Mirror of roytam1's UXP fork just in case Moonchild and Tobin decide to go after him
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

259 lines
8.3 KiB

/* 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
*/
/* These are required because Mozilla's moz.build system doesn't pass
* -DDefined macros to the assembler.
*/
#define PNG_READ_SUPPORTED
#define MOZ_PNG_HAVE_ARM_NEON
/* 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 */