This will allow for easier implementation of Altivec functions in libraries other than libavcodec.
Showing the changes to libavcodec (15 of 15 files). Show all
Unified · Split
| @@ -24,9 +24,9 @@ · 23 unmodified lines | |||
| 24 | #if HAVE_ALTIVEC_H | 24 | #if HAVE_ALTIVEC_H |
| 25 | #include <altivec.h> | 25 | #include <altivec.h> |
| 26 | #endif | 26 | #endif |
| 27 | #include "libavutil/ppc/types_altivec.h" | ||
| 28 | #include "libavutil/ppc/util_altivec.h" | ||
| 27 | #include "libavcodec/dsputil.h" | 29 | #include "libavcodec/dsputil.h" |
| 28 | #include "util_altivec.h" | ||
| 29 | #include "types_altivec.h" | ||
| 30 | #include "dsputil_altivec.h" | 30 | #include "dsputil_altivec.h" |
| 31 | 31 | ||
| 32 | static int sad16_x2_altivec(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) | 32 | static int sad16_x2_altivec(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
| @@ -19,9 +19,10 @@ · 18 unmodified lines | |||
| 19 | * License along with Libav; if not, write to the Free Software | 19 | * License along with Libav; if not, write to the Free Software |
| 20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | 20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 21 | */ | 21 | */ |
| 22 | |||
| 23 | #include "libavutil/ppc/types_altivec.h" | ||
| 24 | #include "libavutil/ppc/util_altivec.h" | ||
| 22 | #include "libavcodec/fft.h" | 25 | #include "libavcodec/fft.h" |
| 23 | #include "util_altivec.h" | ||
| 24 | #include "types_altivec.h" | ||
| 25 | 26 | ||
| 26 | /** | 27 | /** |
| 27 | * Do a complex FFT with the parameters defined in ff_fft_init(). The | 28 | * Do a complex FFT with the parameters defined in ff_fft_init(). The |
| @@ -18,10 +18,10 @@ · 17 unmodified lines | |||
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 19 | */ | 19 | */ |
| 20 | 20 | ||
| 21 | #include "libavutil/ppc/util_altivec.h" | ||
| 21 | #include "libavcodec/dsputil.h" | 22 | #include "libavcodec/dsputil.h" |
| 22 | 23 | ||
| 23 | #include "dsputil_altivec.h" | 24 | #include "dsputil_altivec.h" |
| 24 | #include "util_altivec.h" | ||
| 25 | 25 | ||
| 26 | static void vector_fmul_altivec(float *dst, const float *src0, const float *src1, int len) | 26 | static void vector_fmul_altivec(float *dst, const float *src0, const float *src1, int len) |
| 27 | { | 27 | { |
| @@ -20,8 +20,8 @@ · 19 unmodified lines | |||
| 20 | 20 | ||
| 21 | #include "libavcodec/fmtconvert.h" | 21 | #include "libavcodec/fmtconvert.h" |
| 22 | 22 | ||
| 23 | #include "libavutil/ppc/util_altivec.h" | ||
| 23 | #include "dsputil_altivec.h" | 24 | #include "dsputil_altivec.h" |
| 24 | #include "util_altivec.h" | ||
| 25 | 25 | ||
| 26 | static void int32_to_float_fmul_scalar_altivec(float *dst, const int *src, float mul, int len) | 26 | static void int32_to_float_fmul_scalar_altivec(float *dst, const int *src, float mul, int len) |
| 27 | { | 27 | { |
| @@ -20,9 +20,9 @@ · 19 unmodified lines | |||
| 20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | 20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 21 | */ | 21 | */ |
| 22 | 22 | ||
