From 547b6a570dbad844e79b4ba5eb934f043bab6318 Mon Sep 17 00:00:00 2001 From: Solomon Peachy Date: Wed, 8 May 2024 10:36:38 -0400 Subject: codecs: Update libspeex from 1.2beta3 to 1.2rc1 This is a relatively minor bump, but it's the first step towards bringing this current. Change-Id: Iab6c9b0c77f0ba705280434ea74b513364719499 --- lib/rbcodec/codecs/libspeex/resample.c | 735 ++++++++++++++++----------------- 1 file changed, 363 insertions(+), 372 deletions(-) (limited to 'lib/rbcodec/codecs/libspeex/resample.c') diff --git a/lib/rbcodec/codecs/libspeex/resample.c b/lib/rbcodec/codecs/libspeex/resample.c index 65dfef28a3..5f5b9c6b4d 100644 --- a/lib/rbcodec/codecs/libspeex/resample.c +++ b/lib/rbcodec/codecs/libspeex/resample.c @@ -1,5 +1,6 @@ -/* Copyright (C) 2007 Jean-Marc Valin - +/* Copyright (C) 2007-2008 Jean-Marc Valin + Copyright (C) 2008 Thorvald Natvig + File: resample.c Arbitrary resampling code @@ -37,27 +38,27 @@ - Low memory requirement - Good *perceptual* quality (and not best SNR) - Warning: This resampler is relatively new. Although I think I got rid of + Warning: This resampler is relatively new. Although I think I got rid of all the major bugs and I don't expect the API to change anymore, there may be something I've missed. So use with caution. This algorithm is based on this original resampling algorithm: Smith, Julius O. Digital Audio Resampling Home Page - Center for Computer Research in Music and Acoustics (CCRMA), + Center for Computer Research in Music and Acoustics (CCRMA), Stanford University, 2007. Web published at http://www-ccrma.stanford.edu/~jos/resample/. - There is one main difference, though. This resampler uses cubic + There is one main difference, though. This resampler uses cubic interpolation instead of linear interpolation in the above paper. This makes the table much smaller and makes it possible to compute that table - on a per-stream basis. In turn, being able to tweak the table for each - stream makes it possible to both reduce complexity on simple ratios - (e.g. 2/3), and get rid of the rounding operations in the inner loop. + on a per-stream basis. In turn, being able to tweak the table for each + stream makes it possible to both reduce complexity on simple ratios + (e.g. 2/3), and get rid of the rounding operations in the inner loop. The latter both reduces CPU time and makes the algorithm more SIMD-friendly. */ #ifdef HAVE_CONFIG_H -#include "config-speex.h" +#include "config.h" #endif #ifdef OUTSIDE_SPEEX @@ -68,12 +69,13 @@ static void speex_free (void *ptr) {free(ptr);} #include "speex_resampler.h" #include "arch.h" #else /* OUTSIDE_SPEEX */ - + #include "speex/speex_resampler.h" #include "arch.h" #include "os_support.h" #endif /* OUTSIDE_SPEEX */ +#include "stack_alloc.h" #include #ifndef M_PI @@ -81,14 +83,10 @@ static void speex_free (void *ptr) {free(ptr);} #endif #ifdef FIXED_POINT -#define WORD2INT(x) ((x) < -32767 ? -32768 : ((x) > 32766 ? 32767 : (x))) +#define WORD2INT(x) ((x) < -32767 ? -32768 : ((x) > 32766 ? 32767 : (x))) #else -#define WORD2INT(x) ((x) < -32767.5f ? -32768 : ((x) > 32766.5f ? 32767 : floor(.5+(x)))) +#define WORD2INT(x) ((x) < -32767.5f ? -32768 : ((x) > 32766.5f ? 32767 : floor(.5+(x)))) #endif - -/*#define float double*/ -#define FILTER_SIZE 64 -#define OVERSAMPLE 8 #define IMAX(a,b) ((a) > (b) ? (a) : (b)) #define IMIN(a,b) ((a) < (b) ? (a) : (b)) @@ -97,6 +95,17 @@ static void speex_free (void *ptr) {free(ptr);} #define NULL 0 #endif +#ifdef _USE_SSE +#include "resample_sse.h" +#endif + +/* Numer of elements to allocate on the stack */ +#ifdef VAR_ARRAYS +#define FIXED_STACK_ALLOC 8192 +#else +#define FIXED_STACK_ALLOC 1024 +#endif + typedef int (*resampler_basic_func)(SpeexResamplerState *, spx_uint32_t , const spx_word16_t *, spx_uint32_t *, spx_word16_t *, spx_uint32_t *); struct SpeexResamplerState_ { @@ -104,28 +113,29 @@ struct SpeexResamplerState_ { spx_uint32_t out_rate; spx_uint32_t num_rate; spx_uint32_t den_rate; - + int quality; spx_uint32_t nb_channels; spx_uint32_t filt_len; spx_uint32_t mem_alloc_size; + spx_uint32_t buffer_size; int int_advance; int frac_advance; float cutoff; spx_uint32_t oversample; int initialised; int started; - + /* These are per-channel */ spx_int32_t *last_sample; spx_uint32_t *samp_frac_num; spx_uint32_t *magic_samples; - + spx_word16_t *mem; spx_word16_t *sinc_table; spx_uint32_t sinc_table_length; resampler_basic_func resampler_ptr; - + int in_stride; int out_stride; } ; @@ -167,7 +177,7 @@ static double kaiser8_table[36] = { 0.32108304, 0.27619388, 0.23465776, 0.19672670, 0.16255380, 0.13219758, 0.10562887, 0.08273982, 0.06335451, 0.04724088, 0.03412321, 