From a06f651c3e5f52820e5c7233fa06ef21f793c812 Mon Sep 17 00:00:00 2001 From: Michael Giacomelli Date: Sat, 28 Jun 2008 18:56:19 +0000 Subject: Put the WMA windowing and output buffer into IRAM on targets with > 96KB of it. Improves 192k WMA decoding speed by about 2MHz on PP5024, and maybe saves some more when applying DSP. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@17853 a1c6a512-1295-4272-9138-f99709370657 --- apps/codecs/libwma/wmadec.h | 14 +++++++++++++- apps/codecs/libwma/wmadeci.c | 17 ++++++++--------- 2 files changed, 21 insertions(+), 10 deletions(-) (limited to 'apps/codecs/libwma') diff --git a/apps/codecs/libwma/wmadec.h b/apps/codecs/libwma/wmadec.h index f7434ed759..30ba35b533 100644 --- a/apps/codecs/libwma/wmadec.h +++ b/apps/codecs/libwma/wmadec.h @@ -54,6 +54,17 @@ #define LSP_POW_BITS 7 +/*define IRAM for targets with 48k/80k IRAM split*/ +#ifndef IBSS_ATTR_WMA_LARGE_IRAM +#if (CONFIG_CPU == PP5022) || (CONFIG_CPU == PP5024) || (CONFIG_CPU == MCF5250) +/* PP5022/24 and MCF5250 have 128KB of IRAM, with 80KB allocated for codecs */ +#define IBSS_ATTR_WMA_LARGE_IRAM IBSS_ATTR +#else +/* other PP's and MCF5249 have 96KB of IRAM */ +#define IBSS_ATTR_WMA_LARGE_IRAM +#endif +#endif + typedef struct WMADecodeContext { GetBitContext gb; @@ -110,7 +121,8 @@ typedef struct WMADecodeContext MDCTContext mdct_ctx[BLOCK_NB_SIZES]; fixed32 *windows[BLOCK_NB_SIZES]; /* output buffer for one frame and the last for IMDCT windowing */ - fixed32 frame_out[MAX_CHANNELS][BLOCK_MAX_SIZE * 2]; + fixed32 (*frame_out)[MAX_CHANNELS][BLOCK_MAX_SIZE*2]; + /* last frame info */ uint8_t last_superframe[MAX_CODED_SUPERFRAME_SIZE + 4]; /* padding added */ int last_bitoffset; diff --git a/apps/codecs/libwma/wmadeci.c b/apps/codecs/libwma/wmadeci.c index 20c22187b2..23916f6dc8 100644 --- a/apps/codecs/libwma/wmadeci.c +++ b/apps/codecs/libwma/wmadeci.c @@ -52,6 +52,8 @@ CoefVLCTable; static void wma_lsp_to_curve_init(WMADecodeContext *s, int frame_len); fixed32 coefsarray[MAX_CHANNELS][BLOCK_MAX_SIZE] IBSS_ATTR; +/*decode and window into IRAM on targets with at least 80KB of codec IRAM*/ +fixed32 frame_out_buf[MAX_CHANNELS][BLOCK_MAX_SIZE * 2] IBSS_ATTR_WMA_LARGE_IRAM; //static variables that replace malloced stuff fixed32 stat0[2048], stat1[1024], stat2[512], stat3[256], stat4[128]; //these are the MDCT reconstruction windows @@ -329,6 +331,7 @@ int wma_decode_init(WMADecodeContext* s, asf_waveformatex_t *wfx) s->block_align = wfx->blockalign; s->coefs = &coefsarray; + s->frame_out = &frame_out_buf; if (wfx->codec_id == ASF_CODEC_ID_WMAV1) { s->version = 1; @@ -1369,7 +1372,6 @@ static int wma_decode_block(WMADecodeContext *s) if (s->channel_coded[ch]) { static fixed32 output[BLOCK_MAX_SIZE * 2] IBSS_ATTR; - int n4, index, n; n = s->block_len; @@ -1382,8 +1384,7 @@ static int wma_decode_block(WMADecodeContext *s) /* add in the frame */ index = (s->frame_len / 2) + s->block_pos - n4; - - wma_window(s, output, &s->frame_out[ch][index]); + wma_window(s, output, &((*s->frame_out)[ch][index])); @@ -1391,7 +1392,7 @@ static int wma_decode_block(WMADecodeContext *s) channel if it is not coded */ if (s->ms_stereo && !s->channel_coded[1]) { - wma_window(s, output, &s->frame_out[1][index]); + wma_window(s, output, &((*s->frame_out)[1][index])); } } } @@ -1415,7 +1416,6 @@ static int wma_decode_frame(WMADecodeContext *s, int32_t *samples) int ret, i, n, ch, incr; int32_t *ptr; fixed32 *iptr; - // rb->splash(HZ, "in wma_decode_frame"); /* read each block */ s->block_num = 0; @@ -1443,17 +1443,16 @@ static int wma_decode_frame(WMADecodeContext *s, int32_t *samples) for(ch = 0; ch < s->nb_channels; ++ch) { ptr = samples + ch; - iptr = s->frame_out[ch]; + iptr = &((*s->frame_out)[ch][0]); for (i=0;iframe_out[ch][0], &s->frame_out[ch][s->frame_len], - s->frame_len * sizeof(fixed32)); + memmove(&((*s->frame_out)[ch][0]), &((*s->frame_out)[ch][s->frame_len]), + s->frame_len * sizeof(fixed32)); } return 0; -- cgit v1.2.3