[FFmpeg-cvslog] yadif: Port inline assembly to yasm
Daniel Kang
git at videolan.org
Thu Jan 10 12:32:42 CET 2013
ffmpeg | branch: master | Daniel Kang <daniel.d.kang at gmail.com> | Wed Jan 9 09:34:46 2013 -0600| [899157b30800bb986867235714f4eba353fead07] | committer: Luca Barbato
yadif: Port inline assembly to yasm
Signed-off-by: Luca Barbato <lu_zero at gentoo.org>
> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=899157b30800bb986867235714f4eba353fead07
---
libavfilter/x86/Makefile | 3 +-
libavfilter/x86/yadif.asm | 242 ++++++++++++++++++++++++++
libavfilter/x86/{yadif.c => yadif_init.c} | 60 +++----
libavfilter/x86/yadif_template.c | 261 -----------------------------
4 files changed, 264 insertions(+), 302 deletions(-)
diff --git a/libavfilter/x86/Makefile b/libavfilter/x86/Makefile
index 0f08e39..47569cf 100644
--- a/libavfilter/x86/Makefile
+++ b/libavfilter/x86/Makefile
@@ -1,6 +1,7 @@
OBJS-$(CONFIG_GRADFUN_FILTER) += x86/gradfun.o
OBJS-$(CONFIG_VOLUME_FILTER) += x86/af_volume_init.o
-OBJS-$(CONFIG_YADIF_FILTER) += x86/yadif.o
+OBJS-$(CONFIG_YADIF_FILTER) += x86/yadif_init.o
YASM-OBJS-$(CONFIG_HQDN3D_FILTER) += x86/hqdn3d.o
YASM-OBJS-$(CONFIG_VOLUME_FILTER) += x86/af_volume.o
+YASM-OBJS-$(CONFIG_YADIF_FILTER) += x86/yadif.o
diff --git a/libavfilter/x86/yadif.asm b/libavfilter/x86/yadif.asm
new file mode 100644
index 0000000..5e406a4
--- /dev/null
+++ b/libavfilter/x86/yadif.asm
@@ -0,0 +1,242 @@
+;*****************************************************************************
+;* x86-optimized functions for yadif filter
+;*
+;* Copyright (C) 2006 Michael Niedermayer <michaelni at gmx.at>
+;* Copyright (c) 2013 Daniel Kang <daniel.d.kang at gmail.com>
+;*
+;* This file is part of Libav.
+;*
+;* Libav is free software; you can redistribute it and/or modify
+;* it under the terms of the GNU General Public License as published by
+;* the Free Software Foundation; either version 2 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 General Public License for more details.
+;*
+;* You should have received a copy of the GNU 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
+
+pb_1: times 16 db 1
+pw_1: times 8 dw 1
+
+SECTION .text
+
+%macro CHECK 2
+ movu m2, [curq+mrefsq+%1]
+ movu m3, [curq+prefsq+%2]
+ mova m4, m2
+ mova m5, m2
+ pxor m4, m3
+ pavgb m5, m3
+ pand m4, [pb_1]
+ psubusb m5, m4
+%if mmsize == 16
+ psrldq m5, 1
+%else
+ psrlq m5, 8
+%endif
+ punpcklbw m5, m7
+ mova m4, m2
+ psubusb m2, m3
+ psubusb m3, m4
+ pmaxub m2, m3
+ mova m3, m2
+ mova m4, m2
+%if mmsize == 16
+ psrldq m3, 1
+ psrldq m4, 2
+%else
+ psrlq m3, 8
+ psrlq m4, 16
+%endif
+ punpcklbw m2, m7
+ punpcklbw m3, m7
+ punpcklbw m4, m7
+ paddw m2, m3
+ paddw m2, m4
+%endmacro
+
+%macro CHECK1 0
+ mova m3, m0
+ pcmpgtw m3, m2
+ pminsw m0, m2
+ mova m6, m3