0.02369490, 0.01563093, 0.00959968, 0.00527363, 0.00233883, 0.00050000, 0.00000000}; - + static double kaiser6_table[36] = { 0.99733006, 1.00000000, 0.99733006, 0.98935595, 0.97618418, 0.95799003, 0.93501423, 0.90755855, 0.87598009, 0.84068475, 0.80211977, 0.76076565, @@ -180,7 +190,7 @@ struct FuncDef { double *table; int oversample; }; - + static struct FuncDef _KAISER12 = {kaiser12_table, 64}; #define KAISER12 (&_KAISER12) /*static struct FuncDef _KAISER12 = {kaiser12_table, 32}; @@ -202,7 +212,7 @@ struct QualityMapping { /* This table maps conversion quality to internal parameters. There are two - reasons that explain why the up-sampling bandwidth is larger than the + reasons that explain why the up-sampling bandwidth is larger than the down-sampling bandwidth: 1) When up-sampling, we can assume that the spectrum is already attenuated close to the Nyquist rate (from an A/D or a previous resampling filter) @@ -228,7 +238,7 @@ static double compute_func(float x, struct FuncDef *func) { float y, frac; double interp[4]; - int ind; + int ind; y = x*func->oversample; ind = (int)floor(y); frac = (y-ind); @@ -239,7 +249,7 @@ static double compute_func(float x, struct FuncDef *func) interp[0] = -0.3333333333*frac + 0.5*(frac*frac) - 0.1666666667*(frac*frac*frac); /* Just to make sure we don't have rounding problems */ interp[1] = 1.f-interp[3]-interp[2]-interp[0]; - + /*sum = frac*accum[1] + (1-frac)*accum[2];*/ return interp[0]*func->table[ind] + interp[1]*func->table[ind+1] + interp[2]*func->table[ind+2] + interp[3]*func->table[ind+3]; } @@ -317,47 +327,47 @@ static void cubic_coef(spx_word16_t frac, spx_word16_t interp[4]) static int resampler_basic_direct_single(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_word16_t *in, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len) { - int N = st->filt_len; + const int N = st->filt_len; int out_sample = 0; - spx_word16_t *mem; int last_sample = st->last_sample[channel_index]; spx_uint32_t samp_frac_num = st->samp_frac_num[channel_index]; - mem = st->mem + channel_index * st->mem_alloc_size; + const spx_word16_t *sinc_table = st->sinc_table; + const int out_stride = st->out_stride; + const int int_advance = st->int_advance; + const int frac_advance = st->frac_advance; + const spx_uint32_t den_rate = st->den_rate; + spx_word32_t sum; + int j; + while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len)) { - int j; - spx_word32_t sum=0; - - /* We already have all the filter coefficients pre-computed in the table */ - const spx_word16_t *ptr; - /* Do the memory part */ - for (j=0;last_sample-N+1+j < 0;j++) - { - sum += MULT16_16(mem[last_sample+j],st->sinc_table[samp_frac_num*st->filt_len+j]); - } - - /* Do the new part */ - if (in != NULL) - { - ptr = in+st->in_stride*(last_sample-N+1+j); - for (;jsinc_table[samp_frac_num*st->filt_len+j]); - ptr += st->in_stride; - } + const spx_word16_t *sinc = & sinc_table[samp_frac_num*N]; + const spx_word16_t *iptr = & in[last_sample]; + +#ifndef OVERRIDE_INNER_PRODUCT_SINGLE + float accum[4] = {0,0,0,0}; + + for(j=0;jout_stride; - out_sample++; - last_sample += st->int_advance; - samp_frac_num += st->frac_advance; - if (samp_frac_num >= st->den_rate) + sum = accum[0] + accum[1] + accum[2] + accum[3]; +#else + sum = inner_product_single(sinc, iptr, N); +#endif + + out[out_stride * out_sample++] = PSHR32(sum, 15); + last_sample += int_advance; + samp_frac_num += frac_advance; + if (samp_frac_num >= den_rate) { - samp_frac_num -= st->den_rate; + samp_frac_num -= den_rate; last_sample++; } } + st->last_sample[channel_index] = last_sample; st->samp_frac_num[channel_index] = samp_frac_num; return out_sample; @@ -368,47 +378,47 @@ static int resampler_basic_direct_single(SpeexResamplerState *st, spx_uint32_t c /* This is the same as the previous function, except with a double-precision accumulator */ static int resampler_basic_direct_double(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_word16_t *in, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len) { - int N = st->filt_len; + const int N = st->filt_len; int out_sample = 0; - spx_word16_t *mem; int last_sample = st->last_sample[channel_index]; spx_uint32_t samp_frac_num = st->samp_frac_num[channel_index]; - mem = st->mem + channel_index * st->mem_alloc_size; + const spx_word16_t *sinc_table = st->sinc_table; + const int out_stride = st->out_stride; + const int int_advance = st->int_advance; + const int frac_advance = st->frac_advance; + const spx_uint32_t den_rate = st->den_rate; + double sum; + int j; + while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len)) { - int j; - double sum=0; - - /* We already have all the filter