long)line_size)
|
|
|
|
|
:REG_a, "memory");
|
2002-09-11 14:39:53 +02:00
|
|
|
}
|
|
|
|
|
|
2005-09-06 23:25:35 +02:00
|
|
|
static void attribute_unused DEF(put, pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
|
2003-01-05 16:57:10 +01:00
|
|
|
{
|
|
|
|
|
MOVQ_BFE(mm6);
|
|
|
|
|
__asm __volatile(
|
2005-12-22 02:10:11 +01:00
|
|
|
"testl $1, %0 \n\t"
|
|
|
|
|
" jz 1f \n\t"
|
|
|
|
|
"movq (%1), %%mm0 \n\t"
|
|
|
|
|
"movq (%2), %%mm1 \n\t"
|
|
|
|
|
"movq 8(%1), %%mm2 \n\t"
|
|
|
|
|
"movq 8(%2), %%mm3 \n\t"
|
|
|
|
|
"add %4, %1 \n\t"
|
|
|
|
|
"add $16, %2 \n\t"
|
|
|
|
|
PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
|
|
|
|
|
"movq %%mm4, (%3) \n\t"
|
|
|
|
|
"movq %%mm5, 8(%3) \n\t"
|
|
|
|
|
"add %5, %3 \n\t"
|
|
|
|
|
"decl %0 \n\t"
|
2006-08-12 18:37:31 +02:00
|
|
|
ASMALIGN(3)
|
2005-12-22 02:10:11 +01:00
|
|
|
"1: \n\t"
|
|
|
|
|
"movq (%1), %%mm0 \n\t"
|
|
|
|
|
"movq (%2), %%mm1 \n\t"
|
|
|
|
|
"movq 8(%1), %%mm2 \n\t"
|
|
|
|
|
"movq 8(%2), %%mm3 \n\t"
|
|
|
|
|
"add %4, %1 \n\t"
|
|
|
|
|
PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
|
|
|
|
|
"movq %%mm4, (%3) \n\t"
|
|
|
|
|
"movq %%mm5, 8(%3) \n\t"
|
|
|
|
|
"add %5, %3 \n\t"
|
|
|
|
|
"movq (%1), %%mm0 \n\t"
|
|
|
|
|
"movq 16(%2), %%mm1 \n\t"
|
|
|
|
|
"movq 8(%1), %%mm2 \n\t"
|
|
|
|
|
"movq 24(%2), %%mm3 \n\t"
|
|
|
|
|
"add %4, %1 \n\t"
|
|
|
|
|
PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
|
|
|
|
|
"movq %%mm4, (%3) \n\t"
|
|
|
|
|
"movq %%mm5, 8(%3) \n\t"
|
|
|
|
|
"add %5, %3 \n\t"
|
|
|
|
|
"add $32, %2 \n\t"
|
|
|
|
|
"subl $2, %0 \n\t"
|
|
|
|
|
"jnz 1b \n\t"
|
2003-01-07 14:00:38 +01:00
|
|
|
#ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cant be used
|
2005-12-22 02:10:11 +01:00
|
|
|
:"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
|
2003-01-07 14:00:38 +01:00
|
|
|
#else
|
2005-12-22 02:10:11 +01:00
|
|
|
:"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
|
2003-01-07 14:00:38 +01:00
|
|
|
#endif
|
2005-12-22 02:10:11 +01:00
|
|
|
:"S"((long)src1Stride), "D"((long)dstStride)
|
|
|
|
|
:"memory");
|
2003-01-05 16:57:10 +01:00
|
|
|
}
|
|
|
|
|
|
2003-02-11 17:35:48 +01:00
|
|
|
static void DEF(put, pixels8_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
|
2002-05-29 19:16:22 +02:00
|
|
|
{
|
2002-05-30 17:14:56 +02:00
|
|
|
MOVQ_BFE(mm6);
|
|
|
|
|
__asm __volatile(
|
2005-12-22 02:10:11 +01:00
|
|
|
"lea (%3, %3), %%"REG_a" \n\t"
|
|
|
|
|
"movq (%1), %%mm0 \n\t"
|
2006-08-12 18:37:31 +02:00
|
|
|
ASMALIGN(3)
|
2005-12-22 02:10:11 +01:00
|
|
|
"1: \n\t"
|
|
|
|
|
"movq (%1, %3), %%mm1 \n\t"
