diff options
Diffstat (limited to 'third_party/libpng16')
30 files changed, 2302 insertions, 305 deletions
diff --git a/third_party/libpng16/0001-disable-arm-neon.patch b/third_party/libpng16/0001-disable-arm-neon.patch deleted file mode 100644 index 91616bbf84..0000000000 --- a/third_party/libpng16/0001-disable-arm-neon.patch +++ /dev/null @@ -1,12 +0,0 @@ -diff a/third_party/libpng16/pngpriv.h b/third_party/libpng16/pngpriv.h ---- a/third_party/libpng16/pngpriv.h -+++ b/third_party/libpng16/pngpriv.h -@@ -127,7 +127,7 @@ - */ - # if (defined(__ARM_NEON__) || defined(__ARM_NEON)) && \ - defined(PNG_ALIGNED_MEMORY_SUPPORTED) --# define PNG_ARM_NEON_OPT 2 -+# define PNG_ARM_NEON_OPT 0 - # else - # define PNG_ARM_NEON_OPT 0 - # endif diff --git a/third_party/libpng16/README.pdfium b/third_party/libpng16/README.pdfium index b0e6805537..733867b782 100644 --- a/third_party/libpng16/README.pdfium +++ b/third_party/libpng16/README.pdfium @@ -1,16 +1,19 @@ Name: libpng -URL: http://www.libpng.org/pub/png/libpng.html -Version: 1.6.20 +URL: http://libpng.org/ +Version: 1.6.22 Security Critical: yes License: libpng license +License Android Compatible: yes Description: PNG library. Local Modifications: +Local changes in Chromium's copy of libpng as of https://crrev.com/404379 - see +README.chromium. + pnglibconf.h: a copy of libpng's scripts/pnglibconf.h.prebuilt. -pngprefix.h: manually-created redefinitions to avoid conflicts with chome. +pngprefix.h: manually-created redefinitions to avoid conflicts with Chromium. 0000-build-config.patch: Local build configuration changes. -0001-disable-arm-neon.diff: Disable ARM NEON optimizations. 0002-static-png-gt.patch: Unconditionally use static png_gt() in png.c to avoid compilation warning. diff --git a/third_party/libpng16/arm/arm_init.c b/third_party/libpng16/arm/arm_init.c new file mode 100644 index 0000000000..02df812e77 --- /dev/null +++ b/third_party/libpng16/arm/arm_init.c @@ -0,0 +1,135 @@ + +/* arm_init.c - NEON optimised filter functions + * + * Copyright (c) 2014,2016 Glenn Randers-Pehrson + * Written by Mans Rullgard, 2011. + * Last changed in libpng 1.6.22 [May 26, 2016] + * + * This code is released under the libpng license. + * For conditions of distribution and use, see the disclaimer + * and license in png.h + */ +/* Below, after checking __linux__, various non-C90 POSIX 1003.1 functions are + * called. + */ +#define _POSIX_SOURCE 1 + +#include "../pngpriv.h" + +#ifdef PNG_READ_SUPPORTED + +#if PNG_ARM_NEON_OPT > 0 +#ifdef PNG_ARM_NEON_CHECK_SUPPORTED /* Do run-time checks */ +/* WARNING: it is strongly recommended that you do not build libpng with + * run-time checks for CPU features if at all possible. In the case of the ARM + * NEON instructions there is no processor-specific way of detecting the + * presence of the required support, therefore run-time detection is extremely + * OS specific. + * + * You may set the macro PNG_ARM_NEON_FILE to the file name of file containing + * a fragment of C source code which defines the png_have_neon function. There + * are a number of implementations in contrib/arm-neon, but the only one that + * has partial support is contrib/arm-neon/linux.c - a generic Linux + * implementation which reads /proc/cpufino. + */ +#ifndef PNG_ARM_NEON_FILE +# ifdef __linux__ +# define PNG_ARM_NEON_FILE "contrib/arm-neon/linux.c" +# endif +#endif + +#ifdef PNG_ARM_NEON_FILE + +#include <signal.h> /* for sig_atomic_t */ +static int png_have_neon(png_structp png_ptr); +#include PNG_ARM_NEON_FILE + +#else /* PNG_ARM_NEON_FILE */ +# error "PNG_ARM_NEON_FILE undefined: no support for run-time ARM NEON checks" +#endif /* PNG_ARM_NEON_FILE */ +#endif /* PNG_ARM_NEON_CHECK_SUPPORTED */ + +#ifndef PNG_ALIGNED_MEMORY_SUPPORTED +# error "ALIGNED_MEMORY is required; set: -DPNG_ALIGNED_MEMORY_SUPPORTED" +#endif + +void +png_init_filter_functions_neon(png_structp pp, unsigned int bpp) +{ + /* The switch statement is compiled in for ARM_NEON_API, the call to + * png_have_neon is compiled in for ARM_NEON_CHECK. If both are defined + * the check is only performed if the API has not set the NEON option on + * or off explicitly. In this case the check controls what happens. + * + * If the CHECK is not compiled in and the option is UNSET the behavior prior + * to 1.6.7 was to use the NEON code - this was a bug caused by having the + * wrong order of the 'ON' and 'default' cases. UNSET now defaults to OFF, + * as documented in png.h + */ + png_debug(1, "in png_init_filter_functions_neon"); +#ifdef PNG_ARM_NEON_API_SUPPORTED + switch ((pp->options >> PNG_ARM_NEON) & 3) + { + case PNG_OPTION_UNSET: + /* Allow the run-time check to execute if it has been enabled - + * thus both API and CHECK can be turned on. If it isn't supported + * this case will fall through to the 'default' below, which just + * returns. + */ +#endif /* PNG_ARM_NEON_API_SUPPORTED */ +#ifdef PNG_ARM_NEON_CHECK_SUPPORTED + { + static volatile sig_atomic_t no_neon = -1; /* not checked */ + + if (no_neon < 0) + no_neon = !png_have_neon(pp); + + if (no_neon) + return; + } +#ifdef PNG_ARM_NEON_API_SUPPORTED + break; +#endif +#endif /* PNG_ARM_NEON_CHECK_SUPPORTED */ + +#ifdef PNG_ARM_NEON_API_SUPPORTED + default: /* OFF or INVALID */ + return; + + case PNG_OPTION_ON: + /* Option turned on */ + break; + } +#endif + + /* IMPORTANT: any new external functions used here must be declared using + * PNG_INTERNAL_FUNCTION in ../pngpriv.h. This is required so that the + * 'prefix' option to configure works: + * + * ./configure --with-libpng-prefix=foobar_ + * + * Verify you have got this right by running the above command, doing a build + * and examining pngprefix.h; it must contain a #define for every external + * function you add. (Notice that this happens automatically for the + * initialization function.) + */ + pp->read_filter[PNG_FILTER_VALUE_UP-1] = png_read_filter_row_up_neon; + + if (bpp == 3) + { + pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub3_neon; + pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg3_neon; + pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = + png_read_filter_row_paeth3_neon; + } + + else if (bpp == 4) + { + pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub4_neon; + pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg4_neon; + pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = + png_read_filter_row_paeth4_neon; + } +} +#endif /* PNG_ARM_NEON_OPT > 0 */ +#endif /* READ */ diff --git a/third_party/libpng16/arm/filter_neon.S b/third_party/libpng16/arm/filter_neon.S new file mode 100644 index 0000000000..3b061d6bbf --- /dev/null +++ b/third_party/libpng16/arm/filter_neon.S @@ -0,0 +1,253 @@ + +/* filter_neon.S - NEON optimised filter functions + * + * Copyright (c) 2014 Glenn Randers-Pehrson + * Written by Mans Rullgard, 2011. + * Last changed in libpng 1.6.16 [December 22, 2014] + * + * This code is released under the libpng license. + * For conditions of distribution and use, see the disclaimer + * and license in png.h + */ + +/* This is required to get the symbol renames, which are #defines, and the + * definitions (or not) of PNG_ARM_NEON_OPT and PNG_ARM_NEON_IMPLEMENTATION. + */ +#define PNG_VERSION_INFO_ONLY +#include "../pngpriv.h" + +#if defined(__linux__) && defined(__ELF__) +.section .note.GNU-stack,"",%progbits /* mark stack as non-executable */ +#endif + +#ifdef PNG_READ_SUPPORTED + +/* Assembler NEON support - only works for 32-bit ARM (i.e. it does not work for + * ARM64). The code in arm/filter_neon_intrinsics.c supports ARM64, however it + * only works if -mfpu=neon is specified on the GCC command line. See pngpriv.h + * for the logic which sets PNG_USE_ARM_NEON_ASM: + */ +#if PNG_ARM_NEON_IMPLEMENTATION == 2 /* hand-coded assembler */ + +#if PNG_ARM_NEON_OPT > 0 + +#ifdef __ELF__ +# define ELF +#else +# define ELF @ +#endif + + .arch armv7-a + .fpu neon + +.macro func name, export=0 + .macro endfunc +ELF .size \name, . - \name + .endfunc + .purgem endfunc + .endm + .text + + /* Explicitly specifying alignment here because some versions of + * GAS don't align code correctly. This is harmless in correctly + * written versions of GAS. + */ + .align 2 + + .if \export + .global \name + .endif +ELF .type \name, STT_FUNC + .func \name +\name: +.endm + +func png_read_filter_row_sub4_neon, export=1 + ldr r3, [r0, #4] @ rowbytes + vmov.i8 d3, #0 +1: + vld4.32 {d4[],d5[],d6[],d7[]}, [r1,:128] + vadd.u8 d0, d3, d4 + vadd.u8 d1, d0, d5 + vadd.u8 d2, d1, d6 + vadd.u8 d3, d2, d7 + vst4.32 {d0[0],d1[0],d2[0],d3[0]},[r1,:128]! + subs r3, r3, #16 + bgt 1b + + bx lr +endfunc + +func png_read_filter_row_sub3_neon, export=1 + ldr r3, [r0, #4] @ rowbytes + vmov.i8 d3, #0 + mov r0, r1 + mov r2, #3 + mov r12, #12 + vld1.8 {q11}, [r0], r12 +1: + vext.8 d5, d22, d23, #3 + vadd.u8 d0, d3, d22 + vext.8 d6, d22, d23, #6 + vadd.u8 d1, d0, d5 + vext.8 d7, d23, d23, #1 + vld1.8 {q11}, [r0], r12 + vst1.32 {d0[0]}, [r1,:32], r2 + vadd.u8 d2, d1, d6 + vst1.32 {d1[0]}, [r1], r2 + vadd.u8 d3, d2, d7 + vst1.32 {d2[0]}, [r1], r2 + vst1.32 {d3[0]}, [r1], r2 + subs r3, r3, #12 + bgt 1b + + bx lr +endfunc + +func png_read_filter_row_up_neon, export=1 + ldr r3, [r0, #4] @ rowbytes +1: + vld1.8 {q0}, [r1,:128] + vld1.8 {q1}, [r2,:128]! + vadd.u8 q0, q0, q1 + vst1.8 {q0}, [r1,:128]! + subs r3, r3, #16 + bgt 1b + + bx lr +endfunc + +func png_read_filter_row_avg4_neon, export=1 + ldr r12, [r0, #4] @ rowbytes + vmov.i8 d3, #0 +1: + vld4.32 {d4[],d5[],d6[],d7[]}, [r1,:128] + vld4.32 {d16[],d17[],d18[],d19[]},[r2,:128]! + vhadd.u8 d0, d3, d16 + vadd.u8 d0, d0, d4 + vhadd.u8 d1, d0, d17 + vadd.u8 d1, d1, d5 + vhadd.u8 d2, d1, d18 + vadd.u8 d2, d2, d6 + vhadd.u8 d3, d2, d19 + vadd.u8 d3, d3, d7 + vst4.32 {d0[0],d1[0],d2[0],d3[0]},[r1,:128]! + subs r12, r12, #16 + bgt 1b + + bx lr +endfunc + +func png_read_filter_row_avg3_neon, export=1 + push {r4,lr} + ldr r12, [r0, #4] @ rowbytes + vmov.i8 d3, #0 + mov r0, r1 + mov r4, #3 + mov lr, #12 + vld1.8 {q11}, [r0], lr +1: + vld1.8 {q10}, [r2], lr + vext.8 d5, d22, d23, #3 + vhadd.u8 d0, d3, d20 + vext.8 d17, d20, d21, #3 + vadd.u8 d0, d0, d22 + vext.8 d6, d22, d23, #6 + vhadd.u8 d1, d0, d17 + vext.8 d18, d20, d21, #6 + vadd.u8 d1, d1, d5 + vext.8 d7, d23, d23, #1 + vld1.8 {q11}, [r0], lr + vst1.32 {d0[0]}, [r1,:32], r4 + vhadd.u8 d2, d1, d18 + vst1.32 {d1[0]}, [r1], r4 + vext.8 d19, d21, d21, #1 + vadd.u8 d2, d2, d6 + vhadd.u8 d3, d2, d19 + vst1.32 {d2[0]}, [r1], r4 + vadd.u8 d3, d3, d7 + vst1.32 {d3[0]}, [r1], r4 + subs r12, r12, #12 + bgt 1b + + pop {r4,pc} +endfunc + +.macro paeth rx, ra, rb, rc + vaddl.u8 q12, \ra, \rb @ a + b + vaddl.u8 q15, \rc, \rc @ 2*c + vabdl.u8 q13, \rb, \rc @ pa + vabdl.u8 q14, \ra, \rc @ pb + vabd.u16 q15, q12, q15 @ pc + vcle.u16 q12, q13, q14 @ pa <= pb + vcle.u16 q13, q13, q15 @ pa <= pc + vcle.u16 q14, q14, q15 @ pb <= pc + vand q12, q12, q13 @ pa <= pb && pa <= pc + vmovn.u16 d28, q14 + vmovn.u16 \rx, q12 + vbsl d28, \rb, \rc + vbsl \rx, \ra, d28 +.endm + +func png_read_filter_row_paeth4_neon, export=1 + ldr r12, [r0, #4] @ rowbytes + vmov.i8 d3, #0 + vmov.i8 d20, #0 +1: + vld4.32 {d4[],d5[],d6[],d7[]}, [r1,:128] + vld4.32 {d16[],d17[],d18[],d19[]},[r2,:128]! + paeth d0, d3, d16, d20 + vadd.u8 d0, d0, d4 + paeth d1, d0, d17, d16 + vadd.u8 d1, d1, d5 + paeth d2, d1, d18, d17 + vadd.u8 d2, d2, d6 + paeth d3, d2, d19, d18 + vmov d20, d19 + vadd.u8 d3, d3, d7 + vst4.32 {d0[0],d1[0],d2[0],d3[0]},[r1,:128]! + subs r12, r12, #16 + bgt 1b + + bx lr +endfunc + +func png_read_filter_row_paeth3_neon, export=1 + push {r4,lr} + ldr r12, [r0, #4] @ rowbytes + vmov.i8 d3, #0 + vmov.i8 d4, #0 + mov r0, r1 + mov r4, #3 + mov lr, #12 + vld1.8 {q11}, [r0], lr +1: + vld1.8 {q10}, [r2], lr + paeth d0, d3, d20, d4 + vext.8 d5, d22, d23, #3 + vadd.u8 d0, d0, d22 + vext.8 d17, d20, d21, #3 + paeth d1, d0, d17, d20 + vst1.32 {d0[0]}, [r1,:32], r4 + vext.8 d6, d22, d23, #6 + vadd.u8 d1, d1, d5 + vext.8 d18, d20, d21, #6 + paeth d2, d1, d18, d17 + vext.8 d7, d23, d23, #1 + vld1.8 {q11}, [r0], lr + vst1.32 {d1[0]}, [r1], r4 + vadd.u8 d2, d2, d6 + vext.8 d19, d21, d21, #1 + paeth d3, d2, d19, d18 + vst1.32 {d2[0]}, [r1], r4 + vmov d4, d19 + vadd.u8 d3, d3, d7 + vst1.32 {d3[0]}, [r1], r4 + subs r12, r12, #12 + bgt 1b + + pop {r4,pc} +endfunc +#endif /* PNG_ARM_NEON_OPT > 0 */ +#endif /* PNG_ARM_NEON_IMPLEMENTATION == 2 (assembler) */ +#endif /* READ */ diff --git a/third_party/libpng16/arm/filter_neon_intrinsics.c b/third_party/libpng16/arm/filter_neon_intrinsics.c new file mode 100644 index 0000000000..ea7e356bcc --- /dev/null +++ b/third_party/libpng16/arm/filter_neon_intrinsics.c @@ -0,0 +1,387 @@ + +/* filter_neon_intrinsics.c - NEON optimised filter functions + * + * Copyright (c) 2014,2016 Glenn Randers-Pehrson + * Written by James Yu <james.yu at linaro.org>, October 2013. + * Based on filter_neon.S, written by Mans Rullgard, 2011. + * + * Last changed in libpng 1.6.22 [May 26, 2016] + * + * This code is released under the libpng license. + * For conditions of distribution and use, see the disclaimer + * and license in png.h + */ + +#include "../pngpriv.h" + +#ifdef PNG_READ_SUPPORTED + +/* This code requires -mfpu=neon on the command line: */ +#if PNG_ARM_NEON_IMPLEMENTATION == 1 /* intrinsics code from pngpriv.h */ + +#include <arm_neon.h> + +/* libpng row pointers are not necessarily aligned to any particular boundary, + * however this code will only work with appropriate alignment. arm/arm_init.c + * checks for this (and will not compile unless it is done). This code uses + * variants of png_aligncast to avoid compiler warnings. + */ +#define png_ptr(type,pointer) png_aligncast(type *,pointer) +#define png_ptrc(type,pointer) png_aligncastconst(const type *,pointer) + +/* The following relies on a