| 1 | /* |
| 2 | * This file is part of FFmpeg. |
| 3 | * |
| 4 | * FFmpeg is free software; you can redistribute it and/or |
| 5 | * modify it under the terms of the GNU Lesser General Public |
| 6 | * License as published by the Free Software Foundation; either |
| 7 | * version 2.1 of the License, or (at your option) any later version. |
| 8 | * |
| 9 | * FFmpeg is distributed in the hope that it will be useful, |
| 10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 12 | * Lesser General Public License for more details. |
| 13 | * |
| 14 | * You should have received a copy of the GNU Lesser General Public |
| 15 | * License along with FFmpeg; if not, write to the Free Software |
| 16 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 17 | */ |
| 18 | |
| 19 | #include "config.h" |
| 20 | #include "libavutil/attributes.h" |
| 21 | #include "libavutil/cpu.h" |
| 22 | #include "libavutil/x86/asm.h" |
| 23 | #include "libavutil/x86/cpu.h" |
| 24 | #include "libavcodec/mpegvideodsp.h" |
| 25 | #include "libavcodec/videodsp.h" |
| 26 | |
| 27 | #if HAVE_INLINE_ASM |
| 28 | |
| 29 | static void gmc_mmx(uint8_t *dst, uint8_t *src, |
| 30 | int stride, int h, int ox, int oy, |
| 31 | int dxx, int dxy, int dyx, int dyy, |
| 32 | int shift, int r, int width, int height) |
| 33 | { |
| 34 | const int w = 8; |
| 35 | const int ix = ox >> (16 + shift); |
| 36 | const int iy = oy >> (16 + shift); |
| 37 | const int oxs = ox >> 4; |
| 38 | const int oys = oy >> 4; |
| 39 | const int dxxs = dxx >> 4; |
| 40 | const int dxys = dxy >> 4; |
| 41 | const int dyxs = dyx >> 4; |
| 42 | const int dyys = dyy >> 4; |
| 43 | const uint16_t r4[4] = { r, r, r, r }; |
| 44 | const uint16_t dxy4[4] = { dxys, dxys, dxys, dxys }; |
| 45 | const uint16_t dyy4[4] = { dyys, dyys, dyys, dyys }; |
| 46 | const uint64_t shift2 = 2 * shift; |
| 47 | #define MAX_STRIDE 4096U |
| 48 | #define MAX_H 8U |
| 49 | uint8_t edge_buf[(MAX_H + 1) * MAX_STRIDE]; |
| 50 | int x, y; |
| 51 | |
| 52 | const int dxw = (dxx - (1 << (16 + shift))) * (w - 1); |
| 53 | const int dyh = (dyy - (1 << (16 + shift))) * (h - 1); |
| 54 | const int dxh = dxy * (h - 1); |
| 55 | const int dyw = dyx * (w - 1); |
| 56 | int need_emu = (unsigned) ix >= width - w || |
| 57 | (unsigned) iy >= height - h; |
| 58 | |
| 59 | if ( // non-constant fullpel offset (3% of blocks) |
| 60 | ((ox ^ (ox + dxw)) | (ox ^ (ox + dxh)) | (ox ^ (ox + dxw + dxh)) | |
| 61 | (oy ^ (oy + dyw)) | (oy ^ (oy + dyh)) | (oy ^ (oy + dyw + dyh))) >> (16 + shift) || |
| 62 | // uses more than 16 bits of subpel mv (only at huge resolution) |
| 63 | (dxx | dxy | dyx | dyy) & 15 || |
| 64 | (need_emu && (h > MAX_H || stride > MAX_STRIDE))) { |
| 65 | // FIXME could still use mmx for some of the rows |
| 66 | ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, |
| 67 | shift, r, width, height); |
| 68 | return; |
| 69 | } |
| 70 | |
| 71 | src += ix + iy * stride; |
| 72 | if (need_emu) { |
| 73 | ff_emulated_edge_mc_8(edge_buf, src, stride, stride, w + 1, h + 1, ix, iy, width, height); |