|
|
|
|
|
"movq (%1, %%"REG_a"),%%mm2 \n\t"
|
|
|
|
|
PAVGBP(%%mm1, %%mm0, %%mm4, %%mm2, %%mm1, %%mm5)
|
|
|
|
|
"movq %%mm4, (%2) \n\t"
|
|
|
|
|
"movq %%mm5, (%2, %3) \n\t"
|
|
|
|
|
"add %%"REG_a", %1 \n\t"
|
|
|
|
|
"add %%"REG_a", %2 \n\t"
|
|
|
|
|
"movq (%1, %3), %%mm1 \n\t"
|
|
|
|
|
"movq (%1, %%"REG_a"),%%mm0 \n\t"
|
|
|
|
|
PAVGBP(%%mm1, %%mm2, %%mm4, %%mm0, %%mm1, %%mm5)
|
|
|
|
|
"movq %%mm4, (%2) \n\t"
|
|
|
|
|
"movq %%mm5, (%2, %3) \n\t"
|
|
|
|
|
"add %%"REG_a", %1 \n\t"
|
|
|
|
|
"add %%"REG_a", %2 \n\t"
|
|
|
|
|
"subl $4, %0 \n\t"
|
|
|
|
|
"jnz 1b \n\t"
|
|
|
|
|
:"+g"(h), "+S"(pixels), "+D"(block)
|
|
|
|
|
:"r"((long)line_size)
|
|
|
|
|
:REG_a, "memory");
|
2002-05-29 19:16:22 +02:00
|
|
|
}
|
|
|
|
|
|
2003-02-11 17:35:48 +01:00
|
|
|
static void DEF(put, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
|
2002-05-29 19:16:22 +02:00
|
|
|
{
|
2002-05-30 17:14:56 +02:00
|
|
|
MOVQ_ZERO(mm7);
|
|
|
|
|
SET_RND(mm6); // =2 for rnd and =1 for no_rnd version
|
2002-05-29 19:16:22 +02:00
|
|
|
__asm __volatile(
|
2005-12-22 02:10:11 +01:00
|
|
|
"movq (%1), %%mm0 \n\t"
|
|
|
|
|
"movq 1(%1), %%mm4 \n\t"
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"
|
|
|
|
|
"movq %%mm4, %%mm5 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm0 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm4 \n\t"
|
|
|
|
|
"punpckhbw %%mm7, %%mm1 \n\t"
|
|
|
|
|
"punpckhbw %%mm7, %%mm5 \n\t"
|
|
|
|
|
"paddusw %%mm0, %%mm4 \n\t"
|
|
|
|
|
"paddusw %%mm1, %%mm5 \n\t"
|
|
|
|
|
"xor %%"REG_a", %%"REG_a" \n\t"
|
|
|
|
|
"add %3, %1 \n\t"
|
2006-08-12 18:37:31 +02:00
|
|
|
ASMALIGN(3)
|
2005-12-22 02:10:11 +01:00
|
|
|
"1: \n\t"
|
|
|
|
|
"movq (%1, %%"REG_a"), %%mm0 \n\t"
|
|
|
|
|
"movq 1(%1, %%"REG_a"), %%mm2 \n\t"
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"
|
|
|
|
|
"movq %%mm2, %%mm3 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm0 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm2 \n\t"
|
|
|
|
|
"punpckhbw %%mm7, %%mm1 \n\t"
|
|
|
|
|
"punpckhbw %%mm7, %%mm3 \n\t"
|
|
|
|
|
"paddusw %%mm2, %%mm0 \n\t"
|
|
|
|
|
"paddusw %%mm3, %%mm1 \n\t"
|
|
|
|
|
"paddusw %%mm6, %%mm4 \n\t"
|
|
|
|
|
"paddusw %%mm6, %%mm5 \n\t"
|
|
|
|
|
"paddusw %%mm0, %%mm4 \n\t"
|
|
|
|
|
"paddusw %%mm1, %%mm5 \n\t"
|
|
|
|
|
"psrlw $2, %%mm4 \n\t"
|
|
|
|
|
"psrlw $2, %%mm5 \n\t"
|
|
|
|
|
"packuswb %%mm5, %%mm4 \n\t"
|
|
|
|
|
"movq %%mm4, (%2, %%"REG_a") \n\t"
|
|
|
|
|
"add %3, %%"REG_a" \n\t"
|
2002-05-29 19:16:22 +02:00
|
|
|
|
2005-12-22 02:10:11 +01:00
|
|
|
"movq (%1, %%"REG_a"), %%mm2 \n\t" // 0 <-> 2 1 <-> 3
