| 1 | /* |
| 2 | * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at> |
| 3 | * |
| 4 | * This file is part of FFmpeg. |
| 5 | * |
| 6 | * FFmpeg is free software; you can redistribute it and/or modify |
| 7 | * it under the terms of the GNU General Public License as published by |
| 8 | * the Free Software Foundation; either version 2 of the License, or |
| 9 | * (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 |
| 14 | * GNU General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License along |
| 17 | * with FFmpeg; if not, write to the Free Software Foundation, Inc., |
| 18 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. |
| 19 | */ |
| 20 | |
| 21 | |
| 22 | #include "libavutil/attributes.h" |
| 23 | #include "libavutil/cpu.h" |
| 24 | #include "libavutil/mem.h" |
| 25 | #include "libavutil/x86/asm.h" |
| 26 | #include "libavfilter/vf_spp.h" |
| 27 | |
| 28 | #if HAVE_MMX_INLINE |
| 29 | static void hardthresh_mmx(int16_t dst[64], const int16_t src[64], |
| 30 | int qp, const uint8_t *permutation) |
| 31 | { |
| 32 | int bias = 0; //FIXME |
| 33 | unsigned int threshold1; |
| 34 | |
| 35 | threshold1 = qp * ((1<<4) - bias) - 1; |
| 36 | |
| 37 | #define REQUANT_CORE(dst0, dst1, dst2, dst3, src0, src1, src2, src3) \ |
| 38 | "movq " #src0 ", %%mm0 \n" \ |
| 39 | "movq " #src1 ", %%mm1 \n" \ |
| 40 | "movq " #src2 ", %%mm2 \n" \ |
| 41 | "movq " #src3 ", %%mm3 \n" \ |
| 42 | "psubw %%mm4, %%mm0 \n" \ |
| 43 | "psubw %%mm4, %%mm1 \n" \ |
| 44 | "psubw %%mm4, %%mm2 \n" \ |
| 45 | "psubw %%mm4, %%mm3 \n" \ |
| 46 | "paddusw %%mm5, %%mm0 \n" \ |
| 47 | "paddusw %%mm5, %%mm1 \n" \ |
| 48 | "paddusw %%mm5, %%mm2 \n" \ |
| 49 | "paddusw %%mm5, %%mm3 \n" \ |
| 50 | "paddw %%mm6, %%mm0 \n" \ |
| 51 | "paddw %%mm6, %%mm1 \n" \ |
| 52 | "paddw %%mm6, %%mm2 \n" \ |
| 53 | "paddw %%mm6, %%mm3 \n" \ |
| 54 | "psubusw %%mm6, %%mm0 \n" \ |
| 55 | "psubusw %%mm6, %%mm1 \n" \ |
| 56 | "psubusw %%mm6, %%mm2 \n" \ |
| 57 | "psubusw %%mm6, %%mm3 \n" \ |
| 58 | "psraw $3, %%mm0 \n" \ |
| 59 | "psraw $3, %%mm1 \n" \ |
| 60 | "psraw $3, %%mm2 \n" \ |
| 61 | "psraw $3, %%mm3 \n" \ |
| 62 | \ |
| 63 | "movq %%mm0, %%mm7 \n" \ |
| 64 | "punpcklwd %%mm2, %%mm0 \n" /*A*/ \ |
| 65 | "punpckhwd %%mm2, %%mm7 \n" /*C*/ \ |
| 66 | "movq %%mm1, %%mm2 \n" \ |
| 67 | "punpcklwd %%mm3, %%mm1 \n" /*B*/ \ |
| 68 | "punpckhwd %%mm3, %%mm2 \n" /*D*/ \ |
| 69 | "movq %%mm0, %%mm3 \n" \ |
| 70 | "punpcklwd %%mm1, %%mm0 \n" /*A*/ \ |
| 71 | "punpckhwd %%mm7, %%mm3 \n" /*C*/ \ |
| 72 | "punpcklwd %%mm2, %%mm7 \n" /*B*/ \ |
| 73 | "punpckhwd %%mm2, %%mm1 \n" /*D*/ \ |
| 74 | \ |
| 75 | "movq %%mm0, " #dst0 " \n" \ |
| 76 | "movq %%mm7, " #dst1 " \n" \ |
| 77 | "movq %%mm3, " #dst2 " \n" \ |
| 78 | "movq %%mm1, " #dst3 " \n" |
| 79 | |
| 80 | __asm__ volatile( |