| 74 | src = edge_buf; |
| 75 | } |
| 76 | |
| 77 | __asm__ volatile ( |
| 78 | "movd %0, %%mm6 \n\t" |
| 79 | "pxor %%mm7, %%mm7 \n\t" |
| 80 | "punpcklwd %%mm6, %%mm6 \n\t" |
| 81 | "punpcklwd %%mm6, %%mm6 \n\t" |
| 82 | :: "r" (1 << shift)); |
| 83 | |
| 84 | for (x = 0; x < w; x += 4) { |
| 85 | uint16_t dx4[4] = { oxs - dxys + dxxs * (x + 0), |
| 86 | oxs - dxys + dxxs * (x + 1), |
| 87 | oxs - dxys + dxxs * (x + 2), |
| 88 | oxs - dxys + dxxs * (x + 3) }; |
| 89 | uint16_t dy4[4] = { oys - dyys + dyxs * (x + 0), |
| 90 | oys - dyys + dyxs * (x + 1), |
| 91 | oys - dyys + dyxs * (x + 2), |
| 92 | oys - dyys + dyxs * (x + 3) }; |
| 93 | |
| 94 | for (y = 0; y < h; y++) { |
| 95 | __asm__ volatile ( |
| 96 | "movq %0, %%mm4 \n\t" |
| 97 | "movq %1, %%mm5 \n\t" |
| 98 | "paddw %2, %%mm4 \n\t" |
| 99 | "paddw %3, %%mm5 \n\t" |
| 100 | "movq %%mm4, %0 \n\t" |
| 101 | "movq %%mm5, %1 \n\t" |
| 102 | "psrlw $12, %%mm4 \n\t" |
| 103 | "psrlw $12, %%mm5 \n\t" |
| 104 | : "+m" (*dx4), "+m" (*dy4) |
| 105 | : "m" (*dxy4), "m" (*dyy4)); |
| 106 | |
| 107 | __asm__ volatile ( |
| 108 | "movq %%mm6, %%mm2 \n\t" |
| 109 | "movq %%mm6, %%mm1 \n\t" |
| 110 | "psubw %%mm4, %%mm2 \n\t" |
| 111 | "psubw %%mm5, %%mm1 \n\t" |
| 112 | "movq %%mm2, %%mm0 \n\t" |
| 113 | "movq %%mm4, %%mm3 \n\t" |
| 114 | "pmullw %%mm1, %%mm0 \n\t" // (s - dx) * (s - dy) |
| 115 | "pmullw %%mm5, %%mm3 \n\t" // dx * dy |
| 116 | "pmullw %%mm5, %%mm2 \n\t" // (s - dx) * dy |
| 117 | "pmullw %%mm4, %%mm1 \n\t" // dx * (s - dy) |
| 118 | |
| 119 | "movd %4, %%mm5 \n\t" |
| 120 | "movd %3, %%mm4 \n\t" |
| 121 | "punpcklbw %%mm7, %%mm5 \n\t" |
| 122 | "punpcklbw %%mm7, %%mm4 \n\t" |
| 123 | "pmullw %%mm5, %%mm3 \n\t" // src[1, 1] * dx * dy |
| 124 | "pmullw %%mm4, %%mm2 \n\t" // src[0, 1] * (s - dx) * dy |
| 125 | |
| 126 | "movd %2, %%mm5 \n\t" |
| 127 | "movd %1, %%mm4 \n\t" |
| 128 | "punpcklbw %%mm7, %%mm5 \n\t" |
| 129 | "punpcklbw %%mm7, %%mm4 \n\t" |
| 130 | "pmullw %%mm5, %%mm1 \n\t" // src[1, 0] * dx * (s - dy) |
| 131 | "pmullw %%mm4, %%mm0 \n\t" // src[0, 0] * (s - dx) * (s - dy) |
| 132 | "paddw %5, %%mm1 \n\t" |
| 133 | "paddw %%mm3, %%mm2 \n\t" |
| 134 | "paddw %%mm1, %%mm0 \n\t" |
| 135 | "paddw %%mm2, %%mm0 \n\t" |
| 136 | |
| 137 | "psrlw %6, %%mm0 \n\t" |
| 138 | "packuswb %%mm0, %%mm0 \n\t" |
| 139 | "movd %%mm0, %0 \n\t" |
| 140 | |
| 141 | : "=m" (dst[x + y * stride]) |
| 142 | : "m" (src[0]), "m" (src[1]), |
| 143 | "m" (src[stride]), "m" (src[stride + 1]), |
| 144 | "m" (*r4), "m" (shift2)); |
| 145 | src += stride; |
| 146 | } |
| 147 | src += 4 - h * stride; |
| 148 | } |
| 149 | } |
| 150 | |
| 151 | #endif /* HAVE_INLINE_ASM */ |
| 152 | |
| 153 | av_cold void ff_mpegvideodsp_init_x86(MpegVideoDSPContext *c) |
| 154 | { |
| 155 | #if HAVE_INLINE_ASM |
| 156 | int cpu_flags = av_get_cpu_flags(); |
| 157 | |
| 158 | if (INLINE_MMX(cpu_flags)) |
| 159 | c->gmc = gmc_mmx; |
| 160 | #endif /* HAVE_INLINE_ASM */ |
| 161 | } |