1/*
2 * RV40 decoder motion compensation functions x86-optimised
3 * Copyright (c) 2008 Konstantin Shishkov
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg 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 * FFmpeg 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 FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22/**
23 * @file
24 * RV40 decoder motion compensation functions x86-optimised
25 * 2,0 and 0,2 have h264 equivalents.
26 * 3,3 is bugged in the rv40 format and maps to _xy2 version
27 */
28
29#include "libavcodec/rv34dsp.h"
30#include "libavutil/attributes.h"
31#include "libavutil/mem_internal.h"
32#include "libavutil/x86/cpu.h"
33#include "hpeldsp.h"
34
35#define DEFINE_FN(op, size, insn) \
36static void op##_rv40_qpel##size##_mc33_##insn(uint8_t *dst, const uint8_t *src, \
37                                               ptrdiff_t stride) \
38{ \
39    ff_##op##_pixels##size##_xy2_##insn(dst, src, stride, size); \
40}
41
42#if HAVE_X86ASM
43void ff_put_rv40_chroma_mc8_mmx  (uint8_t *dst, uint8_t *src,
44                                  ptrdiff_t stride, int h, int x, int y);
45void ff_avg_rv40_chroma_mc8_mmxext(uint8_t *dst, uint8_t *src,
46                                   ptrdiff_t stride, int h, int x, int y);
47
48void ff_put_rv40_chroma_mc4_mmx  (uint8_t *dst, uint8_t *src,
49                                  ptrdiff_t stride, int h, int x, int y);
50void ff_avg_rv40_chroma_mc4_mmxext(uint8_t *dst, uint8_t *src,
51                                   ptrdiff_t stride, int h, int x, int y);
52
53#define DECLARE_WEIGHT(opt) \
54void ff_rv40_weight_func_rnd_16_##opt(uint8_t *dst, uint8_t *src1, uint8_t *src2, \
55                                      int w1, int w2, ptrdiff_t stride); \
56void ff_rv40_weight_func_rnd_8_##opt (uint8_t *dst, uint8_t *src1, uint8_t *src2, \
57                                      int w1, int w2, ptrdiff_t stride); \
58void ff_rv40_weight_func_nornd_16_##opt(uint8_t *dst, uint8_t *src1, uint8_t *src2, \
59                                        int w1, int w2, ptrdiff_t stride); \
60void ff_rv40_weight_func_nornd_8_##opt (uint8_t *dst, uint8_t *src1, uint8_t *src2, \
61                                        int w1, int w2, ptrdiff_t stride);
62DECLARE_WEIGHT(sse2)
63DECLARE_WEIGHT(ssse3)
64
65/** @{ */
66/**
67 * Define one qpel function.
68 * LOOPSIZE must be already set to the number of pixels processed per
69 * iteration in the inner loop of the called functions.
70 * COFF(x) must be already defined so as to provide the offset into any
71 * array of coeffs used by the called function for the qpel position x.
