Chiebot-Mirror
/
FFmpeg
mirror of https://github.com/FFmpeg/FFmpeg.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

dsputil: x86: Convert mpeg4 qpel and dsputil avg to yasm

Browse Source 
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
pull/9/head
Daniel Kang 12 years ago committed by Luca Barbato
parent f90ff772e7
commit 71155d7b41
6 changed files with 1380 additions and 1312 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 2
      libavcodec/x86/Makefile
  2. 789
      libavcodec/x86/dsputil_avg_template.c
  3. 874
      libavcodec/x86/dsputil_mmx.c
  4. 465
      libavcodec/x86/hpeldsp.asm
  5. 558
      libavcodec/x86/mpeg4qpel.asm
  6. 4
      libavcodec/x86/vc1dsp_mmx.c

2
libavcodec/x86/Makefile
Unescape View File

@ -71,3 +71,5 @@ YASM-OBJS-$(CONFIG_VP8_DECODER) += x86/vp8dsp.o
YASM-OBJS += x86/dsputil.o \
x86/deinterlace.o \
x86/fmtconvert.o \
x86/hpeldsp.o \
x86/mpeg4qpel.o \

789
libavcodec/x86/dsputil_avg_template.c
Unescape View File

@ -24,781 +24,54 @@
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/* XXX: we use explicit registers to avoid a gcc 2.95.2 register asm
clobber bug - now it will work with 2.95.2 and also with -fPIC
*/
static void DEF(put_pixels8_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
__asm__ volatile(
"lea (%3, %3), %%"REG_a" \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"movq (%1, %3), %%mm1 \n\t"
PAVGB" 1(%1), %%mm0 \n\t"
PAVGB" 1(%1, %3), %%mm1 \n\t"
"movq %%mm0, (%2) \n\t"
"movq %%mm1, (%2, %3) \n\t"
"add %%"REG_a", %1 \n\t"
"add %%"REG_a", %2 \n\t"
"movq (%1), %%mm0 \n\t"
"movq (%1, %3), %%mm1 \n\t"
PAVGB" 1(%1), %%mm0 \n\t"
PAVGB" 1(%1, %3), %%mm1 \n\t"
"add %%"REG_a", %1 \n\t"
"movq %%mm0, (%2) \n\t"
"movq %%mm1, (%2, %3) \n\t"
"add %%"REG_a", %2 \n\t"
"subl $4, %0 \n\t"
"jnz 1b \n\t"
:"+g"(h), "+S"(pixels), "+D"(block)
:"r" ((x86_reg)line_size)
:"%"REG_a, "memory");
}
#ifndef SKIP_FOR_3DNOW
static void DEF(put_pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
{
__asm__ volatile(
"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" %%mm1, %%mm0 \n\t"
"movq %%mm0, (%3) \n\t"
"add %5, %3 \n\t"
"decl %0 \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"add %4, %1 \n\t"
"movq (%1), %%mm1 \n\t"
"add %4, %1 \n\t"
PAVGB" (%2), %%mm0 \n\t"
PAVGB" 8(%2), %%mm1 \n\t"
"movq %%mm0, (%3) \n\t"
"add %5, %3 \n\t"
"movq %%mm1, (%3) \n\t"
"add %5, %3 \n\t"
"movq (%1), %%mm0 \n\t"
"add %4, %1 \n\t"
"movq (%1), %%mm1 \n\t"
"add %4, %1 \n\t"
PAVGB" 16(%2), %%mm0 \n\t"
PAVGB" 24(%2), %%mm1 \n\t"
"movq %%mm0, (%3) \n\t"
"add %5, %3 \n\t"
"movq %%mm1, (%3) \n\t"
"add %5, %3 \n\t"
"add $32, %2 \n\t"
"subl $4, %0 \n\t"
"jnz 1b \n\t"
#if !HAVE_EBX_AVAILABLE //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cannot be used
:"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
#else
:"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
#endif
:"S"((x86_reg)src1Stride), "D"((x86_reg)dstStride)
:"memory");
//the following should be used, though better not with gcc ...
/* :"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
:"r"(src1Stride), "r"(dstStride)
:"memory");*/
}
static void DEF(put_no_rnd_pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
{
__asm__ volatile(
"pcmpeqb %%mm6, %%mm6 \n\t"
"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"
"pxor %%mm6, %%mm0 \n\t"
"pxor %%mm6, %%mm1 \n\t"
PAVGB" %%mm1, %%mm0 \n\t"
"pxor %%mm6, %%mm0 \n\t"
"movq %%mm0, (%3) \n\t"
"add %5, %3 \n\t"
"decl %0 \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"add %4, %1 \n\t"
"movq (%1), %%mm1 \n\t"
"add %4, %1 \n\t"
"movq (%2), %%mm2 \n\t"
"movq 8(%2), %%mm3 \n\t"
"pxor %%mm6, %%mm0 \n\t"
"pxor %%mm6, %%mm1 \n\t"
"pxor %%mm6, %%mm2 \n\t"
"pxor %%mm6, %%mm3 \n\t"
PAVGB" %%mm2, %%mm0 \n\t"
PAVGB" %%mm3, %%mm1 \n\t"
"pxor %%mm6, %%mm0 \n\t"
"pxor %%mm6, %%mm1 \n\t"
"movq %%mm0, (%3) \n\t"
"add %5, %3 \n\t"
"movq %%mm1, (%3) \n\t"
"add %5, %3 \n\t"
"movq (%1), %%mm0 \n\t"
"add %4, %1 \n\t"
"movq (%1), %%mm1 \n\t"
"add %4, %1 \n\t"
"movq 16(%2), %%mm2 \n\t"
"movq 24(%2), %%mm3 \n\t"
"pxor %%mm6, %%mm0 \n\t"
"pxor %%mm6, %%mm1 \n\t"
"pxor %%mm6, %%mm2 \n\t"
"pxor %%mm6, %%mm3 \n\t"
PAVGB" %%mm2, %%mm0 \n\t"
PAVGB" %%mm3, %%mm1 \n\t"
"pxor %%mm6, %%mm0 \n\t"
"pxor %%mm6, %%mm1 \n\t"
"movq %%mm0, (%3) \n\t"
"add %5, %3 \n\t"
"movq %%mm1, (%3) \n\t"
"add %5, %3 \n\t"
"add $32, %2 \n\t"
"subl $4, %0 \n\t"
"jnz 1b \n\t"
#if !HAVE_EBX_AVAILABLE //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cannot be used
:"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
#else
:"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
#endif
:"S"((x86_reg)src1Stride), "D"((x86_reg)dstStride)
:"memory");
//the following should be used, though better not with gcc ...
/* :"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
:"r"(src1Stride), "r"(dstStride)
:"memory");*/
}
static void DEF(avg_pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
{
__asm__ volatile(
"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" %%mm1, %%mm0 \n\t"
PAVGB" (%3), %%mm0 \n\t"
"movq %%mm0, (%3) \n\t"
"add %5, %3 \n\t"
"decl %0 \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"add %4, %1 \n\t"
"movq (%1), %%mm1 \n\t"
"add %4, %1 \n\t"
PAVGB" (%2), %%mm0 \n\t"
PAVGB" 8(%2), %%mm1 \n\t"
PAVGB" (%3), %%mm0 \n\t"
"movq %%mm0, (%3) \n\t"
"add %5, %3 \n\t"
PAVGB" (%3), %%mm1 \n\t"
"movq %%mm1, (%3) \n\t"
"add %5, %3 \n\t"
"movq (%1), %%mm0 \n\t"
"add %4, %1 \n\t"
"movq (%1), %%mm1 \n\t"
"add %4, %1 \n\t"
PAVGB" 16(%2), %%mm0 \n\t"
PAVGB" 24(%2), %%mm1 \n\t"
PAVGB" (%3), %%mm0 \n\t"
"movq %%mm0, (%3) \n\t"
"add %5, %3 \n\t"
PAVGB" (%3), %%mm1 \n\t"
"movq %%mm1, (%3) \n\t"
"add %5, %3 \n\t"
"add $32, %2 \n\t"
"subl $4, %0 \n\t"
"jnz 1b \n\t"
#if !HAVE_EBX_AVAILABLE //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cannot be used
