From 4c4319d334de1f91fa2674770d3a54ed03eb227b Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?The=CC=81o=20Monnom?= Date: Sat, 21 Sep 2024 20:03:45 -0700 Subject: [PATCH 1/2] using soxr finally.. --- Cargo.lock | 2 +- soxr-sys/Cargo.lock | 32 + soxr-sys/Cargo.toml | 14 + soxr-sys/build.rs | 40 + soxr-sys/generate_bindings.sh | 1 + soxr-sys/src/LICENCE | 23 + soxr-sys/src/aliases.h | 39 + soxr-sys/src/avfft32.c | 33 + soxr-sys/src/avfft32s.c | 32 + soxr-sys/src/ccrw2.h | 75 ++ soxr-sys/src/cr-core.c | 314 ++++++ soxr-sys/src/cr.c | 588 ++++++++++ soxr-sys/src/cr.h | 178 +++ soxr-sys/src/cr32.c | 8 + soxr-sys/src/cr32s.c | 8 + soxr-sys/src/cr64.c | 8 + soxr-sys/src/cr64s.c | 8 + soxr-sys/src/data-io.c | 223 ++++ soxr-sys/src/data-io.h | 39 + soxr-sys/src/dbesi0.c | 149 +++ soxr-sys/src/dev32s.h | 54 + soxr-sys/src/dev64s.h | 42 + soxr-sys/src/fft4g.c | 1346 +++++++++++++++++++++++ soxr-sys/src/fft4g.h | 23 + soxr-sys/src/fft4g32.c | 36 + soxr-sys/src/fft4g32s.c | 31 + soxr-sys/src/fft4g64.c | 35 + soxr-sys/src/fft4g_cache.h | 92 ++ soxr-sys/src/fifo.h | 125 +++ soxr-sys/src/filter.c | 277 +++++ soxr-sys/src/filter.h | 44 + soxr-sys/src/half-coefs.h | 75 ++ soxr-sys/src/half-fir.h | 61 ++ soxr-sys/src/internal.h | 84 ++ soxr-sys/src/lib.rs | 152 +++ soxr-sys/src/math-wrap.h | 31 + soxr-sys/src/pffft-avx.h | 40 + soxr-sys/src/pffft-wrap.c | 110 ++ soxr-sys/src/pffft.c | 1946 +++++++++++++++++++++++++++++++++ soxr-sys/src/pffft.h | 197 ++++ soxr-sys/src/pffft32.c | 39 + soxr-sys/src/pffft32s.c | 34 + soxr-sys/src/pffft64s.c | 34 + soxr-sys/src/poly-fir.h | 150 +++ soxr-sys/src/poly-fir0.h | 56 + soxr-sys/src/rdft.h | 31 + soxr-sys/src/rdft_t.h | 24 + soxr-sys/src/rint-clip.h | 158 +++ soxr-sys/src/rint.h | 102 ++ soxr-sys/src/samplerate.h | 1 + soxr-sys/src/soxr-config.h | 28 + soxr-sys/src/soxr-lsr.c | 198 ++++ soxr-sys/src/soxr-lsr.h | 78 ++ soxr-sys/src/soxr.c | 842 ++++++++++++++ soxr-sys/src/soxr.h | 344 ++++++ soxr-sys/src/soxr.rs | 403 +++++++ soxr-sys/src/std-types.h | 48 + soxr-sys/src/util-simd.c | 89 ++ soxr-sys/src/util32s.c | 8 + soxr-sys/src/util32s.h | 23 + soxr-sys/src/util64s.c | 8 + soxr-sys/src/util64s.h | 23 + soxr-sys/src/vr-coefs.c | 115 ++ soxr-sys/src/vr-coefs.h | 94 ++ soxr-sys/src/vr32.c | 651 +++++++++++ 65 files changed, 10195 insertions(+), 1 deletion(-) create mode 100644 soxr-sys/Cargo.lock create mode 100644 soxr-sys/Cargo.toml create mode 100644 soxr-sys/build.rs create mode 100755 soxr-sys/generate_bindings.sh create mode 100644 soxr-sys/src/LICENCE create mode 100644 soxr-sys/src/aliases.h create mode 100644 soxr-sys/src/avfft32.c create mode 100644 soxr-sys/src/avfft32s.c create mode 100644 soxr-sys/src/ccrw2.h create mode 100644 soxr-sys/src/cr-core.c create mode 100644 soxr-sys/src/cr.c create mode 100644 soxr-sys/src/cr.h create mode 100644 soxr-sys/src/cr32.c create mode 100644 soxr-sys/src/cr32s.c create mode 100644 soxr-sys/src/cr64.c create mode 100644 soxr-sys/src/cr64s.c create mode 100644 soxr-sys/src/data-io.c create mode 100644 soxr-sys/src/data-io.h create mode 100644 soxr-sys/src/dbesi0.c create mode 100644 soxr-sys/src/dev32s.h create mode 100644 soxr-sys/src/dev64s.h create mode 100644 soxr-sys/src/fft4g.c create mode 100644 soxr-sys/src/fft4g.h create mode 100644 soxr-sys/src/fft4g32.c create mode 100644 soxr-sys/src/fft4g32s.c create mode 100644 soxr-sys/src/fft4g64.c create mode 100644 soxr-sys/src/fft4g_cache.h create mode 100644 soxr-sys/src/fifo.h create mode 100644 soxr-sys/src/filter.c create mode 100644 soxr-sys/src/filter.h create mode 100644 soxr-sys/src/half-coefs.h create mode 100644 soxr-sys/src/half-fir.h create mode 100644 soxr-sys/src/internal.h create mode 100644 soxr-sys/src/lib.rs create mode 100644 soxr-sys/src/math-wrap.h create mode 100644 soxr-sys/src/pffft-avx.h create mode 100644 soxr-sys/src/pffft-wrap.c create mode 100644 soxr-sys/src/pffft.c create mode 100644 soxr-sys/src/pffft.h create mode 100644 soxr-sys/src/pffft32.c create mode 100644 soxr-sys/src/pffft32s.c create mode 100644 soxr-sys/src/pffft64s.c create mode 100644 soxr-sys/src/poly-fir.h create mode 100644 soxr-sys/src/poly-fir0.h create mode 100644 soxr-sys/src/rdft.h create mode 100644 soxr-sys/src/rdft_t.h create mode 100644 soxr-sys/src/rint-clip.h create mode 100644 soxr-sys/src/rint.h create mode 100644 soxr-sys/src/samplerate.h create mode 100644 soxr-sys/src/soxr-config.h create mode 100644 soxr-sys/src/soxr-lsr.c create mode 100644 soxr-sys/src/soxr-lsr.h create mode 100644 soxr-sys/src/soxr.c create mode 100644 soxr-sys/src/soxr.h create mode 100644 soxr-sys/src/soxr.rs create mode 100644 soxr-sys/src/std-types.h create mode 100644 soxr-sys/src/util-simd.c create mode 100644 soxr-sys/src/util32s.c create mode 100644 soxr-sys/src/util32s.h create mode 100644 soxr-sys/src/util64s.c create mode 100644 soxr-sys/src/util64s.h create mode 100644 soxr-sys/src/vr-coefs.c create mode 100644 soxr-sys/src/vr-coefs.h create mode 100644 soxr-sys/src/vr32.c diff --git a/Cargo.lock b/Cargo.lock index 91bcfced1..4cce07499 100644 --- a/Cargo.lock +++ b/Cargo.lock @@ -1591,7 +1591,7 @@ dependencies = [ [[package]] name = "livekit-ffi" -version = "0.10.1" +version = "0.10.2" dependencies = [ "console-subscriber", "dashmap", diff --git a/soxr-sys/Cargo.lock b/soxr-sys/Cargo.lock new file mode 100644 index 000000000..63cf08bf5 --- /dev/null +++ b/soxr-sys/Cargo.lock @@ -0,0 +1,32 @@ +# This file is automatically @generated by Cargo. +# It is not intended for manual editing. +version = 3 + +[[package]] +name = "cc" +version = "1.1.21" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "07b1695e2c7e8fc85310cde85aeaab7e3097f593c91d209d3f9df76c928100f0" +dependencies = [ + "shlex", +] + +[[package]] +name = "hound" +version = "3.5.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "62adaabb884c94955b19907d60019f4e145d091c75345379e70d1ee696f7854f" + +[[package]] +name = "shlex" +version = "1.3.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "0fda2ff0d084019ba4d7c6f371c95d8fd75ce3524c3cb8fb653a3023f6323e64" + +[[package]] +name = "soxr-rs" +version = "0.1.0" +dependencies = [ + "cc", + "hound", +] diff --git a/soxr-sys/Cargo.toml b/soxr-sys/Cargo.toml new file mode 100644 index 000000000..6661f1e24 --- /dev/null +++ b/soxr-sys/Cargo.toml @@ -0,0 +1,14 @@ +[package] +name = "soxr-sys" +version = "0.1.0" +authors = ["Theo Monnom . + + +Notes + +1. Re software in the `examples' directory: works that are not resampling +examples but are based on the given examples -- for example, applications using +the library -- shall not be considered to be derivative works of the examples. + +2. If building with pffft.c, see the licence embedded in that file. diff --git a/soxr-sys/src/aliases.h b/soxr-sys/src/aliases.h new file mode 100644 index 000000000..d1a392f6e --- /dev/null +++ b/soxr-sys/src/aliases.h @@ -0,0 +1,39 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#if defined SOXR_LIB + +#define lsx_bessel_I_0 _soxr_bessel_I_0 +#define lsx_cdft_f _soxr_cdft_f +#define lsx_cdft _soxr_cdft +#define lsx_clear_fft_cache_f _soxr_clear_fft_cache_f +#define lsx_clear_fft_cache _soxr_clear_fft_cache +#define lsx_ddct_f _soxr_ddct_f +#define lsx_ddct _soxr_ddct +#define lsx_ddst_f _soxr_ddst_f +#define lsx_ddst _soxr_ddst +#define lsx_design_lpf _soxr_design_lpf +#define lsx_dfct_f _soxr_dfct_f +#define lsx_dfct _soxr_dfct +#define lsx_dfst_f _soxr_dfst_f +#define lsx_dfst _soxr_dfst +#define lsx_fir_to_phase _soxr_fir_to_phase +#define lsx_f_resp _soxr_f_resp +#define lsx_init_fft_cache_f _soxr_init_fft_cache_f +#define lsx_init_fft_cache _soxr_init_fft_cache +#define lsx_inv_f_resp _soxr_inv_f_resp +#define lsx_kaiser_beta _soxr_kaiser_beta +#define lsx_kaiser_params _soxr_kaiser_params +#define lsx_make_lpf _soxr_make_lpf +#define lsx_ordered_convolve_f _soxr_ordered_convolve_f +#define lsx_ordered_convolve _soxr_ordered_convolve +#define lsx_ordered_partial_convolve_f _soxr_ordered_partial_convolve_f +#define lsx_ordered_partial_convolve _soxr_ordered_partial_convolve +#define lsx_rdft_f _soxr_rdft_f +#define lsx_rdft _soxr_rdft +#define lsx_safe_cdft_f _soxr_safe_cdft_f +#define lsx_safe_cdft _soxr_safe_cdft +#define lsx_safe_rdft_f _soxr_safe_rdft_f +#define lsx_safe_rdft _soxr_safe_rdft + +#endif diff --git a/soxr-sys/src/avfft32.c b/soxr-sys/src/avfft32.c new file mode 100644 index 000000000..fe651f5db --- /dev/null +++ b/soxr-sys/src/avfft32.c @@ -0,0 +1,33 @@ +/* SoX Resampler Library Copyright (c) 2007-13 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#include +#include +#include +#include "filter.h" +#include "rdft_t.h" + +static void * forward_setup(int len) {return av_rdft_init((int)(log(len)/log(2)+.5),DFT_R2C);} +static void * backward_setup(int len) {return av_rdft_init((int)(log(len)/log(2)+.5),IDFT_C2R);} +static void rdft(int length, void * setup, float * h) {av_rdft_calc(setup, h); (void)length;} +static int multiplier(void) {return 2;} +static void nothing(void) {} +static int flags(void) {return 0;} + +fn_t _soxr_rdft32_cb[] = { + (fn_t)forward_setup, + (fn_t)backward_setup, + (fn_t)av_rdft_end, + (fn_t)rdft, + (fn_t)rdft, + (fn_t)rdft, + (fn_t)rdft, + (fn_t)_soxr_ordered_convolve_f, + (fn_t)_soxr_ordered_partial_convolve_f, + (fn_t)multiplier, + (fn_t)nothing, + (fn_t)malloc, + (fn_t)calloc, + (fn_t)free, + (fn_t)flags, +}; diff --git a/soxr-sys/src/avfft32s.c b/soxr-sys/src/avfft32s.c new file mode 100644 index 000000000..5a7e62db2 --- /dev/null +++ b/soxr-sys/src/avfft32s.c @@ -0,0 +1,32 @@ +/* SoX Resampler Library Copyright (c) 2007-13 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#include +#include +#include "util32s.h" +#include "rdft_t.h" + +static void * forward_setup(int len) {return av_rdft_init((int)(log(len)/log(2)+.5),DFT_R2C);} +static void * backward_setup(int len) {return av_rdft_init((int)(log(len)/log(2)+.5),IDFT_C2R);} +static void rdft(int length, void * setup, float * h) {av_rdft_calc(setup, h); (void)length;} +static int multiplier(void) {return 2;} +static void nothing(void) {} +static int flags(void) {return RDFT_IS_SIMD;} + +fn_t _soxr_rdft32s_cb[] = { + (fn_t)forward_setup, + (fn_t)backward_setup, + (fn_t)av_rdft_end, + (fn_t)rdft, + (fn_t)rdft, + (fn_t)rdft, + (fn_t)rdft, + (fn_t)ORDERED_CONVOLVE_SIMD, + (fn_t)ORDERED_PARTIAL_CONVOLVE_SIMD, + (fn_t)multiplier, + (fn_t)nothing, + (fn_t)SIMD_ALIGNED_MALLOC, + (fn_t)SIMD_ALIGNED_CALLOC, + (fn_t)SIMD_ALIGNED_FREE, + (fn_t)flags, +}; diff --git a/soxr-sys/src/ccrw2.h b/soxr-sys/src/ccrw2.h new file mode 100644 index 000000000..09331a4b1 --- /dev/null +++ b/soxr-sys/src/ccrw2.h @@ -0,0 +1,75 @@ +/* SoX Resampler Library Copyright (c) 2007-13 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +/* Concurrent Control with "Readers" and "Writers", P.J. Courtois et al, 1971 */ + +#if !defined soxr_ccrw2_included +#define soxr_ccrw2_included + +#if defined SOXR_LIB +#include "internal.h" +#endif + +#if defined _OPENMP + +#include + +typedef struct { + int readcount, writecount; /* initial value = 0 */ + omp_lock_t mutex_1, mutex_2, mutex_3, w, r; /* initial value = 1 */ +} ccrw2_t; /* Problem #2: `writers-preference' */ + +#define ccrw2_become_reader(p) do {\ + omp_set_lock(&p.mutex_3);\ + omp_set_lock(&p.r);\ + omp_set_lock(&p.mutex_1);\ + if (++p.readcount == 1) omp_set_lock(&p.w);\ + omp_unset_lock(&p.mutex_1);\ + omp_unset_lock(&p.r);\ + omp_unset_lock(&p.mutex_3);\ +} while (0) +#define ccrw2_cease_reading(p) do {\ + omp_set_lock(&p.mutex_1);\ + if (!--p.readcount) omp_unset_lock(&p.w);\ + omp_unset_lock(&p.mutex_1);\ +} while (0) +#define ccrw2_become_writer(p) do {\ + omp_set_lock(&p.mutex_2);\ + if (++p.writecount == 1) omp_set_lock(&p.r);\ + omp_unset_lock(&p.mutex_2);\ + omp_set_lock(&p.w);\ +} while (0) +#define ccrw2_cease_writing(p) do {\ + omp_unset_lock(&p.w);\ + omp_set_lock(&p.mutex_2);\ + if (!--p.writecount) omp_unset_lock(&p.r);\ + omp_unset_lock(&p.mutex_2);\ +} while (0) +#define ccrw2_init(p) do {\ + omp_init_lock(&p.mutex_1);\ + omp_init_lock(&p.mutex_2);\ + omp_init_lock(&p.mutex_3);\ + omp_init_lock(&p.w);\ + omp_init_lock(&p.r);\ +} while (0) +#define ccrw2_clear(p) do {\ + omp_destroy_lock(&p.r);\ + omp_destroy_lock(&p.w);\ + omp_destroy_lock(&p.mutex_3);\ + omp_destroy_lock(&p.mutex_2);\ + omp_destroy_lock(&p.mutex_1);\ +} while (0) + +#else + +typedef int ccrw2_t; +#define ccrw2_become_reader(x) (void)(x) +#define ccrw2_cease_reading(x) (void)(x) +#define ccrw2_become_writer(x) (void)(x) +#define ccrw2_cease_writing(x) (void)(x) +#define ccrw2_init(x) (void)(x) +#define ccrw2_clear(x) (void)(x) + +#endif /* _OPENMP */ + +#endif diff --git a/soxr-sys/src/cr-core.c b/soxr-sys/src/cr-core.c new file mode 100644 index 000000000..159a5d976 --- /dev/null +++ b/soxr-sys/src/cr-core.c @@ -0,0 +1,314 @@ +/* SoX Resampler Library Copyright (c) 2007-18 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. + * + * Constant-rate resampling engine-specific code. */ + +#include +#include +#include +#include + +#include "filter.h" + +#if defined SOXR_LIB + #include "internal.h" + #include "cr.h" + #if CORE_TYPE & CORE_DBL + typedef double sample_t; + #if CORE_TYPE & CORE_SIMD_DFT + #define RDFT_CB _soxr_rdft64s_cb + #else + #define RDFT_CB _soxr_rdft64_cb + #endif + #else + typedef float sample_t; + #if CORE_TYPE & CORE_SIMD_DFT + #define RDFT_CB _soxr_rdft32s_cb + #else + #define RDFT_CB _soxr_rdft32_cb + #endif + #endif + + #if CORE_TYPE & (CORE_SIMD_POLY|CORE_SIMD_HALF|CORE_SIMD_DFT) + #if CORE_TYPE & CORE_DBL + #include "util64s.h" + #include "dev64s.h" + #else + #include "util32s.h" + #include "dev32s.h" + #endif + #endif + + extern fn_t RDFT_CB[]; +#else + #define RDFT_CB 0 +#endif + + + +static void cubic_stage_fn(stage_t * p, fifo_t * output_fifo) +{ + sample_t const * input = stage_read_p(p); + int num_in = min(stage_occupancy(p), p->input_size); + int i, max_num_out = 1 + (int)(num_in * p->out_in_ratio); + sample_t * output = fifo_reserve(output_fifo, max_num_out); + + for (i = 0; p->at.integer < num_in; ++i, p->at.whole += p->step.whole) { + sample_t const * s = input + p->at.integer; + double x = p->at.fraction * (1 / MULT32); + double b = .5*(s[1]+s[-1])-*s, a = (1/6.)*(s[2]-s[1]+s[-1]-*s-4*b); + double c = s[1]-*s-a-b; + output[i] = (sample_t)(p->mult * (((a*x + b)*x + c)*x + *s)); + } + assert(max_num_out - i >= 0); + fifo_trim_by(output_fifo, max_num_out - i); + fifo_read(&p->fifo, p->at.integer, NULL); + p->at.integer = 0; +} + + + +#if defined __AVX__ + #define DEFINED_AVX 1 +#else + #define DEFINED_AVX 0 +#endif + +#if defined __x86_64__ || defined _M_X64 || defined i386 || defined _M_IX86 + #define DEFINED_X86 1 +#else + #define DEFINED_X86 0 +#endif + +#if defined __arm__ + #define DEFINED_ARM 1 +#else + #define DEFINED_ARM 0 +#endif + + + +#if CORE_TYPE & CORE_DBL + #define SIMD_AVX ((CORE_TYPE & CORE_SIMD_HALF) && DEFINED_AVX) + #define SIMD_SSE 0 +#else + #define SIMD_SSE ((CORE_TYPE & CORE_SIMD_HALF) && DEFINED_X86) + #define SIMD_AVX 0 +#endif + +#define SIMD_NEON ((CORE_TYPE & CORE_SIMD_HALF) && DEFINED_ARM) + + + +#include "half-coefs.h" + +#if !(CORE_TYPE & CORE_SIMD_HALF) +#define FUNCTION_H h7 +#define CONVOLVE ____ __ _ +#include "half-fir.h" +#endif + +#define FUNCTION_H h8 +#define CONVOLVE ____ ____ +#include "half-fir.h" + +#define FUNCTION_H h9 +#define CONVOLVE ____ ____ _ +#include "half-fir.h" + +#if CORE_TYPE & CORE_DBL + #define FUNCTION_H h10 + #define CONVOLVE ____ ____ __ + #include "half-fir.h" + + #define FUNCTION_H h11 + #define CONVOLVE ____ ____ __ _ + #include "half-fir.h" + + #define FUNCTION_H h12 + #define CONVOLVE ____ ____ ____ + #include "half-fir.h" + + #define FUNCTION_H h13 + #define CONVOLVE ____ ____ ____ _ + #include "half-fir.h" +#endif + +static half_fir_info_t const half_firs[] = { +#if !(CORE_TYPE & CORE_SIMD_HALF) + { 7, half_fir_coefs_7 , h7 , 0 , 120.65f}, +#endif + { 8, half_fir_coefs_8 , h8 , 0 , 136.51f}, + { 9, half_fir_coefs_9 , h9 , 0 , 152.32f}, +#if CORE_TYPE & CORE_DBL + {10, half_fir_coefs_10, h10, 0 , 168.08f}, + {11, half_fir_coefs_11, h11, 0 , 183.79f}, + {12, half_fir_coefs_12, h12, 0 , 199.46f}, + {13, half_fir_coefs_13, h13, 0 , 215.12f}, +#endif +}; + +#undef SIMD_AVX +#undef SIMD_NEON +#undef SIMD_SSE + + + +#if CORE_TYPE & CORE_DBL + #define SIMD_AVX ((CORE_TYPE & CORE_SIMD_POLY) && DEFINED_AVX) + #define SIMD_SSE 0 +#else + #define SIMD_SSE ((CORE_TYPE & CORE_SIMD_POLY) && DEFINED_X86) + #define SIMD_AVX 0 +#endif + +#define SIMD_NEON ((CORE_TYPE & CORE_SIMD_POLY) && DEFINED_ARM) + + + +#define COEFS (sample_t * __restrict)p->shared->poly_fir_coefs +#define VAR_LENGTH p->n +#define VAR_CONVOLVE(n) while (j < (n)) _ +#define VAR_POLY_PHASE_BITS p->phase_bits + + + +#define FUNCTION vpoly0 +#define FIR_LENGTH VAR_LENGTH +#define CONVOLVE(n) VAR_CONVOLVE(n) +#include "poly-fir0.h" + +#define FUNCTION vpoly1 +#define COEF_INTERP 1 +#define PHASE_BITS VAR_POLY_PHASE_BITS +#define FIR_LENGTH VAR_LENGTH +#define CONVOLVE(n) VAR_CONVOLVE(n) +#include "poly-fir.h" + +#define FUNCTION vpoly2 +#define COEF_INTERP 2 +#define PHASE_BITS VAR_POLY_PHASE_BITS +#define FIR_LENGTH VAR_LENGTH +#define CONVOLVE(n) VAR_CONVOLVE(n) +#include "poly-fir.h" + +#define FUNCTION vpoly3 +#define COEF_INTERP 3 +#define PHASE_BITS VAR_POLY_PHASE_BITS +#define FIR_LENGTH VAR_LENGTH +#define CONVOLVE(n) VAR_CONVOLVE(n) +#include "poly-fir.h" + + + +#if !