History

yadif sse2/ssse3 optimizations

Originally committed as revision 25874 to svn://svn.ffmpeg.org/ffmpeg/trunk

Baptiste Coudurier committed 15 years ago

Commit 1ef64491 parent 23b83423 files · +294 −212Browse files

Unified · Split

3 files
  1. libavfilter/vf_yadif.c +5 −1
  2. libavfilter/x86/yadif.c +21 −211
  3. libavfilter/x86/yadif_template.c +268 −0
libavfilter/vf_yadif.c+5−1
@@ -304,7 +304,11 @@ static av_cold int init(AVFilterContext *ctx, const char *args, void *opaque) · 303 unmodified lines
304304 if (args) sscanf(args, "%d:%d", &yadif->mode, &yadif->parity);
305305
306306 yadif->filter_line = filter_line_c;
307− if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX)
307+ if (HAVE_SSSE3 && cpu_flags & AV_CPU_FLAG_SSSE3)
308+ yadif->filter_line = ff_yadif_filter_line_ssse3;
309+ else if (HAVE_SSE && cpu_flags & AV_CPU_FLAG_SSE2)
310+ yadif->filter_line = ff_yadif_filter_line_sse2;
311+ else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX)
308312 yadif->filter_line = ff_yadif_filter_line_mmx;
309313
310314 av_log(ctx, AV_LOG_INFO, "mode:%d parity:%d\n", yadif->mode, yadif->parity);
libavfilter/x86/yadif.c+21−211
@@ -22,214 +22,24 @@ · 21 unmodified lines
2222 #include "libavutil/x86_cpu.h"
2323 #include "libavfilter/yadif.h"
2424
25−#define LOAD4(mem,dst) \
26− "movd "mem", "#dst" \n\t"\
27− "punpcklbw %%mm7, "#dst" \n\t"
28−
29−#define PABS(tmp,dst) \
30− "pxor "#tmp", "#tmp" \n\t"\
31− "psubw "#dst", "#tmp" \n\t"\
32− "pmaxsw "#tmp", "#dst" \n\t"
33−
34−#define CHECK(pj,mj) \
35− "movq "#pj"(%[cur],%[mrefs]), %%mm2 \n\t" /* cur[x-refs-1+j] */\
36− "movq "#mj"(%[cur],%[prefs]), %%mm3 \n\t" /* cur[x+refs-1-j] */\
37− "movq %%mm2, %%mm4 \n\t"\
38− "movq %%mm2, %%mm5 \n\t"\
39− "pxor %%mm3, %%mm4 \n\t"\
40− "pavgb %%mm3, %%mm5 \n\t"\
41− "pand "MANGLE(pb_1)", %%mm4 \n\t"\
42− "psubusb %%mm4, %%mm5 \n\t"\
43− "psrlq $8, %%mm5 \n\t"\
44− "punpcklbw %%mm7, %%mm5 \n\t" /* (cur[x-refs+j] + cur[x+refs-j])>>1 */\
45− "movq %%mm2, %%mm4 \n\t"\
46− "psubusb %%mm3, %%mm2 \n\t"\
47− "psubusb %%mm4, %%mm3 \n\t"\
48− "pmaxub %%mm3, %%mm2 \n\t"\
49− "movq %%mm2, %%mm3 \n\t"\
50− "movq %%mm2, %%mm4 \n\t" /* ABS(cur[x-refs-1+j] - cur[x+refs-1-j]) */\
51− "psrlq $8, %%mm3 \n\t" /* ABS(cur[x-refs +j] - cur[x+refs -j]) */\
