| 1 | /* |
| 2 | * Copyright (c) 2002 Brian Foley |
| 3 | * Copyright (c) 2002 Dieter Shirley |
| 4 | * Copyright (c) 2003-2004 Romain Dolbeau <romain@dolbeau.org> |
| 5 | * |
| 6 | * This file is part of FFmpeg. |
| 7 | * |
| 8 | * FFmpeg is free software; you can redistribute it and/or |
| 9 | * modify it under the terms of the GNU Lesser General Public |
| 10 | * License as published by the Free Software Foundation; either |
| 11 | * version 2.1 of the License, or (at your option) any later version. |
| 12 | * |
| 13 | * FFmpeg is distributed in the hope that it will be useful, |
| 14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 16 | * Lesser General Public License for more details. |
| 17 | * |
| 18 | * You should have received a copy of the GNU Lesser General Public |
| 19 | * License along with FFmpeg; if not, write to the Free Software |
| 20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 21 | */ |
| 22 | |
| 23 | #include "config.h" |
| 24 | #if HAVE_ALTIVEC_H |
| 25 | #include <altivec.h> |
| 26 | #endif |
| 27 | |
| 28 | #include "libavutil/attributes.h" |
| 29 | #include "libavutil/cpu.h" |
| 30 | #include "libavutil/ppc/cpu.h" |
| 31 | #include "libavutil/ppc/types_altivec.h" |
| 32 | #include "libavutil/ppc/util_altivec.h" |
| 33 | #include "libavcodec/avcodec.h" |
| 34 | #include "libavcodec/pixblockdsp.h" |
| 35 | |
| 36 | #if HAVE_ALTIVEC |
| 37 | |
| 38 | #if HAVE_VSX |
| 39 | static void get_pixels_altivec(int16_t *restrict block, const uint8_t *pixels, |
| 40 | ptrdiff_t line_size) |
| 41 | { |
| 42 | int i; |
| 43 | vector unsigned char perm = |
| 44 | (vector unsigned char) {0x00,0x10, 0x01,0x11,0x02,0x12,0x03,0x13,\ |
| 45 | 0x04,0x14,0x05,0x15,0x06,0x16,0x07,0x17}; |
| 46 | const vector unsigned char zero = |
| 47 | (const vector unsigned char) vec_splat_u8(0); |
| 48 | |
| 49 | for (i = 0; i < 8; i++) { |
| 50 | /* Read potentially unaligned pixels. |
| 51 | * We're reading 16 pixels, and actually only want 8, |
| 52 | * but we simply ignore the extras. */ |
| 53 | vector unsigned char bytes = vec_vsx_ld(0, pixels); |
| 54 | |
| 55 | // Convert the bytes into shorts. |
| 56 | //vector signed short shorts = (vector signed short) vec_perm(zero, bytes, perm); |
| 57 | vector signed short shorts = (vector signed short) vec_perm(bytes, zero, perm); |
| 58 | |
| 59 | // Save the data to the block, we assume the block is 16-byte aligned. |
| 60 | vec_vsx_st(shorts, i * 16, (vector signed short *) block); |
| 61 | |
| 62 | pixels += line_size; |
| 63 | } |
| 64 | } |
| 65 | #else |
| 66 | static void get_pixels_altivec(int16_t *restrict block, const uint8_t *pixels, |
| 67 | ptrdiff_t line_size) |
| 68 | { |
| 69 | int i; |
| 70 | vector unsigned char perm = vec_lvsl(0, pixels); |
| 71 | const vector unsigned char zero = |
| 72 | (const vector unsigned char) vec_splat_u8(0); |
| 73 | |
| 74 | for (i = 0; i < 8; i++) { |
| 75 | /* Read potentially unaligned pixels. |
| 76 | * We're reading 16 pixels, and actually only want 8, |
| 77 | * but we simply ignore the extras. */ |