+ pand m5, m3
+ pandn m3, m1
+ por m3, m5
+ mova m1, m3
+%endmacro
+
+%macro CHECK2 0
+ paddw m6, [pw_1]
+ psllw m6, 14
+ paddsw m2, m6
+ mova m3, m0
+ pcmpgtw m3, m2
+ pminsw m0, m2
+ pand m5, m3
+ pandn m3, m1
+ por m3, m5
+ mova m1, m3
+%endmacro
+
+%macro LOAD 2
+ movh m%1, %2
+ punpcklbw m%1, m7
+%endmacro
+
+%macro FILTER 3
+.loop%1:
+ pxor m7, m7
+ LOAD 0, [curq+mrefsq]
+ LOAD 1, [curq+prefsq]
+ LOAD 2, [%2]
+ LOAD 3, [%3]
+ mova m4, m3
+ paddw m3, m2
+ psraw m3, 1
+ mova [rsp+ 0], m0
+ mova [rsp+16], m3
+ mova [rsp+32], m1
+ psubw m2, m4
+ ABS1 m2, m4
+ LOAD 3, [prevq+mrefsq]
+ LOAD 4, [prevq+prefsq]
+ psubw m3, m0
+ psubw m4, m1
+ ABS1 m3, m5
+ ABS1 m4, m5
+ paddw m3, m4
+ psrlw m2, 1
+ psrlw m3, 1
+ pmaxsw m2, m3
+ LOAD 3, [nextq+mrefsq]
+ LOAD 4, [nextq+prefsq]
+ psubw m3, m0
+ psubw m4, m1
+ ABS1 m3, m5
+ ABS1 m4, m5
+ paddw m3, m4
+ psrlw m3, 1
+ pmaxsw m2, m3
+ mova [rsp+48], m2
+
+ paddw m1, m0
+ paddw m0, m0
+ psubw m0, m1
+ psrlw m1, 1
+ ABS1 m0, m2
+
+ movu m2, [curq+mrefsq-1]
+ movu m3, [curq+prefsq-1]
+ mova m4, m2
+ psubusb m2, m3
+ psubusb m3, m4
+ pmaxub m2, m3
+%if mmsize == 16
+ mova m3, m2
+ psrldq m3, 2
+%else
+ pshufw m3, m2, q0021
+%endif
+ punpcklbw m2, m7
+ punpcklbw m3, m7
+ paddw m0, m2
+ paddw m0, m3
+ psubw m0, [pw_1]
+
+ CHECK -2, 0
+ CHECK1
+ CHECK -3, 1
+ CHECK2
+ CHECK 0, -2
+ CHECK1
+ CHECK 1, -3
+ CHECK2
+
+ mova m6, [rsp+48]
+ cmp DWORD modem, 2
+ jge .end%1
+ LOAD 2, [%2+mrefsq*2]
+ LOAD 4, [%3+mrefsq*2]
+ LOAD 3, [%2+prefsq*2]
+ LOAD 5, [%3+prefsq*2]
+ paddw m2, m4
+ paddw m3, m5
+ psrlw m2, 1
+ psrlw m3, 1
+ mova m4, [rsp+ 0]
+ mova m5, [rsp+16]
+ mova m7, [rsp+32]
+ psubw m2, m4
+ psubw m3, m7
+ mova m0, m5
+ psubw m5, m4
+ psubw m0, m7
+ mova m4, m2
+ pminsw m2, m3
+ pmaxsw m3, m4
+ pmaxsw m2, m5
+ pminsw m3, m5
+ pmaxsw m2, m0
+ pminsw m3, m0
+ pxor m4, m4
+ pmaxsw m6, m3
+ psubw m4, m2
+ pmaxsw m6, m4
+
+.end%1:
+ mova m2, [rsp+16]
+ mova m3, m2
+ psubw m2, m6
+ paddw m3, m6
+ pmaxsw m1, m2
+ pminsw m1, m3
+ packuswb m1, m1
+
+ movh [dstq], m1
+ add dstq, mmsize/2
+ add prevq, mmsize/2
+ add curq, mmsize/2
+ add nextq, mmsize/2
+ sub wd, mmsize/2
+ jg .loop%1
+%endmacro
+
+%macro YADIF 0
+cglobal yadif_filter_line, 7, 7, 8, 16*5, dst, prev, cur, next, w, prefs, \
+ mrefs, parity, mode
+ test wq, wq
+ jle .ret
+ movsxdifnidn prefsq, prefsd
+ movsxdifnidn mrefsq, mrefsd
+
+ cmp DWORD paritym, 0
+ je .parity0
+ FILTER 1, prevq, curq
+ jmp .ret
+
+.parity0:
+ FILTER 0, curq, nextq
+
+.ret:
+ RET
+%endmacro
+
+INIT_XMM ssse3
+YADIF
+INIT_XMM sse2
+YADIF
+%if ARCH_X86_32
+INIT_MMX mmxext
+YADIF
+%endif
diff --git a/libavfilter/x86/yadif.c b/libavfilter/x86/yadif_init.c
similarity index 52%
rename from libavfilter/x86/yadif.c
rename to libavfilter/x86/yadif_init.c
index ab1d282..0cee8e5 100644