72 */
73#define QPEL_FUNC_DECL(OP, SIZE, PH, PV, OPT)                           \
74static void OP ## rv40_qpel ##SIZE ##_mc ##PH ##PV ##OPT(uint8_t *dst,  \
75                                                         const uint8_t *src, \
76                                                         ptrdiff_t stride)  \
77{                                                                       \
78    int i;                                                              \
79    if (PH && PV) {                                                     \
80        LOCAL_ALIGNED(16, uint8_t, tmp, [SIZE * (SIZE + 5)]);           \
81        uint8_t *tmpptr = tmp + SIZE * 2;                               \
82        src -= stride * 2;                                              \
83                                                                        \
84        for (i = 0; i < SIZE; i += LOOPSIZE)                            \
85            ff_put_rv40_qpel_h ##OPT(tmp + i, SIZE, src + i, stride,    \
86                                     SIZE + 5, HCOFF(PH));              \
87        for (i = 0; i < SIZE; i += LOOPSIZE)                            \
88            ff_ ##OP ##rv40_qpel_v ##OPT(dst + i, stride, tmpptr + i,   \
89                                         SIZE, SIZE, VCOFF(PV));        \
90    } else if (PV) {                                                    \
91        for (i = 0; i < SIZE; i += LOOPSIZE)                            \
92            ff_ ##OP ##rv40_qpel_v ## OPT(dst + i, stride, src + i,     \
93                                          stride, SIZE, VCOFF(PV));     \
94    } else {                                                            \
95        for (i = 0; i < SIZE; i += LOOPSIZE)                            \
96            ff_ ##OP ##rv40_qpel_h ## OPT(dst + i, stride, src + i,     \
97                                          stride, SIZE, HCOFF(PH));     \
98    }                                                                   \
99}
100
101/** Declare functions for sizes 8 and 16 and given operations
102 *  and qpel position. */
103#define QPEL_FUNCS_DECL(OP, PH, PV, OPT) \
104    QPEL_FUNC_DECL(OP,  8, PH, PV, OPT)  \
105    QPEL_FUNC_DECL(OP, 16, PH, PV, OPT)
106
107/** Declare all functions for all sizes and qpel positions */
108#define QPEL_MC_DECL(OP, OPT)                                           \
109void ff_ ##OP ##rv40_qpel_h ##OPT(uint8_t *dst, ptrdiff_t dstStride,    \
110                                  const uint8_t *src,                   \
111                                  ptrdiff_t srcStride,                  \
112                                  int len, int m);                      \
113void ff_ ##OP ##rv40_qpel_v ##OPT(uint8_t *dst, ptrdiff_t dstStride,    \
114                                  const uint8_t *src,                   \
115                                  ptrdiff_t srcStride,                  \
116                                  int len, int m);                      \
117QPEL_FUNCS_DECL(OP, 0, 1, OPT)                                          \
118QPEL_FUNCS_DECL(OP, 0, 3, OPT)                                          \
119QPEL_FUNCS_DECL(OP, 1, 0, OPT)                                          \
120QPEL_FUNCS_DECL(OP, 1, 1, OPT)                                          \
121QPEL_FUNCS_DECL(OP, 1, 2, OPT)                                          \
122QPEL_FUNCS_DECL(OP, 1, 3, OPT)                                          \
123QPEL_FUNCS_DECL(OP, 2, 1, OPT)                                          \
124QPEL_FUNCS_DECL(OP, 2, 2, OPT)                                          \
125QPEL_FUNCS_DECL(OP, 2, 3, OPT)                                          \
126QPEL_FUNCS_DECL(OP, 3, 0, OPT)                                          \
127QPEL_FUNCS_DECL(OP, 3, 1, OPT)                                          \
