| 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.h" |
| 32 | #include "libavutil/x86/cpu.h" |
| 33 | #include "hpeldsp.h" |
| 34 | |
| 35 | #define DEFINE_FN(op, size, insn) \ |
| 36 | static 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_YASM |
| 43 | void ff_put_rv40_chroma_mc8_mmx (uint8_t *dst, uint8_t *src, |
| 44 | int stride, int h, int x, int y); |
| 45 | void ff_avg_rv40_chroma_mc8_mmxext(uint8_t *dst, uint8_t *src, |
| 46 | int stride, int h, int x, int y); |
| 47 | void ff_avg_rv40_chroma_mc8_3dnow(uint8_t *dst, uint8_t *src, |
| 48 | int stride, int h, int x, int y); |
| 49 | |
| 50 | void ff_put_rv40_chroma_mc4_mmx (uint8_t *dst, uint8_t *src, |
| 51 | int stride, int h, int x, int y); |
| 52 | void ff_avg_rv40_chroma_mc4_mmxext(uint8_t *dst, uint8_t *src, |
| 53 | int stride, int h, int x, int y); |
| 54 | void ff_avg_rv40_chroma_mc4_3dnow(uint8_t *dst, uint8_t *src, |
| 55 | int stride, int h, int x, int y); |
| 56 | |
| 57 | #define DECLARE_WEIGHT(opt) \ |
| 58 | void ff_rv40_weight_func_rnd_16_##opt(uint8_t *dst, uint8_t *src1, uint8_t *src2, \ |
| 59 | int w1, int w2, ptrdiff_t stride); \ |
| 60 | void ff_rv40_weight_func_rnd_8_##opt (uint8_t *dst, uint8_t *src1, uint8_t *src2, \ |
| 61 | int w1, int w2, ptrdiff_t stride); \ |
| 62 | void ff_rv40_weight_func_nornd_16_##opt(uint8_t *dst, uint8_t *src1, uint8_t *src2, \ |
| 63 | int w1, int w2, ptrdiff_t stride); \ |
| 64 | void ff_rv40_weight_func_nornd_8_##opt (uint8_t *dst, uint8_t *src1, uint8_t *src2, \ |
| 65 | int w1, int w2, ptrdiff_t stride); |
| 66 | DECLARE_WEIGHT(mmxext) |
| 67 | DECLARE_WEIGHT(sse2) |
| 68 | DECLARE_WEIGHT(ssse3) |
| 69 | |
| 70 | /** @{ */ |
| 71 | /** |
| 72 | * Define one qpel function. |
| 73 | * LOOPSIZE must be already set to the number of pixels processed per |
| 74 | * iteration in the inner loop of the called functions. |
| 75 | * COFF(x) must be already defined so as to provide the offset into any |
| 76 | * array of coeffs used by the called function for the qpel position x. |
| 77 | */ |
| 78 | #define QPEL_FUNC_DECL(OP, SIZE, PH, PV, OPT) \ |
| 79 | static void OP ## rv40_qpel ##SIZE ##_mc ##PH ##PV ##OPT(uint8_t *dst, \ |
| 80 | const uint8_t *src, \ |
| 81 | ptrdiff_t stride) \ |
| 82 | { \ |
| 83 | int i; \ |
| 84 | if (PH && PV) { \ |
| 85 | DECLARE_ALIGNED(16, uint8_t, tmp)[SIZE * (SIZE + 5)]; \ |
| 86 | uint8_t *tmpptr = tmp + SIZE * 2; \ |
| 87 | src -= stride * 2; \ |
| 88 | \ |
| 89 | for (i = 0; i < SIZE; i += LOOPSIZE) \ |
| 90 | ff_put_rv40_qpel_h ##OPT(tmp + i, SIZE, src + i, stride, \ |