:"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
#else
:"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
#endif
:"S"((x86_reg)src1Stride), "D"((x86_reg)dstStride)
:"memory");
//the following should be used, though better not with gcc ...
/* :"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
:"r"(src1Stride), "r"(dstStride)
:"memory");*/
}
#endif /* SKIP_FOR_3DNOW */
static void DEF(put_pixels16_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
__asm__ volatile(
"lea (%3, %3), %%"REG_a" \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"movq (%1, %3), %%mm1 \n\t"
"movq 8(%1), %%mm2 \n\t"
"movq 8(%1, %3), %%mm3 \n\t"
PAVGB" 1(%1), %%mm0 \n\t"
PAVGB" 1(%1, %3), %%mm1 \n\t"
PAVGB" 9(%1), %%mm2 \n\t"
PAVGB" 9(%1, %3), %%mm3 \n\t"
"movq %%mm0, (%2) \n\t"
"movq %%mm1, (%2, %3) \n\t"
"movq %%mm2, 8(%2) \n\t"
"movq %%mm3, 8(%2, %3) \n\t"
"add %%"REG_a", %1 \n\t"
"add %%"REG_a", %2 \n\t"
"movq (%1), %%mm0 \n\t"
"movq (%1, %3), %%mm1 \n\t"
"movq 8(%1), %%mm2 \n\t"
"movq 8(%1, %3), %%mm3 \n\t"
PAVGB" 1(%1), %%mm0 \n\t"
PAVGB" 1(%1, %3), %%mm1 \n\t"
PAVGB" 9(%1), %%mm2 \n\t"
PAVGB" 9(%1, %3), %%mm3 \n\t"
"add %%"REG_a", %1 \n\t"
"movq %%mm0, (%2) \n\t"
"movq %%mm1, (%2, %3) \n\t"
"movq %%mm2, 8(%2) \n\t"
"movq %%mm3, 8(%2, %3) \n\t"
"add %%"REG_a", %2 \n\t"
"subl $4, %0 \n\t"
"jnz 1b \n\t"
:"+g"(h), "+S"(pixels), "+D"(block)
:"r" ((x86_reg)line_size)
:"%"REG_a, "memory");
}
#ifndef SKIP_FOR_3DNOW
static void DEF(put_pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
{
__asm__ volatile(
"testl $1, %0 \n\t"
" jz 1f \n\t"
"movq (%1), %%mm0 \n\t"
"movq 8(%1), %%mm1 \n\t"
PAVGB" (%2), %%mm0 \n\t"
PAVGB" 8(%2), %%mm1 \n\t"
"add %4, %1 \n\t"
"add $16, %2 \n\t"
"movq %%mm0, (%3) \n\t"
"movq %%mm1, 8(%3) \n\t"
"add %5, %3 \n\t"
"decl %0 \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"movq 8(%1), %%mm1 \n\t"
"add %4, %1 \n\t"
PAVGB" (%2), %%mm0 \n\t"
PAVGB" 8(%2), %%mm1 \n\t"
"movq %%mm0, (%3) \n\t"
"movq %%mm1, 8(%3) \n\t"
"add %5, %3 \n\t"
"movq (%1), %%mm0 \n\t"
"movq 8(%1), %%mm1 \n\t"
"add %4, %1 \n\t"
PAVGB" 16(%2), %%mm0 \n\t"
PAVGB" 24(%2), %%mm1 \n\t"
"movq %%mm0, (%3) \n\t"
"movq %%mm1, 8(%3) \n\t"
"add %5, %3 \n\t"
"add $32, %2 \n\t"
"subl $2, %0 \n\t"
"jnz 1b \n\t"
#if !HAVE_EBX_AVAILABLE //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cannot be used
:"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
#else
:"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
#endif
:"S"((x86_reg)src1Stride), "D"((x86_reg)dstStride)
:"memory");
//the following should be used, though better not with gcc ...
/* :"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
:"r"(src1Stride), "r"(dstStride)
:"memory");*/
}
static void DEF(avg_pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
{
__asm__ volatile(
"testl $1, %0 \n\t"
" jz 1f \n\t"
"movq (%1), %%mm0 \n\t"
"movq 8(%1), %%mm1 \n\t"
PAVGB" (%2), %%mm0 \n\t"
PAVGB" 8(%2), %%mm1 \n\t"
"add %4, %1 \n\t"
"add $16, %2 \n\t"
PAVGB" (%3), %%mm0 \n\t"
PAVGB" 8(%3), %%mm1 \n\t"
"movq %%mm0, (%3) \n\t"
"movq %%mm1, 8(%3) \n\t"
"add %5, %3 \n\t"
"decl %0 \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"movq 8(%1), %%mm1 \n\t"
"add %4, %1 \n\t"
PAVGB" (%2), %%mm0 \n\t"
PAVGB" 8(%2), %%mm1 \n\t"
PAVGB" (%3), %%mm0 \n\t"
PAVGB" 8(%3), %%mm1 \n\t"
"movq %%mm0, (%3) \n\t"
"movq %%mm1, 8(%3) \n\t"
"add %5, %3 \n\t"
"movq (%1), %%mm0 \n\t"
"movq 8(%1), %%mm1 \n\t"
"add %4, %1 \n\t"
PAVGB" 16(%2), %%mm0 \n\t"
PAVGB" 24(%2), %%mm1 \n\t"
PAVGB" (%3), %%mm0 \n\t"
PAVGB" 8(%3), %%mm1 \n\t"
"movq %%mm0, (%3) \n\t"
"movq %%mm1, 8(%3) \n\t"
"add %5, %3 \n\t"
"add $32, %2 \n\t"
"subl $2, %0 \n\t"
"jnz 1b \n\t"
#if !HAVE_EBX_AVAILABLE //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cannot be used
:"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
#else
:"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
#endif
:"S"((x86_reg)src1Stride), "D"((x86_reg)dstStride)
:"memory");
//the following should be used, though better not with gcc ...
/* :"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
:"r"(src1Stride), "r"(dstStride)
:"memory");*/
}
static void DEF(put_no_rnd_pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
{
__asm__ volatile(
"pcmpeqb %%mm6, %%mm6 \n\t"
"testl $1, %0 \n\t"
" jz 1f \n\t"
"movq (%1), %%mm0 \n\t"
"movq 8(%1), %%mm1 \n\t"
"movq (%2), %%mm2 \n\t"
"movq 8(%2), %%mm3 \n\t"
"pxor %%mm6, %%mm0 \n\t"
"pxor %%mm6, %%mm1 \n\t"
"pxor %%mm6, %%mm2 \n\t"
"pxor %%mm6, %%mm3 \n\t"
PAVGB" %%mm2, %%mm0 \n\t"
PAVGB" %%mm3, %%mm1 \n\t"
"pxor %%mm6, %%mm0 \n\t"
"pxor %%mm6, %%mm1 \n\t"
"add %4, %1 \n\t"
"add $16, %2 \n\t"
"movq %%mm0, (%3) \n\t"
"movq %%mm1, 8(%3) \n\t"
"add %5, %3 \n\t"
"decl %0 \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"movq 8(%1), %%mm1 \n\t"
"add %4, %1 \n\t"
"movq (%2), %%mm2 \n\t"
"movq 8(%2), %%mm3 \n\t"
"pxor %%mm6, %%mm0 \n\t"
"pxor %%mm6, %%mm1 \n\t"
"pxor %%mm6, %%mm2 \n\t"
"pxor %%mm6, %%mm3 \n\t"
PAVGB" %%mm2, %%mm0 \n\t"
PAVGB" %%mm3, %%mm1 \n\t"
"pxor %%mm6, %%mm0 \n\t"
"pxor %%mm6, %%mm1 \n\t"
"movq %%mm0, (%3) \n\t"
"movq %%mm1, 8(%3) \n\t"
"add %5, %3 \n\t"
"movq (%1), %%mm0 \n\t"
"movq 8(%1), %%mm1 \n\t"
"add %4, %1 \n\t"
"movq 16(%2), %%mm2 \n\t"
"movq 24(%2), %%mm3 \n\t"
"pxor %%mm6, %%mm0 \n\t"
"pxor %%mm6, %%mm1 \n\t"
"pxor %%mm6, %%mm2 \n\t"
"pxor %%mm6, %%mm3 \n\t"
PAVGB" %%mm2, %%mm0 \n\t"
PAVGB" %%mm3, %%mm1 \n\t"
"pxor %%mm6, %%mm0 \n\t"
"pxor %%mm6, %%mm1 \n\t"
"movq %%mm0, (%3) \n\t"
"movq %%mm1, 8(%3) \n\t"
"add %5, %3 \n\t"
"add $32, %2 \n\t"
"subl $2, %0 \n\t"
"jnz 1b \n\t"
#if !HAVE_EBX_AVAILABLE //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cannot be used
:"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
#else
:"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
#endif
