FFmpeg: libavcodec/x86/rv40dsp

00001 /*
00002  * RV40 decoder motion compensation functions x86-optimised
00003  * Copyright (c) 2008 Konstantin Shishkov
00004  *
00005  * This file is part of FFmpeg.
00006  *
00007  * FFmpeg is free software; you can redistribute it and/or
00008  * modify it under the terms of the GNU Lesser General Public
00009  * License as published by the Free Software Foundation; either
00010  * version 2.1 of the License, or (at your option) any later version.
00011  *
00012  * FFmpeg is distributed in the hope that it will be useful,
00013  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00014  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00015  * Lesser General Public License for more details.
00016  *
00017  * You should have received a copy of the GNU Lesser General Public
00018  * License along with FFmpeg; if not, write to the Free Software
00019  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00020  */
00021 
00029 #include "libavcodec/rv34dsp.h"
00030 #include "libavutil/mem.h"
00031 #include "libavutil/x86/cpu.h"
00032 #include "dsputil_mmx.h"
00033 
00034 #if HAVE_YASM
00035 void ff_put_rv40_chroma_mc8_mmx  (uint8_t *dst, uint8_t *src,
00036                                   int stride, int h, int x, int y);
00037 void ff_avg_rv40_chroma_mc8_mmx2 (uint8_t *dst, uint8_t *src,
00038                                   int stride, int h, int x, int y);
00039 void ff_avg_rv40_chroma_mc8_3dnow(uint8_t *dst, uint8_t *src,
00040                                   int stride, int h, int x, int y);
00041 
00042 void ff_put_rv40_chroma_mc4_mmx  (uint8_t *dst, uint8_t *src,
00043                                   int stride, int h, int x, int y);
00044 void ff_avg_rv40_chroma_mc4_mmx2 (uint8_t *dst, uint8_t *src,
00045                                   int stride, int h, int x, int y);
00046 void ff_avg_rv40_chroma_mc4_3dnow(uint8_t *dst, uint8_t *src,
00047                                   int stride, int h, int x, int y);
00048 
00049 #define DECLARE_WEIGHT(opt) \
00050 void ff_rv40_weight_func_rnd_16_##opt(uint8_t *dst, uint8_t *src1, uint8_t *src2, \
00051                                       int w1, int w2, ptrdiff_t stride); \
00052 void ff_rv40_weight_func_rnd_8_##opt (uint8_t *dst, uint8_t *src1, uint8_t *src2, \
00053                                       int w1, int w2, ptrdiff_t stride); \
00054 void ff_rv40_weight_func_nornd_16_##opt(uint8_t *dst, uint8_t *src1, uint8_t *src2, \
00055                                         int w1, int w2, ptrdiff_t stride); \
00056 void ff_rv40_weight_func_nornd_8_##opt (uint8_t *dst, uint8_t *src1, uint8_t *src2, \
00057                                         int w1, int w2, ptrdiff_t stride);
00058 DECLARE_WEIGHT(mmx2)
00059 DECLARE_WEIGHT(sse2)
00060 DECLARE_WEIGHT(ssse3)
00061 
00070 #define QPEL_FUNC_DECL(OP, SIZE, PH, PV, OPT)                           \
00071 static void OP ## rv40_qpel ##SIZE ##_mc ##PH ##PV ##OPT(uint8_t *dst,  \
00072                                                          uint8_t *src,  \
00073                                                          int stride)    \
00074 {                                                                       \
00075     int i;                                                              \
00076     if (PH && PV) {                                                     \
00077         DECLARE_ALIGNED(16, uint8_t, tmp)[SIZE * (SIZE + 5)];           \
00078         uint8_t *tmpptr = tmp + SIZE * 2;                               \
00079         src -= stride * 2;                                              \
00080                                                                         \