variable 'temp_pointer' being declared with type + * 'type'. This is written this way just to hide the GCC strict aliasing + * warning; note that the code is safe because there never is an alias between + * the input and output pointers. + */ +#define png_ldr(type,pointer)\ + (temp_pointer = png_ptr(type,pointer), *temp_pointer) + +#if PNG_ARM_NEON_OPT > 0 + +void +png_read_filter_row_up_neon(png_row_infop row_info, png_bytep row, + png_const_bytep prev_row) +{ + png_bytep rp = row; + png_bytep rp_stop = row + row_info->rowbytes; + png_const_bytep pp = prev_row; + + png_debug(1, "in png_read_filter_row_up_neon"); + + for (; rp < rp_stop; rp += 16, pp += 16) + { + uint8x16_t qrp, qpp; + + qrp = vld1q_u8(rp); + qpp = vld1q_u8(pp); + qrp = vaddq_u8(qrp, qpp); + vst1q_u8(rp, qrp); + } +} + +void +png_read_filter_row_sub3_neon(png_row_infop row_info, png_bytep row, + png_const_bytep prev_row) +{ + png_bytep rp = row; + png_bytep rp_stop = row + row_info->rowbytes; + + uint8x16_t vtmp = vld1q_u8(rp); + uint8x8x2_t *vrpt = png_ptr(uint8x8x2_t, &vtmp); + uint8x8x2_t vrp = *vrpt; + + uint8x8x4_t vdest; + vdest.val[3] = vdup_n_u8(0); + + png_debug(1, "in png_read_filter_row_sub3_neon"); + + for (; rp < rp_stop;) + { + uint8x8_t vtmp1, vtmp2; + uint32x2_t *temp_pointer; + + vtmp1 = vext_u8(vrp.val[0], vrp.val[1], 3); + vdest.val[0] = vadd_u8(vdest.val[3], vrp.val[0]); + vtmp2 = vext_u8(vrp.val[0], vrp.val[1], 6); + vdest.val[1] = vadd_u8(vdest.val[0], vtmp1); + + vtmp1 = vext_u8(vrp.val[1], vrp.val[1], 1); + vdest.val[2] = vadd_u8(vdest.val[1], vtmp2); + vdest.val[3] = vadd_u8(vdest.val[2], vtmp1); + + vtmp = vld1q_u8(rp + 12); + vrpt = png_ptr(uint8x8x2_t, &vtmp); + vrp = *vrpt; + + vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[0]), 0); + rp += 3; + vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[1]), 0); + rp += 3; + vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[2]), 0); + rp += 3; + vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[3]), 0); + rp += 3; + } + + PNG_UNUSED(prev_row) +} + +void +png_read_filter_row_sub4_neon(png_row_infop row_info, png_bytep row, + png_const_bytep prev_row) +{ + png_bytep rp = row; + png_bytep rp_stop = row + row_info->rowbytes; + + uint8x8x4_t vdest; + vdest.val[3] = vdup_n_u8(0); + + png_debug(1, "in png_read_filter_row_sub4_neon"); + + for (; rp < rp_stop; rp += 16) + { + uint32x2x4_t vtmp = vld4_u32(png_ptr(uint32_t,rp)); + uint8x8x4_t *vrpt = png_ptr(uint8x8x4_t,&vtmp); + uint8x8x4_t vrp = *vrpt; + uint32x2x4_t *temp_pointer; + + vdest.val[0] = vadd_u8(vdest.val[3], vrp.val[0]); + vdest.val[1] = vadd_u8(vdest.val[0], vrp.val[1]); + vdest.val[2] = vadd_u8(vdest.val[1], vrp.val[2]); + vdest.val[3] = vadd_u8(vdest.val[2], vrp.val[3]); + vst4_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2x4_t,&vdest), 0); + } + + PNG_UNUSED(prev_row) +} + +void +png_read_filter_row_avg3_neon(png_row_infop row_info, png_bytep row, + png_const_bytep prev_row) +{ + png_bytep rp = row; + png_const_bytep pp = prev_row; + png_bytep rp_stop = row + row_info->rowbytes; + + uint8x16_t vtmp; + uint8x8x2_t *vrpt; + uint8x8x2_t vrp; + uint8x8x4_t vdest; + vdest.val[3] = vdup_n_u8(0); + + vtmp = vld1q_u8(rp); + vrpt = png_ptr(uint8x8x2_t,&vtmp); + vrp = *vrpt; + + png_debug(1, "in png_read_filter_row_avg3_neon"); + + for (; rp < rp_stop; pp += 12) + { + uint8x8_t vtmp1, vtmp2, vtmp3; + + uint8x8x2_t *vppt; + uint8x8x2_t vpp; + + uint32x2_t *temp_pointer; + + vtmp = vld1q_u8(pp); + vppt = png_ptr(uint8x8x2_t,&vtmp); + vpp = *vppt; + + vtmp1 = vext_u8(vrp.val[0], vrp.val[1], 3); + vdest.val[0] = vhadd_u8(vdest.val[3], vpp.val[0]); + vdest.val[0] = vadd_u8(vdest.val[0], vrp.val[0]); + + vtmp2 = vext_u8(vpp.val[0], vpp.val[1], 3); + vtmp3 = vext_u8(vrp.val[0], vrp.val[1], 6); + vdest.val[1] = vhadd_u8(vdest.val[0], vtmp2); + vdest.val[1] = vadd_u8(vdest.val[1], vtmp1); + + vtmp2 = vext_u8(vpp.val[0], vpp.val[1], 6); + vtmp1 = vext_u8(vrp.val[1], vrp.val[1], 1); + + vtmp = vld1q_u8(rp + 12); + vrpt = png_ptr(uint8x8x2_t,&vtmp); + vrp = *vrpt; + + vdest.val[2] = vhadd_u8(vdest.val[1], vtmp2); + vdest.val[2] = vadd_u8(vdest.val[2], vtmp3); + + vtmp2 = vext_u8(vpp.val[1], vpp.val[1], 1); + + vdest.val[3] = vhadd_u8(vdest.val[2], vtmp2); + vdest.val[3] = vadd_u8(vdest.val[3], vtmp1); + + vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[0]), 0); + rp += 3; + vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[1]), 0); + rp += 3; + vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[2]), 0); + rp += 3; + vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[3]), 0); + rp += 3; + } +} + +void +png_read_filter_row_avg4_neon(png_row_infop row_info, png_bytep row, + png_const_bytep prev_row) +{ + png_bytep rp = row; + png_bytep rp_stop = row + row_info->rowbytes; + png_const_bytep pp = prev_row; + + uint8x8x4_t vdest; + vdest.val[3] = vdup_n_u8(0); + + png_debug(1, "in png_read_filter_row_avg4_neon"); + + for (; rp < rp_stop; rp += 16, pp += 16) + { + uint32x2x4_t vtmp; + uint8x8x4_t *vrpt, *vppt; + uint8x8x4_t vrp, vpp; + uint32x2x4_t *temp_pointer; + + vtmp = vld4_u32(png_ptr(uint32_t,rp)); + vrpt = png_ptr(uint8x8x4_t,&vtmp); + vrp = *vrpt; + vtmp = vld4_u32(png_ptrc(uint32_t,pp)); + vppt = png_ptr(uint8x8x4_t,&vtmp); + vpp = *vppt; + + vdest.val[0] = vhadd_u8(vdest.val[3], vpp.val[0]); + vdest.val[0] = vadd_u8(vdest.val[0], vrp.val[0]); + vdest.val[1] = vhadd_u8(vdest.val[0], vpp.val[1]); + vdest.val[1] = vadd_u8(vdest.val[1], vrp.val[1]); + vdest.val[2] = vhadd_u8(vdest.val[1], vpp.val[2]); + vdest.val[2] = vadd_u8(vdest.val[2], vrp.val[2]); + vdest.val[3] = vhadd_u8(vdest.val[2], vpp.val[3]); + vdest.val[3] = vadd_u8(vdest.val[3], vrp.val[3]); + + vst4_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2x4_t,&vdest), 0); + } +} + +static uint8x8_t +paeth(uint8x8_t a, uint8x8_t b, uint8x8_t c) +{ + uint8x8_t d, e; + uint16x8_t p1, pa, pb, pc; + + p1 = vaddl_u8(a, b); /* a + b */ + pc = vaddl_u8(c, c); /* c * 2 */ + pa = vabdl_u8(b, c); /* pa */ + pb = vabdl_u8(a, c); /* pb */ + pc = vabdq_u16(p1, pc); /* pc */ + + p1 = vcleq_u16(pa, pb); /* pa <= pb */ + pa = vcleq_u16(pa, pc); /* pa <= pc */ + pb = vcleq_u16(pb, pc); /* pb <= pc */ + + p1 = vandq_u16(p1, pa); /* pa <= pb && pa <= pc */ + + d = vmovn_u16(pb); + e = vmovn_u16(p1); + + d = vbsl_u8(d, b, c); + e = vbsl_u8(e, a, d); + + return e; +} + +void +png_read_filter_row_paeth3_neon(png_row_infop row_info, png_bytep row, + png_const_bytep prev_row) +{ + png_bytep rp = row; + png_const_bytep pp = prev_row; + png_bytep rp_stop = row + row_info->rowbytes; + + uint8x16_t vtmp; + uint8x8x2_t *vrpt; + uint8x8x2_t vrp; + uint8x8_t vlast = vdup_n_u8(0); + uint8x8x4_t vdest; + vdest.val[3] = vdup_n_u8(0); + + vtmp = vld1q_u8(rp); + vrpt = png_ptr(uint8x8x2_t,&vtmp); + vrp = *vrpt; + + png_debug(1, "in png_read_filter_row_paeth3_neon"); + + for (; rp < rp_stop; pp += 12) + { + uint8x8x2_t *vppt; + uint8x8x2_t vpp; + uint8x8_t vtmp1, vtmp2, vtmp3; + uint32x2_t *temp_pointer; + + vtmp = vld1q_u8(pp); + vppt = png_ptr(uint8x8x2_t,&vtmp); + vpp = *vppt; + + vdest.val[0] = paeth(vdest.val[3], vpp.val[0], vlast); + vdest.val[0] = vadd_u8(vdest.val[0], vrp.val[0]); + + vtmp1 = vext_u8(vrp.val[0], vrp.val[1], 3); + vtmp2 = vext_u8(vpp.val[0], vpp.val[1], 3); + vdest.val[1] = paeth(vdest.val[0], vtmp2, vpp.val[0]); + vdest.val[1] = vadd_u8(vdest.val[1], vtmp1); + + vtmp1 = vext_u8(vrp.val[0], vrp.val[1], 6); + vtmp3 = vext_u8(vpp.val[0], vpp.val[1], 6); + vdest.val[2] = paeth(vdest.val[1], vtmp3, vtmp2); + vdest.val[2] = vadd_u8(vdest.val[2], vtmp1); + + vtmp1 = vext_u8(vrp.val[1], vrp.val[1], 1); + vtmp2 = vext_u8(vpp.val[1], vpp.val[1], 1); + + vtmp = vld1q_u8(rp + 12); + vrpt = png_ptr(uint8x8x2_t,&vtmp); + vrp = *vrpt; + + vdest.val[3] = paeth(vdest.val[2], vtmp2, vtmp3); + vdest.val[3] = vadd_u8(vdest.val[3], vtmp1); + + vlast = vtmp2; + + vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[0]), 0); + rp += 3; + vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[1]), 0); + rp += 3; + vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[2]), 0); + rp += 3; + vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[3]), 0); + rp += 3; + } +} + +void +png_read_filter_row_paeth4_neon(png_row_infop row_info, png_bytep row, + png_const_bytep prev_row) +{ + png_bytep rp = row; + png_bytep rp_stop = row + row_info->rowbytes; + png_const_bytep pp = prev_row; + + uint8x8_t vlast = vdup_n_u8(0); + uint8x8x4_t vdest; + vdest.val[3] = vdup_n_u8(0); + + png_debug(1, "in png_read_filter_row_paeth4_neon"); + + for (; rp < rp_stop; rp += 16, pp += 16) + { + uint32x2x4_t vtmp; + uint8x8x4_t *vrpt, *vppt; + uint8x8x4_t vrp, vpp; + uint32x2x4_t *temp_pointer; + + vtmp = vld4_u32(png_ptr(uint32_t,rp)); + vrpt = png_ptr(uint8x8x4_t,&vtmp); + vrp = *vrpt; + vtmp = vld4_u32(png_ptrc(uint32_t,pp)); + vppt = png_ptr(uint8x8x4_t,&vtmp); + vpp = *vppt; + + vdest.val[0] = paeth(vdest.val[3], vpp.val[0], vlast); + vdest.val[0] = vadd_u8(vdest.val[0], vrp.val[0]); + vdest.val[1] = paeth(vdest.val[0], vpp.val[1], vpp.val[0]); + vdest.val[1] = vadd_u8(vdest.val[1], vrp.val[1]); + vdest.val[2] = paeth(vdest.val[1], vpp.val[2], vpp.val[1]); + vdest.val[2] = vadd_u8(vdest.val[2], vrp.val[2]); + vdest.val[3] = paeth(vdest.val[2], vpp.val[3], vpp.val[2]); + vdest.val[3] = vadd_u8(vdest.val[3], vrp.val[3]); + + vlast = vpp.val[3]; + + vst4_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2x4_t,&vdest), 0); + } +} + +#endif /* PNG_ARM_NEON_OPT > 0 */ +#endif /* PNG_ARM_NEON_IMPLEMENTATION == 1 (intrinsics) */ +#endif /* READ */ diff --git a/third_party/libpng16/contrib/intel/INSTALL b/third_party/libpng16/contrib/intel/INSTALL new file mode 100644 index 0000000000..cd5cdd94eb --- /dev/null +++ b/third_party/libpng16/contrib/intel/INSTALL @@ -0,0 +1,158 @@ +Enabling SSE support + +Copyright (c) 2016 Google, Inc. +Written by Mike Klein, Matt Sarett + +This INSTALL file written by Glenn Randers-Pehrson, 2016. + +If you have moved intel_init.c and filter_sse2_intrinsics.c to a different +directory, be sure to update the '#include "../../pngpriv.h"' line in both +files if necessary to point to the correct relative location of pngpriv.h +with respect to the new location of those files. + +To enable SSE support in libpng, follow the instructions in I, II, or III, +below: + +I. Using patched "configure" scripts: + +First, apply intel_sse.patch in your build directory. + + patch -i contrib/intel/intel_sse.patch -p1 + +Then, if you are not building in a new GIT clone, e.g., in a tar +distribution, remove any existing pre-built configure scripts: + + ./configure --enable-maintainer-mode + make maintainer-clean + ./autogen.sh --maintainer --clean + +Finally, configure libpng with -DPNG_INTEL_SSE in CPPFLAGS: + + ./autogen.sh --maintainer + CPPFLAGS="-DPNG_INTEL_SSE" ./configure [options] + make CPPFLAGS="-DPNG_INTEL_SSE" [options] + make + +II. Using a custom makefile: + +If you are using a custom makefile makefile, you will have to update it +manually to include contrib/intel/*.o in the dependencies, and to define +PNG_INTEL_SSE. + +III. Using manually updated "configure" scripts: + +If you prefer, manually edit pngpriv.h, configure.ac, and Makefile.am, +following the instructions below, then follow the instructions in +section II of INSTALL in the main libpng directory, then configure libpng +with -DPNG_INTEL_SSE in CPPFLAGS. + +1. Add the following code to configure.ac under HOST SPECIFIC OPTIONS +directly beneath the section for ARM: + +-----------------cut---------------- +# INTEL +# ===== +# +# INTEL SSE (SIMD) support. + +AC_ARG_ENABLE([intel-sse], + AS_HELP_STRING([[[--enable-intel-sse]]], + [Enable Intel SSE optimizations: =no/off, yes/on:] + [no/off: disable the optimizations;] + [yes/on: enable the optimizations.] + [If not specified: determined by the compiler.]), + [case "$enableval" in + no|off) + # disable the default enabling: + AC_DEFINE([PNG_INTEL_SSE_OPT], [0], + [Disable Intel SSE optimizations]) + # Prevent inclusion of the assembler files below: + enable_intel_sse=no;; + yes|on) + AC_DEFINE([PNG_INTEL_SSE_OPT], [1], + [Enable Intel SSE optimizations]);; + *) + AC_MSG_ERROR([--enable-intel-sse=${enable_intel_sse}: invalid value]) + esac]) + +# Add Intel specific files to all builds where the host_cpu is Intel ('x86*') +# or where Intel optimizations were explicitly requested (this allows a +# fallback if a future host CPU does not match 'x86*') +AM_CONDITIONAL([PNG_INTEL_SSE], + [test "$enable_intel_sse" != 'no' && + case "$host_cpu" in + i?86|x86_64) :;; + *) test "$enable_intel_sse" != '';; + esac]) +-----------------cut---------------- + +2. Add the following code to Makefile.am under HOST SPECIFIC OPTIONS +directly beneath the "if PNG_ARM_NEON ... endif" statement: + +-----------------cut---------------- +if PNG_INTEL_SSE +libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES += contrib/intel/intel_init.c\ + contrib/intel/filter_sse2_intrinsics.c +endif +-----------------cut---------------- + +3. Add the following lines to pngpriv.h, following the PNG_ARM_NEON_OPT +code: + +-----------------cut---------------- +#ifndef PNG_INTEL_SSE_OPT +# ifdef PNG_INTEL_SSE + /* Only check for SSE if the build configuration has been modified to + * enable SSE optimizations. This means that these optimizations will + * be off by default. See contrib/intel for more details. + */ +# if defined(__SSE4_1__) || defined(__AVX__) || defined(__SSSE3__) || \ + defined(__SSE2__) || defined(_M_X64) || defined(_M_AMD64) || \ + (defined(_M_IX86_FP) && _M_IX86_FP >= 2) +# define PNG_INTEL_SSE_OPT 1 +# endif +# endif +#endif + +#if PNG_INTEL_SSE_OPT > 0 +# ifndef PNG_INTEL_SSE_IMPLEMENTATION +# if defined(__SSE4_1__) || defined(__AVX__) + /* We are not