| Line | Branch | Exec | Source |
|---|---|---|---|
| 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 <stdint.h> | ||
| 20 | |||
| 21 | #include "libavutil/attributes.h" | ||
| 22 | #include "libavutil/avassert.h" | ||
| 23 | #include "libavutil/common.h" | ||
| 24 | #include "libavutil/cpu.h" | ||
| 25 | #include "libavutil/x86/asm.h" | ||
| 26 | #include "libavutil/x86/cpu.h" | ||
| 27 | #include "libavcodec/avcodec.h" | ||
| 28 | #include "libavcodec/mpegvideoencdsp.h" | ||
| 29 | |||
| 30 | int ff_pix_sum16_sse2(const uint8_t *pix, ptrdiff_t line_size); | ||
| 31 | int ff_pix_sum16_xop(const uint8_t *pix, ptrdiff_t line_size); | ||
| 32 | int ff_pix_norm1_sse2(const uint8_t *pix, ptrdiff_t line_size); | ||
| 33 | |||
| 34 | #if HAVE_INLINE_ASM | ||
| 35 | #if HAVE_SSSE3_INLINE | ||
| 36 | #define SCALE_OFFSET -1 | ||
| 37 | |||
| 38 | #define MAX_ABS 512 | ||
| 39 | |||
| 40 | ✗ | static int try_8x8basis_ssse3(const int16_t rem[64], const int16_t weight[64], const int16_t basis[64], int scale) | |
| 41 | { | ||
| 42 | ✗ | x86_reg i=0; | |
| 43 | |||
| 44 | av_assert2(FFABS(scale) < MAX_ABS); | ||
| 45 | ✗ | scale <<= 16 + SCALE_OFFSET - BASIS_SHIFT + RECON_SHIFT; | |
| 46 | |||
| 47 | ✗ | __asm__ volatile( | |
| 48 | "pxor %%xmm2, %%xmm2 \n\t" | ||
| 49 | "movd %4, %%xmm3 \n\t" | ||
| 50 | "punpcklwd %%xmm3, %%xmm3 \n\t" | ||
| 51 | "pshufd $0, %%xmm3, %%xmm3 \n\t" | ||
| 52 | ".p2align 4 \n\t" | ||
| 53 | "1: \n\t" | ||
| 54 | "movdqa (%1, %0), %%xmm0 \n\t" | ||
| 55 | "movdqa 16(%1, %0), %%xmm1 \n\t" | ||
| 56 | "pmulhrsw %%xmm3, %%xmm0 \n\t" | ||
| 57 | "pmulhrsw %%xmm3, %%xmm1 \n\t" | ||
| 58 | "paddw (%2, %0), %%xmm0 \n\t" | ||
| 59 | "paddw 16(%2, %0), %%xmm1 \n\t" | ||
| 60 | "psraw $6, %%xmm0 \n\t" | ||
| 61 | "psraw $6, %%xmm1 \n\t" | ||
| 62 | "pmullw (%3, %0), %%xmm0 \n\t" | ||
| 63 | "pmullw 16(%3, %0), %%xmm1 \n\t" | ||
| 64 | "pmaddwd %%xmm0, %%xmm0 \n\t" | ||
| 65 | "pmaddwd %%xmm1, %%xmm1 \n\t" | ||
| 66 | "paddd %%xmm1, %%xmm0 \n\t" | ||
| 67 | "psrld $4, %%xmm0 \n\t" | ||
| 68 | "paddd %%xmm0, %%xmm2 \n\t" | ||
| 69 | "add $32, %0 \n\t" | ||
| 70 | "cmp $128, %0 \n\t" //FIXME optimize & bench | ||
| 71 | " jb 1b \n\t" | ||
| 72 | "pshufd $0x0E, %%xmm2, %%xmm0 \n\t" | ||
| 73 | "paddd %%xmm0, %%xmm2 \n\t" | ||
| 74 | "pshufd $0x01, %%xmm2, %%xmm0 \n\t" | ||
| 75 | "paddd %%xmm0, %%xmm2 \n\t" | ||
| 76 | "psrld $2, %%xmm2 \n\t" | ||
| 77 | "movd %%xmm2, %0 \n\t" | ||
| 78 | : "+r" (i) | ||
| 79 | : "r"(basis), "r"(rem), "r"(weight), "g"(scale) | ||
| 80 | XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3") | ||
| 81 | ); | ||
| 82 | ✗ | return i; | |
| 83 | } | ||
| 84 | |||
| 85 | 1 | static void add_8x8basis_ssse3(int16_t rem[64], const int16_t basis[64], int scale) | |
| 86 | { | ||
| 87 | 1 | x86_reg i=0; | |
| 88 | |||
| 89 |
2/4✓ Branch 0 taken 1 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 1 times.
