Chris Craik | b50c217 | 2013-07-29 15:28:30 -0700 | [diff] [blame] | 1 | |
| 2 | /* arm_init.c - NEON optimised filter functions |
| 3 | * |
Sireesh Tripurari | b478e66 | 2014-05-09 15:15:10 +0530 | [diff] [blame^] | 4 | * Copyright (c) 2014 Glenn Randers-Pehrson |
Chris Craik | b50c217 | 2013-07-29 15:28:30 -0700 | [diff] [blame] | 5 | * Written by Mans Rullgard, 2011. |
Sireesh Tripurari | b478e66 | 2014-05-09 15:15:10 +0530 | [diff] [blame^] | 6 | * Last changed in libpng 1.6.10 [March 6, 2014] |
Chris Craik | b50c217 | 2013-07-29 15:28:30 -0700 | [diff] [blame] | 7 | * |
| 8 | * This code is released under the libpng license. |
| 9 | * For conditions of distribution and use, see the disclaimer |
| 10 | * and license in png.h |
| 11 | */ |
| 12 | /* Below, after checking __linux__, various non-C90 POSIX 1003.1 functions are |
| 13 | * called. |
| 14 | */ |
| 15 | #define _POSIX_SOURCE 1 |
| 16 | |
| 17 | #include "../pngpriv.h" |
| 18 | |
| 19 | #ifdef PNG_READ_SUPPORTED |
| 20 | #if PNG_ARM_NEON_OPT > 0 |
| 21 | #ifdef PNG_ARM_NEON_CHECK_SUPPORTED /* Do run-time checks */ |
Sireesh Tripurari | b478e66 | 2014-05-09 15:15:10 +0530 | [diff] [blame^] | 22 | /* WARNING: it is strongly recommended that you do not build libpng with |
| 23 | * run-time checks for CPU features if at all possible. In the case of the ARM |
| 24 | * NEON instructions there is no processor-specific way of detecting the |
| 25 | * presense of the required support, therefore run-time detectioon is extremely |
| 26 | * OS specific. |
Chris Craik | b50c217 | 2013-07-29 15:28:30 -0700 | [diff] [blame] | 27 | * |
Sireesh Tripurari | b478e66 | 2014-05-09 15:15:10 +0530 | [diff] [blame^] | 28 | * You may set the macro PNG_ARM_NEON_FILE to the file name of file containing |
| 29 | * a fragment of C source code which defines the png_have_neon function. There |
| 30 | * are a number of implementations in contrib/arm-neon, but the only one that |
| 31 | * has partial support is contrib/arm-neon/linux.c - a generic Linux |
| 32 | * implementation which reads /proc/cpufino. |
Chris Craik | b50c217 | 2013-07-29 15:28:30 -0700 | [diff] [blame] | 33 | */ |
Sireesh Tripurari | b478e66 | 2014-05-09 15:15:10 +0530 | [diff] [blame^] | 34 | #ifndef PNG_ARM_NEON_FILE |
| 35 | # ifdef __linux__ |
| 36 | # define PNG_ARM_NEON_FILE "contrib/arm-neon/linux.c" |
| 37 | # endif |
| 38 | #endif |
Chris Craik | b50c217 | 2013-07-29 15:28:30 -0700 | [diff] [blame] | 39 | |
Sireesh Tripurari | b478e66 | 2014-05-09 15:15:10 +0530 | [diff] [blame^] | 40 | #ifdef PNG_ARM_NEON_FILE |
Chris Craik | b50c217 | 2013-07-29 15:28:30 -0700 | [diff] [blame] | 41 | |
Sireesh Tripurari | b478e66 | 2014-05-09 15:15:10 +0530 | [diff] [blame^] | 42 | #include <signal.h> /* for sig_atomic_t */ |
| 43 | static int png_have_neon(png_structp png_ptr); |
| 44 | #include PNG_ARM_NEON_FILE |
Chris Craik | b50c217 | 2013-07-29 15:28:30 -0700 | [diff] [blame] | 45 | |
Sireesh Tripurari | b478e66 | 2014-05-09 15:15:10 +0530 | [diff] [blame^] | 46 | #else /* PNG_ARM_NEON_FILE */ |
| 47 | # error "PNG_ARM_NEON_FILE undefined: no support for run-time ARM NEON checks" |
| 48 | #endif /* PNG_ARM_NEON_FILE */ |
Chris Craik | b50c217 | 2013-07-29 15:28:30 -0700 | [diff] [blame] | 49 | #endif /* PNG_ARM_NEON_CHECK_SUPPORTED */ |
| 50 | |
| 51 | #ifndef PNG_ALIGNED_MEMORY_SUPPORTED |
| 52 | # error "ALIGNED_MEMORY is required; set: -DPNG_ALIGNED_MEMORY_SUPPORTED" |
| 53 | #endif |
| 54 | |
| 55 | void |
| 56 | png_init_filter_functions_neon(png_structp pp, unsigned int bpp) |
| 57 | { |
