1 /*****************************************************************************
2 * util.h: x86 inline asm
3 *****************************************************************************
4 * Copyright (C) 2008-2016 x264 project
6 * Authors: Fiona Glaser <fiona@x264.com>
7 * Loren Merritt <lorenm@u.washington.edu>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
23 * This program is also available under a commercial proprietary license.
24 * For more information, contact us at licensing@x264.com.
25 *****************************************************************************/
27 #ifndef X264_X86_UTIL_H
28 #define X264_X86_UTIL_H
31 #include <xmmintrin.h>
34 #define M128_ZERO ((__m128){0,0,0,0})
35 #define x264_union128_t x264_union128_sse_t
36 typedef union { __m128 i; uint64_t a[2]; uint32_t b[4]; uint16_t c[8]; uint8_t d[16]; } MAY_ALIAS x264_union128_sse_t;
38 typedef uint32_t v4si __attribute__((vector_size (16)));
42 #if HAVE_X86_INLINE_ASM && HAVE_MMX
44 #define x264_median_mv x264_median_mv_mmx2
45 static ALWAYS_INLINE void x264_median_mv_mmx2( int16_t *dst, int16_t *a, int16_t *b, int16_t *c )
50 "movq %%mm0, %%mm3 \n"
52 "pmaxsw %%mm1, %%mm0 \n"
53 "pminsw %%mm3, %%mm1 \n"
54 "pminsw %%mm2, %%mm0 \n"
55 "pmaxsw %%mm1, %%mm0 \n"
57 :"=m"(*(x264_union32_t*)dst)
58 :"m"(M32( a )), "m"(M32( b )), "m"(M32( c ))
62 #define x264_predictor_difference x264_predictor_difference_mmx2
63 static ALWAYS_INLINE int x264_predictor_difference_mmx2( int16_t (*mvc)[2], intptr_t i_mvc )
66 static const uint64_t pw_1 = 0x0001000100010001ULL;
69 "pxor %%mm4, %%mm4 \n"
72 "movd -8(%2,%1,4), %%mm0 \n"
73 "movd -4(%2,%1,4), %%mm3 \n"
74 "psubw %%mm3, %%mm0 \n"
79 "movq -8(%2,%1,4), %%mm0 \n"
80 "psubw -4(%2,%1,4), %%mm0 \n"
83 "pxor %%mm2, %%mm2 \n"
84 "psubw %%mm0, %%mm2 \n"
85 "pmaxsw %%mm2, %%mm0 \n"
86 "paddusw %%mm0, %%mm4 \n"
88 "pmaddwd %4, %%mm4 \n"
89 "pshufw $14, %%mm4, %%mm0 \n"
90 "paddd %%mm0, %%mm4 \n"
92 :"=r"(sum), "+r"(i_mvc)
93 :"r"(mvc), "m"(M64( mvc )), "m"(pw_1)
98 #define x264_cabac_mvd_sum x264_cabac_mvd_sum_mmx2
99 static ALWAYS_INLINE uint16_t x264_cabac_mvd_sum_mmx2(uint8_t *mvdleft, uint8_t *mvdtop)
101 static const uint64_t pb_2 = 0x0202020202020202ULL;
102 static const uint64_t pb_32 = 0x2020202020202020ULL;
103 static const uint64_t pb_33 = 0x2121212121212121ULL;
108 "paddusb %%mm1, %%mm0 \n"
109 "pminub %5, %%mm0 \n"
110 "pxor %%mm2, %%mm2 \n"
111 "movq %%mm0, %%mm1 \n"
112 "pcmpgtb %3, %%mm0 \n"
113 "pcmpgtb %4, %%mm1 \n"
114 "psubb %%mm0, %%mm2 \n"
115 "psubb %%mm1, %%mm2 \n"
118 :"m"(M16( mvdleft )),"m"(M16( mvdtop )),
119 "m"(pb_2),"m"(pb_32),"m"(pb_33)
124 #define x264_predictor_clip x264_predictor_clip_mmx2
125 static int ALWAYS_INLINE x264_predictor_clip_mmx2( int16_t (*dst)[2], int16_t (*mvc)[2], int i_mvc, int16_t mv_limit[2][2], uint32_t pmv )
127 static const uint32_t pd_32 = 0x20;