52− "psrlq $16, %%mm4 \n\t" /* ABS(cur[x-refs+1+j] - cur[x+refs+1-j]) */\
53− "punpcklbw %%mm7, %%mm2 \n\t"\
54− "punpcklbw %%mm7, %%mm3 \n\t"\
55− "punpcklbw %%mm7, %%mm4 \n\t"\
56− "paddw %%mm3, %%mm2 \n\t"\
57− "paddw %%mm4, %%mm2 \n\t" /* score */
58−
59−#define CHECK1 \
60− "movq %%mm0, %%mm3 \n\t"\
61− "pcmpgtw %%mm2, %%mm3 \n\t" /* if(score < spatial_score) */\
62− "pminsw %%mm2, %%mm0 \n\t" /* spatial_score= score; */\
63− "movq %%mm3, %%mm6 \n\t"\
64− "pand %%mm3, %%mm5 \n\t"\
65− "pandn %%mm1, %%mm3 \n\t"\
66− "por %%mm5, %%mm3 \n\t"\
67− "movq %%mm3, %%mm1 \n\t" /* spatial_pred= (cur[x-refs+j] + cur[x+refs-j])>>1; */
68−
69−#define CHECK2 /* pretend not to have checked dir=2 if dir=1 was bad.\
70− hurts both quality and speed, but matches the C version. */\
71− "paddw "MANGLE(pw_1)", %%mm6 \n\t"\
72− "psllw $14, %%mm6 \n\t"\
73− "paddsw %%mm6, %%mm2 \n\t"\
74− "movq %%mm0, %%mm3 \n\t"\
75− "pcmpgtw %%mm2, %%mm3 \n\t"\
76− "pminsw %%mm2, %%mm0 \n\t"\
77− "pand %%mm3, %%mm5 \n\t"\
78− "pandn %%mm1, %%mm3 \n\t"\
79− "por %%mm5, %%mm3 \n\t"\
80− "movq %%mm3, %%mm1 \n\t"
81−
82−DECLARE_ASM_CONST(16, uint64_t, pw_1) = 0x0001000100010001ULL;
83−DECLARE_ASM_CONST(16, uint64_t, pb_1) = 0x0101010101010101ULL;
84−
85−void ff_yadif_filter_line_mmx(uint8_t *dst,
86− uint8_t *prev, uint8_t *cur, uint8_t *next,
87− int w, int refs, int parity, int mode)
88−{
89− uint64_t tmp0, tmp1, tmp2, tmp3;
90− int x;
91−
92−#define FILTER\
93− for(x=0; x<w; x+=4){\
94− __asm__ volatile(\
95− "pxor %%mm7, %%mm7 \n\t"\
96− LOAD4("(%[cur],%[mrefs])", %%mm0) /* c = cur[x-refs] */\
97− LOAD4("(%[cur],%[prefs])", %%mm1) /* e = cur[x+refs] */\
98− LOAD4("(%["prev2"])", %%mm2) /* prev2[x] */\
99− LOAD4("(%["next2"])", %%mm3) /* next2[x] */\
100− "movq %%mm3, %%mm4 \n\t"\
101− "paddw %%mm2, %%mm3 \n\t"\
102− "psraw $1, %%mm3 \n\t" /* d = (prev2[x] + next2[x])>>1 */\
103− "movq %%mm0, %[tmp0] \n\t" /* c */\
104− "movq %%mm3, %[tmp1] \n\t" /* d */\
105− "movq %%mm1, %[tmp2] \n\t" /* e */\
106− "psubw %%mm4, %%mm2 \n\t"\
107− PABS( %%mm4, %%mm2) /* temporal_diff0 */\
108− LOAD4("(%[prev],%[mrefs])", %%mm3) /* prev[x-refs] */\
109− LOAD4("(%[prev],%[prefs])", %%mm4) /* prev[x+refs] */\
110− "psubw %%mm0, %%mm3 \n\t"\