actually using AVX, but checking for AVX is the best + way we can detect SSE4.1 and SSSE3 on MSVC. + */ +# define PNG_INTEL_SSE_IMPLEMENTATION 3 +# elif defined(__SSSE3__) +# define PNG_INTEL_SSE_IMPLEMENTATION 2 +# elif defined(__SSE2__) || defined(_M_X64) || defined(_M_AMD64) || \ + (defined(_M_IX86_FP) && _M_IX86_FP >= 2) +# define PNG_INTEL_SSE_IMPLEMENTATION 1 +# else +# define PNG_INTEL_SSE_IMPLEMENTATION 0 +# endif +# endif + +# if PNG_INTEL_SSE_IMPLEMENTATION > 0 +# define PNG_FILTER_OPTIMIZATIONS png_init_filter_functions_sse2 +# endif +#endif + +-----------------cut---------------- + +4. Add the following lines to pngpriv.h, following the prototype for +png_read_filter_row_paeth4_neon: + +-----------------cut---------------- +PNG_INTERNAL_FUNCTION(void,png_read_filter_row_sub3_sse2,(png_row_infop + row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); +PNG_INTERNAL_FUNCTION(void,png_read_filter_row_sub4_sse2,(png_row_infop + row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); +PNG_INTERNAL_FUNCTION(void,png_read_filter_row_avg3_sse2,(png_row_infop + row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); +PNG_INTERNAL_FUNCTION(void,png_read_filter_row_avg4_sse2,(png_row_infop + row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); +PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth3_sse2,(png_row_infop + row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); +PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth4_sse2,(png_row_infop + row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); + +-----------------cut---------------- diff --git a/third_party/libpng16/contrib/intel/filter_sse2_intrinsics.c b/third_party/libpng16/contrib/intel/filter_sse2_intrinsics.c new file mode 100644 index 0000000000..aea3f86af5 --- /dev/null +++ b/third_party/libpng16/contrib/intel/filter_sse2_intrinsics.c @@ -0,0 +1,379 @@ + +/* filter_sse2_intrinsics.c - SSE2 optimized filter functions + * + * Copyright (c) 2016 Google, Inc. + * Written by Mike Klein and Matt Sarett + * Derived from arm/filter_neon_intrinsics.c, which was + * Copyright (c) 2014,2016 Glenn Randers-Pehrson + * + * Last changed in libpng 1.6.22 [May 26, 2016] + * + * This code is released under the libpng license. + * For conditions of distribution and use, see the disclaimer + * and license in png.h + */ + +#include "../../pngpriv.h" + +#ifdef PNG_READ_SUPPORTED + +#if PNG_INTEL_SSE_IMPLEMENTATION > 0 + +#include <immintrin.h> + +/* Functions in this file look at most 3 pixels (a,b,c) to predict the 4th (d). + * They're positioned like this: + * prev: c b + * row: a d + * The Sub filter predicts d=a, Avg d=(a+b)/2, and Paeth predicts d to be + * whichever of a, b, or c is closest to p=a+b-c. + */ + +static __m128i load4(const void* p) { + return _mm_cvtsi32_si128(*(const int*)p); +} + +static void store4(void* p, __m128i v) { + *(int*)p = _mm_cvtsi128_si32(v); +} + +static __m128i load3(const void* p) { + /* We'll load 2 bytes, then 1 byte, + * then mask them together, and finally load into SSE. + */ + const png_uint_16* p01 = p; + const png_byte* p2 = (const png_byte*)(p01+1); + + png_uint_32 v012 = (png_uint_32)(*p01) + | (png_uint_32)(*p2) << 16; + return load4(&v012); +} + +static void store3(void* p, __m128i v) { + /* We'll pull from SSE as a 32-bit int, then write + * its bottom two bytes, then its third byte. + */ + png_uint_32 v012; + store4(&v012, v); + + png_uint_16* p01 = p; + png_byte* p2 = (png_byte*)(p01+1); + *p01 = v012; + *p2 = v012 >> 16; +} + +void png_read_filter_row_sub3_sse2(png_row_infop row_info, png_bytep row, + png_const_bytep prev) +{ + /* The Sub filter predicts each pixel as the previous pixel, a. + * There is no pixel to the left of the first pixel. It's encoded directly. + * That works with our main loop if we just say that left pixel was zero. + */ + png_debug(1, "in png_read_filter_row_sub3_sse2"); + __m128i a, d = _mm_setzero_si128(); + + int rb = row_info->rowbytes; + while (rb >= 4) { + a = d; d = load4(row); + d = _mm_add_epi8(d, a); + store3(row, d); + + row += 3; + rb -= 3; + } + if (rb > 0) { + a = d; d = load3(row); + d = _mm_add_epi8(d, a); + store3(row, d); + + row += 3; + rb -= 3; + } +} + +void png_read_filter_row_sub4_sse2(png_row_infop row_info, png_bytep row, + png_const_bytep prev) +{ + /* The Sub filter predicts each pixel as the previous pixel, a. + * There is no pixel to the left of the first pixel. It's encoded directly. + * That works with our main loop if we just say that left pixel was zero. + */ + png_debug(1, "in png_read_filter_row_sub4_sse2"); + __m128i a, d = _mm_setzero_si128(); + + int rb = row_info->rowbytes; + while (rb > 0) { + a = d; d = load4(row); + d = _mm_add_epi8(d, a); + store4(row, d); + + row += 4; + rb -= 4; + } +} + +void png_read_filter_row_avg3_sse2(png_row_infop row_info, png_bytep row, + png_const_bytep prev) +{ + /* The Avg filter predicts each pixel as the (truncated) average of a and b. + * There's no pixel to the left of the first pixel. Luckily, it's + * predicted to be half of the pixel above it. So again, this works + * perfectly with our loop if we make sure a starts at zero. + */ + png_debug(1, "in png_read_filter_row_avg3_sse2"); + const __m128i zero = _mm_setzero_si128(); + __m128i b; + __m128i a, d = zero; + + int rb = row_info->rowbytes; + while (rb >= 4) { + b = load4(prev); + a = d; d = load4(row ); + + /* PNG requires a truncating average, so we can't just use _mm_avg_epu8 */ + __m128i avg = _mm_avg_epu8(a,b); + /* ...but we can fix it up by subtracting off 1 if it rounded up. */ + avg = _mm_sub_epi8(avg, _mm_and_si128(_mm_xor_si128(a,b), + _mm_set1_epi8(1))); + d = _mm_add_epi8(d, avg); + store3(row, d); + + prev += 3; + row += 3; + rb -= 3; + } + if (rb > 0) { + b = load3(prev); + a = d; d = load3(row ); + + /* PNG requires a truncating average, so we can't just use _mm_avg_epu8 */ + __m128i avg = _mm_avg_epu8(a,b); + /* ...but we can fix it up by subtracting off 1 if it rounded up. */ + avg = _mm_sub_epi8(avg, _mm_and_si128(_mm_xor_si128(a,b), + _mm_set1_epi8(1))); + + d = _mm_add_epi8(d, avg); + store3(row, d); + + prev += 3; + row += 3; + rb -= 3; + } +} + +void png_read_filter_row_avg4_sse2(png_row_infop row_info, png_bytep row, + png_const_bytep prev) +{ + /* The Avg filter predicts each pixel as the (truncated) average of a and b. + * There's no pixel to the left of the first pixel. Luckily, it's + * predicted to be half of the pixel above it. So again, this works + * perfectly with our loop if we make sure a starts at zero. + */ + png_debug(1, "in png_read_filter_row_avg4_sse2"); + const __m128i zero = _mm_setzero_si128(); + __m128i b; + __m128i a, d = zero; + + int rb = row_info->rowbytes; + while (rb > 0) { + b = load4(prev); + a = d; d = load4(row ); + + /* PNG requires a truncating average, so we can't just use _mm_avg_epu8 */ + __m128i avg = _mm_avg_epu8(a,b); + /* ...but we can fix it up by subtracting off 1 if it rounded up. */ + avg = _mm_sub_epi8(avg, _mm_and_si128(_mm_xor_si128(a,b), + _mm_set1_epi8(1))); + + d = _mm_add_epi8(d, avg); + store4(row, d); + + prev += 4; + row += 4; + rb -= 4; + } +} + +/* Returns |x| for 16-bit lanes. */ +static __m128i abs_i16(__m128i x) { +#if PNG_INTEL_SSE_IMPLEMENTATION >= 2 + return _mm_abs_epi16(x); +#else + /* Read this all as, return x<0 ? -x : x. + * To negate two's complement, you flip all the bits then add 1. + */ + __m128i is_negative = _mm_cmplt_epi16(x, _mm_setzero_si128()); + + /* Flip negative lanes. */ + x = _mm_xor_si128(x, is_negative); + + /* +1 to negative lanes, else +0. */ + x = _mm_add_epi16(x, _mm_srli_epi16(is_negative, 15)); + return x; +#endif +} + +/* Bytewise c ? t : e. */ +static __m128i if_then_else(__m128i c, __m128i t, __m128i e) { +#if PNG_INTEL_SSE_IMPLEMENTATION >= 3 + return _mm_blendv_epi8(e,t,c); +#else + return _mm_or_si128(_mm_and_si128(c, t), _mm_andnot_si128(c, e)); +#endif +} + +void png_read_filter_row_paeth3_sse2(png_row_infop row_info, png_bytep row, + png_const_bytep prev) +{ + /* Paeth tries to predict pixel d using the pixel to the left of it, a, + * and two pixels from the previous row, b and c: + * prev: c b + * row: a d + * The Paeth function predicts d to be whichever of a, b, or c is nearest to + * p=a+b-c. + * + * The first pixel has no left context, and so uses an Up filter, p = b. + * This works naturally with our main loop's p = a+b-c if we force a and c + * to zero. + * Here we zero b and d, which become c and a respectively at the start of + * the loop. + */ + png_debug(1, "in png_read_filter_row_paeth3_sse2"); + const __m128i zero = _mm_setzero_si128(); + __m128i c, b = zero, + a, d = zero; + + int rb = row_info->rowbytes; + while (rb >= 4) { + /* It's easiest to do this math (particularly, deal with pc) with 16-bit + * intermediates. + */ + c = b; b = _mm_unpacklo_epi8(load4(prev), zero); + a = d; d = _mm_unpacklo_epi8(load4(row ), zero); + + /* (p-a) == (a+b-c - a) == (b-c) */ + __m128i pa = _mm_sub_epi16(b,c); + + /* (p-b) == (a+b-c - b) == (a-c) */ + __m128i pb = _mm_sub_epi16(a,c); + + /* (p-c) == (a+b-c - c) == (a+b-c-c) == (b-c)+(a-c) */ + __m128i pc = _mm_add_epi16(pa,pb); + + pa = abs_i16(pa); /* |p-a| */ + pb = abs_i16(pb); /* |p-b| */ + pc = abs_i16(pc); /* |p-c| */ + + __m128i smallest = _mm_min_epi16(pc, _mm_min_epi16(pa, pb)); + + /* Paeth breaks ties favoring a over b over c. */ + __m128i nearest = if_then_else(_mm_cmpeq_epi16(smallest, pa), a, + if_then_else(_mm_cmpeq_epi16(smallest, pb), b, + c)); + + /* Note `_epi8`: we need addition to wrap modulo 255. */ + d = _mm_add_epi8(d, nearest); + store3(row, _mm_packus_epi16(d,d)); + + prev += 3; + row += 3; + rb -= 3; + } + if (rb > 0) { + /* It's easiest to do this math (particularly, deal with pc) with 16-bit + * intermediates. + */ + c = b; b = _mm_unpacklo_epi8(load3(prev), zero); + a = d; d = _mm_unpacklo_epi8(load3(row ), zero); + + /* (p-a) == (a+b-c - a) == (b-c) */ + __m128i pa = _mm_sub_epi16(b,c); + + /* (p-b) == (a+b-c - b) == (a-c) */ + __m128i pb = _mm_sub_epi16(a,c); + + /* (p-c) == (a+b-c - c) == (a+b-c-c) == (b-c)+(a-c) */ + __m128i pc = _mm_add_epi16(pa,pb); + + pa = abs_i16(pa); /* |p-a| */ + pb = abs_i16(pb); /* |p-b| */ + pc = abs_i16(pc); /* |p-c| */ + + __m128i smallest = _mm_min_epi16(pc, _mm_min_epi16(pa, pb)); + + /* Paeth breaks ties favoring a over b over c. */ + __m128i nearest = if_then_else(_mm_cmpeq_epi16(smallest, pa), a, + if_then_else(_mm_cmpeq_epi16(smallest, pb), b, + c)); + + /* Note `_epi8`: we need addition to wrap modulo 255. */ + d = _mm_add_epi8(d, nearest); + store3(row, _mm_packus_epi16(d,d)); + + prev += 3; + row += 3; + rb -= 3; + } +} + +void png_read_filter_row_paeth4_sse2(png_row_infop row_info, png_bytep row, + png_const_bytep prev) +{ + /* Paeth tries to predict pixel d using the pixel to the left of it, a, + * and two pixels from the previous row, b and c: + * prev: c b + * row: a d + * The Paeth function predicts d to be whichever of a, b, or c is nearest to + * p=a+b-c. + * + * The first pixel has no left context, and so uses an Up filter, p = b. + * This works naturally with our main loop's p = a+b-c if we force a and c + * to zero. + * Here we zero b and d, which become c and a respectively at the start of + * the loop. + */ + png_debug(1, "in png_read_filter_row_paeth4_sse2"); + const __m128i zero = _mm_setzero_si128(); + __m128i c, b = zero, + a, d = zero; + + int rb = row_info->rowbytes; + while (rb > 0) { + /* It's easiest to do this math (particularly, deal with pc) with 16-bit + * intermediates. + */ + c = b; b = _mm_unpacklo_epi8(load4(prev), zero); + a = d; d = _mm_unpacklo_epi8(load4(row ), zero); + + /* (p-a) == (a+b-c - a) == (b-c) */ + __m128i pa = _mm_sub_epi16(b,c); + + /* (p-b) == (a+b-c - b) == (a-c) */ + __m128i pb = _mm_sub_epi16(a,c); + + /* (p-c) == (a+b-c - c) == (a+b-c-c) == (b-c)+(a-c) */ + __m128i pc = _mm_add_epi16(pa,pb); + + pa = abs_i16(pa); /* |p-a| */ + pb = abs_i16(pb); /* |p-b| */ + pc = abs_i16(pc); /* |p-c| */ + + __m128i smallest = _mm_min_epi16(pc, _mm_min_epi16(pa, pb)); + + /* Paeth breaks ties favoring a over b over c. */ + __m128i nearest = if_then_else(_mm_cmpeq_epi16(smallest, pa), a, + if_then_else(_mm_cmpeq_epi16(smallest, pb), b, + c)); + + /* Note `_epi8`: we need addition to wrap modulo 255. */ + d = _mm_add_epi8(d, nearest); + store4(row, _mm_packus_epi16(d,d)); + + prev += 4; + row += 4; + rb -= 4; + } +} + +#endif /* PNG_INTEL_SSE_IMPLEMENTATION > 0 */ +#endif /* READ */ diff --git a/third_party/libpng16/contrib/intel/intel_init.c b/third_party/libpng16/contrib/intel/intel_init.c new file mode 100644 index 0000000000..328e90e9af --- /dev/null +++ b/third_party/libpng16/contrib/intel/intel_init.c @@ -0,0 +1,54 @@ + +/* intel_init.c - SSE2 optimized filter functions + * + * Copyright (c) 2016 Google, Inc. + * Written by Mike Klein and Matt Sarett + * Derived from arm/arm_init.c, which was + * Copyright (c) 2014,2016 Glenn Randers-Pehrson + * + * Last changed in libpng 1.6.22 [May 26, 2016] + * + * This code is released under the libpng license. + * For conditions of distribution and use, see the disclaimer + * and license in png.h + */ + +#include "../../pngpriv.h" + +#ifdef PNG_READ_SUPPORTED +#if PNG_INTEL_SSE_IMPLEMENTATION > 0 + +void +png_init_filter_functions_sse2(png_structp pp, unsigned int bpp) +{ + /* The techniques used to implement each of these filters in SSE operate on + * one pixel at a time. + * So they generally speed up 3bpp images about 3x, 4bpp images about 4x. + * They can scale up to 6 and 8 bpp images and down to 2 bpp images, + * but they'd not likely have any benefit for 1bpp images. + * Most of these can be implemented using only MMX and 64-bit registers, + * but they end up a bit slower than using