:"S"((x86_reg)src1Stride), "D"((x86_reg)dstStride)
:"memory");
//the following should be used, though better not with gcc ...
/* :"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
:"r"(src1Stride), "r"(dstStride)
:"memory");*/
}
#endif /* SKIP_FOR_3DNOW */
/* GL: this function does incorrect rounding if overflow */
static void DEF(put_no_rnd_pixels8_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
MOVQ_BONE(mm6);
__asm__ volatile(
"lea (%3, %3), %%"REG_a" \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"movq (%1, %3), %%mm2 \n\t"
"movq 1(%1), %%mm1 \n\t"
"movq 1(%1, %3), %%mm3 \n\t"
"add %%"REG_a", %1 \n\t"
"psubusb %%mm6, %%mm0 \n\t"
"psubusb %%mm6, %%mm2 \n\t"
PAVGB" %%mm1, %%mm0 \n\t"
PAVGB" %%mm3, %%mm2 \n\t"
"movq %%mm0, (%2) \n\t"
"movq %%mm2, (%2, %3) \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"
"add %%"REG_a", %2 \n\t"
"add %%"REG_a", %1 \n\t"
"psubusb %%mm6, %%mm0 \n\t"
"psubusb %%mm6, %%mm2 \n\t"
PAVGB" %%mm1, %%mm0 \n\t"
PAVGB" %%mm3, %%mm2 \n\t"
"movq %%mm0, (%2) \n\t"
"movq %%mm2, (%2, %3) \n\t"
"add %%"REG_a", %2 \n\t"
"subl $4, %0 \n\t"
"jnz 1b \n\t"
:"+g"(h), "+S"(pixels), "+D"(block)
:"r" ((x86_reg)line_size)
:"%"REG_a, "memory");
}
static void DEF(put_no_rnd_pixels8_x2_exact)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
__asm__ volatile (
"pcmpeqb %%mm6, %%mm6 \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"movq (%1, %3), %%mm2 \n\t"
"movq 1(%1), %%mm1 \n\t"
"movq 1(%1, %3), %%mm3 \n\t"
"pxor %%mm6, %%mm0 \n\t"
"pxor %%mm6, %%mm2 \n\t"
"pxor %%mm6, %%mm1 \n\t"
"pxor %%mm6, %%mm3 \n\t"
PAVGB" %%mm1, %%mm0 \n\t"
PAVGB" %%mm3, %%mm2 \n\t"
"pxor %%mm6, %%mm0 \n\t"
"pxor %%mm6, %%mm2 \n\t"
"movq %%mm0, (%2) \n\t"
"movq %%mm2, (%2, %3) \n\t"
"movq (%1, %3,2), %%mm0 \n\t"
"movq 1(%1, %3,2), %%mm1 \n\t"
"movq (%1, %4), %%mm2 \n\t"
"movq 1(%1, %4), %%mm3 \n\t"
"pxor %%mm6, %%mm0 \n\t"
"pxor %%mm6, %%mm1 \n\t"
"pxor %%mm6, %%mm2 \n\t"
"pxor %%mm6, %%mm3 \n\t"
PAVGB" %%mm1, %%mm0 \n\t"
PAVGB" %%mm3, %%mm2 \n\t"
"pxor %%mm6, %%mm0 \n\t"
"pxor %%mm6, %%mm2 \n\t"
"movq %%mm0, (%2, %3,2) \n\t"
"movq %%mm2, (%2, %4) \n\t"
"lea (%1, %3,4), %1 \n\t"
"lea (%2, %3,4), %2 \n\t"
"subl $4, %0 \n\t"
"jg 1b \n\t"
: "+g"(h), "+r"(pixels), "+r"(block)
: "r" ((x86_reg)line_size), "r"((x86_reg)3*line_size)
: "memory"
);
}
static void DEF(put_pixels8_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
//FIXME the following could be optimized too ...
static void DEF(ff_put_no_rnd_pixels16_x2)(uint8_t *block,
const uint8_t *pixels,
int line_size, int h)
{
__asm__ volatile(
"lea (%3, %3), %%"REG_a" \n\t"
"movq (%1), %%mm0 \n\t"
"sub %3, %2 \n\t"
"1: \n\t"
"movq (%1, %3), %%mm1 \n\t"
"movq (%1, %%"REG_a"), %%mm2 \n\t"
"add %%"REG_a", %1 \n\t"
PAVGB" %%mm1, %%mm0 \n\t"
PAVGB" %%mm2, %%mm1 \n\t"
"movq %%mm0, (%2, %3) \n\t"
"movq %%mm1, (%2, %%"REG_a") \n\t"
"movq (%1, %3), %%mm1 \n\t"
"movq (%1, %%"REG_a"), %%mm0 \n\t"
"add %%"REG_a", %2 \n\t"
"add %%"REG_a", %1 \n\t"
PAVGB" %%mm1, %%mm2 \n\t"
PAVGB" %%mm0, %%mm1 \n\t"
"movq %%mm2, (%2, %3) \n\t"
"movq %%mm1, (%2, %%"REG_a") \n\t"
"add %%"REG_a", %2 \n\t"
"subl $4, %0 \n\t"
"jnz 1b \n\t"
:"+g"(h), "+S"(pixels), "+D" (block)
:"r" ((x86_reg)line_size)
:"%"REG_a, "memory");
DEF(ff_put_no_rnd_pixels8_x2)(block, pixels, line_size, h);
DEF(ff_put_no_rnd_pixels8_x2)(block + 8, pixels + 8, line_size, h);
}
/* GL: this function does incorrect rounding if overflow */
static void DEF(put_no_rnd_pixels8_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
static void DEF(ff_put_pixels16_y2)(uint8_t *block, const uint8_t *pixels,
int line_size, int h)
{
MOVQ_BONE(mm6);
__asm__ volatile(
"lea (%3, %3), %%"REG_a" \n\t"
"movq (%1), %%mm0 \n\t"
"sub %3, %2 \n\t"
"1: \n\t"
"movq (%1, %3), %%mm1 \n\t"
"movq (%1, %%"REG_a"), %%mm2 \n\t"
"add %%"REG_a", %1 \n\t"
"psubusb %%mm6, %%mm1 \n\t"
PAVGB" %%mm1, %%mm0 \n\t"
PAVGB" %%mm2, %%mm1 \n\t"
"movq %%mm0, (%2, %3) \n\t"
"movq %%mm1, (%2, %%"REG_a") \n\t"
"movq (%1, %3), %%mm1 \n\t"
"movq (%1, %%"REG_a"), %%mm0 \n\t"
"add %%"REG_a", %2 \n\t"
"add %%"REG_a", %1 \n\t"
"psubusb %%mm6, %%mm1 \n\t"
PAVGB" %%mm1, %%mm2 \n\t"
PAVGB" %%mm0, %%mm1 \n\t"
"movq %%mm2, (%2, %3) \n\t"
"movq %%mm1, (%2, %%"REG_a") \n\t"
"add %%"REG_a", %2 \n\t"
"subl $4, %0 \n\t"
"jnz 1b \n\t"
:"+g"(h), "+S"(pixels), "+D" (block)
:"r" ((x86_reg)line_size)
:"%"REG_a, "memory");
DEF(ff_put_pixels8_y2)(block, pixels, line_size, h);
DEF(ff_put_pixels8_y2)(block + 8, pixels + 8, line_size, h);
}
static void DEF(put_no_rnd_pixels8_y2_exact)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
static void DEF(ff_put_no_rnd_pixels16_y2)(uint8_t *block,
const uint8_t *pixels,
int line_size, int h)
{
__asm__ volatile (
"movq (%1), %%mm0 \n\t"
"pcmpeqb %%mm6, %%mm6 \n\t"
"add %3, %1 \n\t"
"pxor %%mm6, %%mm0 \n\t"
"1: \n\t"
"movq (%1), %%mm1 \n\t"
"movq (%1, %3), %%mm2 \n\t"
"pxor %%mm6, %%mm1 \n\t"
"pxor %%mm6, %%mm2 \n\t"
PAVGB" %%mm1, %%mm0 \n\t"
PAVGB" %%mm2, %%mm1 \n\t"
"pxor %%mm6, %%mm0 \n\t"
"pxor %%mm6, %%mm1 \n\t"
"movq %%mm0, (%2) \n\t"
"movq %%mm1, (%2, %3) \n\t"
"movq (%1, %3,2), %%mm1 \n\t"
"movq (%1, %4), %%mm0 \n\t"
"pxor %%mm6, %%mm1 \n\t"
"pxor %%mm6, %%mm0 \n\t"
PAVGB" %%mm1, %%mm2 \n\t"
PAVGB" %%mm0, %%mm1 \n\t"
"pxor %%mm6, %%mm2 \n\t"
"pxor %%mm6, %%mm1 \n\t"
"movq %%mm2, (%2, %3,2) \n\t"
"movq %%mm1, (%2, %4) \n\t"
"lea (%1, %3,4), %1 \n\t"
"lea (%2, %3,4), %2 \n\t"
"subl $4, %0 \n\t"
"jg 1b \n\t"
:"+g"(h), "+r"(pixels), "+r" (block)
:"r" ((x86_reg)line_size), "r"((x86_reg)3*line_size)
:"memory"
);
DEF(ff_put_no_rnd_pixels8_y2)(block, pixels, line_size, h);
DEF(ff_put_no_rnd_pixels8_y2)(block + 8, pixels + 8, line_size, h);
}
static void DEF(avg_pixels8)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
static void DEF(ff_avg_pixels16)(uint8_t *block, const uint8_t *pixels,
