/*
* Copyright (C) 2006 Michael Niedermayer <michaelni@gmx.at>
*
- * This file is part of MPlayer.
+ * This file is part of FFmpeg.
*
- * MPlayer is free software; you can redistribute it and/or modify
+ * FFmpeg 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.
*
- * MPlayer is distributed in the hope that it will be useful,
+ * FFmpeg 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 MPlayer; if not, write to the Free Software Foundation, Inc.,
+ * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
-/* */
-#if defined(CAN_COMPILE_SSE2) && ((__GNUC__ > 3) || (__GNUC__ == 3 && __GNUC_MINOR__ > 0))
-
-#define HAVE_YADIF_SSE2
-
-#define LOAD4(mem,dst) \
- "movd "mem", "#dst" \n\t"\
- "punpcklbw %%mm7, "#dst" \n\t"
-
-#define PABS(tmp,dst) \
- "pxor "#tmp", "#tmp" \n\t"\
- "psubw "#dst", "#tmp" \n\t"\
- "pmaxsw "#tmp", "#dst" \n\t"
-
-#define CHECK(pj,mj) \
- "movq "#pj"(%[cur],%[mrefs]), %%mm2 \n\t" /* cur[x-refs-1+j] */\
- "movq "#mj"(%[cur],%[prefs]), %%mm3 \n\t" /* cur[x+refs-1-j] */\
- "movq %%mm2, %%mm4 \n\t"\
- "movq %%mm2, %%mm5 \n\t"\
- "pxor %%mm3, %%mm4 \n\t"\
- "pavgb %%mm3, %%mm5 \n\t"\
- "pand %[pb1], %%mm4 \n\t"\
- "psubusb %%mm4, %%mm5 \n\t"\
- "psrlq $8, %%mm5 \n\t"\
- "punpcklbw %%mm7, %%mm5 \n\t" /* (cur[x-refs+j] + cur[x+refs-j])>>1 */\
- "movq %%mm2, %%mm4 \n\t"\
- "psubusb %%mm3, %%mm2 \n\t"\
- "psubusb %%mm4, %%mm3 \n\t"\
- "pmaxub %%mm3, %%mm2 \n\t"\
- "movq %%mm2, %%mm3 \n\t"\
- "movq %%mm2, %%mm4 \n\t" /* ABS(cur[x-refs-1+j] - cur[x+refs-1-j]) */\
- "psrlq $8, %%mm3 \n\t" /* ABS(cur[x-refs +j] - cur[x+refs -j]) */\
- "psrlq $16, %%mm4 \n\t" /* ABS(cur[x-refs+1+j] - cur[x+refs+1-j]) */\
- "punpcklbw %%mm7, %%mm2 \n\t"\
- "punpcklbw %%mm7, %%mm3 \n\t"\
- "punpcklbw %%mm7, %%mm4 \n\t"\
- "paddw %%mm3, %%mm2 \n\t"\
- "paddw %%mm4, %%mm2 \n\t" /* score */
-
-#define CHECK1 \
- "movq %%mm0, %%mm3 \n\t"\
- "pcmpgtw %%mm2, %%mm3 \n\t" /* if(score < spatial_score) */\
- "pminsw %%mm2, %%mm0 \n\t" /* spatial_score= score; */\
- "movq %%mm3, %%mm6 \n\t"\
- "pand %%mm3, %%mm5 \n\t"\
- "pandn %%mm1, %%mm3 \n\t"\
- "por %%mm5, %%mm3 \n\t"\
- "movq %%mm3, %%mm1 \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 %[pw1], %%mm6 \n\t"\
- "psllw $14, %%mm6 \n\t"\
- "paddsw %%mm6, %%mm2 \n\t"\
- "movq %%mm0, %%mm3 \n\t"\
- "pcmpgtw %%mm2, %%mm3 \n\t"\
- "pminsw %%mm2, %%mm0 \n\t"\
- "pand %%mm3, %%mm5 \n\t"\
- "pandn %%mm1, %%mm3 \n\t"\
- "por %%mm5, %%mm3 \n\t"\
- "movq %%mm3, %%mm1 \n\t"
+#ifdef HAVE_CONFIG_H
+# include "config.h"
+#endif