| 78 | vector unsigned char pixl = vec_ld(0, pixels); |
| 79 | vector unsigned char pixr = vec_ld(7, pixels); |
| 80 | vector unsigned char bytes = vec_perm(pixl, pixr, perm); |
| 81 | |
| 82 | // Convert the bytes into shorts. |
| 83 | vector signed short shorts = (vector signed short) vec_mergeh(zero, |
| 84 | bytes); |
| 85 | |
| 86 | // Save the data to the block, we assume the block is 16-byte aligned. |
| 87 | vec_st(shorts, i * 16, (vector signed short *) block); |
| 88 | |
| 89 | pixels += line_size; |
| 90 | } |
| 91 | } |
| 92 | |
| 93 | #endif /* HAVE_VSX */ |
| 94 | |
| 95 | #if HAVE_VSX |
| 96 | static void diff_pixels_altivec(int16_t *restrict block, const uint8_t *s1, |
| 97 | const uint8_t *s2, int stride) |
| 98 | { |
| 99 | int i; |
| 100 | const vector unsigned char zero = |
| 101 | (const vector unsigned char) vec_splat_u8(0); |
| 102 | vector signed short shorts1, shorts2; |
| 103 | |
| 104 | for (i = 0; i < 4; i++) { |
| 105 | /* Read potentially unaligned pixels. |
| 106 | * We're reading 16 pixels, and actually only want 8, |
| 107 | * but we simply ignore the extras. */ |
| 108 | vector unsigned char bytes = vec_vsx_ld(0, s1); |
| 109 | |
| 110 | // Convert the bytes into shorts. |
| 111 | shorts1 = (vector signed short) vec_mergeh(bytes, zero); |
| 112 | |
| 113 | // Do the same for the second block of pixels. |
| 114 | bytes =vec_vsx_ld(0, s2); |
| 115 | |
| 116 | // Convert the bytes into shorts. |
| 117 | shorts2 = (vector signed short) vec_mergeh(bytes, zero); |
| 118 | |
| 119 | // Do the subtraction. |
| 120 | shorts1 = vec_sub(shorts1, shorts2); |
| 121 | |
| 122 | // Save the data to the block, we assume the block is 16-byte aligned. |
| 123 | vec_vsx_st(shorts1, 0, (vector signed short *) block); |
| 124 | |
| 125 | s1 += stride; |
| 126 | s2 += stride; |
| 127 | block += 8; |
| 128 | |
| 129 | /* The code below is a copy of the code above... |
| 130 | * This is a manual unroll. */ |
| 131 | |
| 132 | /* Read potentially unaligned pixels. |
| 133 | * We're reading 16 pixels, and actually only want 8, |
| 134 | * but we simply ignore the extras. */ |
| 135 | bytes = vec_vsx_ld(0, s1); |
| 136 | |
| 137 | // Convert the bytes into shorts. |
| 138 | shorts1 = (vector signed short) vec_mergeh(bytes, zero); |
| 139 | |
| 140 | // Do the same for the second block of pixels. |
| 141 | bytes = vec_vsx_ld(0, s2); |
| 142 | |
| 143 | // Convert the bytes into shorts. |
| 144 | shorts2 = (vector signed short) vec_mergeh(bytes, zero); |
| 145 | |
| 146 | // Do the subtraction. |
| 147 | shorts1 = vec_sub(shorts1, shorts2); |
| 148 | |
| 149 | // Save the data to the block, we assume the block is 16-byte aligned. |
| 150 | vec_vsx_st(shorts1, 0, (vector signed short *) block); |
| 151 | |
| 152 | s1 += stride; |
| 153 | s2 += stride; |
| 154 | block += 8; |
| 155 | } |
| 156 | } |
| 157 | #else |
| 158 | static void diff_pixels_altivec(int16_t *restrict block, const uint8_t *s1, |
| 159 | const uint8_t *s2, int stride) |
| 160 | { |
| 161 | int i; |
| 162 | vector unsigned char perm1 = vec_lvsl(0, s1); |