✗ Branch 3 not taken.
|
1 | if (FFABS(scale) < 1024) { |
| 90 | 1 | scale <<= 16 + SCALE_OFFSET - BASIS_SHIFT + RECON_SHIFT; | |
| 91 | 1 | __asm__ volatile( | |
| 92 | "movd %3, %%xmm2 \n\t" | ||
| 93 | "punpcklwd %%xmm2, %%xmm2 \n\t" | ||
| 94 | "pshufd $0, %%xmm2, %%xmm2 \n\t" | ||
| 95 | ".p2align 4 \n\t" | ||
| 96 | "1: \n\t" | ||
| 97 | "movdqa (%1, %0), %%xmm0 \n\t" | ||
| 98 | "movdqa 16(%1, %0), %%xmm1 \n\t" | ||
| 99 | "pmulhrsw %%xmm2, %%xmm0 \n\t" | ||
| 100 | "pmulhrsw %%xmm2, %%xmm1 \n\t" | ||
| 101 | "paddw (%2, %0), %%xmm0 \n\t" | ||
| 102 | "paddw 16(%2, %0), %%xmm1 \n\t" | ||
| 103 | "movdqa %%xmm0, (%2, %0) \n\t" | ||
| 104 | "movdqa %%xmm1, 16(%2, %0) \n\t" | ||
| 105 | "add $32, %0 \n\t" | ||
| 106 | "cmp $128, %0 \n\t" // FIXME optimize & bench | ||
| 107 | " jb 1b \n\t" | ||
| 108 | : "+r" (i) | ||
| 109 | : "r"(basis), "r"(rem), "g"(scale) | ||
| 110 | XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2") | ||
| 111 | ); | ||
| 112 | } else { | ||
| 113 | ✗ | for (i=0; i<8*8; i++) { | |
| 114 | ✗ | rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT); | |
| 115 | } | ||
| 116 | } | ||
| 117 | 1 | } | |
| 118 | |||
| 119 | #endif /* HAVE_SSSE3_INLINE */ | ||
| 120 | |||
| 121 | /* Draw the edges of width 'w' of an image of size width, height */ | ||
| 122 | 165 | static void draw_edges_mmx(uint8_t *buf, ptrdiff_t wrap, int width, int height, | |
| 123 | int w, int h, int sides) | ||
| 124 | { | ||
| 125 | uint8_t *ptr, *last_line; | ||
| 126 | int i; | ||
| 127 | |||
| 128 | /* left and right */ | ||
| 129 | 165 | ptr = buf; | |
| 130 |
2/2✓ Branch 0 taken 105 times.
✓ Branch 1 taken 60 times.
|
165 | if (w == 8) { |
| 131 | 105 | __asm__ volatile ( | |
| 132 | "1: \n\t" | ||
| 133 | "movd (%0), %%mm0 \n\t" | ||
| 134 | "punpcklbw %%mm0, %%mm0 \n\t" | ||
| 135 | "punpcklwd %%mm0, %%mm0 \n\t" | ||
| 136 | "punpckldq %%mm0, %%mm0 \n\t" | ||
| 137 | "movq %%mm0, -8(%0) \n\t" | ||
| 138 | "movq -8(%0, %2), %%mm1 \n\t" | ||
| 139 | "punpckhbw %%mm1, %%mm1 \n\t" | ||
| 140 | "punpckhwd %%mm1, %%mm1 \n\t" | ||
| 141 | "punpckhdq %%mm1, %%mm1 \n\t" | ||
| 142 | "movq %%mm1, (%0, %2) \n\t" | ||
| 143 | "add %1, %0 \n\t" | ||
| 144 | "cmp %3, %0 \n\t" | ||
| 145 | "jnz 1b \n\t" | ||
| 146 | : "+r" (ptr) | ||
| 147 | 105 | : "r" ((x86_reg) wrap), "r" ((x86_reg) width), | |
| 148 | 105 | "r" (ptr + wrap * height)); | |
| 149 |
2/2✓ Branch 0 taken 55 times.
✓ Branch 1 taken 5 times.