111− "psubw %%mm1, %%mm4 \n\t"\
112− PABS( %%mm5, %%mm3)\
113− PABS( %%mm5, %%mm4)\
114− "paddw %%mm4, %%mm3 \n\t" /* temporal_diff1 */\
115− "psrlw $1, %%mm2 \n\t"\
116− "psrlw $1, %%mm3 \n\t"\
117− "pmaxsw %%mm3, %%mm2 \n\t"\
118− LOAD4("(%[next],%[mrefs])", %%mm3) /* next[x-refs] */\
119− LOAD4("(%[next],%[prefs])", %%mm4) /* next[x+refs] */\
120− "psubw %%mm0, %%mm3 \n\t"\
121− "psubw %%mm1, %%mm4 \n\t"\
122− PABS( %%mm5, %%mm3)\
123− PABS( %%mm5, %%mm4)\
124− "paddw %%mm4, %%mm3 \n\t" /* temporal_diff2 */\
125− "psrlw $1, %%mm3 \n\t"\
126− "pmaxsw %%mm3, %%mm2 \n\t"\
127− "movq %%mm2, %[tmp3] \n\t" /* diff */\
128−\
129− "paddw %%mm0, %%mm1 \n\t"\
130− "paddw %%mm0, %%mm0 \n\t"\
131− "psubw %%mm1, %%mm0 \n\t"\
132− "psrlw $1, %%mm1 \n\t" /* spatial_pred */\
133− PABS( %%mm2, %%mm0) /* ABS(c-e) */\
134−\
135− "movq -1(%[cur],%[mrefs]), %%mm2 \n\t" /* cur[x-refs-1] */\
136− "movq -1(%[cur],%[prefs]), %%mm3 \n\t" /* cur[x+refs-1] */\
137− "movq %%mm2, %%mm4 \n\t"\
138− "psubusb %%mm3, %%mm2 \n\t"\
139− "psubusb %%mm4, %%mm3 \n\t"\
140− "pmaxub %%mm3, %%mm2 \n\t"\
141− "pshufw $9,%%mm2, %%mm3 \n\t"\
142− "punpcklbw %%mm7, %%mm2 \n\t" /* ABS(cur[x-refs-1] - cur[x+refs-1]) */\
143− "punpcklbw %%mm7, %%mm3 \n\t" /* ABS(cur[x-refs+1] - cur[x+refs+1]) */\
144− "paddw %%mm2, %%mm0 \n\t"\
145− "paddw %%mm3, %%mm0 \n\t"\
146− "psubw "MANGLE(pw_1)", %%mm0 \n\t" /* spatial_score */\
147−\
148− CHECK(-2,0)\
149− CHECK1\
150− CHECK(-3,1)\
151− CHECK2\
152− CHECK(0,-2)\
153− CHECK1\
154− CHECK(1,-3)\
155− CHECK2\
156−\
157− /* if(p->mode<2) ... */\
158− "movq %[tmp3], %%mm6 \n\t" /* diff */\
159− "cmpl $2, %[mode] \n\t"\
160− "jge 1f \n\t"\
161− LOAD4("(%["prev2"],%[mrefs],2)", %%mm2) /* prev2[x-2*refs] */\
162− LOAD4("(%["next2"],%[mrefs],2)", %%mm4) /* next2[x-2*refs] */\
163− LOAD4("(%["prev2"],%[prefs],2)", %%mm3) /* prev2[x+2*refs] */\
164− LOAD4("(%["next2"],%[prefs],2)", %%mm5) /* next2[x+2*refs] */\
165− "paddw %%mm4, %%mm2 \n\t"\
166− "paddw %%mm5, %%mm3 \n\t"\
167− "psrlw $1, %%mm2 \n\t" /* b */\
168− "psrlw $1, %%mm3 \n\t" /* f */\
169− "movq %[tmp0], %%mm4 \n\t" /* c */\
170− "movq %[tmp1], %%mm5 \n\t" /* d */\
171− "movq %[tmp2], %%mm7 \n\t" /* e */\
172− "psubw %%mm4, %%mm2 \n\t" /* b-c */\
173− "psubw %%mm7, %%mm3 \n\t" /* f-e */\