00081         for (i = 0; i < SIZE; i += LOOPSIZE)                            \
00082             ff_put_rv40_qpel_h ##OPT(tmp + i, SIZE, src + i, stride,    \
00083                                      SIZE + 5, HCOFF(PH));              \
00084         for (i = 0; i < SIZE; i += LOOPSIZE)                            \
00085             ff_ ##OP ##rv40_qpel_v ##OPT(dst + i, stride, tmpptr + i,   \
00086                                          SIZE, SIZE, VCOFF(PV));        \
00087     } else if (PV) {                                                    \
00088         for (i = 0; i < SIZE; i += LOOPSIZE)                            \
00089             ff_ ##OP ##rv40_qpel_v ## OPT(dst + i, stride, src + i,     \
00090                                           stride, SIZE, VCOFF(PV));     \
00091     } else {                                                            \
00092         for (i = 0; i < SIZE; i += LOOPSIZE)                            \
00093             ff_ ##OP ##rv40_qpel_h ## OPT(dst + i, stride, src + i,     \
00094                                           stride, SIZE, HCOFF(PH));     \
00095     }                                                                   \
00096 };
00097 
00100 #define QPEL_FUNCS_DECL(OP, PH, PV, OPT) \
00101     QPEL_FUNC_DECL(OP,  8, PH, PV, OPT)  \
00102     QPEL_FUNC_DECL(OP, 16, PH, PV, OPT)
00103 
00105 #define QPEL_MC_DECL(OP, OPT)                                           \
00106 void ff_ ##OP ##rv40_qpel_h ##OPT(uint8_t *dst, ptrdiff_t dstStride,    \
00107                                   const uint8_t *src,                   \
00108                                   ptrdiff_t srcStride,                  \
00109                                   int len, int m);                      \
00110 void ff_ ##OP ##rv40_qpel_v ##OPT(uint8_t *dst, ptrdiff_t dstStride,    \
00111                                   const uint8_t *src,                   \
00112                                   ptrdiff_t srcStride,                  \
00113                                   int len, int m);                      \
00114 QPEL_FUNCS_DECL(OP, 0, 1, OPT)                                          \
00115 QPEL_FUNCS_DECL(OP, 0, 3, OPT)                                          \
00116 QPEL_FUNCS_DECL(OP, 1, 0, OPT)                                          \
00117 QPEL_FUNCS_DECL(OP, 1, 1, OPT)                                          \
00118 QPEL_FUNCS_DECL(OP, 1, 2, OPT)                                          \
00119 QPEL_FUNCS_DECL(OP, 1, 3, OPT)                                          \
00120 QPEL_FUNCS_DECL(OP, 2, 1, OPT)                                          \
00121 QPEL_FUNCS_DECL(OP, 2, 2, OPT)                                          \
00122 QPEL_FUNCS_DECL(OP, 2, 3, OPT)                                          \
00123 QPEL_FUNCS_DECL(OP, 3, 0, OPT)                                          \
00124 QPEL_FUNCS_DECL(OP, 3, 1, OPT)                                          \
00125 QPEL_FUNCS_DECL(OP, 3, 2, OPT)
00126 
00128 #define LOOPSIZE  8
00129 #define HCOFF(x)  (32 * (x - 1))
00130 #define VCOFF(x)  (32 * (x - 1))
00131 QPEL_MC_DECL(put_, _ssse3)
00132 QPEL_MC_DECL(avg_, _ssse3)
00133 
00134 #undef LOOPSIZE
00135 #undef HCOFF
00136 #undef VCOFF
00137 #define LOOPSIZE  8
00138 #define HCOFF(x)  (64 * (x - 1))
00139 #define VCOFF(x)  (64 * (x - 1))
00140 QPEL_MC_DECL(put_, _sse2)
00141 QPEL_MC_DECL(avg_, _sse2)
00142 
00143 #if ARCH_X86_32
00144 #undef LOOPSIZE
00145 #undef HCOFF
00146 #undef VCOFF
00147 #define LOOPSIZE  4
00148 #define HCOFF(x)  (64 * (x - 1))
00149 #define VCOFF(x)  (64 * (x - 1))
00150 
00151 QPEL_MC_DECL(put_, _mmx)
00152 
00153 #define ff_put_rv40_qpel_h_mmx2  ff_put_rv40_qpel_h_mmx
