| 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 | #include "mpegvideoencdsp.h" | ||
| 30 | |||
| 31 | int ff_pix_sum16_sse2(const uint8_t *pix, ptrdiff_t line_size); | ||
| 32 | int ff_pix_sum16_xop(const uint8_t *pix, ptrdiff_t line_size); | ||
| 33 | int ff_pix_norm1_sse2(const uint8_t *pix, ptrdiff_t line_size); | ||
| 34 | void ff_add_8x8basis_ssse3(int16_t rem[64], const int16_t basis[64], int scale); | ||
| 35 | |||
| 36 | #if HAVE_SSSE3_INLINE | ||
| 37 | #define SCALE_OFFSET -1 | ||
| 38 | |||
| 39 | #define MAX_ABS 512 | ||
| 40 | |||
| 41 | ✗ | static int try_8x8basis_ssse3(const int16_t rem[64], const int16_t weight[64], const int16_t basis[64], int scale) | |
| 42 | { | ||
| 43 | ✗ | x86_reg i=0; | |
| 44 | |||
| 45 | av_assert2(FFABS(scale) < MAX_ABS); | ||
| 46 | ✗ | scale *= 1 << (16 + SCALE_OFFSET - BASIS_SHIFT + RECON_SHIFT); | |
| 47 | |||
| 48 | ✗ | __asm__ volatile( | |
| 49 | "pxor %%xmm2, %%xmm2 \n\t" | ||
| 50 | "movd %4, %%xmm3 \n\t" | ||
| 51 | "punpcklwd %%xmm3, %%xmm3 \n\t" | ||
| 52 | "pshufd $0, %%xmm3, %%xmm3 \n\t" | ||
| 53 | ".p2align 4 \n\t" | ||
| 54 | "1: \n\t" | ||
| 55 | "movdqa (%1, %0), %%xmm0 \n\t" | ||
| 56 | "movdqa 16(%1, %0), %%xmm1 \n\t" | ||
| 57 | "pmulhrsw %%xmm3, %%xmm0 \n\t" | ||
| 58 | "pmulhrsw %%xmm3, %%xmm1 \n\t" | ||
| 59 | "paddw (%2, %0), %%xmm0 \n\t" | ||
| 60 | "paddw 16(%2, %0), %%xmm1 \n\t" | ||
| 61 | "psraw $6, %%xmm0 \n\t" | ||
| 62 | "psraw $6, %%xmm1 \n\t" | ||
| 63 | "pmullw (%3, %0), %%xmm0 \n\t" | ||
| 64 | "pmullw 16(%3, %0), %%xmm1 \n\t" | ||
| 65 | "pmaddwd %%xmm0, %%xmm0 \n\t" | ||
| 66 | "pmaddwd %%xmm1, %%xmm1 \n\t" | ||
| 67 | "paddd %%xmm1, %%xmm0 \n\t" | ||
| 68 | "psrld $4, %%xmm0 \n\t" | ||
| 69 | "paddd %%xmm0, %%xmm2 \n\t" | ||
| 70 | "add $32, %0 \n\t" | ||
| 71 | "cmp $128, %0 \n\t" //FIXME optimize & bench | ||
| 72 | " jb 1b \n\t" | ||
| 73 | "pshufd $0x0E, %%xmm2, %%xmm0 \n\t" | ||
| 74 | "paddd %%xmm0, %%xmm2 \n\t" | ||
| 75 | "pshufd $0x01, %%xmm2, %%xmm0 \n\t" | ||
| 76 | "paddd %%xmm0, %%xmm2 \n\t" | ||
| 77 | "psrld $2, %%xmm2 \n\t" | ||
| 78 | "movd %%xmm2, %0 \n\t" | ||
| 79 | : "+r" (i) | ||
| 80 | : "r"(basis), "r"(rem), "r"(weight), "g"(scale) | ||
| 81 | XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3") | ||
| 82 | ); | ||
| 83 | ✗ | return i; | |
| 84 | } | ||
| 85 | |||
| 86 | /* Draw the edges of width 'w' of an image of size width, height */ | ||
| 87 | 294 | static void draw_edges_ssse3(uint8_t *buf, ptrdiff_t wrap, int width, int height, | |
| 88 | int w, int h, int sides) | ||
| 89 | { | ||
| 90 | 294 | uint8_t *ptr = buf, *last_line; | |
| 91 | int i; | ||
| 92 | |||
| 93 | av_assert1(w == 16 || w == 8 || w == 4); | ||
| 94 | |||
| 95 | /* left and right */ | ||
| 96 | 294 | __asm__ volatile ( | |
| 97 | "pcmpeqw %%xmm3, %%xmm3 \n\t" | ||
| 98 | "pxor %%xmm2, %%xmm2 \n\t" | ||
| 99 | "psrlw $14, %%xmm3 \n\t" // pw_3 | ||
| 100 | "pshufb %%xmm2, %%xmm3 \n\t" // pb_3 | ||
| 101 | "cmp $8, %4 \n\t" | ||
| 102 | "jg 16f \n\t" | ||
| 103 | "jl 4f \n\t" | ||
| 104 | "8: \n\t" | ||
| 105 | "movd (%0), %%xmm0 \n\t" | ||
| 106 | "movd -4(%0, %2), %%xmm1 \n\t" | ||
| 107 | "pshufb %%xmm2, %%xmm0 \n\t" | ||
| 108 | "pshufb %%xmm3, %%xmm1 \n\t" | ||
| 109 | "movq %%xmm0, -8(%0) \n\t" | ||
| 110 | "movq %%xmm1, (%0, %2) \n\t" | ||