(CORE_TYPE & CORE_SIMD_POLY) + +#define poly_fir_convolve_U100 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ +#define FUNCTION U100_0 +#define FIR_LENGTH U100_l +#define CONVOLVE(n) poly_fir_convolve_U100 +#include "poly-fir0.h" + +#define u100_l 11 +#define poly_fir_convolve_u100 _ _ _ _ _ _ _ _ _ _ _ +#define FUNCTION u100_0 +#define FIR_LENGTH u100_l +#define CONVOLVE(n) poly_fir_convolve_u100 +#include "poly-fir0.h" + +#define FUNCTION u100_1 +#define COEF_INTERP 1 +#define PHASE_BITS 8 +#define FIR_LENGTH u100_l +#define CONVOLVE(n) poly_fir_convolve_u100 +#include "poly-fir.h" + +#define FUNCTION u100_2 +#define COEF_INTERP 2 +#define PHASE_BITS 6 +#define FIR_LENGTH u100_l +#define CONVOLVE(n) poly_fir_convolve_u100 +#include "poly-fir.h" + +#endif + +#define u100_1_b 8 +#define u100_2_b 6 + + + +static poly_fir_t const poly_firs[] = { + {-1, {{0, vpoly0}, { 7.2f, vpoly1}, {5.0f, vpoly2}}}, + {-1, {{0, vpoly0}, { 9.4f, vpoly1}, {6.7f, vpoly2}}}, + {-1, {{0, vpoly0}, {12.4f, vpoly1}, {7.8f, vpoly2}}}, + {-1, {{0, vpoly0}, {13.6f, vpoly1}, {9.3f, vpoly2}}}, + {-1, {{0, vpoly0}, {10.5f, vpoly2}, {8.4f, vpoly3}}}, + {-1, {{0, vpoly0}, {11.85f,vpoly2}, {9.0f, vpoly3}}}, + + {-1, {{0, vpoly0}, { 8.0f, vpoly1}, {5.3f, vpoly2}}}, + {-1, {{0, vpoly0}, { 8.6f, vpoly1}, {5.7f, vpoly2}}}, + {-1, {{0, vpoly0}, {10.6f, vpoly1}, {6.75f,vpoly2}}}, + {-1, {{0, vpoly0}, {12.6f, vpoly1}, {8.6f, vpoly2}}}, + {-1, {{0, vpoly0}, { 9.6f, vpoly2}, {7.6f, vpoly3}}}, + {-1, {{0, vpoly0}, {11.4f, vpoly2}, {8.65f,vpoly3}}}, + +#if CORE_TYPE & CORE_SIMD_POLY + {10.62f, {{0, vpoly0}, {0, 0}, {0, 0}}}, + {-1, {{0, vpoly0}, {u100_1_b, vpoly1}, {u100_2_b, vpoly2}}}, +#else + {10.62f, {{U100_l, U100_0}, {0, 0}, {0, 0}}}, + {11.28f, {{u100_l, u100_0}, {u100_1_b, u100_1}, {u100_2_b, u100_2}}}, +#endif + {-1, {{0, vpoly0}, { 9, vpoly1}, { 6, vpoly2}}}, + {-1, {{0, vpoly0}, { 11, vpoly1}, { 7, vpoly2}}}, + {-1, {{0, vpoly0}, { 13, vpoly1}, { 8, vpoly2}}}, + {-1, {{0, vpoly0}, { 10, vpoly2}, { 8, vpoly3}}}, + {-1, {{0, vpoly0}, { 12, vpoly2}, { 9, vpoly3}}}, +}; + + + +static cr_core_t const cr_core = { + +#if CORE_TYPE & CORE_SIMD_POLY + {SIMD_ALIGNED_MALLOC, SIMD_ALIGNED_CALLOC, SIMD_ALIGNED_FREE}, +#else + {malloc, calloc, free}, +#endif + half_firs, array_length(half_firs), + 0, 0, + cubic_stage_fn, + poly_firs, RDFT_CB +}; + + + +#if defined SOXR_LIB + +#include "soxr.h" + +static char const * rate_create(void * channel, void * shared, double io_ratio, + soxr_quality_spec_t * q_spec, soxr_runtime_spec_t * r_spec, double scale) +{ + return _soxr_init(channel, shared, io_ratio, q_spec, r_spec, scale, + &cr_core, CORE_TYPE); +} + + + +static char const * id(void) {return CORE_STR;} + +fn_t RATE_CB[] = { + (fn_t)_soxr_input, + (fn_t)_soxr_process, + (fn_t)_soxr_output, + (fn_t)_soxr_flush, + (fn_t)_soxr_close, + (fn_t)_soxr_delay, + (fn_t)_soxr_sizes, + (fn_t)rate_create, + (fn_t)0, + (fn_t)id, +}; + +#endif diff --git a/soxr-sys/src/cr.c b/soxr-sys/src/cr.c new file mode 100644 index 000000000..4122db3ce --- /dev/null +++ b/soxr-sys/src/cr.c @@ -0,0 +1,588 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. + * + * Constant-rate resampling common code. */ + +#include +#include +#include +#include + +#include "filter.h" + +#if defined SOXR_LIB + #include "internal.h" + #define STATIC +#endif + +#include "cr.h" + +#define num_coefs4 ((core_flags&CORE_SIMD_POLY)? ((num_coefs+3)&~3) : num_coefs) + +#define coef_coef(C,T,x) \ + C((T*)result, interp_order, num_coefs4, j, x, num_coefs4 - 1 - i) + +#define STORE(C,T) { \ + if (interp_order > 2) coef_coef(C,T,3) = (T)d; \ + if (interp_order > 1) coef_coef(C,T,2) = (T)c; \ + if (interp_order > 0) coef_coef(C,T,1) = (T)b; \ + coef_coef(C,T,0) = (T)f0;} + +static real * prepare_poly_fir_coefs(double const * coefs, int num_coefs, + int num_phases, int interp_order, double multiplier, + core_flags_t core_flags, alloc_t const * mem) +{ + int i, j, length = num_coefs4 * num_phases * (interp_order + 1); + real * result = mem->calloc(1,(size_t)length << LOG2_SIZEOF_REAL(core_flags)); + double fm1 = coefs[0], f1 = 0, f2 = 0; + + for (i = num_coefs - 1; i >= 0; --i) + for (j = num_phases - 1; j >= 0; --j) { + double f0 = fm1, b = 0, c = 0, d = 0; /* = 0 to kill compiler warning */ + int pos = i * num_phases + j - 1; + fm1 = pos > 0 ? coefs[pos - 1] * multiplier : 0; + switch (interp_order) { + case 1: b = f1 - f0; break; + case 2: b = f1 - (.5 * (f2+f0) - f1) - f0; c = .5 * (f2+f0) - f1; break; + case 3: c=.5*(f1+fm1)-f0;d=(1/6.)*(f2-f1+fm1-f0-4*c);b=f1-f0-d-c; break; + default: assert(!interp_order); + } + switch (core_flags & 3) { + case 0: if (WITH_CR32 ) STORE(coef , float ); break; + case 1: if (WITH_CR64 ) STORE(coef , double); break; + case 2: if (WITH_CR32S) STORE(coef4, float ); break; + default:if (WITH_CR64S) STORE(coef4, double); break; + } + f2 = f1, f1 = f0; + } + return result; +} + +#undef STORE +#undef coef_coef + +#define IS_FLOAT32 (WITH_CR32 || WITH_CR32S) && \ + (!(WITH_CR64 || WITH_CR64S) || sizeof_real == sizeof(float)) +#define WITH_FLOAT64 WITH_CR64 || WITH_CR64S + +static void dft_stage_fn(stage_t * p, fifo_t * output_fifo) +{ + real * output, * dft_out; + int i, j, num_in = max(0, fifo_occupancy(&p->fifo)); + rate_shared_t const * s = p->shared; + dft_filter_t const * f = &s->dft_filter[p->dft_filter_num]; + int const overlap = f->num_taps - 1; + + if (p->at.integer + p->L * num_in >= f->dft_length) { + fn_t const * const RDFT_CB = p->rdft_cb; + size_t const sizeof_real = sizeof(char) << LOG2_SIZEOF_REAL(p->core_flags); + div_t divd = div(f->dft_length - overlap - p->at.integer + p->L - 1, p->L); + real const * input = fifo_read_ptr(&p->fifo); + fifo_read(&p->fifo, divd.quot, NULL); + num_in -= divd.quot; + + output = fifo_reserve(output_fifo, f->dft_length); + dft_out = (p->core_flags & CORE_SIMD_DFT)? p->dft_out : output; + + if (lsx_is_power_of_2(p->L)) { /* F-domain */ + int portion = f->dft_length / p->L; + memcpy(dft_out, input, (unsigned)portion * sizeof_real); + rdft_oforward(portion, f->dft_forward_setup, dft_out, p->dft_scratch); + if (IS_FLOAT32) { +#define dft_out ((float *)dft_out) + for (i = portion + 2; i < (portion << 1); i += 2) /* Mirror image. */ + dft_out[i] = dft_out[(portion << 1) - i], + dft_out[i+1] = -dft_out[(portion << 1) - i + 1]; + dft_out[portion] = dft_out[1]; + dft_out[portion + 1] = 0; + dft_out[1] = dft_out[0]; +#undef dft_out + } + else if (WITH_FLOAT64) { +#define dft_out ((double *)dft_out) + for (i = portion + 2; i < (portion << 1); i += 2) /* Mirror image. */ + dft_out[i] = dft_out[(portion << 1) - i], + dft_out[i+1] = -dft_out[(portion << 1) - i + 1]; + dft_out[portion] = dft_out[1]; + dft_out[portion + 1] = 0; + dft_out[1] = dft_out[0]; +#undef dft_out + } + + for (portion <<= 1; i < f->dft_length; i += portion, portion <<= 1) { + memcpy((char *)dft_out + (size_t)i * sizeof_real, dft_out, (size_t)portion * sizeof_real); + if (IS_FLOAT32) + #define dft_out ((float *)dft_out) + dft_out[i + 1] = 0; + #undef dft_out + else if (WITH_FLOAT64) + #define dft_out ((double *)dft_out) + dft_out[i + 1] = 0; + #undef dft_out + } + if (p->step.integer > 0) + rdft_reorder_back(f->dft_length, f->dft_backward_setup, dft_out, p->dft_scratch); + } else { + if (p->L == 1) + memcpy(dft_out, input, (size_t)f->dft_length * sizeof_real); + else { + memset(dft_out, 0, (size_t)f->dft_length * sizeof_real); + if (IS_FLOAT32) + for (j = 0, i = p->at.integer; i < f->dft_length; ++j, i += p->L) + ((float *)dft_out)[i] = ((float *)input)[j]; + else if (WITH_FLOAT64) + for (j = 0, i = p->at.integer; i < f->dft_length; ++j, i += p->L) + ((double *)dft_out)[i] = ((double *)input)[j]; + p->at.integer = p->L - 1 - divd.rem; + } + if (p->step.integer > 0) + rdft_forward(f->dft_length, f->dft_forward_setup, dft_out, p->dft_scratch); + else + rdft_oforward(f->dft_length, f->dft_forward_setup, dft_out, p->dft_scratch); + } + + if (p->step.integer > 0) { + rdft_convolve(f->dft_length, f->dft_backward_setup, dft_out, f->coefs); + rdft_backward(f->dft_length, f->dft_backward_setup, dft_out, p->dft_scratch); + if ((p->core_flags & CORE_SIMD_DFT) && p->step.integer == 1) + memcpy(output, dft_out, (size_t)f->dft_length * sizeof_real); + if (p->step.integer != 1) { + if (IS_FLOAT32) + for (j = 0, i = p->remM; i < f->dft_length - overlap; ++j, + i += p->step.integer) + ((float *)output)[j] = ((float *)dft_out)[i]; + else if (WITH_FLOAT64) + for (j = 0, i = p->remM; i < f->dft_length - overlap; ++j, + i += p->step.integer) + ((double *)output)[j] = ((double *)dft_out)[i]; + p->remM = i - (f->dft_length - overlap); + fifo_trim_by(output_fifo, f->dft_length - j); + } + else fifo_trim_by(output_fifo, overlap); + } + else { /* F-domain */ + int m = -p->step.integer; + rdft_convolve_portion(f->dft_length >> m, dft_out, f->coefs); + rdft_obackward(f->dft_length >> m, f->dft_backward_setup, dft_out, p->dft_scratch); + if (p->core_flags & CORE_SIMD_DFT) + memcpy(output, dft_out, (size_t)(f->dft_length >> m) * sizeof_real); + fifo_trim_by(output_fifo, (((1 << m) - 1) * f->dft_length + overlap) >>m); + } + (void)RDFT_CB; + } + p->input_size = (f->dft_length - p->at.integer + p->L - 1) / p->L; +} + +/* Set to 4 x nearest power of 2 or half of that */ +/* if danger of causing too many cache misses. */ +static int set_dft_length(int num_taps, int min, int large) +{ + double d = log((double)num_taps) / log(2.); + return 1 << range_limit((int)(d + 2.77), min, max((int)(d + 1.77), large)); +} + +static void dft_stage_init( + unsigned instance, double Fp, double Fs, double Fn, double att, + double phase_response, stage_t * p, int L, int M, double * multiplier, + unsigned min_dft_size, unsigned large_dft_size, core_flags_t core_flags, + fn_t const * RDFT_CB) +{ + dft_filter_t * f = &p->shared->dft_filter[instance]; + int num_taps = 0, dft_length = f->dft_length, i, offset; + bool f_domain_m = abs(3-M) == 1 && Fs <= 1; + size_t const sizeof_real = sizeof(char) << LOG2_SIZEOF_REAL(core_flags); + + if (!dft_length) { + int k = phase_response == 50 && lsx_is_power_of_2(L) && Fn == L? L << 1 : 4; + double m, * h = lsx_design_lpf(Fp, Fs, Fn, att, &num_taps, -k, -1.); + + if (phase_response != 50) + lsx_fir_to_phase(&h, &num_taps, &f->post_peak, phase_response); + else f->post_peak = num_taps / 2; + + dft_length = set_dft_length(num_taps, (int)min_dft_size, (int)large_dft_size); + f->coefs = rdft_calloc((size_t)dft_length, sizeof_real); + offset = dft_length - num_taps + 1; + m = (1. / dft_length) * rdft_multiplier() * L * *multiplier; + if (IS_FLOAT32) for (i = 0; i < num_taps; ++i) + ((float *)f->coefs)[(i + offset) & (dft_length - 1)] =(float)(h[i] * m); + else if (WITH_FLOAT64) for (i = 0; i < num_taps; ++i) + ((double *)f->coefs)[(i + offset) & (dft_length - 1)] = h[i] * m; + free(h); + } + + if (rdft_flags() & RDFT_IS_SIMD) + p->dft_out = rdft_malloc(sizeof_real * (size_t)dft_length); + if (rdft_flags() & RDFT_NEEDS_SCRATCH) + p->dft_scratch = rdft_malloc(2 * sizeof_real * (size_t)dft_length); + + if (!f->dft_length) { + void * coef_setup = rdft_forward_setup(dft_length); + int Lp = lsx_is_power_of_2(L)? L : 1; + int Mp = f_domain_m? M : 1; + f->dft_forward_setup = rdft_forward_setup(dft_length / Lp); + f->dft_backward_setup = rdft_backward_setup(dft_length / Mp); + if (Mp == 1) + rdft_forward(dft_length, coef_setup, f->coefs, p->dft_scratch); + else + rdft_oforward(dft_length, coef_setup, f->coefs, p->dft_scratch); + rdft_delete_setup(coef_setup); + f->num_taps = num_taps; + f->dft_length = dft_length; + lsx_debug("fir_len=%i dft_length=%i Fp=%g Fs=%g Fn=%g att=%g %i/%i", + num_taps, dft_length, Fp, Fs, Fn, att, L, M); + } + *multiplier = 1; + p->out_in_ratio = (double)L / M; + p->core_flags = core_flags; + p->rdft_cb = RDFT_CB; + p->fn = dft_stage_fn; + p->preload = f->post_peak / L; + p->at.integer = f->post_peak % L; + p->L = L; + p->step.integer = f_domain_m? -M/2 : M; + p->dft_filter_num = instance; + p->block_len = f->dft_length - (f->num_taps - 1); + p->phase0 = p->at.integer / p->L; + p->input_size = (f->dft_length - p->at.integer + p->L - 1) / p->L; +} + +static struct half_fir_info const * find_half_fir( + struct half_fir_info const * firs, size_t len, double att) +{ + size_t i; + for (i = 0; i + 1 < len && att > firs[i].att; ++i); + return &firs[i]; +} + +#define have_pre_stage (preM * preL != 1) +#define have_arb_stage (arbM * arbL != 1) +#define have_post_stage (postM * postL != 1) + +#include "soxr.h" + +STATIC char const * _soxr_init( + rate_t * const p, /* Per audio channel. */ + rate_shared_t * const shared, /* By channels undergoing same rate change. */ + double const io_ratio, /* Input rate divided by output rate. */ + soxr_quality_spec_t const * const q_spec, + soxr_runtime_spec_t const * const r_spec, + double multiplier, /* Linear gain to apply during conversion. */ + cr_core_t const * const core, + core_flags_t const core_flags) +{ + size_t const sizeof_real = sizeof(char) << LOG2_SIZEOF_REAL(core_flags); + double const tolerance = 1 + 1e-5; + + double bits = q_spec->precision; + rolloff_t const rolloff = (rolloff_t)(q_spec->flags & 3); + int interpolator = (int)(r_spec->flags & 3) - 1; + double const Fp0 = q_spec->passband_end, Fs0 = q_spec->stopband_begin; + double const phase_response = q_spec->phase_response, tbw0 = Fs0-Fp0; + + bool const maintain_3dB_pt = !!(q_spec->flags & SOXR_MAINTAIN_3DB_PT); + double tbw_tighten = 1, alpha; + #define tighten(x) (Fs0-(Fs0-(x))*tbw_tighten) + + double arbM = io_ratio, Fn1, Fp1 = Fp0, Fs1 = Fs0, bits1 = min(bits,33); + double att = (bits1 + 1) * linear_to_dB(2.), attArb = att; /* +1: pass+stop */ + int preL = 1, preM = 1, shr = 0, arbL = 1, postL = 1, postM = 1; + bool upsample=false, rational=false, iOpt=!(r_spec->flags&SOXR_NOSMALLINTOPT); + bool lq_bits= (q_spec->flags & SOXR_PROMOTE_TO_LQ)? bits <= 16 : bits == 16; + bool lq_Fp0 = (q_spec->flags & SOXR_PROMOTE_TO_LQ)? Fp0<=lq_bw0 : Fp0==lq_bw0; + int n = 0, i, mode = lq_bits && rolloff == rolloff_medium? io_ratio > 1 || + phase_response != 50 || !lq_Fp0 || Fs0 != 1 : ((int)ceil(bits1) - 6) / 4; + struct half_fir_info const * half_fir_info; + stage_t * s; + + if (io_ratio < 1 && Fs0 - 1 > 1 - Fp0 / tolerance) + return "imaging greater than rolloff"; + if (.002 / tolerance > tbw0 || tbw0 > .5 * tolerance) + return "transition bandwidth not in [0.2,50] % of nyquist"; + if (.5 / tolerance > Fp0 || Fs0 > 1.5 * tolerance) + return "transition band not within [50,150] % of nyquist"; + if (bits!=0 && (15 > bits || bits > 33)) + return "precision not in [15,33] bits"; + if (io_ratio <= 0) + return "resampling factor not positive"; + if (0 > phase_response || phase_response > 100) + return "phase response not in [0=min-phase,100=max-phase] %"; + + p->core = core; + p->io_ratio = io_ratio; + if (bits!=0) while (!n++) { /* Determine stages: */ + int try, L, M, x, maxL = interpolator > 0? 1 : mode? 2048 : + (int)ceil(r_spec->coef_size_kbytes * 1000. / (U100_l * (int)sizeof_real)); + double d, epsilon = 0, frac; + upsample = arbM < 1; + for (i = (int)(.5 * arbM), shr = 0; i >>= 1; arbM *= .5, ++shr); + preM = upsample || (arbM > 1.5 && arbM < 2); + postM = 1 + (arbM > 1 && preM), arbM /= postM; + preL = 1 + (!preM && arbM < 2) + (upsample && mode), arbM *= preL; + if ((frac = arbM - (int)arbM)!=0) + epsilon = fabs(floor(frac * MULT32 + .5) / (frac * MULT32) - 1); + for (i = 1, rational = frac==0; i <= maxL && !rational; ++i) { + d = frac * i, try = (int)(d + .5); + if ((rational = fabs(try / d - 1) <= epsilon)) { /* No long doubles! */ + if (try == i) + arbM = ceil(arbM), shr += x = arbM > 3, arbM /= 1 + x; + else arbM = i * (int)arbM + try, arbL = i; + } + } + L = preL * arbL, M = (int)(arbM * postM), x = (L|M)&1, L >>= !x, M >>= !x; + if (iOpt && postL == 1 && (d = preL * arbL / arbM) > 4 && d != 5) { + for (postL = 4, i = (int)(d / 16); (i >>= 1) && postL < 256; postL <<= 1); + arbM = arbM * postL / arbL / preL, arbL = 1, n = 0; + } else if (rational && (max(L, M) < 3 + 2 * iOpt || L * M < 6 * iOpt)) + preL = L, preM = M, arbM = arbL = postM = 1; + if (!mode && (!rational || !n)) + ++mode, n = 0; + } + + p->num_stages = shr + have_pre_stage + have_arb_stage + have_post_stage; + if (!p->num_stages && multiplier != 1) { + bits = arbL = 0; /* Use cubic_stage in this case. */ + ++p->num_stages; + } + p->stages = calloc((size_t)p->num_stages + 1, sizeof(*p->stages)); + if (!p->stages) + return "out of memory"; + for (i = 0; i < p->num_stages; ++i) { + p->stages[i].num = i; + p->stages[i].shared = shared; + p->stages[i].input_size = 8192; + } + p->stages[0].is_input = true; + + alpha = postM / (io_ratio * (postL << 0)); + + if ((n = p->num_stages) > 1) { /* Att. budget: */ + if (have_arb_stage) + att += linear_to_dB(2.), attArb = att, --n; + att += linear_to_dB((double)n); + } + + half_fir_info = find_half_fir(core->half_firs, core->half_firs_len, att); + for (i = 0, s = p->stages; i < shr; ++i, ++s) { + s->fn = half_fir_info->fn; + s->coefs = half_fir_info->coefs; + s->n = half_fir_info->num_coefs; + s->pre_post = 4 * s->n; + s->preload = s->pre = s->pre_post >> 1; + } + + if (have_pre_stage) { + if (maintain_3dB_pt && have_post_stage) { /* Trans. bands overlapping. */ + double x = tbw0 * lsx_inv_f_resp(-3., att); + x = -lsx_f_resp(x / (max(2 * alpha - Fs0, alpha) - Fp0), att); + if (x > .035) { + tbw_tighten = ((4.3074e-3 - 3.9121e-4 * x) * x - .040009) * x + 1.0014; + lsx_debug("tbw_tighten=%g (%gdB)", tbw_tighten, x); + } + } + Fn1 = preM? max(preL, preM) : arbM / arbL; + dft_stage_init(0, tighten(Fp1), Fs1, Fn1, att, phase_response, s++, preL, + max(preM, 1), &multiplier, r_spec->log2_min_dft_size, + r_spec->log2_large_dft_size, core_flags, core->rdft_cb); + Fp1 /= Fn1, Fs1 /= Fn1; + } + + if (bits==0 && have_arb_stage) { /* `Quick' cubic arb stage: */ + s->fn = core->cubic_stage_fn; + s->mult = multiplier, multiplier = 1; + s->step.whole = (int64_t)(arbM * MULT32 + .5); + s->pre_post = max(3, s->step.integer); + s->preload = s->pre = 1; + s->out_in_ratio = MULT32 / (double)s->step.whole; + } + else if (have_arb_stage) { /* Higher quality arb stage: */ + static const float rolloffs[] = {-.01f, -.3f, 0, -.103f}; + poly_fir_t const * f = &core->poly_firs[6*(upsample+!!preM)+mode-!upsample]; + int order, num_coefs = (int)f->interp[0].scalar, phase_bits, phases; + size_t coefs_size; + double at, Fp = Fp1, Fs, Fn, mult = upsample? 