| 163 | vector unsigned char perm2 = vec_lvsl(0, s2); |
| 164 | const vector unsigned char zero = |
| 165 | (const vector unsigned char) vec_splat_u8(0); |
| 166 | vector signed short shorts1, shorts2; |
| 167 | |
| 168 | for (i = 0; i < 4; i++) { |
| 169 | /* Read potentially unaligned pixels. |
| 170 | * We're reading 16 pixels, and actually only want 8, |
| 171 | * but we simply ignore the extras. */ |
| 172 | vector unsigned char pixl = vec_ld(0, s1); |
| 173 | vector unsigned char pixr = vec_ld(15, s1); |
| 174 | vector unsigned char bytes = vec_perm(pixl, pixr, perm1); |
| 175 | |
| 176 | // Convert the bytes into shorts. |
| 177 | shorts1 = (vector signed short) vec_mergeh(zero, bytes); |
| 178 | |
| 179 | // Do the same for the second block of pixels. |
| 180 | pixl = vec_ld(0, s2); |
| 181 | pixr = vec_ld(15, s2); |
| 182 | bytes = vec_perm(pixl, pixr, perm2); |
| 183 | |
| 184 | // Convert the bytes into shorts. |
| 185 | shorts2 = (vector signed short) vec_mergeh(zero, bytes); |
| 186 | |
| 187 | // Do the subtraction. |
| 188 | shorts1 = vec_sub(shorts1, shorts2); |
| 189 | |
| 190 | // Save the data to the block, we assume the block is 16-byte aligned. |
| 191 | vec_st(shorts1, 0, (vector signed short *) block); |
| 192 | |
| 193 | s1 += stride; |
| 194 | s2 += stride; |
| 195 | block += 8; |
| 196 | |
| 197 | /* The code below is a copy of the code above... |
| 198 | * This is a manual unroll. */ |
| 199 | |
| 200 | /* Read potentially unaligned pixels. |
| 201 | * We're reading 16 pixels, and actually only want 8, |
| 202 | * but we simply ignore the extras. */ |
| 203 | pixl = vec_ld(0, s1); |
| 204 | pixr = vec_ld(15, s1); |
| 205 | bytes = vec_perm(pixl, pixr, perm1); |
| 206 | |
| 207 | // Convert the bytes into shorts. |
| 208 | shorts1 = (vector signed short) vec_mergeh(zero, bytes); |
| 209 | |
| 210 | // Do the same for the second block of pixels. |
| 211 | pixl = vec_ld(0, s2); |
| 212 | pixr = vec_ld(15, s2); |
| 213 | bytes = vec_perm(pixl, pixr, perm2); |
| 214 | |
| 215 | // Convert the bytes into shorts. |
| 216 | shorts2 = (vector signed short) vec_mergeh(zero, bytes); |
| 217 | |
| 218 | // Do the subtraction. |
| 219 | shorts1 = vec_sub(shorts1, shorts2); |
| 220 | |
| 221 | // Save the data to the block, we assume the block is 16-byte aligned. |
| 222 | vec_st(shorts1, 0, (vector signed short *) block); |
| 223 | |
| 224 | s1 += stride; |
| 225 | s2 += stride; |
| 226 | block += 8; |
| 227 | } |
| 228 | } |
| 229 | |
| 230 | #endif /* HAVE_VSX */ |
| 231 | |
| 232 | #endif /* HAVE_ALTIVEC */ |
| 233 | |
| 234 | av_cold void ff_pixblockdsp_init_ppc(PixblockDSPContext *c, |
| 235 | AVCodecContext *avctx, |
| 236 | unsigned high_bit_depth) |
| 237 | { |
| 238 | #if HAVE_ALTIVEC |
| 239 | if (!PPC_ALTIVEC(av_get_cpu_flags())) |
| 240 | return; |
| 241 | |
| 242 | c->diff_pixels = diff_pixels_altivec; |
| 243 | |
| 244 | if (!high_bit_depth) { |
| 245 | c->get_pixels = get_pixels_altivec; |
| 246 | } |
| 247 | #endif /* HAVE_ALTIVEC */ |
| 248 | } |