From 9c8913437bf5eecd35a88ad0d7f0bcefd9333c0f Mon Sep 17 00:00:00 2001 From: Wim Taymans Date: Tue, 26 Mar 2019 12:58:26 +0100 Subject: [PATCH] audioconvert: add simple and faster resampler Add a sinc based resampler that, unlike speex, avoids memcpy and works directly on the source data. It also allows for ssse3 optimizations and aligned loads. It will later switch to table interpolation when doing variable rate. --- spa/include/spa/utils/defs.h | 3 +- spa/plugins/audioconvert/meson.build | 1 + spa/plugins/audioconvert/resample-native-c.h | 51 ++++ .../audioconvert/resample-native-impl.h | 112 +++++++ .../audioconvert/resample-native-sse.h | 74 +++++ .../audioconvert/resample-native-ssse3.h | 110 +++++++ spa/plugins/audioconvert/resample-native.h | 273 ++++++++++++++++++ spa/plugins/audioconvert/resample.c | 3 + spa/plugins/audioconvert/resample.h | 14 +- spa/plugins/audioconvert/test-resample.c | 86 ++++++ 10 files changed, 720 insertions(+), 7 deletions(-) create mode 100644 spa/plugins/audioconvert/resample-native-c.h create mode 100644 spa/plugins/audioconvert/resample-native-impl.h create mode 100644 spa/plugins/audioconvert/resample-native-sse.h create mode 100644 spa/plugins/audioconvert/resample-native-ssse3.h create mode 100644 spa/plugins/audioconvert/resample-native.h create mode 100644 spa/plugins/audioconvert/test-resample.c diff --git a/spa/include/spa/utils/defs.h b/spa/include/spa/utils/defs.h index eb9b9917e..fb1e3ddb7 100644 --- a/spa/include/spa/utils/defs.h +++ b/spa/include/spa/utils/defs.h @@ -171,7 +171,8 @@ struct spa_param_info { #define SPA_ROUND_DOWN_N(num,align) ((num) & ~((align) - 1)) #define SPA_ROUND_UP_N(num,align) SPA_ROUND_DOWN_N((num) + ((align) - 1),align) -#define SPA_IS_ALIGNED(p,align) (((intptr_t)(p) & ((align)-1)) == 0) +#define SPA_PTR_ALIGNMENT(p,align) ((intptr_t)(p) & ((align)-1)) +#define SPA_IS_ALIGNED(p,align) (SPA_PTR_ALIGNMENT(p,align) == 0) #define SPA_PTR_ALIGN(p,align,type) (type*)SPA_ROUND_UP_N((intptr_t)(p), (intptr_t)(align)) #ifndef SPA_LIKELY diff --git a/spa/plugins/audioconvert/meson.build b/spa/plugins/audioconvert/meson.build index 93d76a446..e323a2870 100644 --- a/spa/plugins/audioconvert/meson.build +++ b/spa/plugins/audioconvert/meson.build @@ -15,6 +15,7 @@ audioconvertlib = shared_library('spa-audioconvert', test_apps = [ 'test-fmt-ops', + 'test-resample', ] foreach a : test_apps diff --git a/spa/plugins/audioconvert/resample-native-c.h b/spa/plugins/audioconvert/resample-native-c.h new file mode 100644 index 000000000..c2ff6d289 --- /dev/null +++ b/spa/plugins/audioconvert/resample-native-c.h @@ -0,0 +1,51 @@ +/* Spa + * + * Copyright © 2019 Wim Taymans + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ + +static void inner_product_c(float *d, const float * SPA_RESTRICT s, + const float * SPA_RESTRICT taps, uint32_t n_taps) +{ + float sum = 0.0f; + uint32_t i; + + for (i = 0; i < n_taps; i++) + sum += s[i] * taps[i]; + *d = sum; +} + +static void inner_product_ip_c(float *d, const float * SPA_RESTRICT s, + const float * SPA_RESTRICT t0, const float * SPA_RESTRICT t1, float x, + uint32_t n_taps) +{ + float sum[2] = { 0.0f, 0.0f }; + uint32_t i; + + for (i = 0; i < n_taps; i++) { + sum[0] += s[i] * t0[i]; + sum[1] += s[i] * t1[i]; + } + *d = (sum[1] - sum[0]) * x + sum[0]; +} + +MAKE_RESAMPLER_FULL(c); +MAKE_RESAMPLER_INTER(c); diff --git a/spa/plugins/audioconvert/resample-native-impl.h b/spa/plugins/audioconvert/resample-native-impl.h new file mode 100644 index 000000000..1a25c2909 --- /dev/null +++ b/spa/plugins/audioconvert/resample-native-impl.h @@ -0,0 +1,112 @@ +/* Spa + * + * Copyright © 2019 Wim Taymans + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ + +#define MAKE_RESAMPLER_FULL(arch) \ +static void do_resample_full_##arch(struct resample *r, \ + const void * SPA_RESTRICT src[], uint32_t *in_len, \ + void * SPA_RESTRICT dst[], uint32_t offs, uint32_t *out_len) \ +{ \ + struct native_data *data = r->data; \ + uint32_t index, phase; \ + uint32_t out_rate = data->out_rate; \ + uint32_t n_taps = data->n_taps; \ + uint32_t c, o, olen = *out_len, ilen = *in_len; \ + \ + if (r->channels == 0) \ + return; \ + \ + for (c = 0; c < r->channels; c++) { \ + const float *s = src[c]; \ + float *d = dst[c]; \ + \ + index = data->index; \ + phase = data->phase; \ + \ + for (o = offs; o < olen && index + n_taps <= ilen; o++) { \ + const float *ip, *taps; \ + \ + ip = &s[index]; \ + taps = &data->filter[phase * n_taps]; \ + index += data->inc; \ + phase += data->frac; \ + if (phase >= out_rate) { \ + phase -= out_rate; \ + index += 1; \ + } \ + inner_product_##arch(&d[o], ip, taps, n_taps); \ + } \ + } \ + *in_len = index - data->index; \ + *out_len = o; \ + data->index = index; \ + data->phase = phase; \ +} + +#define MAKE_RESAMPLER_INTER(arch) \ +static void do_resample_inter_##arch(struct resample *r, \ + const void * SPA_RESTRICT src[], uint32_t *in_len, \ + void * SPA_RESTRICT dst[], uint32_t offs, uint32_t *out_len) \ +{ \ + struct native_data *data = r->data; \ + uint32_t index, phase; \ + uint32_t n_phases = data->n_phases, out_rate = data->out_rate; \ + uint32_t n_taps = data->n_taps; \ + uint32_t c, o, olen = *out_len, ilen = *in_len; \ + \ + if (r->channels == 0) \ + return; \ + \ + for (c = 0; c < r->channels; c++) { \ + const float *s = src[c]; \ + float *d = dst[c]; \ + \ + index = data->index; \ + phase = data->phase; \ + \ + for (o = offs; o < olen && index + n_taps <= ilen; o++) { \ + const float *ip; \ + float ph, x, *t0, *t1; \ + uint32_t offset; \ + \ + ip = &s[index]; \ + ph = (float)phase * n_phases / out_rate; \ + offset = floor(ph); \ + x = ph - (float)offset; \ + \ + t0 = &data->filter[(offset + 0) * n_taps]; \ + t1 = &data->filter[(offset + 1) * n_taps]; \ + index += data->inc; \ + phase += data->frac; \ + if (phase >= out_rate) { \ + phase -= out_rate; \ + index += 1; \ + } \ + inner_product_ip_##arch(&d[o], ip, t0, t1, x, n_taps); \ + } \ + } \ + *in_len = index - data->index; \ + *out_len = o; \ + data->index = index; \ + data->phase = phase; \ +} diff --git a/spa/plugins/audioconvert/resample-native-sse.h b/spa/plugins/audioconvert/resample-native-sse.h new file mode 100644 index 000000000..d52c14753 --- /dev/null +++ b/spa/plugins/audioconvert/resample-native-sse.h @@ -0,0 +1,74 @@ +/* Spa + * + * Copyright © 2019 Wim Taymans + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ + +#include + +#pragma GCC target("ssse3") +#include + +static void inner_product_sse(float *d, const float * SPA_RESTRICT s, + const float * SPA_RESTRICT taps, uint32_t n_taps) +{ + __m128 sum = _mm_setzero_ps(); + uint32_t