1 : arbM / arbL; + poly_fir1_t const * f1; + + if (!upsample && preM) + Fn = 2 * mult, Fs = 3 + fabs(Fs1 - 1); + else Fn = 1, Fs = 2 - (mode? Fp1 + (Fs1 - Fp1) * .7 : Fs1); + + if (mode) + Fp = Fs - (Fs - Fp) / (1 - lsx_inv_f_resp(rolloffs[rolloff], attArb)); + + i = (interpolator < 0? !rational : max(interpolator, !rational)) - 1; + do { + f1 = &f->interp[++i]; + assert(f1->fn); + if (i) + arbM /= arbL, arbL = 1, rational = false; + phase_bits = (int)ceil(f1->scalar - log(mult)/log(2.)); + phases = !rational? (1 << phase_bits) : arbL; + if (f->interp[0].scalar==0) { + int phases0 = max(phases, 19), n0 = 0; + lsx_design_lpf(Fp, Fs, -Fn, attArb, &n0, phases0, f->beta); + num_coefs = n0 / phases0 + 1, num_coefs += num_coefs & !preM; + } + if ((num_coefs & 1) && rational && (arbL & 1)) + phases <<= 1, arbL <<= 1, arbM *= 2; + at = arbL * (s->phase0 = .5 * (num_coefs & 1)); + order = i + (i && mode > 4); + coefs_size = (size_t)(num_coefs4 * phases * (order+1)) * sizeof_real; + } while (interpolator < 0 && i < 2 && f->interp[i+1].fn && + coefs_size / 1000 > r_spec->coef_size_kbytes); + + if (!s->shared->poly_fir_coefs) { + int num_taps = num_coefs * phases - 1; + double * coefs = lsx_design_lpf( + Fp, Fs, Fn, attArb, &num_taps, phases, f->beta); + s->shared->poly_fir_coefs = prepare_poly_fir_coefs( + coefs, num_coefs, phases, order, multiplier, core_flags, &core->mem); + lsx_debug("fir_len=%i phases=%i coef_interp=%i size=%.3gk", + num_coefs, phases, order, (double)coefs_size / 1000.); + free(coefs); + } + multiplier = 1; + s->fn = f1->fn; + s->pre_post = num_coefs4 - 1; + s->preload = ((num_coefs - 1) >> 1) + (num_coefs4 - num_coefs); + s->n = num_coefs4; + s->phase_bits = phase_bits; + s->L = arbL; + s->use_hi_prec_clock = + mode>1 && (q_spec->flags & SOXR_HI_PREC_CLOCK) && !rational; +#if WITH_FLOAT_STD_PREC_CLOCK + if (order && !s->use_hi_prec_clock) { + s->at.flt = at; + s->step.flt = arbM; + s->out_in_ratio = (double)(arbL / s->step.flt); + } else +#endif + { + s->at.whole = (int64_t)(at * MULT32 + .5); +#if WITH_HI_PREC_CLOCK + if (s->use_hi_prec_clock) { + double M = arbM * MULT32; + s->at.fix.ls.parts.ms = 0x80000000ul; + s->step.whole = (int64_t)M; + M -= (double)s->step.whole; + M *= MULT32 * MULT32; + s->step.fix.ls.all = (uint64_t)M; + } else +#endif + s->step.whole = (int64_t)(arbM * MULT32 + .5); + s->out_in_ratio = MULT32 * arbL / (double)s->step.whole; + } + ++s; + } + + if (have_post_stage) + dft_stage_init(1, tighten(Fp0 / (upsample? alpha : 1)), upsample? max(2 - + Fs0 / alpha, 1) : Fs0, (double)max(postL, postM), att, phase_response, + s++, postL, postM, &multiplier, r_spec->log2_min_dft_size, + r_spec->log2_large_dft_size, core_flags, core->rdft_cb); + + lsx_debug("%g: >>%i %i/%i %i/%g %i/%i (%x)", 1/io_ratio, + shr, preL, preM, arbL, arbM, postL, postM, core_flags); + + for (i = 0, s = p->stages; i < p->num_stages; ++i, ++s) { + fifo_create(&s->fifo, (int)sizeof_real); + memset(fifo_reserve(&s->fifo, s->preload), 0, + sizeof_real * (size_t)s->preload); + lsx_debug_more("%5i|%-5i preload=%i remL=%i", + s->pre, s->pre_post-s->pre, s->preload, s->at.integer); + } + fifo_create(&s->fifo, (int)sizeof_real); + return 0; +} + +static bool stage_process(stage_t * stage, bool flushing) +{ + fifo_t * fifo = &stage->fifo; + bool done = false; + int want; + while (!done && (want = stage->input_size - fifo_occupancy(fifo)) > 0) { + if (stage->is_input) { + if (flushing) + memset(fifo_reserve(fifo, want), 0, fifo->item_size * (size_t)want); + else done = true; + } + else done = stage_process(stage - 1, flushing); + } + stage->fn(stage, &stage[1].fifo); + return done && fifo_occupancy(fifo) < stage->input_size; +} + +STATIC void _soxr_process(rate_t * p, size_t olen) +{ + int const n = p->flushing? min(-(int)p->samples_out, (int)olen) : (int)olen; + stage_t * stage = &p->stages[p->num_stages]; + fifo_t * fifo = &stage->fifo; + bool done = false; + while (!done && fifo_occupancy(fifo) < (int)n) + done = stage->is_input || stage_process(stage - 1, p->flushing); +} + +STATIC real * _soxr_input(rate_t * p, real const * samples, size_t n) +{ + if (p->flushing) + return 0; + p->samples_in += (int64_t)n; + return fifo_write(&p->stages[0].fifo, (int)n, samples); +} + +STATIC real const * _soxr_output(rate_t * p, real * samples, size_t * n0) +{ + fifo_t * fifo = &p->stages[p->num_stages].fifo; + int n = p->flushing? min(-(int)p->samples_out, (int)*n0) : (int)*n0; + p->samples_out += n = min(n, fifo_occupancy(fifo)); + return fifo_read(fifo, (int)(*n0 = (size_t)n), samples); +} + +STATIC void _soxr_flush(rate_t * p) +{ + if (p->flushing) return; + p->samples_out -= (int64_t)((double)p->samples_in / p->io_ratio + .5); + p->samples_in = 0; + p->flushing = true; +} + +STATIC void _soxr_close(rate_t * p) +{ + if (p->stages) { + fn_t const * const RDFT_CB = p->core->rdft_cb; + rate_shared_t * shared = p->stages[0].shared; + int i; + + for (i = 0; i <= p->num_stages; ++i) { + stage_t * s = &p->stages[i]; + rdft_free(s->dft_scratch); + rdft_free(s->dft_out); + fifo_delete(&s->fifo); + } + if (shared) { + for (i = 0; i < 2; ++i) { + dft_filter_t * f= &shared->dft_filter[i]; + rdft_free(f->coefs); + rdft_delete_setup(f->dft_forward_setup); + rdft_delete_setup(f->dft_backward_setup); + } + p->core->mem.free(shared->poly_fir_coefs); + memset(shared, 0, sizeof(*shared)); + } + free(p->stages); + (void)RDFT_CB; + } +} + +#if defined SOXR_LIB +STATIC double _soxr_delay(rate_t * p) +{ + return (double)p->samples_in / p->io_ratio - (double)p->samples_out; +} + +STATIC void _soxr_sizes(size_t * shared, size_t * channel) +{ + *shared = sizeof(rate_shared_t); + *channel = sizeof(rate_t); +} +#endif diff --git a/soxr-sys/src/cr.h b/soxr-sys/src/cr.h new file mode 100644 index 000000000..d6e863799 --- /dev/null +++ b/soxr-sys/src/cr.h @@ -0,0 +1,178 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#if !defined soxr_cr_included +#define soxr_cr_included + +#define FIFO_SIZE_T int +#include "fifo.h" + +typedef void real; /* float or double */ +struct stage; +typedef void (* stage_fn_t)(struct stage * input, fifo_t * output); +typedef struct half_fir_info { + int num_coefs; + real const * coefs; + stage_fn_t fn, dfn; + float att; +} half_fir_info_t; +typedef struct {float scalar; stage_fn_t fn;} poly_fir1_t; +typedef struct {float beta; poly_fir1_t interp[3];} poly_fir_t; + +#define U100_l 42 +#define MULT32 (65536. * 65536.) + +/* Conceptually: coef_p is &coefs[num_phases][fir_len][interp_order+1]: */ +#define coef(coef_p, interp_order, fir_len, phase_num, coef_interp_num, fir_coef_num) (coef_p)[\ + (fir_len) * ((interp_order) + 1) * (phase_num) + \ + ((interp_order) + 1) * (fir_coef_num) + \ + ((interp_order) - (coef_interp_num))] + +/* Conceptually: coef_p is &coefs[num_phases][fir_len/4][interp_order+1][4]: */ +#define coef4(coef_p, interp_order, fir_len, phase_num, coef_interp_num, fir_coef_num) (coef_p)[\ + (fir_len) * ((interp_order) + 1) * (phase_num) + \ + ((interp_order) + 1) * ((fir_coef_num) & ~3) + \ + 4 * ((interp_order) - (coef_interp_num)) + \ + ((fir_coef_num) & 3)] + +typedef union { /* Int64 in parts */ + #if HAVE_BIGENDIAN + struct {int32_t ms; uint32_t ls;} parts; + #else + struct {uint32_t ls; int32_t ms;} parts; + #endif + int64_t all; +} int64p_t; + +typedef union { /* Uint64 in parts */ + #if HAVE_BIGENDIAN + struct {uint32_t ms, ls;} parts; + #else + struct {uint32_t ls, ms;} parts; + #endif + uint64_t all; +} uint64p_t; + +typedef struct { + int dft_length, num_taps, post_peak; + void * dft_forward_setup, * dft_backward_setup; + real * coefs; +} dft_filter_t; + +typedef struct { /* So generated filter coefs may be shared between channels */ + real * poly_fir_coefs; + dft_filter_t dft_filter[2]; +} rate_shared_t; + +typedef double float_step_t; /* Or long double or __float128. */ + +typedef union { /* Fixed point arithmetic */ + struct {uint64p_t ls; int64p_t ms;} fix; /* Hi-prec has ~96 bits. */ + float_step_t flt; +} step_t; + +#define integer fix.ms.parts.ms +#define fraction fix.ms.parts.ls +#define whole fix.ms.all + +#define CORE_DBL 1 +#define CORE_SIMD_POLY 2 +#define CORE_SIMD_HALF 4 +#define CORE_SIMD_DFT 8 +#define LOG2_SIZEOF_REAL(core_flags) (2 + ((core_flags) & 1)) + +typedef int core_flags_t; + +#if defined SOXR_LIB +#include "rdft_t.h" +#else +typedef void fn_t; +#endif + +typedef struct stage { + int num; + + /* Common to all stage types: */ + core_flags_t core_flags; + stage_fn_t fn; + fifo_t fifo; + int pre; /* Number of past samples to store */ + int pre_post; /* pre + number of future samples to store */ + int preload; /* Number of zero samples to pre-load the fifo */ + double out_in_ratio; /* For buffer management. */ + int input_size; + bool is_input; + + /* For a stage with variable (run-time generated) filter coefs: */ + fn_t const * rdft_cb; + rate_shared_t * shared; + unsigned dft_filter_num; /* Which, if any, of the 2 DFT filters to use */ + real * dft_scratch; + float * dft_out; + real const * coefs; + + /* For a stage with variable L/M: */ + step_t at, step; + bool use_hi_prec_clock; + int L, remM; + int n, phase_bits, block_len; + double mult, phase0; +} stage_t; + +#define stage_occupancy(s) max(0, fifo_occupancy(&(s)->fifo) - (s)->pre_post) +#define stage_read_p(s) ((sample_t *)fifo_read_ptr(&(s)->fifo) + (s)->pre) + +#define lq_bw0 (1385/2048.) /* ~.67625, FP exact. */ + +typedef enum {rolloff_small, rolloff_medium, rolloff_none} rolloff_t; + +typedef struct { + void * (* alloc)(size_t); + void * (* calloc)(size_t, size_t); + void (* free)(void *); +} alloc_t; + +typedef struct { + alloc_t mem; + half_fir_info_t const * half_firs; + size_t half_firs_len; + half_fir_info_t const * doub_firs; + size_t doub_firs_len; + stage_fn_t cubic_stage_fn; + poly_fir_t const * poly_firs; + fn_t * rdft_cb; +} cr_core_t; + +typedef struct rate rate_t; +struct rate { + cr_core_t const * core; + double io_ratio; + int64_t samples_in, samples_out; + int num_stages, flushing; + stage_t * stages; +}; + +#if defined SOXR_LIB + +#include "soxr.h" + +char const * _soxr_init( + rate_t * const p, /* Per audio channel. */ + rate_shared_t * const shared, /* Between channels (undergoing same rate change)*/ + double const io_ratio, /* Input rate divided by output rate. */ + soxr_quality_spec_t const * const q_spec, + soxr_runtime_spec_t const * const r_spec, + double multiplier, /* Linear gain to apply during conversion. 1 */ + cr_core_t const * const core, + core_flags_t const); + +void _soxr_process(struct rate * p, size_t olen); +real * _soxr_input(struct rate * p, real const * samples, size_t n); +real const * _soxr_output(struct rate * p, real * samples, size_t * n0); +void _soxr_flush(struct rate * p); +void _soxr_close(struct rate * p); +double _soxr_delay(struct rate * p); +void _soxr_sizes(size_t * shared, size_t * channel); +#endif + +#endif diff --git a/soxr-sys/src/cr32.c b/soxr-sys/src/cr32.c new file mode 100644 index 000000000..b9eb264d0 --- /dev/null +++ b/soxr-sys/src/cr32.c @@ -0,0 +1,8 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#define RATE_CB _soxr_rate32_cb +#define CORE_STR "cr32" + +#define CORE_TYPE 0 +#include "cr-core.c" diff --git a/soxr-sys/src/cr32s.c b/soxr-sys/src/cr32s.c new file mode 100644 index 000000000..5de2a4336 --- /dev/null +++ b/soxr-sys/src/cr32s.c @@ -0,0 +1,8 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#define RATE_CB _soxr_rate32s_cb +#define CORE_STR "cr32s" + +#define CORE_TYPE (CORE_SIMD_POLY|CORE_SIMD_HALF|CORE_SIMD_DFT) +#include "cr-core.c" diff --git a/soxr-sys/src/cr64.c b/soxr-sys/src/cr64.c new file mode 100644 index 000000000..518cdd761 --- /dev/null +++ b/soxr-sys/src/cr64.c @@ -0,0 +1,8 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#define RATE_CB _soxr_rate64_cb +#define CORE_STR "cr64" + +#define CORE_TYPE CORE_DBL +#include "cr-core.c" diff --git a/soxr-sys/src/cr64s.c b/soxr-sys/src/cr64s.c new file mode 100644 index 000000000..5dcd6f100 --- /dev/null +++ b/soxr-sys/src/cr64s.c @@ -0,0 +1,8 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#define RATE_CB _soxr_rate64s_cb +#define CORE_STR "cr64s" + +#define CORE_TYPE (CORE_DBL|CORE_SIMD_POLY|CORE_SIMD_HALF|CORE_SIMD_DFT) +#include "cr-core.c" diff --git a/soxr-sys/src/data-io.c b/soxr-sys/src/data-io.c new file mode 100644 index 000000000..fb6167583 --- /dev/null +++ b/soxr-sys/src/data-io.c @@ -0,0 +1,223 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#include +#include +#include + +#include "data-io.h" +#include "internal.h" + + + +#define DEINTERLEAVE_FROM(T,flag) do { \ + unsigned i; \ + size_t j; \ + T const * src = *src0; \ + if (ch > 1) for (j = 0; j < n; ++j) \ + for (i = 0; i < ch; ++i) dest[i][j] = (DEINTERLEAVE_TO)*src++; \ + else if (flag) memcpy(dest[0], src, n * sizeof(T)), src = &src[n]; \ + else for (j = 0; j < n; dest[0][j++] = (DEINTERLEAVE_TO)*src++); \ + *src0 = src; \ +} while (0) + + + +#if WITH_CR64 || WITH_CR64S +void _soxr_deinterleave(double * * dest, /* Round/clipping not needed here */ + soxr_datatype_t data_type, void const * * src0, size_t n, unsigned ch) +{ +#define DEINTERLEAVE_TO double + switch (data_type & 3) { + case SOXR_FLOAT32: DEINTERLEAVE_FROM(float, 0); break; + case SOXR_FLOAT64: DEINTERLEAVE_FROM(double, 1); break; + case SOXR_INT32: DEINTERLEAVE_FROM(int32_t, 0); break; + case SOXR_INT16: DEINTERLEAVE_FROM(int16_t, 0); break; + default: break; + } +} +#endif + + + +#if WITH_CR32 || WITH_CR32S || WITH_VR32 +void _soxr_deinterleave_f(float * * dest, /* Round/clipping not needed here */ + soxr_datatype_t data_type, void const * * src0, size_t n, unsigned ch) +{ +#undef DEINTERLEAVE_TO +#define DEINTERLEAVE_TO float + switch (data_type & 3) { + case SOXR_FLOAT32: DEINTERLEAVE_FROM(float, 1); break; + case SOXR_FLOAT64: DEINTERLEAVE_FROM(double, 0); break; + case SOXR_INT32: DEINTERLEAVE_FROM(int32_t, 0); break; + case SOXR_INT16: DEINTERLEAVE_FROM(int16_t, 0); break; + default: break; + } +} +#endif + + + +#include "rint.h" + + + +#if defined FE_INVALID && defined FPU_RINT32 && defined __STDC_VERSION__ + #if __STDC_VERSION__ >= 199901L + #pragma STDC FENV_ACCESS ON + #endif +#endif + +#if WITH_CR64 || WITH_CR64S +#define FLOATX double + +#define LSX_RINT_CLIP_2 lsx_rint32_clip_2 +#define LSX_RINT_CLIP lsx_rint32_clip +#define RINT_CLIP rint32_clip +#define RINT rint32D +#if defined FPU_RINT32 + #define FPU_RINT +#endif +#define RINT_T int32_t +#define RINT_MAX 2147483647L +#include "rint-clip.h" + +#define LSX_RINT_CLIP_2 lsx_rint16_clip_2 +#define LSX_RINT_CLIP lsx_rint16_clip +#define RINT_CLIP rint16_clip +#define RINT rint16D +#if defined FPU_RINT16 + #define FPU_RINT +#endif +#define RINT_T int16_t +#define RINT_MAX 32767 +#include "rint-clip.h" + +#define LSX_RINT_CLIP_2 lsx_rint16_clip_2_dither +#define LSX_RINT_CLIP lsx_rint16_clip_dither +#define RINT_CLIP rint16_clip_dither +#define RINT rint16D +#if defined FPU_RINT16 + #define FPU_RINT +#endif +#define RINT_T int16_t +#define RINT_MAX 32767 +#define DITHER +#include "rint-clip.h" + +#undef FLOATX +#endif + + + +#if WITH_CR32 || WITH_CR32S || WITH_VR32 +#define FLOATX float + +#define LSX_RINT_CLIP_2 lsx_rint32_clip_2_f +#define LSX_RINT_CLIP lsx_rint32_clip_f +#define RINT_CLIP rint32_clip_f +#define RINT rint32F +#if defined FPU_RINT32 + #define FPU_RINT +#endif +#define RINT_T int32_t +#define RINT_MAX 2147483647L +#include "rint-clip.h" + +#define LSX_RINT_CLIP_2 lsx_rint16_clip_2_f +#define LSX_RINT_CLIP lsx_rint16_clip_f +#define RINT_CLIP rint16_clip_f +#define RINT rint16F +#if defined FPU_RINT16 + #define FPU_RINT +#endif +#define RINT_T int16_t +#define RINT_MAX 32767 +#include "rint-clip.h" + +#define LSX_RINT_CLIP_2 lsx_rint16_clip_2_dither_f +#define LSX_RINT_CLIP lsx_rint16_clip_dither_f +#define RINT_CLIP rint16_clip_dither_f +#define RINT rint16D +#if defined FPU_RINT16 + #define FPU_RINT +#endif +#define RINT_T int16_t +#define RINT_MAX 32767 +#define DITHER +#include "rint-clip.h" + +#undef FLOATX +#endif + +#if defined FE_INVALID && defined FPU_RINT32 && defined __STDC_VERSION__ + #if __STDC_VERSION__ >= 199901L + #pragma STDC FENV_ACCESS OFF + #endif +#endif + + + +#define INTERLEAVE_TO(T,flag) do { \ + unsigned i; \ + size_t j; \ + T * dest = *dest0; \ + if (ch > 1) \ + for (j = 0; j < n; ++j) for (i = 0; i < ch; ++i) *dest++ = (T)src[i][j]; \ + else if (flag) memcpy(dest, src[0], n * sizeof(T)), dest = &dest[n]; \ + else for (j = 0; j < n; *dest++ = (T)src[0][j++]); \ + *dest0 = dest; \ + return 0; \ +} while (0) + +#if WITH_CR64 || WITH_CR64S +size_t /* clips */ _soxr_interleave(soxr_datatype_t data_type, void * * dest0, + double const * const * src, size_t n, unsigned ch, unsigned long * seed) +{ + switch (data_type & 3) { + case SOXR_FLOAT32: INTERLEAVE_TO(float, 0); + case SOXR_FLOAT64: INTERLEAVE_TO(double, 