int line_size, int h)
{
__asm__ volatile(
"lea (%3, %3), %%"REG_a" \n\t"
"1: \n\t"
"movq (%2), %%mm0 \n\t"
"movq (%2, %3), %%mm1 \n\t"
PAVGB" (%1), %%mm0 \n\t"
PAVGB" (%1, %3), %%mm1 \n\t"
"movq %%mm0, (%2) \n\t"
"movq %%mm1, (%2, %3) \n\t"
"add %%"REG_a", %1 \n\t"
"add %%"REG_a", %2 \n\t"
"movq (%2), %%mm0 \n\t"
"movq (%2, %3), %%mm1 \n\t"
PAVGB" (%1), %%mm0 \n\t"
PAVGB" (%1, %3), %%mm1 \n\t"
"add %%"REG_a", %1 \n\t"
"movq %%mm0, (%2) \n\t"
"movq %%mm1, (%2, %3) \n\t"
"add %%"REG_a", %2 \n\t"
"subl $4, %0 \n\t"
"jnz 1b \n\t"
:"+g"(h), "+S"(pixels), "+D"(block)
:"r" ((x86_reg)line_size)
:"%"REG_a, "memory");
DEF(ff_avg_pixels8)(block, pixels, line_size, h);
DEF(ff_avg_pixels8)(block + 8, pixels + 8, line_size, h);
}
static void DEF(avg_pixels8_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
static void DEF(ff_avg_pixels16_x2)(uint8_t *block, const uint8_t *pixels,
int line_size, int h)
{
__asm__ volatile(
"lea (%3, %3), %%"REG_a" \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"movq (%1, %3), %%mm2 \n\t"
PAVGB" 1(%1), %%mm0 \n\t"
PAVGB" 1(%1, %3), %%mm2 \n\t"
PAVGB" (%2), %%mm0 \n\t"
PAVGB" (%2, %3), %%mm2 \n\t"
"add %%"REG_a", %1 \n\t"
"movq %%mm0, (%2) \n\t"
"movq %%mm2, (%2, %3) \n\t"
"movq (%1), %%mm0 \n\t"
"movq (%1, %3), %%mm2 \n\t"
PAVGB" 1(%1), %%mm0 \n\t"
PAVGB" 1(%1, %3), %%mm2 \n\t"
"add %%"REG_a", %2 \n\t"
"add %%"REG_a", %1 \n\t"
PAVGB" (%2), %%mm0 \n\t"
PAVGB" (%2, %3), %%mm2 \n\t"
"movq %%mm0, (%2) \n\t"
"movq %%mm2, (%2, %3) \n\t"
"add %%"REG_a", %2 \n\t"
"subl $4, %0 \n\t"
"jnz 1b \n\t"
:"+g"(h), "+S"(pixels), "+D"(block)
:"r" ((x86_reg)line_size)
:"%"REG_a, "memory");
DEF(ff_avg_pixels8_x2)(block, pixels, line_size, h);
DEF(ff_avg_pixels8_x2)(block + 8, pixels + 8, line_size, h);
}
static void DEF(avg_pixels8_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
static void DEF(ff_avg_pixels16_y2)(uint8_t *block, const uint8_t *pixels,
int line_size, int h)
{
__asm__ volatile(
"lea (%3, %3), %%"REG_a" \n\t"
"movq (%1), %%mm0 \n\t"
"sub %3, %2 \n\t"
"1: \n\t"
"movq (%1, %3), %%mm1 \n\t"
"movq (%1, %%"REG_a"), %%mm2 \n\t"
"add %%"REG_a", %1 \n\t"
PAVGB" %%mm1, %%mm0 \n\t"
PAVGB" %%mm2, %%mm1 \n\t"
"movq (%2, %3), %%mm3 \n\t"
"movq (%2, %%"REG_a"), %%mm4 \n\t"
PAVGB" %%mm3, %%mm0 \n\t"
PAVGB" %%mm4, %%mm1 \n\t"
"movq %%mm0, (%2, %3) \n\t"
"movq %%mm1, (%2, %%"REG_a") \n\t"
"movq (%1, %3), %%mm1 \n\t"
"movq (%1, %%"REG_a"), %%mm0 \n\t"
PAVGB" %%mm1, %%mm2 \n\t"
PAVGB" %%mm0, %%mm1 \n\t"
"add %%"REG_a", %2 \n\t"
"add %%"REG_a", %1 \n\t"
"movq (%2, %3), %%mm3 \n\t"
"movq (%2, %%"REG_a"), %%mm4 \n\t"
PAVGB" %%mm3, %%mm2 \n\t"
PAVGB" %%mm4, %%mm1 \n\t"
"movq %%mm2, (%2, %3) \n\t"
"movq %%mm1, (%2, %%"REG_a") \n\t"
"add %%"REG_a", %2 \n\t"
"subl $4, %0 \n\t"
"jnz 1b \n\t"
:"+g"(h), "+S"(pixels), "+D"(block)
:"r" ((x86_reg)line_size)
:"%"REG_a, "memory");
DEF(ff_avg_pixels8_y2)(block, pixels, line_size, h);
DEF(ff_avg_pixels8_y2)(block + 8, pixels + 8, line_size, h);
}
/* Note this is not correctly rounded, but this function is only
* used for B-frames so it does not matter. */
static void DEF(avg_pixels8_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
static void DEF(ff_avg_pixels16_xy2)(uint8_t *block, const uint8_t *pixels,
int line_size, int h)
{
MOVQ_BONE(mm6);
__asm__ volatile(
"lea (%3, %3), %%"REG_a" \n\t"
"movq (%1), %%mm0 \n\t"
PAVGB" 1(%1), %%mm0 \n\t"
".p2align 3 \n\t"
"1: \n\t"
"movq (%1, %%"REG_a"), %%mm2 \n\t"
"movq (%1, %3), %%mm1 \n\t"
"psubusb %%mm6, %%mm2 \n\t"
PAVGB" 1(%1, %3), %%mm1 \n\t"
PAVGB" 1(%1, %%"REG_a"), %%mm2 \n\t"
"add %%"REG_a", %1 \n\t"
PAVGB" %%mm1, %%mm0 \n\t"
PAVGB" %%mm2, %%mm1 \n\t"
PAVGB" (%2), %%mm0 \n\t"
PAVGB" (%2, %3), %%mm1 \n\t"
"movq %%mm0, (%2) \n\t"
"movq %%mm1, (%2, %3) \n\t"
"movq (%1, %3), %%mm1 \n\t"
"movq (%1, %%"REG_a"), %%mm0 \n\t"
PAVGB" 1(%1, %3), %%mm1 \n\t"
PAVGB" 1(%1, %%"REG_a"), %%mm0 \n\t"
"add %%"REG_a", %2 \n\t"
"add %%"REG_a", %1 \n\t"
PAVGB" %%mm1, %%mm2 \n\t"
PAVGB" %%mm0, %%mm1 \n\t"
PAVGB" (%2), %%mm2 \n\t"
PAVGB" (%2, %3), %%mm1 \n\t"
"movq %%mm2, (%2) \n\t"
"movq %%mm1, (%2, %3) \n\t"
"add %%"REG_a", %2 \n\t"
"subl $4, %0 \n\t"
"jnz 1b \n\t"
:"+g"(h), "+S"(pixels), "+D"(block)
:"r" ((x86_reg)line_size)
:"%"REG_a, "memory");
}
//FIXME the following could be optimized too ...
static void DEF(put_no_rnd_pixels16_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
DEF(put_no_rnd_pixels8_x2)(block , pixels , line_size, h);
DEF(put_no_rnd_pixels8_x2)(block+8, pixels+8, line_size, h);
}
static void DEF(put_pixels16_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
DEF(put_pixels8_y2)(block , pixels , line_size, h);
DEF(put_pixels8_y2)(block+8, pixels+8, line_size, h);
}
static void DEF(put_no_rnd_pixels16_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
DEF(put_no_rnd_pixels8_y2)(block , pixels , line_size, h);
DEF(put_no_rnd_pixels8_y2)(block+8, pixels+8, line_size, h);
}
static void DEF(avg_pixels16)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
DEF(avg_pixels8)(block , pixels , line_size, h);
DEF(avg_pixels8)(block+8, pixels+8, line_size, h);
}
static void DEF(avg_pixels16_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
DEF(avg_pixels8_x2)(block , pixels , line_size, h);
DEF(avg_pixels8_x2)(block+8, pixels+8, line_size, h);
}
static void DEF(avg_pixels16_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
DEF(avg_pixels8_y2)(block , pixels , line_size, h);
DEF(avg_pixels8_y2)(block+8, pixels+8, line_size, h);
}
static void DEF(avg_pixels16_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
DEF(avg_pixels8_xy2)(block , pixels , line_size, h);
DEF(avg_pixels8_xy2)(block+8, pixels+8, line_size, h);
DEF(ff_avg_pixels8_xy2)(block, pixels, line_size, h);
DEF(ff_avg_pixels8_xy2)(block + 8, pixels + 8, line_size, h);
}

