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/x86/cpu.h"
26 #include "libavcodec/vp9dsp.h"
27 #include "libavcodec/x86/vp9dsp_init.h"
28 
29 #if HAVE_X86ASM
30 
31 extern const int16_t ff_filters_16bpp[3][15][4][16];
32 
33 decl_mc_funcs(4, sse2, int16_t, 16, BPC);
34 decl_mc_funcs(8, sse2, int16_t, 16, BPC);
35 decl_mc_funcs(16, avx2, int16_t, 16, BPC);
36 
37 mc_rep_funcs(16,  8, 16, sse2, int16_t, 16, BPC)
38 mc_rep_funcs(32, 16, 32, sse2, int16_t, 16, BPC)
39 mc_rep_funcs(64, 32, 64, sse2, int16_t, 16, BPC)
40 #if HAVE_AVX2_EXTERNAL
41 mc_rep_funcs(32, 16, 32, avx2, int16_t, 16, BPC)
42 mc_rep_funcs(64, 32, 64, avx2, int16_t, 16, BPC)
43 #endif
44 
45 filters_8tap_2d_fn2(put, 16, BPC, 2, sse2, sse2, 16bpp)
46 filters_8tap_2d_fn2(avg, 16, BPC, 2, sse2, sse2, 16bpp)
47 #if HAVE_AVX2_EXTERNAL
48 filters_8tap_2d_fn(put, 64, 32, BPC, 2, avx2, 16bpp)
49 filters_8tap_2d_fn(avg, 64, 32, BPC, 2, avx2, 16bpp)
50 filters_8tap_2d_fn(put, 32, 32, BPC, 2, avx2, 16bpp)
51 filters_8tap_2d_fn(avg, 32, 32, BPC, 2, avx2, 16bpp)
52 filters_8tap_2d_fn(put, 16, 32, BPC, 2, avx2, 16bpp)
53 filters_8tap_2d_fn(avg, 16, 32, BPC, 2, avx2, 16bpp)
54 #endif
55 
56 filters_8tap_1d_fn3(put, BPC, sse2, sse2, 16bpp)
57 filters_8tap_1d_fn3(avg, BPC, sse2, sse2, 16bpp)
58 #if HAVE_AVX2_EXTERNAL
59 filters_8tap_1d_fn2(put, 64, BPC, avx2, 16bpp)
60 filters_8tap_1d_fn2(avg, 64, BPC, avx2, 16bpp)
61 filters_8tap_1d_fn2(put, 32, BPC, avx2, 16bpp)
62 filters_8tap_1d_fn2(avg, 32, BPC, avx2, 16bpp)
63 filters_8tap_1d_fn2(put, 16, BPC, avx2, 16bpp)
64 filters_8tap_1d_fn2(avg, 16, BPC, avx2, 16bpp)
65 #endif
66 
67 #define decl_lpf_func(dir, wd, bpp, opt) \
68 void ff_vp9_loop_filter_##dir##_##wd##_##bpp##_##opt(uint8_t *dst, ptrdiff_t stride, \
69                                                      int E, int I, int H)
70 
71 #define decl_lpf_funcs(dir, wd, bpp) \
72 decl_lpf_func(dir, wd, bpp, sse2); \
73 decl_lpf_func(dir, wd, bpp, ssse3); \
74 decl_lpf_func(dir, wd, bpp, avx)
75 
76 #define decl_lpf_funcs_wd(dir) \
77 decl_lpf_funcs(dir,  4, BPC); \
78 decl_lpf_funcs(dir,  8, BPC); \
79 decl_lpf_funcs(dir, 16, BPC)
80 
81 decl_lpf_funcs_wd(h);
82 decl_lpf_funcs_wd(v);
83 
84 #define lpf_16_wrapper(dir, off, bpp, opt) \
85 static void loop_filter_##dir##_16_##bpp##_##opt(uint8_t *dst, ptrdiff_t stride, \
86                                                  int E, int I, int H) \
87 { \
88     ff_vp9_loop_filter_##dir##_16_##bpp##_##opt(dst,       stride, E, I, H); \
89     ff_vp9_loop_filter_##dir##_16_##bpp##_##opt(dst + off, stride, E, I, H); \
90 }
91 
92 #define lpf_16_wrappers(bpp, opt) \
93 lpf_16_wrapper(h, 8 * stride, bpp, opt) \
94 lpf_16_wrapper(v, 16,         bpp, opt)
95 
96 lpf_16_wrappers(BPC, sse2)
97 lpf_16_wrappers(BPC, ssse3)
98 lpf_16_wrappers(BPC, avx)
99 
100 #define lpf_mix2_wrapper(dir, off, wd1, wd2, bpp, opt) \
