dsputil_mmx.h
Go to the documentation of this file.
1 /*
2  * MMX optimized DSP utils
3  * Copyright (c) 2007 Aurelien Jacobs <aurel@gnuage.org>
4  *
5  * This file is part of Libav.
6  *
7  * Libav is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * Libav is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with Libav; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 #ifndef AVCODEC_X86_DSPUTIL_MMX_H
23 #define AVCODEC_X86_DSPUTIL_MMX_H
24 
25 #include <stdint.h>
26 #include "libavcodec/dsputil.h"
27 #include "libavutil/x86_cpu.h"
28 
29 typedef struct { uint64_t a, b; } xmm_reg;
30 
31 extern const uint64_t ff_bone;
32 extern const uint64_t ff_wtwo;
33 
34 extern const uint64_t ff_pdw_80000000[2];
35 
36 extern const xmm_reg ff_pw_3;
37 extern const xmm_reg ff_pw_4;
38 extern const xmm_reg ff_pw_5;
39 extern const xmm_reg ff_pw_8;
40 extern const uint64_t ff_pw_15;
41 extern const xmm_reg ff_pw_16;
42 extern const xmm_reg ff_pw_18;
43 extern const uint64_t ff_pw_20;
44 extern const xmm_reg ff_pw_27;
45 extern const xmm_reg ff_pw_28;
46 extern const xmm_reg ff_pw_32;
47 extern const uint64_t ff_pw_42;
48 extern const uint64_t ff_pw_53;
49 extern const xmm_reg ff_pw_63;
50 extern const xmm_reg ff_pw_64;
51 extern const uint64_t ff_pw_96;
52 extern const uint64_t ff_pw_128;
53 extern const uint64_t ff_pw_255;
54 
55 extern const xmm_reg ff_pb_1;
56 extern const xmm_reg ff_pb_3;
57 extern const uint64_t ff_pb_7;
58 extern const uint64_t ff_pb_1F;
59 extern const uint64_t ff_pb_3F;
60 extern const uint64_t ff_pb_81;
61 extern const xmm_reg ff_pb_A1;
62 extern const xmm_reg ff_pb_F8;
63 extern const uint64_t ff_pb_FC;
64 extern const xmm_reg ff_pb_FE;
65 
66 extern const double ff_pd_1[2];
67 extern const double ff_pd_2[2];
68 
69 #define LOAD4(stride,in,a,b,c,d)\
70  "movq 0*"#stride"+"#in", "#a"\n\t"\
71  "movq 1*"#stride"+"#in", "#b"\n\t"\
72  "movq 2*"#stride"+"#in", "#c"\n\t"\
73  "movq 3*"#stride"+"#in", "#d"\n\t"
74 
75 #define STORE4(stride,out,a,b,c,d)\
76  "movq "#a", 0*"#stride"+"#out"\n\t"\
77  "movq "#b", 1*"#stride"+"#out"\n\t"\
78  "movq "#c", 2*"#stride"+"#out"\n\t"\
79  "movq "#d", 3*"#stride"+"#out"\n\t"
80 
81 /* in/out: mma=mma+mmb, mmb=mmb-mma */
82 #define SUMSUB_BA( a, b ) \
83  "paddw "#b", "#a" \n\t"\
84  "paddw "#b", "#b" \n\t"\
85  "psubw "#a", "#b" \n\t"
86 
87 #define SBUTTERFLY(a,b,t,n,m)\
88  "mov" #m " " #a ", " #t " \n\t" /* abcd */\
89  "punpckl" #n " " #b ", " #a " \n\t" /* aebf */\
90  "punpckh" #n " " #b ", " #t " \n\t" /* cgdh */\
91 
92 #define TRANSPOSE4(a,b,c,d,t)\
93  SBUTTERFLY(a,b,t,wd,q) /* a=aebf t=cgdh */\
94  SBUTTERFLY(c,d,b,wd,q) /* c=imjn b=kolp */\
95  SBUTTERFLY(a,c,d,dq,q) /* a=aeim d=bfjn */\
96  SBUTTERFLY(t,b,c,dq,q) /* t=cgko c=dhlp */
97 
98 static inline void transpose4x4(uint8_t *dst, uint8_t *src, x86_reg dst_stride, x86_reg src_stride){
99  __asm__ volatile( //FIXME could save 1 instruction if done as 8x4 ...