128 intptr_t tmp = (intptr_t)mv_limit, mvc_max = i_mvc, i = 0;
131 "movq (%2), %%mm5 \n"
133 "psllw $2, %%mm5 \n" // Convert to subpel
134 "pshufw $0xEE, %%mm5, %%mm6 \n"
136 "jz 2f \n" // if( i_mvc == 1 ) {do the last iteration}
137 "punpckldq %%mm3, %%mm3 \n"
138 "punpckldq %%mm5, %%mm5 \n"
140 "lea (%0,%3,4), %3 \n"
142 "movq (%0), %%mm0 \n"
144 "movq %%mm3, %%mm1 \n"
145 "pxor %%mm2, %%mm2 \n"
146 "pcmpeqd %%mm0, %%mm1 \n" // mv == pmv
147 "pcmpeqd %%mm0, %%mm2 \n" // mv == 0
148 "por %%mm1, %%mm2 \n" // (mv == pmv || mv == 0) * -1
149 "pmovmskb %%mm2, %k2 \n" // (mv == pmv || mv == 0) * 0xf
150 "pmaxsw %%mm5, %%mm0 \n"
151 "pminsw %%mm6, %%mm0 \n"
152 "pand %%mm4, %%mm2 \n" // (mv0 == pmv || mv0 == 0) * 32
153 "psrlq %%mm2, %%mm0 \n" // drop mv0 if it's skipped
154 "movq %%mm0, (%5,%4,4) \n"
158 "shr $4, %k2 \n" // (4-val)>>1
159 "sub %2, %4 \n" // +1 for each valid motion vector
162 "jg 3f \n" // if( i == i_mvc - 1 ) {do the last iteration}
164 /* Do the last iteration */
166 "movd (%0), %%mm0 \n"
167 "pxor %%mm2, %%mm2 \n"
168 "pcmpeqd %%mm0, %%mm3 \n"
169 "pcmpeqd %%mm0, %%mm2 \n"
170 "por %%mm3, %%mm2 \n"
171 "pmovmskb %%mm2, %k2 \n"
172 "pmaxsw %%mm5, %%mm0 \n"
173 "pminsw %%mm6, %%mm0 \n"
174 "movd %%mm0, (%5,%4,4) \n"
177 "sub %2, %4 \n" // output += !(mv == pmv || mv == 0)
179 :"+r"(mvc), "=m"(M64( dst )), "+r"(tmp), "+r"(mvc_max), "+r"(i)
180 :"r"(dst), "g"(pmv), "m"(pd_32), "m"(M64( mvc ))
185 /* Same as the above, except we do (mv + 2) >> 2 on the input. */
186 #define x264_predictor_roundclip x264_predictor_roundclip_mmx2
187 static int ALWAYS_INLINE x264_predictor_roundclip_mmx2( int16_t (*dst)[2], int16_t (*mvc)[2], int i_mvc, int16_t mv_limit[2][2], uint32_t pmv )
189 static const uint64_t pw_2 = 0x0002000200020002ULL;
190 static const uint32_t pd_32 = 0x20;
191 intptr_t tmp = (intptr_t)mv_limit, mvc_max = i_mvc, i = 0;
194 "movq (%2), %%mm5 \n"
197 "pshufw $0xEE, %%mm5, %%mm6 \n"
200 "punpckldq %%mm3, %%mm3 \n"
201 "punpckldq %%mm5, %%mm5 \n"
203 "lea (%0,%3,4), %3 \n"
205 "movq (%0), %%mm0 \n"
207 "paddw %%mm7, %%mm0 \n"
209 "movq %%mm3, %%mm1 \n"
210 "pxor %%mm2, %%mm2 \n"
211 "pcmpeqd %%mm0, %%mm1 \n"
212 "pcmpeqd %%mm0, %%mm2 \n"
213 "por %%mm1, %%mm2 \n"
214 "pmovmskb %%mm2, %k2 \n"
215 "pmaxsw %%mm5, %%mm0 \n"
216 "pminsw %%mm6, %%mm0 \n"
217 "pand %%mm4, %%mm2 \n"
218 "psrlq %%mm2, %%mm0 \n"
219 "movq %%mm0, (%5,%4,4) \n"
229 /* Do the last iteration */
231 "movd (%0), %%mm0 \n"
232 "paddw %%mm7, %%mm0 \n"
234 "pxor %%mm2, %%mm2 \n"
235 "pcmpeqd %%mm0, %%mm3 \n"
236 "pcmpeqd %%mm0, %%mm2 \n"
237 "por %%mm3, %%mm2 \n"
238 "pmovmskb %%mm2, %k2 \n"
239 "pmaxsw %%mm5, %%mm0 \n"
240 "pminsw %%mm6, %%mm0 \n"
241 "movd %%mm0, (%5,%4,4) \n"
246 :"+r"(mvc), "=m"(M64( dst )), "+r"(tmp), "+r"(mvc_max), "+r"(i)
247 :"r"(dst), "m"(pw_2), "g"(pmv), "m"(pd_32), "m"(M64( mvc ))