-static void yadif_filter_line_mmx2(struct vf_priv_s *p, uint8_t *dst, uint8_t *prev, uint8_t *cur, uint8_t *next, int w, int refs, int parity){
- static const uint64_t pw_1 = 0x0001000100010001ULL;
- static const uint64_t pb_1 = 0x0101010101010101ULL;
- const int mode = p->mode;
- uint64_t tmp0, tmp1, tmp2, tmp3;
- int x;
+#if defined(__GNUC__)
+# define DECLARE_ALIGNED(n,t,v) t __attribute__ ((aligned (n))) v
+# if VLC_GCC_VERSION(3,1)
+# define DECLARE_ASM_CONST(n,t,v) static const t __attribute__((used)) __attribute__ ((aligned (n))) v
+# else
+# define DECLARE_ASM_CONST(n,t,v) static const t __attribute__ ((aligned (n))) v
+# endif
+#endif
-#define FILTER\
- for(x=0; x<w; x+=4){\
- __asm__ volatile(\
- "pxor %%mm7, %%mm7 \n\t"\
- LOAD4("(%[cur],%[mrefs])", %%mm0) /* c = cur[x-refs] */\
- LOAD4("(%[cur],%[prefs])", %%mm1) /* e = cur[x+refs] */\
- LOAD4("(%["prev2"])", %%mm2) /* prev2[x] */\
- LOAD4("(%["next2"])", %%mm3) /* next2[x] */\
- "movq %%mm3, %%mm4 \n\t"\
- "paddw %%mm2, %%mm3 \n\t"\
- "psraw $1, %%mm3 \n\t" /* d = (prev2[x] + next2[x])>>1 */\
- "movq %%mm0, %[tmp0] \n\t" /* c */\
- "movq %%mm3, %[tmp1] \n\t" /* d */\
- "movq %%mm1, %[tmp2] \n\t" /* e */\
- "psubw %%mm4, %%mm2 \n\t"\
- PABS( %%mm4, %%mm2) /* temporal_diff0 */\
- LOAD4("(%[prev],%[mrefs])", %%mm3) /* prev[x-refs] */\
- LOAD4("(%[prev],%[prefs])", %%mm4) /* prev[x+refs] */\
- "psubw %%mm0, %%mm3 \n\t"\
- "psubw %%mm1, %%mm4 \n\t"\
- PABS( %%mm5, %%mm3)\
- PABS( %%mm5, %%mm4)\
- "paddw %%mm4, %%mm3 \n\t" /* temporal_diff1 */\
- "psrlw $1, %%mm2 \n\t"\
- "psrlw $1, %%mm3 \n\t"\
- "pmaxsw %%mm3, %%mm2 \n\t"\
- LOAD4("(%[next],%[mrefs])", %%mm3) /* next[x-refs] */\
- LOAD4("(%[next],%[prefs])", %%mm4) /* next[x+refs] */\
- "psubw %%mm0, %%mm3 \n\t"\
- "psubw %%mm1, %%mm4 \n\t"\
- PABS( %%mm5, %%mm3)\
- PABS( %%mm5, %%mm4)\
- "paddw %%mm4, %%mm3 \n\t" /* temporal_diff2 */\
- "psrlw $1, %%mm3 \n\t"\
- "pmaxsw %%mm3, %%mm2 \n\t"\
- "movq %%mm2, %[tmp3] \n\t" /* diff */\
-\
- "paddw %%mm0, %%mm1 \n\t"\
- "paddw %%mm0, %%mm0 \n\t"\
- "psubw %%mm1, %%mm0 \n\t"\
- "psrlw $1, %%mm1 \n\t" /* spatial_pred */\
- PABS( %%mm2, %%mm0) /* ABS(c-e) */\
-\
- "movq -1(%[cur],%[mrefs]), %%mm2 \n\t" /* cur[x-refs-1] */\
- "movq -1(%[cur],%[prefs]), %%mm3 \n\t" /* cur[x+refs-1] */\
- "movq %%mm2, %%mm4 \n\t"\
- "psubusb %%mm3, %%mm2 \n\t"\
- "psubusb %%mm4, %%mm3 \n\t"\
- "pmaxub %%mm3, %%mm2 \n\t"\
- "pshufw $9,%%mm2, %%mm3 \n\t"\
- "punpcklbw %%mm7, %%mm2 \n\t" /* ABS(cur[x-refs-1] - cur[x+refs-1]) */\