1); + + case SOXR_INT32: if (ch == 1) + return lsx_rint32_clip(dest0, src[0], n); + return lsx_rint32_clip_2(dest0, src, ch, n); + + case SOXR_INT16: if (seed) { + if (ch == 1) + return lsx_rint16_clip_dither(dest0, src[0], n, seed); + return lsx_rint16_clip_2_dither(dest0, src, ch, n, seed); + } + if (ch == 1) + return lsx_rint16_clip(dest0, src[0], n); + return lsx_rint16_clip_2(dest0, src, ch, n); + default: break; + } + return 0; +} +#endif + +#if WITH_CR32 || WITH_CR32S || WITH_VR32 +size_t /* clips */ _soxr_interleave_f(soxr_datatype_t data_type, void * * dest0, + float const * const * src, size_t n, unsigned ch, unsigned long * seed) +{ + switch (data_type & 3) { + case SOXR_FLOAT32: INTERLEAVE_TO(float, 1); + case SOXR_FLOAT64: INTERLEAVE_TO(double, 0); + + case SOXR_INT32: if (ch == 1) + return lsx_rint32_clip_f(dest0, src[0], n); + return lsx_rint32_clip_2_f(dest0, src, ch, n); + + case SOXR_INT16: if (seed) { + if (ch == 1) + return lsx_rint16_clip_dither_f(dest0, src[0], n, seed); + return lsx_rint16_clip_2_dither_f(dest0, src, ch, n, seed); + } + if (ch == 1) + return lsx_rint16_clip_f(dest0, src[0], n); + return lsx_rint16_clip_2_f(dest0, src, ch, n); + default: break; + } + return 0; +} +#endif diff --git a/soxr-sys/src/data-io.h b/soxr-sys/src/data-io.h new file mode 100644 index 000000000..83a0a133d --- /dev/null +++ b/soxr-sys/src/data-io.h @@ -0,0 +1,39 @@ +/* SoX Resampler Library Copyright (c) 2007-13 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#if !defined soxr_data_io_included +#define soxr_data_io_included + +#include "soxr.h" + +void _soxr_deinterleave( + double * * dest, + soxr_datatype_t data_type, + void const * * src0, + size_t n, + unsigned ch); + +void _soxr_deinterleave_f( + float * * dest, + soxr_datatype_t data_type, + void const * * src0, + size_t n, + unsigned ch); + +size_t /* clips */ _soxr_interleave( + soxr_datatype_t data_type, + void * * dest, + double const * const * src, + size_t n, + unsigned ch, + unsigned long * seed); + +size_t /* clips */ _soxr_interleave_f( + soxr_datatype_t data_type, + void * * dest, + float const * const * src, + size_t n, + unsigned ch, + unsigned long * seed); + +#endif diff --git a/soxr-sys/src/dbesi0.c b/soxr-sys/src/dbesi0.c new file mode 100644 index 000000000..654216eb4 --- /dev/null +++ b/soxr-sys/src/dbesi0.c @@ -0,0 +1,149 @@ +/* Copyright(C) 1996 Takuya OOURA + +You may use, copy, modify this code for any purpose and +without fee. + +Package home: http://www.kurims.kyoto-u.ac.jp/~ooura/bessel.html +*/ + +#include "filter.h" +#define dbesi0 lsx_bessel_I_0 + +/* Bessel I_0(x) function in double precision */ + +#include + +double dbesi0(double x) +{ + int k; + double w, t, y; + static double a[65] = { + 8.5246820682016865877e-11, 2.5966600546497407288e-9, + 7.9689994568640180274e-8, 1.9906710409667748239e-6, + 4.0312469446528002532e-5, 6.4499871606224265421e-4, + 0.0079012345761930579108, 0.071111111109207045212, + 0.444444444444724909, 1.7777777777777532045, + 4.0000000000000011182, 3.99999999999999998, + 1.0000000000000000001, + 1.1520919130377195927e-10, 2.2287613013610985225e-9, + 8.1903951930694585113e-8, 1.9821560631611544984e-6, + 4.0335461940910133184e-5, 6.4495330974432203401e-4, + 0.0079013012611467520626, 0.071111038160875566622, + 0.44444450319062699316, 1.7777777439146450067, + 4.0000000132337935071, 3.9999999968569015366, + 1.0000000003426703174, + 1.5476870780515238488e-10, 1.2685004214732975355e-9, + 9.2776861851114223267e-8, 1.9063070109379044378e-6, + 4.0698004389917945832e-5, 6.4370447244298070713e-4, + 0.0079044749458444976958, 0.071105052411749363882, + 0.44445280640924755082, 1.7777694934432109713, + 4.0000055808824003386, 3.9999977081165740932, + 1.0000004333949319118, + 2.0675200625006793075e-10, -6.1689554705125681442e-10, + 1.2436765915401571654e-7, 1.5830429403520613423e-6, + 4.2947227560776583326e-5, 6.3249861665073441312e-4, + 0.0079454472840953930811, 0.070994327785661860575, + 0.44467219586283000332, 1.7774588182255374745, + 4.0003038986252717972, 3.9998233869142057195, + 1.0000472932961288324, + 2.7475684794982708655e-10, -3.8991472076521332023e-9, + 1.9730170483976049388e-7, 5.9651531561967674521e-7, + 5.1992971474748995357e-5, 5.7327338675433770752e-4, + 0.0082293143836530412024, 0.069990934858728039037, + 0.44726764292723985087, 1.7726685170014087784, + 4.0062907863712704432, 3.9952750700487845355, + 1.0016354346654179322 + }; + static double b[70] = { + 6.7852367144945531383e-8, 4.6266061382821826854e-7, + 6.9703135812354071774e-6, 7.6637663462953234134e-5, + 7.9113515222612691636e-4, 0.0073401204731103808981, + 0.060677114958668837046, 0.43994941411651569622, + 2.7420017097661750609, 14.289661921740860534, + 59.820609640320710779, 188.78998681199150629, + 399.8731367825601118, 427.56411572180478514, + 1.8042097874891098754e-7, 1.2277164312044637357e-6, + 1.8484393221474274861e-5, 2.0293995900091309208e-4, + 0.0020918539850246207459, 0.019375315654033949297, + 0.15985869016767185908, 1.1565260527420641724, + 7.1896341224206072113, 37.354773811947484532, + 155.80993164266268457, 489.5211371158540918, + 1030.9147225169564806, 1093.5883545113746958, + 4.8017305613187493564e-7, 3.261317843912380074e-6, + 4.9073137508166159639e-5, 5.3806506676487583755e-4, + 0.0055387918291051866561, 0.051223717488786549025, + 0.42190298621367914765, 3.0463625987357355872, + 18.895299447327733204, 97.915189029455461554, + 407.13940115493494659, 1274.3088990480582632, + 2670.9883037012547506, 2815.7166284662544712, + 1.2789926338424623394e-6, 8.6718263067604918916e-6, + 1.3041508821299929489e-4, 0.001428224737372747892, + 0.014684070635768789378, 0.13561403190404185755, + 1.1152592585977393953, 8.0387088559465389038, + 49.761318895895479206, 257.2684232313529138, + 1066.8543146269566231, 3328.3874581009636362, + 6948.8586598121634874, 7288.4893398212481055, + 3.409350368197032893e-6, 2.3079025203103376076e-5, + 3.4691373283901830239e-4, 0.003794994977222908545, + 0.038974209677945602145, 0.3594948380414878371, + 2.9522878893539528226, 21.246564609514287056, + 131.28727387146173141, 677.38107093296675421, + 2802.3724744545046518, 8718.5731420798254081, + 18141.348781638832286, 18948.925349296308859 + }; + static double c[45] = { + 2.5568678676452702768e-15, 3.0393953792305924324e-14, + 6.3343751991094840009e-13, 1.5041298011833009649e-11, + 4.4569436918556541414e-10, 1.746393051427167951e-8, + 1.0059224011079852317e-6, 1.0729838945088577089e-4, + 0.05150322693642527738, + 5.2527963991711562216e-15, 7.202118481421005641e-15, + 7.2561421229904797156e-13, 1.482312146673104251e-11, + 4.4602670450376245434e-10, 1.7463600061788679671e-8, + 1.005922609132234756e-6, 1.0729838937545111487e-4, + 0.051503226936437300716, + 1.3365917359358069908e-14, -1.2932643065888544835e-13, + 1.7450199447905602915e-12, 1.0419051209056979788e-11, + 4.58047881980598326e-10, 1.7442405450073548966e-8, + 1.0059461453281292278e-6, 1.0729837434500161228e-4, + 0.051503226940658446941, + 5.3771611477352308649e-14, -1.1396193006413731702e-12, + 1.2858641335221653409e-11, -5.9802086004570057703e-11, + 7.3666894305929510222e-10, 1.6731837150730356448e-8, + 1.0070831435812128922e-6, 1.0729733111203704813e-4, + 0.051503227360726294675, + 3.7819492084858931093e-14, -4.8600496888588034879e-13, + 1.6898350504817224909e-12, 4.5884624327524255865e-11, + 1.2521615963377513729e-10, 1.8959658437754727957e-8, + 1.0020716710561353622e-6, 1.073037119856927559e-4, + 0.05150322383300230775 + }; + + w = fabs(x); + if (w < 8.5) { + t = w * w * 0.0625; + k = 13 * ((int) t); + y = (((((((((((a[k] * t + a[k + 1]) * t + + a[k + 2]) * t + a[k + 3]) * t + a[k + 4]) * t + + a[k + 5]) * t + a[k + 6]) * t + a[k + 7]) * t + + a[k + 8]) * t + a[k + 9]) * t + a[k + 10]) * t + + a[k + 11]) * t + a[k + 12]; + } else if (w < 12.5) { + k = (int) w; + t = w - k; + k = 14 * (k - 8); + y = ((((((((((((b[k] * t + b[k + 1]) * t + + b[k + 2]) * t + b[k + 3]) * t + b[k + 4]) * t + + b[k + 5]) * t + b[k + 6]) * t + b[k + 7]) * t + + b[k + 8]) * t + b[k + 9]) * t + b[k + 10]) * t + + b[k + 11]) * t + b[k + 12]) * t + b[k + 13]; + } else { + t = 60 / w; + k = 9 * ((int) t); + y = ((((((((c[k] * t + c[k + 1]) * t + + c[k + 2]) * t + c[k + 3]) * t + c[k + 4]) * t + + c[k + 5]) * t + c[k + 6]) * t + c[k + 7]) * t + + c[k + 8]) * sqrt(t) * exp(w); + } + return y; +} diff --git a/soxr-sys/src/dev32s.h b/soxr-sys/src/dev32s.h new file mode 100644 index 000000000..7edae868d --- /dev/null +++ b/soxr-sys/src/dev32s.h @@ -0,0 +1,54 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#if !defined soxr_dev32s_included +#define soxr_dev32s_included + +#if defined __GNUC__ + #define SIMD_INLINE(T) static __inline T __attribute__((always_inline)) + #define vAlign __attribute__((aligned (16))) +#elif defined _MSC_VER + #define SIMD_INLINE(T) static __forceinline T + #define vAlign __declspec(align(16)) +#endif + +#if defined __x86_64__ || defined _M_X64 || defined i386 || defined _M_IX86 + +#include + +#define vZero() _mm_setzero_ps() +#define vSet1(a) _mm_set_ss(a) +#define vMul(a,b) _mm_mul_ps(a,b) +#define vAdd(a,b) _mm_add_ps(a,b) +#define vMac(a,b,c) vAdd(vMul(a,b),c) +#define vLds(a) _mm_set1_ps(a) +#define vLd(a) _mm_load_ps(a) +#define vLdu(a) _mm_loadu_ps(a) + +typedef __m128 v4_t; + +SIMD_INLINE(void) vStorSum(float * a, v4_t b) { + v4_t t = vAdd(_mm_movehl_ps(b, b), b); + _mm_store_ss(a, vAdd(t, _mm_shuffle_ps(t,t,1)));} + +#elif defined __arm__ + +#include + +#define vZero() vdupq_n_f32(0) +#define vMul(a,b) vmulq_f32(a,b) +#define vAdd(a,b) vaddq_f32(a,b) +#define vMac(a,b,c) vmlaq_f32(c,a,b) +#define vLds(a) vld1q_dup_f32(&(a)) +#define vLd(a) vld1q_f32(a) +#define vLdu(a) vld1q_f32(a) + +typedef float32x4_t v4_t; + +SIMD_INLINE(void) vStorSum(float * a, v4_t b) { + float32x2_t t = vadd_f32(vget_high_f32(b), vget_low_f32(b)); + *a = vget_lane_f32(vpadd_f32(t, t), 0);} + +#endif + +#endif diff --git a/soxr-sys/src/dev64s.h b/soxr-sys/src/dev64s.h new file mode 100644 index 000000000..4672210d1 --- /dev/null +++ b/soxr-sys/src/dev64s.h @@ -0,0 +1,42 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#if !defined soxr_dev64s_included +#define soxr_dev64s_included + +#if defined __GNUC__ + #define SIMD_INLINE(T) static __inline T __attribute__((always_inline)) + #define vAlign __attribute__((aligned (32))) +#elif defined _MSC_VER + #define SIMD_INLINE(T) static __forceinline T + #define vAlign __declspec(align(32)) +#else + #define SIMD_INLINE(T) static __inline T +#endif + +#if defined __x86_64__ || defined _M_X64 || defined i386 || defined _M_IX86 + +#include + +#if defined __AVX__ + +#define vZero() _mm256_setzero_pd() +#define vSet1(a) _mm256_set_pd(0,0,0,a) +#define vMul(a,b) _mm256_mul_pd(a,b) +#define vAdd(a,b) _mm256_add_pd(a,b) +#define vMac(a,b,c) vAdd(vMul(a,b),c) /* Note: gcc -mfma will `fuse' these */ +#define vLds(a) _mm256_set1_pd(a) +#define vLd(a) _mm256_load_pd(a) +#define vLdu(a) _mm256_loadu_pd(a) + +typedef __m256d v4_t; + +SIMD_INLINE(void) vStorSum(double * a, v4_t b) { + b = _mm256_hadd_pd(b, _mm256_permute2f128_pd(b,b,1)); + _mm_store_sd(a, _mm256_castpd256_pd128(_mm256_hadd_pd(b,b)));} + +#endif + +#endif + +#endif diff --git a/soxr-sys/src/fft4g.c b/soxr-sys/src/fft4g.c new file mode 100644 index 000000000..cf6293a04 --- /dev/null +++ b/soxr-sys/src/fft4g.c @@ -0,0 +1,1346 @@ +/* Copyright Takuya OOURA, 1996-2001. + +You may use, copy, modify and distribute this code for any +purpose (include commercial use) and without fee. Please +refer to this package when you modify this code. + +Package home: http://www.kurims.kyoto-u.ac.jp/~ooura/fft.html + +Fast Fourier/Cosine/Sine Transform + dimension :one + data length :power of 2 + decimation :frequency + radix :4, 2 + data :inplace + table :use +functions + cdft: Complex Discrete Fourier Transform + rdft: Real Discrete Fourier Transform + ddct: Discrete Cosine Transform + ddst: Discrete Sine Transform + dfct: Cosine Transform of RDFT (Real Symmetric DFT) + dfst: Sine Transform of RDFT (Real Anti-symmetric DFT) +function prototypes + void cdft(int, int, double *, int *, double *); + void rdft(int, int, double *, int *, double *); + void ddct(int, int, double *, int *, double *); + void ddst(int, int, double *, int *, double *); + void dfct(int, double *, double *, int *, double *); + void dfst(int, double *, double *, int *, double *); + + +-------- Complex DFT (Discrete Fourier Transform) -------- + [definition] + + X[k] = sum_j=0^n-1 x[j]*exp(2*pi*i*j*k/n), 0<=k + X[k] = sum_j=0^n-1 x[j]*exp(-2*pi*i*j*k/n), 0<=k + ip[0] = 0; // first time only + cdft(2*n, 1, a, ip, w); + + ip[0] = 0; // first time only + cdft(2*n, -1, a, ip, w); + [parameters] + 2*n :data length (int) + n >= 1, n = power of 2 + a[0...2*n-1] :input/output data (double *) + input data + a[2*j] = Re(x[j]), + a[2*j+1] = Im(x[j]), 0<=j= 2+sqrt(n) + strictly, + length of ip >= + 2+(1<<(int)(log(n+0.5)/log(2))/2). + ip[0],ip[1] are pointers of the cos/sin table. + w[0...n/2-1] :cos/sin table (double *) + w[],ip[] are initialized if ip[0] == 0. + [remark] + Inverse of + cdft(2*n, -1, a, ip, w); + is + cdft(2*n, 1, a, ip, w); + for (j = 0; j <= 2 * n - 1; j++) { + a[j] *= 1.0 / n; + } + . + + +-------- Real DFT / Inverse of Real DFT -------- + [definition] + RDFT + R[k] = sum_j=0^n-1 a[j]*cos(2*pi*j*k/n), 0<=k<=n/2 + I[k] = sum_j=0^n-1 a[j]*sin(2*pi*j*k/n), 0 IRDFT (excluding scale) + a[k] = (R[0] + R[n/2]*cos(pi*k))/2 + + sum_j=1^n/2-1 R[j]*cos(2*pi*j*k/n) + + sum_j=1^n/2-1 I[j]*sin(2*pi*j*k/n), 0<=k + ip[0] = 0; // first time only + rdft(n, 1, a, ip, w); + + ip[0] = 0; // first time only + rdft(n, -1, a, ip, w); + [parameters] + n :data length (int) + n >= 2, n = power of 2 + a[0...n-1] :input/output data (double *) + + output data + a[2*k] = R[k], 0<=k + input data + a[2*j] = R[j], 0<=j= 2+sqrt(n/2) + strictly, + length of ip >= + 2+(1<<(int)(log(n/2+0.5)/log(2))/2). + ip[0],ip[1] are pointers of the cos/sin table. + w[0...n/2-1] :cos/sin table (double *) + w[],ip[] are initialized if ip[0] == 0. + [remark] + Inverse of + rdft(n, 1, a, ip, w); + is + rdft(n, -1, a, ip, w); + for (j = 0; j <= n - 1; j++) { + a[j] *= 2.0 / n; + } + . + + +-------- DCT (Discrete Cosine Transform) / Inverse of DCT -------- + [definition] + IDCT (excluding scale) + C[k] = sum_j=0^n-1 a[j]*cos(pi*j*(k+1/2)/n), 0<=k DCT + C[k] = sum_j=0^n-1 a[j]*cos(pi*(j+1/2)*k/n), 0<=k + ip[0] = 0; // first time only + ddct(n, 1, a, ip, w); + + ip[0] = 0; // first time only + ddct(n, -1, a, ip, w); + [parameters] + n :data length (int) + n >= 2, n = power of 2 + a[0...n-1] :input/output data (double *) + output data + a[k] = C[k], 0<=k= 2+sqrt(n/2) + strictly, + length of ip >= + 2+(1<<(int)(log(n/2+0.5)/log(2))/2). + ip[0],ip[1] are pointers of the cos/sin table. + w[0...n*5/4-1] :cos/sin table (double *) + w[],ip[] are initialized if ip[0] == 0. + [remark] + Inverse of + ddct(n, -1, a, ip, w); + is + a[0] *= 0.5; + ddct(n, 1, a, ip, w); + for (j = 0; j <= n - 1; j++) { + a[j] *= 2.0 / n; + } + . + + +-------- DST (Discrete Sine Transform) / Inverse of DST -------- + [definition] + IDST (excluding scale) + S[k] = sum_j=1^n A[j]*sin(pi*j*(k+1/2)/n), 0<=k DST + S[k] = sum_j=0^n-1 a[j]*sin(pi*(j+1/2)*k/n), 0 + ip[0] = 0; // first time only + ddst(n, 1, a, ip, w); + + ip[0] = 0; // first time only + ddst(n, -1, a, ip, w); + [parameters] + n :data length (int) + n >= 2, n = power of 2 + a[0...n-1] :input/output data (double *) + + input data + a[j] = A[j], 0 + output data + a[k] = S[k], 0= 2+sqrt(n/2) + strictly, + length of ip >= + 2+(1<<(int)(log(n/2+0.5)/log(2))/2). + ip[0],ip[1] are pointers of the cos/sin table. + w[0...n*5/4-1] :cos/sin table (double *) + w[],ip[] are initialized if ip[0] == 0. + [remark] + Inverse of + ddst(n, -1, a, ip, w); + is + a[0] *= 0.5; + ddst(n, 1, a, ip, w); + for (j = 0; j <= n - 1; j++) { + a[j] *= 2.0 / n; + } + . + + +-------- Cosine Transform of RDFT (Real Symmetric DFT) -------- + [definition] + C[k] = sum_j=0^n a[j]*cos(pi*j*k/n), 0<=k<=n + [usage] + ip[0] = 0; // first time only + dfct(n, a, t, ip, w); + [parameters] + n :data length - 1 (int) + n >= 2, n = power of 2 + a[0...n] :input/output data (double *) + output data + a[k] = C[k], 0<=k<=n + t[0...n/2] :work area (double *) + ip[0...