1b50c217251b086440efcdb273c22f86a06c80cbaChris Craik
2b50c217251b086440efcdb273c22f86a06c80cbaChris Craik/* arm_init.c - NEON optimised filter functions
3b50c217251b086440efcdb273c22f86a06c80cbaChris Craik *
47a055fdaacbcb54d3606638017fb3381f05d96acAlex Naidis * Copyright (c) 2014,2016 Glenn Randers-Pehrson
5b50c217251b086440efcdb273c22f86a06c80cbaChris Craik * Written by Mans Rullgard, 2011.
67a055fdaacbcb54d3606638017fb3381f05d96acAlex Naidis * Last changed in libpng 1.6.22 [May 26, 2016]
7b50c217251b086440efcdb273c22f86a06c80cbaChris Craik *
8b50c217251b086440efcdb273c22f86a06c80cbaChris Craik * This code is released under the libpng license.
9b50c217251b086440efcdb273c22f86a06c80cbaChris Craik * For conditions of distribution and use, see the disclaimer
10b50c217251b086440efcdb273c22f86a06c80cbaChris Craik * and license in png.h
11b50c217251b086440efcdb273c22f86a06c80cbaChris Craik */
12b50c217251b086440efcdb273c22f86a06c80cbaChris Craik/* Below, after checking __linux__, various non-C90 POSIX 1003.1 functions are
13b50c217251b086440efcdb273c22f86a06c80cbaChris Craik * called.
14b50c217251b086440efcdb273c22f86a06c80cbaChris Craik */
15b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#define _POSIX_SOURCE 1
16b50c217251b086440efcdb273c22f86a06c80cbaChris Craik
17b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#include "../pngpriv.h"
18b50c217251b086440efcdb273c22f86a06c80cbaChris Craik
19b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#ifdef PNG_READ_SUPPORTED
209b1fe63dcc7ba076b9730b7bfa031cc0dbc25561Matt Sarett
21b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#if PNG_ARM_NEON_OPT > 0
22b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#ifdef PNG_ARM_NEON_CHECK_SUPPORTED /* Do run-time checks */
23b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari/* WARNING: it is strongly recommended that you do not build libpng with
24b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari * run-time checks for CPU features if at all possible.  In the case of the ARM
25b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari * NEON instructions there is no processor-specific way of detecting the
269b1fe63dcc7ba076b9730b7bfa031cc0dbc25561Matt Sarett * presence of the required support, therefore run-time detection is extremely
27b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari * OS specific.
28b50c217251b086440efcdb273c22f86a06c80cbaChris Craik *
29b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari * You may set the macro PNG_ARM_NEON_FILE to the file name of file containing
30b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari * a fragment of C source code which defines the png_have_neon function.  There
31b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari * are a number of implementations in contrib/arm-neon, but the only one that
32b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari * has partial support is contrib/arm-neon/linux.c - a generic Linux
33b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari * implementation which reads /proc/cpufino.
34b50c217251b086440efcdb273c22f86a06c80cbaChris Craik */
35b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari#ifndef PNG_ARM_NEON_FILE
36b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari#  ifdef __linux__
37b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari#     define PNG_ARM_NEON_FILE "contrib/arm-neon/linux.c"
38b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari#  endif
39b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari#endif
40b50c217251b086440efcdb273c22f86a06c80cbaChris Craik
41b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari#ifdef PNG_ARM_NEON_FILE
42b50c217251b086440efcdb273c22f86a06c80cbaChris Craik
43b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari#include <signal.h> /* for sig_atomic_t */
44b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripuraristatic int png_have_neon(png_structp png_ptr);
45b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari#include PNG_ARM_NEON_FILE
46b50c217251b086440efcdb273c22f86a06c80cbaChris Craik
47b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari#else  /* PNG_ARM_NEON_FILE */
48b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari#  error "PNG_ARM_NEON_FILE undefined: no support for run-time ARM NEON checks"
49b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari#endif /* PNG_ARM_NEON_FILE */
50b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#endif /* PNG_ARM_NEON_CHECK_SUPPORTED */
51b50c217251b086440efcdb273c22f86a06c80cbaChris Craik
52b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#ifndef PNG_ALIGNED_MEMORY_SUPPORTED
53b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#  error "ALIGNED_MEMORY is required; set: -DPNG_ALIGNED_MEMORY_SUPPORTED"
54b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#endif
55b50c217251b086440efcdb273c22f86a06c80cbaChris Craik
56b50c217251b086440efcdb273c22f86a06c80cbaChris Craikvoid
57b50c217251b086440efcdb273c22f86a06c80cbaChris Craikpng_init_filter_functions_neon(png_structp pp, unsigned int bpp)
58b50c217251b086440efcdb273c22f86a06c80cbaChris Craik{
59b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari   /* The switch statement is compiled in for ARM_NEON_API, the call to
60b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari    * png_have_neon is compiled in for ARM_NEON_CHECK.  If both are defined
61b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari    * the check is only performed if the API has not set the NEON option on
62b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari    * or off explicitly.  In this case the check controls what happens.
