| 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 "config.h" | ||
| 20 | #include "config_components.h" | ||
| 21 | |||
| 22 | #include "libavutil/cpu.h" | ||
| 23 | #include "libavutil/mem_internal.h" | ||
| 24 | |||
| 25 | #include "libavcodec/x86/fdct.h" | ||
| 26 | #include "libavcodec/x86/xvididct.h" | ||
| 27 | #include "libavcodec/x86/simple_idct.h" | ||
| 28 | |||
| 29 | #if CONFIG_PRORES_DECODER && ARCH_X86_64 && HAVE_X86ASM | ||
| 30 | void ff_prores_idct_put_10_sse2(uint16_t *dst, int linesize, | ||
| 31 | int16_t *block, int16_t *qmat); | ||
| 32 | |||
| 33 | #define PR_WRAP(INSN) \ | ||
| 34 | static void ff_prores_idct_put_10_##INSN##_wrap(int16_t *dst){ \ | ||
| 35 | LOCAL_ALIGNED(16, int16_t, qmat, [64]); \ | ||
| 36 | LOCAL_ALIGNED(16, int16_t, tmp, [64]); \ | ||
| 37 | int i; \ | ||
| 38 | \ | ||
| 39 | for(i=0; i<64; i++){ \ | ||
| 40 | qmat[i]=4; \ | ||
| 41 | tmp[i]= dst[i]; \ | ||
| 42 | } \ | ||
| 43 | ff_prores_idct_put_10_##INSN (dst, 16, tmp, qmat); \ | ||
| 44 | \ | ||
| 45 | for(i=0; i<64; i++) { \ | ||
| 46 | dst[i] -= 512; \ | ||
| 47 | } \ | ||
| 48 | } | ||
| 49 | |||
| 50 |
4/4✓ Branch 0 taken 3840000 times.
✓ Branch 1 taken 60000 times.
✓ Branch 3 taken 3840000 times.
✓ Branch 4 taken 60000 times.
|
7740000 | PR_WRAP(sse2) |
| 51 | |||
| 52 | # if HAVE_AVX_EXTERNAL | ||
| 53 | void ff_prores_idct_put_10_avx(uint16_t *dst, int linesize, | ||
| 54 | int16_t *block, int16_t *qmat); | ||
| 55 |
4/4✓ Branch 0 taken 3840000 times.
✓ Branch 1 taken 60000 times.
✓ Branch 3 taken 3840000 times.
✓ Branch 4 taken 60000 times.
|
7740000 | PR_WRAP(avx) |
| 56 | # endif | ||
| 57 | |||
| 58 | #endif | ||
| 59 | |||
| 60 | static const struct algo fdct_tab_arch[] = { | ||
| 61 | #if HAVE_SSE2_INLINE | ||
| 62 | { "SSE2", ff_fdct_sse2, FF_IDCT_PERM_NONE, AV_CPU_FLAG_SSE2 }, | ||
| 63 | #endif | ||
| 64 | { 0 } | ||
| 65 | }; | ||
| 66 | |||
| 67 | static const struct algo idct_tab_arch[] = { | ||
| 68 | #if CONFIG_MPEG4_DECODER && HAVE_X86ASM | ||
| 69 | #if HAVE_SSE2_EXTERNAL | ||
| 70 | { "XVID-SSE2", ff_xvid_idct_sse2, FF_IDCT_PERM_SSE2, AV_CPU_FLAG_SSE2, 1 }, | ||
| 71 | #endif | ||
| 72 | #endif /* CONFIG_MPEG4_DECODER && HAVE_X86ASM */ | ||
| 73 | #if CONFIG_PRORES_DECODER && ARCH_X86_64 && HAVE_X86ASM | ||
| 74 | { "PR-SSE2", ff_prores_idct_put_10_sse2_wrap, FF_IDCT_PERM_TRANSPOSE, AV_CPU_FLAG_SSE2, 1 }, | ||
| 75 | # if HAVE_AVX_EXTERNAL | ||
| 76 | { "PR-AVX", ff_prores_idct_put_10_avx_wrap, FF_IDCT_PERM_TRANSPOSE, AV_CPU_FLAG_AVX, 1 }, | ||
| 77 | # endif | ||
| 78 | #endif | ||