the equally-ubiquitous SSE2. + */ + png_debug(1, "in png_init_filter_functions_sse2"); + if (bpp == 3) + { + pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub3_sse2; + pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg3_sse2; + pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = + png_read_filter_row_paeth3_sse2; + } + else if (bpp == 4) + { + pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub4_sse2; + pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg4_sse2; + pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = + png_read_filter_row_paeth4_sse2; + } + + /* No need optimize PNG_FILTER_VALUE_UP. The compiler should + * autovectorize. + */ +} + +#endif /* PNG_INTEL_SSE_IMPLEMENTATION > 0 */ +#endif /* PNG_READ_SUPPORTED */ diff --git a/third_party/libpng16/contrib/intel/intel_sse.patch b/third_party/libpng16/contrib/intel/intel_sse.patch new file mode 100644 index 0000000000..d9d02bb8fc --- /dev/null +++ b/third_party/libpng16/contrib/intel/intel_sse.patch @@ -0,0 +1,164 @@ +diff --git libpng-1.6.22-orig/configure.ac libpng-1.6.22/configure.ac +--- libpng-1.6.22-orig/configure.ac 2016-05-25 18:59:10.000000000 -0400 ++++ libpng-1.6.22/configure.ac 2016-05-25 19:48:10.631751170 -0400 +@@ -341,16 +341,50 @@ AC_ARG_ENABLE([arm-neon], + + AM_CONDITIONAL([PNG_ARM_NEON], + [test "$enable_arm_neon" != 'no' && + case "$host_cpu" in + arm*|aarch64*) :;; + *) test "$enable_arm_neon" != '';; + esac]) + ++# INTEL ++# ===== ++# ++# INTEL SSE (SIMD) support. ++ ++AC_ARG_ENABLE([intel-sse], ++ AS_HELP_STRING([[[--enable-intel-sse]]], ++ [Enable Intel SSE optimizations: =no/off, yes/on:] ++ [no/off: disable the optimizations;] ++ [yes/on: enable the optimizations.] ++ [If not specified: determined by the compiler.]), ++ [case "$enableval" in ++ no|off) ++ # disable the default enabling: ++ AC_DEFINE([PNG_INTEL_SSE_OPT], [0], ++ [Disable Intel SSE optimizations]) ++ # Prevent inclusion of the assembler files below: ++ enable_intel_sse=no;; ++ yes|on) ++ AC_DEFINE([PNG_INTEL_SSE_OPT], [1], ++ [Enable Intel SSE optimizations]);; ++ *) ++ AC_MSG_ERROR([--enable-intel-sse=${enable_intel_sse}: invalid value]) ++ esac]) ++ ++# Add Intel specific files to all builds where the host_cpu is Intel ('x86*') ++# or where Intel optimizations were explicitly requested (this allows a ++# fallback if a future host CPU does not match 'x86*') ++AM_CONDITIONAL([PNG_INTEL_SSE], ++ [test "$enable_intel_sse" != 'no' && ++ case "$host_cpu" in ++ i?86|x86_64) :;; ++ *) test "$enable_intel_sse" != '';; ++ esac]) + AC_MSG_NOTICE([[Extra options for compiler: $PNG_COPTS]]) + + # Config files, substituting as above + AC_CONFIG_FILES([Makefile libpng.pc:libpng.pc.in]) + AC_CONFIG_FILES([libpng-config:libpng-config.in], + [chmod +x libpng-config]) + + AC_OUTPUT +diff --git libpng-1.6.22-orig/Makefile.am libpng-1.6.22/Makefile.am +--- libpng-1.6.22-orig/Makefile.am 2016-05-17 18:15:12.000000000 -0400 ++++ libpng-1.6.22/Makefile.am 2016-05-25 19:48:10.631751170 -0400 +@@ -89,16 +89,20 @@ libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SO + pngset.c pngtrans.c pngwio.c pngwrite.c pngwtran.c pngwutil.c\ + png.h pngconf.h pngdebug.h pnginfo.h pngpriv.h pngstruct.h pngusr.dfa + + if PNG_ARM_NEON + libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES += arm/arm_init.c\ + arm/filter_neon.S arm/filter_neon_intrinsics.c + endif + ++if PNG_INTEL_SSE ++libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES += contrib/intel/intel_init.c\ ++ contrib/intel/filter_sse2_intrinsics.c ++endif + nodist_libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES = pnglibconf.h + + libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_LDFLAGS = -no-undefined -export-dynamic \ + -version-number @PNGLIB_MAJOR@@PNGLIB_MINOR@:@PNGLIB_RELEASE@:0 + + if HAVE_LD_VERSION_SCRIPT + # Versioned symbols and restricted exports + if HAVE_SOLARIS_LD +diff --git libpng-1.6.22-orig/pngpriv.h libpng-1.6.22/pngpriv.h +--- libpng-1.6.22-orig/pngpriv.h 2016-05-25 18:59:10.000000000 -0400 ++++ libpng-1.6.22/pngpriv.h 2016-05-25 19:48:10.635751171 -0400 +@@ -177,16 +177,52 @@ + # endif /* !PNG_ARM_NEON_IMPLEMENTATION */ + + # ifndef PNG_ARM_NEON_IMPLEMENTATION + /* Use the intrinsics code by default. */ + # define PNG_ARM_NEON_IMPLEMENTATION 1 + # endif + #endif /* PNG_ARM_NEON_OPT > 0 */ + ++#ifndef PNG_INTEL_SSE_OPT ++# ifdef PNG_INTEL_SSE ++ /* Only check for SSE if the build configuration has been modified to ++ * enable SSE optimizations. This means that these optimizations will ++ * be off by default. See contrib/intel for more details. ++ */ ++# if defined(__SSE4_1__) || defined(__AVX__) || defined(__SSSE3__) || \ ++ defined(__SSE2__) || defined(_M_X64) || defined(_M_AMD64) || \ ++ (defined(_M_IX86_FP) && _M_IX86_FP >= 2) ++# define PNG_INTEL_SSE_OPT 1 ++# endif ++# endif ++#endif ++ ++#if PNG_INTEL_SSE_OPT > 0 ++# ifndef PNG_INTEL_SSE_IMPLEMENTATION ++# if defined(__SSE4_1__) || defined(__AVX__) ++ /* We are not actually using AVX, but checking for AVX is the best ++ way we can detect SSE4.1 and SSSE3 on MSVC. ++ */ ++# define PNG_INTEL_SSE_IMPLEMENTATION 3 ++# elif defined(__SSSE3__) ++# define PNG_INTEL_SSE_IMPLEMENTATION 2 ++# elif defined(__SSE2__) || defined(_M_X64) || defined(_M_AMD64) || \ ++ (defined(_M_IX86_FP) && _M_IX86_FP >= 2) ++# define PNG_INTEL_SSE_IMPLEMENTATION 1 ++# else ++# define PNG_INTEL_SSE_IMPLEMENTATION 0 ++# endif ++# endif ++ ++# if PNG_INTEL_SSE_IMPLEMENTATION > 0 ++# define PNG_FILTER_OPTIMIZATIONS png_init_filter_functions_sse2 ++# endif ++#endif ++ + /* Is this a build of a DLL where compilation of the object modules requires + * different preprocessor settings to those required for a simple library? If + * so PNG_BUILD_DLL must be set. + * + * If libpng is used inside a DLL but that DLL does not export the libpng APIs + * PNG_BUILD_DLL must not be set. To avoid the code below kicking in build a + * static library of libpng then link the DLL against that. + */ +@@ -1184,16 +1220,29 @@ PNG_INTERNAL_FUNCTION(void,png_read_filt + row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); + PNG_INTERNAL_FUNCTION(void,png_read_filter_row_avg4_neon,(png_row_infop + row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); + PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth3_neon,(png_row_infop + row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); + PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth4_neon,(png_row_infop + row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); + ++PNG_INTERNAL_FUNCTION(void,png_read_filter_row_sub3_sse2,(png_row_infop ++ row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); ++PNG_INTERNAL_FUNCTION(void,png_read_filter_row_sub4_sse2,(png_row_infop ++ row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); ++PNG_INTERNAL_FUNCTION(void,png_read_filter_row_avg3_sse2,(png_row_infop ++ row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); ++PNG_INTERNAL_FUNCTION(void,png_read_filter_row_avg4_sse2,(png_row_infop ++ row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); ++PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth3_sse2,(png_row_infop ++ row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); ++PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth4_sse2,(png_row_infop ++ row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); ++ + /* Choose the best filter to use and filter the row data */ + PNG_INTERNAL_FUNCTION(void,png_write_find_filter,(png_structrp png_ptr, + png_row_infop row_info),PNG_EMPTY); + + #ifdef PNG_SEQUENTIAL_READ_SUPPORTED + PNG_INTERNAL_FUNCTION(void,png_read_IDAT_data,(png_structrp png_ptr, + png_bytep output, png_alloc_size_t avail_out),PNG_EMPTY); + /* Read 'avail_out' bytes of data from the IDAT stream. If the output buffer diff --git a/third_party/libpng16/png.c b/third_party/libpng16/png.c index 1cf5271046..35e14f63d8 100644 --- a/third_party/libpng16/png.c +++ b/third_party/libpng16/png.c @@ -2,7 +2,7 @@ /* png.c - location for general purpose libpng functions * * Last changed in libpng 1.6.19 [November 12, 2015] - * Copyright (c) 1998-2015 Glenn Randers-Pehrson + * Copyright (c) 1998-2002,2004,2006-2015 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * @@ -14,7 +14,7 @@ #include "pngpriv.h" /* Generate a compiler error if there is an old png.h in the search path. */ -typedef png_libpng_version_1_6_20 Your_png_h_is_not_version_1_6_20; +typedef png_libpng_version_1_6_22 Your_png_h_is_not_version_1_6_22; /* Tells libpng that we have already handled the first "num_bytes" bytes * of the PNG file signature. If the PNG data is embedded into another @@ -775,14 +775,15 @@ png_get_copyright(png_const_structrp png_ptr) #else # ifdef __STDC__ return PNG_STRING_NEWLINE \ - "libpng version 1.6.20 - December 3, 2015" PNG_STRING_NEWLINE \ - "Copyright (c) 1998-2015 Glenn Randers-Pehrson" PNG_STRING_NEWLINE \ + "libpng version 1.6.22 - May 26, 2016" PNG_STRING_NEWLINE \ + "Copyright (c) 1998-2002,2004,2006-2016 Glenn Randers-Pehrson" \ + PNG_STRING_NEWLINE \ "Copyright (c) 1996-1997 Andreas Dilger" PNG_STRING_NEWLINE \ "Copyright (c) 1995-1996 Guy Eric Schalnat, Group 42, Inc." \ PNG_STRING_NEWLINE; # else - return "libpng version 1.6.20 - December 3, 2015\ - Copyright (c) 1998-2015 Glenn Randers-Pehrson\ + return "libpng version 1.6.22 - May 26, 2016\ + Copyright (c) 1998-2002,2004,2006-2016 Glenn Randers-Pehrson\ Copyright (c) 1996-1997 Andreas Dilger\ Copyright (c) 1995-1996 Guy Eric Schalnat, Group 42, Inc."; # endif diff --git a/third_party/libpng16/png.h b/third_party/libpng16/png.h index 4d03dfc136..08c039df0b 100644 --- a/third_party/libpng16/png.h +++ b/third_party/libpng16/png.h @@ -1,9 +1,9 @@ /* png.h - header file for PNG reference library * - * libpng version 1.6.20, December 3, 2015 + * libpng version 1.6.22, May 26, 2016 * - * Copyright (c) 1998-2015 Glenn Randers-Pehrson + * Copyright (c) 1998-2002,2004,2006-2016 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * @@ -12,7 +12,7 @@ * Authors and maintainers: * libpng versions 0.71, May 1995, through 0.88, January 1996: Guy Schalnat * libpng versions 0.89, June 1996, through 0.96, May 1997: Andreas Dilger - * libpng versions 0.97, January 1998, through 1.6.20, December 3, 2015: + * libpng versions 0.97, January 1998, through 1.6.22, May 26, 2016: * Glenn Randers-Pehrson. * See also "Contributing Authors", below. */ @@ -25,8 +25,12 @@ * * This code is released under the libpng license. * - * libpng versions 1.0.7, July 1, 2000, through 1.6.20, December 3, 2015, are - * Copyright (c) 2000-2002, 2004, 2006-2015 Glenn Randers-Pehrson, are + * Some files in the "contrib" directory and some configure-generated + * files that are distributed with libpng have other copyright owners and + * are released under other open source licenses. + * + * libpng versions 1.0.7, July 1, 2000 through 1.6.22, May 26, 2016 are + * Copyright (c) 2000-2002, 2004, 2006-2016 Glenn Randers-Pehrson, are * derived from libpng-1.0.6, and are distributed according to the same * disclaimer and license as libpng-1.0.6 with the following individuals * added to the list of Contributing Authors: @@ -47,6 +51,10 @@ * risk of satisfactory quality, performance, accuracy, and effort is with * the user. * + * Some files in the "contrib" directory have other copyright owners and + * are released under other open source licenses. + * + * * libpng versions 0.97, January 1998, through 1.0.6, March 20, 2000, are * Copyright (c) 1998-2000 Glenn Randers-Pehrson, are derived from * libpng-0.96, and are distributed according to the same disclaimer and @@ -57,6 +65,9 @@ * Glenn Randers-Pehrson * Willem van Schaik * + * Some files in the "scripts" directory have different copyright owners + * but are also released under this license. + * * libpng versions 0.89, June 1996, through 0.96, May 1997, are * Copyright (c) 1996-1997 Andreas Dilger, are derived from libpng-0.88, * and are distributed according to the same disclaimer and license as @@ -70,6 +81,9 @@ * Greg Roelofs * Tom Tanner * + * Some files in the "scripts" directory have other copyright owners + * but are released under this license. + * * libpng versions 0.5, May 1995, through 0.88, January 1996, are * Copyright (c) 1995-1996 Guy Eric Schalnat, Group 42, Inc. * @@ -109,6 +123,29 @@ * appreciated. * * END OF COPYRIGHT NOTICE, DISCLAIMER, and LICENSE. + * + * TRADEMARK: + * + * The name "libpng" has not been registered by the Copyright owner + * as a trademark in any jurisdiction. However, because libpng has + * been distributed and maintained world-wide, continually since 1995, + * the Copyright owner claims "common-law trademark protection" in any + * jurisdiction where common-law trademark is recognized. + * + * OSI CERTIFICATION: + * + * Libpng is OSI Certified Open Source Software. OSI Certified Open Source is + * a certification mark of the Open Source Initiative. OSI has not addressed + * the additional disclaimers inserted at version 1.0.7. + * + * EXPORT CONTROL: + * + * The Copyright owner believes that the Export Control Classification + * Number (ECCN) for libpng is EAR99, which means not subject to export + * controls or International Traffic in Arms Regulations (ITAR) because + * it is open source, publicly available software, that does not contain + * any encryption software. See the EAR, paragraphs 734.3(b)(3) and + * 734.7(b). */ /* @@ -122,12 +159,6 @@ */ /* - * Libpng is OSI Certified Open Source Software. OSI Certified Open Source is - * a certification mark of the Open Source Initiative. OSI has not addressed - * the additional disclaimers inserted at version 1.0.7. - */ - -/* * The contributing authors would like to thank all those who helped * with testing, bug fixes, and patience. This wouldn't