|
|
|
|
|
"movq 1(%1, %%"REG_a"), %%mm4 \n\t"
|
|
|
|
|
"movq %%mm2, %%mm3 \n\t"
|
|
|
|
|
"movq %%mm4, %%mm5 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm2 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm4 \n\t"
|
|
|
|
|
"punpckhbw %%mm7, %%mm3 \n\t"
|
|
|
|
|
"punpckhbw %%mm7, %%mm5 \n\t"
|
|
|
|
|
"paddusw %%mm2, %%mm4 \n\t"
|
|
|
|
|
"paddusw %%mm3, %%mm5 \n\t"
|
|
|
|
|
"paddusw %%mm6, %%mm0 \n\t"
|
|
|
|
|
"paddusw %%mm6, %%mm1 \n\t"
|
|
|
|
|
"paddusw %%mm4, %%mm0 \n\t"
|
|
|
|
|
"paddusw %%mm5, %%mm1 \n\t"
|
|
|
|
|
"psrlw $2, %%mm0 \n\t"
|
|
|
|
|
"psrlw $2, %%mm1 \n\t"
|
|
|
|
|
"packuswb %%mm1, %%mm0 \n\t"
|
|
|
|
|
"movq %%mm0, (%2, %%"REG_a") \n\t"
|
|
|
|
|
"add %3, %%"REG_a" \n\t"
|
2002-05-29 19:16:22 +02:00
|
|
|
|
2005-12-22 02:10:11 +01:00
|
|
|
"subl $2, %0 \n\t"
|
|
|
|
|
"jnz 1b \n\t"
|
|
|
|
|
:"+g"(h), "+S"(pixels)
|
|
|
|
|
:"D"(block), "r"((long)line_size)
|
|
|
|
|
:REG_a, "memory");
|
2002-05-29 19:16:22 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// avg_pixels
|
2005-09-06 23:25:35 +02:00
|
|
|
static void attribute_unused DEF(avg, pixels4)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
|
2004-09-07 03:48:45 +02:00
|
|
|
{
|
|
|
|
|
MOVQ_BFE(mm6);
|
|
|
|
|
JUMPALIGN();
|
|
|
|
|
do {
|
2005-12-22 02:10:11 +01:00
|
|
|
__asm __volatile(
|
|
|
|
|
"movd %0, %%mm0 \n\t"
|
|
|
|
|
"movd %1, %%mm1 \n\t"
|
|
|
|
|
PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
|
|
|
|
|
"movd %%mm2, %0 \n\t"
|
|
|
|
|
:"+m"(*block)
|
|
|
|
|
:"m"(*pixels)
|
|
|
|
|
:"memory");
|
|
|
|
|
pixels += line_size;
|
|
|
|
|
block += line_size;
|
2004-09-07 03:48:45 +02:00
|
|
|
}
|
|
|
|
|
while (--h);
|
|
|
|
|
}
|
|
|
|
|
|
2002-06-03 19:31:46 +02:00
|
|
|
// in case more speed is needed - unroling would certainly help
|
2003-02-11 17:35:48 +01:00
|
|
|
static void DEF(avg, pixels8)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
|
2002-06-03 19:31:46 +02:00
|
|
|
{
|
|
|
|
|
MOVQ_BFE(mm6);
|
|
|
|
|
JUMPALIGN();
|
|
|
|
|
do {
|
2005-12-22 02:10:11 +01:00
|
|
|
__asm __volatile(
|
|
|
|
|
"movq %0, %%mm0 \n\t"
|
|
|
|
|
"movq %1, %%mm1 \n\t"
|
|
|
|
|
PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
|
|
|
|
|
"movq %%mm2, %0 \n\t"
|
|
|
|
|
:"+m"(*block)
|
|
|
|
|
:"m"(*pixels)
|
|
|
|
|
:"memory");
|
|
|
|
|
pixels += line_size;
|
|
|
|
|
block += line_size;
|
2002-06-03 19:31:46 +02:00
|
|
|
}
|
|
|
|
|
while (--h);
|
|
|
|
|
}
|
|
|
|
|
|
2003-02-11 17:35:48 +01:00
|
|
|
static void DEF(avg, pixels16)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
|
