2 * Copyright (C) 2009 Loren Merritt <lorenm@u.washignton.edu>
4 * This file is part of Libav.
6 * Libav is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * Libav 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 GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with Libav; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "libavutil/attributes.h"
22 #include "libavutil/cpu.h"
23 #include "libavutil/mem.h"
24 #include "libavutil/x86/asm.h"
25 #include "libavfilter/gradfun.h"
29 DECLARE_ALIGNED(16, static const uint16_t, pw_7f)[8] = {0x7F,0x7F,0x7F,0x7F,0x7F,0x7F,0x7F,0x7F};
30 DECLARE_ALIGNED(16, static const uint16_t, pw_ff)[8] = {0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF};
32 #if HAVE_MMXEXT_INLINE
33 static void gradfun_filter_line_mmxext(uint8_t *dst, uint8_t *src, uint16_t *dc,
34 int width, int thresh,
35 const uint16_t *dithers)
40 ff_gradfun_filter_line_c(dst + x, src + x, dc + x / 2, width - x, thresh, dithers);
46 "pxor %%mm7, %%mm7 \n"
47 "pshufw $0, %%mm5, %%mm5 \n"
51 "movd (%2,%0), %%mm0 \n"
52 "movd (%3,%0), %%mm1 \n"
53 "punpcklbw %%mm7, %%mm0 \n"
54 "punpcklwd %%mm1, %%mm1 \n"
56 "pxor %%mm2, %%mm2 \n"
57 "psubw %%mm0, %%mm1 \n" // delta = dc - pix
58 "psubw %%mm1, %%mm2 \n"
59 "pmaxsw %%mm1, %%mm2 \n"
60 "pmulhuw %%mm5, %%mm2 \n" // m = abs(delta) * thresh >> 16
61 "psubw %%mm6, %%mm2 \n"
62 "pminsw %%mm7, %%mm2 \n" // m = -max(0, 127-m)
63 "pmullw %%mm2, %%mm2 \n"
64 "paddw %%mm4, %%mm0 \n" // pix += dither
65 "pmulhw %%mm2, %%mm1 \n"
66 "psllw $2, %%mm1 \n" // m = m*m*delta >> 14
67 "paddw %%mm1, %%mm0 \n" // pix += m
69 "packuswb %%mm0, %%mm0 \n"
70 "movd %%mm0, (%1,%0) \n" // dst = clip(pix>>7)
75 :"r"(dst+width), "r"(src+width), "r"(dc+width/2),
76 "rm"(thresh), "m"(*dithers), "m"(*pw_7f)
83 static void gradfun_filter_line_ssse3(uint8_t *dst, uint8_t *src, uint16_t *dc, int width, int thresh, const uint16_t *dithers)
87 // could be 10% faster if I somehow eliminated this
89 ff_gradfun_filter_line_c(dst + x, src + x, dc + x / 2, width - x, thresh, dithers);
95 "pxor %%xmm7, %%xmm7 \n"
96 "pshuflw $0,%%xmm5, %%xmm5 \n"
97 "movdqa %6, %%xmm6 \n"
98 "punpcklqdq %%xmm5, %%xmm5 \n"
99 "movdqa %5, %%xmm4 \n"
101 "movq (%2,%0), %%xmm0 \n"
102 "movq (%3,%0), %%xmm1 \n"
103 "punpcklbw %%xmm7, %%xmm0 \n"
104 "punpcklwd %%xmm1, %%xmm1 \n"
105 "psllw $7, %%xmm0 \n"
106 "psubw %%xmm0, %%xmm1 \n" // delta = dc - pix
107 "pabsw %%xmm1, %%xmm2 \n"
108 "pmulhuw %%xmm5, %%xmm2 \n" // m = abs(delta) * thresh >> 16
109 "psubw %%xmm6, %%xmm2 \n"
110 "pminsw %%xmm7, %%xmm2 \n" // m = -max(0, 127-m)
111 "pmullw %%xmm2, %%xmm2 \n"
112 "psllw $1, %%xmm2 \n"
113 "paddw %%xmm4, %%xmm0 \n" // pix += dither
114 "pmulhrsw %%xmm2, %%xmm1 \n" // m = m*m*delta >> 14
115 "paddw %%xmm1, %%xmm0 \n" // pix += m
116 "psraw $7, %%xmm0 \n"
117 "packuswb %%xmm0, %%xmm0 \n"
118 "movq %%xmm0, (%1,%0) \n" // dst = clip(pix>>7)
122 :"r"(dst+width), "r"(src+width), "r"(dc+width/2),
123 "rm"(thresh), "m"(*dithers), "m"(*pw_7f)
127 #endif /* HAVE_SSSE3_INLINE */
130 static void gradfun_blur_line_sse2(uint16_t *dc, uint16_t *buf, uint16_t *buf1, uint8_t *src, int src_linesize, int width)
133 intptr_t x = -2*width;\
135 "movdqa %6, %%xmm7 \n"\
137 load" (%4,%0), %%xmm0 \n"\
138 load" (%5,%0), %%xmm1 \n"\
139 "movdqa %%xmm0, %%xmm2 \n"\
140 "movdqa %%xmm1, %%xmm3 \n"\
141 "psrlw $8, %%xmm0 \n"\
142 "psrlw $8, %%xmm1 \n"\
143 "pand %%xmm7, %%xmm2 \n"\
144 "pand %%xmm7, %%xmm3 \n"\
145 "paddw %%xmm1, %%xmm0 \n"\
146 "paddw %%xmm3, %%xmm2 \n"\
147 "paddw %%xmm2, %%xmm0 \n"\
148 "paddw (%2,%0), %%xmm0 \n"\
149 "movdqa (%1,%0), %%xmm1 \n"\
150 "movdqa %%xmm0, (%1,%0) \n"\
151 "psubw %%xmm1, %%xmm0 \n"\
152 "movdqa %%xmm0, (%3,%0) \n"\
160 "r"(src+width*2+src_linesize),\
164 if (((intptr_t) src | src_linesize) & 15) {
170 #endif /* HAVE_SSE2_INLINE */
172 #endif /* HAVE_INLINE_ASM */
174 av_cold void ff_gradfun_init_x86(GradFunContext *gf)
176 int cpu_flags = av_get_cpu_flags();
178 #if HAVE_MMXEXT_INLINE
179 if (cpu_flags & AV_CPU_FLAG_MMXEXT)
180 gf->filter_line = gradfun_filter_line_mmxext;
182 #if HAVE_SSSE3_INLINE
183 if (cpu_flags & AV_CPU_FLAG_SSSE3)
184 gf->filter_line = gradfun_filter_line_ssse3;
187 if (cpu_flags & AV_CPU_FLAG_SSE2)
188 gf->blur_line = gradfun_blur_line_sse2;