- "punpcklbw %%mm7, %%mm3 \n\t" /* ABS(cur[x-refs+1] - cur[x+refs+1]) */\
- "paddw %%mm2, %%mm0 \n\t"\
- "paddw %%mm3, %%mm0 \n\t"\
- "psubw %[pw1], %%mm0 \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 %[tmp3], %%mm6 \n\t" /* diff */\
- "cmpl $2, %[mode] \n\t"\
- "jge 1f \n\t"\
- LOAD4("(%["prev2"],%[mrefs],2)", %%mm2) /* prev2[x-2*refs] */\
- LOAD4("(%["next2"],%[mrefs],2)", %%mm4) /* next2[x-2*refs] */\
- LOAD4("(%["prev2"],%[prefs],2)", %%mm3) /* prev2[x+2*refs] */\
- LOAD4("(%["next2"],%[prefs],2)", %%mm5) /* next2[x+2*refs] */\
- "paddw %%mm4, %%mm2 \n\t"\
- "paddw %%mm5, %%mm3 \n\t"\
- "psrlw $1, %%mm2 \n\t" /* b */\
- "psrlw $1, %%mm3 \n\t" /* f */\
- "movq %[tmp0], %%mm4 \n\t" /* c */\
- "movq %[tmp1], %%mm5 \n\t" /* d */\
- "movq %[tmp2], %%mm7 \n\t" /* e */\
- "psubw %%mm4, %%mm2 \n\t" /* b-c */\
- "psubw %%mm7, %%mm3 \n\t" /* f-e */\
- "movq %%mm5, %%mm0 \n\t"\
- "psubw %%mm4, %%mm5 \n\t" /* d-c */\
- "psubw %%mm7, %%mm0 \n\t" /* d-e */\
- "movq %%mm2, %%mm4 \n\t"\
- "pminsw %%mm3, %%mm2 \n\t"\
- "pmaxsw %%mm4, %%mm3 \n\t"\
- "pmaxsw %%mm5, %%mm2 \n\t"\
- "pminsw %%mm5, %%mm3 \n\t"\
- "pmaxsw %%mm0, %%mm2 \n\t" /* max */\
- "pminsw %%mm0, %%mm3 \n\t" /* min */\
- "pxor %%mm4, %%mm4 \n\t"\
- "pmaxsw %%mm3, %%mm6 \n\t"\
- "psubw %%mm2, %%mm4 \n\t" /* -max */\
- "pmaxsw %%mm4, %%mm6 \n\t" /* diff= MAX3(diff, min, -max); */\
- "1: \n\t"\
-\
- "movq %[tmp1], %%mm2 \n\t" /* d */\
- "movq %%mm2, %%mm3 \n\t"\
- "psubw %%mm6, %%mm2 \n\t" /* d-diff */\
- "paddw %%mm6, %%mm3 \n\t" /* d+diff */\
- "pmaxsw %%mm2, %%mm1 \n\t"\
- "pminsw %%mm3, %%mm1 \n\t" /* d = clip(spatial_pred, d-diff, d+diff); */\
- "packuswb %%mm1, %%mm1 \n\t"\
-\
- :[tmp0]"=m"(tmp0),\
- [tmp1]"=m"(tmp1),\
- [tmp2]"=m"(tmp2),\
- [tmp3]"=m"(tmp3)\
- :[prev] "r"(prev),\
- [cur] "r"(cur),\
- [next] "r"(next),\
- [prefs]"r"((x86_reg)refs),\
- [mrefs]"r"((x86_reg)-refs),\
- [pw1] "m"(pw_1),\
- [pb1] "m"(pb_1),\
- [mode] "g"(mode)\
- );\
- __asm__ volatile("movd %%mm1, %0" :"=m"(*dst));\
- dst += 4;\
- prev+= 4;\
- cur += 4;\
- next+= 4;\
- }
+typedef intptr_t x86_reg;
+typedef struct { uint64_t a, b; } xmm_reg;
+
+DECLARE_ASM_CONST(16, const xmm_reg, pb_1) = {0x0101010101010101ULL, 0x0101010101010101ULL};
+DECLARE_ASM_CONST(16, const xmm_reg, pw_1) = {0x0001000100010001ULL, 0x0001000100010001ULL};
+
+
+#ifdef CAN_COMPILE_SSSE3
+#if defined(__SSE__) || VLC_GCC_VERSION(4, 4)
+// ================ SSSE3 =================
+#define HAVE_YADIF_SSSE3
+#define COMPILE_TEMPLATE_SSE 1
+#define COMPILE_TEMPLATE_SSSE3 1
+#define VLC_TARGET VLC_SSE