coefficients pre-computed in the table */ - const spx_word16_t *ptr; - /* Do the memory part */ - for (j=0;last_sample-N+1+j < 0;j++) - { - sum += MULT16_16(mem[last_sample+j],(double)st->sinc_table[samp_frac_num*st->filt_len+j]); - } - - /* Do the new part */ - if (in != NULL) - { - ptr = in+st->in_stride*(last_sample-N+1+j); - for (;jsinc_table[samp_frac_num*st->filt_len+j]); - ptr += st->in_stride; - } + const spx_word16_t *sinc = & sinc_table[samp_frac_num*N]; + const spx_word16_t *iptr = & in[last_sample]; + +#ifndef OVERRIDE_INNER_PRODUCT_DOUBLE + double accum[4] = {0,0,0,0}; + + for(j=0;jout_stride; - out_sample++; - last_sample += st->int_advance; - samp_frac_num += st->frac_advance; - if (samp_frac_num >= st->den_rate) + sum = accum[0] + accum[1] + accum[2] + accum[3]; +#else + sum = inner_product_double(sinc, iptr, N); +#endif + + out[out_stride * out_sample++] = PSHR32(sum, 15); + last_sample += int_advance; + samp_frac_num += frac_advance; + if (samp_frac_num >= den_rate) { - samp_frac_num -= st->den_rate; + samp_frac_num -= den_rate; last_sample++; } } + st->last_sample[channel_index] = last_sample; st->samp_frac_num[channel_index] = samp_frac_num; return out_sample; @@ -417,69 +427,58 @@ static int resampler_basic_direct_double(SpeexResamplerState *st, spx_uint32_t c static int resampler_basic_interpolate_single(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_word16_t *in, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len) { - int N = st->filt_len; + const int N = st->filt_len; int out_sample = 0; - spx_word16_t *mem; int last_sample = st->last_sample[channel_index]; spx_uint32_t samp_frac_num = st->samp_frac_num[channel_index]; - mem = st->mem + channel_index * st->mem_alloc_size; + const int out_stride = st->out_stride; + const int int_advance = st->int_advance; + const int frac_advance = st->frac_advance; + const spx_uint32_t den_rate = st->den_rate; + int j; + spx_word32_t sum; + while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len)) { - int j; - spx_word32_t sum=0; - - /* We need to interpolate the sinc filter */ - spx_word32_t accum[4] = {0.f,0.f, 0.f, 0.f}; - spx_word16_t interp[4]; - const spx_word16_t *ptr; - int offset; - spx_word16_t frac; - offset = samp_frac_num*st->oversample/st->den_rate; + const spx_word16_t *iptr = & in[last_sample]; + + const int offset = samp_frac_num*st->oversample/st->den_rate; #ifdef FIXED_POINT - frac = PDIV32(SHL32((samp_frac_num*st->oversample) % st->den_rate,15),st->den_rate); + const spx_word16_t frac = PDIV32(SHL32((samp_frac_num*st->oversample) % st->den_rate,15),st->den_rate); #else - frac = ((float)((samp_frac_num*st->oversample) % st->den_rate))/st->den_rate; + const spx_word16_t frac = ((float)((samp_frac_num*st->oversample) % st->den_rate))/st->den_rate; #endif - /* This code is written like this to make it easy to optimise with SIMD. - For most DSPs, it would be best to split the loops in two because most DSPs - have only two accumulators */ - for (j=0;last_sample-N+1+j < 0;j++) - { - spx_word16_t curr_mem = mem[last_sample+j]; - accum[0] += MULT16_16(curr_mem,st->sinc_table[4+(j+1)*st->oversample-offset-2]); - accum[1] += MULT16_16(curr_mem,st->sinc_table[4+(j+1)*st->oversample-offset-1]); - accum[2] += MULT16_16(curr_mem,st->sinc_table[4+(j+1)*st->oversample-offset]); - accum[3] += MULT16_16(curr_mem,st->sinc_table[4+(j+1)*st->oversample-offset+1]); - } - - if (in != NULL) - { - ptr = in+st->in_stride*(last_sample-N+1+j); - /* Do the new part */ - for (;jin_stride; - accum[0] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-2]); - accum[1] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-1]); - accum[2] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset]); - accum[3] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset+1]); - } + spx_word16_t interp[4]; + + +#ifndef OVERRIDE_INTERPOLATE_PRODUCT_SINGLE + spx_word32_t accum[4] = {0,0,0,0}; + + for(j=0;jsinc_table[4+(j+1)*st->oversample-offset-2]); + accum[1] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-1]); + accum[2] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset]); + accum[3] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset+1]); } + cubic_coef(frac, interp); sum = MULT16_32_Q15(interp[0],accum[0]) + MULT16_32_Q15(interp[1],accum[1]) + MULT16_32_Q15(interp[2],accum[2]) + MULT16_32_Q15(interp[3],accum[3]); - - *out = PSHR32(sum,15); - out += st->out_stride; - out_sample++; - last_sample += st->int_advance; - samp_frac_num += st->frac_advance; - if (samp_frac_num >= st->den_rate) +#else + cubic_coef(frac, interp); + sum = interpolate_product_single(iptr, st->sinc_table + st->oversample + 4 - offset - 