have been * possible without all of you. @@ -182,11 +213,11 @@ * ... * 1.0.19 10 10019 10.so.0.19[.0] * ... - * 1.2.53 13 10253 12.so.0.53[.0] + * 1.2.56 13 10256 12.so.0.56[.0] * ... - * 1.5.23 15 10523 15.so.15.23[.0] + * 1.5.27 15 10527 15.so.15.27[.0] * ... - * 1.6.20 16 10620 16.so.16.20[.0] + * 1.6.22 16 10622 16.so.16.22[.0] * * Henceforth the source version will match the shared-library major * and minor numbers; the shared-library major version number will be @@ -214,13 +245,13 @@ * Y2K compliance in libpng: * ========================= * - * December 3, 2015 + * May 26, 2016 * * Since the PNG Development group is an ad-hoc body, we can't make * an official declaration. * * This is your unofficial assurance that libpng from version 0.71 and - * upward through 1.6.20 are Y2K compliant. It is my belief that + * upward through 1.6.22 are Y2K compliant. It is my belief that * earlier versions were also Y2K compliant. * * Libpng only has two year fields. One is a 2-byte unsigned integer @@ -282,9 +313,9 @@ */ /* Version information for png.h - this should match the version in png.c */ -#define PNG_LIBPNG_VER_STRING "1.6.20" +#define PNG_LIBPNG_VER_STRING "1.6.22" #define PNG_HEADER_VERSION_STRING \ - " libpng version 1.6.20 - December 3, 2015\n" + " libpng version 1.6.22 - May 26, 2016\n" #define PNG_LIBPNG_VER_SONUM 16 #define PNG_LIBPNG_VER_DLLNUM 16 @@ -292,7 +323,7 @@ /* These should match the first 3 components of PNG_LIBPNG_VER_STRING: */ #define PNG_LIBPNG_VER_MAJOR 1 #define PNG_LIBPNG_VER_MINOR 6 -#define PNG_LIBPNG_VER_RELEASE 20 +#define PNG_LIBPNG_VER_RELEASE 22 /* This should match the numeric part of the final component of * PNG_LIBPNG_VER_STRING, omitting any leading zero: @@ -323,7 +354,7 @@ * version 1.0.0 was mis-numbered 100 instead of 10000). From * version 1.0.1 it's xxyyzz, where x=major, y=minor, z=release */ -#define PNG_LIBPNG_VER 10620 /* 1.6.20 */ +#define PNG_LIBPNG_VER 10622 /* 1.6.22 */ /* Library configuration: these options cannot be changed after * the library has been built. @@ -433,7 +464,7 @@ extern "C" { /* This triggers a compiler error in png.c, if png.c and png.h * do not agree upon the version number. */ -typedef char* png_libpng_version_1_6_20; +typedef char* png_libpng_version_1_6_22; /* Basic control structions. Read libpng-manual.txt or libpng.3 for more info. * @@ -734,24 +765,22 @@ typedef png_unknown_chunk * * png_unknown_chunkpp; * data in the info_struct to be written into the output file. The values * of the PNG_INFO_<chunk> defines should NOT be changed. */ -#define PNG_INFO_gAMA 0x0001 -#define PNG_INFO_sBIT 0x0002 -#define PNG_INFO_cHRM 0x0004 -#define PNG_INFO_PLTE 0x0008 -#define PNG_INFO_tRNS 0x0010 -#define PNG_INFO_bKGD 0x0020 -#define PNG_INFO_hIST 0x0040 -#define PNG_INFO_pHYs 0x0080 -#define PNG_INFO_oFFs 0x0100 -#define PNG_INFO_tIME 0x0200 -#define PNG_INFO_pCAL 0x0400 -#define PNG_INFO_sRGB 0x0800 /* GR-P, 0.96a */ -#define PNG_INFO_iCCP 0x1000 /* ESR, 1.0.6 */ -#define PNG_INFO_sPLT 0x2000 /* ESR, 1.0.6 */ -#define PNG_INFO_sCAL 0x4000 /* ESR, 1.0.6 */ -#if INT_MAX >= 0x8000 /* else this might break */ -#define PNG_INFO_IDAT 0x8000 /* ESR, 1.0.6 */ -#endif +#define PNG_INFO_gAMA 0x0001U +#define PNG_INFO_sBIT 0x0002U +#define PNG_INFO_cHRM 0x0004U +#define PNG_INFO_PLTE 0x0008U +#define PNG_INFO_tRNS 0x0010U +#define PNG_INFO_bKGD 0x0020U +#define PNG_INFO_hIST 0x0040U +#define PNG_INFO_pHYs 0x0080U +#define PNG_INFO_oFFs 0x0100U +#define PNG_INFO_tIME 0x0200U +#define PNG_INFO_pCAL 0x0400U +#define PNG_INFO_sRGB 0x0800U /* GR-P, 0.96a */ +#define PNG_INFO_iCCP 0x1000U /* ESR, 1.0.6 */ +#define PNG_INFO_sPLT 0x2000U /* ESR, 1.0.6 */ +#define PNG_INFO_sCAL 0x4000U /* ESR, 1.0.6 */ +#define PNG_INFO_IDAT 0x8000U /* ESR, 1.0.6 */ /* This is used for the transformation routines, as some of them * change these values for the row. It also should enable using @@ -1322,7 +1351,7 @@ PNG_EXPORT(229, void, png_set_scale_16, (png_structrp png_ptr)); #endif #ifdef PNG_READ_STRIP_16_TO_8_SUPPORTED -#define PNG_READ_16_TO_8 SUPPORTED /* Name prior to 1.5.4 */ +#define PNG_READ_16_TO_8_SUPPORTED /* Name prior to 1.5.4 */ /* Strip the second byte of information from a 16-bit depth file. */ PNG_EXPORT(48, void, png_set_strip_16, (png_structrp png_ptr)); #endif @@ -1473,8 +1502,8 @@ PNG_EXPORT(67, void, png_set_filter, (png_structrp png_ptr, int method, #define PNG_FILTER_UP 0x20 #define PNG_FILTER_AVG 0x40 #define PNG_FILTER_PAETH 0x80 -#define PNG_ALL_FILTERS (PNG_FILTER_NONE | PNG_FILTER_SUB | PNG_FILTER_UP | \ - PNG_FILTER_AVG | PNG_FILTER_PAETH) +#define PNG_FAST_FILTERS (PNG_FILTER_NONE | PNG_FILTER_SUB | PNG_FILTER_UP) +#define PNG_ALL_FILTERS (PNG_FAST_FILTERS | PNG_FILTER_AVG | PNG_FILTER_PAETH) /* Filter values (not flags) - used in pngwrite.c, pngwutil.c for now. * These defines should NOT be changed. @@ -1751,21 +1780,21 @@ PNG_EXPORT(99, void, png_data_freer, (png_const_structrp png_ptr, #define PNG_SET_WILL_FREE_DATA 1 #define PNG_USER_WILL_FREE_DATA 2 /* Flags for png_ptr->free_me and info_ptr->free_me */ -#define PNG_FREE_HIST 0x0008 -#define PNG_FREE_ICCP 0x0010 -#define PNG_FREE_SPLT 0x0020 -#define PNG_FREE_ROWS 0x0040 -#define PNG_FREE_PCAL 0x0080 -#define PNG_FREE_SCAL 0x0100 +#define PNG_FREE_HIST 0x0008U +#define PNG_FREE_ICCP 0x0010U +#define PNG_FREE_SPLT 0x0020U +#define PNG_FREE_ROWS 0x0040U +#define PNG_FREE_PCAL 0x0080U +#define PNG_FREE_SCAL 0x0100U #ifdef PNG_STORE_UNKNOWN_CHUNKS_SUPPORTED -# define PNG_FREE_UNKN 0x0200 +# define PNG_FREE_UNKN 0x0200U #endif -/* PNG_FREE_LIST 0x0400 removed in 1.6.0 because it is ignored */ -#define PNG_FREE_PLTE 0x1000 -#define PNG_FREE_TRNS 0x2000 -#define PNG_FREE_TEXT 0x4000 -#define PNG_FREE_ALL 0x7fff -#define PNG_FREE_MUL 0x4220 /* PNG_FREE_SPLT|PNG_FREE_TEXT|PNG_FREE_UNKN */ +/* PNG_FREE_LIST 0x0400U removed in 1.6.0 because it is ignored */ +#define PNG_FREE_PLTE 0x1000U +#define PNG_FREE_TRNS 0x2000U +#define PNG_FREE_TEXT 0x4000U +#define PNG_FREE_ALL 0x7fffU +#define PNG_FREE_MUL 0x4220U /* PNG_FREE_SPLT|PNG_FREE_TEXT|PNG_FREE_UNKN */ #ifdef PNG_USER_MEM_SUPPORTED PNG_EXPORTA(100, png_voidp, png_malloc_default, (png_const_structrp png_ptr, @@ -2889,12 +2918,19 @@ typedef struct * is the minimum 'row stride', the minimum count of components between each * row. For a color-mapped image this is the minimum number of bytes in a * row. + * + * WARNING: this macro overflows for some images with more than one component + * and very large image widths. libpng will refuse to process an image where + * this macro would overflow. */ #define PNG_IMAGE_BUFFER_SIZE(image, row_stride)\ (PNG_IMAGE_PIXEL_COMPONENT_SIZE((image).format)*(image).height*(row_stride)) /* Return the size, in bytes, of an image buffer given a png_image and a row * stride - the number of components to leave space for in each row. + * + * WARNING: this macro overflows a 32-bit integer for some large PNG images, + * libpng will refuse to process an image where such an overflow would occur. */ #define PNG_IMAGE_SIZE(image)\ @@ -3015,7 +3051,6 @@ PNG_EXPORT(238, void, png_image_free, (png_imagep image)); #endif /* SIMPLIFIED_READ */ #ifdef PNG_SIMPLIFIED_WRITE_SUPPORTED -#ifdef PNG_STDIO_SUPPORTED /* WRITE APIS * ---------- * For write you must initialize a png_image structure to describe the image to @@ -3032,6 +3067,7 @@ PNG_EXPORT(238, void, png_image_free, (png_imagep image)); * values do not correspond to the colors in sRGB. * colormap_entries: set to the number of entries in the color-map (0 to 256) */ +#ifdef PNG_SIMPLIFIED_WRITE_STDIO_SUPPORTED PNG_EXPORT(239, int, png_image_write_to_file, (png_imagep image, const char *file, int convert_to_8bit, const void *buffer, png_int_32 row_stride, const void *colormap)); @@ -3041,8 +3077,9 @@ PNG_EXPORT(240, int, png_image_write_to_stdio, (png_imagep image, FILE *file, int convert_to_8_bit, const void *buffer, png_int_32 row_stride, const void *colormap)); /* Write the image to the given (FILE*). */ +#endif /* SIMPLIFIED_WRITE_STDIO */ -/* With both write APIs if image is in one of the linear formats with 16-bit +/* With all write APIs if image is in one of the linear formats with 16-bit * data then setting convert_to_8_bit will cause the output to be an 8-bit PNG * gamma encoded according to the sRGB specification, otherwise a 16-bit linear * encoded PNG file is written. @@ -3054,13 +3091,103 @@ PNG_EXPORT(240, int, png_image_write_to_stdio, (png_imagep image, FILE *file, * * With all APIs row_stride is handled as in the read APIs - it is the spacing * from one row to the next in component sized units (1 or 2 bytes) and if - * negative indicates a bottom-up row layout in the buffer. If row_stride is zero, - * libpng will calculate it for you from the image width and number of channels. + * negative indicates a bottom-up row layout in the buffer. If row_stride is + * zero, libpng will calculate it for you from the image width and number of + * channels. * - * Note that the write API does not support interlacing, sub-8-bit pixels, indexed - * PNG (color_type 3) or most ancillary chunks. + * Note that the write API does not support interlacing, sub-8-bit pixels or + * most ancillary chunks. If you need to write text chunks (e.g. for copyright + * notices) you need to use one of the other APIs. */ -#endif /* STDIO */ + +PNG_EXPORT(245, int, png_image_write_to_memory, (png_imagep image, void *memory, + png_alloc_size_t * PNG_RESTRICT memory_bytes, int convert_to_8_bit, + const void *buffer, png_int_32 row_stride, const void *colormap)); + /* Write the image to the given memory buffer. The function both writes the + * whole PNG data stream to *memory and updates *memory_bytes with the count + * of bytes written. + * + * 'memory' may be NULL. In this case *memory_bytes is not read however on + * success the number of bytes which would have been written will still be + * stored in *memory_bytes. On failure *memory_bytes will contain 0. + * + * If 'memory' is not NULL it must point to memory[*memory_bytes] of + * writeable memory. + * + * If the function returns success memory[*memory_bytes] (if 'memory' is not + * NULL) contains the written PNG data. *memory_bytes will always be less + * than or equal to the original value. + * + * If the function returns false and *memory_bytes was not changed an error + * occured during write. If *memory_bytes was changed, or is not 0 if + * 'memory' was NULL, the write would have succeeded but for the memory + * buffer being too small. *memory_bytes contains the required number of + * bytes and will be bigger that the original value. + */ + +#define png_image_write_get_memory_size(image, size, convert_to_8_bit, buffer,\ + row_stride, colormap)\ + png_image_write_to_memory(&(image), 0, &(size), convert_to_8_bit, buffer,\ + row_stride, colormap) + /* Return the amount of memory in 'size' required to compress this image. + * The png_image structure 'image' must be filled in as in the above + * function and must not be changed before the actual write call, the buffer + * and all other parameters must also be identical to that in the final + * write call. The 'size' variable need not be initialized. + * + * NOTE: the macro returns true/false, if false is returned 'size' will be + * set to zero and the write failed and probably will fail if tried again. + */ + +/* You can pre-allocate the buffer by making sure it is of sufficient size + * regardless of the amount of compression achieved. The buffer size will + * always be bigger than the original image and it will never be filled. The + * following macros are provided to assist in allocating the buffer. + */ +#define PNG_IMAGE_DATA_SIZE(image) (PNG_IMAGE_SIZE(image)+(image).height) + /* The number of uncompressed bytes in the PNG byte encoding of the image; + * uncompressing the PNG IDAT data will give this number of bytes. + * + * NOTE: while PNG_IMAGE_SIZE cannot overflow for an image in memory this + * macro can because of the extra bytes used in the PNG byte encoding. You + * need to avoid this macro if your image size approaches 2^30 in width or + * height. The same goes for the remainder of these macros; they all produce + * bigger numbers than the actual in-memory image size. + */ +#ifndef PNG_ZLIB_MAX_SIZE +# define PNG_ZLIB_MAX_SIZE(b) ((b)+(((b)+7U)>>3)+(((b)+63U)>>6)+11U) + /* An upper bound on the number of compressed bytes given 'b' uncompressed + * bytes. This is based on deflateBounds() in zlib; different + * implementations of zlib compression may conceivably produce more data so + * if your zlib implementation is not zlib itself redefine this macro + * appropriately. + */ +#endif + +#define PNG_IMAGE_COMPRESSED_SIZE_MAX(image)\ + PNG_ZLIB_MAX_SIZE((png_alloc_size_t)PNG_IMAGE_DATA_SIZE(image)) + /* An upper bound on the size of the data in the PNG IDAT chunks. */ + +#define PNG_IMAGE_PNG_SIZE_MAX_(image, image_size)\ + ((8U/*sig*/+25U/*IHDR*/+16U/*gAMA*/+44U/*cHRM*/+12U/*IEND*/+\ + (((image).format&PNG_FORMAT_FLAG_COLORMAP)?/*colormap: PLTE, tRNS*/\ + 12U+3U*(image).colormap_entries/*PLTE data*/+\ + (((image).format&PNG_FORMAT_FLAG_ALPHA)?