| 79 | #if HAVE_X86ASM | ||
| 80 | #if ARCH_X86_64 | ||
| 81 | #if HAVE_SSE2_EXTERNAL | ||
| 82 | { "SIMPLE8-SSE2", ff_simple_idct8_sse2, FF_IDCT_PERM_TRANSPOSE, AV_CPU_FLAG_SSE2}, | ||
| 83 | { "SIMPLE10-SSE2", ff_simple_idct10_sse2, FF_IDCT_PERM_TRANSPOSE, AV_CPU_FLAG_SSE2}, | ||
| 84 | { "SIMPLE12-SSE2", ff_simple_idct12_sse2, FF_IDCT_PERM_TRANSPOSE, AV_CPU_FLAG_SSE2, 1 }, | ||
| 85 | #endif | ||
| 86 | #if HAVE_AVX_EXTERNAL | ||
| 87 | { "SIMPLE8-AVX", ff_simple_idct8_avx, FF_IDCT_PERM_TRANSPOSE, AV_CPU_FLAG_AVX}, | ||
| 88 | { "SIMPLE10-AVX", ff_simple_idct10_avx, FF_IDCT_PERM_TRANSPOSE, AV_CPU_FLAG_AVX}, | ||
| 89 | { "SIMPLE12-AVX", ff_simple_idct12_avx, FF_IDCT_PERM_TRANSPOSE, AV_CPU_FLAG_AVX, 1 }, | ||
| 90 | #endif | ||
| 91 | #else | ||
| 92 | #if HAVE_SSE2_EXTERNAL | ||
| 93 | { "SIMPLE-SSE2", ff_simple_idct_sse2, FF_IDCT_PERM_SIMPLE, AV_CPU_FLAG_SSE2}, | ||
| 94 | #endif | ||
| 95 | #endif | ||
| 96 | #endif | ||
| 97 | { 0 } | ||
| 98 | }; | ||
| 99 | |||
| 100 | static const uint8_t idct_simple_mmx_perm[64] = { | ||
| 101 | 0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D, | ||
| 102 | 0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D, | ||
| 103 | 0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D, | ||
| 104 | 0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F, | ||
| 105 | 0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F, | ||
| 106 | 0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D, | ||
| 107 | 0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F, | ||
| 108 | 0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F, | ||
| 109 | }; | ||
| 110 | |||
| 111 | static const uint8_t idct_sse2_row_perm[8] = { 0, 4, 1, 5, 2, 6, 3, 7 }; | ||
| 112 | |||
| 113 | 1120000 | static int permute_x86(int16_t dst[64], const int16_t src[64], | |
| 114 | enum idct_permutation_type perm_type) | ||
| 115 | { | ||
| 116 | int i; | ||
| 117 | |||
| 118 |
2/3✗ Branch 0 not taken.
✓ Branch 1 taken 60000 times.
✓ Branch 2 taken 1060000 times.
|
1120000 | switch (perm_type) { |
| 119 | ✗ | case FF_IDCT_PERM_SIMPLE: | |
| 120 | ✗ | for (i = 0; i < 64; i++) | |
| 121 | ✗ | dst[idct_simple_mmx_perm[i]] = src[i]; | |
| 122 | ✗ | return 1; | |
| 123 | 60000 | case FF_IDCT_PERM_SSE2: | |
| 124 |
2/2✓ Branch 0 taken 3840000 times.
✓ Branch 1 taken 60000 times.
|
3900000 | for (i = 0; i < 64; i++) |
| 125 | 3840000 | dst[(i & 0x38) | idct_sse2_row_perm[i & 7]] = src[i]; | |
| 126 | 60000 | return 1; | |
| 127 | } | ||
| 128 | |||
| 129 | 1060000 | return 0; | |
| 130 | } | ||
| 131 |