--- a/libavfilter/x86/yadif.c
+++ b/libavfilter/x86/yadif_init.c
@@ -23,52 +23,32 @@
#include "libavutil/internal.h"
#include "libavutil/mem.h"
#include "libavutil/x86/asm.h"
+#include "libavutil/x86/cpu.h"
#include "libavcodec/x86/dsputil_mmx.h"
#include "libavfilter/yadif.h"
-#if HAVE_INLINE_ASM
-
-DECLARE_ASM_CONST(16, const xmm_reg, pb_1) = {0x0101010101010101ULL, 0x0101010101010101ULL};
-DECLARE_ASM_CONST(16, const xmm_reg, pw_1) = {0x0001000100010001ULL, 0x0001000100010001ULL};
-
-#if HAVE_SSSE3_INLINE
-#define COMPILE_TEMPLATE_SSE2 1
-#define COMPILE_TEMPLATE_SSSE3 1
-#undef RENAME
-#define RENAME(a) a ## _ssse3
-#include "yadif_template.c"
-#undef COMPILE_TEMPLATE_SSSE3
-#endif
-
-#if HAVE_SSE2_INLINE
-#undef RENAME
-#define RENAME(a) a ## _sse2
-#include "yadif_template.c"
-#undef COMPILE_TEMPLATE_SSE2
-#endif
-
-#if HAVE_MMXEXT_INLINE
-#undef RENAME
-#define RENAME(a) a ## _mmxext
-#include "yadif_template.c"
-#endif
-
-#endif /* HAVE_INLINE_ASM */
+void ff_yadif_filter_line_mmxext(uint8_t *dst, uint8_t *prev, uint8_t *cur,
+ uint8_t *next, int w, int prefs,
+ int mrefs, int parity, int mode);
+void ff_yadif_filter_line_sse2(uint8_t *dst, uint8_t *prev, uint8_t *cur,
+ uint8_t *next, int w, int prefs,
+ int mrefs, int parity, int mode);
+void ff_yadif_filter_line_ssse3(uint8_t *dst, uint8_t *prev, uint8_t *cur,
+ uint8_t *next, int w, int prefs,
+ int mrefs, int parity, int mode);
av_cold void ff_yadif_init_x86(YADIFContext *yadif)
{
int cpu_flags = av_get_cpu_flags();
-#if HAVE_MMXEXT_INLINE
- if (cpu_flags & AV_CPU_FLAG_MMXEXT)
- yadif->filter_line = yadif_filter_line_mmxext;
-#endif
-#if HAVE_SSE2_INLINE
- if (cpu_flags & AV_CPU_FLAG_SSE2)
- yadif->filter_line = yadif_filter_line_sse2;
-#endif
-#if HAVE_SSSE3_INLINE
- if (cpu_flags & AV_CPU_FLAG_SSSE3)
- yadif->filter_line = yadif_filter_line_ssse3;
-#endif
+#if HAVE_YASM
+#if ARCH_X86_32
+ if (EXTERNAL_MMXEXT(cpu_flags))
+ yadif->filter_line = ff_yadif_filter_line_mmxext;
+#endif /* ARCH_X86_32 */
+ if (EXTERNAL_SSE2(cpu_flags))
+ yadif->filter_line = ff_yadif_filter_line_sse2;
+ if (EXTERNAL_SSSE3(cpu_flags))
+ yadif->filter_line = ff_yadif_filter_line_ssse3;
+#endif /* HAVE_YASM */
}
diff --git a/libavfilter/x86/yadif_template.c b/libavfilter/x86/yadif_template.c
deleted file mode 100644
index 02b0c9f..0000000
--- a/libavfilter/x86/yadif_template.c
+++ /dev/null
@@ -1,261 +0,0 @@
-/*
- * Copyright (C) 2006 Michael Niedermayer <michaelni at gmx.at>
- *
- * This file is part of Libav.
- *
- * Libav is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation; either version 2 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 General Public License for more details.
- *
- * You should have received a copy of the GNU 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.
