Skip to content

Commit

Permalink
PNG 2024-09-12 (a7e0c0c8)
Browse files Browse the repository at this point in the history 
Run the UpdateFromUpstream.sh script to extract upstream PNG
using the following shell commands.

$ git archive --prefix=upstream-png/ a7e0c0c8 -- 
  png*.c
  png*.h
  arm/*
  LICENSE
  scripts/pnglibconf.h.prebuilt
   | tar x
$ git shortlog --perl-regexp --author='^((?!Kitware Robot).*)$' --no-merges --abbrev=8 --format='%h %s' 7a614829..a7e0c0c8

Ben Wagner (1):
      40878fd6 fix: Restore STDERR in pngtest.c

Benjamin Buch (2):
      7b888092 build: Mark the installed libpng headers as system headers in CMake
      0e204b73 build: Add a CMake config file compatible with the FindPNG module

Cosmin Truta (35):
      8120345c ci: Update (again) the ci_verify_*.sh scripts; update .shellcheckrc
      1cdde118 build: Update the makefiles for the benefit of cross-compilation
      63c715b7 ci: Remove the workaround for CI_AR from ci_verify_makefiles.sh
      fb65436e ci: Fix the check of `PNG_LIBPNG_VER_BUILD` in ci_verify_version.sh
      f74d5ecc ci: Pacify shellcheck version 0.8 and apply other linting improvements
      42c8fcbf Add a GitHub Action for linting
      4191872d chore: Update the .editorconfig files and pacify editorconfig-checker
      0fa3c0f6 chore: Clean up the spurious uses of `sizeof(png_byte)`; fix the manual
      72c4520d ci: Allow the user to force an in-tree cleanup before verification
      dddaf0c6 ci: Fix the reporting in ci_lint.sh
      6b5a2da0 Fix "ci: Fix the reporting in ci_lint.sh"
      aa95dee6 build: Update the CMake build options PNG_TOOLS and PNG_FRAMEWORK
      d165a20a build: Improve the search for an AWK processor in the CMake build
      29e31f62 build: Add an explicit declaration of the AWK variable to configure.ac
      14a348dd build: Checking for compiler support of LoongArch LSX should be guarded
      ec2e58c1 pngexif: Import pngexifinfo as an externally-contributed project
      e05ebfba doc: Update the README file
      3b9a73ed doc: Review the libpng history and update scripts/cmake/AUTHORS.md
      89023102 chore: Delete comments and config settings and stuff from here and there
      80691b9d test: Fix a compiler warning in pngtest.c
      e1fa61da ci: Add the libpng release tags to the list of exclusions
      ed217e3e Release libpng version 1.6.43
      f1848a3b Bump version to 1.6.44.git
      e4a31f02 arm: Add a placeholder file in lieu of the former `filter_neon.S`
      532fec02 ci: Fix the verification of the msys2 toolchain on AppVeyor CI
      33ef48b6 cmake: Fix the handling of PNG_HARDWARE_OPTIMIZATIONS on FreeBSD/amd64
      7e18d142 ci: Add the targets/ subdirectory to facilitate cross-platform testing
      22208658 chore: Pacify editorconfig-checker version 3.0
      fcdec9c6 chore: Delete contrib/tools/chkfmt.sh
      53a7f4e3 ci: Simplify the Travis CI configuration matrix
      1964d560 Deprecate PNGARG and remove all of its remaining uses
      843dbb75 Revert "cmake: Fix an error in the declaration of target include directories"
      3117b5f9 oss-fuzz: Update the README file, the Docker file and the build script
      88ab4f59 chore: Rerun `./autogen.sh --maintainer`
      f5e92d76 Release libpng version 1.6.44

Dan Rosser (1):
      e7ba9c0d build: Fix a CMake build regression introduced in version 1.6.41

Eric Riff (1):
      43d6ad3e cmake: Honor CMAKE_SYSROOT if set

Green Sky (2):
      1d1cc9ae cmake: Fix an error in the declaration of target include directories
      8cc22a8c cmake: Fix an error in the declaration of target include directories

John Bowler (3):
      ceed2a3c SECURITY: disable build of filter_neon.S on arm
      20f819c2 fix: Remove cHRM check to accomodate ACES AP1
      5a7e87fc fix: Prevent overflow in chromaticity calculations

Mikhail Khachayants (1):
      68ba4f1f oss-fuzz: Add fuzzing targets for simplified READ API

Ross Burton (1):
      9e538750 arm: Remove obsolete assembler implementation filter_neon.S

Change-Id: I5a75ec7e571b5b316397bd60c5d4e1038945b447
  • Loading branch information
@blowekamp
LIBPNG Upstream authored and blowekamp committed Nov 18, 2024
1 parent 55292c9 commit 2b57d16
Show file tree
Hide file tree
Showing 9 changed files with 144 additions and 369 deletions.
266 changes: 37 additions & 229 deletions arm/filter_neon.S
View file
Original file line number Diff line number Diff line change
@@ -1,253 +1,61 @@

/* filter_neon.S - NEON optimised filter functions
/* filter_neon.S - placeholder file
*
* Copyright (c) 2018 Cosmin Truta
* Copyright (c) 2014,2017 Glenn Randers-Pehrson
* Written by Mans Rullgard, 2011.
* Copyright (c) 2024 Cosmin Truta
*
* This code is released under the libpng license.
* For conditions of distribution and use, see the disclaimer
* and license in png.h
*/

/* IMPORTANT NOTE:
*
* Historically, the hand-coded assembler implementation of Neon optimizations
* in this module had not been in sync with the intrinsics-based implementation
* in filter_neon_intrinsics.c and palette_neon_intrinsics.c, at least since
* the introduction of riffled palette optimizations. Moreover, the assembler
* code used to work on 32-bit ARM only, and it caused problems, even if empty,
* on 64-bit ARM.
*
* All references to this module from our internal build scripts and projects
* have been removed.
*
* For the external projects that might still expect this module to be present,
* we leave this stub in place, for the remaining lifetime of libpng-1.6.x.
* Everything should continue to function normally, as long as there are no
* deliberate attempts to use the old hand-made assembler code. A build error
* will be raised otherwise.
*/

/* 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(__FreeBSD__)) && 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
#if defined(__clang__)
#define GNUC_VERSION 0 /* not gcc, although it might pretend to be */
#elif defined(__GNUC__)
#define GNUC_MAJOR (__GNUC__ + 0)
#define GNUC_MINOR (__GNUC_MINOR__ + 0)
#define GNUC_PATCHLEVEL (__GNUC_PATCHLEVEL__ + 0)
#define GNUC_VERSION (GNUC_MAJOR * 10000 + GNUC_MINOR * 100 + GNUC_PATCHLEVEL)
#else
# define ELF @
#define GNUC_VERSION 0 /* not gcc */
#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
#if (GNUC_VERSION > 0) && (GNUC_VERSION < 40300)
#error "PNG_ARM_NEON is not supported with gcc versions earlier than 4.3.0"
#elif GNUC_VERSION == 40504
#error "PNG_ARM_NEON is not supported with gcc version 4.5.4"
#else
#error "Please use 'arm/*_neon_intrinsics.c' for PNG_ARM_NEON support"
#endif

pop {r4,pc}
endfunc
#endif /* PNG_ARM_NEON_OPT > 0 */
#endif /* PNG_ARM_NEON_IMPLEMENTATION == 2 (assembler) */
#endif /* PNG_ARM_NEON_IMPLEMENTATION == 2 */
#endif /* READ */
Loading

0 comments on commit 2b57d16

Please sign in to comment.