| 23 | #include "libavutil/ppc/types_altivec.h" | ||
| 24 | #include "libavutil/ppc/util_altivec.h" | ||
| 23 | #include "libavcodec/dsputil.h" | 25 | #include "libavcodec/dsputil.h" |
| 24 | #include "util_altivec.h" | ||
| 25 | #include "types_altivec.h" | ||
| 26 | #include "dsputil_altivec.h" | 26 | #include "dsputil_altivec.h" |
| 27 | 27 | ||
| 28 | /* | 28 | /* |
| @@ -19,13 +19,13 @@ · 18 unmodified lines | |||
| 19 | */ | 19 | */ |
| 20 | 20 | ||
| 21 | #include "libavutil/cpu.h" | 21 | #include "libavutil/cpu.h" |
| 22 | #include "libavutil/ppc/types_altivec.h" | ||
| 23 | #include "libavutil/ppc/util_altivec.h" | ||
| 22 | #include "libavcodec/dsputil.h" | 24 | #include "libavcodec/dsputil.h" |
| 23 | #include "libavcodec/h264data.h" | 25 | #include "libavcodec/h264data.h" |
| 24 | #include "libavcodec/h264dsp.h" | 26 | #include "libavcodec/h264dsp.h" |
| 25 | 27 | ||
| 26 | #include "dsputil_altivec.h" | 28 | #include "dsputil_altivec.h" |
| 27 | #include "util_altivec.h" | ||
| 28 | #include "types_altivec.h" | ||
| 29 | 29 | ||
| 30 | #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s | 30 | #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s |
| 31 | #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s) | 31 | #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s) |
| @@ -41,8 +41,8 @@ · 40 unmodified lines | |||
| 41 | #if HAVE_ALTIVEC_H | 41 | #if HAVE_ALTIVEC_H |
| 42 | #include <altivec.h> | 42 | #include <altivec.h> |
| 43 | #endif | 43 | #endif |
| 44 | #include "libavutil/ppc/types_altivec.h" | ||
| 44 | #include "libavcodec/dsputil.h" | 45 | #include "libavcodec/dsputil.h" |
| 45 | #include "types_altivec.h" | ||
| 46 | #include "dsputil_altivec.h" | 46 | #include "dsputil_altivec.h" |
| 47 | 47 | ||
| 48 | #define IDCT_HALF \ | 48 | #define IDCT_HALF \ |
| @@ -28,12 +28,11 @@ · 27 unmodified lines | |||
| 28 | #include <altivec.h> | 28 | #include <altivec.h> |
| 29 | #endif | 29 | #endif |
| 30 | 30 | ||
| 31 | #include "libavutil/ppc/types_altivec.h" | ||
| 31 | #include "libavcodec/dsputil.h" | 32 | #include "libavcodec/dsputil.h" |
| 32 | 33 | ||
| 33 | #include "dsputil_altivec.h" | 34 | #include "dsputil_altivec.h" |
| 34 | 35 | ||
| 35 | #include "types_altivec.h" | ||
| 36 | |||
| 37 | static int ssd_int8_vs_int16_altivec(const int8_t *pix1, const int16_t *pix2, | 36 | static int ssd_int8_vs_int16_altivec(const int8_t *pix1, const int16_t *pix2, |
| 38 | int size) { | 37 | int size) { |
| 39 | int i, size16; | 38 | int i, size16; |
| @@ -20,7 +20,7 @@ · 19 unmodified lines | |||
| 20 | */ | 20 | */ |
| 21 | 21 | ||
| 22 | #include "dsputil_altivec.h" | 22 | #include "dsputil_altivec.h" |
| 23 | #include "util_altivec.h" | 23 | #include "libavutil/ppc/util_altivec.h" |
| 24 | #include "libavcodec/dsputil.h" | 24 | #include "libavcodec/dsputil.h" |
| 25 | #include "libavcodec/mpegaudiodsp.h" | 25 | #include "libavcodec/mpegaudiodsp.h" |
| 26 | 26 | ||
| @@ -23,12 +23,13 @@ · 22 unmodified lines | |||
| 23 | 23 | ||
| 24 | #include <stdlib.h> | 24 | #include <stdlib.h> |