101 static void loop_filter_##dir##_##wd1##wd2##_##bpp##_##opt(uint8_t *dst, ptrdiff_t stride, \
102                                                            int E, int I, int H) \
103 { \
104     ff_vp9_loop_filter_##dir##_##wd1##_##bpp##_##opt(dst,       stride, \
105                                                      E & 0xff, I & 0xff, H & 0xff); \
106     ff_vp9_loop_filter_##dir##_##wd2##_##bpp##_##opt(dst + off, stride, \
107                                                      E >> 8,   I >> 8,   H >> 8); \
108 }
109 
110 #define lpf_mix2_wrappers(wd1, wd2, bpp, opt) \
111 lpf_mix2_wrapper(h, 8 * stride, wd1, wd2, bpp, opt) \
112 lpf_mix2_wrapper(v, 16,         wd1, wd2, bpp, opt)
113 
114 #define lpf_mix2_wrappers_set(bpp, opt) \
115 lpf_mix2_wrappers(4, 4, bpp, opt) \
116 lpf_mix2_wrappers(4, 8, bpp, opt) \
117 lpf_mix2_wrappers(8, 4, bpp, opt) \
118 lpf_mix2_wrappers(8, 8, bpp, opt) \
119 
120 lpf_mix2_wrappers_set(BPC, sse2)
121 lpf_mix2_wrappers_set(BPC, ssse3)
122 lpf_mix2_wrappers_set(BPC, avx)
123 
124 decl_ipred_fns(tm, BPC, mmxext, sse2);
125 
126 decl_itxfm_func(iwht, iwht, 4, BPC, mmxext);
127 #if BPC == 10
128 decl_itxfm_func(idct,  idct,  4, BPC, mmxext);
129 decl_itxfm_funcs(4, BPC, ssse3);
130 #else
131 decl_itxfm_func(idct,  idct,  4, BPC, sse2);
132 #endif
133 decl_itxfm_func(idct,  iadst, 4, BPC, sse2);
134 decl_itxfm_func(iadst, idct,  4, BPC, sse2);
135 decl_itxfm_func(iadst, iadst, 4, BPC, sse2);
136 decl_itxfm_funcs(8, BPC, sse2);
137 decl_itxfm_funcs(16, BPC, sse2);
138 decl_itxfm_func(idct,  idct, 32, BPC, sse2);
139 #endif /* HAVE_X86ASM */
140 
INIT_FUNC(VP9DSPContext *dsp, int bitexact)141 av_cold void INIT_FUNC(VP9DSPContext *dsp, int bitexact)
142 {
143 #if HAVE_X86ASM
144     int cpu_flags = av_get_cpu_flags();
145 
146 #define init_lpf_8_func(idx1, idx2, dir, wd, bpp, opt) \
147     dsp->loop_filter_8[idx1][idx2] = ff_vp9_loop_filter_##dir##_##wd##_##bpp##_##opt
148 #define init_lpf_16_func(idx, dir, bpp, opt) \
149     dsp->loop_filter_16[idx] = loop_filter_##dir##_16_##bpp##_##opt
150 #define init_lpf_mix2_func(idx1, idx2, idx3, dir, wd1, wd2, bpp, opt) \
151     dsp->loop_filter_mix2[idx1][idx2][idx3] = loop_filter_##dir##_##wd1##wd2##_##bpp##_##opt
152 
153 #define init_lpf_funcs(bpp, opt) \
154     init_lpf_8_func(0, 0, h,  4, bpp, opt); \
155     init_lpf_8_func(0, 1, v,  4, bpp, opt); \
156     init_lpf_8_func(1, 0, h,  8, bpp, opt); \
157     init_lpf_8_func(1, 1, v,  8, bpp, opt); \
158     init_lpf_8_func(2, 0, h, 16, bpp, opt); \
159     init_lpf_8_func(2, 1, v, 16, bpp, opt); \
160     init_lpf_16_func(0, h, bpp, opt); \
161     init_lpf_16_func(1, v, bpp, opt); \
162     init_lpf_mix2_func(0, 0, 0, h, 4, 4, bpp, opt); \
163     init_lpf_mix2_func(0, 1, 0, h, 4, 8, bpp, opt); \
164     init_lpf_mix2_func(1, 0, 0, h, 8, 4, bpp, opt); \
165     init_lpf_mix2_func(1, 1, 0, h, 8, 8, bpp, opt); \
166     init_lpf_mix2_func(0, 0, 1, v, 4, 4, bpp, opt); \