+#define RENAME(a) a ## _ssse3
+#include "yadif_template.h"
+#undef COMPILE_TEMPLATE_SSE
+#undef COMPILE_TEMPLATE_SSSE3
+#undef VLC_TARGET
+#undef RENAME
+#endif
+#endif
- 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 LOAD4
-#undef PABS
-#undef CHECK
-#undef CHECK1
-#undef CHECK2
-#undef FILTER
+#ifdef CAN_COMPILE_SSE2
+#if defined(__SSE__) || VLC_GCC_VERSION(4, 4)
+// ================= SSE2 =================
+#define HAVE_YADIF_SSE2
+#define COMPILE_TEMPLATE_SSE 1
+#define VLC_TARGET VLC_SSE
+#define RENAME(a) a ## _sse2
+#include "yadif_template.h"
+#undef COMPILE_TEMPLATE_SSE
+#undef VLC_TARGET
+#undef RENAME
+#endif
+#endif
+#ifdef CAN_COMPILE_MMX
+#if defined(__MMX__) || VLC_GCC_VERSION(4, 4)
+// ================ MMX =================
+#define HAVE_YADIF_MMX
+#define VLC_TARGET VLC_MMX
+#define RENAME(a) a ## _mmx
+#include "yadif_template.h"
+#undef VLC_TARGET
+#undef RENAME
+#endif
#endif
-static void yadif_filter_line_c(struct vf_priv_s *p, uint8_t *dst, uint8_t *prev, uint8_t *cur, uint8_t *next, int w, int refs, int parity){
+static void yadif_filter_line_c(uint8_t *dst, uint8_t *prev, uint8_t *cur, uint8_t *next, int w, int prefs, int mrefs, int parity, int mode) {
int x;
uint8_t *prev2= parity ? prev : cur ;
uint8_t *next2= parity ? cur : next;
for(x=0; x<w; x++){
- int c= cur[-refs];
+ int c= cur[mrefs];
int d= (prev2[0] + next2[0])>>1;
- int e= cur[+refs];
+ int e= cur[prefs];
int temporal_diff0= FFABS(prev2[0] - next2[0]);
- int temporal_diff1=( FFABS(prev[-refs] - c) + FFABS(prev[+refs] - e) )>>1;
- int temporal_diff2=( FFABS(next[-refs] - c) + FFABS(next[+refs] - e) )>>1;
+ int temporal_diff1=( FFABS(prev[mrefs] - c) + FFABS(prev[prefs] - e) )>>1;
+ int temporal_diff2=( FFABS(next[mrefs] - c) + FFABS(next[prefs] - e) )>>1;
int diff= FFMAX3(temporal_diff0>>1, temporal_diff1, temporal_diff2);
int spatial_pred= (c+e)>>1;
- int spatial_score= FFABS(cur[-refs-1] - cur[+refs-1]) + FFABS(c-e)
- + FFABS(cur[-refs+1] - cur[+refs+1]) - 1;
+ int spatial_score= FFABS(cur[mrefs-1] - cur[prefs-1]) + FFABS(c-e)
+ + FFABS(cur[mrefs+1] - cur[prefs+1]) - 1;
#define CHECK(j)\
- { int score= FFABS(cur[-refs-1+j] - cur[+refs-1-j])\
- + FFABS(cur[-refs +j] - cur[+refs -j])\
- + FFABS(cur[-refs+1+j] - cur[+refs+1-j]);\
+ { int score= FFABS(cur[mrefs-1+j] - cur[prefs-1-j])\
+ + FFABS(cur[mrefs +j] - cur[prefs -j])\
+ + FFABS(cur[mrefs+1+j] - cur[prefs+1-j]);\
if(score < spatial_score){\
spatial_score= score;\
- spatial_pred= (cur[-refs +j] + cur[+refs -j])>>1;\
