From 2640bdb262d07bf910a6ff614834d73713bdf4a4 Mon Sep 17 00:00:00 2001 From: Jens Arnold Date: Thu, 18 Oct 2007 22:37:33 +0000 Subject: [PATCH] APE codec: Assembler optimised vector math routines for coldfire. -c2000 is now usable at 130% realtime (was 107%), -c3000 is near realtime (93%, was 64%). -c1000 doesn't change. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@15194 a1c6a512-1295-4272-9138-f99709370657 --- apps/codecs/demac/libdemac/decoder.c | 11 +- apps/codecs/demac/libdemac/filter.c | 13 ++ apps/codecs/demac/libdemac/vector_math16_cf.h | 180 ++++++++++++++++++ 3 files changed, 200 insertions(+), 4 deletions(-) create mode 100644 apps/codecs/demac/libdemac/vector_math16_cf.h diff --git a/apps/codecs/demac/libdemac/decoder.c b/apps/codecs/demac/libdemac/decoder.c index 4f4a583d00..326e893ec4 100644 --- a/apps/codecs/demac/libdemac/decoder.c +++ b/apps/codecs/demac/libdemac/decoder.c @@ -32,12 +32,15 @@ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110, USA /* Statically allocate the filter buffers */ -static int16_t filterbuf32[(32*3 + HISTORY_SIZE) * 2] IBSS_ATTR; /* 4480 bytes */ -static int16_t filterbuf256[(256*3 + HISTORY_SIZE) * 2] IBSS_ATTR; /* 5120 bytes */ +static int16_t filterbuf32[(32*3 + HISTORY_SIZE) * 2] /* 4480 bytes */ + IBSS_ATTR __attribute__((aligned(16))); +static int16_t filterbuf256[(256*3 + HISTORY_SIZE) * 2] /* 5120 bytes */ + IBSS_ATTR __attribute__((aligned(16))); /* This is only needed for "insane" files, and no Rockbox targets can hope to decode them in realtime anyway. */ -static int16_t filterbuf1280[(1280*3 + HISTORY_SIZE) * 2]; /* 17408 bytes */ +static int16_t filterbuf1280[(1280*3 + HISTORY_SIZE) * 2] /* 17408 bytes */ + __attribute__((aligned(16))); void init_frame_decoder(struct ape_ctx_t* ape_ctx, unsigned char* inbuffer, int* firstbyte, @@ -163,7 +166,7 @@ int decode_chunk(struct ape_ctx_t* ape_ctx, } /* Now apply the predictor decoding */ - predictor_decode_stereo(&ape_ctx->predictor,decoded0,decoded1,count); + predictor_decode_stereo(&ape_ctx->predictor,decoded0,decoded1,count); if (ape_ctx->bps == 8) { /* TODO: Handle 8-bit streams */ diff --git a/apps/codecs/demac/libdemac/filter.c b/apps/codecs/demac/libdemac/filter.c index 131c152590..ac12959241 100644 --- a/apps/codecs/demac/libdemac/filter.c +++ b/apps/codecs/demac/libdemac/filter.c @@ -25,10 +25,15 @@ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110, USA #include #include +#include "codecs.h" /* for Rockbox CPU definitions etc */ #include "demac.h" #include "filter.h" +#ifdef CPU_COLDFIRE +#include "vector_math16_cf.h" +#else #include "vector_math16.h" +#endif struct filter_t { int16_t* coeffs; /* ORDER entries */ @@ -84,6 +89,10 @@ static inline void do_apply_filter_3980(struct filter_t* f, int32_t* data, int c int res; int absres; +#ifdef PREPARE_SCALARPRODUCT + PREPARE_SCALARPRODUCT +#endif + while(count--) { res = FP_TO_INT(scalarproduct(f->delay - ORDER, f->coeffs)); @@ -135,6 +144,10 @@ static inline void do_apply_filter_3980(struct filter_t* f, int32_t* data, int c static inline void do_apply_filter_3970(struct filter_t* f, int32_t* data, int count) { int res; + +#ifdef PREPARE_SCALARPRODUCT + PREPARE_SCALARPRODUCT +#endif while(count--) { diff --git a/apps/codecs/demac/libdemac/vector_math16_cf.h b/apps/codecs/demac/libdemac/vector_math16_cf.h new file mode 100644 index 0000000000..85258c97a5 --- /dev/null +++ b/apps/codecs/demac/libdemac/vector_math16_cf.h @@ -0,0 +1,180 @@ +/* + +libdemac - A Monkey's Audio decoder + +$Id$ + +Copyright (C) Dave Chapman 2007 + +Coldfire vector math copyright (C) 2007 Jens Arnold + +This program is free software; you can redistribute it and/or modify +it under the terms of the GNU General Public License as published by +the Free Software Foundation; either version 2 of the License, or +(at your option) any later version. + +This program 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 General Public