| 81 | "movd %2, %%mm4 \n" |
| 82 | "movd %3, %%mm5 \n" |
| 83 | "movd %4, %%mm6 \n" |
| 84 | "packssdw %%mm4, %%mm4 \n" |
| 85 | "packssdw %%mm5, %%mm5 \n" |
| 86 | "packssdw %%mm6, %%mm6 \n" |
| 87 | "packssdw %%mm4, %%mm4 \n" |
| 88 | "packssdw %%mm5, %%mm5 \n" |
| 89 | "packssdw %%mm6, %%mm6 \n" |
| 90 | REQUANT_CORE( (%1), 8(%1), 16(%1), 24(%1), (%0), 8(%0), 64(%0), 72(%0)) |
| 91 | REQUANT_CORE(32(%1), 40(%1), 48(%1), 56(%1),16(%0),24(%0), 48(%0), 56(%0)) |
| 92 | REQUANT_CORE(64(%1), 72(%1), 80(%1), 88(%1),32(%0),40(%0), 96(%0),104(%0)) |
| 93 | REQUANT_CORE(96(%1),104(%1),112(%1),120(%1),80(%0),88(%0),112(%0),120(%0)) |
| 94 | : : "r" (src), "r" (dst), "g" (threshold1+1), "g" (threshold1+5), "g" (threshold1-4) //FIXME maybe more accurate then needed? |
| 95 | ); |
| 96 | dst[0] = (src[0] + 4) >> 3; |
| 97 | } |
| 98 | |
| 99 | static void softthresh_mmx(int16_t dst[64], const int16_t src[64], |
| 100 | int qp, const uint8_t *permutation) |
| 101 | { |
| 102 | int bias = 0; //FIXME |
| 103 | unsigned int threshold1; |
| 104 | |
| 105 | threshold1 = qp*((1<<4) - bias) - 1; |
| 106 | |
| 107 | #undef REQUANT_CORE |
| 108 | #define REQUANT_CORE(dst0, dst1, dst2, dst3, src0, src1, src2, src3) \ |
| 109 | "movq " #src0 ", %%mm0 \n" \ |
| 110 | "movq " #src1 ", %%mm1 \n" \ |
| 111 | "pxor %%mm6, %%mm6 \n" \ |
| 112 | "pxor %%mm7, %%mm7 \n" \ |
| 113 | "pcmpgtw %%mm0, %%mm6 \n" \ |
| 114 | "pcmpgtw %%mm1, %%mm7 \n" \ |
| 115 | "pxor %%mm6, %%mm0 \n" \ |
| 116 | "pxor %%mm7, %%mm1 \n" \ |
| 117 | "psubusw %%mm4, %%mm0 \n" \ |
| 118 | "psubusw %%mm4, %%mm1 \n" \ |
| 119 | "pxor %%mm6, %%mm0 \n" \ |
| 120 | "pxor %%mm7, %%mm1 \n" \ |
| 121 | "movq " #src2 ", %%mm2 \n" \ |
| 122 | "movq " #src3 ", %%mm3 \n" \ |
| 123 | "pxor %%mm6, %%mm6 \n" \ |
| 124 | "pxor %%mm7, %%mm7 \n" \ |
| 125 | "pcmpgtw %%mm2, %%mm6 \n" \ |
| 126 | "pcmpgtw %%mm3, %%mm7 \n" \ |
| 127 | "pxor %%mm6, %%mm2 \n" \ |
| 128 | "pxor %%mm7, %%mm3 \n" \ |
| 129 | "psubusw %%mm4, %%mm2 \n" \ |
| 130 | "psubusw %%mm4, %%mm3 \n" \ |
| 131 | "pxor %%mm6, %%mm2 \n" \ |
| 132 | "pxor %%mm7, %%mm3 \n" \ |
| 133 | \ |
| 134 | "paddsw %%mm5, %%mm0 \n" \ |
| 135 | "paddsw %%mm5, %%mm1 \n" \ |
| 136 | "paddsw %%mm5, %%mm2 \n" \ |
| 137 | "paddsw %%mm5, %%mm3 \n" \ |
| 138 | "psraw $3, %%mm0 \n" \ |
| 139 | "psraw $3, %%mm1 \n" \ |
| 140 | "psraw $3, %%mm2 \n" \ |
| 141 | "psraw $3, %%mm3 \n" \ |
| 142 | \ |
| 143 | "movq %%mm0, %%mm7 \n" \ |
| 144 | "punpcklwd %%mm2, %%mm0 \n" /*A*/ \ |
| 145 | "punpckhwd %%mm2, %%mm7 \n" /*C*/ \ |
| 146 | "movq %%mm1, %%mm2 \n" \ |
| 147 | "punpcklwd %%mm3, %%mm1 \n" /*B*/ \ |
| 148 | "punpckhwd %%mm3, %%mm2 \n" /*D*/ \ |
| 149 | "movq %%mm0, %%mm3 \n" \ |
| 150 | "punpcklwd %%mm1, %%mm0 \n" /*A*/ \ |
| 151 | "punpckhwd %%mm7, %%mm3 \n" /*C*/ \ |