174− "movq %%mm5, %%mm0 \n\t"\
175− "psubw %%mm4, %%mm5 \n\t" /* d-c */\
176− "psubw %%mm7, %%mm0 \n\t" /* d-e */\
177− "movq %%mm2, %%mm4 \n\t"\
178− "pminsw %%mm3, %%mm2 \n\t"\
179− "pmaxsw %%mm4, %%mm3 \n\t"\
180− "pmaxsw %%mm5, %%mm2 \n\t"\
181− "pminsw %%mm5, %%mm3 \n\t"\
182− "pmaxsw %%mm0, %%mm2 \n\t" /* max */\
183− "pminsw %%mm0, %%mm3 \n\t" /* min */\
184− "pxor %%mm4, %%mm4 \n\t"\
185− "pmaxsw %%mm3, %%mm6 \n\t"\
186− "psubw %%mm2, %%mm4 \n\t" /* -max */\
187− "pmaxsw %%mm4, %%mm6 \n\t" /* diff= MAX3(diff, min, -max); */\
188− "1: \n\t"\
189−\
190− "movq %[tmp1], %%mm2 \n\t" /* d */\
191− "movq %%mm2, %%mm3 \n\t"\
192− "psubw %%mm6, %%mm2 \n\t" /* d-diff */\
193− "paddw %%mm6, %%mm3 \n\t" /* d+diff */\
194− "pmaxsw %%mm2, %%mm1 \n\t"\
195− "pminsw %%mm3, %%mm1 \n\t" /* d = clip(spatial_pred, d-diff, d+diff); */\
196− "packuswb %%mm1, %%mm1 \n\t"\
197−\
198− :[tmp0]"=m"(tmp0),\
199− [tmp1]"=m"(tmp1),\
200− [tmp2]"=m"(tmp2),\
201− [tmp3]"=m"(tmp3)\
202− :[prev] "r"(prev),\
203− [cur] "r"(cur),\
204− [next] "r"(next),\
205− [prefs]"r"((x86_reg)refs),\
206− [mrefs]"r"((x86_reg)-refs),\
207− [mode] "g"(mode)\
208− );\
209− __asm__ volatile("movd %%mm1, %0" :"=m"(*dst));\
210− dst += 4;\
211− prev+= 4;\
212− cur += 4;\
213− next+= 4;\
214− }
215−
216− if (parity) {
217−#define prev2 "prev"
218−#define next2 "cur"
219− FILTER
220−#undef prev2
221−#undef next2
222− } else {
223−#define prev2 "cur"
224−#define next2 "next"
225− FILTER
226−#undef prev2
227−#undef next2
228− }
229−}
230−#undef LOAD4
231−#undef PABS
232−#undef CHECK
233−#undef CHECK1
234−#undef CHECK2
235−#undef FILTER
25+#if HAVE_SSSE3
26+#define COMPILE_TEMPLATE_SSE 1
27+#define COMPILE_TEMPLATE_SSSE3 1
28+#undef RENAME
29+#define RENAME(a) a ## _ssse3
30+#include "yadif_template.c"
31+#undef COMPILE_TEMPLATE_SSSE3
32+#endif
33+
34+#if HAVE_SSE
35+#undef RENAME
36+#define RENAME(a) a ## _sse2
37+#include "yadif_template.c"
38+#undef COMPILE_TEMPLATE_SSE
39+#endif
40+
41+#if HAVE_MMX
42+#undef RENAME
43+#define RENAME(a) a ## _mmx
44+#include "yadif_template.c"
45+#endif
libavfilter/x86/yadif_template.c+268−0
@@ -0,0 +1,268 @@
1+/*
2+ * Copyright (C) 2006 Michael Niedermayer <michaelni@gmx.at>
3+ *
4+ * This file is part of FFmpeg.
5+ *