i; + + for (i = 0; i < n_taps; i += 8) { + sum = _mm_add_ps(sum, + _mm_mul_ps( + _mm_loadu_ps(s + i + 0), + _mm_load_ps(taps + i + 0))); + sum = _mm_add_ps(sum, + _mm_mul_ps( + _mm_loadu_ps(s + i + 4), + _mm_load_ps(taps + i + 4))); + } + sum = _mm_add_ps(sum, _mm_movehl_ps(sum, sum)); + sum = _mm_add_ss(sum, _mm_shuffle_ps(sum, sum, 0x55)); + _mm_store_ss(d, sum); +} + +static void inner_product_ip_sse(float *d, const float * SPA_RESTRICT s, + const float * SPA_RESTRICT t0, const float * SPA_RESTRICT t1, float x, + uint32_t n_taps) +{ + __m128 sum[2] = { _mm_setzero_ps (), _mm_setzero_ps () }, t; + uint32_t i; + + for (i = 0; i < n_taps; i += 8) { + t = _mm_loadu_ps(s + i + 0); + sum[0] = _mm_add_ps(sum[0], _mm_mul_ps(t, _mm_load_ps(t0 + i + 0))); + sum[1] = _mm_add_ps(sum[1], _mm_mul_ps(t, _mm_load_ps(t1 + i + 0))); + t = _mm_loadu_ps(s + i + 4); + sum[0] = _mm_add_ps(sum[0], _mm_mul_ps(t, _mm_load_ps(t0 + i + 4))); + sum[1] = _mm_add_ps(sum[1], _mm_mul_ps(t, _mm_load_ps(t1 + i + 4))); + } + sum[1] = _mm_mul_ps(_mm_sub_ps(sum[1], sum[0]), _mm_load1_ps(&x)); + sum[0] = _mm_add_ps(sum[0], sum[1]); + sum[0] = _mm_add_ps(sum[0], _mm_movehl_ps(sum[0], sum[0])); + sum[0] = _mm_add_ss(sum[0], _mm_shuffle_ps(sum[0], sum[0], 0x55)); + _mm_store_ss(d, sum[0]); +} + +MAKE_RESAMPLER_FULL(sse); +MAKE_RESAMPLER_INTER(sse); diff --git a/spa/plugins/audioconvert/resample-native-ssse3.h b/spa/plugins/audioconvert/resample-native-ssse3.h new file mode 100644 index 000000000..4812d891f --- /dev/null +++ b/spa/plugins/audioconvert/resample-native-ssse3.h @@ -0,0 +1,110 @@ +/* Spa + * + * Copyright © 2019 Wim Taymans + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ + +#include + +static void inner_product_ssse3(float *d, const float * SPA_RESTRICT s, + const float * SPA_RESTRICT taps, uint32_t n_taps) +{ + __m128 sum = _mm_setzero_ps(); + __m128 t0, t1; + uint32_t i; + + switch (SPA_PTR_ALIGNMENT(s, 16)) { + case 0: + for (i = 0; i < n_taps; i += 8) { + sum = _mm_add_ps(sum, + _mm_mul_ps( + _mm_load_ps(s + i + 0), + _mm_load_ps(taps + i + 0))); + sum = _mm_add_ps(sum, + _mm_mul_ps( + _mm_load_ps(s + i + 4), + _mm_load_ps(taps + i + 4))); + } + break; + case 4: + t0 = _mm_load_ps(s - 1); + for (i = 0; i < n_taps; i += 8) { + t1 = _mm_load_ps(s + i + 3); + t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 4); + sum = _mm_add_ps(sum, + _mm_mul_ps(t0, _mm_load_ps(taps + i + 0))); + t0 = t1; + t1 = _mm_load_ps(s + i + 7); + t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 4); + sum = _mm_add_ps(sum, + _mm_mul_ps(t0, _mm_load_ps(taps + i + 4))); + t0 = t1; + } + break; + case 8: + t0 = _mm_load_ps(s - 2); + for (i = 0; i < n_taps; i += 8) { + t1 = _mm_load_ps(s + i + 2); + t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 8); + sum = _mm_add_ps(sum, + _mm_mul_ps(t0, _mm_load_ps(taps + i + 0))); + t0 = t1; + t1 = _mm_load_ps(s + i + 6); + t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 8); + sum = _mm_add_ps(sum, + _mm_mul_ps(t0, _mm_load_ps(taps + i + 4))); + t0 = t1; + } + break; + case 12: + t0 = _mm_load_ps(s - 3); + for (i = 0; i < n_taps; i += 8) { + t1 = _mm_load_ps(s + i + 1); + t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 12); + sum = _mm_add_ps(sum, + _mm_mul_ps(t0, _mm_load_ps(taps + i + 0))); + t0 = t1; + t1 = _mm_load_ps(s + i + 5); + t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 12); + sum = _mm_add_ps(sum, + _mm_mul_ps(t0, _mm_load_ps(taps + i + 4))); + t0 = t1; + } + break; + } + sum = _mm_add_ps(sum, _mm_movehl_ps(sum, sum)); + sum = _mm_add_ss(sum, _mm_shuffle_ps(sum, sum, 0x55)); + _mm_store_ss(d, sum); +} + +static void inner_product_ip_ssse3(float *d, const float * SPA_RESTRICT s, + const float * SPA_RESTRICT t0, const float * SPA_RESTRICT t1, float x, + uint32_t n_taps) +{ +#if defined (__SSE__) + inner_product_ip_sse(d, s, t0, t1, x, n_taps); +#else + inner_product_ip_c(d, s, t0, t1, x, n_taps); +#endif +} + +MAKE_RESAMPLER_FULL(ssse3); +MAKE_RESAMPLER_INTER(ssse3); diff --git a/spa/plugins/audioconvert/resample-native.h b/spa/plugins/audioconvert/resample-native.h new file mode 100644 index 000000000..4c5628f60 --- /dev/null +++ b/spa/plugins/audioconvert/resample-native.h @@ -0,0 +1,273 @@ +/* Spa + * + * Copyright © 2019 Wim Taymans + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ + +#include + +typedef void (*resample_func_t)(struct resample *r, + const void * SPA_RESTRICT src[], uint32_t *in_len, + void * SPA_RESTRICT dst[], uint32_t offs, uint32_t *out_len); + +struct native_data { + double rate; + uint32_t n_taps; + uint32_t n_phases; + uint32_t in_rate; + uint32_t out_rate; + uint32_t index; + uint32_t phase; + uint32_t inc; + uint32_t frac; + uint32_t hist; + float **history; + resample_func_t func; + float *filter; + float *hist_mem; +}; + +#include "resample-native-impl.h" +#include "resample-native-c.h" +#if defined (__SSE__) +#include "resample-native-sse.h" +#endif +#if defined (__SSSE3__) +#include "resample-native-ssse3.h" +#endif + +struct quality { + uint32_t n_taps; + double cutoff; +}; + +#define DEFAULT_QUALITY 4 + +static const struct quality blackman_qualities[] = { + { 8, 0.5, }, + { 16, 0.6, }, + { 24, 0.72, }, + { 32, 0.8, }, + { 48, 0.85, }, /* default */ + { 64, 0.90, }, + { 80, 0.92, }, + { 96, 0.933, }, + { 128, 0.950, }, + { 144, 0.955, }, + { 160, 0.960, } +}; + +static inline double sinc(double x) +{ + if (x < 1e-6) return 1.0; + x *= M_PI; + return sin(x) / x; +} + +static inline double blackman(double x, double n_taps) +{ + double w = 2.0 * x * M_PI / n_taps + M_PI; + return 0.3635819 - 0.4891775 * cos(w) + + 0.1365995 * cos(2 * w) - 0.0106411 * cos(3 * w); +} + +static int build_filter(float *taps, uint32_t stride, uint32_t n_taps, uint32_t n_phases, double cutoff) +{ + uint32_t i, j, n_taps12 = n_taps/2; + + for (i = 0; i <= n_phases; i++) { + double t = (double) i / (double) n_phases; + for (j = 0; j < n_taps12; j++, t += 1.0) + taps[i * stride + (n_taps12 - j - 1)] = + cutoff * sinc(t * cutoff) * blackman(t, n_taps); + } + for (i = 0; i <= n_phases; i++) { + for (j = n_taps12; j < n_taps; j++) + taps[i * stride + j] = taps[(n_phases - i) * stride + n_taps - j - 1]; + } + return 0; +} + +static void impl_native_free(struct resample *r) +{ + free(r->data); + r->data = NULL; +} + +static void impl_native_update_rate(struct resample *r, double rate) +{ +} + +static void impl_native_process(struct resample *r, + const void * SPA_RESTRICT src[], uint32_t *in_len, + void * SPA_RESTRICT dst[], uint32_t *out_len) +{ + struct native_data *data = r->data; + uint32_t n_taps = data->n_taps; + float **history = data->history; + const