# HG changeset patch # User mru # Date 1250468153 0 # Node ID c2556eaf12caf762713bea35734f51cd668202cf # Parent 7ee08d73ac3aaecf6a4e6febec17b08ca71535c2 AVR32: optimisations for intreadwrite.h diff -r 7ee08d73ac3a -r c2556eaf12ca avr32/intreadwrite.h --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/avr32/intreadwrite.h Mon Aug 17 00:15:53 2009 +0000 @@ -0,0 +1,182 @@ +/* + * Copyright (c) 2009 Mans Rullgard + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifndef AVUTIL_AVR32_INTREADWRITE_H +#define AVUTIL_AVR32_INTREADWRITE_H + +#include +#include "config.h" +#include "libavutil/bswap.h" + +/* + * AVR32 does not support unaligned memory accesses, except for the AP + * series which suppports unaligned 32-bit loads and stores. 16-bit + * and 64-bit accesses must be aligned to 16 and 32 bits, respectively. + * This means we cannot use the byte-swapping load/store instructions + * here. + * + * For 16-bit, 24-bit, and (on UC series) 32-bit loads, we instead use + * the LDINS.B instruction, which gcc fails to utilise with the + * generic code. GCC also fails to use plain LD.W and ST.W even for + * AP processors, so we override the generic code. The 64-bit + * versions are improved by using our optimised 32-bit functions. + */ + +#define AV_RL16 AV_RL16 +static inline uint16_t AV_RL16(const void *p) +{ + uint16_t v; + __asm__ ("ld.ub %0, %1 \n\t" + "ldins.b %0:l, %2 \n\t" + : "=&r"(v) + : "m"(*(const uint8_t*)p), "RKs12"(*((const uint8_t*)p+1))); + return v; +} + +#define AV_RB16 AV_RB16 +static inline uint16_t AV_RB16(const void *p) +{ + uint16_t v; + __asm__ ("ld.ub %0, %2 \n\t" + "ldins.b %0:l, %1 \n\t" + : "=&r"(v) + : "RKs12"(*(const uint8_t*)p), "m"(*((const uint8_t*)p+1))); + return v; +} + +#define AV_RB24 AV_RB24 +static inline uint32_t AV_RB24(const void *p) +{ + uint32_t v; + __asm__ ("ld.ub %0, %3 \n\t" + "ldins.b %0:l, %2 \n\t" + "ldins.b %0:u, %1 \n\t" + : "=&r"(v) + : "RKs12"(* (const uint8_t*)p), + "RKs12"(*((const uint8_t*)p+1)), + "m" (*((const uint8_t*)p+2))); + return v; +} + +#define AV_RL24 AV_RL24 +static inline uint32_t AV_RL24(const void *p) +{ + uint32_t v; + __asm__ ("ld.ub %0, %1 \n\t" + "ldins.b %0:l, %2 \n\t" + "ldins.b %0:u, %3 \n\t" + : "=&r"(v) + : "m" (* (const uint8_t*)p), + "RKs12"(*((const uint8_t*)p+1)), + "RKs12"(*((const uint8_t*)p+2))); + return v; +} + +#if ARCH_AVR32_AP + +#define AV_RB32 AV_RB32 +static inline uint32_t AV_RB32(const void *p) +{ + uint32_t v; + __asm__ ("ld.w %0, %1" : "=r"(v) : "m"(*(const uint32_t*)p)); + return v; +} + +#define AV_WB32 AV_WB32 +static inline void AV_WB32(void *p, uint32_t v) +{ + __asm__ ("st.w %0, %1" : "=m"(*(uint32_t*)p) : "r"(v)); +} + +/* These two would be defined by generic code, but we need them sooner. */ +#define AV_RL32(p) bswap_32(AV_RB32(p)) +#define AV_WL32(p, v) AV_WB32(p, bswap_32(v)) + +#define AV_WB64 AV_WB64 +static inline void AV_WB64(void *p, uint64_t v) +{ + union { uint64_t v; uint32_t hl[2]; } vv = { v }; + AV_WB32(p, vv.hl[0]); + AV_WB32((uint32_t*)p+1, vv.hl[1]); +} + +#define AV_WL64 AV_WL64 +static inline void AV_WL64(void *p, uint64_t v) +{ + union { uint64_t v; uint32_t hl[2]; } vv = { v }; + AV_WL32(p, vv.hl[1]); + AV_WL32((uint32_t*)p+1, vv.hl[0]); +} + +#else /* ARCH_AVR32_AP */ + +#define AV_RB32 AV_RB32 +static inline uint32_t AV_RB32(const void *p) +{ + uint32_t v; + __asm__ ("ld.ub %0, %4 \n\t" + "ldins.b %0:l, %3 \n\t" + "ldins.b %0:u, %2 \n\t" + "ldins.b %0:t, %1 \n\t" + : "=&r"(v) + : "RKs12"(* (const uint8_t*)p), + "RKs12"(*((const uint8_t*)p+1)), + "RKs12"(*((const uint8_t*)p+2)), + "m" (*((const uint8_t*)p+3))); + return v; +} + +#define AV_RL32 AV_RL32 +static inline uint32_t AV_RL32(const void *p) +{ + uint32_t v; + __asm__ ("ld.ub %0, %1 \n\t" + "ldins.b %0:l, %2 \n\t" + "ldins.b %0:u, %3 \n\t" + "ldins.b %0:t, %4 \n\t" + : "=&r"(v) + : "m" (* (const uint8_t*)p), + "RKs12"(*((const uint8_t*)p+1)), + "RKs12"(*((const uint8_t*)p+2)), + "RKs12"(*((const uint8_t*)p+3))); + return v; +} + +#endif /* ARCH_AVR32_AP */ + +#define AV_RB64 AV_RB64 +static inline uint64_t AV_RB64(const void *p) +{ + union { uint64_t v; uint32_t hl[2]; } v; + v.hl[0] = AV_RB32(p); + v.hl[1] = AV_RB32((const uint32_t*)p+1); + return v.v; +} + +#define AV_RL64 AV_RL64 +static inline uint64_t AV_RL64(const void *p) +{ + union { uint64_t v; uint32_t hl[2]; } v; + v.hl[1] = AV_RL32(p); + v.hl[0] = AV_RL32((const uint32_t*)p+1); + return v.v; +} + +#endif /* AVUTIL_AVR32_INTREADWRITE_H */ diff -r 7ee08d73ac3a -r c2556eaf12ca intreadwrite.h --- a/intreadwrite.h Mon Aug 17 00:15:51 2009 +0000 +++ b/intreadwrite.h Mon Aug 17 00:15:53 2009 +0000 @@ -31,6 +31,8 @@ #if ARCH_ARM # include "arm/intreadwrite.h" +#elif ARCH_AVR32 +# include "avr32/intreadwrite.h" #elif ARCH_MIPS # include "mips/intreadwrite.h" #elif ARCH_PPC