| 111 | "add %1, %0 \n\t" | ||
| 112 | "cmp %3, %0 \n\t" | ||
| 113 | "jnz 8b \n\t" | ||
| 114 | "jmp 1f \n\t" | ||
| 115 | "4: \n\t" | ||
| 116 | "movd (%0), %%xmm0 \n\t" | ||
| 117 | "movd -4(%0, %2), %%xmm1 \n\t" | ||
| 118 | "pshufb %%xmm2, %%xmm0 \n\t" | ||
| 119 | "pshufb %%xmm3, %%xmm1 \n\t" | ||
| 120 | "movd %%xmm0, -4(%0) \n\t" | ||
| 121 | "movd %%xmm1, (%0, %2) \n\t" | ||
| 122 | "add %1, %0 \n\t" | ||
| 123 | "cmp %3, %0 \n\t" | ||
| 124 | "jnz 4b \n\t" | ||
| 125 | "jmp 1f \n\t" | ||
| 126 | "16: \n\t" | ||
| 127 | "movd (%0), %%xmm0 \n\t" | ||
| 128 | "movd -4(%0, %2), %%xmm1 \n\t" | ||
| 129 | "pshufb %%xmm2, %%xmm0 \n\t" | ||
| 130 | "pshufb %%xmm3, %%xmm1 \n\t" | ||
| 131 | "movdqu %%xmm0, -16(%0) \n\t" | ||
| 132 | "movdqu %%xmm1, (%0, %2) \n\t" | ||
| 133 | "add %1, %0 \n\t" | ||
| 134 | "cmp %3, %0 \n\t" | ||
| 135 | "jnz 16b \n\t" | ||
| 136 | "1: \n\t" | ||
| 137 | : "+r" (ptr) | ||
| 138 | 294 | : "r" ((x86_reg) wrap), "r" ((x86_reg) width), "r"(ptr + wrap * height), "r" (w) | |
| 139 | XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3") | ||
| 140 | ); | ||
| 141 | |||
| 142 | /* top and bottom + corners */ | ||
| 143 | 294 | buf -= w; | |
| 144 | 294 | last_line = buf + (height - 1) * wrap; | |
| 145 |
2/2✓ Branch 0 taken 240 times.
✓ Branch 1 taken 54 times.
|
294 | if (sides & EDGE_TOP) |
| 146 |
2/2✓ Branch 0 taken 2540 times.
✓ Branch 1 taken 240 times.
|
2780 | for (i = 0; i < h; i++) |
| 147 | // top | ||
| 148 | 2540 | memcpy(buf - (i + 1) * wrap, buf, width + w + w); | |
| 149 |
1/2✓ Branch 0 taken 294 times.
✗ Branch 1 not taken.
|
294 | if (sides & EDGE_BOTTOM) |
| 150 |
2/2✓ Branch 0 taken 3116 times.
✓ Branch 1 taken 294 times.
|
3410 | for (i = 0; i < h; i++) |
| 151 | // bottom | ||
| 152 | 3116 | memcpy(last_line + (i + 1) * wrap, last_line, width + w + w); | |
| 153 | 294 | } | |
| 154 | #endif /* HAVE_SSSE3_INLINE */ | ||
| 155 | |||
| 156 | 387 | av_cold void ff_mpegvideoencdsp_init_x86(MpegvideoEncDSPContext *c, | |
| 157 | AVCodecContext *avctx) | ||
| 158 | { | ||
| 159 | 387 | int cpu_flags = av_get_cpu_flags(); | |
| 160 | |||
| 161 |
2/2✓ Branch 0 taken 20 times.
✓ Branch 1 taken 367 times.
|
387 | if (EXTERNAL_SSE2(cpu_flags)) { |
| 162 | 20 | c->denoise_dct = ff_mpv_denoise_dct_sse2; | |
| 163 | 20 | c->pix_sum = ff_pix_sum16_sse2; | |
| 164 | 20 | c->pix_norm1 = ff_pix_norm1_sse2; | |
| 165 | } | ||
| 166 | |||
| 167 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 387 times.
|
387 | if (EXTERNAL_XOP(cpu_flags)) { |
| 168 | ✗ | c->pix_sum = ff_pix_sum16_xop; | |
| 169 | } | ||
| 170 | |||
| 171 |
2/2✓ Branch 0 taken 18 times.
✓ Branch 1 taken 369 times.
|
387 | if (X86_SSSE3(cpu_flags)) { |
| 172 | #if HAVE_SSSE3_INLINE | ||
| 173 |
2/2✓ Branch 0 taken 16 times.
✓ Branch 1 taken 2 times.
|
18 | if (!(avctx->flags & AV_CODEC_FLAG_BITEXACT)) { |
| 174 | 16 | c->try_8x8basis = try_8x8basis_ssse3; | |
| 175 | } | ||
| 176 |
1/2✓ Branch 0 taken 18 times.
✗ Branch 1 not taken.
|
18 | if (avctx->bits_per_raw_sample <= 8) { |
| 177 | 18 | c->draw_edges = draw_edges_ssse3; | |
| 178 | } | ||
| 179 | #endif /* HAVE_SSSE3_INLINE */ | ||
| 180 | #if HAVE_SSSE3_EXTERNAL | ||
| 181 | 18 | c->add_8x8basis = ff_add_8x8basis_ssse3; | |
| 182 | #endif | ||
| 183 | } | ||
| 184 | |||
| 185 | 387 | } | |
| 186 |