*] :work area for bit reversal (int *) + length of ip >= 2+sqrt(n/4) + strictly, + length of ip >= + 2+(1<<(int)(log(n/4+0.5)/log(2))/2). + ip[0],ip[1] are pointers of the cos/sin table. + w[0...n*5/8-1] :cos/sin table (double *) + w[],ip[] are initialized if ip[0] == 0. + [remark] + Inverse of + a[0] *= 0.5; + a[n] *= 0.5; + dfct(n, a, t, ip, w); + is + a[0] *= 0.5; + a[n] *= 0.5; + dfct(n, a, t, ip, w); + for (j = 0; j <= n; j++) { + a[j] *= 2.0 / n; + } + . + + +-------- Sine Transform of RDFT (Real Anti-symmetric DFT) -------- + [definition] + S[k] = sum_j=1^n-1 a[j]*sin(pi*j*k/n), 0= 2, n = power of 2 + a[0...n-1] :input/output data (double *) + output data + a[k] = S[k], 0= 2+sqrt(n/4) + strictly, + length of ip >= + 2+(1<<(int)(log(n/4+0.5)/log(2))/2). + ip[0],ip[1] are pointers of the cos/sin table. + w[0...n*5/8-1] :cos/sin table (double *) + w[],ip[] are initialized if ip[0] == 0. + [remark] + Inverse of + dfst(n, a, t, ip, w); + is + dfst(n, a, t, ip, w); + for (j = 1; j <= n - 1; j++) { + a[j] *= 2.0 / n; + } + . + + +Appendix : + The cos/sin table is recalculated when the larger table required. + w[] and ip[] are compatible with all routines. +*/ + + +#include "math-wrap.h" +#include "fft4g.h" + +#ifdef FFT4G_FLOAT + #define double float + #define one_half 0.5f + + #define sin(x) sinf(x) + #define cos(x) cosf(x) + #define atan(x) atanf(x) + + #define cdft lsx_cdft_f + #define rdft lsx_rdft_f + #define ddct lsx_ddct_f + #define ddst lsx_ddst_f + #define dfct lsx_dfct_f + #define dfst lsx_dfst_f +#else + #define one_half 0.5 + #define cdft lsx_cdft + #define rdft lsx_rdft + #define ddct lsx_ddct + #define ddst lsx_ddst + #define dfct lsx_dfct + #define dfst lsx_dfst +#endif + +static void bitrv2conj(int n, int *ip, double *a); +static void bitrv2(int n, int *ip, double *a); +static void cft1st(int n, double *a, double const *w); +static void cftbsub(int n, double *a, double const *w); +static void cftfsub(int n, double *a, double const *w); +static void cftmdl(int n, int l, double *a, double const *w); +static void dctsub(int n, double *a, int nc, double const *c); +static void dstsub(int n, double *a, int nc, double const *c); +static void makect(int nc, int *ip, double *c); +static void makewt(int nw, int *ip, double *w); +static void rftbsub(int n, double *a, int nc, double const *c); +static void rftfsub(int n, double *a, int nc, double const *c); + + +void cdft(int n, int isgn, double *a, int *ip, double *w) +{ + if (n > (ip[0] << 2)) { + makewt(n >> 2, ip, w); + } + if (n > 4) { + if (isgn >= 0) { + bitrv2(n, ip + 2, a); + cftfsub(n, a, w); + } else { + bitrv2conj(n, ip + 2, a); + cftbsub(n, a, w); + } + } else if (n == 4) { + cftfsub(n, a, w); + } +} + + +void rdft(int n, int isgn, double *a, int *ip, double *w) +{ + int nw, nc; + double xi; + + nw = ip[0]; + if (n > (nw << 2)) { + nw = n >> 2; + makewt(nw, ip, w); + } + nc = ip[1]; + if (n > (nc << 2)) { + nc = n >> 2; + makect(nc, ip, w + nw); + } + if (isgn >= 0) { + if (n > 4) { + bitrv2(n, ip + 2, a); + cftfsub(n, a, w); + rftfsub(n, a, nc, w + nw); + } else if (n == 4) { + cftfsub(n, a, w); + } + xi = a[0] - a[1]; + a[0] += a[1]; + a[1] = xi; + } else { + a[1] = one_half * (a[0] - a[1]); + a[0] -= a[1]; + if (n > 4) { + rftbsub(n, a, nc, w + nw); + bitrv2(n, ip + 2, a); + cftbsub(n, a, w); + } else if (n == 4) { + cftfsub(n, a, w); + } + } +} + + +void ddct(int n, int isgn, double *a, int *ip, double *w) +{ + int j, nw, nc; + double xr; + + nw = ip[0]; + if (n > (nw << 2)) { + nw = n >> 2; + makewt(nw, ip, w); + } + nc = ip[1]; + if (n > nc) { + nc = n; + makect(nc, ip, w + nw); + } + if (isgn < 0) { + xr = a[n - 1]; + for (j = n - 2; j >= 2; j -= 2) { + a[j + 1] = a[j] - a[j - 1]; + a[j] += a[j - 1]; + } + a[1] = a[0] - xr; + a[0] += xr; + if (n > 4) { + rftbsub(n, a, nc, w + nw); + bitrv2(n, ip + 2, a); + cftbsub(n, a, w); + } else if (n == 4) { + cftfsub(n, a, w); + } + } + dctsub(n, a, nc, w + nw); + if (isgn >= 0) { + if (n > 4) { + bitrv2(n, ip + 2, a); + cftfsub(n, a, w); + rftfsub(n, a, nc, w + nw); + } else if (n == 4) { + cftfsub(n, a, w); + } + xr = a[0] - a[1]; + a[0] += a[1]; + for (j = 2; j < n; j += 2) { + a[j - 1] = a[j] - a[j + 1]; + a[j] += a[j + 1]; + } + a[n - 1] = xr; + } +} + + +void ddst(int n, int isgn, double *a, int *ip, double *w) +{ + int j, nw, nc; + double xr; + + nw = ip[0]; + if (n > (nw << 2)) { + nw = n >> 2; + makewt(nw, ip, w); + } + nc = ip[1]; + if (n > nc) { + nc = n; + makect(nc, ip, w + nw); + } + if (isgn < 0) { + xr = a[n - 1]; + for (j = n - 2; j >= 2; j -= 2) { + a[j + 1] = -a[j] - a[j - 1]; + a[j] -= a[j - 1]; + } + a[1] = a[0] + xr; + a[0] -= xr; + if (n > 4) { + rftbsub(n, a, nc, w + nw); + bitrv2(n, ip + 2, a); + cftbsub(n, a, w); + } else if (n == 4) { + cftfsub(n, a, w); + } + } + dstsub(n, a, nc, w + nw); + if (isgn >= 0) { + if (n > 4) { + bitrv2(n, ip + 2, a); + cftfsub(n, a, w); + rftfsub(n, a, nc, w + nw); + } else if (n == 4) { + cftfsub(n, a, w); + } + xr = a[0] - a[1]; + a[0] += a[1]; + for (j = 2; j < n; j += 2) { + a[j - 1] = -a[j] - a[j + 1]; + a[j] -= a[j + 1]; + } + a[n - 1] = -xr; + } +} + + +void dfct(int n, double *a, double *t, int *ip, double *w) +{ + int j, k, l, m, mh, nw, nc; + double xr, xi, yr, yi; + + nw = ip[0]; + if (n > (nw << 3)) { + nw = n >> 3; + makewt(nw, ip, w); + } + nc = ip[1]; + if (n > (nc << 1)) { + nc = n >> 1; + makect(nc, ip, w + nw); + } + m = n >> 1; + yi = a[m]; + xi = a[0] + a[n]; + a[0] -= a[n]; + t[0] = xi - yi; + t[m] = xi + yi; + if (n > 2) { + mh = m >> 1; + for (j = 1; j < mh; j++) { + k = m - j; + xr = a[j] - a[n - j]; + xi = a[j] + a[n - j]; + yr = a[k] - a[n - k]; + yi = a[k] + a[n - k]; + a[j] = xr; + a[k] = yr; + t[j] = xi - yi; + t[k] = xi + yi; + } + t[mh] = a[mh] + a[n - mh]; + a[mh] -= a[n - mh]; + dctsub(m, a, nc, w + nw); + if (m > 4) { + bitrv2(m, ip + 2, a); + cftfsub(m, a, w); + rftfsub(m, a, nc, w + nw); + } else if (m == 4) { + cftfsub(m, a, w); + } + a[n - 1] = a[0] - a[1]; + a[1] = a[0] + a[1]; + for (j = m - 2; j >= 2; j -= 2) { + a[2 * j + 1] = a[j] + a[j + 1]; + a[2 * j - 1] = a[j] - a[j + 1]; + } + l = 2; + m = mh; + while (m >= 2) { + dctsub(m, t, nc, w + nw); + if (m > 4) { + bitrv2(m, ip + 2, t); + cftfsub(m, t, w); + rftfsub(m, t, nc, w + nw); + } else if (m == 4) { + cftfsub(m, t, w); + } + a[n - l] = t[0] - t[1]; + a[l] = t[0] + t[1]; + k = 0; + for (j = 2; j < m; j += 2) { + k += l << 2; + a[k - l] = t[j] - t[j + 1]; + a[k + l] = t[j] + t[j + 1]; + } + l <<= 1; + mh = m >> 1; + for (j = 0; j < mh; j++) { + k = m - j; + t[j] = t[m + k] - t[m + j]; + t[k] = t[m + k] + t[m + j]; + } + t[mh] = t[m + mh]; + m = mh; + } + a[l] = t[0]; + a[n] = t[2] - t[1]; + a[0] = t[2] + t[1]; + } else { + a[1] = a[0]; + a[2] = t[0]; + a[0] = t[1]; + } +} + + +void dfst(int n, double *a, double *t, int *ip, double *w) +{ + int j, k, l, m, mh, nw, nc; + double xr, xi, yr, yi; + + nw = ip[0]; + if (n > (nw << 3)) { + nw = n >> 3; + makewt(nw, ip, w); + } + nc = ip[1]; + if (n > (nc << 1)) { + nc = n >> 1; + makect(nc, ip, w + nw); + } + if (n > 2) { + m = n >> 1; + mh = m >> 1; + for (j = 1; j < mh; j++) { + k = m - j; + xr = a[j] + a[n - j]; + xi = a[j] - a[n - j]; + yr = a[k] + a[n - k]; + yi = a[k] - a[n - k]; + a[j] = xr; + a[k] = yr; + t[j] = xi + yi; + t[k] = xi - yi; + } + t[0] = a[mh] - a[n - mh]; + a[mh] += a[n - mh]; + a[0] = a[m]; + dstsub(m, a, nc, w + nw); + if (m > 4) { + bitrv2(m, ip + 2, a); + cftfsub(m, a, w); + rftfsub(m, a, nc, w + nw); + } else if (m == 4) { + cftfsub(m, a, w); + } + a[n - 1] = a[1] - a[0]; + a[1] = a[0] + a[1]; + for (j = m - 2; j >= 2; j -= 2) { + a[2 * j + 1] = a[j] - a[j + 1]; + a[2 * j - 1] = -a[j] - a[j + 1]; + } + l = 2; + m = mh; + while (m >= 2) { + dstsub(m, t, nc, w + nw); + if (m > 4) { + bitrv2(m, ip + 2, t); + cftfsub(m, t, w); + rftfsub(m, t, nc, w + nw); + } else if (m == 4) { + cftfsub(m, t, w); + } + a[n - l] = t[1] - t[0]; + a[l] = t[0] + t[1]; + k = 0; + for (j = 2; j < m; j += 2) { + k += l << 2; + a[k - l] = -t[j] - t[j + 1]; + a[k + l] = t[j] - t[j + 1]; + } + l <<= 1; + mh = m >> 1; + for (j = 1; j < mh; j++) { + k = m - j; + t[j] = t[m + k] + t[m + j]; + t[k] = t[m + k] - t[m + j]; + } + t[0] = t[m + mh]; + m = mh; + } + a[l] = t[0]; + } + a[0] = 0; +} + + +/* -------- initializing routines -------- */ + + +static void makewt(int nw, int *ip, double *w) +{ + int j, nwh; + double delta, x, y; + + ip[0] = nw; + ip[1] = 1; + if (nw > 2) { + nwh = nw >> 1; + delta = atan(1.0) / (double)nwh; + w[0] = 1; + w[1] = 0; + w[nwh] = cos(delta * (double)nwh); + w[nwh + 1] = w[nwh]; + if (nwh > 2) { + for (j = 2; j < nwh; j += 2) { + x = cos(delta * (double)j); + y = sin(delta * (double)j); + w[j] = x; + w[j + 1] = y; + w[nw - j] = y; + w[nw - j + 1] = x; + } + bitrv2(nw, ip + 2, w); + } + } +} + + +static void makect(int nc, int *ip, double *c) +{ + int j, nch; + double delta; + + ip[1] = nc; + if (nc > 1) { + nch = nc >> 1; + delta = atan(1.0) / (double)nch; + c[0] = cos(delta * (double)nch); + c[nch] = one_half * c[0]; + for (j = 1; j < nch; j++) { + c[j] = one_half * cos(delta * (double)j); + c[nc - j] = one_half * sin(delta * (double)j); + } + } +} + + +/* -------- child routines -------- */ + + +static void bitrv2(int n, int *ip0, double *a) +{ + int j, j1, k, k1, l, m, m2, ip[1024]; + double xr, xi, yr, yi; + + (void)ip0; + ip[0] = 0; + l = n; + m = 1; + while ((m << 3) < l) { + l >>= 1; + for (j = 0; j < m; j++) { + ip[m + j] = ip[j] + l; + } + m <<= 1; + } + m2 = 2 * m; + if ((m << 3) == l) { + for (k = 0; k < m; k++) { + for (j = 0; j < k; j++) { + j1 = 2 * j + ip[k]; + k1 = 2 * k + ip[j]; + xr = a[j1]; + xi = a[j1 + 1]; + yr = a[k1]; + yi = a[k1 + 1]; + a[j1] = yr; + a[j1 + 1] = yi; + a[k1] = xr; + a[k1 + 1] = xi; + j1 += m2; + k1 += 2 * m2; + xr = a[j1]; + xi = a[j1 + 1]; + yr = a[k1]; + yi = a[k1 + 1]; + a[j1] = yr; + a[j1 + 1] = yi; + a[k1] = xr; + a[k1 + 1] = xi; + j1 += m2; + k1 -= m2; + xr = a[j1]; + xi = a[j1 + 1]; + yr = a[k1]; + yi = a[k1 + 1]; + a[j1] = yr; + a[j1 + 1] = yi; + a[k1] = xr; + a[k1 + 1] = xi; + j1 += m2; + k1 += 2 * m2; + xr = a[j1]; + xi = a[j1 + 1]; + yr = a[k1]; + yi = a[k1 + 1]; + a[j1] = yr; + a[j1 + 1] = yi; + a[k1] = xr; + a[k1 + 1] = xi; + } + j1 = 2 * k + m2 + ip[k]; + k1 = j1 + m2; + xr = a[j1]; + xi = a[j1 + 1]; + yr = a[k1]; + yi = a[k1 + 1]; + a[j1] = yr; + a[j1 + 1] = yi; + a[k1] = xr; + a[k1 + 1] = xi; + } + } else { + for (k = 1; k < m; k++) { + for (j = 0; j < k; j++) { + j1 = 2 * j + ip[k]; + k1 = 2 * k + ip[j]; + xr = a[j1]; + xi = a[j1 + 1]; + yr = a[k1]; + yi = a[k1 + 1]; + a[j1] = yr; + a[j1 + 1] = yi; + a[k1] = xr; + a[k1 + 1] = xi; + j1 += m2; + k1 += m2; + xr = a[j1]; + xi = a[j1 + 1]; + yr = a[k1]; + yi = a[k1 + 1]; + a[j1] = yr; + a[j1 + 1] = yi; + a[k1] = xr; + a[k1 + 1] = xi; + } + } + } +} + + +static void bitrv2conj(int n, int *ip0, double *a) +{ + int j, j1, k, k1, l, m, m2, ip[512]; + double xr, xi, yr, yi; + + (void)ip0; + ip[0] = 0; + l = n; + m = 1; + while ((m << 3) < l) { + l >>= 1; + for (j = 0; j < m; j++) { + ip[m + j] = ip[j] + l; + } + m <<= 1; + } + m2 = 2 * m; + if ((m << 3) == l) { + for (k = 0; k < m; k++) { + for (j = 0; j < k; j++) { + j1 = 2 * j + ip[k]; + k1 = 2 * k + ip[j]; + xr = a[j1]; + xi = -a[j1 + 1]; + yr = a[k1]; + yi = -a[k1 + 1]; + a[j1] = yr; + a[j1 + 1] = yi; + a[k1] = xr; + a[k1 + 1] = xi; + j1 += m2; + k1 += 2 * m2; + xr = a[j1]; + xi = -a[j1 + 1]; + yr = a[k1]; + yi = -a[k1 + 1]; + a[j1] = yr; + a[j1 + 1] = yi; + a[k1] = xr; + a[k1 + 1] = xi; + j1 += m2; + k1 -= m2; + xr = a[j1]; + xi = -a[j1 + 1]; + yr = a[k1]; + yi = -a[k1 + 1]; + a[j1] = yr; + a[j1 + 1] = yi; + a[k1] = xr; + a[k1 + 1] = xi; + j1 += m2; + k1 += 2 * m2; + xr = a[j1]; + xi = -a[j1 + 1]; + yr = a[k1]; + yi = -a[k1 + 1]; + a[j1] = yr; + a[j1 + 1] = yi; + a[k1] = xr; + a[k1 + 1] = xi; + } + k1 = 2 * k + ip[k]; + a[k1 + 1] = -a[k1 + 1]; + j1 = k1 + m2; + k1 = j1 + m2; + xr = a[j1]; + xi = -a[j1 + 1]; + yr = a[k1]; + yi = -a[k1 + 1]; + a[j1] = yr; + a[j1 + 1] = yi; + a[k1] = xr; + a[k1 + 1] = xi; + k1 += m2; + a[k1 + 1] = -a[k1 + 1]; + } + } else { + a[1] = -a[1]; + a[m2 + 1] = -a[m2 + 1]; + for (k = 1; k < m; k++) { + for (j = 0; j < k; j++) { + j1 = 2 * j + ip[k]; + k1 = 2 * k + ip[j]; + xr = a[j1]; + xi = -a[j1 + 1]; + yr = a[k1]; + yi = -a[k1 + 1]; + a[j1] = yr; + a[j1 + 1] = yi; + a[k1] = xr; + a[k1 + 1] = xi; + j1 += m2; + k1 += m2; + xr = a[j1]; + xi = -a[j1 + 1]; + yr = a[k1]; + yi = -a[k1 + 1]; + a[j1] = yr; + a[j1 + 1] = yi; + a[k1] = xr; + a[k1 + 1] = xi; + } + k1 = 2 * k + ip[k]; + a[k1 + 1] = -a[k1 + 1]; + a[k1 + m2 + 1] = -a[k1 + m2 + 1]; + } + } +} + + +static void cftfsub(int n, double *a, double const *w) +{ + int j, j1, j2, j3, l; + double x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i; + + l = 2; + if (n > 8) { + cft1st(n, a, w); + l = 8; + while ((l << 2) < n) { + cftmdl(n, l, a, w); + l <<= 2; + } + } + if ((l << 2) == n) { + for (j = 0; j < l; j += 2) { + j1 = j + l; + j2 = j1 + l; + j3 = j2 + l; + x0r = a[j] + a[j1]; + x0i = a[j + 1] + a[j1 + 1]; + x1r = a[j] - a[j1]; + x1i = a[j + 1] - a[j1 + 1]; + x2r = a[j2] + a[j3]; + x2i = a[j2 + 1] + a[j3 + 1]; + x3r = a[j2] - a[j3]; + x3i = a[j2 + 1] - a[j3 + 1]; + a[j] = x0r + x2r; + a[j + 1] = x0i + x2i; + a[j2] = x0r - x2r; + a[j2 + 1] = x0i - x2i; + a[j1] = x1r - x3i; + a[j1 + 1] = x1i + x3r; + a[j3] = x1r + x3i; + a[j3 + 1] = x1i - x3r; + } + } else { + for (j = 0; j < l; j += 2) { + j1 = j + l; + x0r = a[j] - a[j1]; + x0i = a[j + 1] - a[j1 + 1]; + a[j] += a[j1]; + a[j + 1] += a[j1 + 1]; + a[j1] = x0r; + a[j1 + 1] = x0i; + } + } +} + + +static void cftbsub(int n, double *a, double const *w) +{ + int j, j1, j2, j3, l; + double x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i; + + l = 2; + if (n > 8) { + cft1st(n, a, w); + l = 8; + while ((l << 2) < n) { + cftmdl(n, l, a, w); + l <<= 2; + } + } + if ((l << 2) == n) { + for (j = 0; j < l; j += 2) { + j1 = j + l; + j2 = j1 + l; + j3 = j2 + l; + x0r = a[j] + a[j1]; + x0i = -a[j + 1] - a[j1 + 1]; + x1r = a[j] - a[j1]; + x1i = -a[j + 1] + a[j1 + 1]; + x2r = a[j2] + a[j3]; + x2i = a[j2 + 1] + a[j3 + 1]; + x3r = a[j2] - a[j3]; + x3i = a[j2 + 1] - a[j3 + 1]; + a[j] = x0r + x2r; + a[j + 1] = x0i - x2i; + a[j2] = x0r - x2r; + a[j2 + 1] = x0i + x2i; + a[j1] = x1r - x3i; + a[j1 + 1] = x1i - x3r; + a[j3] = x1r + x3i; + a[j3 + 1] = x1i + x3r; + } + } else { + for (j = 0; j < l; j += 2) { + j1 = j + l; + x0r = a[j] - a[j1]; + x0i = -a[j + 1] + a[j1 + 1]; + a[j] += a[j1]; + a[j + 1] = -a[j + 1] - a[j1 + 1]; + a[j1] = x0r; + a[j1 + 1] = x0i; + } + } +} + + +static void cft1st(int n, double *a, double const *w) +{ + int j, k1, k2; + double wk1r, wk1i, wk2r, wk2i, wk3r, wk3i; + double x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i; + + x0r = a[0] + a[2]; + x0i = a[1] + a[3]; + x1r = a[0] - a[2]; + x1i = a[1] - a[3]; + x2r = a[4] + a[6]; + x2i = a[5] + a[7]; + x3r = a[4] - a[6]; + x3i = a[5] - a[7]; + a[0] = x0r + x2r; + a[1] = x0i + x2i; + a[4] = x0r - x2r; + a[5] = x0i - x2i; + a[2] = x1r - x3i; + a[3] = x1i + x3r; + a[6] = x1r + x3i; + a[7] = x1i - x3r; + wk1r = w[2]; + x0r = a[8] + a[10]; + x0i = a[9] + a[11]; + x1r = a[8] - a[10]; + x1i = a[9] - a[11]; + x2r = a[12] + a[14]; + x2i = a[13] + a[15]; + x3r = a[12] - a[14]; + x3i = a[13] - a[15]; + a[8] = x0r + x2r; + a[9] = x0i + x2i; + a[12] = x2i - x0i; + a[13] = x0r - x2r; + x0r = x1r - x3i; + x0i = x1i + x3r; + a[10] = wk1r * (x0r - x0i); + a[11] = wk1r * (x0r + x0i); + x0r = x3i + x1r; + x0i = x3r - x1i; + a[14] = wk1r * (x0i - x0r); + a[15] = wk1r * (x0i + x0r); + k1 = 0; + for (j = 16; j < n; j += 16) { + k1 += 2; + k2 = 2 * k1; + wk2r = w[k1]; + wk2i = w[k1 + 1]; + wk1r = w[k2]; + wk1i = w[k2 + 1]; + wk3r = wk1r - 2 * wk2i * wk1i; + wk3i = 2 * wk2i * wk1r - wk1i; + x0r = a[j] + a[j + 2]; + x0i = a[j + 1] + a[j + 3]; + x1r = a[j] - a[j + 2]; + x1i = a[j + 1] - a[j + 3]; + x2r = a[j + 4] + a[j + 6]; + x2i = a[j + 5] + a[j + 7]; + x3r = a[j + 4] - a[j + 6]; + x3i = a[j + 5] - a[j + 7]; + a[j] = x0r + x2r; + a[j + 1] = x0i + x2i; + x0r -= x2r; + x0i -= x2i; + a[j + 4] = wk2r * x0r - wk2i * x0i; + a[j + 5] = wk2r * x0i + wk2i * x0r; + x0r = x1r - x3i; + x0i = x1i + x3r; + a[j + 2] = wk1r * x0r - wk1i * x0i; + a[j + 3] = wk1r * x0i + wk1i * x0r; + x0r = x1r + x3i; + x0i = x1i - x3r; + a[j + 6] = wk3r * x0r - wk3i * x0i; + a[j + 7] = wk3r * x0i + wk3i * x0r; + wk1r = w[k2 + 2]; + wk1i = w[k2 + 3]; + wk3r = wk1r - 2 * wk2r * wk1i; + wk3i = 2 * wk2r * wk1r - wk1i; + x0r = a[j + 8] + a[j + 10]; + x0i = a[j + 9] + a[j + 11]; + x1r = a[j + 8] - a[j + 10]; + x1i = a[j + 9] - a[j + 11]; + x2r = a[j + 12] + a[j + 14]; + x2i = a[j + 13] + a[j + 15]; + x3r = a[j + 12] - a[j + 14]; + x3i = a[j + 13] - a[j + 15]; + a[j + 8] = x0r + x2r; + a[j + 9] = x0i + x2i; + x0r -= x2r; + x0i -= x2i; + a[j + 12] = -wk2i * x0r - wk2r * x0i; + a[j + 13] = -wk2i * x0i + wk2r * x0r; + x0r = x1r - x3i; + x0i = x1i + x3r; + a[j + 10] = wk1r * x0r - wk1i * x0i; + a[j + 11] = wk1r * x0i + wk1i * x0r; + x0r = x1r + x3i; + x0i = x1i - x3r; + a[j + 14] = wk3r * x0r - wk3i * x0i; + a[j + 15] = wk3r * x0i + wk3i * x0r; + } +} + + +static void cftmdl(int n, int l, double *a, double const *w) +{ + int j, j1, j2, j3, k, k1, k2, m, m2; + double wk1r, wk1i, wk2r, wk2i, wk3r, wk3i; + double x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i; + + m = l << 2; + for (j = 0; j < l; j += 2) { + j1 = j + l; + j2 = j1 + l; + j3 = j2 + l; + x0r = a[j] + a[j1]; + x0i = a[j + 1] + a[j1 + 1]; + x1r = a[j] - a[j1]; + x1i = a[j + 1] - a[j1 + 1]; + x2r = a[j2] + a[j3]; + x2i = a[j2 + 1] + a[j3 + 1]; + x3r = a[j2] - a[j3]; + x3i = a[j2 + 1] - a[j3 + 1]; + a[j] = x0r + x2r; + a[j + 1] = x0i + x2i; + a[j2] = x0r - x2r; + a[j2 + 1] = x0i - x2i; + a[j1] = x1r - x3i; + a[j1 + 1] = x1i + x3r; + a[j3] = x1r + x3i; + a[j3 + 1] = x1i - x3r; + } + wk1r = w[2]; + for (j = m; j < l + m; j += 2) { + j1 = j + l; + j2 = j1 + l; + j3 = j2 + l; + x0r = a[j] + a[j1]; + x0i = a[j + 1] + a[j1 + 1]; + x1r = a[j] - a[j1]; + x1i = a[j + 1] - a[j1 + 1]; + x2r = a[j2] + a[j3]; + x2i = a[j2 + 1] + a[j3 + 1]; + x3r = a[j2] - a[j3]; + x3i = a[j2 + 1] - a[j3 + 1]; + a[j] = x0r + x2r; + a[j + 1] = x0i + x2i; + a[j2] = x2i - x0i; + a[j2 + 1] = x0r - x2r; + x0r = x1r - x3i; + x0i = x1i + x3r; + a[j1] = wk1r * (x0r - x0i); + a[j1 + 1] = wk1r * (x0r + x0i); + x0r = x3i + x1r; + x0i = x3r - x1i; + a[j3] = wk1r * (x0i - x0r); + a[j3 + 1] = wk1r * (x0i + x0r); + } + k1 = 0; + m2 = 2 * m; + for (k = m2; k < n; k += m2) { + k1 += 2; + k2 = 2 * k1; + wk2r = w[k1]; + wk2i = w[k1 + 1]; + wk1r = w[k2]; + wk1i = w[k2 + 1]; + wk3r = wk1r - 2 * wk2i * wk1i; + wk3i = 2 * wk2i * wk1r - wk1i; + for (j = k; j < l + k; j += 2) { + j1 = j + l; + j2 = j1 + l; + j3 = j2 + l; + x0r = a[j] + a[j1]; + x0i = a[j + 1] + a[j1 + 1]; + x1r = a[j] - a[j1]; + x1i = a[j + 1] - a[j1 + 