| 25 | #include <stdio.h> | 25 | #include <stdio.h> |
| 26 | |||
| 26 | #include "libavutil/cpu.h" | 27 | #include "libavutil/cpu.h" |
| 28 | #include "libavutil/ppc/types_altivec.h" | ||
| 29 | #include "libavutil/ppc/util_altivec.h" | ||
| 27 | #include "libavcodec/dsputil.h" | 30 | #include "libavcodec/dsputil.h" |
| 28 | #include "libavcodec/mpegvideo.h" | 31 | #include "libavcodec/mpegvideo.h" |
| 29 | 32 | ||
| 30 | #include "util_altivec.h" | ||
| 31 | #include "types_altivec.h" | ||
| 32 | #include "dsputil_altivec.h" | 33 | #include "dsputil_altivec.h" |
| 33 | 34 | ||
| 34 | /* AltiVec version of dct_unquantize_h263 | 35 | /* AltiVec version of dct_unquantize_h263 |
| @@ -1,47 +0,0 @@ | |||
| 1 | /* | ||
| 2 | * Copyright (c) 2006 Guillaume Poirier <gpoirier@mplayerhq.hu> | ||
| 3 | * | ||
| 4 | * This file is part of Libav. | ||
| 5 | * | ||
| 6 | * Libav is free software; you can redistribute it and/or | ||
| 7 | * modify it under the terms of the GNU Lesser General Public | ||
| 8 | * License as published by the Free Software Foundation; either | ||
| 9 | * version 2.1 of the License, or (at your option) any later version. | ||
| 10 | * | ||
| 11 | * Libav is distributed in the hope that it will be useful, | ||
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 14 | * Lesser General Public License for more details. | ||
| 15 | * | ||
| 16 | * You should have received a copy of the GNU Lesser General Public | ||
| 17 | * License along with Libav; if not, write to the Free Software | ||
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 19 | */ | ||
| 20 | |||
| 21 | #ifndef AVCODEC_PPC_TYPES_ALTIVEC_H | ||
| 22 | #define AVCODEC_PPC_TYPES_ALTIVEC_H | ||
| 23 | |||
| 24 | /*********************************************************************** | ||
| 25 | * Vector types | ||
| 26 | **********************************************************************/ | ||
| 27 | #define vec_u8 vector unsigned char | ||
| 28 | #define vec_s8 vector signed char | ||
| 29 | #define vec_u16 vector unsigned short | ||
| 30 | #define vec_s16 vector signed short | ||
| 31 | #define vec_u32 vector unsigned int | ||
| 32 | #define vec_s32 vector signed int | ||
| 33 | #define vec_f vector float | ||
| 34 | |||
| 35 | /*********************************************************************** | ||
| 36 | * Null vector | ||
| 37 | **********************************************************************/ | ||
| 38 | #define LOAD_ZERO const vec_u8 zerov = vec_splat_u8( 0 ) | ||
| 39 | |||
| 40 | #define zero_u8v (vec_u8) zerov | ||
| 41 | #define zero_s8v (vec_s8) zerov | ||
| 42 | #define zero_u16v (vec_u16) zerov | ||
| 43 | #define zero_s16v (vec_s16) zerov | ||
| 44 | #define zero_u32v (vec_u32) zerov | ||
| 45 | #define zero_s32v (vec_s32) zerov | ||
| 46 | |||
| 47 | #endif /* AVCODEC_PPC_TYPES_ALTIVEC_H */ | ||
| @@ -1,118 +0,0 @@ | |||
| 1 | /* | ||
| 2 | * This file is part of Libav. | ||
| 3 | * | ||
| 4 | * Libav 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 | * Libav 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 Libav; if not, write to the Free Software | ||
