xref: /aosp_15_r20/external/libdav1d/src/x86/looprestoration.h (revision c09093415860a1c2373dacd84c4fde00c507cdfd)
1*c0909341SAndroid Build Coastguard Worker /*
2*c0909341SAndroid Build Coastguard Worker  * Copyright © 2018, VideoLAN and dav1d authors
3*c0909341SAndroid Build Coastguard Worker  * Copyright © 2018, Two Orioles, LLC
4*c0909341SAndroid Build Coastguard Worker  * All rights reserved.
5*c0909341SAndroid Build Coastguard Worker  *
6*c0909341SAndroid Build Coastguard Worker  * Redistribution and use in source and binary forms, with or without
7*c0909341SAndroid Build Coastguard Worker  * modification, are permitted provided that the following conditions are met:
8*c0909341SAndroid Build Coastguard Worker  *
9*c0909341SAndroid Build Coastguard Worker  * 1. Redistributions of source code must retain the above copyright notice, this
10*c0909341SAndroid Build Coastguard Worker  *    list of conditions and the following disclaimer.
11*c0909341SAndroid Build Coastguard Worker  *
12*c0909341SAndroid Build Coastguard Worker  * 2. Redistributions in binary form must reproduce the above copyright notice,
13*c0909341SAndroid Build Coastguard Worker  *    this list of conditions and the following disclaimer in the documentation
14*c0909341SAndroid Build Coastguard Worker  *    and/or other materials provided with the distribution.
15*c0909341SAndroid Build Coastguard Worker  *
16*c0909341SAndroid Build Coastguard Worker  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
17*c0909341SAndroid Build Coastguard Worker  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
18*c0909341SAndroid Build Coastguard Worker  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
19*c0909341SAndroid Build Coastguard Worker  * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
20*c0909341SAndroid Build Coastguard Worker  * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
21*c0909341SAndroid Build Coastguard Worker  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
22*c0909341SAndroid Build Coastguard Worker  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
23*c0909341SAndroid Build Coastguard Worker  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24*c0909341SAndroid Build Coastguard Worker  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
25*c0909341SAndroid Build Coastguard Worker  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26*c0909341SAndroid Build Coastguard Worker  */
27*c0909341SAndroid Build Coastguard Worker 
28*c0909341SAndroid Build Coastguard Worker #include "src/cpu.h"
29*c0909341SAndroid Build Coastguard Worker #include "src/looprestoration.h"
30*c0909341SAndroid Build Coastguard Worker 
31*c0909341SAndroid Build Coastguard Worker #include "common/intops.h"
32*c0909341SAndroid Build Coastguard Worker 
33*c0909341SAndroid Build Coastguard Worker #define decl_wiener_filter_fns(ext) \
34*c0909341SAndroid Build Coastguard Worker decl_lr_filter_fn(BF(dav1d_wiener_filter7, ext)); \
35*c0909341SAndroid Build Coastguard Worker decl_lr_filter_fn(BF(dav1d_wiener_filter5, ext))
36*c0909341SAndroid Build Coastguard Worker 
37*c0909341SAndroid Build Coastguard Worker #define decl_sgr_filter_fns(ext) \
38*c0909341SAndroid Build Coastguard Worker decl_lr_filter_fn(BF(dav1d_sgr_filter_5x5, ext)); \
39*c0909341SAndroid Build Coastguard Worker decl_lr_filter_fn(BF(dav1d_sgr_filter_3x3, ext)); \
40*c0909341SAndroid Build Coastguard Worker decl_lr_filter_fn(BF(dav1d_sgr_filter_mix, ext))
41*c0909341SAndroid Build Coastguard Worker 
42*c0909341SAndroid Build Coastguard Worker decl_wiener_filter_fns(sse2);
43*c0909341SAndroid Build Coastguard Worker decl_wiener_filter_fns(ssse3);
44*c0909341SAndroid Build Coastguard Worker decl_wiener_filter_fns(avx2);
45*c0909341SAndroid Build Coastguard Worker decl_wiener_filter_fns(avx512icl);
46*c0909341SAndroid Build Coastguard Worker decl_sgr_filter_fns(ssse3);