6+ * FFmpeg is free software; you can redistribute it and/or modify
7+ * it under the terms of the GNU General Public License as published by
8+ * the Free Software Foundation; either version 2 of the License, or
9+ * (at your option) any later version.
10+ *
11+ * FFmpeg is distributed in the hope that it will be useful,
12+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
13+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14+ * GNU General Public License for more details.
15+ *
16+ * You should have received a copy of the GNU General Public License along
17+ * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
18+ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
19+ */
20+
21+#ifdef COMPILE_TEMPLATE_SSE
22+#define MM "%%xmm"
23+#define MOV "movq"
24+#define MOVQ "movdqa"
25+#define MOVQU "movdqu"
26+#define STEP 8
27+#define LOAD(mem,dst) \
28+ MOV" "mem", "dst" \n\t"\
29+ "punpcklbw "MM"7, "dst" \n\t"
30+#define PSRL1(reg) "psrldq $1, "reg" \n\t"
31+#define PSRL2(reg) "psrldq $2, "reg" \n\t"
32+#define PSHUF(src,dst) "movdqa "dst", "src" \n\t"\
33+ "psrldq $2, "src" \n\t"
34+#else
35+#define MM "%%mm"
36+#define MOV "movd"
37+#define MOVQ "movq"
38+#define MOVQU "movq"
39+#define STEP 4
40+#define LOAD(mem,dst) \
41+ MOV" "mem", "dst" \n\t"\
42+ "punpcklbw "MM"7, "dst" \n\t"
43+#define PSRL1(reg) "psrlq $8, "reg" \n\t"
44+#define PSRL2(reg) "psrlq $16, "reg" \n\t"
45+#define PSHUF(src,dst) "pshufw $9, "dst", "src" \n\t"
46+#endif
47+
48+#ifdef COMPILE_TEMPLATE_SSSE3
49+#define PABS(tmp,dst) \
50+ "pabsw "dst", "dst" \n\t"
51+#else
52+#define PABS(tmp,dst) \
53+ "pxor "tmp", "tmp" \n\t"\
54+ "psubw "dst", "tmp" \n\t"\
55+ "pmaxsw "tmp", "dst" \n\t"
56+#endif
57+
58+#define CHECK(pj,mj) \
59+ MOVQU" "#pj"(%[cur],%[mrefs]), "MM"2 \n\t" /* cur[x-refs-1+j] */\
60+ MOVQU" "#mj"(%[cur],%[prefs]), "MM"3 \n\t" /* cur[x+refs-1-j] */\
61+ MOVQ" "MM"2, "MM"4 \n\t"\
62+ MOVQ" "MM"2, "MM"5 \n\t"\
63+ "pxor "MM"3, "MM"4 \n\t"\
64+ "pavgb "MM"3, "MM"5 \n\t"\
65+ "pand "MANGLE(ff_pb_1)", "MM"4 \n\t"\
66+ "psubusb "MM"4, "MM"5 \n\t"\
67+ PSRL1(MM"5") \
68+ "punpcklbw "MM"7, "MM"5 \n\t" /* (cur[x-refs+j] + cur[x+refs-j])>>1 */\
69+ MOVQ" "MM"2, "MM"4 \n\t"\
70+ "psubusb "MM"3, "MM"2 \n\t"\
71+ "psubusb "MM"4, "MM"3 \n\t"\