2002-09-11 14:39:53 +02:00
|
|
|
{
|
|
|
|
|
MOVQ_BFE(mm6);
|
|
|
|
|
JUMPALIGN();
|
|
|
|
|
do {
|
2005-12-22 02:10:11 +01:00
|
|
|
__asm __volatile(
|
|
|
|
|
"movq %0, %%mm0 \n\t"
|
|
|
|
|
"movq %1, %%mm1 \n\t"
|
|
|
|
|
PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
|
|
|
|
|
"movq %%mm2, %0 \n\t"
|
|
|
|
|
"movq 8%0, %%mm0 \n\t"
|
|
|
|
|
"movq 8%1, %%mm1 \n\t"
|
|
|
|
|
PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
|
|
|
|
|
"movq %%mm2, 8%0 \n\t"
|
|
|
|
|
:"+m"(*block)
|
|
|
|
|
:"m"(*pixels)
|
|
|
|
|
:"memory");
|
|
|
|
|
pixels += line_size;
|
|
|
|
|
block += line_size;
|
2002-09-11 14:39:53 +02:00
|
|
|
}
|
|
|
|
|
while (--h);
|
|
|
|
|
}
|
|
|
|
|
|
2003-02-11 17:35:48 +01:00
|
|
|
static void DEF(avg, pixels8_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
|
2002-06-03 19:31:46 +02:00
|
|
|
{
|
|
|
|
|
MOVQ_BFE(mm6);
|
|
|
|
|
JUMPALIGN();
|
|
|
|
|
do {
|
2005-12-22 02:10:11 +01:00
|
|
|
__asm __volatile(
|
|
|
|
|
"movq %1, %%mm0 \n\t"
|
|
|
|
|
"movq 1%1, %%mm1 \n\t"
|
|
|
|
|
"movq %0, %%mm3 \n\t"
|
|
|
|
|
PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
|
|
|
|
|
PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
|
|
|
|
|
"movq %%mm0, %0 \n\t"
|
|
|
|
|
:"+m"(*block)
|
|
|
|
|
:"m"(*pixels)
|
|
|
|
|
:"memory");
|
|
|
|
|
pixels += line_size;
|
|
|
|
|
block += line_size;
|
2002-06-03 19:31:46 +02:00
|
|
|
} while (--h);
|
|
|
|
|
}
|
|
|
|
|
|
2004-05-18 19:09:46 +02:00
|
|
|
static __attribute__((unused)) void DEF(avg, pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
|
2003-01-05 16:57:10 +01:00
|
|
|
{
|
|
|
|
|
MOVQ_BFE(mm6);
|
|
|
|
|
JUMPALIGN();
|
|
|
|
|
do {
|
2005-12-22 02:10:11 +01:00
|
|
|
__asm __volatile(
|
|
|
|
|
"movq %1, %%mm0 \n\t"
|
|
|
|
|
"movq %2, %%mm1 \n\t"
|
|
|
|
|
"movq %0, %%mm3 \n\t"
|
|
|
|
|
PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
|
|
|
|
|
PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
|
|
|
|
|
"movq %%mm0, %0 \n\t"
|
|
|
|
|
:"+m"(*dst)
|
|
|
|
|
:"m"(*src1), "m"(*src2)
|
|
|
|
|
:"memory");
|
|
|
|
|
dst += dstStride;
|
2003-01-05 16:57:10 +01:00
|
|
|
src1 += src1Stride;
|
|
|
|
|
src2 += 8;
|
|
|
|
|
} while (--h);
|
|
|
|
|
}
|
|
|
|
|
|
2003-02-11 17:35:48 +01:00
|
|
|
static void DEF(avg, pixels16_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
|
2002-09-11 14:39:53 +02:00
|
|
|
{
|
|
|
|
|
MOVQ_BFE(mm6);
|
|
|
|
|
JUMPALIGN();
|
|
|
|
|
do {
|
2005-12-22 02:10:11 +01:00
|
|
|
__asm __volatile(
|
|
|
|
|
"movq %1, %%mm0 \n\t"
|
|
|
|
|
"movq 1%1, %%mm1 \n\t"
|
|
|
|
|
"movq %0, %%mm3 \n\t"
|
|
|
|
|
PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
|
|
|
|
|
PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
|
|
|
|
|
"movq %%mm0, %0 \n\t"
|
|
|
|
|
"movq 8%1, %%mm0 \n\t"
|
|
|
|
|
"movq 9%1, %%mm1 \n\t"
|
|
|
|
|
"movq 8%0, %%mm3 \n\t"
|
|
|
|
|
PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
|
|
|
|
|
PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
|
|
|
|
|
"movq %%mm0, 8%0 \n\t"
|
|
|
|
|
:"+m"(*block)
|
|
|
|
|
:"m"(*pixels)
|
|
|
|
|
:"memory");
|
|
|
|
|
pixels += line_size;
|
|
|
|
|
block += line_size;
|
2002-09-11 14:39:53 +02:00
|
|
|
} while (--h);
|
|
|
|
|
}
|
|
|
|
|
|
2004-05-18 19:09:46 +02:00
|
|
|
static __attribute__((unused)) void DEF(avg, pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
|
2003-01-05 16:57:10 +01:00
|
|
|
{
|
|
|
|
|
MOVQ_BFE(mm6);
|
|
|
|
|
JUMPALIGN();
|
|
|
|
|
do {
|
2005-12-22 02:10:11 +01:00
|
|
|
__asm __volatile(
|
|
|
|
|
"movq %1, %%mm0 \n\t"
|
|
|
|
|
"movq %2, %%mm1 \n\t"
|
|
|
|
|
"movq %0, %%mm3 \n\t"
|
|
|
|
|
PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
|
|
|
|
|
PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
|
|
|
|
|
"movq %%mm0, %0 \n\t"
|
|
|
|
|
"movq 8%1, %%mm0 \n\t"
|
|
|
|
|
"movq 8%2, %%mm1 \n\t"
|
|
|
|
|
"movq 8%0, %%mm3 \n\t"
|
|
|
|
|
PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
|
|
|
|
|
PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
|
|
|
|
|
"movq %%mm0, 8%0 \n\t"
|
|
|
|
|
:"+m"(*dst)
|
|
|
|
|
:"m"(*src1), "m"(*src2)
|
|
|
|
|
:"memory");
|
|
|
|
|
dst += dstStride;
|
2003-01-05 16:57:10 +01:00
|
|
|
src1 += src1Stride;
|
|
|
|
|
src2 += 16;
|
|
|
|
|
} while (--h);
|
|
|
|
|
}
|
|
|
|
|
|
2003-02-11 17:35:48 +01:00
|
|
|
static void DEF(avg, pixels8_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
|
2002-06-03 19:31:46 +02:00
|
|
|
{
|
|
|
|
|
MOVQ_BFE(mm6);
|
|
|
|
|
__asm __volatile(
|
2005-12-22 02:10:11 +01:00
|
|
|
"lea (%3, %3), %%"REG_a" \n\t"
|
|
|
|
|
"movq (%1), %%mm0 \n\t"
|
2006-08-12 18:37:31 +02:00
|
|
|
ASMALIGN(3)
|
2005-12-22 02:10:11 +01:00
|
|
|
"1: \n\t"
|
|
|
|
|
"movq (%1, %3), %%mm1 \n\t"
|
|
|
|
|
"movq (%1, %%"REG_a"), %%mm2 \n\t"
|
|
|
|
|
PAVGBP(%%mm1, %%mm0, %%mm4, %%mm2, %%mm1, %%mm5)
|
|
|
|
|
"movq (%2), %%mm3 \n\t"
|
|
|
|
|
PAVGB(%%mm3, %%mm4, %%mm0, %%mm6)
|
|
|
|
|
"movq (%2, %3), %%mm3 \n\t"
|
|
|
|
|
PAVGB(%%mm3, %%mm5, %%mm1, %%mm6)
|
|
|
|
|
"movq %%mm0, (%2) \n\t"
|
|
|
|
|
"movq %%mm1, (%2, %3) \n\t"
|
|
|
|
|
"add %%"REG_a", %1 \n\t"
|
|
|
|
|
"add %%"REG_a", %2 \n\t"
|
2002-06-03 19:31:46 +02:00
|
|
|
|
2005-12-22 02:10:11 +01:00
|
|
|
