| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /* | ||
| 2 | * VP9 SIMD optimizations | ||
| 3 | * | ||
| 4 | * Copyright (c) 2013 Ronald S. Bultje <rsbultje gmail com> | ||
| 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 "libavutil/attributes.h" | ||
| 24 | #include "libavutil/cpu.h" | ||
| 25 | #include "libavutil/x86/cpu.h" | ||
| 26 | #include "libavcodec/vp9dsp.h" | ||
| 27 | #include "libavcodec/x86/vp9dsp_init.h" | ||
| 28 | |||
| 29 | #if HAVE_X86ASM | ||
| 30 | |||
| 31 | decl_fpel_func(put, 8, , mmx); | ||
| 32 | decl_fpel_func(avg, 8, _16, mmxext); | ||
| 33 | decl_fpel_func(put, 16, , sse); | ||
| 34 | decl_fpel_func(put, 32, , sse); | ||
| 35 | decl_fpel_func(put, 64, , sse); | ||
| 36 | decl_fpel_func(put, 128, , sse); | ||
| 37 | decl_fpel_func(avg, 16, _16, sse2); | ||
| 38 | decl_fpel_func(avg, 32, _16, sse2); | ||
| 39 | decl_fpel_func(avg, 64, _16, sse2); | ||
| 40 | decl_fpel_func(avg, 128, _16, sse2); | ||
| 41 | decl_fpel_func(put, 32, , avx); | ||
| 42 | decl_fpel_func(put, 64, , avx); | ||
| 43 | decl_fpel_func(put, 128, , avx); | ||
| 44 | decl_fpel_func(avg, 32, _16, avx2); | ||
| 45 | decl_fpel_func(avg, 64, _16, avx2); | ||
| 46 | decl_fpel_func(avg, 128, _16, avx2); | ||
| 47 | |||
| 48 | decl_ipred_fns(v, 16, mmx, sse); | ||
| 49 | decl_ipred_fns(h, 16, mmxext, sse2); | ||
| 50 | decl_ipred_fns(dc, 16, mmxext, sse2); | ||
| 51 | decl_ipred_fns(dc_top, 16, mmxext, sse2); | ||
| 52 | decl_ipred_fns(dc_left, 16, mmxext, sse2); | ||
| 53 | decl_ipred_fn(dl, 16, 16, avx2); | ||
| 54 | decl_ipred_fn(dl, 32, 16, avx2); | ||
| 55 | decl_ipred_fn(dr, 16, 16, avx2); | ||
| 56 | decl_ipred_fn(dr, 32, 16, avx2); | ||
| 57 | decl_ipred_fn(vl, 16, 16, avx2); | ||
| 58 | decl_ipred_fn(hd, 16, 16, avx2); | ||
| 59 | |||
| 60 | #define decl_ipred_dir_funcs(type) \ | ||
| 61 | decl_ipred_fns(type, 16, sse2, sse2); \ | ||
| 62 | decl_ipred_fns(type, 16, ssse3, ssse3); \ | ||
| 63 | decl_ipred_fns(type, 16, avx, avx) | ||
| 64 | |||
| 65 | decl_ipred_dir_funcs(dl); | ||
| 66 | decl_ipred_dir_funcs(dr); | ||
| 67 | decl_ipred_dir_funcs(vl); | ||
| 68 | decl_ipred_dir_funcs(vr); | ||
| 69 | decl_ipred_dir_funcs(hu); | ||
| 70 | decl_ipred_dir_funcs(hd); | ||
| 71 | #endif /* HAVE_X86ASM */ | ||
| 72 | |||
| 73 | 156 | av_cold void ff_vp9dsp_init_16bpp_x86(VP9DSPContext *dsp) | |
| 74 | { | ||
| 75 | #if HAVE_X86ASM | ||
| 76 | 156 | int cpu_flags = av_get_cpu_flags(); | |
| 77 | |||
| 78 |
2/2✓ Branch 0 taken 123 times.
✓ Branch 1 taken 33 times.
|
156 | if (EXTERNAL_MMX(cpu_flags)) { |
| 79 | 123 | init_fpel_func(4, 0, 8, put, , mmx); | |
| 80 | 123 | init_ipred_func(v, VERT, 4, 16, mmx); | |
| 81 | } | ||
| 82 | |||
| 83 |
2/2✓ Branch 0 taken 113 times.
✓ Branch 1 taken 43 times.
|
156 | if (EXTERNAL_MMXEXT(cpu_flags)) { |
| 84 | 113 | init_fpel_func(4, 1, 8, avg, _16, mmxext); | |
| 85 | 113 | init_ipred_func(h, HOR, 4, 16, mmxext); | |
| 86 | 113 | init_ipred_func(dc, DC, 4, 16, mmxext); | |
| 87 | 113 | init_ipred_func(dc_top, TOP_DC, 4, 16, mmxext); | |
| 88 | 113 | init_ipred_func(dc_left, LEFT_DC, 4, 16, mmxext); | |
| 89 | } | ||
| 90 | |||
| 91 |
2/2✓ Branch 0 taken 103 times.
✓ Branch 1 taken 53 times.
|
156 | if (EXTERNAL_SSE(cpu_flags)) { |
| 92 | 103 | init_fpel_func(3, 0, 16, put, , sse); | |
| 93 | 103 | init_fpel_func(2, 0, 32, put, , sse); | |
| 94 | 103 | init_fpel_func(1, 0, 64, put, , sse); | |
| 95 | 103 | init_fpel_func(0, 0, 128, put, , sse); | |
| 96 | 103 | init_8_16_32_ipred_funcs(v, VERT, 16, sse); | |
| 97 | } | ||
| 98 | |||
| 99 |
2/2✓ Branch 0 taken 93 times.