| 16 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 17 | */ | ||
| 18 | |||
| 19 | /** | ||
| 20 | * @file | ||
| 21 | * Contains misc utility macros and inline functions | ||
| 22 | */ | ||
| 23 | |||
| 24 | #ifndef AVCODEC_PPC_UTIL_ALTIVEC_H | ||
| 25 | #define AVCODEC_PPC_UTIL_ALTIVEC_H | ||
| 26 | |||
| 27 | #include <stdint.h> | ||
| 28 | |||
| 29 | #include "config.h" | ||
| 30 | |||
| 31 | #if HAVE_ALTIVEC_H | ||
| 32 | #include <altivec.h> | ||
| 33 | #endif | ||
| 34 | |||
| 35 | #include "types_altivec.h" | ||
| 36 | |||
| 37 | // used to build registers permutation vectors (vcprm) | ||
| 38 | // the 's' are for words in the _s_econd vector | ||
| 39 | #define WORD_0 0x00,0x01,0x02,0x03 | ||
| 40 | #define WORD_1 0x04,0x05,0x06,0x07 | ||
| 41 | #define WORD_2 0x08,0x09,0x0a,0x0b | ||
| 42 | #define WORD_3 0x0c,0x0d,0x0e,0x0f | ||
| 43 | #define WORD_s0 0x10,0x11,0x12,0x13 | ||
| 44 | #define WORD_s1 0x14,0x15,0x16,0x17 | ||
| 45 | #define WORD_s2 0x18,0x19,0x1a,0x1b | ||
| 46 | #define WORD_s3 0x1c,0x1d,0x1e,0x1f | ||
| 47 | |||
| 48 | #define vcprm(a,b,c,d) (const vector unsigned char){WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d} | ||
| 49 | #define vcii(a,b,c,d) (const vector float){FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d} | ||
| 50 | |||
| 51 | // vcprmle is used to keep the same index as in the SSE version. | ||
| 52 | // it's the same as vcprm, with the index inversed | ||
| 53 | // ('le' is Little Endian) | ||
| 54 | #define vcprmle(a,b,c,d) vcprm(d,c,b,a) | ||
| 55 | |||
| 56 | // used to build inverse/identity vectors (vcii) | ||
| 57 | // n is _n_egative, p is _p_ositive | ||
| 58 | #define FLOAT_n -1. | ||
| 59 | #define FLOAT_p 1. | ||
| 60 | |||
| 61 | |||
| 62 | // Transpose 8x8 matrix of 16-bit elements (in-place) | ||
| 63 | #define TRANSPOSE8(a,b,c,d,e,f,g,h) \ | ||
| 64 | do { \ | ||
| 65 | vector signed short A1, B1, C1, D1, E1, F1, G1, H1; \ | ||
| 66 | vector signed short A2, B2, C2, D2, E2, F2, G2, H2; \ | ||
| 67 | \ | ||
| 68 | A1 = vec_mergeh (a, e); \ | ||
| 69 | B1 = vec_mergel (a, e); \ | ||
| 70 | C1 = vec_mergeh (b, f); \ | ||
| 71 | D1 = vec_mergel (b, f); \ | ||
| 72 | E1 = vec_mergeh (c, g); \ | ||
| 73 | F1 = vec_mergel (c, g); \ | ||
| 74 | G1 = vec_mergeh (d, h); \ | ||
| 75 | H1 = vec_mergel (d, h); \ | ||
| 76 | \ | ||
| 77 | A2 = vec_mergeh (A1, E1); \ | ||
| 78 | B2 = vec_mergel (A1, E1); \ | ||
| 79 | C2 = vec_mergeh (B1, F1); \ | ||
| 80 | D2 = vec_mergel (B1, F1); \ | ||
| 81 | E2 = vec_mergeh (C1, G1); \ | ||
| 82 | F2 = vec_mergel (C1, G1); \ | ||
| 83 | G2 = vec_mergeh (D1, H1); \ | ||
| 84 | H2 = vec_mergel (D1, H1); \ | ||
| 85 | \ | ||
| 86 | a = vec_mergeh (A2, E2); \ | ||
| 87 | b = vec_mergel (A2, E2); \ | ||
| 88 | c = vec_mergeh (B2, F2); \ | ||
| 89 | d = vec_mergel (B2, F2); \ | ||
| 90 | e = vec_mergeh (C2, G2); \ | ||
| 91 | f = vec_mergel (C2, G2); \ | ||
| 92 | g = vec_mergeh (D2, H2); \ | ||
| 93 | h = vec_mergel (D2, H2); \ | ||
| 94 | } while (0) | ||
| 95 | |||
| 96 | |||