"movq (%1, %3), %%mm1 \n\t"
|
|
|
|
|
"movq (%1, %%"REG_a"), %%mm0 \n\t"
|
|
|
|
|
PAVGBP(%%mm1, %%mm2, %%mm4, %%mm0, %%mm1, %%mm5)
|
|
|
|
|
"movq (%2), %%mm3 \n\t"
|
|
|
|
|
PAVGB(%%mm3, %%mm4, %%mm2, %%mm6)
|
|
|
|
|
"movq (%2, %3), %%mm3 \n\t"
|
|
|
|
|
PAVGB(%%mm3, %%mm5, %%mm1, %%mm6)
|
|
|
|
|
"movq %%mm2, (%2) \n\t"
|
|
|
|
|
"movq %%mm1, (%2, %3) \n\t"
|
|
|
|
|
"add %%"REG_a", %1 \n\t"
|
|
|
|
|
"add %%"REG_a", %2 \n\t"
|
2002-06-03 19:31:46 +02:00
|
|
|
|
2005-12-22 02:10:11 +01:00
|
|
|
"subl $4, %0 \n\t"
|
|
|
|
|
"jnz 1b \n\t"
|
|
|
|
|
:"+g"(h), "+S"(pixels), "+D"(block)
|
|
|
|
|
:"r"((long)line_size)
|
|
|
|
|
:REG_a, "memory");
|
2002-06-03 19:31:46 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// this routine is 'slightly' suboptimal but mostly unused
|
2003-02-11 17:35:48 +01:00
|
|
|
static void DEF(avg, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
|
2002-06-03 19:31:46 +02:00
|
|
|
{
|
|
|
|
|
MOVQ_ZERO(mm7);
|
|
|
|
|
SET_RND(mm6); // =2 for rnd and =1 for no_rnd version
|
|
|
|
|
__asm __volatile(
|
2005-12-22 02:10:11 +01:00
|
|
|
"movq (%1), %%mm0 \n\t"
|
|
|
|
|
"movq 1(%1), %%mm4 \n\t"
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"
|
|
|
|
|
"movq %%mm4, %%mm5 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm0 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm4 \n\t"
|
|
|
|
|
"punpckhbw %%mm7, %%mm1 \n\t"
|
|
|
|
|
"punpckhbw %%mm7, %%mm5 \n\t"
|
|
|
|
|
"paddusw %%mm0, %%mm4 \n\t"
|
|
|
|
|
"paddusw %%mm1, %%mm5 \n\t"
|
|
|
|
|
"xor %%"REG_a", %%"REG_a" \n\t"
|
|
|
|
|
"add %3, %1 \n\t"
|
2006-08-12 18:37:31 +02:00
|
|
|
ASMALIGN(3)
|
2005-12-22 02:10:11 +01:00
|
|
|
"1: \n\t"
|
|
|
|
|
"movq (%1, %%"REG_a"), %%mm0 \n\t"
|
|
|
|
|
"movq 1(%1, %%"REG_a"), %%mm2 \n\t"
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"
|
|
|
|
|
"movq %%mm2, %%mm3 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm0 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm2 \n\t"
|
|
|
|
|
"punpckhbw %%mm7, %%mm1 \n\t"
|
|
|
|
|
"punpckhbw %%mm7, %%mm3 \n\t"
|
|
|
|
|
"paddusw %%mm2, %%mm0 \n\t"
|
|
|
|
|
"paddusw %%mm3, %%mm1 \n\t"
|
|
|
|
|
"paddusw %%mm6, %%mm4 \n\t"
|
|
|
|
|
"paddusw %%mm6, %%mm5 \n\t"
|
|
|
|
|
"paddusw %%mm0, %%mm4 \n\t"
|
|
|
|
|
"paddusw %%mm1, %%mm5 \n\t"
|
|
|
|
|
"psrlw $2, %%mm4 \n\t"
|
|
|
|
|
"psrlw $2, %%mm5 \n\t"
|
|
|
|
|
"movq (%2, %%"REG_a"), %%mm3 \n\t"
|
|
|
|
|
"packuswb %%mm5, %%mm4 \n\t"
|
|
|
|
|
"pcmpeqd %%mm2, %%mm2 \n\t"
|
|
|
|
|
"paddb %%mm2, %%mm2 \n\t"
|
|
|
|
|
PAVGB(%%mm3, %%mm4, %%mm5, %%mm2)
|
|
|
|
|
"movq %%mm5, (%2, %%"REG_a") \n\t"