float **s; + uint32_t c, refill, hist, in, out, remain; + + out = refill = in = 0; + hist = data->hist; + + if (hist) { + if (hist < n_taps) { + refill = SPA_MIN(*in_len, n_taps); + for (c = 0; c < r->channels; c++) + memcpy(&history[c][hist], src[c], refill * sizeof(float)); + + if (hist + refill < n_taps) { + data->hist = hist + refill; + *in_len = refill; + *out_len = 0; + return; + } + } + in = hist + refill; + out = *out_len; + data->func(r, (const void**)history, &in, dst, 0, &out); + data->index -= hist; + } + + if (out < *out_len) { + in = *in_len; + s = (const float **)src; + data->func(r, src, &in, dst, out, out_len); + remain = *in_len - in; + } else { + s = (const float **)history; + remain = hist - in; + if (*in_len < n_taps) { + remain += refill; + *in_len = refill; + } else { + *in_len = 0; + } + } + if (remain < n_taps) { + for (c = 0; c < r->channels; c++) + memmove(history[c], &s[c][in], remain * sizeof(float)); + } else { + remain = 0; + *in_len = in; + } + data->index = 0; + data->hist = remain; + return; +} + +static void impl_native_reset (struct resample *r) +{ + struct native_data *d = r->data; + memset(d->hist_mem, 0, r->channels * sizeof(float) * d->n_taps * 2); + d->hist = d->n_taps / 2; + d->index = 0; + d->phase = 0; +} + +static uint32_t impl_native_delay (struct resample *r) +{ + struct native_data *d = r->data; + return d->n_taps; +} + +static inline uint32_t calc_gcd(uint32_t a, uint32_t b) +{ + while (b != 0) { + uint32_t temp = a; + a = b; + b = b % temp; + } + return a; +} + +static int impl_native_init(struct resample *r) +{ + struct native_data *d; + const struct quality *q = &blackman_qualities[DEFAULT_QUALITY]; + double scale; + uint32_t c, n_taps, n_phases, filter_size, in_rate, out_rate, gcd, stride; + uint32_t history_stride, history_size; + + r->free = impl_native_free; + r->update_rate = impl_native_update_rate; + r->process = impl_native_process; + r->reset = impl_native_reset; + r->delay = impl_native_delay; + + gcd = calc_gcd(r->i_rate, r->o_rate); + + in_rate = r->i_rate /= gcd; + out_rate = r->o_rate /= gcd; + + scale = SPA_MIN(q->cutoff * r->o_rate / r->i_rate, 1.0); + n_taps = SPA_ROUND_UP_N((uint32_t)ceil(q->n_taps / scale), 8); + stride = n_taps * sizeof(float); + n_phases = out_rate; + + fprintf(stderr, "in %d out %d %d %d %d %f\n", in_rate, out_rate, gcd, n_taps, n_phases, scale); + + filter_size = stride * (n_phases + 1); + history_stride = 2 * n_taps * sizeof(float); + history_size = r->channels * history_stride; + + d = malloc(sizeof(struct native_data) + + filter_size + + history_size + + (r->channels * sizeof(float*)) + + 32); + + if (d == NULL) + return -ENOMEM; + + r->data = d; + d->rate = 1.0f; + d->filter = SPA_MEMBER(d, sizeof(struct native_data), float); + d->filter = SPA_PTR_ALIGN(d->filter, 16, float); + d->hist_mem = SPA_MEMBER(d->filter, filter_size, float); + d->hist_mem = SPA_PTR_ALIGN(d->hist_mem, 16, float); + d->history = SPA_MEMBER(d->hist_mem, history_size, float*); + for (c = 0; c < r->channels; c++) + d->history[c] = SPA_MEMBER(d->hist_mem, c * history_stride, float); + + d->in_rate = in_rate; + d->out_rate = out_rate; + d->inc = in_rate / out_rate; + d->frac = in_rate % out_rate; + d->n_taps = n_taps; + d->n_phases = n_phases; + + build_filter(d->filter, n_taps, n_taps, n_phases, scale); + + impl_native_reset (r); + + d->func = d->rate == 1.0f ? do_resample_full_c : do_resample_inter_c; +#if defined (__SSE__) + if (r->cpu_flags & SPA_CPU_FLAG_SSE) + d->func = d->rate == 1.0f ? do_resample_full_sse : do_resample_inter_sse; +#endif +#if defined (__SSSE3__) + if (r->cpu_flags & SPA_CPU_FLAG_SSSE3) + d->func = d->rate == 1.0f ? do_resample_full_ssse3 : do_resample_inter_ssse3; +#endif + return 0; +} diff --git a/spa/plugins/audioconvert/resample.c b/spa/plugins/audioconvert/resample.c index 0c930be5c..08f78ae76 100644 --- a/spa/plugins/audioconvert/resample.c +++ b/spa/plugins/audioconvert/resample.c @@ -40,6 +40,7 @@ #include "resample-speex.h" #include "resample-peaks.h" +#include "resample-native.h" #define NAME "resample" @@ -155,6 +156,8 @@ static int setup_convert(struct impl *this, if (this->monitor) err = impl_peaks_init(&this->resample); + else if (1) + err = impl_native_init(&this->resample); else err = impl_speex_init(&this->resample); diff --git a/spa/plugins/audioconvert/resample.h b/spa/plugins/audioconvert/resample.h index 96d738c7e..4f46e6d3d 100644 --- a/spa/plugins/audioconvert/resample.h +++ b/spa/plugins/audioconvert/resample.h @@ -30,12 +30,13 @@ struct resample { uint32_t i_rate; uint32_t o_rate; - void (*free) (struct resample *r); - void (*update_rate) (struct resample *r, double rate); - void (*process) (struct resample *r, - const void * SPA_RESTRICT src[], uint32_t *in_len, - void * SPA_RESTRICT dst[], uint32_t *out_len); - void (*reset) (struct resample *r); + void (*free) (struct resample *r); + void (*update_rate) (struct resample *r, double rate); + void (*process) (struct resample *r, + const void * SPA_RESTRICT src[], uint32_t *in_len, + void * SPA_RESTRICT dst[], uint32_t *out_len); + void (*reset) (struct resample *r); + uint32_t (*delay) (struct resample *r); void *data; }; @@ -43,3 +44,4 @@ struct resample { #define resample_update_rate(r,...) (r)->update_rate(r,__VA_ARGS__) #define resample_process(r,...) (r)->process(r,__VA_ARGS__) #define resample_reset(r) (r)->reset(r) +#define resample_delay(r) (r)->delay(r) diff --git a/spa/plugins/audioconvert/test-resample.c b/spa/plugins/audioconvert/test-resample.c new file mode 100644 index 000000000..0a82060d2 --- /dev/null +++ b/spa/plugins/audioconvert/test-resample.c @@ -0,0 +1,86 @@ +/* Spa + * + * Copyright © 2019 Wim Taymans + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ + +#include +#include +#include +#include +#include +#include + +#include + +#include "resample.h" +#include "resample-native.h" + +#define N_SAMPLES 253 +#define N_CHANNELS 11 + +static float samp_in[N_SAMPLES * 4]; +static float samp_out[N_SAMPLES * 4]; + +static void feed_1(struct resample *r) +{ + uint32_t i; + const void *src[1]; + void *dst[1]; + + spa_zero(samp_out); + src[0] = samp_in; + dst[0] = samp_out; + + for (i = 0; i < 500; i++) { + uint32_t in, out; + + in = out = 1; + samp_in[0] = i; + resample_process(r, src, &in, dst, &out); + fprintf(stderr, "%d %d %f %d\n", i, in, samp_out[0], out); + } +} + +static void test_native(void) +{ + struct resample r; + + r.channels = 1; + r.i_rate = 44100; + r.o_rate = 44100; + + impl_native_init(&r); + + feed_1(&r); + + + + + +} + +int main(int argc, char *argv[]) +{ + test_native(); + + return 0; +}