License for more details. + +You should have received a copy of the GNU General Public License +along with this program; if not, write to the Free Software +Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110, USA + +*/ + +static inline void vector_add(int16_t* v1, int16_t* v2) +{ +#define ADDHALFREGS(s1, sum) /* 's1' can be an A or D reg */ \ + "move.l " #s1 ", %%d4 \n" /* 'sum' must be a D reg */ \ + "add.l " #sum ", " #s1 "\n" /* 's1' and %%d4 are clobbered! */ \ + "clr.w %%d4 \n" \ + "add.l %%d4 , " #sum "\n" \ + "move.w " #s1 ", " #sum "\n" + + asm volatile ( +#if ORDER > 16 + "moveq.l %[cnt], %%d5 \n" + "1: \n" +#endif + "movem.l (%[v1]), %%d0-%%d3 \n" + "movem.l (%[v2]), %%a0-%%a3 \n" + + ADDHALFREGS(%%a0, %%d0) + ADDHALFREGS(%%a1, %%d1) + ADDHALFREGS(%%a2, %%d2) + ADDHALFREGS(%%a3, %%d3) + + "movem.l %%d0-%%d3, (%[v1]) \n" + "lea.l (16, %[v1]), %[v1] \n" + "movem.l (%[v1]), %%d0-%%d3 \n" + "lea.l (16, %[v2]), %[v2] \n" + "movem.l (%[v2]), %%a0-%%a3 \n" + + ADDHALFREGS(%%a0, %%d0) + ADDHALFREGS(%%a1, %%d1) + ADDHALFREGS(%%a2, %%d2) + ADDHALFREGS(%%a3, %%d3) + + "movem.l %%d0-%%d3, (%[v1]) \n" +#if ORDER > 16 + "lea.l (16, %[v1]), %[v1] \n" + "lea.l (16, %[v2]), %[v2] \n" + "subq.l #1, %%d5 \n" + "bne.s 1b \n" +#endif + : /* outputs */ + [v1]"+a"(v1), + [v2]"+a"(v2) + : /* inputs */ + [cnt]"n"(ORDER>>4) + : /* clobbers */ + "d0", "d1", "d2", "d3", "d4", "d5", + "a0", "a1", "a2", "a3", "memory" + ); +} + +static inline void vector_sub(int16_t* v1, int16_t* v2) +{ +#define SUBHALFREGS(min, sub, dif) /* 'min' can be an A or D reg */ \ + "move.l " #min ", " #dif "\n" /* 'sub' and 'dif' must be D regs */ \ + "sub.l " #sub ", " #min "\n" /* 'min' and 'sub' are clobbered! */ \ + "clr.w " #sub "\n" \ + "sub.l " #sub ", " #dif "\n" \ + "move.w " #min ", " #dif "\n" + + asm volatile ( +#if ORDER > 16 + "moveq.l %[cnt], %%d5 \n" + "1: \n" +#endif + "movem.l (%[v1]), %%a0-%%a3 \n" + "movem.l (%[v2]), %%d1-%%d4 \n" + + SUBHALFREGS(%%a0, %%d1, %%d0) + SUBHALFREGS(%%a1, %%d2, %%d1) + SUBHALFREGS(%%a2, %%d3, %%d2) + SUBHALFREGS(%%a3, %%d4, %%d3) + + "movem.l %%d0-%%d3, (%[v1]) \n" + "lea.l (16, %[v1]), %[v1] \n" + "movem.l (%[v1]), %%a0-%%a3 \n" + "lea.l (16, %[v2]), %[v2] \n" + "movem.l (%[v2]), %%d1-%%d4 \n" + + SUBHALFREGS(%%a0, %%d1, %%d0) + SUBHALFREGS(%%a1, %%d2, %%d1) + SUBHALFREGS(%%a2, %%d3, %%d2) + SUBHALFREGS(%%a3, %%d4, %%d3) + + "movem.l %%d0-%%d3, (%[v1]) \n" +#if ORDER > 16 + "lea.l (16, %[v1]), %[v1] \n" + "lea.l (16, %[v2]), %[v2] \n" + "subq.l #1, %%d5 \n" + "bne.s 1b \n" +#endif + : /* outputs */ + [v1]"+a"(v1), + [v2]"+a"(v2) + : /* inputs */ + [cnt]"n"(ORDER>>4) + : /* clobbers */ + "d0", "d1", "d2", "d3", "d4", "d5", + "a0", "a1", "a2", "a3", "memory" + ); +} + +#define PREPARE_SCALARPRODUCT coldfire_set_macsr(0); /* signed integer mode */ + +/* Needs EMAC in signed integer mode! */ +static inline int32_t scalarproduct(int16_t* v1, int16_t* v2) +{ + int res = 0; + +#define MACBLOCK4 \ + "mac.w %%d0u, %%d1u, (%[v1])+, %%d2, %%acc0\n" \ + "mac.w %%d0l, %%d1l, (%[v2])+, %%d3, %%acc0\n" \ + "mac.w %%d2u, %%d3u, (%[v1])+, %%d0, %%acc0\n" \ + "mac.w %%d2l, %%d3l, (%[v2])+, %%d1, %%acc0\n" + + asm volatile ( +#if ORDER > 32 + "moveq.l %[cnt], %[res] \n" +#endif + "move.l (%[v1])+, %%d0 \n" + "move.l (%[v2])+, %%d1 \n" + "1: \n" +#if ORDER > 16 + MACBLOCK4 + MACBLOCK4 + MACBLOCK4 + MACBLOCK4 +#endif + MACBLOCK4 + MACBLOCK4 + MACBLOCK4 + "mac.w %%d0u, %%d1u, (%[v1])+, %%d2, %%acc0\n" + "mac.w %%d0l, %%d1l, (%[v2])+, %%d3, %%acc0\n" +#if ORDER > 32 + "mac.w %%d2u, %%d3u, (%[v1])+, %%d0, %%acc0\n" + "mac.w %%d2l, %%d3l, (%[v2])+, %%d1, %%acc0\n" + "subq.l #1, %[res] \n" + "bne.w 1b \n" +#else + "mac.w %%d2u, %%d3u, %%acc0 \n" + "mac.w %%d2l, %%d3l, %%acc0 \n" +#endif + "movclr.l %%acc0, %[res] \n" + : /* outputs */ + [v1]"+a"(v1), + [v2]"+a"(v2), + [res]"=&d"(res) + : /* inputs */ + [cnt]"n"(ORDER>>5) + : /* clobbers */ + "d0", "d1", "d2", "d3" + ); + return res; +}