+ spatial_pred= (cur[mrefs +j] + cur[prefs -j])>>1;\
CHECK(-1) CHECK(-2) }} }}
CHECK( 1) CHECK( 2) }} }}
- if(p->mode<2){
- int b= (prev2[-2*refs] + next2[-2*refs])>>1;
- int f= (prev2[+2*refs] + next2[+2*refs])>>1;
+ if(mode<2){
+ int b= (prev2[2*mrefs] + next2[2*mrefs])>>1;
+ int f= (prev2[2*prefs] + next2[2*prefs])>>1;
#if 0
- int a= cur[-3*refs];
- int g= cur[+3*refs];
+ int a= cur[3*mrefs];
+ int g= cur[3*prefs];
int max= FFMAX3(d-e, d-c, FFMIN3(FFMAX(b-c,f-e),FFMAX(b-c,b-a),FFMAX(f-g,f-e)) );
int min= FFMIN3(d-e, d-c, FFMAX3(FFMIN(b-c,f-e),FFMIN(b-c,b-a),FFMIN(f-g,f-e)) );
#else
--- /dev/null
+/*
+ * Copyright (C) 2006 Michael Niedermayer <michaelni@gmx.at>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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.
+ *
+ * FFmpeg 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 FFmpeg; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#ifdef COMPILE_TEMPLATE_SSE
+#define REGMM "xmm"
+#define MM "%%"REGMM
+#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 REGMM "mm"
+#define MM "%%"REGMM
+#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 %[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 %[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"
+
+VLC_TARGET 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, tmp0)[16];
+ DECLARE_ALIGNED(16, uint8_t, tmp1)[16];
+ DECLARE_ALIGNED(16, uint8_t, tmp2)[16];
+ DECLARE_ALIGNED(16, uint8_t, tmp3)[16];
+ 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, %[tmp0] \n\t" /* c */\
+ MOVQ" "MM"3, %[tmp1] \n\t" /* d */\
+ MOVQ" "MM"1, %[tmp2] \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, %[tmp3] \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 %[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" %[tmp3], "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" %[tmp0], "MM"4 \n\t" /* c */\
+ MOVQ" %[tmp1], "MM"5 \n\t" /* d */\
+ MOVQ" %[tmp2], "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" %[tmp1], "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"\
+\
+ :[tmp0]"=m"(tmp0),\
+ [tmp1]"=m"(tmp1),\
+ [tmp2]"=m"(tmp2),\
+ [tmp3]"=m"(tmp3)\
+ :[prev] "r"(prev),\
+ [cur] "r"(cur),\
+ [next] "r"(next),\
+ [prefs]"r"((x86_reg)prefs),\
+ [mrefs]"r"((x86_reg)mrefs),\
+ [pw_1] "m"(pw_1),\
+ [pb_1] "m"(pb_1),\
+ [mode] "g"(mode)\
+ :REGMM"0",REGMM"1",REGMM"2",REGMM"3",REGMM"4",REGMM"5",REGMM"6",REGMM"7"\
+ );\
+ __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 REGMM
+#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
+