| 91 | SIZE + 5, HCOFF(PH)); \ |
| 92 | for (i = 0; i < SIZE; i += LOOPSIZE) \ |
| 93 | ff_ ##OP ##rv40_qpel_v ##OPT(dst + i, stride, tmpptr + i, \ |
| 94 | SIZE, SIZE, VCOFF(PV)); \ |
| 95 | } else if (PV) { \ |
| 96 | for (i = 0; i < SIZE; i += LOOPSIZE) \ |
| 97 | ff_ ##OP ##rv40_qpel_v ## OPT(dst + i, stride, src + i, \ |
| 98 | stride, SIZE, VCOFF(PV)); \ |
| 99 | } else { \ |
| 100 | for (i = 0; i < SIZE; i += LOOPSIZE) \ |
| 101 | ff_ ##OP ##rv40_qpel_h ## OPT(dst + i, stride, src + i, \ |
| 102 | stride, SIZE, HCOFF(PH)); \ |
| 103 | } \ |
| 104 | }; |
| 105 | |
| 106 | /** Declare functions for sizes 8 and 16 and given operations |
| 107 | * and qpel position. */ |
| 108 | #define QPEL_FUNCS_DECL(OP, PH, PV, OPT) \ |
| 109 | QPEL_FUNC_DECL(OP, 8, PH, PV, OPT) \ |
| 110 | QPEL_FUNC_DECL(OP, 16, PH, PV, OPT) |
| 111 | |
| 112 | /** Declare all functions for all sizes and qpel positions */ |
| 113 | #define QPEL_MC_DECL(OP, OPT) \ |
| 114 | void ff_ ##OP ##rv40_qpel_h ##OPT(uint8_t *dst, ptrdiff_t dstStride, \ |
| 115 | const uint8_t *src, \ |
| 116 | ptrdiff_t srcStride, \ |
| 117 | int len, int m); \ |
| 118 | void ff_ ##OP ##rv40_qpel_v ##OPT(uint8_t *dst, ptrdiff_t dstStride, \ |
| 119 | const uint8_t *src, \ |
| 120 | ptrdiff_t srcStride, \ |
| 121 | int len, int m); \ |
| 122 | QPEL_FUNCS_DECL(OP, 0, 1, OPT) \ |
| 123 | QPEL_FUNCS_DECL(OP, 0, 3, OPT) \ |
| 124 | QPEL_FUNCS_DECL(OP, 1, 0, OPT) \ |
| 125 | QPEL_FUNCS_DECL(OP, 1, 1, OPT) \ |
| 126 | QPEL_FUNCS_DECL(OP, 1, 2, OPT) \ |
| 127 | QPEL_FUNCS_DECL(OP, 1, 3, OPT) \ |
| 128 | QPEL_FUNCS_DECL(OP, 2, 1, OPT) \ |
| 129 | QPEL_FUNCS_DECL(OP, 2, 2, OPT) \ |
| 130 | QPEL_FUNCS_DECL(OP, 2, 3, OPT) \ |
| 131 | QPEL_FUNCS_DECL(OP, 3, 0, OPT) \ |
| 132 | QPEL_FUNCS_DECL(OP, 3, 1, OPT) \ |
| 133 | QPEL_FUNCS_DECL(OP, 3, 2, OPT) |
| 134 | /** @} */ |
| 135 | |
| 136 | #define LOOPSIZE 8 |
| 137 | #define HCOFF(x) (32 * ((x) - 1)) |
| 138 | #define VCOFF(x) (32 * ((x) - 1)) |
| 139 | QPEL_MC_DECL(put_, _ssse3) |
| 140 | QPEL_MC_DECL(avg_, _ssse3) |
| 141 | |
| 142 | #undef LOOPSIZE |
| 143 | #undef HCOFF |
| 144 | #undef VCOFF |
| 145 | #define LOOPSIZE 8 |
| 146 | #define HCOFF(x) (64 * ((x) - 1)) |
| 147 | #define VCOFF(x) (64 * ((x) - 1)) |
| 148 | QPEL_MC_DECL(put_, _sse2) |
| 149 | QPEL_MC_DECL(avg_, _sse2) |
| 150 | |
| 151 | #if ARCH_X86_32 |
| 152 | #undef LOOPSIZE |
| 153 | #undef HCOFF |
| 154 | #undef VCOFF |
| 155 | #define LOOPSIZE 4 |
| 156 | #define HCOFF(x) (64 * ((x) - 1)) |