✓ Branch 1 taken 63 times.
|
156 | if (EXTERNAL_SSE2(cpu_flags)) { |
| 100 | 93 | init_fpel_func(3, 1, 16, avg, _16, sse2); | |
| 101 | 93 | init_fpel_func(2, 1, 32, avg, _16, sse2); | |
| 102 | 93 | init_fpel_func(1, 1, 64, avg, _16, sse2); | |
| 103 | 93 | init_fpel_func(0, 1, 128, avg, _16, sse2); | |
| 104 | 93 | init_8_16_32_ipred_funcs(h, HOR, 16, sse2); | |
| 105 | 93 | init_8_16_32_ipred_funcs(dc, DC, 16, sse2); | |
| 106 | 93 | init_8_16_32_ipred_funcs(dc_top, TOP_DC, 16, sse2); | |
| 107 | 93 | init_8_16_32_ipred_funcs(dc_left, LEFT_DC, 16, sse2); | |
| 108 | 93 | init_ipred_funcs(dl, DIAG_DOWN_LEFT, 16, sse2); | |
| 109 | 93 | init_ipred_funcs(dr, DIAG_DOWN_RIGHT, 16, sse2); | |
| 110 | 93 | init_ipred_funcs(vl, VERT_LEFT, 16, sse2); | |
| 111 | 93 | init_ipred_funcs(vr, VERT_RIGHT, 16, sse2); | |
| 112 | 93 | init_ipred_funcs(hu, HOR_UP, 16, sse2); | |
| 113 | 93 | init_ipred_funcs(hd, HOR_DOWN, 16, sse2); | |
| 114 | } | ||
| 115 | |||
| 116 |
2/2✓ Branch 0 taken 73 times.
✓ Branch 1 taken 83 times.
|
156 | if (EXTERNAL_SSSE3(cpu_flags)) { |
| 117 | 73 | init_ipred_funcs(dl, DIAG_DOWN_LEFT, 16, ssse3); | |
| 118 | 73 | init_ipred_funcs(dr, DIAG_DOWN_RIGHT, 16, ssse3); | |
| 119 | 73 | init_ipred_funcs(vl, VERT_LEFT, 16, ssse3); | |
| 120 | 73 | init_ipred_funcs(vr, VERT_RIGHT, 16, ssse3); | |
| 121 | 73 | init_ipred_funcs(hu, HOR_UP, 16, ssse3); | |
| 122 | 73 | init_ipred_funcs(hd, HOR_DOWN, 16, ssse3); | |
| 123 | } | ||
| 124 | |||
| 125 |
3/4✓ Branch 0 taken 33 times.
✓ Branch 1 taken 123 times.
✓ Branch 2 taken 33 times.
✗ Branch 3 not taken.
|
156 | if (EXTERNAL_AVX_FAST(cpu_flags)) { |
| 126 | 33 | init_fpel_func(2, 0, 32, put, , avx); | |
| 127 | 33 | init_fpel_func(1, 0, 64, put, , avx); | |
| 128 | 33 | init_fpel_func(0, 0, 128, put, , avx); | |
| 129 | 33 | init_ipred_funcs(dl, DIAG_DOWN_LEFT, 16, avx); | |
| 130 | 33 | init_ipred_funcs(dr, DIAG_DOWN_RIGHT, 16, avx); | |
| 131 | 33 | init_ipred_funcs(vl, VERT_LEFT, 16, avx); | |
| 132 | 33 | init_ipred_funcs(vr, VERT_RIGHT, 16, avx); | |
| 133 | 33 | init_ipred_funcs(hu, HOR_UP, 16, avx); | |
| 134 | 33 | init_ipred_funcs(hd, HOR_DOWN, 16, avx); | |
| 135 | } | ||
| 136 | |||
| 137 |
3/4✓ Branch 0 taken 13 times.
✓ Branch 1 taken 143 times.
✓ Branch 2 taken 13 times.
✗ Branch 3 not taken.
|
156 | if (EXTERNAL_AVX2_FAST(cpu_flags)) { |
| 138 | 13 | init_fpel_func(2, 1, 32, avg, _16, avx2); | |
| 139 | 13 | init_fpel_func(1, 1, 64, avg, _16, avx2); | |
| 140 | 13 | init_fpel_func(0, 1, 128, avg, _16, avx2); | |
| 141 | 13 | init_ipred_func(dl, DIAG_DOWN_LEFT, 16, 16, avx2); | |
| 142 | 13 | init_ipred_func(dl, DIAG_DOWN_LEFT, 32, 16, avx2); | |
| 143 | 13 | init_ipred_func(dr, DIAG_DOWN_RIGHT, 16, 16, avx2); | |
| 144 | 13 | init_ipred_func(vl, VERT_LEFT, 16, 16, avx2); | |
| 145 | 13 | init_ipred_func(hd, HOR_DOWN, 16, 16, avx2); | |
| 146 | #if ARCH_X86_64 | ||
| 147 | 13 | init_ipred_func(dr, DIAG_DOWN_RIGHT, 32, 16, avx2); | |
| 148 | #endif | ||
| 149 | } | ||
| 150 | |||
| 151 | #endif /* HAVE_X86ASM */ | ||
| 152 | 156 | } | |
| 153 |