72+ "pmaxub "MM"3, "MM"2 \n\t"\
73+ MOVQ" "MM"2, "MM"3 \n\t"\
74+ MOVQ" "MM"2, "MM"4 \n\t" /* ABS(cur[x-refs-1+j] - cur[x+refs-1-j]) */\
75+ PSRL1(MM"3") /* ABS(cur[x-refs +j] - cur[x+refs -j]) */\
76+ PSRL2(MM"4") /* ABS(cur[x-refs+1+j] - cur[x+refs+1-j]) */\
77+ "punpcklbw "MM"7, "MM"2 \n\t"\
78+ "punpcklbw "MM"7, "MM"3 \n\t"\
79+ "punpcklbw "MM"7, "MM"4 \n\t"\
80+ "paddw "MM"3, "MM"2 \n\t"\
81+ "paddw "MM"4, "MM"2 \n\t" /* score */
82+
83+#define CHECK1 \
84+ MOVQ" "MM"0, "MM"3 \n\t"\
85+ "pcmpgtw "MM"2, "MM"3 \n\t" /* if(score < spatial_score) */\
86+ "pminsw "MM"2, "MM"0 \n\t" /* spatial_score= score; */\
87+ MOVQ" "MM"3, "MM"6 \n\t"\
88+ "pand "MM"3, "MM"5 \n\t"\
89+ "pandn "MM"1, "MM"3 \n\t"\
90+ "por "MM"5, "MM"3 \n\t"\
91+ MOVQ" "MM"3, "MM"1 \n\t" /* spatial_pred= (cur[x-refs+j] + cur[x+refs-j])>>1; */
92+
93+#define CHECK2 /* pretend not to have checked dir=2 if dir=1 was bad.\
94+ hurts both quality and speed, but matches the C version. */\
95+ "paddw "MANGLE(ff_pw_1)", "MM"6 \n\t"\
96+ "psllw $14, "MM"6 \n\t"\
97+ "paddsw "MM"6, "MM"2 \n\t"\
98+ MOVQ" "MM"0, "MM"3 \n\t"\
99+ "pcmpgtw "MM"2, "MM"3 \n\t"\
100+ "pminsw "MM"2, "MM"0 \n\t"\
101+ "pand "MM"3, "MM"5 \n\t"\
102+ "pandn "MM"1, "MM"3 \n\t"\
103+ "por "MM"5, "MM"3 \n\t"\
104+ MOVQ" "MM"3, "MM"1 \n\t"
105+
106+void RENAME(ff_yadif_filter_line)(uint8_t *dst,
107+ uint8_t *prev, uint8_t *cur, uint8_t *next,
108+ int w, int refs, int parity, int mode)
109+{
110+ DECLARE_ALIGNED(16, uint8_t, tmp0[16]);
111+ DECLARE_ALIGNED(16, uint8_t, tmp1[16]);
112+ DECLARE_ALIGNED(16, uint8_t, tmp2[16]);
113+ DECLARE_ALIGNED(16, uint8_t, tmp3[16]);
114+ int x;
115+
116+#define FILTER\
117+ for(x=0; x<w; x+=STEP){\
118+ __asm__ volatile(\
119+ "pxor "MM"7, "MM"7 \n\t"\
120+ LOAD("(%[cur],%[mrefs])", MM"0") /* c = cur[x-refs] */\
121+ LOAD("(%[cur],%[prefs])", MM"1") /* e = cur[x+refs] */\
122+ LOAD("(%["prev2"])", MM"2") /* prev2[x] */\
123+ LOAD("(%["next2"])", MM"3") /* next2[x] */\
124+ MOVQ" "MM"3, "MM"4 \n\t"\
125+ "paddw "MM"2, "MM"3 \n\t"\
126+ "psraw $1, "MM"3 \n\t" /* d = (prev2[x] + next2[x])>>1 */\
127+ MOVQ" "MM"0, %[tmp0] \n\t" /* c */\
128+ MOVQ" "MM"3, %[tmp1] \n\t" /* d */\
129+ MOVQ" "MM"1, %[tmp2] \n\t" /* e */\
130+ "psubw "MM"4, "MM"2 \n\t"\