874
libavcodec/x86/dsputil_mmx.c
View File

File diff suppressed because it is too large Load Diff

465
libavcodec/x86/hpeldsp.asm
Unescape View File

@ -0,0 +1,465 @@
;******************************************************************************
;* MMX optimized hpel functions
;*
;* This file is part of Libav.
;*
;* Libav is free software; you can redistribute it and/or
;* modify it under the terms of the GNU Lesser General Public
;* License as published by the Free Software Foundation; either
;* version 2.1 of the License, or (at your option) any later version.
;*
;* Libav is distributed in the hope that it will be useful,
;* 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
;* License along with Libav; if not, write to the Free Software
;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
;******************************************************************************
%include "libavutil/x86/x86util.asm"
SECTION_RODATA
cextern pb_1
SECTION_TEXT
; put_pixels8_x2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
%macro PUT_PIXELS8_X2 0
cglobal put_pixels8_x2, 4,5
movsxdifnidn r2, r2d
lea r4, [r2*2]
.loop:
mova m0, [r1]
mova m1, [r1+r2]
PAVGB m0, [r1+1]
PAVGB m1, [r1+r2+1]
mova [r0], m0
mova [r0+r2], m1
add r1, r4
add r0, r4
mova m0, [r1]
mova m1, [r1+r2]
PAVGB m0, [r1+1]
PAVGB m1, [r1+r2+1]
add r1, r4
mova [r0], m0
mova [r0+r2], m1
add r0, r4
sub r3d, 4
jne .loop
REP_RET
%endmacro
INIT_MMX mmxext
PUT_PIXELS8_X2
INIT_MMX 3dnow
PUT_PIXELS8_X2
; put_pixels16_x2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
%macro PUT_PIXELS_16 0
cglobal put_pixels16_x2, 4,5
movsxdifnidn r2, r2d
lea r4, [r2*2]
.loop:
mova m0, [r1]
mova m1, [r1+r2]
mova m2, [r1+8]
mova m3, [r1+r2+8]
PAVGB m0, [r1+1]
PAVGB m1, [r1+r2+1]
PAVGB m2, [r1+9]
PAVGB m3, [r1+r2+9]
mova [r0], m0
mova [r0+r2], m1
mova [r0+8], m2
mova [r0+r2+8], m3
add r1, r4
add r0, r4
mova m0, [r1]
mova m1, [r1+r2]
mova m2, [r1+8]
mova m3, [r1+r2+8]
PAVGB m0, [r1+1]
PAVGB m1, [r1+r2+1]
PAVGB m2, [r1+9]
PAVGB m3, [r1+r2+9]
add r1, r4
mova [r0], m0
mova [r0+r2], m1
mova [r0+8], m2
mova [r0+r2+8], m3
add r0, r4
sub r3d, 4
jne .loop
REP_RET
%endmacro
INIT_MMX mmxext
PUT_PIXELS_16
INIT_MMX 3dnow
PUT_PIXELS_16
; put_no_rnd_pixels8_x2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
%macro PUT_NO_RND_PIXELS8_X2 0
cglobal put_no_rnd_pixels8_x2, 4,5
mova m6, [pb_1]
movsxdifnidn r2, r2d
lea r4, [r2*2]
.loop:
mova m0, [r1]
mova m2, [r1+r2]
mova m1, [r1+1]
mova m3, [r1+r2+1]
add r1, r4
psubusb m0, m6
psubusb m2, m6
PAVGB m0, m1
PAVGB m2, m3
mova [r0], m0
mova [r0+r2], m2
mova m0, [r1]
mova m1, [r1+1]
mova m2, [r1+r2]
mova m3, [r1+r2+1]
add r0, r4
add r1, r4
psubusb m0, m6
psubusb m2, m6
PAVGB m0, m1
PAVGB m2, m3
mova [r0], m0
mova [r0+r2], m2
add r0, r4
sub r3d, 4
jne .loop
REP_RET
%endmacro
INIT_MMX mmxext
PUT_NO_RND_PIXELS8_X2
INIT_MMX 3dnow
PUT_NO_RND_PIXELS8_X2
; put_no_rnd_pixels8_x2_exact(uint8_t *block, const uint8_t *pixels, int line_size, int h)
%macro PUT_NO_RND_PIXELS8_X2_EXACT 0
cglobal put_no_rnd_pixels8_x2_exact, 4,5
movsxdifnidn r2, r2d
lea r4, [r2*3]
pcmpeqb m6, m6
.loop:
mova m0, [r1]
mova m2, [r1+r2]
mova m1, [r1+1]
mova m3, [r1+r2+1]
pxor m0, m6
pxor m2, m6
pxor m1, m6
pxor m3, m6
PAVGB m0, m1
PAVGB m2, m3
pxor m0, m6
pxor m2, m6
mova [r0], m0
mova [r0+r2], m2
mova m0, [r1+r2*2]
mova m1, [r1+r2*2+1]
mova m2, [r1+r4]
mova m3, [r1+r4+1]
pxor m0, m6
pxor m1, m6
pxor m2, m6
pxor m3, m6
PAVGB m0, m1
PAVGB m2, m3
pxor m0, m6
pxor m2, m6
mova [r0+r2*2], m0
mova [r0+r4], m2
lea r1, [r1+r2*4]
lea r0, [r0+r2*4]
sub r3d, 4
jg .loop
REP_RET
%endmacro
INIT_MMX mmxext
PUT_NO_RND_PIXELS8_X2_EXACT
INIT_MMX 3dnow
PUT_NO_RND_PIXELS8_X2_EXACT
; put_pixels8_y2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
%macro PUT_PIXELS8_Y2 0
cglobal put_pixels8_y2, 4,5
movsxdifnidn r2, r2d
lea r4, [r2*2]
mova m0, [r1]
sub r0, r2
.loop:
mova m1, [r1+r2]
mova m2, [r1+r4]
add r1, r4
PAVGB m0, m1
PAVGB m1, m2
mova [r0+r2], m0
mova [r0+r4], m1
mova m1, [r1+r2]
mova m0, [r1+r4]
add r0, r4
add r1, r4
PAVGB m2, m1
PAVGB m1, m0
mova [r0+r2], m2
mova [r0+r4], m1
add r0, r4
sub r3d, 4
jne .loop
REP_RET
%endmacro
INIT_MMX mmxext
PUT_PIXELS8_Y2
INIT_MMX 3dnow
PUT_PIXELS8_Y2
; put_no_rnd_pixels8_y2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
%macro PUT_NO_RND_PIXELS8_Y2 0
cglobal put_no_rnd_pixels8_y2, 4,5
mova m6, [pb_1]
movsxdifnidn r2, r2d
lea r4, [r2+r2]
mova m0, [r1]
sub r0, r2
.loop:
mova m1, [r1+r2]
mova m2, [r1+r4]
add r1, r4
psubusb m1, m6
PAVGB m0, m1
PAVGB m1, m2
mova [r0+r2], m0
mova [r0+r4], m1
mova m1, [r1+r2]
mova m0, [r1+r4]
add r0, r4
add r1, r4
psubusb m1, m6
PAVGB m2, m1
PAVGB m1, m0
mova [r0+r2], m2
mova [r0+r4], m1
add r0, r4
sub r3d, 4
jne .loop
REP_RET
%endmacro
INIT_MMX mmxext
PUT_NO_RND_PIXELS8_Y2
INIT_MMX 3dnow
PUT_NO_RND_PIXELS8_Y2
; put_no_rnd_pixels8_y2_exact(uint8_t *block, const uint8_t *pixels, int line_size, int h)
%macro PUT_NO_RND_PIXELS8_Y2_EXACT 0
cglobal put_no_rnd_pixels8_y2_exact, 4,5
movsxdifnidn r2, r2d
lea r4, [r2*3]
mova m0, [r1]
pcmpeqb m6, m6
add r1, r2
pxor m0, m6
.loop:
mova m1, [r1]
mova m2, [r1+r2]
pxor m1, m6
pxor m2, m6
PAVGB m0, m1
PAVGB m1, m2
pxor m0, m6
pxor m1, m6
mova [r0], m0
mova [r0+r2], m1
mova m1, [r1+r2*2]
mova m0, [r1+r4]
pxor m1, m6
pxor m0, m6
PAVGB m2, m1
PAVGB m1, m0
pxor m2, m6
pxor m1, m6
mova [r0+r2*2], m2
mova [r0+r4], m1
lea r1, [r1+r2*4]
lea r0, [r0+r2*4]
sub r3d, 4
jg .loop
REP_RET
%endmacro
INIT_MMX mmxext
PUT_NO_RND_PIXELS8_Y2_EXACT
INIT_MMX 3dnow
PUT_NO_RND_PIXELS8_Y2_EXACT
; avg_pixels8(uint8_t *block, const uint8_t *pixels, int line_size, int h)
%macro AVG_PIXELS8 0
cglobal avg_pixels8, 4,5
movsxdifnidn r2, edx
lea r4, [r2+r2]
.loop:
mova m0, [r0]
mova m1, [r0+r2]
PAVGB m0, [r1]
PAVGB m1, [r1+r2]
mova [r0], m0
mova [r0+r2], m1
add r1, r4
add r0, r4
mova m0, [r0]
mova m1, [r0+r2]
PAVGB m0, [r1]
PAVGB m1, [r1+r2]
add r1, r4
mova [r0], m0
mova [r0+r2], m1
add r0, r4
sub r3d, 4
jne .loop
REP_RET
%endmacro
INIT_MMX 3dnow
AVG_PIXELS8
; avg_pixels8_x2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
%macro AVG_PIXELS8_X2 0
cglobal avg_pixels8_x2, 4,5
movsxdifnidn r2, edx
lea r4, [r2*2]
.loop:
mova m0, [r1]
mova m2, [r1+r2]
PAVGB m0, [r1+1]
PAVGB m2, [r1+r2+1]
PAVGB m0, [r0]
PAVGB m2, [r0+r2]
add r1, r4
mova [r0], m0
mova [r0+r2], m2
mova m0, [r1]
mova m2, [r1+r2]
PAVGB m0, [r1+1]
PAVGB m2, [r1+r2+1]
add r0, r4
add r1, r4
PAVGB m0, [r0]
PAVGB m2, [r0+r2]
mova [r0], m0
mova [r0+r2], m2
add r0, r4
sub r3d, 4
jne .loop
REP_RET
%endmacro
INIT_MMX mmxext
AVG_PIXELS8_X2
INIT_MMX 3dnow
AVG_PIXELS8_X2
; avg_pixels8_y2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
%macro AVG_PIXELS8_Y2 0
cglobal avg_pixels8_y2, 4,5
movsxdifnidn r2, r2d
lea r4, [r2*2]
mova m0, [r1]
sub r0, r2
.loop:
mova m1, [r1+r2]
mova m2, [r1+r4]
add r1, r4
PAVGB m0, m1
PAVGB m1, m2
mova m3, [r0+r2]
mova m4, [r0+r4]
PAVGB m0, m3
PAVGB m1, m4
mova [r0+r2], m0
mova [r0+r4], m1
mova m1, [r1+r2]
mova m0, [r1+r4]
PAVGB m2, m1
PAVGB m1, m0
add r0, r4
add r1, r4
mova m3, [r0+r2]
mova m4, [r0+r4]
PAVGB m2, m3
PAVGB m1, m4
mova [r0+r2], m2
mova [r0+r4], m1
add r0, r4
sub r3d, 4
jne .loop
REP_RET
%endmacro
INIT_MMX mmxext
AVG_PIXELS8_Y2
INIT_MMX 3dnow
AVG_PIXELS8_Y2
; avg_pixels8_xy2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
%macro AVG_PIXELS8_XY2 0
cglobal avg_pixels8_xy2, 4,5
mova m6, [pb_1]
movsxdifnidn r2, r2d
lea r4, [r2*2]
mova m0, [r1]
pavgb m0, [r1+1]
.loop:
mova m2, [r1+r4]
mova m1, [r1+r2]
psubusb m2, m6
pavgb m1, [r1+r2+1]
pavgb m2, [r1+r4+1]
add r1, r4
pavgb m0, m1
pavgb m1, m2
pavgb m0, [r0]
pavgb m1, [r0+r2]
mova [r0], m0
mova [r0+r2], m1
mova m1, [r1+r2]
mova m0, [r1+r4]
pavgb m1, [r1+r2+1]
pavgb m0, [r1+r4+1]
add r0, r4
add r1, r4
pavgb m2, m1
pavgb m1, m0
pavgb m2, [r0]
pavgb m1, [r0+r2]
mova [r0], m2
mova [r0+r2], m2
add r0, r4
sub r3d, 4
jne .loop
REP_RET
%endmacro
INIT_MMX mmxext
AVG_PIXELS8_XY2
INIT_MMX 3dnow
AVG_PIXELS8_XY2