2, N, st->oversample, interp); +#endif + + out[out_stride * out_sample++] = PSHR32(sum,15); + last_sample += int_advance; + samp_frac_num += frac_advance; + if (samp_frac_num >= den_rate) { - samp_frac_num -= st->den_rate; + samp_frac_num -= den_rate; last_sample++; } } + st->last_sample[channel_index] = last_sample; st->samp_frac_num[channel_index] = samp_frac_num; return out_sample; @@ -490,63 +489,58 @@ static int resampler_basic_interpolate_single(SpeexResamplerState *st, spx_uint3 /* This is the same as the previous function, except with a double-precision accumulator */ static int resampler_basic_interpolate_double(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_word16_t *in, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len) { - int N = st->filt_len; + const int N = st->filt_len; int out_sample = 0; - spx_word16_t *mem; int last_sample = st->last_sample[channel_index]; spx_uint32_t samp_frac_num = st->samp_frac_num[channel_index]; - mem = st->mem + channel_index * st->mem_alloc_size; + const int out_stride = st->out_stride; + const int int_advance = st->int_advance; + const int frac_advance = st->frac_advance; + const spx_uint32_t den_rate = st->den_rate; + int j; + spx_word32_t sum; + while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len)) { - int j; - spx_word32_t sum=0; - - /* We need to interpolate the sinc filter */ - double accum[4] = {0.f,0.f, 0.f, 0.f}; - float interp[4]; - const spx_word16_t *ptr; - float alpha = ((float)samp_frac_num)/st->den_rate; - int offset = samp_frac_num*st->oversample/st->den_rate; - float frac = alpha*st->oversample - offset; - /* This code is written like this to make it easy to optimise with SIMD. - For most DSPs, it would be best to split the loops in two because most DSPs - have only two accumulators */ - for (j=0;last_sample-N+1+j < 0;j++) - { - double curr_mem = mem[last_sample+j]; - accum[0] += MULT16_16(curr_mem,st->sinc_table[4+(j+1)*st->oversample-offset-2]); - accum[1] += MULT16_16(curr_mem,st->sinc_table[4+(j+1)*st->oversample-offset-1]); - accum[2] += MULT16_16(curr_mem,st->sinc_table[4+(j+1)*st->oversample-offset]); - accum[3] += MULT16_16(curr_mem,st->sinc_table[4+(j+1)*st->oversample-offset+1]); - } - if (in != NULL) - { - ptr = in+st->in_stride*(last_sample-N+1+j); - /* Do the new part */ - for (;jin_stride; - accum[0] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-2]); - accum[1] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-1]); - accum[2] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset]); - accum[3] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset+1]); - } + const spx_word16_t *iptr = & in[last_sample]; + + const int offset = samp_frac_num*st->oversample/st->den_rate; +#ifdef FIXED_POINT + const spx_word16_t frac = PDIV32(SHL32((samp_frac_num*st->oversample) % st->den_rate,15),st->den_rate); +#else + const spx_word16_t frac = ((float)((samp_frac_num*st->oversample) % st->den_rate))/st->den_rate; +#endif + spx_word16_t interp[4]; + + +#ifndef OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE + double accum[4] = {0,0,0,0}; + + for(j=0;jsinc_table[4+(j+1)*st->oversample-offset-2]); + accum[1] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-1]); + accum[2] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset]); + accum[3] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset+1]); } + cubic_coef(frac, interp); - sum = interp[0]*accum[0] + interp[1]*accum[1] + interp[2]*accum[2] + interp[3]*accum[3]; - - *out = PSHR32(sum,15); - out += st->out_stride; - out_sample++; - last_sample += st->int_advance; - samp_frac_num += st->frac_advance; - if (samp_frac_num >= st->den_rate) + sum = MULT16_32_Q15(interp[0],accum[0]) + MULT16_32_Q15(interp[1],accum[1]) + MULT16_32_Q15(interp[2],accum[2]) + MULT16_32_Q15(interp[3],accum[3]); +#else + cubic_coef(frac, interp); + sum = interpolate_product_double(iptr, st->sinc_table + st->oversample + 4 - offset - 2, N, st->oversample, interp); +#endif + + out[out_stride * out_sample++] = PSHR32(sum,15); + last_sample += int_advance; + samp_frac_num += frac_advance; + if (samp_frac_num >= den_rate) { - samp_frac_num -= st->den_rate; + samp_frac_num -= den_rate; last_sample++; } } + st->last_sample[channel_index] = last_sample; st->samp_frac_num[channel_index] = samp_frac_num; return out_sample; @@ -556,11 +550,11 @@ static int resampler_basic_interpolate_double(SpeexResamplerState *st, spx_uint3 static void update_filter(SpeexResamplerState *st) { spx_uint32_t old_length; - + old_length = st->filt_len; st->oversample = quality_map[st->quality].oversample; st->filt_len = quality_map[st->quality].base_length; - + if (st->num_rate > st->den_rate) { /* down-sampling */ @@ -636,25 +630,25 @@ static void update_filter(SpeexResamplerState *st) st->int_advance = st->num_rate/st->den_rate; st->frac_advance = st->num_rate%st->den_rate; - + /* Here's the place where we update the filter memory to take into account the change in filter length. It's probably the messiest part of the code due to handling of lots of corner cases. */ if (!st->mem) { spx_uint32_t i; - st->mem = (spx_word16_t*)speex_alloc(st->nb_channels*(st->filt_len-1) * sizeof(spx_word16_t)); - for (i=0;inb_channels*(st->filt_len-1);i++) + st->mem_alloc_size = st->filt_len-1 + st->buffer_size; + st->mem = (spx_word16_t*)speex_alloc(st->nb_channels*st->mem_alloc_size * sizeof(spx_word16_t)); + for (i=0;inb_channels*st->mem_alloc_size;i++) st->mem[i] = 0; - st->mem_alloc_size = st->filt_len-1; /*speex_warning("init filter");*/ } else if (!st->started) { spx_uint32_t i; - st->mem = (spx_word16_t*)speex_realloc(st->mem, st->nb_channels*(st->filt_len-1) * sizeof(spx_word16_t)); - for (i=0;inb_channels*(st->filt_len-1);i++) + st->mem_alloc_size = st->filt_len-1 + st->buffer_size; + st->mem = (spx_word16_t*)speex_realloc(st->mem, st->nb_channels*st->mem_alloc_size * sizeof(spx_word16_t)); + for (i=0;inb_channels*st->mem_alloc_size;i++) st->mem[i] = 0; - st->mem_alloc_size = st->filt_len-1; /*speex_warning("reinit filter");*/ } else if (st->filt_len > old_length) { @@ -662,10 +656,10 @@ static void update_filter(SpeexResamplerState *st) /* Increase the filter length */ /*speex_warning("increase filter size");*/ int old_alloc_size = st->mem_alloc_size; - if (st->filt_len-1 > st->mem_alloc_size) + if ((st->filt_len-1 + st->buffer_size) > st->mem_alloc_size) { - st->mem = (spx_word16_t*)speex_realloc(st->mem, st->nb_channels*(st->filt_len-1) * sizeof(spx_word16_t)); - st->mem_alloc_size = st->filt_len-1; + st->mem_alloc_size = st->filt_len-1 + st->buffer_size; + st->mem = (spx_word16_t*)speex_realloc(st->mem, st->nb_channels*st->mem_alloc_size * sizeof(spx_word16_t)); } for (i=st->nb_channels-1;i>=0;i--) { @@ -674,7 +668,7 @@ static void update_filter(SpeexResamplerState *st) /*if (st->magic_samples[i])*/ { /* Try and remove the magic samples as if nothing had happened */ - + /* FIXME: This is wrong but for now we need it to avoid going over the array bounds */ olen = old_length + 2*st->magic_samples[i]; for (j=old_length-2+st->magic_samples[i];j>=0;j--) @@ -721,12 +715,12 @@ static void update_filter(SpeexResamplerState *st) } -SpeexResamplerState *speex_resampler_init(spx_uint32_t nb_channels, spx_uint32_t in_rate, spx_uint32_t out_rate, int quality, int *err) +EXPORT SpeexResamplerState *speex_resampler_init(spx_uint32_t nb_channels, spx_uint32_t in_rate, spx_uint32_t out_rate, int quality, int *err) { return speex_resampler_init_frac(nb_channels, in_rate, out_rate, in_rate, out_rate, quality, err); } -SpeexResamplerState *speex_resampler_init_frac(spx_uint32_t nb_channels, spx_uint32_t ratio_num, spx_uint32_t ratio_den, spx_uint32_t in_rate, spx_uint32_t out_rate, int quality, int *err) +EXPORT SpeexResamplerState *speex_resampler_init_frac(spx_uint32_t nb_channels, spx_uint32_t ratio_num, spx_uint32_t ratio_den, spx_uint32_t in_rate, spx_uint32_t out_rate, int quality, int *err) { spx_uint32_t i; SpeexResamplerState *st; @@ -749,12 +743,18 @@ SpeexResamplerState *speex_resampler_init_frac(spx_uint32_t nb_channels, spx_uin st->filt_len = 0; st->mem = 0; st->resampler_ptr = 0; - + st->cutoff = 1.f; st->nb_channels = nb_channels; st->in_stride = 1; st->out_stride = 1; - + +#ifdef FIXED_POINT + st->buffer_size = 160; +#else + st->buffer_size = 160; +#endif + /* Per channel data */ st->last_sample = (spx_int32_t*)speex_alloc(nb_channels*sizeof(int)); st->magic_samples = (spx_uint32_t*)speex_alloc(nb_channels*sizeof(int)); @@ -769,9 +769,9 @@ SpeexResamplerState *speex_resampler_init_frac(spx_uint32_t nb_channels, spx_uin speex_resampler_set_quality(st, quality); speex_resampler_set_rate_frac(st, ratio_num, ratio_den, in_rate, out_rate); - + update_filter(st); - + st->initialised = 1; if (err) *err = RESAMPLER_ERR_SUCCESS; @@ -779,7 +779,7 @@ SpeexResamplerState *speex_resampler_init_frac(spx_uint32_t nb_channels, spx_uin return st; } -void speex_resampler_destroy(SpeexResamplerState *st) +EXPORT void speex_resampler_destroy(SpeexResamplerState *st) { speex_free(st->mem); speex_free(st->sinc_table); @@ -789,186 +789,168 @@ void speex_resampler_destroy(SpeexResamplerState *st) speex_free(st); } - - -static int speex_resampler_process_native(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_word16_t *in, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len) +static int