131+ PABS( MM"4", MM"2") /* temporal_diff0 */\
132+ LOAD("(%[prev],%[mrefs])", MM"3") /* prev[x-refs] */\
133+ LOAD("(%[prev],%[prefs])", MM"4") /* prev[x+refs] */\
134+ "psubw "MM"0, "MM"3 \n\t"\
135+ "psubw "MM"1, "MM"4 \n\t"\
136+ PABS( MM"5", MM"3")\
137+ PABS( MM"5", MM"4")\
138+ "paddw "MM"4, "MM"3 \n\t" /* temporal_diff1 */\
139+ "psrlw $1, "MM"2 \n\t"\
140+ "psrlw $1, "MM"3 \n\t"\
141+ "pmaxsw "MM"3, "MM"2 \n\t"\
142+ LOAD("(%[next],%[mrefs])", MM"3") /* next[x-refs] */\
143+ LOAD("(%[next],%[prefs])", MM"4") /* next[x+refs] */\
144+ "psubw "MM"0, "MM"3 \n\t"\
145+ "psubw "MM"1, "MM"4 \n\t"\
146+ PABS( MM"5", MM"3")\
147+ PABS( MM"5", MM"4")\
148+ "paddw "MM"4, "MM"3 \n\t" /* temporal_diff2 */\
149+ "psrlw $1, "MM"3 \n\t"\
150+ "pmaxsw "MM"3, "MM"2 \n\t"\
151+ MOVQ" "MM"2, %[tmp3] \n\t" /* diff */\
152+\
153+ "paddw "MM"0, "MM"1 \n\t"\
154+ "paddw "MM"0, "MM"0 \n\t"\
155+ "psubw "MM"1, "MM"0 \n\t"\
156+ "psrlw $1, "MM"1 \n\t" /* spatial_pred */\
157+ PABS( MM"2", MM"0") /* ABS(c-e) */\
158+\
159+ MOVQU" -1(%[cur],%[mrefs]), "MM"2 \n\t" /* cur[x-refs-1] */\
160+ MOVQU" -1(%[cur],%[prefs]), "MM"3 \n\t" /* cur[x+refs-1] */\
161+ MOVQ" "MM"2, "MM"4 \n\t"\
162+ "psubusb "MM"3, "MM"2 \n\t"\
163+ "psubusb "MM"4, "MM"3 \n\t"\
164+ "pmaxub "MM"3, "MM"2 \n\t"\
165+ PSHUF(MM"3", MM"2") \
166+ "punpcklbw "MM"7, "MM"2 \n\t" /* ABS(cur[x-refs-1] - cur[x+refs-1]) */\
167+ "punpcklbw "MM"7, "MM"3 \n\t" /* ABS(cur[x-refs+1] - cur[x+refs+1]) */\
168+ "paddw "MM"2, "MM"0 \n\t"\
169+ "paddw "MM"3, "MM"0 \n\t"\
170+ "psubw "MANGLE(ff_pw_1)", "MM"0 \n\t" /* spatial_score */\
171+\
172+ CHECK(-2,0)\
173+ CHECK1\
174+ CHECK(-3,1)\
175+ CHECK2\
176+ CHECK(0,-2)\
177+ CHECK1\
178+ CHECK(1,-3)\
179+ CHECK2\
180+\
181+ /* if(p->mode<2) ... */\
182+ MOVQ" %[tmp3], "MM"6 \n\t" /* diff */\
183+ "cmpl $2, %[mode] \n\t"\
184+ "jge 1f \n\t"\
185+ LOAD("(%["prev2"],%[mrefs],2)", MM"2") /* prev2[x-2*refs] */\
186+ LOAD("(%["next2"],%[mrefs],2)", MM"4") /* next2[x-2*refs] */\
187+ LOAD("(%["prev2"],%[prefs],2)", MM"3") /* prev2[x+2*refs] */\
188+ LOAD("(%["next2"],%[prefs],2)", MM"5") /* next2[x+2*refs] */\
189+ "paddw "MM"4, "MM"2 \n\t"\
190+ "paddw "MM"5, "MM"3 \n\t"\
191+ "psrlw $1, "MM"2 \n\t" /* b */\
192+ "psrlw $1, "MM"3 \n\t" /* f */\