- */
-
-#ifdef COMPILE_TEMPLATE_SSE2
-#define MM "%%xmm"
-#define MOV "movq"
-#define MOVQ "movdqa"
-#define MOVQU "movdqu"
-#define STEP 8
-#define LOAD(mem,dst) \
- MOV" "mem", "dst" \n\t"\
- "punpcklbw "MM"7, "dst" \n\t"
-#define PSRL1(reg) "psrldq $1, "reg" \n\t"
-#define PSRL2(reg) "psrldq $2, "reg" \n\t"
-#define PSHUF(src,dst) "movdqa "dst", "src" \n\t"\
- "psrldq $2, "src" \n\t"
-#else
-#define MM "%%mm"
-#define MOV "movd"
-#define MOVQ "movq"
-#define MOVQU "movq"
-#define STEP 4
-#define LOAD(mem,dst) \
- MOV" "mem", "dst" \n\t"\
- "punpcklbw "MM"7, "dst" \n\t"
-#define PSRL1(reg) "psrlq $8, "reg" \n\t"
-#define PSRL2(reg) "psrlq $16, "reg" \n\t"
-#define PSHUF(src,dst) "pshufw $9, "dst", "src" \n\t"
-#endif
-
-#ifdef COMPILE_TEMPLATE_SSSE3
-#define PABS(tmp,dst) \
- "pabsw "dst", "dst" \n\t"
-#else
-#define PABS(tmp,dst) \
- "pxor "tmp", "tmp" \n\t"\
- "psubw "dst", "tmp" \n\t"\
- "pmaxsw "tmp", "dst" \n\t"
-#endif
-
-#define CHECK(pj,mj) \
- MOVQU" "#pj"(%[cur],%[mrefs]), "MM"2 \n\t" /* cur[x-refs-1+j] */\
- MOVQU" "#mj"(%[cur],%[prefs]), "MM"3 \n\t" /* cur[x+refs-1-j] */\
- MOVQ" "MM"2, "MM"4 \n\t"\
- MOVQ" "MM"2, "MM"5 \n\t"\
- "pxor "MM"3, "MM"4 \n\t"\
- "pavgb "MM"3, "MM"5 \n\t"\
- "pand "MANGLE(pb_1)", "MM"4 \n\t"\
- "psubusb "MM"4, "MM"5 \n\t"\
- PSRL1(MM"5") \
- "punpcklbw "MM"7, "MM"5 \n\t" /* (cur[x-refs+j] + cur[x+refs-j])>>1 */\
- MOVQ" "MM"2, "MM"4 \n\t"\
- "psubusb "MM"3, "MM"2 \n\t"\
- "psubusb "MM"4, "MM"3 \n\t"\
- "pmaxub "MM"3, "MM"2 \n\t"\
- MOVQ" "MM"2, "MM"3 \n\t"\
- MOVQ" "MM"2, "MM"4 \n\t" /* ABS(cur[x-refs-1+j] - cur[x+refs-1-j]) */\
- PSRL1(MM"3") /* ABS(cur[x-refs +j] - cur[x+refs -j]) */\
- PSRL2(MM"4") /* ABS(cur[x-refs+1+j] - cur[x+refs+1-j]) */\
- "punpcklbw "MM"7, "MM"2 \n\t"\
- "punpcklbw "MM"7, "MM"3 \n\t"\
- "punpcklbw "MM"7, "MM"4 \n\t"\
- "paddw "MM"3, "MM"2 \n\t"\
- "paddw "MM"4, "MM"2 \n\t" /* score */
-
-#define CHECK1 \
- MOVQ" "MM"0, "MM"3 \n\t"\
- "pcmpgtw "MM"2, "MM"3 \n\t" /* if(score < spatial_score) */\
- "pminsw "MM"2, "MM"0 \n\t" /* spatial_score= score; */\
- MOVQ" "MM"3, "MM"6 \n\t"\
- "pand "MM"3, "MM"5 \n\t"\
- "pandn "MM"1, "MM"3 \n\t"\
- "por "MM"5, "MM"3 \n\t"\
- MOVQ" "MM"3, "MM"1 \n\t" /* spatial_pred= (cur[x-refs+j] + cur[x+refs-j])>>1; */
-
-#define CHECK2 /* pretend not to have checked dir=2 if dir=1 was bad.\
- hurts both quality and speed, but matches the C version. */\
- "paddw "MANGLE(pw_1)", "MM"6 \n\t"\