00154 #define ff_put_rv40_qpel_v_mmx2  ff_put_rv40_qpel_v_mmx
00155 QPEL_MC_DECL(avg_, _mmx2)
00156 
00157 #define ff_put_rv40_qpel_h_3dnow  ff_put_rv40_qpel_h_mmx
00158 #define ff_put_rv40_qpel_v_3dnow  ff_put_rv40_qpel_v_mmx
00159 QPEL_MC_DECL(avg_, _3dnow)
00160 #endif
00161 
00164 #define QPEL_FUNC_SET(OP, SIZE, PH, PV, OPT)                            \
00165     c-> OP ## pixels_tab[2 - SIZE / 8][4 * PV + PH] = OP ## rv40_qpel ##SIZE ## _mc ##PH ##PV ##OPT;
00166 
00168 #define QPEL_FUNCS_SET(OP, PH, PV, OPT)         \
00169     QPEL_FUNC_SET(OP,  8, PH, PV, OPT)          \
00170     QPEL_FUNC_SET(OP, 16, PH, PV, OPT)
00171 
00173 #define QPEL_MC_SET(OP, OPT)   \
00174 QPEL_FUNCS_SET (OP, 0, 1, OPT) \
00175 QPEL_FUNCS_SET (OP, 0, 3, OPT) \
00176 QPEL_FUNCS_SET (OP, 1, 0, OPT) \
00177 QPEL_FUNCS_SET (OP, 1, 1, OPT) \
00178 QPEL_FUNCS_SET (OP, 1, 2, OPT) \
00179 QPEL_FUNCS_SET (OP, 1, 3, OPT) \
00180 QPEL_FUNCS_SET (OP, 2, 1, OPT) \
00181 QPEL_FUNCS_SET (OP, 2, 2, OPT) \
00182 QPEL_FUNCS_SET (OP, 2, 3, OPT) \
00183 QPEL_FUNCS_SET (OP, 3, 0, OPT) \
00184 QPEL_FUNCS_SET (OP, 3, 1, OPT) \
00185 QPEL_FUNCS_SET (OP, 3, 2, OPT)
00186 
00188 #endif /* HAVE_YASM */
00189 
00190 void ff_rv40dsp_init_x86(RV34DSPContext *c, DSPContext *dsp)
00191 {
00192 #if HAVE_YASM
00193     int mm_flags = av_get_cpu_flags();
00194 
00195     if (EXTERNAL_MMX(mm_flags)) {
00196         c->put_chroma_pixels_tab[0] = ff_put_rv40_chroma_mc8_mmx;
00197         c->put_chroma_pixels_tab[1] = ff_put_rv40_chroma_mc4_mmx;
00198 #if HAVE_INLINE_ASM
00199         c->put_pixels_tab[0][15] = ff_put_rv40_qpel16_mc33_mmx;
00200         c->put_pixels_tab[1][15] = ff_put_rv40_qpel8_mc33_mmx;
00201         c->avg_pixels_tab[0][15] = ff_avg_rv40_qpel16_mc33_mmx;
00202         c->avg_pixels_tab[1][15] = ff_avg_rv40_qpel8_mc33_mmx;
00203 #endif /* HAVE_INLINE_ASM */
00204 #if ARCH_X86_32
00205         QPEL_MC_SET(put_, _mmx)
00206 #endif
00207     }
00208     if (EXTERNAL_MMXEXT(mm_flags)) {
00209         c->avg_chroma_pixels_tab[0] = ff_avg_rv40_chroma_mc8_mmx2;
00210         c->avg_chroma_pixels_tab[1] = ff_avg_rv40_chroma_mc4_mmx2;
00211         c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_mmx2;
00212         c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_mmx2;
00213         c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_mmx2;
00214         c->rv40_weight_pixels_tab[1][1] = ff_rv40_weight_func_nornd_8_mmx2;
00215 #if ARCH_X86_32
00216         QPEL_MC_SET(avg_, _mmx2)
00217 #endif
00218     } else if (EXTERNAL_AMD3DNOW(mm_flags)) {
00219         c->avg_chroma_pixels_tab[0] = ff_avg_rv40_chroma_mc8_3dnow;
00220         c->avg_chroma_pixels_tab[1] = ff_avg_rv40_chroma_mc4_3dnow;
00221 #if ARCH_X86_32
00222         QPEL_MC_SET(avg_, _3dnow)
00223 #endif
00224     }
00225     if (EXTERNAL_SSE2(mm_flags)) {
00226         c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_sse2;
00227         c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_sse2;
00228         c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_sse2;
00229         c->rv40_weight_pixels_tab[1][1] = ff_rv40_weight_func_nornd_8_sse2;
00230         QPEL_MC_SET(put_, _sse2)
00231         QPEL_MC_SET(avg_, _sse2)
00232     }
00233     if (EXTERNAL_SSSE3(mm_flags)) {
00234         c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_ssse3;
00235         c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_ssse3;
00236         c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_ssse3;
00237         c->rv40_weight_pixels_tab[1][1] = ff_rv40_weight_func_nornd_8_ssse3;
00238         QPEL_MC_SET(put_, _ssse3)
00239         QPEL_MC_SET(avg_, _ssse3)
00240     }
00241 #endif /* HAVE_YASM */
00242 }
libavcodec/x86/rv40dsp_init.c