167     init_lpf_mix2_func(0, 1, 1, v, 4, 8, bpp, opt); \
168     init_lpf_mix2_func(1, 0, 1, v, 8, 4, bpp, opt); \
169     init_lpf_mix2_func(1, 1, 1, v, 8, 8, bpp, opt)
170 
171 #define init_itx_func(idxa, idxb, typea, typeb, size, bpp, opt) \
172     dsp->itxfm_add[idxa][idxb] = \
173         cat(ff_vp9_##typea##_##typeb##_##size##x##size##_add_, bpp, _##opt);
174 #define init_itx_func_one(idx, typea, typeb, size, bpp, opt) \
175     init_itx_func(idx, DCT_DCT,   typea, typeb, size, bpp, opt); \
176     init_itx_func(idx, ADST_DCT,  typea, typeb, size, bpp, opt); \
177     init_itx_func(idx, DCT_ADST,  typea, typeb, size, bpp, opt); \
178     init_itx_func(idx, ADST_ADST, typea, typeb, size, bpp, opt)
179 #define init_itx_funcs(idx, size, bpp, opt) \
180     init_itx_func(idx, DCT_DCT,   idct,  idct,  size, bpp, opt); \
181     init_itx_func(idx, ADST_DCT,  idct,  iadst, size, bpp, opt); \
182     init_itx_func(idx, DCT_ADST,  iadst, idct,  size, bpp, opt); \
183     init_itx_func(idx, ADST_ADST, iadst, iadst, size, bpp, opt); \
184 
185     if (EXTERNAL_MMXEXT(cpu_flags)) {
186         init_ipred_func(tm, TM_VP8, 4, BPC, mmxext);
187         if (!bitexact) {
188             init_itx_func_one(4 /* lossless */, iwht, iwht, 4, BPC, mmxext);
189 #if BPC == 10
190             init_itx_func(TX_4X4, DCT_DCT, idct, idct, 4, 10, mmxext);
191 #endif
192         }
193     }
194 
195     if (EXTERNAL_SSE2(cpu_flags)) {
196         init_subpel3(0, put, BPC, sse2);
197         init_subpel3(1, avg, BPC, sse2);
198         init_lpf_funcs(BPC, sse2);
199         init_8_16_32_ipred_funcs(tm, TM_VP8, BPC, sse2);
200 #if BPC == 10
201         if (!bitexact) {
202             init_itx_func(TX_4X4, ADST_DCT,  idct,  iadst, 4, 10, sse2);
203             init_itx_func(TX_4X4, DCT_ADST,  iadst, idct,  4, 10, sse2);
204             init_itx_func(TX_4X4, ADST_ADST, iadst, iadst, 4, 10, sse2);
205         }
206 #else
207         init_itx_funcs(TX_4X4, 4, 12, sse2);
208 #endif
209         init_itx_funcs(TX_8X8, 8, BPC, sse2);
210         init_itx_funcs(TX_16X16, 16, BPC, sse2);
211         init_itx_func_one(TX_32X32, idct, idct, 32, BPC, sse2);
212     }
213 
214     if (EXTERNAL_SSSE3(cpu_flags)) {
215         init_lpf_funcs(BPC, ssse3);
216 #if BPC == 10
217         if (!bitexact) {
218             init_itx_funcs(TX_4X4, 4, BPC, ssse3);
219         }
220 #endif
221     }
222 
223     if (EXTERNAL_AVX(cpu_flags)) {
224         init_lpf_funcs(BPC, avx);
225     }
226 
227     if (EXTERNAL_AVX2_FAST(cpu_flags)) {
228 #if HAVE_AVX2_EXTERNAL
229         init_subpel3_32_64(0,  put, BPC, avx2);
230         init_subpel3_32_64(1,  avg, BPC, avx2);
231         init_subpel2(2, 0, 16, put, BPC, avx2);
232         init_subpel2(2, 1, 16, avg, BPC, avx2);
233 #endif
234     }
235 
236 #endif /* HAVE_X86ASM */
237 
238     ff_vp9dsp_init_16bpp_x86(dsp);
239 }
240