193+ MOVQ" %[tmp0], "MM"4 \n\t" /* c */\
194+ MOVQ" %[tmp1], "MM"5 \n\t" /* d */\
195+ MOVQ" %[tmp2], "MM"7 \n\t" /* e */\
196+ "psubw "MM"4, "MM"2 \n\t" /* b-c */\
197+ "psubw "MM"7, "MM"3 \n\t" /* f-e */\
198+ MOVQ" "MM"5, "MM"0 \n\t"\
199+ "psubw "MM"4, "MM"5 \n\t" /* d-c */\
200+ "psubw "MM"7, "MM"0 \n\t" /* d-e */\
201+ MOVQ" "MM"2, "MM"4 \n\t"\
202+ "pminsw "MM"3, "MM"2 \n\t"\
203+ "pmaxsw "MM"4, "MM"3 \n\t"\
204+ "pmaxsw "MM"5, "MM"2 \n\t"\
205+ "pminsw "MM"5, "MM"3 \n\t"\
206+ "pmaxsw "MM"0, "MM"2 \n\t" /* max */\
207+ "pminsw "MM"0, "MM"3 \n\t" /* min */\
208+ "pxor "MM"4, "MM"4 \n\t"\
209+ "pmaxsw "MM"3, "MM"6 \n\t"\
210+ "psubw "MM"2, "MM"4 \n\t" /* -max */\
211+ "pmaxsw "MM"4, "MM"6 \n\t" /* diff= MAX3(diff, min, -max); */\
212+ "1: \n\t"\
213+\
214+ MOVQ" %[tmp1], "MM"2 \n\t" /* d */\
215+ MOVQ" "MM"2, "MM"3 \n\t"\
216+ "psubw "MM"6, "MM"2 \n\t" /* d-diff */\
217+ "paddw "MM"6, "MM"3 \n\t" /* d+diff */\
218+ "pmaxsw "MM"2, "MM"1 \n\t"\
219+ "pminsw "MM"3, "MM"1 \n\t" /* d = clip(spatial_pred, d-diff, d+diff); */\
220+ "packuswb "MM"1, "MM"1 \n\t"\
221+\
222+ :[tmp0]"=m"(tmp0),\
223+ [tmp1]"=m"(tmp1),\
224+ [tmp2]"=m"(tmp2),\
225+ [tmp3]"=m"(tmp3)\
226+ :[prev] "r"(prev),\
227+ [cur] "r"(cur),\
228+ [next] "r"(next),\
229+ [prefs]"r"((x86_reg)refs),\
230+ [mrefs]"r"((x86_reg)-refs),\
231+ [mode] "g"(mode)\
232+ );\
233+ __asm__ volatile(MOV" "MM"1, %0" :"=m"(*dst));\
234+ dst += STEP;\
235+ prev+= STEP;\
236+ cur += STEP;\
237+ next+= STEP;\
238+ }
239+
240+ if (parity) {
241+#define prev2 "prev"
242+#define next2 "cur"
243+ FILTER
244+#undef prev2
245+#undef next2
246+ } else {
247+#define prev2 "cur"
248+#define next2 "next"
249+ FILTER
250+#undef prev2
251+#undef next2
252+ }
253+}
254+#undef STEP
255+#undef MM
256+#undef MOV
257+#undef MOVQ
258+#undef MOVQU
259+#undef PSHUF
260+#undef PSRL1
261+#undef PSRL2
262+#undef LOAD
263+#undef PABS
264+#undef CHECK
265+#undef CHECK1
266+#undef CHECK2
267+#undef FILTER
268+

Keyboard shortcuts

General

Show keyboard shortcuts
?
Focus the current search
/

Issues and pull requests

Open filters
f

Issues, pull requests and Inbox

Select next item
j↓
Select previous item
k↑
Open selected item
Enter
Dismiss filters or clear selection
Esc

Repository code

Go to file
t