1 /*
2 * VP9 SIMD optimizations
3 *
4 * Copyright (c) 2013 Ronald S. Bultje <rsbultje gmail com>
5 *
6 * This file is part of FFmpeg.
7 *
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23 #include "libavutil/attributes.h"
24 #include "libavutil/cpu.h"
25 #include "libavutil/mem.h"
26 #include "libavutil/x86/cpu.h"
27 #include "libavcodec/vp9dsp.h"
28 #include "libavcodec/x86/vp9dsp_init.h"
29
30 #if HAVE_X86ASM
31
32 decl_fpel_func(put, 8, , mmx);
33 decl_fpel_func(avg, 8, _16, mmxext);
34 decl_fpel_func(put, 16, , sse);
35 decl_fpel_func(put, 32, , sse);
36 decl_fpel_func(put, 64, , sse);
37 decl_fpel_func(put, 128, , sse);
38 decl_fpel_func(avg, 16, _16, sse2);
39 decl_fpel_func(avg, 32, _16, sse2);
40 decl_fpel_func(avg, 64, _16, sse2);
41 decl_fpel_func(avg, 128, _16, sse2);
42 decl_fpel_func(put, 32, , avx);
43 decl_fpel_func(put, 64, , avx);
44 decl_fpel_func(put, 128, , avx);
45 decl_fpel_func(avg, 32, _16, avx2);
46 decl_fpel_func(avg, 64, _16, avx2);
47 decl_fpel_func(avg, 128, _16, avx2);
48
49 decl_ipred_fns(v, 16, mmx, sse);
50 decl_ipred_fns(h, 16, mmxext, sse2);
51 decl_ipred_fns(dc, 16, mmxext, sse2);
52 decl_ipred_fns(dc_top, 16, mmxext, sse2);
53 decl_ipred_fns(dc_left, 16, mmxext, sse2);
54 decl_ipred_fn(dl, 16, 16, avx2);
55 decl_ipred_fn(dl, 32, 16, avx2);
56 decl_ipred_fn(dr, 16, 16, avx2);
57 decl_ipred_fn(dr, 32, 16, avx2);
58
59 #define decl_ipred_dir_funcs(type) \
60 decl_ipred_fns(type, 16, sse2, sse2); \
61 decl_ipred_fns(type, 16, ssse3, ssse3); \
62 decl_ipred_fns(type, 16, avx, avx)
63
64 decl_ipred_dir_funcs(dl);
65 decl_ipred_dir_funcs(dr);
66 decl_ipred_dir_funcs(vl);
67 decl_ipred_dir_funcs(vr);
68 decl_ipred_dir_funcs(hu);
69 decl_ipred_dir_funcs(hd);
70 #endif /* HAVE_X86ASM */
71
ff_vp9dsp_init_16bpp_x86(VP9DSPContext * dsp)72 av_cold void ff_vp9dsp_init_16bpp_x86(VP9DSPContext *dsp)
73 {
74 #if HAVE_X86ASM
75 int cpu_flags = av_get_cpu_flags();
76
77 if (EXTERNAL_MMX(cpu_flags)) {
78 init_fpel_func(4, 0, 8, put, , mmx);
79 init_ipred_func(v, VERT, 4, 16, mmx);
80 }
81
82 if (EXTERNAL_MMXEXT(cpu_flags)) {
83 init_fpel_func(4, 1, 8, avg, _16, mmxext);
84 init_ipred_func(h, HOR, 4, 16, mmxext);
85 init_ipred_func(dc, DC, 4, 16, mmxext);
86 init_ipred_func(dc_top, TOP_DC, 4, 16, mmxext);
87 init_ipred_func(dc_left, LEFT_DC, 4, 16, mmxext);
88 }
89
90 if (EXTERNAL_SSE(cpu_flags)) {
91 init_fpel_func(3, 0, 16, put, , sse);
92 init_fpel_func(2, 0, 32, put, , sse);
93 init_fpel_func(1, 0, 64, put, , sse);
94 init_fpel_func(0, 0, 128, put, , sse);
95 init_8_16_32_ipred_funcs(v, VERT, 16, sse);
96 }
97
98 if (EXTERNAL_SSE2(cpu_flags)) {
99 init_fpel_func(3, 1, 16, avg, _16, sse2);
100 init_fpel_func(2, 1, 32, avg, _16, sse2);
101 init_fpel_func(1, 1, 64, avg, _16, sse2);
102 init_fpel_func(0, 1, 128, avg, _16, sse2);
103 init_8_16_32_ipred_funcs(h, HOR, 16, sse2);
104 init_8_16_32_ipred_funcs(dc, DC, 16, sse2);
105 init_8_16_32_ipred_funcs(dc_top, TOP_DC, 16, sse2);
106 init_8_16_32_ipred_funcs(dc_left, LEFT_DC, 16, sse2);
107 init_ipred_funcs(dl, DIAG_DOWN_LEFT, 16, sse2);
108 init_ipred_funcs(dr, DIAG_DOWN_RIGHT, 16, sse2);
109 init_ipred_funcs(vl, VERT_LEFT, 16, sse2);
110 init_ipred_funcs(vr, VERT_RIGHT, 16, sse2);
111 init_ipred_funcs(hu, HOR_UP, 16, sse2);
112 init_ipred_funcs(hd, HOR_DOWN, 16, sse2);
113 }
114
115 if (EXTERNAL_SSSE3(cpu_flags)) {
116 init_ipred_funcs(dl, DIAG_DOWN_LEFT, 16, ssse3);
117 init_ipred_funcs(dr, DIAG_DOWN_RIGHT, 16, ssse3);
118 init_ipred_funcs(vl, VERT_LEFT, 16, ssse3);
119 init_ipred_funcs(vr, VERT_RIGHT, 16, ssse3);
120 init_ipred_funcs(hu, HOR_UP, 16, ssse3);
121 init_ipred_funcs(hd, HOR_DOWN, 16, ssse3);
122 }
123
124 if (EXTERNAL_AVX_FAST(cpu_flags)) {
125 init_fpel_func(2, 0, 32, put, , avx);
126 init_fpel_func(1, 0, 64, put, , avx);
127 init_fpel_func(0, 0, 128, put, , avx);
128 init_ipred_funcs(dl, DIAG_DOWN_LEFT, 16, avx);
129 init_ipred_funcs(dr, DIAG_DOWN_RIGHT, 16, avx);
130 init_ipred_funcs(vl, VERT_LEFT, 16, avx);
131 init_ipred_funcs(vr, VERT_RIGHT, 16, avx);
132 init_ipred_funcs(hu, HOR_UP, 16, avx);
133 init_ipred_funcs(hd, HOR_DOWN, 16, avx);
134 }
135
136 if (EXTERNAL_AVX2_FAST(cpu_flags)) {
137 init_fpel_func(2, 1, 32, avg, _16, avx2);
138 init_fpel_func(1, 1, 64, avg, _16, avx2);
139 init_fpel_func(0, 1, 128, avg, _16, avx2);
140 init_ipred_func(dl, DIAG_DOWN_LEFT, 16, 16, avx2);
141 init_ipred_func(dl, DIAG_DOWN_LEFT, 32, 16, avx2);
142 init_ipred_func(dr, DIAG_DOWN_RIGHT, 16, 16, avx2);
143 #if ARCH_X86_64
144 init_ipred_func(dr, DIAG_DOWN_RIGHT, 32, 16, avx2);
145 #endif
146 }
147
148 #endif /* HAVE_X86ASM */
149 }
150