|
fukasawa |
e60969 |
|
|
fukasawa |
e60969 |
/* arm_init.c - NEON optimised filter functions
|
|
fukasawa |
e60969 |
*
|
|
fukasawa |
e60969 |
* Copyright (c) 2014 Glenn Randers-Pehrson
|
|
fukasawa |
e60969 |
* Written by Mans Rullgard, 2011.
|
|
fukasawa |
e60969 |
* Last changed in libpng 1.6.16 [December 22, 2014]
|
|
fukasawa |
e60969 |
*
|
|
fukasawa |
e60969 |
* This code is released under the libpng license.
|
|
fukasawa |
e60969 |
* For conditions of distribution and use, see the disclaimer
|
|
fukasawa |
e60969 |
* and license in png.h
|
|
fukasawa |
e60969 |
*/
|
|
fukasawa |
e60969 |
/* Below, after checking __linux__, various non-C90 POSIX 1003.1 functions are
|
|
fukasawa |
e60969 |
* called.
|
|
fukasawa |
e60969 |
*/
|
|
fukasawa |
e60969 |
#define _POSIX_SOURCE 1
|
|
fukasawa |
e60969 |
|
|
fukasawa |
e60969 |
#include "../pngpriv.h"
|
|
fukasawa |
e60969 |
|
|
fukasawa |
e60969 |
#ifdef PNG_READ_SUPPORTED
|
|
fukasawa |
e60969 |
|
|
fukasawa |
e60969 |
#if PNG_ARM_NEON_OPT > 0
|
|
fukasawa |
e60969 |
#ifdef PNG_ARM_NEON_CHECK_SUPPORTED /* Do run-time checks */
|
|
fukasawa |
e60969 |
/* WARNING: it is strongly recommended that you do not build libpng with
|
|
fukasawa |
e60969 |
* run-time checks for CPU features if at all possible. In the case of the ARM
|
|
fukasawa |
e60969 |
* NEON instructions there is no processor-specific way of detecting the
|
|
fukasawa |
e60969 |
* presence of the required support, therefore run-time detection is extremely
|
|
fukasawa |
e60969 |
* OS specific.
|
|
fukasawa |
e60969 |
*
|
|
fukasawa |
e60969 |
* You may set the macro PNG_ARM_NEON_FILE to the file name of file containing
|
|
fukasawa |
e60969 |
* a fragment of C source code which defines the png_have_neon function. There
|
|
fukasawa |
e60969 |
* are a number of implementations in contrib/arm-neon, but the only one that
|
|
fukasawa |
e60969 |
* has partial support is contrib/arm-neon/linux.c - a generic Linux
|
|
fukasawa |
e60969 |
* implementation which reads /proc/cpufino.
|
|
fukasawa |
e60969 |
*/
|
|
fukasawa |
e60969 |
#ifndef PNG_ARM_NEON_FILE
|
|
fukasawa |
e60969 |
# ifdef __linux__
|
|
fukasawa |
e60969 |
# define PNG_ARM_NEON_FILE "contrib/arm-neon/linux.c"
|
|
fukasawa |
e60969 |
# endif
|
|
fukasawa |
e60969 |
#endif
|
|
fukasawa |
e60969 |
|
|
fukasawa |
e60969 |
#ifdef PNG_ARM_NEON_FILE
|
|
fukasawa |
e60969 |
|
|
fukasawa |
e60969 |
#include <signal.h> /* for sig_atomic_t */</signal.h>
|
|
fukasawa |
e60969 |
static int png_have_neon(png_structp png_ptr);
|
|
fukasawa |
e60969 |
#include PNG_ARM_NEON_FILE
|
|
fukasawa |
e60969 |
|
|
fukasawa |
e60969 |
#else /* PNG_ARM_NEON_FILE */
|
|
fukasawa |
e60969 |
# error "PNG_ARM_NEON_FILE undefined: no support for run-time ARM NEON checks"
|
|
fukasawa |
e60969 |
#endif /* PNG_ARM_NEON_FILE */
|
|
fukasawa |
e60969 |
#endif /* PNG_ARM_NEON_CHECK_SUPPORTED */
|
|
fukasawa |
e60969 |
|
|
fukasawa |
e60969 |
#ifndef PNG_ALIGNED_MEMORY_SUPPORTED
|
|
fukasawa |
e60969 |
# error "ALIGNED_MEMORY is required; set: -DPNG_ALIGNED_MEMORY_SUPPORTED"
|
|
fukasawa |
e60969 |
#endif
|
|
fukasawa |
e60969 |
|
|
fukasawa |
e60969 |
void
|
|
fukasawa |
e60969 |
png_init_filter_functions_neon(png_structp pp, unsigned int bpp)
|
|
fukasawa |
e60969 |
{
|
|
fukasawa |
e60969 |
/* The switch statement is compiled in for ARM_NEON_API, the call to
|
|
fukasawa |
e60969 |
* png_have_neon is compiled in for ARM_NEON_CHECK. If both are defined
|
|
fukasawa |
e60969 |
* the check is only performed if the API has not set the NEON option on
|
|
fukasawa |
e60969 |
* or off explicitly. In this case the check controls what happens.
