diff options
-rw-r--r-- | libavcodec/alacenc.c | 8 | ||||
-rw-r--r-- | libavcodec/flacenc.c | 10 | ||||
-rw-r--r-- | libavcodec/lpc.c | 57 | ||||
-rw-r--r-- | libavcodec/lpc.h | 31 | ||||
-rw-r--r-- | libavcodec/ra144enc.c | 13 | ||||
-rw-r--r-- | libavcodec/x86/lpc_mmx.c | 19 |
6 files changed, 99 insertions, 39 deletions
diff --git a/libavcodec/alacenc.c b/libavcodec/alacenc.c index 5706ae9f4d..d9ff2b8e0a 100644 --- a/libavcodec/alacenc.c +++ b/libavcodec/alacenc.c @@ -378,6 +378,7 @@ static void write_compressed_frame(AlacEncodeContext *s) static av_cold int alac_encode_init(AVCodecContext *avctx) { AlacEncodeContext *s = avctx->priv_data; + int ret; uint8_t *alac_extradata = av_mallocz(ALAC_EXTRADATA_SIZE+1); avctx->frame_size = DEFAULT_FRAME_SIZE; @@ -455,9 +456,10 @@ static av_cold int alac_encode_init(AVCodecContext *avctx) avctx->coded_frame->key_frame = 1; s->avctx = avctx; - ff_lpc_init(&s->lpc_ctx); + ret = ff_lpc_init(&s->lpc_ctx, avctx->frame_size, s->max_prediction_order, + AV_LPC_TYPE_LEVINSON); - return 0; + return ret; } static int alac_encode_frame(AVCodecContext *avctx, uint8_t *frame, @@ -513,6 +515,8 @@ verbatim: static av_cold int alac_encode_close(AVCodecContext *avctx) { + AlacEncodeContext *s = avctx->priv_data; + ff_lpc_end(&s->lpc_ctx); av_freep(&avctx->extradata); avctx->extradata_size = 0; av_freep(&avctx->coded_frame); diff --git a/libavcodec/flacenc.c b/libavcodec/flacenc.c index d6f0f87376..2bfe56e8eb 100644 --- a/libavcodec/flacenc.c +++ b/libavcodec/flacenc.c @@ -211,13 +211,11 @@ static av_cold int flac_encode_init(AVCodecContext *avctx) int freq = avctx->sample_rate; int channels = avctx->channels; FlacEncodeContext *s = avctx->priv_data; - int i, level; + int i, level, ret; uint8_t *streaminfo; s->avctx = avctx; - ff_lpc_init(&s->lpc_ctx); - if (avctx->sample_fmt != AV_SAMPLE_FMT_S16) return -1; @@ -438,9 +436,12 @@ static av_cold int flac_encode_init(AVCodecContext *avctx) if (!avctx->coded_frame) return AVERROR(ENOMEM); + ret = ff_lpc_init(&s->lpc_ctx, avctx->frame_size, + s->options.max_prediction_order, AV_LPC_TYPE_LEVINSON); + dprint_compression_options(s); - return 0; + return ret; } @@ -1316,6 +1317,7 @@ static av_cold int flac_encode_close(AVCodecContext *avctx) if (avctx->priv_data) { FlacEncodeContext *s = avctx->priv_data; av_freep(&s->md5ctx); + ff_lpc_end(&s->lpc_ctx); } av_freep(&avctx->extradata); avctx->extradata_size = 0; diff --git a/libavcodec/lpc.c b/libavcodec/lpc.c index 3a93c9f673..fd51491716 100644 --- a/libavcodec/lpc.c +++ b/libavcodec/lpc.c @@ -28,7 +28,7 @@ /** * Apply Welch window function to audio block */ -static void apply_welch_window(const int32_t *data, int len, double *w_data) +static void apply_welch_window_c(const int32_t *data, int len, double *w_data) { int i, n2; double w; @@ -54,24 +54,16 @@ static void apply_welch_window(const int32_t *data, int len, double *w_data) * Calculate autocorrelation data from audio samples * A Welch window function is applied before calculation. */ -static void lpc_compute_autocorr_c(const int32_t *data, int len, int lag, +static void lpc_compute_autocorr_c(const double *data, int len, int lag, double *autoc) { int i, j; - double tmp[len + lag + 1]; - double *data1= tmp + lag; - - apply_welch_window(data, len, data1); - - for(j=0; j<lag; j++) - data1[j-lag]= 0.0; - data1[len] = 0.0; for(j=0; j<lag; j+=2){ double sum0 = 1.0, sum1 = 1.0; for(i=j; i<len; i++){ - sum0 += data1[i] * data1[i-j]; - sum1 += data1[i] * data1[i-j-1]; + sum0 += data[i] * data[i-j]; + sum1 += data[i] * data[i-j-1]; } autoc[j ] = sum0; autoc[j+1] = sum1; @@ -80,8 +72,8 @@ static void lpc_compute_autocorr_c(const int32_t *data, int len, int lag, if(j==lag){ double sum = 1.0; for(i=j-1; i<len; i+=2){ - sum += data1[i ] * data1[i-j ] - + data1[i+1] * data1[i-j+1]; + sum += data[i ] * data[i-j ] + + data[i+1] * data[i-j+1]; } autoc[j] = sum; } @@ -177,8 +169,19 @@ int ff_lpc_calc_coefs(LPCContext *s, assert(max_order >= MIN_LPC_ORDER && max_order <= MAX_LPC_ORDER && lpc_type > AV_LPC_TYPE_FIXED); + /* reinit LPC context if parameters have changed */ + if (blocksize != s->blocksize || max_order != s->max_order || + lpc_type != s->lpc_type) { + ff_lpc_end(s); + ff_lpc_init(s, blocksize, max_order, lpc_type); + } + if (lpc_type == AV_LPC_TYPE_LEVINSON) { - s->lpc_compute_autocorr(samples, blocksize, max_order, autoc); + double *windowed_samples = s->windowed_samples + max_order; + + s->lpc_apply_welch_window(samples, blocksize, windowed_samples); + + s->lpc_compute_autocorr(windowed_samples, blocksize, max_order, autoc); compute_lpc_coefs(autoc, max_order, &lpc[0][0], MAX_LPC_ORDER, 0, 1); @@ -236,10 +239,32 @@ int ff_lpc_calc_coefs(LPCContext *s, return opt_order; } -av_cold void ff_lpc_init(LPCContext *s) +av_cold int ff_lpc_init(LPCContext *s, int blocksize, int max_order, + enum AVLPCType lpc_type) { + s->blocksize = blocksize; + s->max_order = max_order; + s->lpc_type = lpc_type; + + if (lpc_type == AV_LPC_TYPE_LEVINSON) { + s->windowed_samples = av_mallocz((blocksize + max_order + 2) * + sizeof(*s->windowed_samples)); + if (!s->windowed_samples) + return AVERROR(ENOMEM); + } else { + s->windowed_samples = NULL; + } + + s->lpc_apply_welch_window = apply_welch_window_c; s->lpc_compute_autocorr = lpc_compute_autocorr_c; if (HAVE_MMX) ff_lpc_init_x86(s); + + return 0; +} + +av_cold void ff_lpc_end(LPCContext *s) +{ + av_freep(&s->windowed_samples); } diff --git a/libavcodec/lpc.h b/libavcodec/lpc.h index a307793374..96b66df909 100644 --- a/libavcodec/lpc.h +++ b/libavcodec/lpc.h @@ -37,16 +37,35 @@ typedef struct LPCContext { + int blocksize; + int max_order; + enum AVLPCType lpc_type; + double *windowed_samples; + + /** + * Apply a Welch window to an array of input samples. + * The output samples have the same scale as the input, but are in double + * sample format. + * @param data input samples + * @param len number of input samples + * @param w_data output samples + */ + void (*lpc_apply_welch_window)(const int32_t *data, int len, + double *w_data); /** * Perform autocorrelation on input samples with delay of 0 to lag. * @param data input samples. - * no alignment needed. + * constraints: no alignment needed, but must have have at + * least lag*sizeof(double) valid bytes preceeding it, and + * size must be at least (len+1)*sizeof(double) if data is + * 16-byte aligned or (len+2)*sizeof(double) if data is + * unaligned. * @param len number of input samples to process * @param lag maximum delay to calculate * @param autoc output