|
60 | } else if (w == 16) { |
| 150 | 55 | __asm__ volatile ( | |
| 151 | "1: \n\t" | ||
| 152 | "movd (%0), %%mm0 \n\t" | ||
| 153 | "punpcklbw %%mm0, %%mm0 \n\t" | ||
| 154 | "punpcklwd %%mm0, %%mm0 \n\t" | ||
| 155 | "punpckldq %%mm0, %%mm0 \n\t" | ||
| 156 | "movq %%mm0, -8(%0) \n\t" | ||
| 157 | "movq %%mm0, -16(%0) \n\t" | ||
| 158 | "movq -8(%0, %2), %%mm1 \n\t" | ||
| 159 | "punpckhbw %%mm1, %%mm1 \n\t" | ||
| 160 | "punpckhwd %%mm1, %%mm1 \n\t" | ||
| 161 | "punpckhdq %%mm1, %%mm1 \n\t" | ||
| 162 | "movq %%mm1, (%0, %2) \n\t" | ||
| 163 | "movq %%mm1, 8(%0, %2) \n\t" | ||
| 164 | "add %1, %0 \n\t" | ||
| 165 | "cmp %3, %0 \n\t" | ||
| 166 | "jnz 1b \n\t" | ||
| 167 | : "+r"(ptr) | ||
| 168 | 55 | : "r"((x86_reg)wrap), "r"((x86_reg)width), "r"(ptr + wrap * height) | |
| 169 | ); | ||
| 170 | } else { | ||
| 171 | av_assert1(w == 4); | ||
| 172 | 5 | __asm__ volatile ( | |
| 173 | "1: \n\t" | ||
| 174 | "movd (%0), %%mm0 \n\t" | ||
| 175 | "punpcklbw %%mm0, %%mm0 \n\t" | ||
| 176 | "punpcklwd %%mm0, %%mm0 \n\t" | ||
| 177 | "movd %%mm0, -4(%0) \n\t" | ||
| 178 | "movd -4(%0, %2), %%mm1 \n\t" | ||
| 179 | "punpcklbw %%mm1, %%mm1 \n\t" | ||
| 180 | "punpckhwd %%mm1, %%mm1 \n\t" | ||
| 181 | "punpckhdq %%mm1, %%mm1 \n\t" | ||
| 182 | "movd %%mm1, (%0, %2) \n\t" | ||
| 183 | "add %1, %0 \n\t" | ||
| 184 | "cmp %3, %0 \n\t" | ||
| 185 | "jnz 1b \n\t" | ||
| 186 | : "+r" (ptr) | ||
| 187 | 5 | : "r" ((x86_reg) wrap), "r" ((x86_reg) width), | |
| 188 | 5 | "r" (ptr + wrap * height)); | |
| 189 | } | ||
| 190 | |||
| 191 | /* top and bottom + corners */ | ||
| 192 | 165 | buf -= w; | |
| 193 | 165 | last_line = buf + (height - 1) * wrap; | |
| 194 |
1/2✓ Branch 0 taken 165 times.
✗ Branch 1 not taken.
|
165 | if (sides & EDGE_TOP) |
| 195 |
2/2✓ Branch 0 taken 1740 times.
✓ Branch 1 taken 165 times.
|
1905 | for (i = 0; i < h; i++) |
| 196 | // top | ||
| 197 | 1740 | memcpy(buf - (i + 1) * wrap, buf, width + w + w); | |
| 198 |
1/2✓ Branch 0 taken 165 times.
✗ Branch 1 not taken.
|
165 | if (sides & EDGE_BOTTOM) |
| 199 |
2/2✓ Branch 0 taken 1740 times.
✓ Branch 1 taken 165 times.
|
1905 | for (i = 0; i < h; i++) |
| 200 | // bottom | ||
| 201 | 1740 | memcpy(last_line + (i + 1) * wrap, last_line, width + w + w); | |
| 202 | 165 | } | |
| 203 | |||
| 204 | #endif /* HAVE_INLINE_ASM */ | ||
| 205 | |||
| 206 | 377 | av_cold void ff_mpegvideoencdsp_init_x86(MpegvideoEncDSPContext *c, | |
| 207 | AVCodecContext *avctx) | ||
| 208 | { | ||
| 209 | 377 | int cpu_flags = av_get_cpu_flags(); | |
| 210 | |||
| 211 |
2/2✓ Branch 0 taken 10 times.
✓ Branch 1 taken 367 times.
|
377 | if (EXTERNAL_SSE2(cpu_flags)) { |
| 212 | 10 | c->pix_sum = ff_pix_sum16_sse2; | |
| 213 | 10 | c->pix_norm1 = ff_pix_norm1_sse2; | |
| 214 | } | ||
| 215 | |||
| 216 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 377 times.
|
377 | if (EXTERNAL_XOP(cpu_flags)) { |
| 217 | ✗ | c->pix_sum = ff_pix_sum16_xop; | |
| 218 | } | ||
| 219 | |||
| 220 | #if HAVE_INLINE_ASM | ||
| 221 | |||
| 222 |
2/2✓ Branch 0 taken 13 times.
✓ Branch 1 taken 364 times.
|
377 | if (INLINE_MMX(cpu_flags)) { |
| 223 |
1/2✓ Branch 0 taken 13 times.
✗ Branch 1 not taken.
|
13 | if (avctx->bits_per_raw_sample <= 8) { |
| 224 | 13 | c->draw_edges = draw_edges_mmx; | |
| 225 | } | ||
| 226 | } | ||
| 227 | |||
| 228 | #if HAVE_SSSE3_INLINE | ||
| 229 |
2/2✓ Branch 0 taken 8 times.
✓ Branch 1 taken 369 times.
|
377 | if (INLINE_SSSE3(cpu_flags)) { |
| 230 |
2/2✓ Branch 0 taken 7 times.
✓ Branch 1 taken 1 times.
|
8 | if (!(avctx->flags & AV_CODEC_FLAG_BITEXACT)) { |
| 231 | 7 | c->try_8x8basis = try_8x8basis_ssse3; | |
| 232 | } | ||
| 233 | 8 | c->add_8x8basis = add_8x8basis_ssse3; | |
| 234 | } | ||
| 235 | #endif /* HAVE_SSSE3_INLINE */ | ||
| 236 | |||
| 237 | #endif /* HAVE_INLINE_ASM */ | ||
| 238 | 377 | } | |
| 239 |