speex_resampler_process_native(SpeexResamplerState *st, spx_uint32_t channel_index, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len) { int j=0; - int N = st->filt_len; + const int N = st->filt_len; int out_sample = 0; - spx_word16_t *mem; - spx_uint32_t tmp_out_len = 0; - mem = st->mem + channel_index * st->mem_alloc_size; + spx_word16_t *mem = st->mem + channel_index * st->mem_alloc_size; + spx_uint32_t ilen; + st->started = 1; - - /* Handle the case where we have samples left from a reduction in filter length */ - if (st->magic_samples[channel_index]) - { - int istride_save; - spx_uint32_t tmp_in_len; - spx_uint32_t tmp_magic; - - istride_save = st->in_stride; - tmp_in_len = st->magic_samples[channel_index]; - tmp_out_len = *out_len; - /* magic_samples needs to be set to zero to avoid infinite recursion */ - tmp_magic = st->magic_samples[channel_index]; - st->magic_samples[channel_index] = 0; - st->in_stride = 1; - speex_resampler_process_native(st, channel_index, mem+N-1, &tmp_in_len, out, &tmp_out_len); - st->in_stride = istride_save; - /*speex_warning_int("extra samples:", tmp_out_len);*/ - /* If we couldn't process all "magic" input samples, save the rest for next time */ - if (tmp_in_len < tmp_magic) - { - spx_uint32_t i; - st->magic_samples[channel_index] = tmp_magic-tmp_in_len; - for (i=0;imagic_samples[channel_index];i++) - mem[N-1+i]=mem[N-1+i+tmp_in_len]; - } - out += tmp_out_len*st->out_stride; - *out_len -= tmp_out_len; - } - + /* Call the right resampler through the function ptr */ - out_sample = st->resampler_ptr(st, channel_index, in, in_len, out, out_len); - + out_sample = st->resampler_ptr(st, channel_index, mem, in_len, out, out_len); + if (st->last_sample[channel_index] < (spx_int32_t)*in_len) *in_len = st->last_sample[channel_index]; - *out_len = out_sample+tmp_out_len; + *out_len = out_sample; st->last_sample[channel_index] -= *in_len; - - for (j=0;jin_stride*(j+*in_len-N+1)]; - + + ilen = *in_len; + + for(j=0;jmagic_samples[channel_index]; + spx_word16_t *mem = st->mem + channel_index * st->mem_alloc_size; + const int N = st->filt_len; + + speex_resampler_process_native(st, channel_index, &tmp_in_len, *out, &out_len); + + st->magic_samples[channel_index] -= tmp_in_len; + + /* If we couldn't process all "magic" input samples, save the rest for next time */ + if (st->magic_samples[channel_index]) + { + spx_uint32_t i; + for (i=0;imagic_samples[channel_index];i++) + mem[N-1+i]=mem[N-1+i+tmp_in_len]; + } + *out += out_len*st->out_stride; + return out_len; +} #ifdef FIXED_POINT -int speex_resampler_process_float(SpeexResamplerState *st, spx_uint32_t channel_index, const float *in, spx_uint32_t *in_len, float *out, spx_uint32_t *out_len) -{ - spx_uint32_t i; - int istride_save, ostride_save; -#ifdef VAR_ARRAYS - spx_word16_t x[*in_len]; - spx_word16_t y[*out_len]; - /*VARDECL(spx_word16_t *x); - VARDECL(spx_word16_t *y); - ALLOC(x, *in_len, spx_word16_t); - ALLOC(y, *out_len, spx_word16_t);*/ - istride_save = st->in_stride; - ostride_save = st->out_stride; - for (i=0;i<*in_len;i++) - x[i] = WORD2INT(in[i*st->in_stride]); - st->in_stride = st->out_stride = 1; - speex_resampler_process_native(st, channel_index, x, in_len, y, out_len); - st->in_stride = istride_save; - st->out_stride = ostride_save; - for (i=0;i<*out_len;i++) - out[i*st->out_stride] = y[i]; +EXPORT int speex_resampler_process_int(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_int16_t *in, spx_uint32_t *in_len, spx_int16_t *out, spx_uint32_t *out_len) #else - spx_word16_t x[FIXED_STACK_ALLOC]; - spx_word16_t y[FIXED_STACK_ALLOC]; - spx_uint32_t ilen=*in_len, olen=*out_len; - istride_save = st->in_stride; - ostride_save = st->out_stride; - while (ilen && olen) - { - spx_uint32_t ichunk, ochunk; - ichunk = ilen; - ochunk = olen; - if (ichunk>FIXED_STACK_ALLOC) - ichunk=FIXED_STACK_ALLOC; - if (ochunk>FIXED_STACK_ALLOC) - ochunk=FIXED_STACK_ALLOC; - for (i=0;iin_stride]); - st->in_stride = st->out_stride = 1; - speex_resampler_process_native(st, channel_index, x, &ichunk, y, &ochunk); - st->in_stride = istride_save; - st->out_stride = ostride_save; - for (i=0;iout_stride] = y[i]; - out += ochunk; - in += ichunk; - ilen -= ichunk; - olen -= ochunk; +EXPORT int speex_resampler_process_float(SpeexResamplerState *st, spx_uint32_t channel_index, const float *in, spx_uint32_t *in_len, float *out, spx_uint32_t *out_len) +#endif +{ + int j; + spx_uint32_t ilen = *in_len; + spx_uint32_t olen = *out_len; + spx_word16_t *x = st->mem + channel_index * st->mem_alloc_size; + const int filt_offs = st->filt_len - 1; + const spx_uint32_t xlen = st->mem_alloc_size - filt_offs; + const int istride = st->in_stride; + + if (st->magic_samples[channel_index]) + olen -= speex_resampler_magic(st, channel_index, &out, olen); + if (! st->magic_samples[channel_index]) { + while (ilen && olen) { + spx_uint32_t ichunk = (ilen > xlen) ? xlen : ilen; + spx_uint32_t ochunk = olen; + + if (in) { + for(j=0;jout_stride; + if (in) + in += ichunk * istride; + } } *in_len -= ilen; - *out_len -= olen; -#endif + *out_len -= olen; return RESAMPLER_ERR_SUCCESS; } -int speex_resampler_process_int(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_int16_t *in, spx_uint32_t *in_len, spx_int16_t *out, spx_uint32_t *out_len) -{ - return speex_resampler_process_native(st, channel_index, in, in_len, out, out_len); -} + +#ifdef FIXED_POINT +EXPORT int speex_resampler_process_float(SpeexResamplerState *st, spx_uint32_t channel_index, const float *in, spx_uint32_t *in_len, float *out, spx_uint32_t *out_len) #else -int speex_resampler_process_float(SpeexResamplerState *st, spx_uint32_t channel_index, const float *in, spx_uint32_t *in_len, float *out, spx_uint32_t *out_len) -{ - return speex_resampler_process_native(st, channel_index, in, in_len, out, out_len); -} -int speex_resampler_process_int(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_int16_t *in, spx_uint32_t *in_len, spx_int16_t *out, spx_uint32_t *out_len) +EXPORT int speex_resampler_process_int(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_int16_t *in, spx_uint32_t *in_len, spx_int16_t *out, spx_uint32_t *out_len) +#endif { - spx_uint32_t i; - int istride_save, ostride_save; + int j; + const int istride_save = st->in_stride; + const int ostride_save = st->out_stride; + spx_uint32_t ilen = *in_len; + spx_uint32_t olen = *out_len; + spx_word16_t *x = st->mem + channel_index * st->mem_alloc_size; + const spx_uint32_t xlen = st->mem_alloc_size - (st->filt_len - 1); #ifdef VAR_ARRAYS - spx_word16_t x[*in_len]; - spx_word16_t y[*out_len]; - /*VARDECL(spx_word16_t *x); - VARDECL(spx_word16_t *y); - ALLOC(x, *in_len, spx_word16_t); - ALLOC(y, *out_len, spx_word16_t);*/ - istride_save = st->in_stride; - ostride_save = st->out_stride; - for (i=0;i<*in_len;i++) - x[i] = in[i*st->in_stride]; - st->in_stride = st->out_stride = 1; - speex_resampler_process_native(st, channel_index, x, in_len, y, out_len); - st->in_stride = istride_save; - st->out_stride = ostride_save; - for (i=0;i<*out_len;i++) - out[i*st->out_stride] = WORD2INT(y[i]); + const unsigned int ylen = (olen < FIXED_STACK_ALLOC) ? olen : FIXED_STACK_ALLOC; + VARDECL(spx_word16_t *ystack); + ALLOC(ystack, ylen, spx_word16_t); #else - spx_word16_t x[FIXED_STACK_ALLOC]; - spx_word16_t y[FIXED_STACK_ALLOC]; - spx_uint32_t ilen=*in_len, olen=*out_len; - istride_save = st->in_stride; - ostride_save = st->out_stride; - while (ilen && olen) - { - spx_uint32_t ichunk, ochunk; - ichunk = ilen; - ochunk = olen; - if (ichunk>FIXED_STACK_ALLOC) - ichunk=FIXED_STACK_ALLOC; - if (ochunk>FIXED_STACK_ALLOC) - ochunk=FIXED_STACK_ALLOC; - for (i=0;iin_stride]; - st->in_stride = st->out_stride = 1; - speex_resampler_process_native(st, channel_index, x, &ichunk, y, &ochunk); - st->in_stride = istride_save; - st->out_stride = ostride_save; - for (i=0;iout_stride] = WORD2INT(y[i]); - out += ochunk; - in += ichunk; - ilen -= ichunk; - olen -= ochunk; + const unsigned int ylen = FIXED_STACK_ALLOC; + spx_word16_t ystack[FIXED_STACK_ALLOC]; +#endif + + st->out_stride = 1; + + while (ilen && olen) { + spx_word16_t *y = ystack; + spx_uint32_t ichunk = (ilen > xlen) ? xlen : ilen; + spx_uint32_t ochunk = (olen > ylen) ? ylen : olen; + spx_uint32_t omagic = 0; + + if (st->magic_samples[channel_index]) { + omagic = speex_resampler_magic(st, channel_index, &y, ochunk); + ochunk -= omagic; + olen -= omagic; + } + if (! st->magic_samples[channel_index]) { + if (in) { + for(j=0;jfilt_len-1]=WORD2INT(in[j*istride_save]); +#else + x[j+st->filt_len-1]=in[j*istride_save]; +#endif + } else { + for(j=0;jfilt_len-1]=0; + } + + speex_resampler_process_native(st, channel_index, &ichunk, y, &ochunk); + } else { + ichunk = 0; + ochunk = 0; + } + + for (j=0;jout_stride = ostride_save; *in_len -= ilen; - *out_len -= olen; -#endif + *out_len -= olen; + return RESAMPLER_ERR_SUCCESS; } -#endif -int speex_resampler_process_interleaved_float(SpeexResamplerState *st, const float *in, spx_uint32_t *in_len, float *out, spx_uint32_t *out_len) +EXPORT int speex_resampler_process_interleaved_float(SpeexResamplerState *st, const float *in, spx_uint32_t *in_len, float *out, spx_uint32_t *out_len) { spx_uint32_t i; int istride_save, ostride_save; @@ -989,8 +971,7 @@ int speex_resampler_process_interleaved_float(SpeexResamplerState *st, const flo return RESAMPLER_ERR_SUCCESS; } - -int