openSUSE Build Service

Overview Repositories Revisions Requests Users Attributes Meta

File 506.patch of Package libpng16 (Revision 47c78e9585219206ed8bd7e837d13acd)

Currently displaying revision 47c78e9585219206ed8bd7e837d13acd , Show latest

From 20917321cbab5cf82b1855af1d1b02cdaa281d85 Mon Sep 17 00:00:00 2001
From: Ross Burton <ross.burton@arm.com>
Date: Thu, 14 Dec 2023 20:25:46 +0000
Subject: [PATCH] arm/filter_neon.S: remove obsolete assembler

This file contains hand-coded assembler implementations of the filter
functions for 32-bit Arm platforms.  These are only used when the
compiler doesn't support neon intrinsics (added to GCC 4.3 in 2008) or
is exactly GCC 4.5.4 (released 2012), both of which are sufficiently
unlikely to be true that it's fair to say the assembler is no longer
used.

Instrinics can be optimised by the compiler, and also has the benefit
that the compiler can add branch protection instructions (such as
PAC/BTI on aarch64) to improve resilience against malicious input.

This commit deletes filter_neon.S and removes the now obsolete
preprocessor logic in pngpriv.h.

Closes #505

Upstream-Status: Submitted [https://github.com/glennrp/libpng/pull/506]
Signed-off-by: Ross Burton <ross.burton@arm.com>
---
 CMakeLists.txt    |   3 -
 Makefile.am       |   2 +-
 arm/filter_neon.S | 253 ----------------------------------------------
 pngpriv.h         |  41 --------
 4 files changed, 1 insertion(+), 298 deletions(-)
 delete mode 100644 arm/filter_neon.S

diff --git a/Makefile.am b/Makefile.am
index ba51d91b4..debddc9af 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -108,7 +108,7 @@ libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES = png.c pngerror.c\
 
 if PNG_ARM_NEON
 libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES += arm/arm_init.c\
-	arm/filter_neon.S arm/filter_neon_intrinsics.c \
+	arm/filter_neon_intrinsics.c \
 	arm/palette_neon_intrinsics.c
 endif
 
diff --git a/arm/filter_neon.S b/arm/filter_neon.S
deleted file mode 100644
index 2308aad13..000000000
--- a/arm/filter_neon.S
+++ /dev/null
@@ -1,253 +0,0 @@
-
-/* filter_neon.S - NEON optimised filter functions
- *
- * Copyright (c) 2018 Cosmin Truta
- * Copyright (c) 2014,2017 Glenn Randers-Pehrson
- * Written by Mans Rullgard, 2011.
- *
- * 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(__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
-#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/pngpriv.h b/pngpriv.h
index 7c19373f0..51d26b978 100644
--- a/pngpriv.h
+++ b/pngpriv.h
@@ -140,47 +140,6 @@
     * callbacks to do this.
     */
 #  define PNG_FILTER_OPTIMIZATIONS png_init_filter_functions_neon
-
-   /* By default the 'intrinsics' code in arm/filter_neon_intrinsics.c is used
-    * if possible - if __ARM_NEON__ is set and the compiler version is not known
-    * to be broken.  This is controlled by PNG_ARM_NEON_IMPLEMENTATION which can
-    * be:
-    *
-    *    1  The intrinsics code (the default with __ARM_NEON__)
-    *    2  The hand coded assembler (the default without __ARM_NEON__)
-    *
-    * It is possible to set PNG_ARM_NEON_IMPLEMENTATION in CPPFLAGS, however
-    * this is *NOT* supported and may cease to work even after a minor revision
-    * to libpng.  It *is* valid to do this for testing purposes, e.g. speed
-    * testing or a new compiler, but the results should be communicated to the
-    * libpng implementation list for incorporation in the next minor release.
-    */
-#  ifndef PNG_ARM_NEON_IMPLEMENTATION
-#     if defined(__ARM_NEON__) || defined(__ARM_NEON)
-#        if defined(__clang__)
-            /* At present it is unknown by the libpng developers which versions
-             * of clang support the intrinsics, however some or perhaps all
-             * versions do not work with the assembler so this may be
-             * irrelevant, so just use the default (do nothing here.)
-             */
-#        elif defined(__GNUC__)
-            /* GCC 4.5.4 NEON support is known to be broken.  4.6.3 is known to
-             * work, so if this *is* GCC, or G++, look for a version >4.5
-             */
-#           if __GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 6)
-#              define PNG_ARM_NEON_IMPLEMENTATION 2
-#           endif /* no GNUC support */
-#        endif /* __GNUC__ */
-#     else /* !defined __ARM_NEON__ */
-         /* The 'intrinsics' code simply won't compile without this -mfpu=neon:
-          */
-#        if !defined(__aarch64__) && !defined(_M_ARM64)
-            /* The assembler code currently does not work on ARM64 */
-#          define PNG_ARM_NEON_IMPLEMENTATION 2
-#        endif /* __aarch64__ */
-#     endif /* __ARM_NEON__ */
-#  endif /* !PNG_ARM_NEON_IMPLEMENTATION */
-
 #  ifndef PNG_ARM_NEON_IMPLEMENTATION
       /* Use the intrinsics code by default. */
 #     define PNG_ARM_NEON_IMPLEMENTATION 1
-- 
2.34.1

Places

File 506.patch of Package libpng16 (Revision 47c78e9585219206ed8bd7e837d13acd)

Places