1]; + x2r = a[j2] + a[j3]; + x2i = a[j2 + 1] + a[j3 + 1]; + x3r = a[j2] - a[j3]; + x3i = a[j2 + 1] - a[j3 + 1]; + a[j] = x0r + x2r; + a[j + 1] = x0i + x2i; + x0r -= x2r; + x0i -= x2i; + a[j2] = wk2r * x0r - wk2i * x0i; + a[j2 + 1] = wk2r * x0i + wk2i * x0r; + x0r = x1r - x3i; + x0i = x1i + x3r; + a[j1] = wk1r * x0r - wk1i * x0i; + a[j1 + 1] = wk1r * x0i + wk1i * x0r; + x0r = x1r + x3i; + x0i = x1i - x3r; + a[j3] = wk3r * x0r - wk3i * x0i; + a[j3 + 1] = wk3r * x0i + wk3i * x0r; + } + wk1r = w[k2 + 2]; + wk1i = w[k2 + 3]; + wk3r = wk1r - 2 * wk2r * wk1i; + wk3i = 2 * wk2r * wk1r - wk1i; + for (j = k + m; j < l + (k + m); j += 2) { + j1 = j + l; + j2 = j1 + l; + j3 = j2 + l; + x0r = a[j] + a[j1]; + x0i = a[j + 1] + a[j1 + 1]; + x1r = a[j] - a[j1]; + x1i = a[j + 1] - a[j1 + 1]; + x2r = a[j2] + a[j3]; + x2i = a[j2 + 1] + a[j3 + 1]; + x3r = a[j2] - a[j3]; + x3i = a[j2 + 1] - a[j3 + 1]; + a[j] = x0r + x2r; + a[j + 1] = x0i + x2i; + x0r -= x2r; + x0i -= x2i; + a[j2] = -wk2i * x0r - wk2r * x0i; + a[j2 + 1] = -wk2i * x0i + wk2r * x0r; + x0r = x1r - x3i; + x0i = x1i + x3r; + a[j1] = wk1r * x0r - wk1i * x0i; + a[j1 + 1] = wk1r * x0i + wk1i * x0r; + x0r = x1r + x3i; + x0i = x1i - x3r; + a[j3] = wk3r * x0r - wk3i * x0i; + a[j3 + 1] = wk3r * x0i + wk3i * x0r; + } + } +} + + +static void rftfsub(int n, double *a, int nc, double const *c) +{ + int j, k, kk, ks, m; + double wkr, wki, xr, xi, yr, yi; + + m = n >> 1; + ks = 2 * nc / m; + kk = 0; + for (j = 2; j < m; j += 2) { + k = n - j; + kk += ks; + wkr = one_half - c[nc - kk]; + wki = c[kk]; + xr = a[j] - a[k]; + xi = a[j + 1] + a[k + 1]; + yr = wkr * xr - wki * xi; + yi = wkr * xi + wki * xr; + a[j] -= yr; + a[j + 1] -= yi; + a[k] += yr; + a[k + 1] -= yi; + } +} + + +static void rftbsub(int n, double *a, int nc, double const *c) +{ + int j, k, kk, ks, m; + double wkr, wki, xr, xi, yr, yi; + + a[1] = -a[1]; + m = n >> 1; + ks = 2 * nc / m; + kk = 0; + for (j = 2; j < m; j += 2) { + k = n - j; + kk += ks; + wkr = one_half - c[nc - kk]; + wki = c[kk]; + xr = a[j] - a[k]; + xi = a[j + 1] + a[k + 1]; + yr = wkr * xr + wki * xi; + yi = wkr * xi - wki * xr; + a[j] -= yr; + a[j + 1] = yi - a[j + 1]; + a[k] += yr; + a[k + 1] = yi - a[k + 1]; + } + a[m + 1] = -a[m + 1]; +} + + +static void dctsub(int n, double *a, int nc, double const *c) +{ + int j, k, kk, ks, m; + double wkr, wki, xr; + + m = n >> 1; + ks = nc / n; + kk = 0; + for (j = 1; j < m; j++) { + k = n - j; + kk += ks; + wkr = c[kk] - c[nc - kk]; + wki = c[kk] + c[nc - kk]; + xr = wki * a[j] - wkr * a[k]; + a[j] = wkr * a[j] + wki * a[k]; + a[k] = xr; + } + a[m] *= c[0]; +} + + +static void dstsub(int n, double *a, int nc, double const *c) +{ + int j, k, kk, ks, m; + double wkr, wki, xr; + + m = n >> 1; + ks = nc / n; + kk = 0; + for (j = 1; j < m; j++) { + k = n - j; + kk += ks; + wkr = c[kk] - c[nc - kk]; + wki = c[kk] + c[nc - kk]; + xr = wki * a[k] - wkr * a[j]; + a[k] = wkr * a[k] + wki * a[j]; + a[j] = xr; + } + a[m] *= c[0]; +} diff --git a/soxr-sys/src/fft4g.h b/soxr-sys/src/fft4g.h new file mode 100644 index 000000000..0f906abcf --- /dev/null +++ b/soxr-sys/src/fft4g.h @@ -0,0 +1,23 @@ +/* SoX Resampler Library Copyright (c) 2007-13 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +void lsx_cdft(int, int, double *, int *, double *); +void lsx_rdft(int, int, double *, int *, double *); +void lsx_ddct(int, int, double *, int *, double *); +void lsx_ddst(int, int, double *, int *, double *); +void lsx_dfct(int, double *, double *, int *, double *); +void lsx_dfst(int, double *, double *, int *, double *); + +void lsx_cdft_f(int, int, float *, int *, float *); +void lsx_rdft_f(int, int, float *, int *, float *); +void lsx_ddct_f(int, int, float *, int *, float *); +void lsx_ddst_f(int, int, float *, int *, float *); +void lsx_dfct_f(int, float *, float *, int *, float *); +void lsx_dfst_f(int, float *, float *, int *, float *); + +#define dft_br_len(l) (2ul + (1ul << (int)(log(l / 2 + .5) / log(2.)) / 2)) +#define dft_sc_len(l) ((unsigned long)l / 2) + +/* Over-allocate h by 2 to use these macros */ +#define LSX_PACK(h, n) h[1] = h[n] +#define LSX_UNPACK(h, n) h[n] = h[1], h[n + 1] = h[1] = 0; diff --git a/soxr-sys/src/fft4g32.c b/soxr-sys/src/fft4g32.c new file mode 100644 index 000000000..7a31ba4bb --- /dev/null +++ b/soxr-sys/src/fft4g32.c @@ -0,0 +1,36 @@ +/* SoX Resampler Library Copyright (c) 2007-13 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#include +#include "filter.h" +#define FFT4G_FLOAT +#include "fft4g.c" +#include "soxr-config.h" + +#if WITH_CR32 +#include "rdft_t.h" +static void * null(void) {return 0;} +static void forward (int length, void * setup, double * H) {lsx_safe_rdft_f(length, 1, H); (void)setup;} +static void backward(int length, void * setup, double * H) {lsx_safe_rdft_f(length, -1, H); (void)setup;} +static int multiplier(void) {return 2;} +static void nothing(void) {} +static int flags(void) {return 0;} + +fn_t _soxr_rdft32_cb[] = { + (fn_t)null, + (fn_t)null, + (fn_t)nothing, + (fn_t)forward, + (fn_t)forward, + (fn_t)backward, + (fn_t)backward, + (fn_t)_soxr_ordered_convolve_f, + (fn_t)_soxr_ordered_partial_convolve_f, + (fn_t)multiplier, + (fn_t)nothing, + (fn_t)malloc, + (fn_t)calloc, + (fn_t)free, + (fn_t)flags, +}; +#endif diff --git a/soxr-sys/src/fft4g32s.c b/soxr-sys/src/fft4g32s.c new file mode 100644 index 000000000..8ce9726ef --- /dev/null +++ b/soxr-sys/src/fft4g32s.c @@ -0,0 +1,31 @@ +/* SoX Resampler Library Copyright (c) 2007-13 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#include "filter.h" +#include "util32s.h" +#include "rdft_t.h" + +static void * null(void) {return 0;} +static void nothing(void) {} +static void forward (int length, void * setup, float * H) {lsx_safe_rdft_f(length, 1, H); (void)setup;} +static void backward(int length, void * setup, float * H) {lsx_safe_rdft_f(length, -1, H); (void)setup;} +static int multiplier(void) {return 2;} +static int flags(void) {return RDFT_IS_SIMD;} + +fn_t _soxr_rdft32s_cb[] = { + (fn_t)null, + (fn_t)null, + (fn_t)nothing, + (fn_t)forward, + (fn_t)forward, + (fn_t)backward, + (fn_t)backward, + (fn_t)ORDERED_CONVOLVE_SIMD, + (fn_t)ORDERED_PARTIAL_CONVOLVE_SIMD, + (fn_t)multiplier, + (fn_t)nothing, + (fn_t)SIMD_ALIGNED_MALLOC, + (fn_t)SIMD_ALIGNED_CALLOC, + (fn_t)SIMD_ALIGNED_FREE, + (fn_t)flags, +}; diff --git a/soxr-sys/src/fft4g64.c b/soxr-sys/src/fft4g64.c new file mode 100644 index 000000000..0018516a0 --- /dev/null +++ b/soxr-sys/src/fft4g64.c @@ -0,0 +1,35 @@ +/* SoX Resampler Library Copyright (c) 2007-13 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#include +#include "filter.h" +#include "fft4g.c" +#include "soxr-config.h" + +#if WITH_CR64 +static void * null(void) {return 0;} +static void nothing(void) {} +static void forward (int length, void * setup, double * H) {lsx_safe_rdft(length, 1, H); (void)setup;} +static void backward(int length, void * setup, double * H) {lsx_safe_rdft(length, -1, H); (void)setup;} +static int multiplier(void) {return 2;} +static int flags(void) {return 0;} + +typedef void (* fn_t)(void); +fn_t _soxr_rdft64_cb[] = { + (fn_t)null, + (fn_t)null, + (fn_t)nothing, + (fn_t)forward, + (fn_t)forward, + (fn_t)backward, + (fn_t)backward, + (fn_t)_soxr_ordered_convolve, + (fn_t)_soxr_ordered_partial_convolve, + (fn_t)multiplier, + (fn_t)nothing, + (fn_t)malloc, + (fn_t)calloc, + (fn_t)free, + (fn_t)flags, +}; +#endif diff --git a/soxr-sys/src/fft4g_cache.h b/soxr-sys/src/fft4g_cache.h new file mode 100644 index 000000000..d776c16c4 --- /dev/null +++ b/soxr-sys/src/fft4g_cache.h @@ -0,0 +1,92 @@ +/* SoX Resampler Library Copyright (c) 2007-13 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +static int * LSX_FFT_BR; +static DFT_FLOAT * LSX_FFT_SC; +static int FFT_LEN = -1; +static ccrw2_t FFT_CACHE_CCRW; + +void LSX_INIT_FFT_CACHE(void) +{ + if (FFT_LEN >= 0) + return; + assert(LSX_FFT_BR == NULL); + assert(LSX_FFT_SC == NULL); + assert(FFT_LEN == -1); + ccrw2_init(FFT_CACHE_CCRW); + FFT_LEN = 0; +} + +void LSX_CLEAR_FFT_CACHE(void) +{ + assert(FFT_LEN >= 0); + ccrw2_clear(FFT_CACHE_CCRW); + free(LSX_FFT_BR); + free(LSX_FFT_SC); + LSX_FFT_SC = NULL; + LSX_FFT_BR = NULL; + FFT_LEN = -1; +} + +static bool UPDATE_FFT_CACHE(int len) +{ + LSX_INIT_FFT_CACHE(); + assert(lsx_is_power_of_2(len)); + assert(FFT_LEN >= 0); + ccrw2_become_reader(FFT_CACHE_CCRW); + if (len > FFT_LEN) { + ccrw2_cease_reading(FFT_CACHE_CCRW); + ccrw2_become_writer(FFT_CACHE_CCRW); + if (len > FFT_LEN) { + int old_n = FFT_LEN; + FFT_LEN = len; + LSX_FFT_BR = realloc(LSX_FFT_BR, dft_br_len(FFT_LEN) * sizeof(*LSX_FFT_BR)); + LSX_FFT_SC = realloc(LSX_FFT_SC, dft_sc_len(FFT_LEN) * sizeof(*LSX_FFT_SC)); + if (!old_n) { + LSX_FFT_BR[0] = 0; +#if SOXR_LIB + atexit(LSX_CLEAR_FFT_CACHE); +#endif + } + return true; + } + ccrw2_cease_writing(FFT_CACHE_CCRW); + ccrw2_become_reader(FFT_CACHE_CCRW); + } + return false; +} + +static void DONE_WITH_FFT_CACHE(bool is_writer) +{ + if (is_writer) + ccrw2_cease_writing(FFT_CACHE_CCRW); + else ccrw2_cease_reading(FFT_CACHE_CCRW); +} + +void LSX_SAFE_RDFT(int len, int type, DFT_FLOAT * d) +{ + bool is_writer = UPDATE_FFT_CACHE(len); + LSX_RDFT(len, type, d, LSX_FFT_BR, LSX_FFT_SC); + DONE_WITH_FFT_CACHE(is_writer); +} + +void LSX_SAFE_CDFT(int len, int type, DFT_FLOAT * d) +{ + bool is_writer = UPDATE_FFT_CACHE(len); + LSX_CDFT(len, type, d, LSX_FFT_BR, LSX_FFT_SC); + DONE_WITH_FFT_CACHE(is_writer); +} + +#undef UPDATE_FFT_CACHE +#undef LSX_SAFE_RDFT +#undef LSX_SAFE_CDFT +#undef LSX_RDFT +#undef LSX_INIT_FFT_CACHE +#undef LSX_FFT_SC +#undef LSX_FFT_BR +#undef LSX_CLEAR_FFT_CACHE +#undef LSX_CDFT +#undef FFT_LEN +#undef FFT_CACHE_CCRW +#undef DONE_WITH_FFT_CACHE +#undef DFT_FLOAT diff --git a/soxr-sys/src/fifo.h b/soxr-sys/src/fifo.h new file mode 100644 index 000000000..33af9fe63 --- /dev/null +++ b/soxr-sys/src/fifo.h @@ -0,0 +1,125 @@ +/* SoX Resampler Library Copyright (c) 2007-13 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#ifndef soxr_fifo_included +#define soxr_fifo_included + +#if !defined FIFO_SIZE_T +#define FIFO_SIZE_T size_t +#endif + +#if !defined FIFO_REALLOC +#include + #define FIFO_REALLOC(a,b,c) realloc(a,b) + #undef FIFO_FREE + #define FIFO_FREE free + #undef FIFO_MALLOC + #define FIFO_MALLOC malloc +#endif + +typedef struct { + char * data; + size_t allocation; /* Number of bytes allocated for data. */ + size_t item_size; /* Size of each item in data */ + size_t begin; /* Offset of the first byte to read. */ + size_t end; /* 1 + Offset of the last byte byte to read. */ +} fifo_t; + +#if !defined FIFO_MIN + #define FIFO_MIN 0x4000 +#endif + +#if !defined UNUSED + #define UNUSED +#endif + +UNUSED static void fifo_clear(fifo_t * f) +{ + f->end = f->begin = 0; +} + +UNUSED static void * fifo_reserve(fifo_t * f, FIFO_SIZE_T n0) +{ + size_t n = (size_t)n0; + n *= f->item_size; + + if (f->begin == f->end) + fifo_clear(f); + + while (1) { + if (f->end + n <= f->allocation) { + void *p = f->data + f->end; + + f->end += n; + return p; + } + if (f->begin > FIFO_MIN) { + memmove(f->data, f->data + f->begin, f->end - f->begin); + f->end -= f->begin; + f->begin = 0; + continue; + } + f->data = FIFO_REALLOC(f->data, f->allocation + n, f->allocation); + f->allocation += n; + if (!f->data) + return 0; + } +} + +UNUSED static void * fifo_write(fifo_t * f, FIFO_SIZE_T n0, void const * data) +{ + size_t n = (size_t)n0; + void * s = fifo_reserve(f, n0); + if (data) + memcpy(s, data, n * f->item_size); + return s; +} + +UNUSED static void fifo_trim_to(fifo_t * f, FIFO_SIZE_T n0) +{ + size_t n = (size_t)n0; + n *= f->item_size; + f->end = f->begin + n; +} + +UNUSED static void fifo_trim_by(fifo_t * f, FIFO_SIZE_T n0) +{ + size_t n = (size_t)n0; + n *= f->item_size; + f->end -= n; +} + +UNUSED static FIFO_SIZE_T fifo_occupancy(fifo_t * f) +{ + return (FIFO_SIZE_T)((f->end - f->begin) / f->item_size); +} + +UNUSED static void * fifo_read(fifo_t * f, FIFO_SIZE_T n0, void * data) +{ + size_t n = (size_t)n0; + char * ret = f->data + f->begin; + n *= f->item_size; + if (n > (f->end - f->begin)) + return NULL; + if (data) + memcpy(data, ret, (size_t)n); + f->begin += n; + return ret; +} + +#define fifo_read_ptr(f) fifo_read(f, (FIFO_SIZE_T)0, NULL) + +UNUSED static void fifo_delete(fifo_t * f) +{ + FIFO_FREE(f->data); +} + +UNUSED static int fifo_create(fifo_t * f, FIFO_SIZE_T item_size) +{ + f->item_size = (size_t)item_size; + f->allocation = FIFO_MIN; + fifo_clear(f); + return !(f->data = FIFO_MALLOC(f->allocation)); +} + +#endif diff --git a/soxr-sys/src/filter.c b/soxr-sys/src/filter.c new file mode 100644 index 000000000..019d24d90 --- /dev/null +++ b/soxr-sys/src/filter.c @@ -0,0 +1,277 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#include "filter.h" + +#include "math-wrap.h" +#include +#include +#include + +#include "fft4g.h" +#include "ccrw2.h" + +#if 1 || WITH_CR64 || WITH_CR64S /* Always need this, for lsx_fir_to_phase. */ +#define DFT_FLOAT double +#define DONE_WITH_FFT_CACHE done_with_fft_cache +#define FFT_CACHE_CCRW fft_cache_ccrw +#define FFT_LEN fft_len +#define LSX_CDFT lsx_cdft +#define LSX_CLEAR_FFT_CACHE lsx_clear_fft_cache +#define LSX_FFT_BR lsx_fft_br +#define LSX_FFT_SC lsx_fft_sc +#define LSX_INIT_FFT_CACHE lsx_init_fft_cache +#define LSX_RDFT lsx_rdft +#define LSX_SAFE_CDFT lsx_safe_cdft +#define LSX_SAFE_RDFT lsx_safe_rdft +#define UPDATE_FFT_CACHE update_fft_cache +#include "fft4g_cache.h" +#endif + +#if (WITH_CR32 && !AVCODEC_FOUND) || (WITH_CR32S && !AVCODEC_FOUND && !WITH_PFFFT) +#define DFT_FLOAT float +#define DONE_WITH_FFT_CACHE done_with_fft_cache_f +#define FFT_CACHE_CCRW fft_cache_ccrw_f +#define FFT_LEN fft_len_f +#define LSX_CDFT lsx_cdft_f +#define LSX_CLEAR_FFT_CACHE lsx_clear_fft_cache_f +#define LSX_FFT_BR lsx_fft_br_f +#define LSX_FFT_SC lsx_fft_sc_f +#define LSX_INIT_FFT_CACHE lsx_init_fft_cache_f +#define LSX_RDFT lsx_rdft_f +#define LSX_SAFE_CDFT lsx_safe_cdft_f +#define LSX_SAFE_RDFT lsx_safe_rdft_f +#define UPDATE_FFT_CACHE update_fft_cache_f +#include "fft4g_cache.h" +#endif + +#if WITH_CR64 || WITH_CR64S || !SOXR_LIB +#define DFT_FLOAT double +#define ORDERED_CONVOLVE lsx_ordered_convolve +#define ORDERED_PARTIAL_CONVOLVE lsx_ordered_partial_convolve +#include "rdft.h" +#endif + +#if WITH_CR32 +#define DFT_FLOAT float +#define ORDERED_CONVOLVE lsx_ordered_convolve_f +#define ORDERED_PARTIAL_CONVOLVE lsx_ordered_partial_convolve_f +#include "rdft.h" +#endif + +double lsx_kaiser_beta(double att, double tr_bw) +{ + if (att >= 60) { + static const double coefs[][4] = { + {-6.784957e-10,1.02856e-05,0.1087556,-0.8988365+.001}, + {-6.897885e-10,1.027433e-05,0.10876,-0.8994658+.002}, + {-1.000683e-09,1.030092e-05,0.1087677,-0.9007898+.003}, + {-3.654474e-10,1.040631e-05,0.1087085,-0.8977766+.006}, + {8.106988e-09,6.983091e-06,0.1091387,-0.9172048+.015}, + {9.519571e-09,7.272678e-06,0.1090068,-0.9140768+.025}, + {-5.626821e-09,1.342186e-05,0.1083999,-0.9065452+.05}, + {-9.965946e-08,5.073548e-05,0.1040967,-0.7672778+.085}, + {1.604808e-07,-5.856462e-05,0.1185998,-1.34824+.1}, + {-1.511964e-07,6.363034e-05,0.1064627,-0.9876665+.18}, + }; + double realm = log(tr_bw/.0005)/log(2.); + double const * c0 = coefs[range_limit( (int)realm, 0, (int)array_length(coefs)-1)]; + double const * c1 = coefs[range_limit(1+(int)realm, 0, (int)array_length(coefs)-1)]; + double b0 = ((c0[0]*att + c0[1])*att + c0[2])*att + c0[3]; + double b1 = ((c1[0]*att + c1[1])*att + c1[2])*att + c1[3]; + return b0 + (b1 - b0) * (realm - (int)realm); + } + if (att > 50 ) return .1102 * (att - 8.7); + if (att > 20.96) return .58417 * pow(att -20.96, .4) + .07886 * (att - 20.96); + return 0; +} + +double * lsx_make_lpf( + int num_taps, double Fc, double beta, double rho, double scale) +{ + int i, m = num_taps - 1; + double * h = malloc((size_t)num_taps * sizeof(*h)); + double mult = scale / lsx_bessel_I_0(beta), mult1 = 1 / (.5 * m + rho); + assert(Fc >= 0 && Fc <= 1); + lsx_debug("make_lpf(n=%i Fc=%.7g beta=%g rho=%g scale=%g)", + num_taps, Fc, beta, rho, scale); + + if (h) for (i = 0; i <= m / 2; ++i) { + double z = i - .5 * m, x = z * M_PI, y = z * mult1; + h[i] = x!=0? sin(Fc * x) / x : Fc; + h[i] *= lsx_bessel_I_0(beta * sqrt(1 - y * y)) * mult; + if (m - i != i) + h[m - i] = h[i]; + } + return h; +} + +void lsx_kaiser_params(double att, double Fc, double tr_bw, double * beta, int * num_taps) +{ + *beta = *beta < 0? lsx_kaiser_beta(att, tr_bw * .5 / Fc): *beta; + att = att < 60? (att - 7.95) / (2.285 * M_PI * 2) : + ((.0007528358-1.577737e-05**beta)**beta+.6248022)**beta+.06186902; + *num_taps = !*num_taps? (int)ceil(att/tr_bw + 1) : *num_taps; +} + +double * lsx_design_lpf( + double Fp, /* End of pass-band */ + double Fs, /* Start of stop-band */ + double Fn, /* Nyquist freq; e.g. 0.5, 1, PI */ + double att, /* Stop-band attenuation in dB */ + int * num_taps, /* 0: value will be estimated */ + int k, /* >0: number of phases; <0: num_taps = 1 (mod -k) */ + double beta) /* <0: value will be estimated */ +{ + int n = *num_taps, phases = max(k, 1), modulo = max(-k, 1); + double tr_bw, Fc, rho = phases == 1? .5 : att < 120? .63 : .75; + + lsx_debug_more("./sinctest %-12.7g %-12.7g %g 0 %-5g %i %i 50 %g %g -4 >1", + Fp, Fs, Fn, att, *num_taps, k, beta, rho); + + Fp /= fabs(Fn), Fs /= fabs(Fn); /* Normalise to Fn = 1 */ + tr_bw = .5 * (Fs - Fp); /* Transition band-width: 6dB to stop points */ + tr_bw /= phases, Fs /= phases; + tr_bw = min(tr_bw, .5 * Fs); + Fc = Fs - tr_bw; + assert(Fc - tr_bw >= 0); + lsx_kaiser_params(att, Fc, tr_bw, &beta, num_taps); + if (!n) + *num_taps = phases > 1? *num_taps / phases * phases + phases - 1 : + (*num_taps + modulo - 2) / modulo * modulo + 1; + return Fn < 0? 