558
libavcodec/x86/mpeg4qpel.asm
Unescape View File

@ -0,0 +1,558 @@
;******************************************************************************
;* mpeg4 qpel
;* Copyright (c) 2008 Loren Merritt
;*
;* This file is part of Libav.
;*
;* Libav is free software; you can redistribute it and/or
;* modify it under the terms of the GNU Lesser General Public
;* License as published by the Free Software Foundation; either
;* version 2.1 of the License, or (at your option) any later version.
;*
;* Libav is distributed in the hope that it will be useful,
;* 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
;* License along with Libav; if not, write to the Free Software
;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
;******************************************************************************
%include "libavutil/x86/x86util.asm"
SECTION_RODATA
cextern pb_1
cextern pw_3
cextern pw_15
cextern pw_16
cextern pw_20
SECTION_TEXT
; put_no_rnd_pixels8_l2(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
%macro PUT_NO_RND_PIXELS8_L2 0
cglobal put_no_rnd_pixels8_l2, 6,6
movsxdifnidn r4, r4d
movsxdifnidn r3, r3d
pcmpeqb m6, m6
test r5d, 1
je .loop
mova m0, [r1]
mova m1, [r2]
add r1, r4
add r2, 8
pxor m0, m6
pxor m1, m6
PAVGB m0, m1
pxor m0, m6
mova [r0], m0
add r0, r3
dec r5d
.loop:
mova m0, [r1]
add r1, r4
mova m1, [r1]
add r1, r4
mova m2, [r2]
mova m3, [r2+8]
pxor m0, m6
pxor m1, m6
pxor m2, m6
pxor m3, m6
PAVGB m0, m2
PAVGB m1, m3
pxor m0, m6
pxor m1, m6
mova [r0], m0
add r0, r3
mova [r0], m1
add r0, r3
mova m0, [r1]
add r1, r4
mova m1, [r1]
add r1, r4
mova m2, [r2+16]
mova m3, [r2+24]
pxor m0, m6
pxor m1, m6
pxor m2, m6
pxor m3, m6
PAVGB m0, m2
PAVGB m1, m3
pxor m0, m6
pxor m1, m6
mova [r0], m0
add r0, r3
mova [r0], m1
add r0, r3
add r2, 32
sub r5d, 4
jne .loop
REP_RET
%endmacro
INIT_MMX mmxext
PUT_NO_RND_PIXELS8_L2
; put_no_rnd_pixels16_l2(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
%macro PUT_NO_RND_PIXELS16_l2 0
cglobal put_no_rnd_pixels16_l2, 5,5
movsxdifnidn r3, r3
movsxdifnidn r4, r4d
pcmpeqb m6, m6
test r5d, 1
je .loop
mova m0, [r1]
mova m1, [r1+8]
mova m2, [r2]
mova m3, [r2+8]
pxor m0, m6
pxor m1, m6
pxor m2, m6
pxor m3, m6
PAVGB m0, m2
PAVGB m1, m3
pxor m0, m6
pxor m1, m6
add r1, r4
add r2, 16
mova [r0], m0
mova [r0+8], m1
add r0, r3
dec r5d
.loop:
mova m0, [r1]
mova m1, [r1+8]
add r1, r4
mova m2, [r2]
mova m3, [r2+8]
pxor m0, m6
pxor m1, m6
pxor m2, m6
pxor m3, m6
PAVGB m0, m2
PAVGB m1, m3
pxor m0, m6
pxor m1, m6
mova [r0], m0
mova [r0+8], m1
add r0, r3
mova m0, [r1]
mova m1, [r1+8]
add r1, r4
mova m2, [r2+16]
mova m3, [r2+24]
pxor m0, m6
pxor m1, m6
pxor m2, m6
pxor m3, m6
PAVGB m0, m2
PAVGB m1, m3
pxor m0, m6
pxor m1, m6
mova [r0], m0
mova [r0+8], m1
add r0, r3
add r2, 32
sub r5d, 2
jne .loop
REP_RET
%endmacro
INIT_MMX mmxext
PUT_NO_RND_PIXELS16_l2
INIT_MMX 3dnow
PUT_NO_RND_PIXELS16_l2
%macro MPEG4_QPEL16_H_LOWPASS 1
cglobal %1_mpeg4_qpel16_h_lowpass, 5, 5, 0, 8
movsxdifnidn r2, r2d
movsxdifnidn r3, r3d
pxor m7, m7
.loop:
mova m0, [r1]
mova m1, m0
mova m2, m0
punpcklbw m0, m7
punpckhbw m1, m7
pshufw m5, m0, 0x90
pshufw m6, m0, 0x41
mova m3, m2
mova m4, m2
psllq m2, 8
psllq m3, 16
psllq m4, 24
punpckhbw m2, m7
punpckhbw m3, m7
punpckhbw m4, m7
paddw m5, m3
paddw m6, m2
paddw m5, m5
psubw m6, m5
pshufw m5, m0, 6
pmullw m6, [pw_3]
paddw m0, m4
paddw m5, m1
pmullw m0, [pw_20]
psubw m0, m5
paddw m6, [PW_ROUND]
paddw m0, m6
psraw m0, 5
mova [rsp-8], m0
mova m0, [r1+5]
mova m5, m0
mova m6, m0
psrlq m0, 8
psrlq m5, 16
punpcklbw m0, m7
punpcklbw m5, m7
paddw m2, m0
paddw m3, m5
paddw m2, m2
psubw m3, m2
mova m2, m6
psrlq m6, 24
punpcklbw m2, m7
punpcklbw m6, m7
pmullw m3, [pw_3]
paddw m1, m2
paddw m4, m6
pmullw m1, [pw_20]
psubw m3, m4
paddw m1, [PW_ROUND]
paddw m3, m1
psraw m3, 5
mova m1, [rsp-8]
packuswb m1, m3
OP_MOV [r0], m1, m4
mova m1, [r1+9]
mova m4, m1
mova m3, m1
psrlq m1, 8
psrlq m4, 16
punpcklbw m1, m7
punpcklbw m4, m7
paddw m5, m1
paddw m0, m4
paddw m5, m5
psubw m0, m5
mova m5, m3
psrlq m3, 24
pmullw m0, [pw_3]
punpcklbw m3, m7
paddw m2, m3
psubw m0, m2
mova m2, m5
punpcklbw m2, m7
punpckhbw m5, m7
paddw m6, m2
pmullw m6, [pw_20]
paddw m0, [PW_ROUND]
paddw m0, m6
psraw m0, 5
paddw m3, m5
pshufw m6, m5, 0xf9
paddw m6, m4
pshufw m4, m5, 0xbe
pshufw m5, m5, 0x6f
paddw m4, m1
paddw m5, m2
paddw m6, m6
psubw m4, m6
pmullw m3, [pw_20]
pmullw m4, [pw_3]
psubw m3, m5
paddw m4, [PW_ROUND]
paddw m4, m3
psraw m4, 5
packuswb m0, m4
OP_MOV [r0+8], m0, m4
add r1, r3
add r0, r2
dec r4d
jne .loop
REP_RET
%endmacro
%macro PUT_OP 2-3
mova %1, %2
%endmacro
%macro AVG_OP 2-3
mova %3, %1
pavgb %2, %3
mova %1, %2
%endmacro
INIT_MMX mmxext
%define PW_ROUND pw_16
%define OP_MOV PUT_OP
MPEG4_QPEL16_H_LOWPASS put
%define PW_ROUND pw_16
%define OP_MOV AVG_OP
MPEG4_QPEL16_H_LOWPASS avg
%define PW_ROUND pw_15
%define OP_MOV PUT_OP
MPEG4_QPEL16_H_LOWPASS put_no_rnd
%macro MPEG4_QPEL8_H_LOWPASS 1
cglobal %1_mpeg4_qpel8_h_lowpass, 5, 5, 0, 8