Sireesh Tripurari | b478e66 | 2014-05-09 15:15:10 +0530 | [diff] [blame^] | 58 | /* The switch statement is compiled in for ARM_NEON_API, the call to |
| 59 | * png_have_neon is compiled in for ARM_NEON_CHECK. If both are defined |
| 60 | * the check is only performed if the API has not set the NEON option on |
| 61 | * or off explicitly. In this case the check controls what happens. |
| 62 | * |
| 63 | * If the CHECK is not compiled in and the option is UNSET the behavior prior |
| 64 | * to 1.6.7 was to use the NEON code - this was a bug caused by having the |
| 65 | * wrong order of the 'ON' and 'default' cases. UNSET now defaults to OFF, |
| 66 | * as documented in png.h |
| 67 | */ |
Chris Craik | b50c217 | 2013-07-29 15:28:30 -0700 | [diff] [blame] | 68 | #ifdef PNG_ARM_NEON_API_SUPPORTED |
| 69 | switch ((pp->options >> PNG_ARM_NEON) & 3) |
| 70 | { |
| 71 | case PNG_OPTION_UNSET: |
| 72 | /* Allow the run-time check to execute if it has been enabled - |
| 73 | * thus both API and CHECK can be turned on. If it isn't supported |
| 74 | * this case will fall through to the 'default' below, which just |
| 75 | * returns. |
| 76 | */ |
| 77 | #endif /* PNG_ARM_NEON_API_SUPPORTED */ |
| 78 | #ifdef PNG_ARM_NEON_CHECK_SUPPORTED |
| 79 | { |
| 80 | static volatile sig_atomic_t no_neon = -1; /* not checked */ |
| 81 | |
| 82 | if (no_neon < 0) |
| 83 | no_neon = !png_have_neon(pp); |
| 84 | |
| 85 | if (no_neon) |
| 86 | return; |
| 87 | } |
| 88 | #ifdef PNG_ARM_NEON_API_SUPPORTED |
| 89 | break; |
| 90 | #endif |
| 91 | #endif /* PNG_ARM_NEON_CHECK_SUPPORTED */ |
Sireesh Tripurari | b478e66 | 2014-05-09 15:15:10 +0530 | [diff] [blame^] | 92 | |
Chris Craik | b50c217 | 2013-07-29 15:28:30 -0700 | [diff] [blame] | 93 | #ifdef PNG_ARM_NEON_API_SUPPORTED |
Sireesh Tripurari | b478e66 | 2014-05-09 15:15:10 +0530 | [diff] [blame^] | 94 | default: /* OFF or INVALID */ |
| 95 | return; |
| 96 | |
Chris Craik | b50c217 | 2013-07-29 15:28:30 -0700 | [diff] [blame] | 97 | case PNG_OPTION_ON: |
| 98 | /* Option turned on */ |
| 99 | break; |
Chris Craik | b50c217 | 2013-07-29 15:28:30 -0700 | [diff] [blame] | 100 | } |
| 101 | #endif |
| 102 | |
| 103 | /* IMPORTANT: any new external functions used here must be declared using |
| 104 | * PNG_INTERNAL_FUNCTION in ../pngpriv.h. This is required so that the |
| 105 | * 'prefix' option to configure works: |
| 106 | * |
| 107 | * ./configure --with-libpng-prefix=foobar_ |
| 108 | * |
| 109 | * Verify you have got this right by running the above command, doing a build |
| 110 | * and examining pngprefix.h; it must contain a #define for every external |
| 111 | * function you add. (Notice that this happens automatically for the |
| 112 | * initialization function.) |
| 113 | */ |
| 114 | pp->read_filter[PNG_FILTER_VALUE_UP-1] = png_read_filter_row_up_neon; |
| 115 | |
| 116 | if (bpp == 3) |
| 117 | { |
| 118 | pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub3_neon; |
| 119 | pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg3_neon; |
| 120 | pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = |
| 121 | png_read_filter_row_paeth3_neon; |
| 122 | } |
| 123 | |
| 124 | else if (bpp == 4) |
| 125 | { |
| 126 | pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub4_neon; |
| 127 | pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg4_neon; |
| 128 | pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = |
| 129 | png_read_filter_row_paeth4_neon; |
| 130 | } |
| 131 | } |
| 132 | #endif /* PNG_ARM_NEON_OPT > 0 */ |
| 133 | #endif /* PNG_READ_SUPPORTED */ |