speex_resampler_process_interleaved_int(SpeexResamplerState *st, const spx_int16_t *in, spx_uint32_t *in_len, spx_int16_t *out, spx_uint32_t *out_len) +EXPORT int speex_resampler_process_interleaved_int(SpeexResamplerState *st, const spx_int16_t *in, spx_uint32_t *in_len, spx_int16_t *out, spx_uint32_t *out_len) { spx_uint32_t i; int istride_save, ostride_save; @@ -1011,25 +992,25 @@ int speex_resampler_process_interleaved_int(SpeexResamplerState *st, const spx_i return RESAMPLER_ERR_SUCCESS; } -int speex_resampler_set_rate(SpeexResamplerState *st, spx_uint32_t in_rate, spx_uint32_t out_rate) +EXPORT int speex_resampler_set_rate(SpeexResamplerState *st, spx_uint32_t in_rate, spx_uint32_t out_rate) { return speex_resampler_set_rate_frac(st, in_rate, out_rate, in_rate, out_rate); } -void speex_resampler_get_rate(SpeexResamplerState *st, spx_uint32_t *in_rate, spx_uint32_t *out_rate) +EXPORT void speex_resampler_get_rate(SpeexResamplerState *st, spx_uint32_t *in_rate, spx_uint32_t *out_rate) { *in_rate = st->in_rate; *out_rate = st->out_rate; } -int speex_resampler_set_rate_frac(SpeexResamplerState *st, spx_uint32_t ratio_num, spx_uint32_t ratio_den, spx_uint32_t in_rate, spx_uint32_t out_rate) +EXPORT int speex_resampler_set_rate_frac(SpeexResamplerState *st, spx_uint32_t ratio_num, spx_uint32_t ratio_den, spx_uint32_t in_rate, spx_uint32_t out_rate) { spx_uint32_t fact; spx_uint32_t old_den; spx_uint32_t i; if (st->in_rate == in_rate && st->out_rate == out_rate && st->num_rate == ratio_num && st->den_rate == ratio_den) return RESAMPLER_ERR_SUCCESS; - + old_den = st->den_rate; st->in_rate = in_rate; st->out_rate = out_rate; @@ -1044,7 +1025,7 @@ int speex_resampler_set_rate_frac(SpeexResamplerState *st, spx_uint32_t ratio_nu st->den_rate /= fact; } } - + if (old_den > 0) { for (i=0;inb_channels;i++) @@ -1055,19 +1036,19 @@ int speex_resampler_set_rate_frac(SpeexResamplerState *st, spx_uint32_t ratio_nu st->samp_frac_num[i] = st->den_rate-1; } } - + if (st->initialised) update_filter(st); return RESAMPLER_ERR_SUCCESS; } -void speex_resampler_get_ratio(SpeexResamplerState *st, spx_uint32_t *ratio_num, spx_uint32_t *ratio_den) +EXPORT void speex_resampler_get_ratio(SpeexResamplerState *st, spx_uint32_t *ratio_num, spx_uint32_t *ratio_den) { *ratio_num = st->num_rate; *ratio_den = st->den_rate; } -int speex_resampler_set_quality(SpeexResamplerState *st, int quality) +EXPORT int speex_resampler_set_quality(SpeexResamplerState *st, int quality) { if (quality > 10 || quality < 0) return RESAMPLER_ERR_INVALID_ARG; @@ -1079,32 +1060,42 @@ int speex_resampler_set_quality(SpeexResamplerState *st, int quality) return RESAMPLER_ERR_SUCCESS; } -void speex_resampler_get_quality(SpeexResamplerState *st, int *quality) +EXPORT void speex_resampler_get_quality(SpeexResamplerState *st, int *quality) { *quality = st->quality; } -void speex_resampler_set_input_stride(SpeexResamplerState *st, spx_uint32_t stride) +EXPORT void speex_resampler_set_input_stride(SpeexResamplerState *st, spx_uint32_t stride) { st->in_stride = stride; } -void speex_resampler_get_input_stride(SpeexResamplerState *st, spx_uint32_t *stride) +EXPORT void speex_resampler_get_input_stride(SpeexResamplerState *st, spx_uint32_t *stride) { *stride = st->in_stride; } -void speex_resampler_set_output_stride(SpeexResamplerState *st, spx_uint32_t stride) +EXPORT void speex_resampler_set_output_stride(SpeexResamplerState *st, spx_uint32_t stride) { st->out_stride = stride; } -void speex_resampler_get_output_stride(SpeexResamplerState *st, spx_uint32_t *stride) +EXPORT void speex_resampler_get_output_stride(SpeexResamplerState *st, spx_uint32_t *stride) { *stride = st->out_stride; } -int speex_resampler_skip_zeros(SpeexResamplerState *st) +EXPORT int speex_resampler_get_input_latency(SpeexResamplerState *st) +{ + return st->filt_len / 2; +} + +EXPORT int speex_resampler_get_output_latency(SpeexResamplerState *st) +{ + return ((st->filt_len / 2) * st->den_rate + (st->num_rate >> 1)) / st->num_rate; +} + +EXPORT int speex_resampler_skip_zeros(SpeexResamplerState *st) { spx_uint32_t i; for (i=0;inb_channels;i++) @@ -1112,7 +1103,7 @@ int speex_resampler_skip_zeros(SpeexResamplerState *st) return RESAMPLER_ERR_SUCCESS; } -int speex_resampler_reset_mem(SpeexResamplerState *st) +EXPORT int speex_resampler_reset_mem(SpeexResamplerState *st) { spx_uint32_t i; for (i=0;inb_channels*(st->filt_len-1);i++) @@ -1120,7 +1111,7 @@ int speex_resampler_reset_mem(SpeexResamplerState *st) return RESAMPLER_ERR_SUCCESS; } -const char *speex_resampler_strerror(int err) +EXPORT const char *speex_resampler_strerror(int err) { switch (err) { -- cgit v1.2.3