| 97 | /** @brief loads unaligned vector @a *src with offset @a offset | ||
| 98 | and returns it */ | ||
| 99 | static inline vector unsigned char unaligned_load(int offset, uint8_t *src) | ||
| 100 | { | ||
| 101 | register vector unsigned char first = vec_ld(offset, src); | ||
| 102 | register vector unsigned char second = vec_ld(offset+15, src); | ||
| 103 | register vector unsigned char mask = vec_lvsl(offset, src); | ||
| 104 | return vec_perm(first, second, mask); | ||
| 105 | } | ||
| 106 | |||
| 107 | /** | ||
| 108 | * loads vector known misalignment | ||
| 109 | * @param perm_vec the align permute vector to combine the two loads from lvsl | ||
| 110 | */ | ||
| 111 | static inline vec_u8 load_with_perm_vec(int offset, uint8_t *src, vec_u8 perm_vec) | ||
| 112 | { | ||
| 113 | vec_u8 a = vec_ld(offset, src); | ||
| 114 | vec_u8 b = vec_ld(offset+15, src); | ||
| 115 | return vec_perm(a, b, perm_vec); | ||
| 116 | } | ||
| 117 | |||
| 118 | #endif /* AVCODEC_PPC_UTIL_ALTIVEC_H */ | ||
| @@ -19,12 +19,11 @@ · 18 unmodified lines | |||
| 19 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | 19 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 20 | */ | 20 | */ |
| 21 | 21 | ||
| 22 | #include "libavutil/ppc/types_altivec.h" | ||
| 23 | #include "libavutil/ppc/util_altivec.h" | ||
| 22 | #include "libavcodec/dsputil.h" | 24 | #include "libavcodec/dsputil.h" |
| 23 | #include "libavcodec/vc1dsp.h" | 25 | #include "libavcodec/vc1dsp.h" |
| 24 | 26 | ||
| 25 | #include "util_altivec.h" | ||
| 26 | #include "dsputil_altivec.h" | ||
| 27 | |||
| 28 | // main steps of 8x8 transform | 27 | // main steps of 8x8 transform |
| 29 | #define STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_rnd) \ | 28 | #define STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_rnd) \ |
| 30 | do { \ | 29 | do { \ |
| @@ -18,9 +18,9 @@ · 17 unmodified lines | |||
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 19 | */ | 19 | */ |
| 20 | 20 | ||
| 21 | #include "libavutil/ppc/types_altivec.h" | ||
| 22 | #include "libavutil/ppc/util_altivec.h" | ||
| 21 | #include "libavcodec/dsputil.h" | 23 | #include "libavcodec/dsputil.h" |
| 22 | #include "util_altivec.h" | ||
| 23 | #include "types_altivec.h" | ||
| 24 | #include "dsputil_altivec.h" | 24 | #include "dsputil_altivec.h" |
| 25 | 25 | ||
| 26 | static const vec_s16 constants = | 26 | static const vec_s16 constants = |
| @@ -21,10 +21,10 @@ · 20 unmodified lines | |||
| 21 | */ | 21 | */ |
| 22 | 22 | ||
| 23 | #include "libavutil/cpu.h" | 23 | #include "libavutil/cpu.h" |
| 24 | #include "libavutil/ppc/types_altivec.h" | ||
| 25 | #include "libavutil/ppc/util_altivec.h" | ||
| 24 | #include "libavcodec/vp8dsp.h" | 26 | #include "libavcodec/vp8dsp.h" |
| 25 | #include "dsputil_altivec.h" | 27 | #include "dsputil_altivec.h" |
| 26 | #include "types_altivec.h" | ||
| 27 | #include "util_altivec.h" | ||
| 28 | 28 | ||
| 29 | #define REPT4(...) { __VA_ARGS__, __VA_ARGS__, __VA_ARGS__, __VA_ARGS__ } | 29 | #define REPT4(...) { __VA_ARGS__, __VA_ARGS__, __VA_ARGS__, __VA_ARGS__ } |
| 30 | 30 | ||