47*c0909341SAndroid Build Coastguard Worker decl_sgr_filter_fns(avx2);
48*c0909341SAndroid Build Coastguard Worker decl_sgr_filter_fns(avx512icl);
49*c0909341SAndroid Build Coastguard Worker 
loop_restoration_dsp_init_x86(Dav1dLoopRestorationDSPContext * const c,const int bpc)50*c0909341SAndroid Build Coastguard Worker static ALWAYS_INLINE void loop_restoration_dsp_init_x86(Dav1dLoopRestorationDSPContext *const c, const int bpc) {
51*c0909341SAndroid Build Coastguard Worker     const unsigned flags = dav1d_get_cpu_flags();
52*c0909341SAndroid Build Coastguard Worker 
53*c0909341SAndroid Build Coastguard Worker     if (!(flags & DAV1D_X86_CPU_FLAG_SSE2)) return;
54*c0909341SAndroid Build Coastguard Worker #if BITDEPTH == 8
55*c0909341SAndroid Build Coastguard Worker     c->wiener[0] = BF(dav1d_wiener_filter7, sse2);
56*c0909341SAndroid Build Coastguard Worker     c->wiener[1] = BF(dav1d_wiener_filter5, sse2);
57*c0909341SAndroid Build Coastguard Worker #endif
58*c0909341SAndroid Build Coastguard Worker 
59*c0909341SAndroid Build Coastguard Worker     if (!(flags & DAV1D_X86_CPU_FLAG_SSSE3)) return;
60*c0909341SAndroid Build Coastguard Worker     c->wiener[0] = BF(dav1d_wiener_filter7, ssse3);
61*c0909341SAndroid Build Coastguard Worker     c->wiener[1] = BF(dav1d_wiener_filter5, ssse3);
62*c0909341SAndroid Build Coastguard Worker     if (BITDEPTH == 8 || bpc == 10) {
63*c0909341SAndroid Build Coastguard Worker         c->sgr[0] = BF(dav1d_sgr_filter_5x5, ssse3);
64*c0909341SAndroid Build Coastguard Worker         c->sgr[1] = BF(dav1d_sgr_filter_3x3, ssse3);
65*c0909341SAndroid Build Coastguard Worker         c->sgr[2] = BF(dav1d_sgr_filter_mix, ssse3);
66*c0909341SAndroid Build Coastguard Worker     }
67*c0909341SAndroid Build Coastguard Worker 
68*c0909341SAndroid Build Coastguard Worker #if ARCH_X86_64
69*c0909341SAndroid Build Coastguard Worker     if (!(flags & DAV1D_X86_CPU_FLAG_AVX2)) return;
70*c0909341SAndroid Build Coastguard Worker 
71*c0909341SAndroid Build Coastguard Worker     c->wiener[0] = BF(dav1d_wiener_filter7, avx2);
72*c0909341SAndroid Build Coastguard Worker     c->wiener[1] = BF(dav1d_wiener_filter5, avx2);
73*c0909341SAndroid Build Coastguard Worker     if (BITDEPTH == 8 || bpc == 10) {
74*c0909341SAndroid Build Coastguard Worker         c->sgr[0] = BF(dav1d_sgr_filter_5x5, avx2);
75*c0909341SAndroid Build Coastguard Worker         c->sgr[1] = BF(dav1d_sgr_filter_3x3, avx2);
76*c0909341SAndroid Build Coastguard Worker         c->sgr[2] = BF(dav1d_sgr_filter_mix, avx2);
77*c0909341SAndroid Build Coastguard Worker     }
78*c0909341SAndroid Build Coastguard Worker 
79*c0909341SAndroid Build Coastguard Worker     if (!(flags & DAV1D_X86_CPU_FLAG_AVX512ICL)) return;
80*c0909341SAndroid Build Coastguard Worker 
81*c0909341SAndroid Build Coastguard Worker     c->wiener[0] = BF(dav1d_wiener_filter7, avx512icl);
82*c0909341SAndroid Build Coastguard Worker #if BITDEPTH == 8
83*c0909341SAndroid Build Coastguard Worker     /* With VNNI we don't need a 5-tap version. */
84*c0909341SAndroid Build Coastguard Worker     c->wiener[1] = c->wiener[0];
85*c0909341SAndroid Build Coastguard Worker #else
86*c0909341SAndroid Build Coastguard Worker     c->wiener[1] = BF(dav1d_wiener_filter5, avx512icl);
87*c0909341SAndroid Build Coastguard Worker #endif
88*c0909341SAndroid Build Coastguard Worker     if (BITDEPTH == 8 || bpc == 10) {
89*c0909341SAndroid Build Coastguard Worker         c->sgr[0] = BF(dav1d_sgr_filter_5x5, avx512icl);
90*c0909341SAndroid Build Coastguard Worker         c->sgr[1] = BF(dav1d_sgr_filter_3x3, avx512icl);
91*c0909341SAndroid Build Coastguard Worker         c->sgr[2] = BF(dav1d_sgr_filter_mix, avx512icl);
92*c0909341SAndroid Build Coastguard Worker     }
93*c0909341SAndroid Build Coastguard Worker #endif
94*c0909341SAndroid Build Coastguard Worker }
95