\ + 12U/*tRNS*/+(image).colormap_entries:0U):0U)+\ + 12U)+(12U*((image_size)/PNG_ZBUF_SIZE))/*IDAT*/+(image_size)) + /* A helper for the following macro; if your compiler cannot handle the + * following macro use this one with the result of + * PNG_IMAGE_COMPRESSED_SIZE_MAX(image) as the second argument (most + * compilers should handle this just fine.) + */ + +#define PNG_IMAGE_PNG_SIZE_MAX(image)\ + PNG_IMAGE_PNG_SIZE_MAX_(image, PNG_IMAGE_COMPRESSED_SIZE_MAX(image)) + /* An upper bound on the total length of the PNG data stream for 'image'. + * The result is of type png_alloc_size_t, on 32-bit systems this may + * overflow even though PNG_IMAGE_DATA_SIZE does not overflow; the write will + * run out of buffer space but return a corrected size which should work. + */ #endif /* SIMPLIFIED_WRITE */ /******************************************************************************* * END OF SIMPLIFIED API @@ -3118,7 +3245,7 @@ PNG_EXPORT(244, int, png_set_option, (png_structrp png_ptr, int option, * one to use is one more than this.) */ #ifdef PNG_EXPORT_LAST_ORDINAL - PNG_EXPORT_LAST_ORDINAL(244); + PNG_EXPORT_LAST_ORDINAL(245); #endif #ifdef __cplusplus diff --git a/third_party/libpng16/pngconf.h b/third_party/libpng16/pngconf.h index 92f250000c..0cabe4b385 100644 --- a/third_party/libpng16/pngconf.h +++ b/third_party/libpng16/pngconf.h @@ -1,9 +1,9 @@ /* pngconf.h - machine configurable file for libpng * - * libpng version 1.6.20, December 3, 2015 + * libpng version 1.6.22, May 26, 2016 * - * Copyright (c) 1998-2015 Glenn Randers-Pehrson + * Copyright (c) 1998-2002,2004,2006-2015 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * diff --git a/third_party/libpng16/pngdebug.h b/third_party/libpng16/pngdebug.h index 6a01b106ed..15a7ed0c95 100644 --- a/third_party/libpng16/pngdebug.h +++ b/third_party/libpng16/pngdebug.h @@ -2,7 +2,7 @@ /* pngdebug.h - Debugging macros for libpng, also used in pngtest.c * * Last changed in libpng 1.6.8 [December 19, 2013] - * Copyright (c) 1998-2013 Glenn Randers-Pehrson + * Copyright (c) 1998-2002,2004,2006-2013 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * diff --git a/third_party/libpng16/pngerror.c b/third_party/libpng16/pngerror.c index bdb959ee51..3fc8092fa6 100644 --- a/third_party/libpng16/pngerror.c +++ b/third_party/libpng16/pngerror.c @@ -2,7 +2,7 @@ /* pngerror.c - stub functions for i/o and memory allocation * * Last changed in libpng 1.6.15 [November 20, 2014] - * Copyright (c) 1998-2014 Glenn Randers-Pehrson + * Copyright (c) 1998-2002,2004,2006-2014 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * @@ -44,7 +44,7 @@ png_error,(png_const_structrp png_ptr, png_const_charp error_message), if (png_ptr != NULL) { if ((png_ptr->flags & - (PNG_FLAG_STRIP_ERROR_NUMBERS|PNG_FLAG_STRIP_ERROR_TEXT)) != 0 + (PNG_FLAG_STRIP_ERROR_NUMBERS|PNG_FLAG_STRIP_ERROR_TEXT)) != 0) { if (*error_message == PNG_LITERAL_SHARP) { diff --git a/third_party/libpng16/pngget.c b/third_party/libpng16/pngget.c index 743a6a9bbc..14fc7be520 100644 --- a/third_party/libpng16/pngget.c +++ b/third_party/libpng16/pngget.c @@ -2,7 +2,7 @@ /* pngget.c - retrieval of values from info struct * * Last changed in libpng 1.6.17 [March 26, 2015] - * Copyright (c) 1998-2015 Glenn Randers-Pehrson + * Copyright (c) 1998-2002,2004,2006-2015 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * @@ -456,11 +456,11 @@ png_get_pHYs_dpi(png_const_structrp png_ptr, png_const_inforp info_ptr, return (retval); } #endif /* pHYs */ -#endif /* INCH_CONVERSIONS */ +#endif /* INCH_CONVERSIONS */ /* png_get_channels really belongs in here, too, but it's been around longer */ -#endif /* EASY_ACCESS */ +#endif /* EASY_ACCESS */ png_byte PNGAPI @@ -1142,19 +1142,19 @@ png_get_compression_buffer_size(png_const_structrp png_ptr) return 0; #ifdef PNG_WRITE_SUPPORTED - if ((png_ptr->mode & PNG_IS_READ_STRUCT) != 0) + if ((png_ptr->mode & PNG_IS_READ_STRUCT) != 0) #endif { #ifdef PNG_SEQUENTIAL_READ_SUPPORTED - return png_ptr->IDAT_read_size; + return png_ptr->IDAT_read_size; #else - return PNG_IDAT_READ_SIZE; + return PNG_IDAT_READ_SIZE; #endif } #ifdef PNG_WRITE_SUPPORTED - else - return png_ptr->zbuffer_size; + else + return png_ptr->zbuffer_size; #endif } diff --git a/third_party/libpng16/pnginfo.h b/third_party/libpng16/pnginfo.h index 4bd264b869..361ed8be70 100644 --- a/third_party/libpng16/pnginfo.h +++ b/third_party/libpng16/pnginfo.h @@ -2,7 +2,7 @@ /* pnginfo.h - header file for PNG reference library * * Last changed in libpng 1.6.1 [March 28, 2013] - * Copyright (c) 1998-2013 Glenn Randers-Pehrson + * Copyright (c) 1998-2002,2004,2006-2013 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * diff --git a/third_party/libpng16/pngmem.c b/third_party/libpng16/pngmem.c index db71219520..b5507134fb 100644 --- a/third_party/libpng16/pngmem.c +++ b/third_party/libpng16/pngmem.c @@ -2,7 +2,7 @@ /* pngmem.c - stub functions for memory allocation * * Last changed in libpng 1.6.15 [November 20, 2014] - * Copyright (c) 1998-2014 Glenn Randers-Pehrson + * Copyright (c) 1998-2002,2004,2006-2014 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * diff --git a/third_party/libpng16/pngpread.c b/third_party/libpng16/pngpread.c index 89ffc4018f..a16fcbef3a 100644 --- a/third_party/libpng16/pngpread.c +++ b/third_party/libpng16/pngpread.c @@ -2,7 +2,7 @@ /* pngpread.c - read a png file in push mode * * Last changed in libpng 1.6.18 [July 23, 2015] - * Copyright (c) 1998-2015 Glenn Randers-Pehrson + * Copyright (c) 1998-2002,2004,2006-2015 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * @@ -210,12 +210,14 @@ png_push_read_chunk(png_structrp png_ptr, png_inforp info_ptr) (png_ptr->mode & PNG_HAVE_PLTE) == 0) png_error(png_ptr, "Missing PLTE before IDAT"); - png_ptr->mode |= PNG_HAVE_IDAT; png_ptr->process_mode = PNG_READ_IDAT_MODE; - if ((png_ptr->mode & PNG_HAVE_CHUNK_AFTER_IDAT) == 0) - if (png_ptr->push_length == 0) - return; + if ((png_ptr->mode & PNG_HAVE_IDAT) != 0) + if ((png_ptr->mode & PNG_HAVE_CHUNK_AFTER_IDAT) == 0) + if (png_ptr->push_length == 0) + return; + + png_ptr->mode |= PNG_HAVE_IDAT; if ((png_ptr->mode & PNG_AFTER_IDAT) != 0) png_benign_error(png_ptr, "Too many IDATs found"); diff --git a/third_party/libpng16/pngpriv.h b/third_party/libpng16/pngpriv.h index 58d3aa8159..9ea023fea7 100644 --- a/third_party/libpng16/pngpriv.h +++ b/third_party/libpng16/pngpriv.h @@ -1,8 +1,8 @@ /* pngpriv.h - private declarations for use inside libpng * - * Last changed in libpng 1.6.18 [July 23, 2015] - * Copyright (c) 1998-2015 Glenn Randers-Pehrson + * Last changed in libpng 1.6.22 [May 26, 2016] + * Copyright (c) 1998-2002,2004,2006-2016 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * @@ -127,7 +127,7 @@ */ # if (defined(__ARM_NEON__) || defined(__ARM_NEON)) && \ defined(PNG_ALIGNED_MEMORY_SUPPORTED) -# define PNG_ARM_NEON_OPT 0 +# define PNG_ARM_NEON_OPT 2 # else # define PNG_ARM_NEON_OPT 0 # endif @@ -182,6 +182,42 @@ # endif #endif /* PNG_ARM_NEON_OPT > 0 */ +#ifndef PNG_INTEL_SSE_OPT +# ifdef PNG_INTEL_SSE + /* Only check for SSE if the build configuration has been modified to + * enable SSE optimizations. This means that these optimizations will + * be off by default. See contrib/intel for more details. + */ +# if defined(__SSE4_1__) || defined(__AVX__) || defined(__SSSE3__) || \ + defined(__SSE2__) || defined(_M_X64) || defined(_M_AMD64) || \ + (defined(_M_IX86_FP) && _M_IX86_FP >= 2) +# define PNG_INTEL_SSE_OPT 1 +# endif +# endif +#endif + +#if PNG_INTEL_SSE_OPT > 0 +# ifndef PNG_INTEL_SSE_IMPLEMENTATION +# if defined(__SSE4_1__) || defined(__AVX__) + /* We are not actually using AVX, but checking for AVX is the best + way we can detect SSE4.1 and SSSE3 on MSVC. + */ +# define PNG_INTEL_SSE_IMPLEMENTATION 3 +# elif defined(__SSSE3__) +# define PNG_INTEL_SSE_IMPLEMENTATION 2 +# elif defined(__SSE2__) || defined(_M_X64) || defined(_M_AMD64) || \ + (defined(_M_IX86_FP) && _M_IX86_FP >= 2) +# define PNG_INTEL_SSE_IMPLEMENTATION 1 +# else +# define PNG_INTEL_SSE_IMPLEMENTATION 0 +# endif +# endif + +# if PNG_INTEL_SSE_IMPLEMENTATION > 0 +# define PNG_FILTER_OPTIMIZATIONS png_init_filter_functions_sse2 +# endif +#endif + /* Is this a build of a DLL where compilation of the object modules requires * different preprocessor settings to those required for a simple library? If * so PNG_BUILD_DLL must be set. @@ -1189,6 +1225,19 @@ PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth3_neon,(png_row_infop PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth4_neon,(png_row_infop row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); +PNG_INTERNAL_FUNCTION(void,png_read_filter_row_sub3_sse2,(png_row_infop + row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); +PNG_INTERNAL_FUNCTION(void,png_read_filter_row_sub4_sse2,(png_row_infop + row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); +PNG_INTERNAL_FUNCTION(void,png_read_filter_row_avg3_sse2,(png_row_infop + row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); +PNG_INTERNAL_FUNCTION(void,png_read_filter_row_avg4_sse2,(png_row_infop + row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); +PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth3_sse2,(png_row_infop + row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); +PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth4_sse2,(png_row_infop + row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); + /* Choose the best filter to use and filter the row data */ PNG_INTERNAL_FUNCTION(void,png_write_find_filter,(png_structrp png_ptr, png_row_infop row_info),PNG_EMPTY); @@ -1915,8 +1964,13 @@ PNG_INTERNAL_FUNCTION(void, PNG_FILTER_OPTIMIZATIONS, (png_structp png_ptr, */ PNG_INTERNAL_FUNCTION(void, png_init_filter_functions_neon, (png_structp png_ptr, unsigned int bpp), PNG_EMPTY); +PNG_INTERNAL_FUNCTION(void, png_init_filter_functions_sse2, + (png_structp png_ptr, unsigned int bpp), PNG_EMPTY); #endif +PNG_INTERNAL_FUNCTION(png_uint_32, png_check_keyword, (png_structrp png_ptr, + png_const_charp key, png_bytep new_key), PNG_EMPTY); + /* Maintainer: Put new private prototypes here ^ */ #include "pngdebug.h" diff --git a/third_party/libpng16/pngread.c b/third_party/libpng16/pngread.c index 9cb4d2e41d..0572c208cc 100644 --- a/third_party/libpng16/pngread.c +++ b/third_party/libpng16/pngread.c @@ -2,7 +2,7 @@ /* pngread.c - read a PNG file * * Last changed in libpng 1.6.17 [March 26, 2015] - * Copyright (c) 1998-2015 Glenn Randers-Pehrson + * Copyright (c) 1998-2002,2004,2006-2015 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * @@ -127,7 +127,10 @@ png_read_info(png_structrp png_ptr, png_inforp info_ptr) } else if ((png_ptr->mode & PNG_HAVE_IDAT) != 0) + { + png_ptr->mode |= PNG_HAVE_CHUNK_AFTER_IDAT; png_ptr->mode |= PNG_AFTER_IDAT; + } /* This should be a binary subdivision search or a hash for * matching the chunk name rather than a linear search. @@ -785,6 +788,9 @@ png_read_end(png_structrp png_ptr, png_inforp info_ptr) png_uint_32 length = png_read_chunk_header(png_ptr); png_uint_32 chunk_name = png_ptr->chunk_name; + if (chunk_name != png_IDAT) + png_ptr->mode |= PNG_HAVE_CHUNK_AFTER_IDAT; + if (chunk_name == png_IEND) png_handle_IEND(png_ptr, info_ptr, length); @@ -799,9 +805,9 @@ png_read_end(png_structrp png_ptr, png_inforp info_ptr) { if (chunk_name == png_IDAT) { - if ((length > 0) || - (png_ptr->mode & PNG_HAVE_CHUNK_AFTER_IDAT) != 0) - png_benign_error(png_ptr, "Too many IDATs found"); + if ((length > 0 && !(png_ptr->flags & PNG_FLAG_ZSTREAM_ENDED)) + || (png_ptr->mode & PNG_HAVE_CHUNK_AFTER_IDAT) != 0) + png_benign_error(png_ptr, ".Too many IDATs found"); } png_handle_unknown(png_ptr, info_ptr, length, keep); if (chunk_name == png_PLTE) @@ -812,10 +818,14 @@ png_read_end(png_structrp png_ptr, png_inforp info_ptr) else if (chunk_name == png_IDAT) { /* Zero length IDATs are legal after the last IDAT has been - * read, but not after other chunks have been read. + * read, but not after other chunks have been read. 1.6 does not + * always read all the deflate data; specifically it cannot be relied + * upon to read the Adler32 at the end. If it doesn't ignore IDAT + * chunks which are longer than zero as well: */ - if ((length > 0) || (png_ptr->mode & PNG_HAVE_CHUNK_AFTER_IDAT) != 0) - png_benign_error(png_ptr, "Too many IDATs found"); + if ((length > 0 && !