| 152 | "punpcklwd %%mm2, %%mm7 \n" /*B*/ \ |
| 153 | "punpckhwd %%mm2, %%mm1 \n" /*D*/ \ |
| 154 | \ |
| 155 | "movq %%mm0, " #dst0 " \n" \ |
| 156 | "movq %%mm7, " #dst1 " \n" \ |
| 157 | "movq %%mm3, " #dst2 " \n" \ |
| 158 | "movq %%mm1, " #dst3 " \n" |
| 159 | |
| 160 | __asm__ volatile( |
| 161 | "movd %2, %%mm4 \n" |
| 162 | "movd %3, %%mm5 \n" |
| 163 | "packssdw %%mm4, %%mm4 \n" |
| 164 | "packssdw %%mm5, %%mm5 \n" |
| 165 | "packssdw %%mm4, %%mm4 \n" |
| 166 | "packssdw %%mm5, %%mm5 \n" |
| 167 | REQUANT_CORE( (%1), 8(%1), 16(%1), 24(%1), (%0), 8(%0), 64(%0), 72(%0)) |
| 168 | REQUANT_CORE(32(%1), 40(%1), 48(%1), 56(%1),16(%0),24(%0), 48(%0), 56(%0)) |
| 169 | REQUANT_CORE(64(%1), 72(%1), 80(%1), 88(%1),32(%0),40(%0), 96(%0),104(%0)) |
| 170 | REQUANT_CORE(96(%1),104(%1),112(%1),120(%1),80(%0),88(%0),112(%0),120(%0)) |
| 171 | : : "r" (src), "r" (dst), "g" (threshold1), "rm" (4) //FIXME maybe more accurate then needed? |
| 172 | ); |
| 173 | |
| 174 | dst[0] = (src[0] + 4) >> 3; |
| 175 | } |
| 176 | |
| 177 | static void store_slice_mmx(uint8_t *dst, const int16_t *src, |
| 178 | int dst_stride, int src_stride, |
| 179 | int width, int height, int log2_scale, |
| 180 | const uint8_t dither[8][8]) |
| 181 | { |
| 182 | int y; |
| 183 | |
| 184 | for (y = 0; y < height; y++) { |
| 185 | uint8_t *dst1 = dst; |
| 186 | const int16_t *src1 = src; |
| 187 | __asm__ volatile( |
| 188 | "movq (%3), %%mm3 \n" |
| 189 | "movq (%3), %%mm4 \n" |
| 190 | "movd %4, %%mm2 \n" |
| 191 | "pxor %%mm0, %%mm0 \n" |
| 192 | "punpcklbw %%mm0, %%mm3 \n" |
| 193 | "punpckhbw %%mm0, %%mm4 \n" |
| 194 | "psraw %%mm2, %%mm3 \n" |
| 195 | "psraw %%mm2, %%mm4 \n" |
| 196 | "movd %5, %%mm2 \n" |
| 197 | "1: \n" |
| 198 | "movq (%0), %%mm0 \n" |
| 199 | "movq 8(%0), %%mm1 \n" |
| 200 | "paddw %%mm3, %%mm0 \n" |
| 201 | "paddw %%mm4, %%mm1 \n" |
| 202 | "psraw %%mm2, %%mm0 \n" |
| 203 | "psraw %%mm2, %%mm1 \n" |
| 204 | "packuswb %%mm1, %%mm0 \n" |
| 205 | "movq %%mm0, (%1) \n" |
| 206 | "add $16, %0 \n" |
| 207 | "add $8, %1 \n" |
| 208 | "cmp %2, %1 \n" |
| 209 | " jb 1b \n" |
| 210 | : "+r" (src1), "+r"(dst1) |
| 211 | : "r"(dst + width), "r"(dither[y]), "g"(log2_scale), "g"(MAX_LEVEL - log2_scale) |
| 212 | ); |
| 213 | src += src_stride; |
| 214 | dst += dst_stride; |
| 215 | } |
| 216 | } |
| 217 | |
| 218 | #endif /* HAVE_MMX_INLINE */ |
| 219 | |
| 220 | av_cold void ff_spp_init_x86(SPPContext *s) |
| 221 | { |
| 222 | #if HAVE_MMX_INLINE |
| 223 | int cpu_flags = av_get_cpu_flags(); |
| 224 | |
| 225 | if (cpu_flags & AV_CPU_FLAG_MMX) { |
| 226 | s->store_slice = store_slice_mmx; |
| 227 | switch (s->mode) { |
| 228 | case 0: s->requantize = hardthresh_mmx; break; |
| 229 | case 1: s->requantize = softthresh_mmx; break; |
| 230 | } |
| 231 | } |
| 232 | #endif |
| 233 | } |