| 157 | #define VCOFF(x) (64 * ((x) - 1)) |
| 158 | |
| 159 | QPEL_MC_DECL(put_, _mmx) |
| 160 | |
| 161 | #define ff_put_rv40_qpel_h_mmxext ff_put_rv40_qpel_h_mmx |
| 162 | #define ff_put_rv40_qpel_v_mmxext ff_put_rv40_qpel_v_mmx |
| 163 | QPEL_MC_DECL(avg_, _mmxext) |
| 164 | |
| 165 | #define ff_put_rv40_qpel_h_3dnow ff_put_rv40_qpel_h_mmx |
| 166 | #define ff_put_rv40_qpel_v_3dnow ff_put_rv40_qpel_v_mmx |
| 167 | QPEL_MC_DECL(avg_, _3dnow) |
| 168 | #endif |
| 169 | |
| 170 | /** @{ */ |
| 171 | /** Set one function */ |
| 172 | #define QPEL_FUNC_SET(OP, SIZE, PH, PV, OPT) \ |
| 173 | c-> OP ## pixels_tab[2 - SIZE / 8][4 * PV + PH] = OP ## rv40_qpel ##SIZE ## _mc ##PH ##PV ##OPT; |
| 174 | |
| 175 | /** Set functions put and avg for sizes 8 and 16 and a given qpel position */ |
| 176 | #define QPEL_FUNCS_SET(OP, PH, PV, OPT) \ |
| 177 | QPEL_FUNC_SET(OP, 8, PH, PV, OPT) \ |
| 178 | QPEL_FUNC_SET(OP, 16, PH, PV, OPT) |
| 179 | |
| 180 | /** Set all functions for all sizes and qpel positions */ |
| 181 | #define QPEL_MC_SET(OP, OPT) \ |
| 182 | QPEL_FUNCS_SET (OP, 0, 1, OPT) \ |
| 183 | QPEL_FUNCS_SET (OP, 0, 3, OPT) \ |
| 184 | QPEL_FUNCS_SET (OP, 1, 0, OPT) \ |
| 185 | QPEL_FUNCS_SET (OP, 1, 1, OPT) \ |
| 186 | QPEL_FUNCS_SET (OP, 1, 2, OPT) \ |
| 187 | QPEL_FUNCS_SET (OP, 1, 3, OPT) \ |
| 188 | QPEL_FUNCS_SET (OP, 2, 1, OPT) \ |
| 189 | QPEL_FUNCS_SET (OP, 2, 2, OPT) \ |
| 190 | QPEL_FUNCS_SET (OP, 2, 3, OPT) \ |
| 191 | QPEL_FUNCS_SET (OP, 3, 0, OPT) \ |
| 192 | QPEL_FUNCS_SET (OP, 3, 1, OPT) \ |
| 193 | QPEL_FUNCS_SET (OP, 3, 2, OPT) |
| 194 | /** @} */ |
| 195 | |
| 196 | DEFINE_FN(put, 8, ssse3) |
| 197 | |
| 198 | DEFINE_FN(put, 16, sse2) |
| 199 | DEFINE_FN(put, 16, ssse3) |
| 200 | |
| 201 | DEFINE_FN(avg, 8, mmxext) |
| 202 | DEFINE_FN(avg, 8, ssse3) |
| 203 | |
| 204 | DEFINE_FN(avg, 16, sse2) |
| 205 | DEFINE_FN(avg, 16, ssse3) |
| 206 | #endif /* HAVE_YASM */ |
| 207 | |
| 208 | #if HAVE_MMX_INLINE |
| 209 | DEFINE_FN(put, 8, mmx) |
| 210 | DEFINE_FN(avg, 8, mmx) |
| 211 | DEFINE_FN(put, 16, mmx) |
| 212 | DEFINE_FN(avg, 16, mmx) |
| 213 | #endif |
| 214 | |
| 215 | av_cold void ff_rv40dsp_init_x86(RV34DSPContext *c) |
| 216 | { |
| 217 | int cpu_flags = av_get_cpu_flags(); |
| 218 | |
| 219 | #if HAVE_MMX_INLINE |
| 220 | if (INLINE_MMX(cpu_flags)) { |
| 221 | c->put_pixels_tab[0][15] = put_rv40_qpel16_mc33_mmx; |
| 222 | c->put_pixels_tab[1][15] = put_rv40_qpel8_mc33_mmx; |
| 223 | c->avg_pixels_tab[0][15] = avg_rv40_qpel16_mc33_mmx; |
| 224 | c->avg_pixels_tab[1][15] = avg_rv40_qpel8_mc33_mmx; |