|
|
fukasawa |
e60969 |
*
|
|
fukasawa |
e60969 |
* If the CHECK is not compiled in and the option is UNSET the behavior prior
|
|
fukasawa |
e60969 |
* to 1.6.7 was to use the NEON code - this was a bug caused by having the
|
|
fukasawa |
e60969 |
* wrong order of the 'ON' and 'default' cases. UNSET now defaults to OFF,
|
|
fukasawa |
e60969 |
* as documented in png.h
|
|
fukasawa |
e60969 |
*/
|
|
fukasawa |
e60969 |
#ifdef PNG_ARM_NEON_API_SUPPORTED
|
|
fukasawa |
e60969 |
switch ((pp->options >> PNG_ARM_NEON) & 3)
|
|
fukasawa |
e60969 |
{
|
|
fukasawa |
e60969 |
case PNG_OPTION_UNSET:
|
|
fukasawa |
e60969 |
/* Allow the run-time check to execute if it has been enabled -
|
|
fukasawa |
e60969 |
* thus both API and CHECK can be turned on. If it isn't supported
|
|
fukasawa |
e60969 |
* this case will fall through to the 'default' below, which just
|
|
fukasawa |
e60969 |
* returns.
|
|
fukasawa |
e60969 |
*/
|
|
fukasawa |
e60969 |
#endif /* PNG_ARM_NEON_API_SUPPORTED */
|
|
fukasawa |
e60969 |
#ifdef PNG_ARM_NEON_CHECK_SUPPORTED
|
|
fukasawa |
e60969 |
{
|
|
fukasawa |
e60969 |
static volatile sig_atomic_t no_neon = -1; /* not checked */
|
|
fukasawa |
e60969 |
|
|
fukasawa |
e60969 |
if (no_neon < 0)
|
|
fukasawa |
e60969 |
no_neon = !png_have_neon(pp);
|
|
fukasawa |
e60969 |
|
|
fukasawa |
e60969 |
if (no_neon)
|
|
fukasawa |
e60969 |
return;
|
|
fukasawa |
e60969 |
}
|
|
fukasawa |
e60969 |
#ifdef PNG_ARM_NEON_API_SUPPORTED
|
|
fukasawa |
e60969 |
break;
|
|
fukasawa |
e60969 |
#endif
|
|
fukasawa |
e60969 |
#endif /* PNG_ARM_NEON_CHECK_SUPPORTED */
|
|
fukasawa |
e60969 |
|
|
fukasawa |
e60969 |
#ifdef PNG_ARM_NEON_API_SUPPORTED
|
|
fukasawa |
e60969 |
default: /* OFF or INVALID */
|
|
fukasawa |
e60969 |
return;
|
|
fukasawa |
e60969 |
|
|
fukasawa |
e60969 |
case PNG_OPTION_ON:
|
|
fukasawa |
e60969 |
/* Option turned on */
|
|
fukasawa |
e60969 |
break;
|
|
fukasawa |
e60969 |
}
|
|
fukasawa |
e60969 |
#endif
|
|
fukasawa |
e60969 |
|
|
fukasawa |
e60969 |
/* IMPORTANT: any new external functions used here must be declared using
|
|
fukasawa |
e60969 |
* PNG_INTERNAL_FUNCTION in ../pngpriv.h. This is required so that the
|
|
fukasawa |
e60969 |
* 'prefix' option to configure works:
|
|
fukasawa |
e60969 |
*
|
|
fukasawa |
e60969 |
* ./configure --with-libpng-prefix=foobar_
|
|
fukasawa |
e60969 |
*
|
|
fukasawa |
e60969 |
* Verify you have got this right by running the above command, doing a build
|
|
fukasawa |
e60969 |
* and examining pngprefix.h; it must contain a #define for every external
|
|
fukasawa |
e60969 |
* function you add. (Notice that this happens automatically for the
|
|
fukasawa |
e60969 |
* initialization function.)
|
|
fukasawa |
e60969 |
*/
|
|
fukasawa |
e60969 |
pp->read_filter[PNG_FILTER_VALUE_UP-1] = png_read_filter_row_up_neon;
|
|
fukasawa |
e60969 |
|
|
fukasawa |
e60969 |
if (bpp == 3)
|
|
fukasawa |
e60969 |
{
|
|
fukasawa |
e60969 |
pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub3_neon;
|
|
fukasawa |
e60969 |
pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg3_neon;
|
|
fukasawa |
e60969 |
pp->read_filter[PNG_FILTER_VALUE_PAETH-1] =
|
|
fukasawa |
e60969 |
png_read_filter_row_paeth3_neon;
|
|
fukasawa |
e60969 |
}
|
|
fukasawa |
e60969 |
|
|
fukasawa |
e60969 |
else if (bpp == 4)
|
|
fukasawa |
e60969 |
{
|
|
fukasawa |
e60969 |
pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub4_neon;
|
|
fukasawa |
e60969 |
pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg4_neon;
|
|
fukasawa |
e60969 |
pp->read_filter[PNG_FILTER_VALUE_PAETH-1] =
|
|
fukasawa |
e60969 |
png_read_filter_row_paeth4_neon;
|
|
fukasawa |
e60969 |
}
|
|
fukasawa |
e60969 |
}
|
|
fukasawa |
e60969 |
#endif /* PNG_ARM_NEON_OPT > 0 */
|
|
fukasawa |
e60969 |
#endif /* READ */
|