| 1 | /* |
| 2 | * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org> |
| 3 | * |
| 4 | * This file is part of FFmpeg. |
| 5 | * |
| 6 | * FFmpeg is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU Lesser General Public |
| 8 | * License as published by the Free Software Foundation; either |
| 9 | * version 2.1 of the License, or (at your option) any later version. |
| 10 | * |
| 11 | * FFmpeg is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 14 | * Lesser General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU Lesser General Public |
| 17 | * License along with FFmpeg; if not, write to the Free Software |
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 19 | */ |
| 20 | |
| 21 | #include "config.h" |
| 22 | #include "libavutil/attributes.h" |
| 23 | #include "libavutil/cpu.h" |
| 24 | #include "libavutil/intreadwrite.h" |
| 25 | #include "libavutil/ppc/cpu.h" |
| 26 | #include "libavutil/ppc/types_altivec.h" |
| 27 | #include "libavutil/ppc/util_altivec.h" |
| 28 | #include "libavcodec/h264qpel.h" |
| 29 | #include "hpeldsp_altivec.h" |
| 30 | |
| 31 | #if HAVE_ALTIVEC |
| 32 | |
| 33 | #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s |
| 34 | #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s) |
| 35 | |
| 36 | #define OP_U8_ALTIVEC PUT_OP_U8_ALTIVEC |
| 37 | #define PREFIX_h264_qpel16_h_lowpass_altivec put_h264_qpel16_h_lowpass_altivec |
| 38 | #define PREFIX_h264_qpel16_h_lowpass_num altivec_put_h264_qpel16_h_lowpass_num |
| 39 | #define PREFIX_h264_qpel16_v_lowpass_altivec put_h264_qpel16_v_lowpass_altivec |
| 40 | #define PREFIX_h264_qpel16_v_lowpass_num altivec_put_h264_qpel16_v_lowpass_num |
| 41 | #define PREFIX_h264_qpel16_hv_lowpass_altivec put_h264_qpel16_hv_lowpass_altivec |
| 42 | #define PREFIX_h264_qpel16_hv_lowpass_num altivec_put_h264_qpel16_hv_lowpass_num |
| 43 | #include "h264qpel_template.c" |
| 44 | #undef OP_U8_ALTIVEC |
| 45 | #undef PREFIX_h264_qpel16_h_lowpass_altivec |
| 46 | #undef PREFIX_h264_qpel16_h_lowpass_num |
| 47 | #undef PREFIX_h264_qpel16_v_lowpass_altivec |
| 48 | #undef PREFIX_h264_qpel16_v_lowpass_num |
| 49 | #undef PREFIX_h264_qpel16_hv_lowpass_altivec |
| 50 | #undef PREFIX_h264_qpel16_hv_lowpass_num |
| 51 | |
| 52 | #define OP_U8_ALTIVEC AVG_OP_U8_ALTIVEC |
| 53 | #define PREFIX_h264_qpel16_h_lowpass_altivec avg_h264_qpel16_h_lowpass_altivec |
| 54 | #define PREFIX_h264_qpel16_h_lowpass_num altivec_avg_h264_qpel16_h_lowpass_num |
| 55 | #define PREFIX_h264_qpel16_v_lowpass_altivec avg_h264_qpel16_v_lowpass_altivec |
| 56 | #define PREFIX_h264_qpel16_v_lowpass_num altivec_avg_h264_qpel16_v_lowpass_num |
| 57 | #define PREFIX_h264_qpel16_hv_lowpass_altivec avg_h264_qpel16_hv_lowpass_altivec |
| 58 | #define PREFIX_h264_qpel16_hv_lowpass_num altivec_avg_h264_qpel16_hv_lowpass_num |
| 59 | #include "h264qpel_template.c" |
| 60 | #undef OP_U8_ALTIVEC |
| 61 | #undef PREFIX_h264_qpel16_h_lowpass_altivec |
| 62 | #undef PREFIX_h264_qpel16_h_lowpass_num |
| 63 | #undef PREFIX_h264_qpel16_v_lowpass_altivec |
| 64 | #undef PREFIX_h264_qpel16_v_lowpass_num |
| 65 | #undef PREFIX_h264_qpel16_hv_lowpass_altivec |
| 66 | #undef PREFIX_h264_qpel16_hv_lowpass_num |
| 67 | |
| 68 | #define H264_MC(OPNAME, SIZE, CODETYPE) \ |
| 69 | static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## CODETYPE (uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 70 | {\ |
| 71 | ff_ ## OPNAME ## pixels ## SIZE ## _ ## CODETYPE(dst, src, stride, SIZE);\ |
| 72 | }\ |
| 73 | \ |
| 74 | static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 75 | { \ |
| 76 | DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\ |
| 77 | put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ |
| 78 | OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\ |
| 79 | }\ |
| 80 | \ |
| 81 | static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 82 | {\ |
| 83 | OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(dst, src, stride, stride);\ |
| 84 | }\ |
| 85 | \ |
| 86 | static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 87 | {\ |
| 88 | DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\ |
| 89 | put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ |
| 90 | OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\ |
| 91 | }\ |
| 92 | \ |
| 93 | static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 94 | {\ |
| 95 | DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\ |