(png_ptr->flags & PNG_FLAG_ZSTREAM_ENDED)) + || (png_ptr->mode & PNG_HAVE_CHUNK_AFTER_IDAT) != 0) + png_benign_error(png_ptr, "..Too many IDATs found"); png_crc_finish(png_ptr, length); } @@ -3848,16 +3858,16 @@ png_image_read_direct(png_voidp argument) else filler = 255; -# ifdef PNG_FORMAT_AFIRST_SUPPORTED - if ((format & PNG_FORMAT_FLAG_AFIRST) != 0) - { - where = PNG_FILLER_BEFORE; - change &= ~PNG_FORMAT_FLAG_AFIRST; - } +#ifdef PNG_FORMAT_AFIRST_SUPPORTED + if ((format & PNG_FORMAT_FLAG_AFIRST) != 0) + { + where = PNG_FILLER_BEFORE; + change &= ~PNG_FORMAT_FLAG_AFIRST; + } - else -# endif - where = PNG_FILLER_AFTER; + else +#endif + where = PNG_FILLER_AFTER; png_set_add_alpha(png_ptr, filler, where); } @@ -3965,12 +3975,12 @@ png_image_read_direct(png_voidp argument) if (info_ptr->bit_depth == 16) info_format |= PNG_FORMAT_FLAG_LINEAR; -# ifdef PNG_FORMAT_BGR_SUPPORTED - if ((png_ptr->transformations & PNG_BGR) != 0) - info_format |= PNG_FORMAT_FLAG_BGR; -# endif +#ifdef PNG_FORMAT_BGR_SUPPORTED + if ((png_ptr->transformations & PNG_BGR) != 0) + info_format |= PNG_FORMAT_FLAG_BGR; +#endif -# ifdef PNG_FORMAT_AFIRST_SUPPORTED +#ifdef PNG_FORMAT_AFIRST_SUPPORTED if (do_local_background == 2) { if ((format & PNG_FORMAT_FLAG_AFIRST) != 0) @@ -4071,58 +4081,84 @@ png_image_finish_read(png_imagep image, png_const_colorp background, { if (image != NULL && image->version == PNG_IMAGE_VERSION) { - png_uint_32 check; + /* Check for row_stride overflow. This check is not performed on the + * original PNG format because it may not occur in the output PNG format + * and libpng deals with the issues of reading the original. + */ + const unsigned int channels = PNG_IMAGE_PIXEL_CHANNELS(image->format); - if (row_stride == 0) - row_stride = PNG_IMAGE_ROW_STRIDE(*image); + if (image->width <= 0x7FFFFFFFU/channels) /* no overflow */ + { + png_uint_32 check; + const png_uint_32 png_row_stride = image->width * channels; - if (row_stride < 0) - check = -row_stride; + if (row_stride == 0) + row_stride = (png_int_32)/*SAFE*/png_row_stride; - else - check = row_stride; + if (row_stride < 0) + check = -row_stride; - if (image->opaque != NULL && buffer != NULL && - check >= PNG_IMAGE_ROW_STRIDE(*image)) - { - if ((image->format & PNG_FORMAT_FLAG_COLORMAP) == 0 || - (image->colormap_entries > 0 && colormap != NULL)) + else + check = row_stride; + + if (image->opaque != NULL && buffer != NULL && check >= png_row_stride) { - int result; - png_image_read_control display; - - memset(&display, 0, (sizeof display)); - display.image = image; - display.buffer = buffer; - display.row_stride = row_stride; - display.colormap = colormap; - display.background = background; - display.local_row = NULL; - - /* Choose the correct 'end' routine; for the color-map case all the - * setup has already been done. + /* Now check for overflow of the image buffer calculation; this + * limits the whole image size to 32 bits for API compatibility with + * the current, 32-bit, PNG_IMAGE_BUFFER_SIZE macro. */ - if ((image->format & PNG_FORMAT_FLAG_COLORMAP) != 0) - result = - png_safe_execute(image, png_image_read_colormap, &display) && - png_safe_execute(image, png_image_read_colormapped, &display); + if (image->height <= 0xFFFFFFFF/png_row_stride) + { + if ((image->format & PNG_FORMAT_FLAG_COLORMAP) == 0 || + (image->colormap_entries > 0 && colormap != NULL)) + { + int result; + png_image_read_control display; + + memset(&display, 0, (sizeof display)); + display.image = image; + display.buffer = buffer; + display.row_stride = row_stride; + display.colormap = colormap; + display.background = background; + display.local_row = NULL; + + /* Choose the correct 'end' routine; for the color-map case + * all the setup has already been done. + */ + if ((image->format & PNG_FORMAT_FLAG_COLORMAP) != 0) + result = png_safe_execute(image, + png_image_read_colormap, &display) && + png_safe_execute(image, + png_image_read_colormapped, &display); - else - result = - png_safe_execute(image, png_image_read_direct, &display); + else + result = + png_safe_execute(image, + png_image_read_direct, &display); + + png_image_free(image); + return result; + } - png_image_free(image); - return result; + else + return png_image_error(image, + "png_image_finish_read[color-map]: no color-map"); + } + + else + return png_image_error(image, + "png_image_finish_read: image too large"); } else return png_image_error(image, - "png_image_finish_read[color-map]: no color-map"); + "png_image_finish_read: invalid argument"); } else return png_image_error(image, - "png_image_finish_read: invalid argument"); + "png_image_finish_read: row_stride too large"); } else if (image != NULL) diff --git a/third_party/libpng16/pngrio.c b/third_party/libpng16/pngrio.c index 38f7fd49d6..5101d54a62 100644 --- a/third_party/libpng16/pngrio.c +++ b/third_party/libpng16/pngrio.c @@ -2,7 +2,7 @@ /* pngrio.c - functions for data input * * Last changed in libpng 1.6.17 [March 26, 2015] - * Copyright (c) 1998-2015 Glenn Randers-Pehrson + * Copyright (c) 1998-2002,2004,2006-2015 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * diff --git a/third_party/libpng16/pngrtran.c b/third_party/libpng16/pngrtran.c index f129ef129c..3138147aff 100644 --- a/third_party/libpng16/pngrtran.c +++ b/third_party/libpng16/pngrtran.c @@ -1,8 +1,8 @@ /* pngrtran.c - transforms the data in a row for PNG readers * - * Last changed in libpng 1.6.19 [November 12, 2015] - * Copyright (c) 1998-2015 Glenn Randers-Pehrson + * Last changed in libpng 1.6.22 [May 26, 2016] + * Copyright (c) 1998-2002,2004,2006-2016 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * @@ -159,7 +159,7 @@ png_set_background(png_structrp png_ptr, png_set_background_fixed(png_ptr, background_color, background_gamma_code, need_expand, png_fixed(png_ptr, background_gamma, "png_set_background")); } -# endif /* FLOATING_POINT */ +# endif /* FLOATING_POINT */ #endif /* READ_BACKGROUND */ /* Scale 16-bit depth files to 8-bit depth. If both of these are set then the @@ -289,9 +289,12 @@ png_set_alpha_mode_fixed(png_structrp png_ptr, int mode, * is expected to be 1 or greater, but this range test allows for some * viewing correction values. The intent is to weed out users of this API * who use the inverse of the gamma value accidentally! Since some of these - * values are reasonable this may have to be changed. + * values are reasonable this may have to be changed: + * + * 1.6.x: changed from 0.07..3 to 0.01..100 (to accomodate the optimal 16-bit + * gamma of 36, and its reciprocal.) */ - if (output_gamma < 70000 || output_gamma > 300000) + if (output_gamma < 1000 || output_gamma > 10000000) png_error(png_ptr, "output gamma out of expected range"); /* The default file gamma is the inverse of the output gamma; the output @@ -1912,7 +1915,7 @@ png_init_read_transformations(png_structrp png_ptr) png_ptr->palette[i].blue = (png_byte)component; } } -#endif /* READ_SHIFT */ +#endif /* READ_SHIFT */ } /* Modify the info structure to reflect the transformations. The diff --git a/third_party/libpng16/pngrutil.c b/third_party/libpng16/pngrutil.c index 6189251352..c9747fc27d 100644 --- a/third_party/libpng16/pngrutil.c +++ b/third_party/libpng16/pngrutil.c @@ -1,8 +1,8 @@ /* pngrutil.c - utilities to read a PNG file * - * Last changed in libpng 1.6.20 [December 3, 2015] - * Copyright (c) 1998-2015 Glenn Randers-Pehrson + * Last changed in libpng 1.6.20 [December 3, 2014] + * Copyright (c) 1998-2002,2004,2006-2015 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * diff --git a/third_party/libpng16/pngset.c b/third_party/libpng16/pngset.c index 8fd7965fca..303328f87e 100644 --- a/third_party/libpng16/pngset.c +++ b/third_party/libpng16/pngset.c @@ -1,7 +1,7 @@ /* pngset.c - storage of image information into info struct * - * Last changed in libpng 1.6.19 [November 12, 2015] + * Last changed in libpng 1.6.21 [January 15, 2016] * Copyright (c) 1998-2015 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) @@ -952,12 +952,15 @@ png_set_tRNS(png_structrp png_ptr, png_inforp info_ptr, png_free_data(png_ptr, info_ptr, PNG_FREE_TRNS, 0); - /* Changed from num_trans to PNG_MAX_PALETTE_LENGTH in version 1.2.1 */ - png_ptr->trans_alpha = info_ptr->trans_alpha = png_voidcast(png_bytep, - png_malloc(png_ptr, PNG_MAX_PALETTE_LENGTH)); - if (num_trans > 0 && num_trans <= PNG_MAX_PALETTE_LENGTH) + { + /* Changed from num_trans to PNG_MAX_PALETTE_LENGTH in version 1.2.1 */ + info_ptr->trans_alpha = png_voidcast(png_bytep, + png_malloc(png_ptr, PNG_MAX_PALETTE_LENGTH)); + memcpy(info_ptr->trans_alpha, trans_alpha, (png_size_t)num_trans); + } + png_ptr->trans_alpha = info_ptr->trans_alpha; } if (trans_color != NULL) @@ -1644,4 +1647,88 @@ png_set_check_for_invalid_index(png_structrp png_ptr, int allowed) png_ptr->num_palette_max = -1; } #endif + +#if defined(PNG_TEXT_SUPPORTED) || defined(PNG_pCAL_SUPPORTED) || \ + defined(PNG_iCCP_SUPPORTED) || defined(PNG_sPLT_SUPPORTED) +/* Check that the tEXt or zTXt keyword is valid per PNG 1.0 specification, + * and if invalid, correct the keyword rather than discarding the entire + * chunk. The PNG 1.0 specification requires keywords 1-79 characters in + * length, forbids leading or trailing whitespace, multiple internal spaces, + * and the non-break space (0x80) from ISO 8859-1. Returns keyword length. + * + * The 'new_key' buffer must be 80 characters in size (for the keyword plus a + * trailing '\0'). If this routine returns 0 then there was no keyword, or a + * valid one could not be generated, and the caller must png_error. + */ +png_uint_32 /* PRIVATE */ +png_check_keyword(png_structrp png_ptr, png_const_charp key, png_bytep new_key) +{ + png_const_charp orig_key = key; + png_uint_32 key_len = 0; + int bad_character = 0; + int space = 1; + + png_debug(1, "in png_check_keyword"); + + if (key == NULL) + { + *new_key = 0; + return 0; + } + + while (*key && key_len < 79) + { + png_byte ch = (png_byte)*key++; + + if ((ch > 32 && ch <= 126) || (ch >= 161 /*&& ch <= 255*/)) + *new_key++ = ch, ++key_len, space = 0; + + else if (space == 0) + { + /* A space or an invalid character when one wasn't seen immediately + * before; output just a space. + */ + *new_key++ = 32, ++key_len, space = 1; + + /* If the character was not a space then it is invalid. */ + if (ch != 32) + bad_character = ch; + } + + else if (bad_character == 0) + bad_character = ch; /* just skip it, record the first error */ + } + + if (key_len > 0 && space != 0) /* trailing space */ + { + --key_len, --new_key; + if (bad_character == 0) + bad_character = 32; + } + + /* Terminate the keyword */ + *new_key = 0; + + if (key_len == 0) + return 0; + +#ifdef PNG_WARNINGS_SUPPORTED + /* Try to only output one warning per keyword: */ + if (*key != 0) /* keyword too long */ + png_warning(png_ptr, "keyword truncated"); + + else if (bad_character != 0) + { + PNG_WARNING_PARAMETERS(p) + + png_warning_parameter(p, 1, orig_key); + png_warning_parameter_signed(p, 2, PNG_NUMBER_FORMAT_02x, bad_character); + + png_formatted_warning(png_ptr, p, "keyword \"@1\": bad character '0x@2'"); + } +#endif /* WARNINGS */ + + return key_len; +} +#endif /* TEXT || pCAL || iCCP || sPLT */ #endif /* READ || WRITE */ diff --git a/third_party/libpng16/pngstruct.h b/third_party/libpng16/pngstruct.h index ad1e7a5adc..0b372de4c6 100644 --- a/third_party/libpng16/pngstruct.h +++ b/third_party/libpng16/pngstruct.h @@ -2,7 +2,7 @@ /* pngstruct.h - header file for PNG reference library * * Last changed in libpng 1.6.18 [July 23, 2015] - * Copyright (c) 1998-2015 Glenn Randers-Pehrson + * Copyright (c) 1998-2002,2004,2006-2015 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * diff --git a/third_party/libpng16/pngtrans.c b/third_party/libpng16/pngtrans.c index 0c0d92d8fd..7f8cc455d3 100644 --- a/third_party/libpng16/pngtrans.c +++ b/third_party/libpng16/pngtrans.c @@ -2,7 +2,7 @@ /* pngtrans.c - transforms the data in a row (used by both readers and writers) * * Last changed in libpng 1.6.18 [July 23, 2015] - * Copyright (c) 1998-2015 Glenn Randers-Pehrson + * Copyright (c) 1998-2002,2004,2006-2015 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * diff --git a/third_party/libpng16/pngwio.c b/third_party/libpng16/pngwio.c index db76e6b8d7..586c03b721 100644 --- a/third_party/libpng16/pngwio.c +++ b/third_party/libpng16/pngwio.c @@ -2,7 +2,7 @@ /* pngwio.c - functions for data output * * Last changed in libpng 1.6.15 [November 20, 2014] - * Copyright (c) 1998-2014 Glenn Randers-Pehrson + * Copyright (c) 1998-2002,2004,2006-2014 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * diff --git a/third_party/libpng16/pngwrite.c b/third_party/libpng16/pngwrite.c index 0d4ee9f7ae..181a899438 100644 --- a/third_party/libpng16/pngwrite.c +++ b/third_party/libpng16/pngwrite.c @@ -2,7 +2,7 @@ /* pngwrite.c - general routines to write a PNG file * * Last changed in libpng 1.6.19 [November 12, 2015] - * Copyright (c) 1998-2015 Glenn Randers-Pehrson + * Copyright (c) 1998-2002,2004,2006-2015 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * @@ -12,9 +12,9 @@ */ #include "pngpriv.h" -#if defined(PNG_SIMPLIFIED_WRITE_SUPPORTED) && defined(PNG_STDIO_SUPPORTED) +#ifdef PNG_SIMPLIFIED_WRITE_STDIO_SUPPORTED # include <errno.h> -#endif +#endif /* SIMPLIFIED_WRITE_STDIO */ #ifdef PNG_WRITE_SUPPORTED @@ -1452,7 +1452,6 @@ png_write_png(png_structrp png_ptr, png_inforp info_ptr, #ifdef PNG_SIMPLIFIED_WRITE_SUPPORTED -# ifdef PNG_STDIO_SUPPORTED /* currently required for png_image_write_* */ /* Initialize the write structure - general purpose utility. */ static int png_image_write_init(png_imagep image) @@ -1504,6 +1503,10 @@ typedef struct png_const_voidp first_row; ptrdiff_t row_bytes; png_voidp local_row; + /* Byte count for memory writing */ + png_bytep memory; + png_alloc_size_t memory_bytes; /* not used for STDIO */ + png_alloc_size_t output_bytes; /* running total */ } png_image_write_control; /* Write png_uint_16 input to a 16-bit PNG; the png_ptr has already been set to @@ -1931,9 +1934,43 @@ png_image_write_main(png_voidp argument) png_set_benign_errors(png_ptr, 0/*error*/); # endif - /* Default the 'row_stride' parameter if required. */ - if (display->row_stride == 0) - display->row_stride = PNG_IMAGE_ROW_STRIDE(*image); + /* Default the 'row_stride' parameter if required, also check the row stride + * and total image size to ensure that they are within the system limits. + */ + { + const unsigned int channels = PNG_IMAGE_PIXEL_CHANNELS(image->format); + + if (image->width <= 0x7FFFFFFFU/channels) /* no overflow */ + { + png_uint_32 check; + const png_uint_32 png_row_stride = image->width * channels; + + if (display->row_stride == 0) + display->row_stride = (png_int_32)/*SAFE*/png_row_stride; + + if (display->row_stride < 0) + check = -display->row_stride; + + else + check = display->row_stride; + + if (check >= png_row_stride) + { + /* Now check for overflow of the image buffer calculation; this + * limits the whole image size to 32 bits for API compatibility with + * the current, 32-bit, PNG_IMAGE_BUFFER_SIZE macro. + */ + if (image->height > 0xFFFFFFFF/png_row_stride) + png_error(image->opaque->png_ptr, "memory image too large"); + } + + else + png_error(image->opaque->png_ptr, "supplied row stride too small"); + } + + else + png_error(image->opaque->png_ptr, "image row stride too large"); + } /* Set the required transforms then write the rows in the correct order. */ if ((format & PNG_FORMAT_FLAG_COLORMAP) != 0) @@ -2110,6 +2147,122 @@ png_image_write_main(png_voidp argument) return 1; } + +static void (PNGCBAPI +image_memory_write)(png_structp png_ptr, png_bytep/*const*/ data, + png_size_t size) +{ + png_image_write_control *display = png_voidcast(png_image_write_control*, + png_ptr->io_ptr/*backdoor: png_get_io_ptr(png_ptr)*/); + const png_alloc_size_t ob = display->output_bytes; + + /* Check for overflow; this should never happen: */ + if (size <= ((png_alloc_size_t)-1) - ob) + { + /* I don't think libpng ever does this, but just in case: */ + if (size > 0) + { + if (display->memory_bytes >= ob+size) /* writing */ + memcpy(display->memory+ob, data, size); + + /* Always update the size: */ + display->output_bytes = ob+size; + } + } + + else + png_error(png_ptr, "png_image_write_to_memory: PNG too big"); +} + +static void (PNGCBAPI +image_memory_flush)(png_structp png_ptr) +{ + PNG_UNUSED(png_ptr) +} + +static int +png_image_write_memory(png_voidp argument) +{ + png_image_write_control *display = png_voidcast(png_image_write_control*, + argument); + + /* The rest of the memory-specific init and write_main in an error protected + * environment. This case needs to use callbacks for the write operations + * since libpng has no built in support for writing to memory. + */ + png_set_write_fn(display->image->opaque->png_ptr, display/*io_ptr*/, + image_memory_write, image_memory_flush); + + return png_image_write_main(display); +} + +int PNGAPI +png_image_write_to_memory(png_imagep image, void *memory, + png_alloc_size_t * PNG_RESTRICT memory_bytes, int convert_to_8bit, + const void *buffer, png_int_32 row_stride, const void *colormap) +{ + /* Write the image to the given buffer, or count the bytes if it is NULL */ + if (image != NULL && image->version == PNG_IMAGE_VERSION) + { + if (memory_bytes != NULL && buffer != NULL) + { + /* This is to give the caller an easier error detection in the NULL + * case and guard against uninitialized variable problems: + */ + if (memory == NULL) + *memory_bytes = 0; + + if (png_image_write_init(image) != 0) + { + png_image_write_control display; + int result; + + memset(&display, 0, (sizeof display)); + display.image = image; + display.buffer = buffer; + display.row_stride = row_stride; + display.colormap = colormap; + display.convert_to_8bit = convert_to_8bit; + display.memory = png_voidcast(png_bytep, memory); + display.memory_bytes = *memory_bytes; + display.output_bytes = 0; + + result = png_safe_execute(image, png_image_write_memory, &display); + png_image_free(image); + + /* write_memory returns true even if we ran out of buffer. */ + if (result) + { + /* On out-of-buffer this function returns '0' but still updates + * memory_bytes: + */ + if (memory != NULL && display.output_bytes > *memory_bytes) + result = 0; + + *memory_bytes = display.output_bytes; + } + + return result; + } + + else + return 0; + } + + else + return png_image_error(image, + "png_image_write_to_memory: invalid argument"); + } + + else if (image != NULL) + return png_image_error(image, + "png_image_write_to_memory: incorrect PNG_IMAGE_VERSION"); + + else + return 0; +} + +#ifdef PNG_SIMPLIFIED_WRITE_STDIO_SUPPORTED int PNGAPI png_image_write_to_stdio(png_imagep image, FILE *file, int convert_to_8bit, const void *buffer, png_int_32 row_stride, const void *colormap) @@ -2117,7 +2270,7 @@ png_image_write_to_stdio(png_imagep image, FILE *file, int convert_to_8bit, /* Write the image to the given (FILE*). */ if (image != NULL && image->version == PNG_IMAGE_VERSION) { - if (file != NULL) + if (file != NULL && buffer != NULL) { if (png_image_write_init(image) != 0) { @@ -2167,7 +2320,7 @@ png_image_write_to_file(png_imagep image, const char *file_name, /* Write the image to the named file. */ if (image != NULL && image->version == PNG_IMAGE_VERSION) { - if (file_name != NULL) + if (file_name != NULL && buffer != NULL) { FILE *fp = fopen(file_name, "wb"); @@ -2225,6 +2378,6 @@ png_image_write_to_file(png_imagep image, const char *file_name, else return 0; } -# endif /* STDIO */ +#endif /* SIMPLIFIED_WRITE_STDIO */ #endif /* SIMPLIFIED_WRITE */ #endif /* WRITE */ diff --git a/third_party/libpng16/pngwtran.c b/third_party/libpng16/pngwtran.c index 5dc949157d..038a2ef5dc 100644 --- a/third_party/libpng16/pngwtran.c +++ b/third_party/libpng16/pngwtran.c @@ -2,7 +2,7 @@ /* pngwtran.c - transforms the data in a row for PNG writers * * Last changed in libpng 1.6.18 [July 23, 2015] - * Copyright (c) 1998-2015 Glenn Randers-Pehrson + * Copyright (c) 1998-2002,2004,2006-2015 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * diff --git a/third_party/libpng16/pngwutil.c b/third_party/libpng16/pngwutil.c index 0ee102b5fb..ca8a03ad5f 100644 --- a/third_party/libpng16/pngwutil.c +++ b/third_party/libpng16/pngwutil.c @@ -1,8 +1,8 @@ /* pngwutil.c - utilities to write a PNG file * - * Last changed in libpng 1.6.19 [November 12, 2015] - * Copyright (c) 1998-2015 Glenn Randers-Pehrson + * Last changed in libpng 1.6.22 [May 26, 2016] + * Copyright (c) 1998-2002,2004,2006-2016 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) * @@ -23,10 +23,10 @@ void PNGAPI png_save_uint_32(png_bytep buf, png_uint_32 i) { - buf[0] = (png_byte)(i >> 24); - buf[1] = (png_byte)(i >> 16); - buf[2] = (png_byte)(i >> 8); - buf[3] = (png_byte)(i ); + buf[0] = (png_byte)((i >> 24) & 0xffU); + buf[1] = (png_byte)((i >> 16) & 0xffU); + buf[2] = (png_byte)((i >> 8) & 0xffU); + buf[3] = (png_byte)( i & 0xffU); } /* Place a 16-bit number into a buffer in PNG byte order. @@ -36,8 +36,8 @@ png_save_uint_32(png_bytep buf, png_uint_32 i) void PNGAPI png_save_uint_16(png_bytep buf, unsigned int i) { - buf[0] = (png_byte)(i >> 8); - buf[1] = (png_byte)(i ); + buf[0] = (png_byte)((i >> 8) & 0xffU); + buf[1] = (png_byte)( i & 0xffU); } #endif @@ -665,90 +665,6 @@ png_write_compressed_data_out(png_structrp png_ptr, compression_state *comp) } #endif /* WRITE_COMPRESSED_TEXT */ -#if defined(PNG_WRITE_TEXT_SUPPORTED) || defined(PNG_WRITE_pCAL_SUPPORTED) || \ - defined(PNG_WRITE_iCCP_SUPPORTED) || defined(PNG_WRITE_sPLT_SUPPORTED) -/* Check that the tEXt or zTXt keyword is valid per PNG 1.0 specification, - * and if invalid, correct the keyword rather than discarding the entire - * chunk. The PNG 1.0 specification requires keywords 1-79 characters in - * length, forbids leading or trailing whitespace, multiple internal spaces, - * and the non-break space (0x80) from ISO 8859-1. Returns keyword length. - * - * The 'new_key' buffer must be 80 characters in size (for the keyword plus a - * trailing '\0'). If this routine returns 0 then there was no keyword, or a - * valid one could not be generated, and the caller must png_error. - */ -static png_uint_32 -png_check_keyword(png_structrp png_ptr, png_const_charp key, png_bytep new_key) -{ - png_const_charp orig_key = key; - png_uint_32 key_len = 0; - int bad_character = 0; - int space = 1; - - png_debug(1, "in png_check_keyword"); - - if (key == NULL) - { - *new_key = 0; - return 0; - } - - while (*key && key_len < 79) - { - png_byte ch = (png_byte)*key++; - - if ((ch > 32 && ch <= 126) || (ch >= 161 /*&& ch <= 255*/)) - *new_key++ = ch, ++key_len, space = 0; - - else if (space == 0) - { - /* A space or an invalid character when one wasn't seen immediately - * before; output just a space. - */ - *new_key++ = 32, ++key_len, space = 1; - - /* If the character was not a space then it is invalid. */ - if (ch != 32) - bad_character = ch; - } - - else if (bad_character == 0) - bad_character = ch; /* just skip it, record the first error */ - } - - if (key_len > 0 && space != 0) /* trailing space */ - { - --key_len, --new_key; - if (bad_character == 0) - bad_character = 32; - } - - /* Terminate the keyword */ - *new_key = 0; - - if (key_len == 0) - return 0; - -#ifdef PNG_WARNINGS_SUPPORTED - /* Try to only output one warning per keyword: */ - if (*key != 0) /* keyword too long */ - png_warning(png_ptr, "keyword truncated"); - - else if (bad_character != 0) - { - PNG_WARNING_PARAMETERS(p) - - png_warning_parameter(p, 1, orig_key); - png_warning_parameter_signed(p, 2, PNG_NUMBER_FORMAT_02x, bad_character); - - png_formatted_warning(png_ptr, p, "keyword \"@1\": bad character '0x@2'"); - } -#endif /* WARNINGS */ - - return key_len; -} -#endif /* WRITE_TEXT || WRITE_pCAL || WRITE_iCCP || WRITE_sPLT */ - /* Write the IHDR chunk, and update the png_struct with the necessary * information. Note that the rest of this code depends upon this * information being correct. @@ -2362,6 +2278,28 @@ png_setup_sub_row(png_structrp png_ptr, const png_uint_32 bpp, return (sum); } +static void /* PRIVATE */ +png_setup_sub_row_only(png_structrp png_ptr, const png_uint_32 bpp, + const png_size_t row_bytes) +{ + png_bytep rp, dp, lp; + png_size_t i; + + png_ptr->try_row[0] = PNG_FILTER_VALUE_SUB; + + for (i = 0, rp = png_ptr->row_buf + 1, dp = png_ptr->try_row + 1; i < bpp; + i++, rp++, dp++) + { + *dp = *rp; + } + + for (lp = png_ptr->row_buf + 1; i < row_bytes; + i++, rp++, lp++, dp++) + { + *dp = (png_byte)(((int)*rp - (int)*lp) & 0xff); + } +} + static png_size_t /* PRIVATE */ png_setup_up_row(png_structrp png_ptr, const png_size_t row_bytes, const png_size_t lmins) @@ -2386,6 +2324,21 @@ png_setup_up_row(png_structrp png_ptr, const png_size_t row_bytes, return (sum); } +static void /* PRIVATE */ +png_setup_up_row_only(png_structrp png_ptr, const png_size_t row_bytes) +{ + png_bytep rp, dp, pp; + png_size_t i; + + png_ptr->try_row[0] = PNG_FILTER_VALUE_UP; + + for (i = 0, rp = png_ptr->row_buf + 1, dp = png_ptr->try_row + 1, + pp = png_ptr->prev_row + 1; i < row_bytes; + i++, rp++, pp++, dp++) + { + *dp = (png_byte)(((int)*rp - (int)*pp) & 0xff); + } +} static png_size_t /* PRIVATE */ png_setup_avg_row(png_structrp png_ptr, const png_uint_32 bpp, @@ -2419,6 +2372,27 @@ png_setup_avg_row(png_structrp png_ptr, const png_uint_32 bpp, return (sum); } +static void /* PRIVATE */ +png_setup_avg_row_only(png_structrp png_ptr, const png_uint_32 bpp, + const png_size_t row_bytes) +{ + png_bytep rp, dp, pp, lp; + png_uint_32 i; + + png_ptr->try_row[0] = PNG_FILTER_VALUE_AVG; + + for (i = 0, rp = png_ptr->row_buf + 1, dp = png_ptr->try_row + 1, + pp = png_ptr->prev_row + 1; i < bpp; i++) + { + *dp++ = (png_byte)(((int)*rp++ - ((int)*pp++ / 2)) & 0xff); + } + + for (lp = png_ptr->row_buf + 1; i < row_bytes; i++) + { + *dp++ = (png_byte)(((int)*rp++ - (((int)*pp++ + (int)*lp++) / 2)) + & 0xff); + } +} static png_size_t /* PRIVATE */ png_setup_paeth_row(png_structrp png_ptr, const png_uint_32 bpp, @@ -2473,6 +2447,48 @@ png_setup_paeth_row(png_structrp png_ptr, const png_uint_32 bpp, return (sum); } +static void /* PRIVATE */ +png_setup_paeth_row_only(png_structrp png_ptr, const png_uint_32 bpp, + const png_size_t row_bytes) +{ + png_bytep rp, dp, pp, cp, lp; + png_size_t i; + + png_ptr->try_row[0] = PNG_FILTER_VALUE_PAETH; + + for (i = 0, rp = png_ptr->row_buf + 1, dp = png_ptr->try_row + 1, + pp = png_ptr->prev_row + 1; i < bpp; i++) + { + *dp++ = (png_byte)(((int)*rp++ - (int)*pp++) & 0xff); + } + + for (lp = png_ptr->row_buf + 1, cp = png_ptr->prev_row + 1; i < row_bytes; + i++) + { + int a, b, c, pa, pb, pc, p; + + b = *pp++; + c = *cp++; + a = *lp++; + + p = b - c; + pc = a - c; + +#ifdef PNG_USE_ABS + pa = abs(p); + pb = abs(pc); + pc = abs(p + pc); +#else + pa = p < 0 ? -p : p; + pb = pc < 0 ? -pc : pc; + pc = (p + pc) < 0 ? -(p + pc) : p + pc; +#endif + + p = (pa <= pb && pa <=pc) ? a : (pb <= pc) ? b : c; + + *dp++ = (png_byte)(((int)*rp++ - p) & 0xff); + } +} #endif /* WRITE_FILTER */ void /* PRIVATE */ @@ -2481,7 +2497,7 @@ png_write_find_filter(png_structrp png_ptr, png_row_infop row_info) #ifndef PNG_WRITE_FILTER_SUPPORTED png_write_filtered_row(png_ptr, png_ptr->row_buf, row_info->rowbytes+1); #else - png_byte filter_to_do = png_ptr->do_filter; + unsigned int filter_to_do = png_ptr->do_filter; png_bytep row_buf; png_bytep best_row; png_uint_32 bpp; @@ -2527,27 +2543,24 @@ png_write_find_filter(png_structrp png_ptr, png_row_infop row_info) */ best_row = png_ptr->row_buf; - - if ((filter_to_do & PNG_FILTER_NONE) != 0 && filter_to_do != PNG_FILTER_NONE) + if (PNG_SIZE_MAX/128 <= row_bytes) + { + /* Overflow can occur in the calculation, just select the lowest set + * filter. + */ + filter_to_do &= -filter_to_do; + } + else if ((filter_to_do & PNG_FILTER_NONE) != 0 && + filter_to_do != PNG_FILTER_NONE) { + /* Overflow not possible and multiple filters in the list, including the + * 'none' filter. + */ png_bytep rp; png_size_t sum = 0; png_size_t i; int v; - if (PNG_SIZE_MAX/128 <= row_bytes) - { - for (i = 0, rp = row_buf + 1; i < row_bytes; i++, rp++) - { - /* Check for overflow */ - if (sum > PNG_SIZE_MAX/128 - 256) - break; - - v = *rp; - sum += (v < 128) ? v : 256 - v; - } - } - else /* Overflow is not possible */ { for (i = 0, rp = row_buf + 1; i < row_bytes; i++, rp++) { @@ -2563,7 +2576,7 @@ png_write_find_filter(png_structrp png_ptr, png_row_infop row_info) if (filter_to_do == PNG_FILTER_SUB) /* It's the only filter so no testing is needed */ { - (void) png_setup_sub_row(png_ptr, bpp, row_bytes, mins); + png_setup_sub_row_only(png_ptr, bpp, row_bytes); best_row = png_ptr->try_row; } @@ -2589,7 +2602,7 @@ png_write_find_filter(png_structrp png_ptr, png_row_infop row_info) /* Up filter */ if (filter_to_do == PNG_FILTER_UP) { - (void) png_setup_up_row(png_ptr, row_bytes, mins); + png_setup_up_row_only(png_ptr, row_bytes); best_row = png_ptr->try_row; } @@ -2615,7 +2628,7 @@ png_write_find_filter(png_structrp png_ptr, png_row_infop row_info) /* Avg filter */ if (filter_to_do == PNG_FILTER_AVG) { - (void) png_setup_avg_row(png_ptr, bpp, row_bytes, mins); + png_setup_avg_row_only(png_ptr, bpp, row_bytes); best_row = png_ptr->try_row; } @@ -2641,7 +2654,7 @@ png_write_find_filter(png_structrp png_ptr, png_row_infop row_info) /* Paeth filter */ if ((filter_to_do == PNG_FILTER_PAETH) != 0) { - (void) png_setup_paeth_row(png_ptr, bpp, row_bytes, mins); + png_setup_paeth_row_only(png_ptr, bpp, row_bytes); best_row = png_ptr->try_row; } |