| 225 | } |
| 226 | #endif /* HAVE_MMX_INLINE */ |
| 227 | |
| 228 | #if HAVE_YASM |
| 229 | if (EXTERNAL_MMX(cpu_flags)) { |
| 230 | c->put_chroma_pixels_tab[0] = ff_put_rv40_chroma_mc8_mmx; |
| 231 | c->put_chroma_pixels_tab[1] = ff_put_rv40_chroma_mc4_mmx; |
| 232 | #if ARCH_X86_32 |
| 233 | QPEL_MC_SET(put_, _mmx) |
| 234 | #endif |
| 235 | } |
| 236 | if (EXTERNAL_AMD3DNOW(cpu_flags)) { |
| 237 | c->avg_chroma_pixels_tab[0] = ff_avg_rv40_chroma_mc8_3dnow; |
| 238 | c->avg_chroma_pixels_tab[1] = ff_avg_rv40_chroma_mc4_3dnow; |
| 239 | #if ARCH_X86_32 |
| 240 | QPEL_MC_SET(avg_, _3dnow) |
| 241 | #endif |
| 242 | } |
| 243 | if (EXTERNAL_MMXEXT(cpu_flags)) { |
| 244 | c->avg_pixels_tab[1][15] = avg_rv40_qpel8_mc33_mmxext; |
| 245 | c->avg_chroma_pixels_tab[0] = ff_avg_rv40_chroma_mc8_mmxext; |
| 246 | c->avg_chroma_pixels_tab[1] = ff_avg_rv40_chroma_mc4_mmxext; |
| 247 | c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_mmxext; |
| 248 | c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_mmxext; |
| 249 | c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_mmxext; |
| 250 | c->rv40_weight_pixels_tab[1][1] = ff_rv40_weight_func_nornd_8_mmxext; |
| 251 | #if ARCH_X86_32 |
| 252 | QPEL_MC_SET(avg_, _mmxext) |
| 253 | #endif |
| 254 | } |
| 255 | if (EXTERNAL_SSE2(cpu_flags)) { |
| 256 | c->put_pixels_tab[0][15] = put_rv40_qpel16_mc33_sse2; |
| 257 | c->avg_pixels_tab[0][15] = avg_rv40_qpel16_mc33_sse2; |
| 258 | c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_sse2; |
| 259 | c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_sse2; |
| 260 | c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_sse2; |
| 261 | c->rv40_weight_pixels_tab[1][1] = ff_rv40_weight_func_nornd_8_sse2; |
| 262 | QPEL_MC_SET(put_, _sse2) |
| 263 | QPEL_MC_SET(avg_, _sse2) |
| 264 | } |
| 265 | if (EXTERNAL_SSSE3(cpu_flags)) { |
| 266 | c->put_pixels_tab[0][15] = put_rv40_qpel16_mc33_ssse3; |
| 267 | c->put_pixels_tab[1][15] = put_rv40_qpel8_mc33_ssse3; |
| 268 | c->avg_pixels_tab[0][15] = avg_rv40_qpel16_mc33_ssse3; |
| 269 | c->avg_pixels_tab[1][15] = avg_rv40_qpel8_mc33_ssse3; |
| 270 | c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_ssse3; |
| 271 | c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_ssse3; |
| 272 | c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_ssse3; |
| 273 | c->rv40_weight_pixels_tab[1][1] = ff_rv40_weight_func_nornd_8_ssse3; |
| 274 | QPEL_MC_SET(put_, _ssse3) |
| 275 | QPEL_MC_SET(avg_, _ssse3) |
| 276 | } |
| 277 | #endif /* HAVE_YASM */ |
| 278 | } |