0 : lsx_make_lpf(*num_taps, Fc, beta, rho, (double)phases); +} + +static double safe_log(double x) +{ + assert(x >= 0); + if (x!=0) + return log(x); + lsx_debug("log(0)"); + return -26; +} + +void lsx_fir_to_phase(double * * h, int * len, int * post_len, double phase) +{ + double * pi_wraps, * work, phase1 = (phase > 50 ? 100 - phase : phase) / 50; + int i, work_len, begin, end, imp_peak = 0, peak = 0; + double imp_sum = 0, peak_imp_sum = 0; + double prev_angle2 = 0, cum_2pi = 0, prev_angle1 = 0, cum_1pi = 0; + + for (i = *len, work_len = 2 * 2 * 8; i > 1; work_len <<= 1, i >>= 1); + + work = calloc((size_t)work_len + 2, sizeof(*work)); /* +2: (UN)PACK */ + pi_wraps = malloc((((size_t)work_len + 2) / 2) * sizeof(*pi_wraps)); + + memcpy(work, *h, (size_t)*len * sizeof(*work)); + lsx_safe_rdft(work_len, 1, work); /* Cepstral: */ + LSX_UNPACK(work, work_len); + + for (i = 0; i <= work_len; i += 2) { + double angle = atan2(work[i + 1], work[i]); + double detect = 2 * M_PI; + double delta = angle - prev_angle2; + double adjust = detect * ((delta < -detect * .7) - (delta > detect * .7)); + prev_angle2 = angle; + cum_2pi += adjust; + angle += cum_2pi; + detect = M_PI; + delta = angle - prev_angle1; + adjust = detect * ((delta < -detect * .7) - (delta > detect * .7)); + prev_angle1 = angle; + cum_1pi += fabs(adjust); /* fabs for when 2pi and 1pi have combined */ + pi_wraps[i >> 1] = cum_1pi; + + work[i] = safe_log(sqrt(sqr(work[i]) + sqr(work[i + 1]))); + work[i + 1] = 0; + } + LSX_PACK(work, work_len); + lsx_safe_rdft(work_len, -1, work); + for (i = 0; i < work_len; ++i) work[i] *= 2. / work_len; + + for (i = 1; i < work_len / 2; ++i) { /* Window to reject acausal components */ + work[i] *= 2; + work[i + work_len / 2] = 0; + } + lsx_safe_rdft(work_len, 1, work); + + for (i = 2; i < work_len; i += 2) /* Interpolate between linear & min phase */ + work[i + 1] = phase1 * i / work_len * pi_wraps[work_len >> 1] + + (1 - phase1) * (work[i + 1] + pi_wraps[i >> 1]) - pi_wraps[i >> 1]; + + work[0] = exp(work[0]), work[1] = exp(work[1]); + for (i = 2; i < work_len; i += 2) { + double x = exp(work[i]); + work[i ] = x * cos(work[i + 1]); + work[i + 1] = x * sin(work[i + 1]); + } + + lsx_safe_rdft(work_len, -1, work); + for (i = 0; i < work_len; ++i) work[i] *= 2. / work_len; + + /* Find peak pos. */ + for (i = 0; i <= (int)(pi_wraps[work_len >> 1] / M_PI + .5); ++i) { + imp_sum += work[i]; + if (fabs(imp_sum) > fabs(peak_imp_sum)) { + peak_imp_sum = imp_sum; + peak = i; + } + if (work[i] > work[imp_peak]) /* For debug check only */ + imp_peak = i; + } + while (peak && fabs(work[peak-1]) > fabs(work[peak]) && work[peak-1] * work[peak] > 0) + --peak; + + if (phase1==0) + begin = 0; + else if (phase1 == 1) + begin = peak - *len / 2; + else { + begin = (int)((.997 - (2 - phase1) * .22) * *len + .5); + end = (int)((.997 + (0 - phase1) * .22) * *len + .5); + begin = peak - (begin & ~3); + end = peak + 1 + ((end + 3) & ~3); + *len = end - begin; + *h = realloc(*h, (size_t)*len * sizeof(**h)); + } + for (i = 0; i < *len; ++i) (*h)[i] = + work[(begin + (phase > 50 ? *len - 1 - i : i) + work_len) & (work_len - 1)]; + *post_len = phase > 50 ? peak - begin : begin + *len - (peak + 1); + + lsx_debug("nPI=%g peak-sum@%i=%g (val@%i=%g); len=%i post=%i (%g%%)", + pi_wraps[work_len >> 1] / M_PI, peak, peak_imp_sum, imp_peak, + work[imp_peak], *len, *post_len, 100 - 100. * *post_len / (*len - 1)); + free(pi_wraps), free(work); +} + +#define F_x(F,expr) static double F(double x) {return expr;} +F_x(sinePhi, ((2.0517e-07*x-1.1303e-04)*x+.023154)*x+.55924 ) +F_x(sinePsi, ((9.0667e-08*x-5.6114e-05)*x+.013658)*x+1.0977 ) +F_x(sinePow, log(.5)/log(sin(x*.5)) ) +#define dB_to_linear(x) exp((x) * (M_LN10 * 0.05)) + +double lsx_f_resp(double t, double a) +{ + double x; + if (t > (a <= 160? .8 : .82)) { + double a1 = a+15; + double p = .00035*a+.375; + double w = 1/(1-.597)*asin(pow((a1-10.6)/a1,1/p)); + double c = 1+asin(pow(1-a/a1,1/p))/w; + return a1*(pow(sin((c-t)*w),p)-1); + } + if (t > .5) + x = sinePsi(a), x = pow(sin((1-t) * x), sinePow(x)); + else + x = sinePhi(a), x = 1 - pow(sin(t * x), sinePow(x)); + return linear_to_dB(x); +} + +double lsx_inv_f_resp(double drop, double a) +{ + double x = sinePhi(a), s; + drop = dB_to_linear(drop); + s = drop > .5 ? 1 - drop : drop; + x = asin(pow(s, 1/sinePow(x))) / x; + return drop > .5? x : 1 -x; +} diff --git a/soxr-sys/src/filter.h b/soxr-sys/src/filter.h new file mode 100644 index 000000000..ccb3ba836 --- /dev/null +++ b/soxr-sys/src/filter.h @@ -0,0 +1,44 @@ +/* SoX Resampler Library Copyright (c) 2007-13 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#if !defined soxr_filter_included +#define soxr_filter_included + +#include "aliases.h" + +double lsx_bessel_I_0(double x); +void lsx_init_fft_cache(void); +void lsx_clear_fft_cache(void); +void lsx_init_fft_cache_f(void); +void lsx_clear_fft_cache_f(void); +#define lsx_is_power_of_2(x) !(x < 2 || (x & (x - 1))) +void lsx_safe_rdft(int len, int type, double * d); +void lsx_safe_cdft(int len, int type, double * d); +void lsx_safe_rdft_f(int len, int type, float * d); +void lsx_safe_cdft_f(int len, int type, float * d); +void lsx_ordered_convolve(int n, void * not_used, double * a, const double * b); +void lsx_ordered_convolve_f(int n, void * not_used, float * a, const float * b); +void lsx_ordered_partial_convolve(int n, double * a, const double * b); +void lsx_ordered_partial_convolve_f(int n, float * a, const float * b); + +double lsx_kaiser_beta(double att, double tr_bw); +double * lsx_make_lpf(int num_taps, double Fc, double beta, double rho, + double scale); +void lsx_kaiser_params(double att, double Fc, double tr_bw, double * beta, int * num_taps); +double * lsx_design_lpf( + double Fp, /* End of pass-band */ + double Fs, /* Start of stop-band */ + double Fn, /* Nyquist freq; e.g. 0.5, 1, PI; < 0: dummy run */ + double att, /* Stop-band attenuation in dB */ + int * num_taps, /* 0: value will be estimated */ + int k, /* >0: number of phases; <0: num_taps = 1 (mod -k) */ + double beta); /* <0: value will be estimated */ + +void lsx_fir_to_phase(double * * h, int * len, + int * post_len, double phase0); + +double lsx_f_resp(double t, double a); +double lsx_inv_f_resp(double drop, double a); +#define lsx_to_3dB(a) (1 - lsx_inv_f_resp(-3., a)) + +#endif diff --git a/soxr-sys/src/half-coefs.h b/soxr-sys/src/half-coefs.h new file mode 100644 index 000000000..a5a0882bc --- /dev/null +++ b/soxr-sys/src/half-coefs.h @@ -0,0 +1,75 @@ +/* SoX Resampler Library Copyright (c) 2007-13 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#if defined __GNUC__ + #pragma GCC system_header +#elif defined __SUNPRO_C + #pragma disable_warn +#elif defined _MSC_VER + #pragma warning(push, 1) +#endif + +#if CORE_TYPE & CORE_SIMD_HALF + #define VALIGN vAlign +#else + #define VALIGN +#endif + +#if !(CORE_TYPE & CORE_SIMD_HALF) +static VALIGN const sample_t half_fir_coefs_7[] = { + 3.1062656496657370e-01, -8.4998810699955796e-02, 3.4007044621123500e-02, +-1.2839903789829387e-02, 3.9899380181723145e-03, -8.9355202017945374e-04, + 1.0918292424806546e-04, +}; +#endif + +static VALIGN const sample_t half_fir_coefs_8[] = { + 3.1154652365332069e-01, -8.7344917685739543e-02, 3.6814458353637280e-02, +-1.5189204581464479e-02, 5.4540855610738801e-03, -1.5643862626630416e-03, + 3.1816575906323303e-04, -3.4799449225005688e-05, +}; + +static VALIGN const sample_t half_fir_coefs_9[] = { + 3.1227034755311189e-01, -8.9221517147969526e-02, 3.9139704015071934e-02, +-1.7250558515852023e-02, 6.8589440230476112e-03, -2.3045049636430419e-03, + 6.0963740543348963e-04, -1.1323803957431231e-04, 1.1197769991000046e-05, +}; + +#if CORE_TYPE & CORE_DBL +static VALIGN const sample_t half_fir_coefs_10[] = { + 3.1285456012000523e-01, -9.0756740799292787e-02, 4.1096398104193160e-02, +-1.9066319572525220e-02, 8.1840569787684902e-03, -3.0766876176359834e-03, + 9.6396524429277980e-04, -2.3585679989922018e-04, 4.0252189026627833e-05, +-3.6298196342497932e-06, +}; + +static VALIGN const sample_t half_fir_coefs_11[] = { + 3.1333588822574199e-01, -9.2035898673019811e-02, 4.2765169698406408e-02, +-2.0673580894964429e-02, 9.4225426824512421e-03, -3.8563379950013192e-03, + 1.3634742159642453e-03, -3.9874150714431009e-04, 9.0586723632664806e-05, +-1.4285617244076783e-05, 1.1834642946400529e-06, +}; + +static VALIGN const sample_t half_fir_coefs_12[] = { + 3.1373928463345568e-01, -9.3118180335301962e-02, 4.4205005881659098e-02, +-2.2103860986973051e-02, 1.0574689371162864e-02, -4.6276428065385065e-03, + 1.7936153397572132e-03, -5.9617527051353237e-04, 1.6314517495669067e-04, +-3.4555126770115446e-05, 5.0617615610782593e-06, -3.8768958592971409e-07, +}; + +static VALIGN const sample_t half_fir_coefs_13[] = { + 3.1408224847888910e-01, -9.4045836332667387e-02, 4.5459878763259978e-02, +-2.3383369012219993e-02, 1.1644273044890753e-02, -5.3806714579057013e-03, + 2.2429072878264022e-03, -8.2204347506606424e-04, 2.5724946477840893e-04, +-6.6072709864248668e-05, 1.3099163296288644e-05, -1.7907147069136000e-06, + 1.2750825595240592e-07, +}; +#endif + +#undef VALIGN + +#if defined __SUNPRO_C + #pragma enable_warn +#elif defined _MSC_VER + #pragma warning(pop) +#endif diff --git a/soxr-sys/src/half-fir.h b/soxr-sys/src/half-fir.h new file mode 100644 index 000000000..782be1bc7 --- /dev/null +++ b/soxr-sys/src/half-fir.h @@ -0,0 +1,61 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +/* Decimate by 2 using a FIR with odd length (LEN). */ +/* Input must be preceded and followed by LEN >> 1 samples. */ + +#define COEFS ((sample_t const *)p->coefs) + +#if SIMD_SSE + #define BEGINNING v4_t sum, q1, q2, t + #define ____ \ + q1 = _mm_shuffle_ps(t=vLdu(input+2*j),vLdu(input+2*j+4),_MM_SHUFFLE(3,1,3,1)); \ + q2 = _mm_shuffle_ps(vLdu(input-2*j-4),vLdu(input-2*j-8),_MM_SHUFFLE(1,3,1,3)); \ + sum = vAdd(j? sum : vMul(vSet1(.5), t), vMul(vAdd(q1, q2), vLd(COEFS+j))); \ + j += 4; + #define __ \ + q1 = _mm_shuffle_ps(vLdu(input+2*j), vLdu(input-2*j-4), _MM_SHUFFLE(1,3,3,1)); \ + q2 = _mm_loadl_pi(q2, (__m64*)(COEFS+j)), q2 = _mm_movelh_ps(q2, q2); \ + sum = vAdd(sum, vMul(q1, q2)); \ + j += 2; + #define _ \ + q1 = _mm_add_ss(_mm_load_ss(input+2*j+1), _mm_load_ss(input-2*j-1)); \ + sum = _mm_add_ss(sum, _mm_mul_ss(q1, _mm_load_ss(COEFS+j))); \ + ++j; + #define END vStorSum(output+i, sum) +/* #elif SIMD_AVX; No good solution found. */ +/* #elif SIMD_NEON; No need: gcc -O3 does a good job by itself. */ +#else + #define BEGINNING sample_t sum = input[0] * .5f + #define ____ __ __ + #define __ _ _ + #define _ sum += (input[-(2*j +1)] + input[(2*j +1)]) * COEFS[j], ++j; + #define END output[i] = sum +#endif + + + +static void FUNCTION_H(stage_t * p, fifo_t * output_fifo) +{ + sample_t const * __restrict input = stage_read_p(p); + int num_in = min(stage_occupancy(p), p->input_size); + int i, num_out = (num_in + 1) >> 1; + sample_t * __restrict output = fifo_reserve(output_fifo, num_out); + + for (i = 0; i < num_out; ++i, input += 2) { + int j = 0; + BEGINNING; CONVOLVE; END; + } + fifo_read(&p->fifo, 2 * num_out, NULL); +} + + + +#undef _ +#undef __ +#undef ____ +#undef BEGINNING +#undef END +#undef COEFS +#undef CONVOLVE +#undef FUNCTION_H diff --git a/soxr-sys/src/internal.h b/soxr-sys/src/internal.h new file mode 100644 index 000000000..08924d500 --- /dev/null +++ b/soxr-sys/src/internal.h @@ -0,0 +1,84 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#if !defined soxr_internal_included +#define soxr_internal_included + +#include "std-types.h" + + + +#undef min +#undef max +#define min(a, b) ((a) <= (b) ? (a) : (b)) +#define max(a, b) ((a) >= (b) ? (a) : (b)) + + + +#define range_limit(x, lower, upper) (min(max(x, lower), upper)) +#define linear_to_dB(x) (log10(x) * 20) +#define array_length(a) (sizeof(a)/sizeof(a[0])) +#if !defined AL +#define AL(a) array_length(a) +#endif +#define iAL(a) (int)AL(a) +#define sqr(a) ((a) * (a)) + + + +#if defined __GNUC__ + #define UNUSED __attribute__ ((unused)) +#else + #define UNUSED +#endif + + + +#if !WITH_DEV_TRACE + #ifdef __GNUC__ + void lsx_dummy(char const *, ...); + #else + static __inline void lsx_dummy(char const * x, ...) {} + #endif + #define lsx_debug if(0) lsx_dummy + #define lsx_debug_more lsx_debug +#else + extern int _soxr_trace_level; + void _soxr_trace(char const * fmt, ...); + #define lsx_debug if (_soxr_trace_level > 0) _soxr_trace + #define lsx_debug_more if (_soxr_trace_level > 1) _soxr_trace +#endif + + + +/* soxr_quality_spec_t.flags: */ + +#define SOXR_ROLLOFF_LSR2Q 3u /* Reserved for internal use. */ +#define SOXR_ROLLOFF_MASK 3u /* For masking these bits. */ +#define SOXR_MAINTAIN_3DB_PT 4u /* Reserved for internal use. */ +#define SOXR_PROMOTE_TO_LQ 64u /* Reserved for internal use. */ + + + +/* soxr_runtime_spec_t.flags: */ + +#define SOXR_STRICT_BUFFERING 4u /* Reserved for future use. */ +#define SOXR_NOSMALLINTOPT 8u /* For test purposes only. */ + + + +/* soxr_quality_spec recipe: */ + +#define SOXR_PRECISIONQ 11 /* Quality specified by the precision parameter. */ + +#define SOXR_PHASE_MASK 0x30 /* For masking these bits. */ + + + +/* soxr_quality_spec flags: */ + +#define RESET_ON_CLEAR (1u<<31) + + + +#endif diff --git a/soxr-sys/src/lib.rs b/soxr-sys/src/lib.rs new file mode 100644 index 000000000..31c0d8dbf --- /dev/null +++ b/soxr-sys/src/lib.rs @@ -0,0 +1,152 @@ +#![allow(non_upper_case_globals)] +#![allow(non_camel_case_types)] + +include!("soxr.rs"); + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn it_works() { + unsafe { + let version = soxr_version(); + let version = std::ffi::CStr::from_ptr(version).to_str().unwrap(); + println!("version: {}", version); + } + } + + #[test] + fn test_stream() { + use std::ffi::c_void; + use std::ffi::CStr; + use std::ptr; + + use hound::{WavReader, WavSpec, WavWriter}; + + let input_wav_path = "input.wav"; + let output_wav_path = "output.wav"; + + let mut reader = WavReader::open(input_wav_path).expect("Failed to open input WAV file"); + + let wav_spec = reader.spec(); + let input_rate = wav_spec.sample_rate as f64; + let output_rate = 44100.0; + + let num_channels = wav_spec.channels as u32; + + let samples: Vec = reader + .samples::() + .map(|s| s.expect("Failed to read sample")) + .collect(); + + let buf_total_len = samples.len(); + let olen = + ((output_rate * buf_total_len as f64) / (input_rate + output_rate) + 0.5) as usize; + let ilen = buf_total_len - olen; + + let mut obuf = vec![0i16; olen]; + + let mut odone: usize = 0; + let mut need_input = true; + + let mut error: soxr_error_t = ptr::null(); + + let io_spec = soxr_io_spec { + itype: SOXR_INT16_I as u32, + otype: SOXR_INT16_I as u32, + scale: 1.0, + e: ptr::null_mut(), + flags: 0, + }; + + let soxr = unsafe { + soxr_create( + input_rate, + output_rate, + num_channels, + &mut error, + &io_spec, + ptr::null(), + ptr::null(), + ) + }; + + if error.is_null() { + let mut input_pos = 0; + let input_len = samples.len(); + + let mut output_samples = Vec::new(); + + while error.is_null() && (need_input || odone > 0) { + let mut ilen1 = 0; + let mut ibuf: Option<&[i16]> = None; + + if need_input { + if input_pos < input_len { + let remaining_samples = input_len - input_pos; + let samples_to_read = std::cmp::min(ilen, remaining_samples); + + ibuf = Some(&samples[input_pos..input_pos + samples_to_read]); + ilen1 = samples_to_read; + input_pos += samples_to_read; + } else { + ibuf = None; + } + } + + let in_ptr = match ibuf { + Some(slice) => slice.as_ptr() as *const c_void, + None => ptr::null(), + }; + + let process_error = unsafe { + soxr_process( + soxr, + in_ptr, + ilen1, + ptr::null_mut(), + obuf.as_mut_ptr() as *mut c_void, + olen, + &mut odone, + ) + }; + + if !process_error.is_null() { + break; + } + + if odone > 0 { + output_samples.extend_from_slice(&obuf[..odone]); + } + + need_input = (odone < olen) && ibuf.is_some(); + } + + let spec = WavSpec { + channels: wav_spec.channels, + sample_rate: output_rate as u32, + bits_per_sample: 16, + sample_format: hound::SampleFormat::Int, + }; + + let mut writer = + WavWriter::create(output_wav_path, spec).expect("Failed to create output WAV file"); + + for sample in output_samples { + writer.write_sample(sample).expect("Failed to write sample"); + } + + writer.finalize().expect("Failed to finalize WAV file"); + + println!("Resampling completed successfully."); + } else { + let error_str = unsafe { CStr::from_ptr(error) }; + eprintln!