100  "movd (%1), %%mm0 \n\t"
101  "add %3, %1 \n\t"
102  "movd (%1), %%mm1 \n\t"
103  "movd (%1,%3,1), %%mm2 \n\t"
104  "movd (%1,%3,2), %%mm3 \n\t"
105  "punpcklbw %%mm1, %%mm0 \n\t"
106  "punpcklbw %%mm3, %%mm2 \n\t"
107  "movq %%mm0, %%mm1 \n\t"
108  "punpcklwd %%mm2, %%mm0 \n\t"
109  "punpckhwd %%mm2, %%mm1 \n\t"
110  "movd %%mm0, (%0) \n\t"
111  "add %2, %0 \n\t"
112  "punpckhdq %%mm0, %%mm0 \n\t"
113  "movd %%mm0, (%0) \n\t"
114  "movd %%mm1, (%0,%2,1) \n\t"
115  "punpckhdq %%mm1, %%mm1 \n\t"
116  "movd %%mm1, (%0,%2,2) \n\t"
117 
118  : "+&r" (dst),
119  "+&r" (src)
120  : "r" (dst_stride),
121  "r" (src_stride)
122  : "memory"
123  );
124 }
125 
126 // e,f,g,h can be memory
127 // out: a,d,t,c
128 #define TRANSPOSE8x4(a,b,c,d,e,f,g,h,t)\
129  "punpcklbw " #e ", " #a " \n\t" /* a0 e0 a1 e1 a2 e2 a3 e3 */\
130  "punpcklbw " #f ", " #b " \n\t" /* b0 f0 b1 f1 b2 f2 b3 f3 */\
131  "punpcklbw " #g ", " #c " \n\t" /* c0 g0 c1 g1 c2 g2 d3 g3 */\
132  "punpcklbw " #h ", " #d " \n\t" /* d0 h0 d1 h1 d2 h2 d3 h3 */\
133  SBUTTERFLY(a, b, t, bw, q) /* a= a0 b0 e0 f0 a1 b1 e1 f1 */\
134  /* t= a2 b2 e2 f2 a3 b3 e3 f3 */\
135  SBUTTERFLY(c, d, b, bw, q) /* c= c0 d0 g0 h0 c1 d1 g1 h1 */\
136  /* b= c2 d2 g2 h2 c3 d3 g3 h3 */\
137  SBUTTERFLY(a, c, d, wd, q) /* a= a0 b0 c0 d0 e0 f0 g0 h0 */\
138  /* d= a1 b1 c1 d1 e1 f1 g1 h1 */\
139  SBUTTERFLY(t, b, c, wd, q) /* t= a2 b2 c2 d2 e2 f2 g2 h2 */\
140  /* c= a3 b3 c3 d3 e3 f3 g3 h3 */
141 
142 #if ARCH_X86_64
143 // permutes 01234567 -> 05736421
144 #define TRANSPOSE8(a,b,c,d,e,f,g,h,t)\
145  SBUTTERFLY(a,b,%%xmm8,wd,dqa)\
146  SBUTTERFLY(c,d,b,wd,dqa)\
147  SBUTTERFLY(e,f,d,wd,dqa)\
148  SBUTTERFLY(g,h,f,wd,dqa)\
149  SBUTTERFLY(a,c,h,dq,dqa)\
150  SBUTTERFLY(%%xmm8,b,c,dq,dqa)\
151  SBUTTERFLY(e,g,b,dq,dqa)\
152  SBUTTERFLY(d,f,g,dq,dqa)\
153  SBUTTERFLY(a,e,f,qdq,dqa)\
154  SBUTTERFLY(%%xmm8,d,e,qdq,dqa)\
155  SBUTTERFLY(h,b,d,qdq,dqa)\
156  SBUTTERFLY(c,g,b,qdq,dqa)\
157  "movdqa %%xmm8, "#g" \n\t"
158 #else
159 #define TRANSPOSE8(a,b,c,d,e,f,g,h,t)\
160  "movdqa "#h", "#t" \n\t"\
161  SBUTTERFLY(a,b,h,wd,dqa)\
162  "movdqa "#h", 16"#t" \n\t"\
163  "movdqa "#t", "#h" \n\t"\
164  SBUTTERFLY(c,d,b,wd,dqa)\
165  SBUTTERFLY(e,f,d,wd,dqa)\
166  SBUTTERFLY(g,h,f,wd,dqa)\
167  SBUTTERFLY(a,c,h,dq,dqa)\
168  "movdqa "#h", "#t" \n\t"\
169  "movdqa 16"#t", "#h" \n\t"\
170  SBUTTERFLY(h,b,c,dq,dqa)\
171  SBUTTERFLY(e,g,b,dq,dqa)\
172  SBUTTERFLY(d,f,g,dq,dqa)\
173  SBUTTERFLY(a,e,f,qdq,dqa)\
174  SBUTTERFLY(h,d,e,qdq,dqa)\
175  "movdqa "#h", 16"#t" \n\t"\
176  "movdqa "#t", "#h" \n\t"\
177  SBUTTERFLY(h,b,d,qdq,dqa)\
178  SBUTTERFLY(c,g,b,qdq,dqa)\
179  "movdqa 16"#t", "#g" \n\t"
180 #endif
181 
182 #define MOVQ_WONE(regd) \
183  __asm__ volatile ( \
184  "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
185  "psrlw $15, %%" #regd ::)
186 
189 
190 void ff_add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
191 void ff_put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
192 void ff_put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
193 
194 void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
195 void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
196 void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
197 void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
198 
199 void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd);
200 void ff_avg_vc1_mspel_mc00_mmx2(uint8_t *dst, const uint8_t *src, int stride, int rnd);
201 
202 void ff_mmx_idct(DCTELEM *block);
204 
205 
206 void ff_deinterlace_line_mmx(uint8_t *dst,
207  const uint8_t *lum_m4, const uint8_t *lum_m3,
208  const uint8_t *lum_m2, const uint8_t *lum_m1,
209  const uint8_t *lum,
210  int size);
211 
212 void ff_deinterlace_line_inplace_mmx(const uint8_t *lum_m4,
213  const uint8_t *lum_m3,
214  const uint8_t *lum_m2,
215  const uint8_t *lum_m1,
216  const uint8_t *lum, int size);
217 
218 #endif /* AVCODEC_X86_DSPUTIL_MMX_H */