movsxdifnidn r2, r2d
movsxdifnidn r3, r3d
pxor m7, m7
.loop:
mova m0, [r1]
mova m1, m0
mova m2, m0
punpcklbw m0, m7
punpckhbw m1, m7
pshufw m5, m0, 0x90
pshufw m6, m0, 0x41
mova m3, m2
mova m4, m2
psllq m2, 8
psllq m3, 16
psllq m4, 24
punpckhbw m2, m7
punpckhbw m3, m7
punpckhbw m4, m7
paddw m5, m3
paddw m6, m2
paddw m5, m5
psubw m6, m5
pshufw m5, m0, 0x6
pmullw m6, [pw_3]
paddw m0, m4
paddw m5, m1
pmullw m0, [pw_20]
psubw m0, m5
paddw m6, [PW_ROUND]
paddw m0, m6
psraw m0, 5
movh m5, [r1+5]
punpcklbw m5, m7
pshufw m6, m5, 0xf9
paddw m1, m5
paddw m2, m6
pshufw m6, m5, 0xbe
pshufw m5, m5, 0x6f
paddw m3, m6
paddw m4, m5
paddw m2, m2
psubw m3, m2
pmullw m1, [pw_20]
pmullw m3, [pw_3]
psubw m3, m4
paddw m1, [PW_ROUND]
paddw m3, m1
psraw m3, 5
packuswb m0, m3
OP_MOV [r0], m0, m4
add r1, r3
add r0, r2
dec r4d
jne .loop
REP_RET
%endmacro
INIT_MMX mmxext
%define PW_ROUND pw_16
%define OP_MOV PUT_OP
MPEG4_QPEL8_H_LOWPASS put
%define PW_ROUND pw_16
%define OP_MOV AVG_OP
MPEG4_QPEL8_H_LOWPASS avg
%define PW_ROUND pw_15
%define OP_MOV PUT_OP
MPEG4_QPEL8_H_LOWPASS put_no_rnd
%macro QPEL_V_LOW 5
paddw m0, m1
mova m4, [pw_20]
pmullw m4, m0
mova m0, %4
mova m5, %1
paddw m5, m0
psubw m4, m5
mova m5, %2
mova m6, %3
paddw m5, m3
paddw m6, m2
paddw m6, m6
psubw m5, m6
pmullw m5, [pw_3]
paddw m4, [PW_ROUND]
paddw m5, m4
psraw m5, 5
packuswb m5, m5
OP_MOV %5, m5, m7
SWAP 0,1,2,3
%endmacro
%macro MPEG4_QPEL16_V_LOWPASS 1
cglobal %1_mpeg4_qpel16_v_lowpass, 4, 6, 0, 544
movsxdifnidn r2, r2d
movsxdifnidn r3, r3d
mov r4d, 17
mov r5, rsp
pxor m7, m7
.looph:
mova m0, [r1]
mova m1, [r1]
mova m2, [r1+8]
mova m3, [r1+8]
punpcklbw m0, m7
punpckhbw m1, m7
punpcklbw m2, m7
punpckhbw m3, m7
mova [r5], m0
mova [r5+0x88], m1
mova [r5+0x110], m2
mova [r5+0x198], m3
add r5, 8
add r1, r3
dec r4d
jne .looph
; NOTE: r1 CHANGES VALUES: r1 -> 4 - 14*dstStride
mov r4d, 4
mov r1, 4
neg r2
lea r1, [r1+r2*8]
lea r1, [r1+r2*4]
lea r1, [r1+r2*2]
neg r2
mov r5, rsp
.loopv:
pxor m7, m7
mova m0, [r5+ 0x0]
mova m1, [r5+ 0x8]
mova m2, [r5+0x10]
mova m3, [r5+0x18]
QPEL_V_LOW [r5+0x10], [r5+ 0x8], [r5+ 0x0], [r5+0x20], [r0]
QPEL_V_LOW [r5+ 0x8], [r5+ 0x0], [r5+ 0x0], [r5+0x28], [r0+r2]
lea r0, [r0+r2*2]
QPEL_V_LOW [r5+ 0x0], [r5+ 0x0], [r5+ 0x8], [r5+0x30], [r0]
QPEL_V_LOW [r5+ 0x0], [r5+ 0x8], [r5+0x10], [r5+0x38], [r0+r2]
lea r0, [r0+r2*2]
QPEL_V_LOW [r5+ 0x8], [r5+0x10], [r5+0x18], [r5+0x40], [r0]
QPEL_V_LOW [r5+0x10], [r5+0x18], [r5+0x20], [r5+0x48], [r0+r2]
lea r0, [r0+r2*2]
QPEL_V_LOW [r5+0x18], [r5+0x20], [r5+0x28], [r5+0x50], [r0]
QPEL_V_LOW [r5+0x20], [r5+0x28], [r5+0x30], [r5+0x58], [r0+r2]
lea r0, [r0+r2*2]
QPEL_V_LOW [r5+0x28], [r5+0x30], [r5+0x38], [r5+0x60], [r0]
QPEL_V_LOW [r5+0x30], [r5+0x38], [r5+0x40], [r5+0x68], [r0+r2]
lea r0, [r0+r2*2]
QPEL_V_LOW [r5+0x38], [r5+0x40], [r5+0x48], [r5+0x70], [r0]
QPEL_V_LOW [r5+0x40], [r5+0x48], [r5+0x50], [r5+0x78], [r0+r2]
lea r0, [r0+r2*2]
QPEL_V_LOW [r5+0x48], [r5+0x50], [r5+0x58], [r5+0x80], [r0]
QPEL_V_LOW [r5+0x50], [r5+0x58], [r5+0x60], [r5+0x80], [r0+r2]
lea r0, [r0+r2*2]
QPEL_V_LOW [r5+0x58], [r5+0x60], [r5+0x68], [r5+0x78], [r0]
QPEL_V_LOW [r5+0x60], [r5+0x68], [r5+0x70], [r5+0x70], [r0+r2]
add r5, 0x88
add r0, r1
dec r4d
jne .loopv
REP_RET
%endmacro
%macro PUT_OPH 2-3
movh %1, %2
%endmacro
%macro AVG_OPH 2-3
movh %3, %1
pavgb %2, %3
movh %1, %2
%endmacro
INIT_MMX mmxext
%define PW_ROUND pw_16
%define OP_MOV PUT_OPH
MPEG4_QPEL16_V_LOWPASS put
%define PW_ROUND pw_16
%define OP_MOV AVG_OPH
MPEG4_QPEL16_V_LOWPASS avg
%define PW_ROUND pw_15
%define OP_MOV PUT_OPH
MPEG4_QPEL16_V_LOWPASS put_no_rnd
%macro MPEG4_QPEL8_V_LOWPASS 1
cglobal %1_mpeg4_qpel8_v_lowpass, 4, 6, 0, 288
movsxdifnidn r2, r2d
movsxdifnidn r3, r3d
mov r4d, 9
mov r5, rsp
pxor m7, m7
.looph:
mova m0, [r1]
mova m1, [r1]
punpcklbw m0, m7
punpckhbw m1, m7
mova [r5], m0
mova [r5+0x48], m1
add r5, 8
add r1, r3
dec r4d
jne .looph
; NOTE: r1 CHANGES VALUES: r1 -> 4 - 6*dstStride
mov r4d, 2
mov r1, 4
neg r2
lea r1, [r1+r2*4]
lea r1, [r1+r2*2]
neg r2
mov r5, rsp
.loopv:
pxor m7, m7
mova m0, [r5+ 0x0]
mova m1, [r5+ 0x8]
mova m2, [r5+0x10]
mova m3, [r5+0x18]
QPEL_V_LOW [r5+0x10], [r5+ 0x8], [r5+ 0x0], [r5+0x20], [r0]
QPEL_V_LOW [r5+ 0x8], [r5+ 0x0], [r5+ 0x0], [r5+0x28], [r0+r2]
lea r0, [r0+r2*2]
QPEL_V_LOW [r5+ 0x0], [r5+ 0x0], [r5+ 0x8], [r5+0x30], [r0]
QPEL_V_LOW [r5+ 0x0], [r5+ 0x8], [r5+0x10], [r5+0x38], [r0+r2]
lea r0, [r0+r2*2]
QPEL_V_LOW [r5+ 0x8], [r5+0x10], [r5+0x18], [r5+0x40], [r0]
QPEL_V_LOW [r5+0x10], [r5+0x18], [r5+0x20], [r5+0x40], [r0+r2]
lea r0, [r0+r2*2]
QPEL_V_LOW [r5+0x18], [r5+0x20], [r5+0x28], [r5+0x38], [r0]
QPEL_V_LOW [r5+0x20], [r5+0x28], [r5+0x30], [r5+0x30], [r0+r2]
add r5, 0x48
add r0, r1
dec r4d
jne .loopv
REP_RET
%endmacro
INIT_MMX mmxext
%define PW_ROUND pw_16
%define OP_MOV PUT_OPH
MPEG4_QPEL8_V_LOWPASS put
%define PW_ROUND pw_16
%define OP_MOV AVG_OPH
MPEG4_QPEL8_V_LOWPASS avg
%define PW_ROUND pw_15
%define OP_MOV PUT_OPH
MPEG4_QPEL8_V_LOWPASS put_no_rnd