- "psllw $14, "MM"6 \n\t"\
- "paddsw "MM"6, "MM"2 \n\t"\
- MOVQ" "MM"0, "MM"3 \n\t"\
- "pcmpgtw "MM"2, "MM"3 \n\t"\
- "pminsw "MM"2, "MM"0 \n\t"\
- "pand "MM"3, "MM"5 \n\t"\
- "pandn "MM"1, "MM"3 \n\t"\
- "por "MM"5, "MM"3 \n\t"\
- MOVQ" "MM"3, "MM"1 \n\t"
-
-static void RENAME(yadif_filter_line)(uint8_t *dst, uint8_t *prev, uint8_t *cur,
- uint8_t *next, int w, int prefs,
- int mrefs, int parity, int mode)
-{
- DECLARE_ALIGNED(16, uint8_t, tmp)[16*4];
- int x;
-
-#define FILTER\
- for(x=0; x<w; x+=STEP){\
- __asm__ volatile(\
- "pxor "MM"7, "MM"7 \n\t"\
- LOAD("(%[cur],%[mrefs])", MM"0") /* c = cur[x-refs] */\
- LOAD("(%[cur],%[prefs])", MM"1") /* e = cur[x+refs] */\
- LOAD("(%["prev2"])", MM"2") /* prev2[x] */\
- LOAD("(%["next2"])", MM"3") /* next2[x] */\
- MOVQ" "MM"3, "MM"4 \n\t"\
- "paddw "MM"2, "MM"3 \n\t"\
- "psraw $1, "MM"3 \n\t" /* d = (prev2[x] + next2[x])>>1 */\
- MOVQ" "MM"0, (%[tmp]) \n\t" /* c */\
- MOVQ" "MM"3, 16(%[tmp]) \n\t" /* d */\
- MOVQ" "MM"1, 32(%[tmp]) \n\t" /* e */\
- "psubw "MM"4, "MM"2 \n\t"\
- PABS( MM"4", MM"2") /* temporal_diff0 */\
- LOAD("(%[prev],%[mrefs])", MM"3") /* prev[x-refs] */\
- LOAD("(%[prev],%[prefs])", MM"4") /* prev[x+refs] */\
- "psubw "MM"0, "MM"3 \n\t"\
- "psubw "MM"1, "MM"4 \n\t"\
- PABS( MM"5", MM"3")\
- PABS( MM"5", MM"4")\
- "paddw "MM"4, "MM"3 \n\t" /* temporal_diff1 */\
- "psrlw $1, "MM"2 \n\t"\
- "psrlw $1, "MM"3 \n\t"\
- "pmaxsw "MM"3, "MM"2 \n\t"\
- LOAD("(%[next],%[mrefs])", MM"3") /* next[x-refs] */\
- LOAD("(%[next],%[prefs])", MM"4") /* next[x+refs] */\
- "psubw "MM"0, "MM"3 \n\t"\
- "psubw "MM"1, "MM"4 \n\t"\
- PABS( MM"5", MM"3")\
- PABS( MM"5", MM"4")\
- "paddw "MM"4, "MM"3 \n\t" /* temporal_diff2 */\
- "psrlw $1, "MM"3 \n\t"\
- "pmaxsw "MM"3, "MM"2 \n\t"\
- MOVQ" "MM"2, 48(%[tmp]) \n\t" /* diff */\
-\
- "paddw "MM"0, "MM"1 \n\t"\
- "paddw "MM"0, "MM"0 \n\t"\
- "psubw "MM"1, "MM"0 \n\t"\
- "psrlw $1, "MM"1 \n\t" /* spatial_pred */\
- PABS( MM"2", MM"0") /* ABS(c-e) */\
-\
- MOVQU" -1(%[cur],%[mrefs]), "MM"2 \n\t" /* cur[x-refs-1] */\
- MOVQU" -1(%[cur],%[prefs]), "MM"3 \n\t" /* cur[x+refs-1] */\
- MOVQ" "MM"2, "MM"4 \n\t"\
- "psubusb "MM"3, "MM"2 \n\t"\
- "psubusb "MM"4, "MM"3 \n\t"\
- "pmaxub "MM"3, "MM"2 \n\t"\
- PSHUF(MM"3", MM"2") \
- "punpcklbw "MM"7, "MM"2 \n\t" /* ABS(cur[x-refs-1] - cur[x+refs-1]) */\
- "punpcklbw "MM"7, "MM"3 \n\t" /* ABS(cur[x-refs+1] - cur[x+refs+1]) */\
- "paddw "MM"2, "MM"0 \n\t"\