|
|
|
|
|
"add %3, %%"REG_a" \n\t"
|
2002-06-03 19:31:46 +02:00
|
|
|
|
2005-12-22 02:10:11 +01:00
|
|
|
"movq (%1, %%"REG_a"), %%mm2 \n\t" // 0 <-> 2 1 <-> 3
|
|
|
|
|
"movq 1(%1, %%"REG_a"), %%mm4 \n\t"
|
|
|
|
|
"movq %%mm2, %%mm3 \n\t"
|
|
|
|
|
"movq %%mm4, %%mm5 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm2 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm4 \n\t"
|
|
|
|
|
"punpckhbw %%mm7, %%mm3 \n\t"
|
|
|
|
|
"punpckhbw %%mm7, %%mm5 \n\t"
|
|
|
|
|
"paddusw %%mm2, %%mm4 \n\t"
|
|
|
|
|
"paddusw %%mm3, %%mm5 \n\t"
|
|
|
|
|
"paddusw %%mm6, %%mm0 \n\t"
|
|
|
|
|
"paddusw %%mm6, %%mm1 \n\t"
|
|
|
|
|
"paddusw %%mm4, %%mm0 \n\t"
|
|
|
|
|
"paddusw %%mm5, %%mm1 \n\t"
|
|
|
|
|
"psrlw $2, %%mm0 \n\t"
|
|
|
|
|
"psrlw $2, %%mm1 \n\t"
|
|
|
|
|
"movq (%2, %%"REG_a"), %%mm3 \n\t"
|
|
|
|
|
"packuswb %%mm1, %%mm0 \n\t"
|
|
|
|
|
"pcmpeqd %%mm2, %%mm2 \n\t"
|
|
|
|
|
"paddb %%mm2, %%mm2 \n\t"
|
|
|
|
|
PAVGB(%%mm3, %%mm0, %%mm1, %%mm2)
|
|
|
|
|
"movq %%mm1, (%2, %%"REG_a") \n\t"
|
|
|
|
|
"add %3, %%"REG_a" \n\t"
|
2002-06-03 19:31:46 +02:00
|
|
|
|
2005-12-22 02:10:11 +01:00
|
|
|
"subl $2, %0 \n\t"
|
|
|
|
|
"jnz 1b \n\t"
|
|
|
|
|
:"+g"(h), "+S"(pixels)
|
|
|
|
|
:"D"(block), "r"((long)line_size)
|
|
|
|
|
:REG_a, "memory");
|
2002-06-03 19:31:46 +02:00
|
|
|
}
|
2002-09-11 14:39:53 +02:00
|
|
|
|
|
|
|
|
//FIXME optimize
|
2003-02-11 17:35:48 +01:00
|
|
|
static void DEF(put, pixels16_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
|
2002-09-11 14:39:53 +02:00
|
|
|
DEF(put, pixels8_y2)(block , pixels , line_size, h);
|
|
|
|
|
DEF(put, pixels8_y2)(block+8, pixels+8, line_size, h);
|
|
|
|
|
}
|
|
|
|
|
|
2003-02-11 17:35:48 +01:00
|
|
|
static void DEF(put, pixels16_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
|
2002-09-11 14:39:53 +02:00
|
|
|
DEF(put, pixels8_xy2)(block , pixels , line_size, h);
|
|
|
|
|
DEF(put, pixels8_xy2)(block+8, pixels+8, line_size, h);
|
|
|
|
|
}
|
|
|
|
|
|
2003-02-11 17:35:48 +01:00
|
|
|
static void DEF(avg, pixels16_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
|
2002-09-11 14:39:53 +02:00
|
|
|
DEF(avg, pixels8_y2)(block , pixels , line_size, h);
|
|
|
|
|
DEF(avg, pixels8_y2)(block+8, pixels+8, line_size, h);
|
|
|
|
|
}
|
|
|
|
|
|
2003-02-11 17:35:48 +01:00
|
|
|
static void DEF(avg, pixels16_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
|
2002-09-11 14:39:53 +02:00
|
|
|
DEF(avg, pixels8_xy2)(block , pixels , line_size, h);
|
|
|
|
|
DEF(avg, pixels8_xy2)(block+8, pixels+8, line_size, h);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