4
libavcodec/x86/vc1dsp_mmx.c
Unescape View File

@ -697,7 +697,9 @@ static void vc1_inv_trans_8x8_dc_mmxext(uint8_t *dest, int linesize,
av_cold void ff_vc1dsp_init_mmx(VC1DSPContext *dsp)
{
#if HAVE_YASM
dsp->put_vc1_mspel_pixels_tab[ 0] = ff_put_vc1_mspel_mc00_mmx;
#endif /* HAVE_YASM */
dsp->put_vc1_mspel_pixels_tab[ 4] = put_vc1_mspel_mc01_mmx;
dsp->put_vc1_mspel_pixels_tab[ 8] = put_vc1_mspel_mc02_mmx;
dsp->put_vc1_mspel_pixels_tab[12] = put_vc1_mspel_mc03_mmx;
@ -720,7 +722,9 @@ av_cold void ff_vc1dsp_init_mmx(VC1DSPContext *dsp)
av_cold void ff_vc1dsp_init_mmxext(VC1DSPContext *dsp)
{
#if HAVE_YASM
dsp->avg_vc1_mspel_pixels_tab[ 0] = ff_avg_vc1_mspel_mc00_mmxext;
#endif /* HAVE_YASM */
dsp->avg_vc1_mspel_pixels_tab[ 4] = avg_vc1_mspel_mc01_mmxext;
dsp->avg_vc1_mspel_pixels_tab[ 8] = avg_vc1_mspel_mc02_mmxext;
dsp->avg_vc1_mspel_pixels_tab[12] = avg_vc1_mspel_mc03_mmxext;

Write Preview
Loading…
Cancel
Save