- "paddw "MM"3, "MM"0 \n\t"\
- "psubw "MANGLE(pw_1)", "MM"0 \n\t" /* spatial_score */\
-\
- CHECK(-2,0)\
- CHECK1\
- CHECK(-3,1)\
- CHECK2\
- CHECK(0,-2)\
- CHECK1\
- CHECK(1,-3)\
- CHECK2\
-\
- /* if(p->mode<2) ... */\
- MOVQ" 48(%[tmp]), "MM"6 \n\t" /* diff */\
- "cmpl $2, %[mode] \n\t"\
- "jge 1f \n\t"\
- LOAD("(%["prev2"],%[mrefs],2)", MM"2") /* prev2[x-2*refs] */\
- LOAD("(%["next2"],%[mrefs],2)", MM"4") /* next2[x-2*refs] */\
- LOAD("(%["prev2"],%[prefs],2)", MM"3") /* prev2[x+2*refs] */\
- LOAD("(%["next2"],%[prefs],2)", MM"5") /* next2[x+2*refs] */\
- "paddw "MM"4, "MM"2 \n\t"\
- "paddw "MM"5, "MM"3 \n\t"\
- "psrlw $1, "MM"2 \n\t" /* b */\
- "psrlw $1, "MM"3 \n\t" /* f */\
- MOVQ" (%[tmp]), "MM"4 \n\t" /* c */\
- MOVQ" 16(%[tmp]), "MM"5 \n\t" /* d */\
- MOVQ" 32(%[tmp]), "MM"7 \n\t" /* e */\
- "psubw "MM"4, "MM"2 \n\t" /* b-c */\
- "psubw "MM"7, "MM"3 \n\t" /* f-e */\
- MOVQ" "MM"5, "MM"0 \n\t"\
- "psubw "MM"4, "MM"5 \n\t" /* d-c */\
- "psubw "MM"7, "MM"0 \n\t" /* d-e */\
- MOVQ" "MM"2, "MM"4 \n\t"\
- "pminsw "MM"3, "MM"2 \n\t"\
- "pmaxsw "MM"4, "MM"3 \n\t"\
- "pmaxsw "MM"5, "MM"2 \n\t"\
- "pminsw "MM"5, "MM"3 \n\t"\
- "pmaxsw "MM"0, "MM"2 \n\t" /* max */\
- "pminsw "MM"0, "MM"3 \n\t" /* min */\
- "pxor "MM"4, "MM"4 \n\t"\
- "pmaxsw "MM"3, "MM"6 \n\t"\
- "psubw "MM"2, "MM"4 \n\t" /* -max */\
- "pmaxsw "MM"4, "MM"6 \n\t" /* diff= MAX3(diff, min, -max); */\
- "1: \n\t"\
-\
- MOVQ" 16(%[tmp]), "MM"2 \n\t" /* d */\
- MOVQ" "MM"2, "MM"3 \n\t"\
- "psubw "MM"6, "MM"2 \n\t" /* d-diff */\
- "paddw "MM"6, "MM"3 \n\t" /* d+diff */\
- "pmaxsw "MM"2, "MM"1 \n\t"\
- "pminsw "MM"3, "MM"1 \n\t" /* d = clip(spatial_pred, d-diff, d+diff); */\
- "packuswb "MM"1, "MM"1 \n\t"\
-\
- ::[prev] "r"(prev),\
- [cur] "r"(cur),\
- [next] "r"(next),\
- [prefs]"r"((x86_reg)prefs),\
- [mrefs]"r"((x86_reg)mrefs),\
- [mode] "g"(mode),\
- [tmp] "r"(tmp)\
- );\
- __asm__ volatile(MOV" "MM"1, %0" :"=m"(*dst));\
- dst += STEP;\
- prev+= STEP;\
- cur += STEP;\
- next+= STEP;\
- }
-
- if (parity) {
-#define prev2 "prev"
-#define next2 "cur"
- FILTER
-#undef prev2
-#undef next2
- } else {
-#define prev2 "cur"
-#define next2 "next"
- FILTER
-#undef prev2
-#undef next2
- }
-}
-#undef STEP
-#undef MM
-#undef MOV
-#undef MOVQ
-#undef MOVQU
-#undef PSHUF
-#undef PSRL1
-#undef PSRL2
-#undef LOAD
-#undef PABS
-#undef CHECK
-#undef CHECK1
-#undef CHECK2
-#undef FILTER
More information about the ffmpeg-cvslog
mailing list