autocorrelation coefficients. * constraints: array size must be at least lag+1. */ - void (*lpc_compute_autocorr)(const int32_t *data, int len, int lag, + void (*lpc_compute_autocorr)(const double *data, int len, int lag, double *autoc); } LPCContext; @@ -64,9 +83,15 @@ int ff_lpc_calc_coefs(LPCContext *s, /** * Initialize LPCContext. */ -void ff_lpc_init(LPCContext *s); +int ff_lpc_init(LPCContext *s, int blocksize, int max_order, + enum AVLPCType lpc_type); void ff_lpc_init_x86(LPCContext *s); +/** + * Uninitialize LPCContext. + */ +void ff_lpc_end(LPCContext *s); + #ifdef LPC_USE_DOUBLE #define LPC_TYPE double #else diff --git a/libavcodec/ra144enc.c b/libavcodec/ra144enc.c index 3f8694eb8f..2c0a6b1fbd 100644 --- a/libavcodec/ra144enc.c +++ b/libavcodec/ra144enc.c @@ -36,6 +36,7 @@ static av_cold int ra144_encode_init(AVCodecContext * avctx) { RA144Context *ractx; + int ret; if (avctx->sample_fmt != AV_SAMPLE_FMT_S16) { av_log(avctx, AV_LOG_ERROR, "invalid sample format\n"); @@ -52,7 +53,16 @@ static av_cold int ra144_encode_init(AVCodecContext * avctx) ractx->lpc_coef[0] = ractx->lpc_tables[0]; ractx->lpc_coef[1] = ractx->lpc_tables[1]; ractx->avctx = avctx; - ff_lpc_init(&ractx->lpc_ctx); + ret = ff_lpc_init(&ractx->lpc_ctx, avctx->frame_size, LPC_ORDER, + AV_LPC_TYPE_LEVINSON); + return ret; +} + + +static av_cold int ra144_encode_close(AVCodecContext *avctx) +{ + RA144Context *ractx = avctx->priv_data; + ff_lpc_end(&ractx->lpc_ctx); return 0; } @@ -506,5 +516,6 @@ AVCodec ra_144_encoder = sizeof(RA144Context), ra144_encode_init, ra144_encode_frame, + ra144_encode_close, .long_name = NULL_IF_CONFIG_SMALL("RealAudio 1.0 (14.4K) encoder"), }; diff --git a/libavcodec/x86/lpc_mmx.c b/libavcodec/x86/lpc_mmx.c index 19aad9860f..3a0a1f0f6f 100644 --- a/libavcodec/x86/lpc_mmx.c +++ b/libavcodec/x86/lpc_mmx.c @@ -69,21 +69,13 @@ static void apply_welch_window_sse2(const int32_t *data, int len, double *w_data #undef WELCH } -static void lpc_compute_autocorr_sse2(const int32_t *data, int len, int lag, +static void lpc_compute_autocorr_sse2(const double *data, int len, int lag, double *autoc) { - double tmp[len + lag + 2]; - double *data1 = tmp + lag; int j; - if((x86_reg)data1 & 15) - data1++; - - apply_welch_window_sse2(data, len, data1); - - for(j=0; j<lag; j++) - data1[j-lag]= 0.0; - data1[len] = 0.0; + if((x86_reg)data & 15) + data++; for(j=0; j<lag; j+=2){ x86_reg i = -len*sizeof(double); @@ -114,7 +106,7 @@ static void lpc_compute_autocorr_sse2(const int32_t *data, int len, int lag, "movsd %%xmm1, 8(%1) \n\t" "movsd %%xmm2, 16(%1) \n\t" :"+&r"(i) - :"r"(autoc+j), "r"(data1+len), "r"(data1+len-j) + :"r"(autoc+j), "r"(data+len), "r"(data+len-j) :"memory" ); } else { @@ -137,7 +129,7 @@ static void lpc_compute_autocorr_sse2(const int32_t *data, int len, int lag, "movsd %%xmm0, %1 \n\t" "movsd %%xmm1, %2 \n\t" :"+&r"(i), "=m"(autoc[j]), "=m"(autoc[j+1]) - :"r"(data1+len), "r"(data1+len-j) + :"r"(data+len), "r"(data+len-j) ); } } @@ -148,6 +140,7 @@ av_cold void ff_lpc_init_x86(LPCContext *c) int mm_flags = av_get_cpu_flags(); if (mm_flags & (AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW)) { + c->lpc_apply_welch_window = apply_welch_window_sse2; c->lpc_compute_autocorr = lpc_compute_autocorr_sse2; } } |