| 96 | put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ |
| 97 | OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\ |
| 98 | }\ |
| 99 | \ |
| 100 | static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 101 | {\ |
| 102 | OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\ |
| 103 | }\ |
| 104 | \ |
| 105 | static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 106 | {\ |
| 107 | DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\ |
| 108 | put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ |
| 109 | OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\ |
| 110 | }\ |
| 111 | \ |
| 112 | static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 113 | {\ |
| 114 | DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ |
| 115 | DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ |
| 116 | put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\ |
| 117 | put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\ |
| 118 | OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ |
| 119 | }\ |
| 120 | \ |
| 121 | static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 122 | {\ |
| 123 | DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ |
| 124 | DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ |
| 125 | put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\ |
| 126 | put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\ |
| 127 | OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ |
| 128 | }\ |
| 129 | \ |
| 130 | static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 131 | {\ |
| 132 | DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ |
| 133 | DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ |
| 134 | put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\ |
| 135 | put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\ |
| 136 | OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ |
| 137 | }\ |
| 138 | \ |
| 139 | static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 140 | {\ |
| 141 | DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ |
| 142 | DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ |
| 143 | put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\ |
| 144 | put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\ |
| 145 | OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ |
| 146 | }\ |
| 147 | \ |
| 148 | static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 149 | {\ |
| 150 | DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ |
| 151 | OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\ |
| 152 | }\ |
| 153 | \ |
| 154 | static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 155 | {\ |
| 156 | DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ |
| 157 | DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\ |
| 158 | DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ |
| 159 | put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\ |
| 160 | put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ |
| 161 | OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\ |
| 162 | }\ |
| 163 | \ |
| 164 | static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 165 | {\ |
| 166 | DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ |
| 167 | DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\ |
| 168 | DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ |
| 169 | put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\ |
| 170 | put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ |
| 171 | OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\ |
| 172 | }\ |
| 173 | \ |
| 174 | static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 175 | {\ |
| 176 | DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ |
| 177 | DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\ |
| 178 | DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ |
| 179 | put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\ |
| 180 | put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ |
| 181 | OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\ |
| 182 | }\ |
| 183 | \ |
| 184 | static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 185 | {\ |
| 186 | DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ |
| 187 | DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\ |
| 188 | DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ |
| 189 | put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\ |
| 190 | put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ |
| 191 | OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\ |
| 192 | }\ |
| 193 | |
| 194 | static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1, |
| 195 | const uint8_t * src2, int dst_stride, |
| 196 | int src_stride1, int h) |
| 197 | { |
| 198 | int i; |
| 199 | vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align; |
| 200 | |
| 201 | mask_ = vec_lvsl(0, src2); |
| 202 | |
| 203 | for (i = 0; i < h; i++) { |
| 204 | |
| 205 | tmp1 = vec_ld(i * src_stride1, src1); |
| 206 | mask = vec_lvsl(i * src_stride1, src1); |
| 207 | tmp2 = vec_ld(i * src_stride1 + 15, src1); |
| 208 | |
| 209 | a = vec_perm(tmp1, tmp2, mask); |
| 210 | |
| 211 | tmp1 = vec_ld(i * 16, src2); |
| 212 | tmp2 = vec_ld(i * 16 + 15, src2); |
| 213 | |
| 214 | b = vec_perm(tmp1, tmp2, mask_); |
| 215 | |
| 216 | tmp1 = vec_ld(0, dst); |
| 217 | mask = vec_lvsl(0, dst); |
| 218 | tmp2 = vec_ld(15, dst); |
| 219 | |
| 220 | d = vec_avg(a, b); |
| 221 | |
| 222 | edges = vec_perm(tmp2, tmp1, mask); |
| 223 | |
| 224 | align = vec_lvsr(0, dst); |
| 225 | |
| 226 | tmp2 = vec_perm(d, edges, align); |
| 227 | tmp1 = vec_perm(edges, d, align); |
| 228 | |
| 229 | vec_st(tmp2, 15, dst); |
| 230 | vec_st(tmp1, 0 , dst); |
| 231 | |
| 232 | dst += dst_stride; |
| 233 | } |
| 234 | } |
| 235 | |
| 236 | static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1, |
| 237 | const uint8_t * src2, int dst_stride, |
| 238 | int src_stride1, int h) |
| 239 | { |
| 240 | int i; |
| 241 | vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align; |
| 242 | |
| 243 | mask_ = vec_lvsl(0, src2); |
| 244 | |
| 245 | for (i = 0; i < h; i++) { |
| 246 | |
| 247 | tmp1 = vec_ld(i * src_stride1, src1); |
| 248 | mask = vec_lvsl(i * src_stride1, src1); |
| 249 | tmp2 = vec_ld(i * src_stride1 + 15, src1); |
| 250 | |
| 251 | a = vec_perm(tmp1, tmp2, mask); |
| 252 | |
| 253 | tmp1 = vec_ld(i * 16, src2); |
| 254 | tmp2 = vec_ld(i * 16 + 15, src2); |
| 255 | |
| 256 | b = vec_perm(tmp1, tmp2, mask_); |
| 257 | |
| 258 | tmp1 = vec_ld(0, dst); |
| 259 | mask = vec_lvsl(0, dst); |
| 260 | tmp2 = vec_ld(15, dst); |
| 261 | |
| 262 | d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b)); |
| 263 | |
| 264 | edges = vec_perm(tmp2, tmp1, mask); |
| 265 | |
| 266 | align = vec_lvsr(0, dst); |
| 267 | |
| 268 | tmp2 = vec_perm(d, edges, align); |
| 269 | tmp1 = vec_perm(edges, d, align); |
| 270 | |
| 271 | vec_st(tmp2, 15, dst); |
| 272 | vec_st(tmp1, 0 , dst); |
| 273 | |
| 274 | dst += dst_stride; |
| 275 | } |
| 276 | } |
| 277 | |
| 278 | /* Implemented but could be faster |
| 279 | #define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h) |
| 280 | #define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h) |
| 281 | */ |
| 282 | |
| 283 | H264_MC(put_, 16, altivec) |
| 284 | H264_MC(avg_, 16, altivec) |
| 285 | #endif /* HAVE_ALTIVEC */ |
| 286 | |
| 287 | av_cold void ff_h264qpel_init_ppc(H264QpelContext *c, int bit_depth) |
| 288 | { |
| 289 | #if HAVE_ALTIVEC |
| 290 | const int high_bit_depth = bit_depth > 8; |
| 291 | |
| 292 | if (!PPC_ALTIVEC(av_get_cpu_flags())) |
| 293 | return; |
| 294 | |
| 295 | if (!high_bit_depth) { |
| 296 | #define dspfunc(PFX, IDX, NUM) \ |
| 297 | c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \ |
| 298 | c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \ |
| 299 | c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \ |
| 300 | c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \ |
| 301 | c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \ |
| 302 | c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \ |
| 303 | c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \ |
| 304 | c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \ |
| 305 | c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \ |
| 306 | c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \ |
| 307 | c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \ |
| 308 | c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \ |
| 309 | c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \ |
| 310 | c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \ |
| 311 | c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \ |
| 312 | c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec |
| 313 | |
| 314 | dspfunc(put_h264_qpel, 0, 16); |
| 315 | dspfunc(avg_h264_qpel, 0, 16); |
| 316 | #undef dspfunc |
| 317 | } |
| 318 | #endif /* HAVE_ALTIVEC */ |
| 319 | } |