128QPEL_FUNCS_DECL(OP, 3, 2, OPT)
129/** @} */
130
131#define LOOPSIZE  8
132#define HCOFF(x)  (32 * ((x) - 1))
133#define VCOFF(x)  (32 * ((x) - 1))
134QPEL_MC_DECL(put_, _ssse3)
135QPEL_MC_DECL(avg_, _ssse3)
136
137#undef LOOPSIZE
138#undef HCOFF
139#undef VCOFF
140#define LOOPSIZE  8
141#define HCOFF(x)  (64 * ((x) - 1))
142#define VCOFF(x)  (64 * ((x) - 1))
143QPEL_MC_DECL(put_, _sse2)
144QPEL_MC_DECL(avg_, _sse2)
145
146/** @{ */
147/** Set one function */
148#define QPEL_FUNC_SET(OP, SIZE, PH, PV, OPT)                            \
149    c-> OP ## pixels_tab[2 - SIZE / 8][4 * PV + PH] = OP ## rv40_qpel ##SIZE ## _mc ##PH ##PV ##OPT;
150
151/** Set functions put and avg for sizes 8 and 16 and a given qpel position */
152#define QPEL_FUNCS_SET(OP, PH, PV, OPT)         \
153    QPEL_FUNC_SET(OP,  8, PH, PV, OPT)          \
154    QPEL_FUNC_SET(OP, 16, PH, PV, OPT)
155
156/** Set all functions for all sizes and qpel positions */
157#define QPEL_MC_SET(OP, OPT)   \
158QPEL_FUNCS_SET (OP, 0, 1, OPT) \
159QPEL_FUNCS_SET (OP, 0, 3, OPT) \
160QPEL_FUNCS_SET (OP, 1, 0, OPT) \
161QPEL_FUNCS_SET (OP, 1, 1, OPT) \
162QPEL_FUNCS_SET (OP, 1, 2, OPT) \
163QPEL_FUNCS_SET (OP, 1, 3, OPT) \
164QPEL_FUNCS_SET (OP, 2, 1, OPT) \
165QPEL_FUNCS_SET (OP, 2, 2, OPT) \
166QPEL_FUNCS_SET (OP, 2, 3, OPT) \
167QPEL_FUNCS_SET (OP, 3, 0, OPT) \
168QPEL_FUNCS_SET (OP, 3, 1, OPT) \
169QPEL_FUNCS_SET (OP, 3, 2, OPT)
170/** @} */
171
172DEFINE_FN(put, 8, ssse3)
173
174DEFINE_FN(put, 16, sse2)
175DEFINE_FN(put, 16, ssse3)
176
177DEFINE_FN(avg, 8, mmxext)
178DEFINE_FN(avg, 8, ssse3)
179
180DEFINE_FN(avg, 16, sse2)
181DEFINE_FN(avg, 16, ssse3)
182#endif /* HAVE_X86ASM */
183
184#if HAVE_MMX_INLINE
185DEFINE_FN(put, 8, mmx)
186#endif
187
188av_cold void ff_rv40dsp_init_x86(RV34DSPContext *c)
189{
190    av_unused int cpu_flags = av_get_cpu_flags();
191
192#if HAVE_MMX_INLINE
193    if (INLINE_MMX(cpu_flags)) {
194        c->put_pixels_tab[1][15] = put_rv40_qpel8_mc33_mmx;
195    }
196#endif /* HAVE_MMX_INLINE */
197
198#if HAVE_X86ASM
199    if (EXTERNAL_MMX(cpu_flags)) {
200        c->put_chroma_pixels_tab[0] = ff_put_rv40_chroma_mc8_mmx;
201        c->put_chroma_pixels_tab[1] = ff_put_rv40_chroma_mc4_mmx;
202    }
203    if (EXTERNAL_MMXEXT(cpu_flags)) {
204        c->avg_pixels_tab[1][15]        = avg_rv40_qpel8_mc33_mmxext;
205        c->avg_chroma_pixels_tab[0]     = ff_avg_rv40_chroma_mc8_mmxext;
206        c->avg_chroma_pixels_tab[1]     = ff_avg_rv40_chroma_mc4_mmxext;
207    }
208    if (EXTERNAL_SSE2(cpu_flags)) {
209        c->put_pixels_tab[0][15]        = put_rv40_qpel16_mc33_sse2;
210        c->avg_pixels_tab[0][15]        = avg_rv40_qpel16_mc33_sse2;
211        c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_sse2;
212        c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_sse2;
213        c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_sse2;
214        c->rv40_weight_pixels_tab[1][1] = ff_rv40_weight_func_nornd_8_sse2;
215        QPEL_MC_SET(put_, _sse2)
216        QPEL_MC_SET(avg_, _sse2)
217    }
218    if (EXTERNAL_SSSE3(cpu_flags)) {
219        c->put_pixels_tab[0][15]        = put_rv40_qpel16_mc33_ssse3;
220        c->put_pixels_tab[1][15]        = put_rv40_qpel8_mc33_ssse3;
221        c->avg_pixels_tab[0][15]        = avg_rv40_qpel16_mc33_ssse3;
222        c->avg_pixels_tab[1][15]        = avg_rv40_qpel8_mc33_ssse3;
223        c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_ssse3;
224        c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_ssse3;
225        c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_ssse3;
226        c->rv40_weight_pixels_tab[1][1] = ff_rv40_weight_func_nornd_8_ssse3;
227        QPEL_MC_SET(put_, _ssse3)
228        QPEL_MC_SET(avg_, _ssse3)
229    }
230#endif /* HAVE_X86ASM */
231}
232