63b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari    *
64b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari    * If the CHECK is not compiled in and the option is UNSET the behavior prior
65b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari    * to 1.6.7 was to use the NEON code - this was a bug caused by having the
66b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari    * wrong order of the 'ON' and 'default' cases.  UNSET now defaults to OFF,
67b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari    * as documented in png.h
68b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari    */
697a055fdaacbcb54d3606638017fb3381f05d96acAlex Naidis   png_debug(1, "in png_init_filter_functions_neon");
70b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#ifdef PNG_ARM_NEON_API_SUPPORTED
71b50c217251b086440efcdb273c22f86a06c80cbaChris Craik   switch ((pp->options >> PNG_ARM_NEON) & 3)
72b50c217251b086440efcdb273c22f86a06c80cbaChris Craik   {
73b50c217251b086440efcdb273c22f86a06c80cbaChris Craik      case PNG_OPTION_UNSET:
74b50c217251b086440efcdb273c22f86a06c80cbaChris Craik         /* Allow the run-time check to execute if it has been enabled -
75b50c217251b086440efcdb273c22f86a06c80cbaChris Craik          * thus both API and CHECK can be turned on.  If it isn't supported
76b50c217251b086440efcdb273c22f86a06c80cbaChris Craik          * this case will fall through to the 'default' below, which just
77b50c217251b086440efcdb273c22f86a06c80cbaChris Craik          * returns.
78b50c217251b086440efcdb273c22f86a06c80cbaChris Craik          */
79b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#endif /* PNG_ARM_NEON_API_SUPPORTED */
80b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#ifdef PNG_ARM_NEON_CHECK_SUPPORTED
81b50c217251b086440efcdb273c22f86a06c80cbaChris Craik         {
82b50c217251b086440efcdb273c22f86a06c80cbaChris Craik            static volatile sig_atomic_t no_neon = -1; /* not checked */
83b50c217251b086440efcdb273c22f86a06c80cbaChris Craik
84b50c217251b086440efcdb273c22f86a06c80cbaChris Craik            if (no_neon < 0)
85b50c217251b086440efcdb273c22f86a06c80cbaChris Craik               no_neon = !png_have_neon(pp);
86b50c217251b086440efcdb273c22f86a06c80cbaChris Craik
87b50c217251b086440efcdb273c22f86a06c80cbaChris Craik            if (no_neon)
88b50c217251b086440efcdb273c22f86a06c80cbaChris Craik               return;
89b50c217251b086440efcdb273c22f86a06c80cbaChris Craik         }
90b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#ifdef PNG_ARM_NEON_API_SUPPORTED
91b50c217251b086440efcdb273c22f86a06c80cbaChris Craik         break;
92b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#endif
93b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#endif /* PNG_ARM_NEON_CHECK_SUPPORTED */
94b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari
95b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#ifdef PNG_ARM_NEON_API_SUPPORTED
96b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari      default: /* OFF or INVALID */
97b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari         return;
98b478e66e7c2621eef5f465e4629ce642db00716bSireesh Tripurari
99b50c217251b086440efcdb273c22f86a06c80cbaChris Craik      case PNG_OPTION_ON:
100b50c217251b086440efcdb273c22f86a06c80cbaChris Craik         /* Option turned on */
101b50c217251b086440efcdb273c22f86a06c80cbaChris Craik         break;
102b50c217251b086440efcdb273c22f86a06c80cbaChris Craik   }
103b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#endif
104b50c217251b086440efcdb273c22f86a06c80cbaChris Craik
105b50c217251b086440efcdb273c22f86a06c80cbaChris Craik   /* IMPORTANT: any new external functions used here must be declared using
106b50c217251b086440efcdb273c22f86a06c80cbaChris Craik    * PNG_INTERNAL_FUNCTION in ../pngpriv.h.  This is required so that the
107b50c217251b086440efcdb273c22f86a06c80cbaChris Craik    * 'prefix' option to configure works:
108b50c217251b086440efcdb273c22f86a06c80cbaChris Craik    *
109b50c217251b086440efcdb273c22f86a06c80cbaChris Craik    *    ./configure --with-libpng-prefix=foobar_
110b50c217251b086440efcdb273c22f86a06c80cbaChris Craik    *
111b50c217251b086440efcdb273c22f86a06c80cbaChris Craik    * Verify you have got this right by running the above command, doing a build
112b50c217251b086440efcdb273c22f86a06c80cbaChris Craik    * and examining pngprefix.h; it must contain a #define for every external
113b50c217251b086440efcdb273c22f86a06c80cbaChris Craik    * function you add.  (Notice that this happens automatically for the
114b50c217251b086440efcdb273c22f86a06c80cbaChris Craik    * initialization function.)
115b50c217251b086440efcdb273c22f86a06c80cbaChris Craik    */
116b50c217251b086440efcdb273c22f86a06c80cbaChris Craik   pp->read_filter[PNG_FILTER_VALUE_UP-1] = png_read_filter_row_up_neon;
117b50c217251b086440efcdb273c22f86a06c80cbaChris Craik
118b50c217251b086440efcdb273c22f86a06c80cbaChris Craik   if (bpp == 3)
119b50c217251b086440efcdb273c22f86a06c80cbaChris Craik   {
120b50c217251b086440efcdb273c22f86a06c80cbaChris Craik      pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub3_neon;
121b50c217251b086440efcdb273c22f86a06c80cbaChris Craik      pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg3_neon;
122b50c217251b086440efcdb273c22f86a06c80cbaChris Craik      pp->read_filter[PNG_FILTER_VALUE_PAETH-1] =
123b50c217251b086440efcdb273c22f86a06c80cbaChris Craik         png_read_filter_row_paeth3_neon;
124b50c217251b086440efcdb273c22f86a06c80cbaChris Craik   }
125b50c217251b086440efcdb273c22f86a06c80cbaChris Craik
126b50c217251b086440efcdb273c22f86a06c80cbaChris Craik   else if (bpp == 4)
127b50c217251b086440efcdb273c22f86a06c80cbaChris Craik   {
128b50c217251b086440efcdb273c22f86a06c80cbaChris Craik      pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub4_neon;
129b50c217251b086440efcdb273c22f86a06c80cbaChris Craik      pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg4_neon;
130b50c217251b086440efcdb273c22f86a06c80cbaChris Craik      pp->read_filter[PNG_FILTER_VALUE_PAETH-1] =
131b50c217251b086440efcdb273c22f86a06c80cbaChris Craik          png_read_filter_row_paeth4_neon;
132b50c217251b086440efcdb273c22f86a06c80cbaChris Craik   }
133b50c217251b086440efcdb273c22f86a06c80cbaChris Craik}
134b50c217251b086440efcdb273c22f86a06c80cbaChris Craik#endif /* PNG_ARM_NEON_OPT > 0 */
1359b1fe63dcc7ba076b9730b7bfa031cc0dbc25561Matt Sarett#endif /* READ */
136