("Error creating resampler: {}", error_str.to_string_lossy()); + } + + unsafe { + soxr_delete(soxr); + } + } +} diff --git a/soxr-sys/src/math-wrap.h b/soxr-sys/src/math-wrap.h new file mode 100644 index 000000000..8a526f13e --- /dev/null +++ b/soxr-sys/src/math-wrap.h @@ -0,0 +1,31 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#if !defined soxr_math_wrap_included +#define soxr_math_wrap_included + +#include + +#if defined __STRICT_ANSI__ + #define sinf(x) (float)sin ((double)(x)) + #define cosf(x) (float)cos ((double)(x)) + #define atanf(x) (float)atan((double)(x)) +#endif + +#if !defined M_PI + #define M_PI 3.141592653589793238462643383279502884 +#endif + +#if !defined M_LN10 + #define M_LN10 2.302585092994045684017991454684364208 +#endif + +#if !defined M_SQRT2 + #define M_SQRT2 1.414213562373095048801688724209698079 +#endif + +#if !defined M_LN2 + #define M_LN2 0.693147180559945309417232121458176568 +#endif + +#endif diff --git a/soxr-sys/src/pffft-avx.h b/soxr-sys/src/pffft-avx.h new file mode 100644 index 000000000..ace19b57d --- /dev/null +++ b/soxr-sys/src/pffft-avx.h @@ -0,0 +1,40 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +/* AVX support macros */ + +#if !defined soxr_avx_included +#define soxr_avx_included + +#include + +typedef __m256d v4sf; +#define VZERO() _mm256_setzero_pd() +#define VMUL(a,b) _mm256_mul_pd(a,b) +#define VADD(a,b) _mm256_add_pd(a,b) +#define VMADD(a,b,c) VADD(VMUL(a,b),c) /* Note: gcc -mfma will `fuse' these */ +#define VSUB(a,b) _mm256_sub_pd(a,b) +#define LD_PS1(p) _mm256_set1_pd(p) +#define INTERLEAVE2(in1, in2, out1, out2) {v4sf \ + t1 = _mm256_unpacklo_pd(in1, in2), \ + t2 = _mm256_unpackhi_pd(in1, in2); \ + out1 = _mm256_permute2f128_pd(t1,t2,0x20); \ + out2 = _mm256_permute2f128_pd(t1,t2,0x31); } +#define UNINTERLEAVE2(in1, in2, out1, out2) {v4sf \ + t1 = _mm256_permute2f128_pd(in1,in2,0x20), \ + t2 = _mm256_permute2f128_pd(in1,in2,0x31); \ + out1 = _mm256_unpacklo_pd(t1, t2); \ + out2 = _mm256_unpackhi_pd(t1, t2);} +#define VTRANSPOSE4(x0,x1,x2,x3) {v4sf \ + t0 = _mm256_shuffle_pd(x0,x1, 0x0), \ + t2 = _mm256_shuffle_pd(x0,x1, 0xf), \ + t1 = _mm256_shuffle_pd(x2,x3, 0x0), \ + t3 = _mm256_shuffle_pd(x2,x3, 0xf); \ + x0 = _mm256_permute2f128_pd(t0,t1, 0x20); \ + x1 = _mm256_permute2f128_pd(t2,t3, 0x20); \ + x2 = _mm256_permute2f128_pd(t0,t1, 0x31); \ + x3 = _mm256_permute2f128_pd(t2,t3, 0x31);} +#define VSWAPHL(a,b) _mm256_permute2f128_pd(b, a, 0x30) +#define VALIGNED(ptr) ((((long)(ptr)) & 0x1F) == 0) + +#endif diff --git a/soxr-sys/src/pffft-wrap.c b/soxr-sys/src/pffft-wrap.c new file mode 100644 index 000000000..c920f06ea --- /dev/null +++ b/soxr-sys/src/pffft-wrap.c @@ -0,0 +1,110 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#if !defined PFFT_MACROS_ONLY + +#include "math-wrap.h" + +#if PFFFT_DOUBLE + #include "util64s.h" +#else + #include "util32s.h" + #define sin(x) sinf(x) + #define cos(x) cosf(x) +#endif + +#define pffft_aligned_free SIMD_ALIGNED_FREE +#define pffft_aligned_malloc SIMD_ALIGNED_MALLOC +#define pffft_aligned_calloc SIMD_ALIGNED_CALLOC + +#undef inline +#define inline __inline + +#endif + + + +#include "pffft.c" + + + +#if !defined PFFT_MACROS_ONLY + +#if !defined PFFFT_SIMD_DISABLE + +static void pffft_zconvolve(PFFFT_Setup *s, const float *a, const float *b, float *ab) { + int i, Ncvec = s->Ncvec; + const v4sf * /*RESTRICT*/ va = (const v4sf*)a; + const v4sf * RESTRICT vb = (const v4sf*)b; + v4sf * /*RESTRICT*/ vab = (v4sf*)ab; + + float ar, ai, br, bi; + +#ifdef __arm__ + __builtin_prefetch(va); + __builtin_prefetch(vb); + __builtin_prefetch(va+2); + __builtin_prefetch(vb+2); + __builtin_prefetch(va+4); + __builtin_prefetch(vb+4); + __builtin_prefetch(va+6); + __builtin_prefetch(vb+6); +#endif + + assert(VALIGNED(a) && VALIGNED(b) && VALIGNED(ab)); + ar = ((v4sf_union*)va)[0].f[0]; + ai = ((v4sf_union*)va)[1].f[0]; + br = ((v4sf_union*)vb)[0].f[0]; + bi = ((v4sf_union*)vb)[1].f[0]; + + for (i=0; i < Ncvec; i += 2) { + v4sf ar, ai, br, bi; + ar = va[2*i+0]; ai = va[2*i+1]; + br = vb[2*i+0]; bi = vb[2*i+1]; + VCPLXMUL(ar, ai, br, bi); + vab[2*i+0] = ar; + vab[2*i+1] = ai; + ar = va[2*i+2]; ai = va[2*i+3]; + br = vb[2*i+2]; bi = vb[2*i+3]; + VCPLXMUL(ar, ai, br, bi); + vab[2*i+2] = ar; + vab[2*i+3] = ai; + } + if (s->transform == PFFFT_REAL) { + ((v4sf_union*)vab)[0].f[0] = ar*br; + ((v4sf_union*)vab)[1].f[0] = ai*bi; + } +} + +#else + +static void pffft_zconvolve(PFFFT_Setup *s, const float *a, const float *b, float *ab) { + int i, Ncvec = s->Ncvec; + + if (s->transform == PFFFT_REAL) { + /* take care of the fftpack ordering */ + ab[0] = a[0]*b[0]; + ab[2*Ncvec-1] = a[2*Ncvec-1]*b[2*Ncvec-1]; + ++ab; ++a; ++b; --Ncvec; + } + for (i=0; i < Ncvec; ++i) { + float ar, ai, br, bi; + ar = a[2*i+0]; ai = a[2*i+1]; + br = b[2*i+0]; bi = b[2*i+1]; + VCPLXMUL(ar, ai, br, bi); + ab[2*i+0] = ar; + ab[2*i+1] = ai; + } +} + +#endif + +#include + +static void pffft_reorder_back(int length, void * setup, float * data, float * work) +{ + memcpy(work, data, (unsigned)length * sizeof(*work)); + pffft_zreorder(setup, work, data, PFFFT_BACKWARD); +} + +#endif diff --git a/soxr-sys/src/pffft.c b/soxr-sys/src/pffft.c new file mode 100644 index 000000000..46c841e74 --- /dev/null +++ b/soxr-sys/src/pffft.c @@ -0,0 +1,1946 @@ +/* https://bitbucket.org/jpommier/pffft/raw/483453d8f7661058e74aa4e7cf5c27bcd7887e7a/pffft.c + * with minor changes for libsoxr. */ + +/* Copyright (c) 2013 Julien Pommier ( pommier@modartt.com ) + + Based on original fortran 77 code from FFTPACKv4 from NETLIB + (http://www.netlib.org/fftpack), authored by Dr Paul Swarztrauber + of NCAR, in 1985. + + As confirmed by the NCAR fftpack software curators, the following + FFTPACKv5 license applies to FFTPACKv4 sources. My changes are + released under the same terms. + + FFTPACK license: + + http://www.cisl.ucar.edu/css/software/fftpack5/ftpk.html + + Copyright (c) 2004 the University Corporation for Atmospheric + Research ("UCAR"). All rights reserved. Developed by NCAR's + Computational and Information Systems Laboratory, UCAR, + www.cisl.ucar.edu. + + Redistribution and use of the Software in source and binary forms, + with or without modification, is permitted provided that the + following conditions are met: + + - Neither the names of NCAR's Computational and Information Systems + Laboratory, the University Corporation for Atmospheric Research, + nor the names of its sponsors or contributors may be used to + endorse or promote products derived from this Software without + specific prior written permission. + + - Redistributions of source code must retain the above copyright + notices, this list of conditions, and the disclaimer below. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions, and the disclaimer below in the + documentation and/or other materials provided with the + distribution. + + THIS 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 CONTRIBUTORS OR COPYRIGHT + HOLDERS BE LIABLE FOR ANY CLAIM, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL 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 WITH THE + SOFTWARE. + + + PFFFT : a Pretty Fast FFT. + + This file is largerly based on the original FFTPACK implementation, modified in + order to take advantage of SIMD instructions of modern CPUs. +*/ + +/* + ChangeLog: + - 2011/10/02, version 1: This is the very first release of this file. +*/ + +#include "pffft.h" +#include +#include +#include +#include + +/* detect compiler flavour */ +#if defined(_MSC_VER) +# define COMPILER_MSVC +#elif defined(__GNUC__) +# define COMPILER_GCC +#endif + +#if defined(COMPILER_GCC) +# define ALWAYS_INLINE(return_type) inline return_type __attribute__ ((always_inline)) +# define NEVER_INLINE(return_type) return_type __attribute__ ((noinline)) +# define RESTRICT __restrict +# define VLA_ARRAY_ON_STACK(type__, varname__, size__) type__ varname__[size__]; +#elif defined(COMPILER_MSVC) +# define ALWAYS_INLINE(return_type) __forceinline return_type +# define NEVER_INLINE(return_type) __declspec(noinline) return_type +# define RESTRICT __restrict +# define VLA_ARRAY_ON_STACK(type__, varname__, size__) type__ *varname__ = (type__*)_alloca(size__ * sizeof(type__)) +#endif + + +/* + vector support macros: the rest of the code is independant of + SSE/Altivec/NEON -- adding support for other platforms with 4-element + vectors should be limited to these macros +*/ + + +/* define PFFFT_SIMD_DISABLE if you want to use scalar code instead of simd code */ +/*#define PFFFT_SIMD_DISABLE */ + +/* + Altivec support macros +*/ +#if !defined(PFFFT_SIMD_DISABLE) && (defined(__ppc__) || defined(__ppc64__)) +typedef vector float v4sf; +# define SIMD_SZ 4 +# define VZERO() ((vector float) vec_splat_u8(0)) +# define VMUL(a,b) vec_madd(a,b, VZERO()) +# define VADD(a,b) vec_add(a,b) +# define VMADD(a,b,c) vec_madd(a,b,c) +# define VSUB(a,b) vec_sub(a,b) +inline v4sf ld_ps1(const float *p) { v4sf v=vec_lde(0,p); return vec_splat(vec_perm(v, v, vec_lvsl(0, p)), 0); } +# define LD_PS1(p) ld_ps1(&p) +# define INTERLEAVE2(in1, in2, out1, out2) { v4sf tmp__ = vec_mergeh(in1, in2); out2 = vec_mergel(in1, in2); out1 = tmp__; } +# define UNINTERLEAVE2(in1, in2, out1, out2) { \ + vector unsigned char vperm1 = (vector unsigned char)(0,1,2,3,8,9,10,11,16,17,18,19,24,25,26,27); \ + vector unsigned char vperm2 = (vector unsigned char)(4,5,6,7,12,13,14,15,20,21,22,23,28,29,30,31); \ + v4sf tmp__ = vec_perm(in1, in2, vperm1); out2 = vec_perm(in1, in2, vperm2); out1 = tmp__; \ + } +# define VTRANSPOSE4(x0,x1,x2,x3) { \ + v4sf y0 = vec_mergeh(x0, x2); \ + v4sf y1 = vec_mergel(x0, x2); \ + v4sf y2 = vec_mergeh(x1, x3); \ + v4sf y3 = vec_mergel(x1, x3); \ + x0 = vec_mergeh(y0, y2); \ + x1 = vec_mergel(y0, y2); \ + x2 = vec_mergeh(y1, y3); \ + x3 = vec_mergel(y1, y3); \ + } +# define VSWAPHL(a,b) vec_perm(a,b, (vector unsigned char)(16,17,18,19,20,21,22,23,8,9,10,11,12,13,14,15)) +# define VALIGNED(ptr) ((((long)(ptr)) & 0xF) == 0) + +/* + SSE1 support macros +*/ +#elif !defined(PFFFT_SIMD_DISABLE) && (defined(__x86_64__) || defined(_M_X64) || defined(i386) || defined(_M_IX86)) + +# define SIMD_SZ 4 /* 4 floats by simd vector -- this is pretty much hardcoded in the preprocess/finalize functions anyway so you will have to work if you want to enable AVX with its 256-bit vectors. */ + +#if !PFFFT_DOUBLE +#include +typedef __m128 v4sf; +# define VZERO() _mm_setzero_ps() +# define VMUL(a,b) _mm_mul_ps(a,b) +# define VADD(a,b) _mm_add_ps(a,b) +# define VMADD(a,b,c) _mm_add_ps(_mm_mul_ps(a,b), c) +# define VSUB(a,b) _mm_sub_ps(a,b) +# define LD_PS1(p) _mm_set1_ps(p) +# define INTERLEAVE2(in1, in2, out1, out2) { v4sf tmp__ = _mm_unpacklo_ps(in1, in2); out2 = _mm_unpackhi_ps(in1, in2); out1 = tmp__; } +# define UNINTERLEAVE2(in1, in2, out1, out2) { v4sf tmp__ = _mm_shuffle_ps(in1, in2, _MM_SHUFFLE(2,0,2,0)); out2 = _mm_shuffle_ps(in1, in2, _MM_SHUFFLE(3,1,3,1)); out1 = tmp__; } +# define VTRANSPOSE4(x0,x1,x2,x3) _MM_TRANSPOSE4_PS(x0,x1,x2,x3) +# define VSWAPHL(a,b) _mm_shuffle_ps(b, a, _MM_SHUFFLE(3,2,1,0)) +# define VALIGNED(ptr) ((((long)(ptr)) & 0xF) == 0) + +#else +#include "pffft-avx.h" +#endif + +/* + ARM NEON support macros +*/ +#elif !defined(PFFFT_SIMD_DISABLE) && defined(__arm__) +# include +typedef float32x4_t v4sf; +# define SIMD_SZ 4 +# define VZERO() vdupq_n_f32(0) +# define VMUL(a,b) vmulq_f32(a,b) +# define VADD(a,b) vaddq_f32(a,b) +# define VMADD(a,b,c) vmlaq_f32(c,a,b) +# define VSUB(a,b) vsubq_f32(a,b) +# define LD_PS1(p) vld1q_dup_f32(&(p)) +# define INTERLEAVE2(in1, in2, out1, out2) { float32x4x2_t tmp__ = vzipq_f32(in1,in2); out1=tmp__.val[0]; out2=tmp__.val[1]; } +# define UNINTERLEAVE2(in1, in2, out1, out2) { float32x4x2_t tmp__ = vuzpq_f32(in1,in2); out1=tmp__.val[0]; out2=tmp__.val[1]; } +# define VTRANSPOSE4(x0,x1,x2,x3) { \ + float32x4x2_t t0_ = vzipq_f32(x0, x2); \ + float32x4x2_t t1_ = vzipq_f32(x1, x3); \ + float32x4x2_t u0_ = vzipq_f32(t0_.val[0], t1_.val[0]); \ + float32x4x2_t u1_ = vzipq_f32(t0_.val[1], t1_.val[1]); \ + x0 = u0_.val[0]; x1 = u0_.val[1]; x2 = u1_.val[0]; x3 = u1_.val[1]; \ + } +/* marginally faster version */ +/*# define VTRANSPOSE4(x0,x1,x2,x3) { asm("vtrn.32 %q0, %q1;\n vtrn.32 %q2,%q3\n vswp %f0,%e2\n vswp %f1,%e3" : "+w"(x0), "+w"(x1), "+w"(x2), "+w"(x3)::); } */ +# define VSWAPHL(a,b) vcombine_f32(vget_low_f32(b), vget_high_f32(a)) +# define VALIGNED(ptr) ((((long)(ptr)) & 0x3) == 0) +#else +# if !defined(PFFFT_SIMD_DISABLE) +# warning "building with simd disabled !\n"; +# define PFFFT_SIMD_DISABLE /* fallback to scalar code */ +# endif +#endif + +#if PFFFT_DOUBLE +#define float double +#endif + +/* fallback mode for situations where SSE/Altivec are not available, use scalar mode instead */ +#ifdef PFFFT_SIMD_DISABLE +typedef float v4sf; +# define SIMD_SZ 1 +# define VZERO() 0.f +# define VMUL(a,b) ((a)*(b)) +# define VADD(a,b) ((a)+(b)) +# define VMADD(a,b,c) ((a)*(b)+(c)) +# define VSUB(a,b) ((a)-(b)) +# define LD_PS1(p) (p) +# define VALIGNED(ptr) ((((long)(ptr)) & 0x3) == 0) +#endif + +/* shortcuts for complex multiplcations */ +#define VCPLXMUL(ar,ai,br,bi) { v4sf tmp; tmp=VMUL(ar,bi); ar=VMUL(ar,br); ar=VSUB(ar,VMUL(ai,bi)); ai=VMUL(ai,br); ai=VADD(ai,tmp); } +#define VCPLXMULCONJ(ar,ai,br,bi) { v4sf tmp; tmp=VMUL(ar,bi); ar=VMUL(ar,br); ar=VADD(ar,VMUL(ai,bi)); ai=VMUL(ai,br); ai=VSUB(ai,tmp); } +#ifndef SVMUL +/* multiply a scalar with a vector */ +#define SVMUL(f,v) VMUL(LD_PS1(f),v) +#endif + +#if !defined PFFT_MACROS_ONLY + +#if !defined(PFFFT_SIMD_DISABLE) +typedef union v4sf_union { + v4sf v; + float f[4]; +} v4sf_union; + +#if 0 +#include + +#define assertv4(v,f0,f1,f2,f3) assert(v.f[0] == (f0) && v.f[1] == (f1) && v.f[2] == (f2) && v.f[3] == (f3)) + +/* detect bugs with the vector support macros */ +void validate_pffft_simd(void); +void validate_pffft_simd(void) { + float f[16] = { 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 }; + v4sf_union a0, a1, a2, a3, t, u; + memcpy(a0.f, f, 4*sizeof(float)); + memcpy(a1.f, f+4, 4*sizeof(float)); + memcpy(a2.f, f+8, 4*sizeof(float)); + memcpy(a3.f, f+12, 4*sizeof(float)); + + t = a0; u = a1; t.v = VZERO(); + printf("VZERO=[%2g %2g %2g %2g]\n", t.f[0], t.f[1], t.f[2], t.f[3]); assertv4(t, 0, 0, 0, 0); + t.v = VADD(a1.v, a2.v); + printf("VADD(4:7,8:11)=[%2g %2g %2g %2g]\n", t.f[0], t.f[1], t.f[2], t.f[3]); assertv4(t, 12, 14, 16, 18); + t.v = VMUL(a1.v, a2.v); + printf("VMUL(4:7,8:11)=[%2g %2g %2g %2g]\n", t.f[0], t.f[1], t.f[2], t.f[3]); assertv4(t, 32, 45, 60, 77); + t.v = VMADD(a1.v, a2.v,a0.v); + printf("VMADD(4:7,8:11,0:3)=[%2g %2g %2g %2g]\n", t.f[0], t.f[1], t.f[2], t.f[3]); assertv4(t, 32, 46, 62, 80); + INTERLEAVE2(a1.v,a2.v,t.v,u.v); + printf("INTERLEAVE2(4:7,8:11)=[%2g %2g %2g %2g] [%2g %2g %2g %2g]\n", t.f[0], t.f[1], t.f[2], t.f[3], u.f[0], u.f[1], u.f[2], u.f[3]); + assertv4(t, 4, 8, 5, 9); assertv4(u, 6, 10, 7, 11); + UNINTERLEAVE2(a1.v,a2.v,t.v,u.v); + printf("UNINTERLEAVE2(4:7,8:11)=[%2g %2g %2g %2g] [%2g %2g %2g %2g]\n", t.f[0], t.f[1], t.f[2], t.f[3], u.f[0], u.f[1], u.f[2], u.f[3]); + assertv4(t, 4, 6, 8, 10); assertv4(u, 5, 7, 9, 11); + + t.v=LD_PS1(f[15]); + printf("LD_PS1(15)=[%2g %2g %2g %2g]\n", t.f[0], t.f[1], t.f[2], t.f[3]); + assertv4(t, 15, 15, 15, 15); + t.v = VSWAPHL(a1.v, a2.v); + printf("VSWAPHL(4:7,8:11)=[%2g %2g %2g %2g]\n", t.f[0], t.f[1], t.f[2], t.f[3]); + assertv4(t, 8, 9, 6, 7); + VTRANSPOSE4(a0.v, a1.v, a2.v, a3.v); + printf("VTRANSPOSE4(0:3,4:7,8:11,12:15)=[%2g %2g %2g %2g] [%2g %2g %2g %2g] [%2g %2g %2g %2g] [%2g %2g %2g %2g]\n", + a0.f[0], a0.f[1], a0.f[2], a0.f[3], a1.f[0], a1.f[1], a1.f[2], a1.f[3], + a2.f[0], a2.f[1], a2.f[2], a2.f[3], a3.f[0], a3.f[1], a3.f[2], a3.f[3]); + assertv4(a0, 0, 4, 8, 12); assertv4(a1, 1, 5, 9, 13); assertv4(a2, 2, 6, 10, 14); assertv4(a3, 3, 7, 11, 15); +} +#endif +#endif /*!PFFFT_SIMD_DISABLE */ + +#if 0 +/* SSE and co like 16-bytes aligned pointers */ +#define MALLOC_V4SF_ALIGNMENT 64 /* with a 64-byte alignment, we are even aligned on L2 cache lines... */ +void *pffft_aligned_malloc(size_t nb_bytes) { + void *p, *p0 = malloc(nb_bytes + MALLOC_V4SF_ALIGNMENT); + if (!p0) return (void *) 0; + p = (void *) (((size_t) p0 + MALLOC_V4SF_ALIGNMENT) & (~((size_t) (MALLOC_V4SF_ALIGNMENT-1)))); + *((void **) p - 1) = p0; + return p; +} + +void pffft_aligned_free(void *p) { + if (p) free(*((void **) p - 1)); +} + +int pffft_simd_size() { return SIMD_SZ; } +#endif + +/* + passf2 and passb2 has been merged here, fsign = -1 for passf2, +1 for passb2 +*/ +static NEVER_INLINE(void) passf2_ps(int ido, int l1, const v4sf *cc, v4sf *ch, const float *wa1, float fsign) { + int k, i; + int l1ido = l1*ido; + if (ido <= 2) { + for (k=0; k < l1ido; k += ido, ch += ido, cc+= 2*ido) { + ch[0] = VADD(cc[0], cc[ido+0]); + ch[l1ido] = VSUB(cc[0], cc[ido+0]); + ch[1] = VADD(cc[1], cc[ido+1]); + ch[l1ido + 1] = VSUB(cc[1], cc[ido+1]); + } + } else { + for (k=0; k < l1ido; k += ido, ch += ido, cc += 2*ido) { + for (i=0; i 2); + for (k=0; k< l1ido; k += ido, cc+= 3*ido, ch +=ido) { + for (i=0; i 2); + for (k = 0; k < l1; ++k, cc += 5*ido, ch += ido) { + for (i = 0; i < ido-1; i += 2) { + ti5 = VSUB(cc_ref(i , 2), cc_ref(i , 5)); + ti2 = VADD(cc_ref(i , 2), cc_ref(i , 5)); + ti4 = VSUB(cc_ref(i , 3), cc_ref(i , 4)); + ti3 = VADD(cc_ref(i , 3), cc_ref(i , 4)); + tr5 = VSUB(cc_ref(i-1, 2), cc_ref(i-1, 5)); + tr2 = VADD(cc_ref(i-1, 2), cc_ref(i-1, 5)); + tr4 = VSUB(cc_ref(i-1, 3), cc_ref(i-1, 4)); + tr3 = VADD(cc_ref(i-1, 3), cc_ref(i-1, 4)); + ch_ref(i-1, 1) = VADD(cc_ref(i-1, 1), VADD(tr2, tr3)); + ch_ref(i , 1) = VADD(cc_ref(i , 1), VADD(ti2, ti3)); + cr2 = VADD(cc_ref(i-1, 1), VADD(SVMUL(tr11, tr2),SVMUL(tr12, tr3))); + ci2 = VADD(cc_ref(i , 1), VADD(SVMUL(tr11, ti2),SVMUL(tr12, ti3))); + cr3 = VADD(cc_ref(i-1, 1), VADD(SVMUL(tr12, tr2),SVMUL(tr11, tr3))); + ci3 = VADD(cc_ref(i , 1), VADD(SVMUL(tr12, ti2),SVMUL(tr11, ti3))); + cr5 = VADD(SVMUL(ti11, tr5), SVMUL(ti12, tr4)); + ci5 = VADD(SVMUL(ti11, ti5), SVMUL(ti12, ti4)); + cr4 = VSUB(SVMUL(ti12, tr5), SVMUL(ti11, tr4)); + ci4 = VSUB(SVMUL(ti12, ti5), SVMUL(ti11, ti4)); + dr3 = VSUB(cr3, ci4); + dr4 = VADD(cr3, ci4); + di3 = VADD(ci3, cr4); + di4 = VSUB(ci3, cr4); + dr5 = VADD(cr2, ci5); + dr2 = VSUB(cr2, ci5); + di5 = VSUB(ci2, cr5); + di2 = VADD(ci2, cr5); + wr1=wa1[i], wi1=fsign*wa1[i+1], wr2=wa2[i], wi2=fsign*wa2[i+1]; + wr3=wa3[i], wi3=fsign*wa3[i+1], wr4=wa4[i], wi4=fsign*wa4[i+1]; + VCPLXMUL(dr2, di2, LD_PS1(wr1), LD_PS1(wi1)); + ch_ref(i - 1, 2) = dr2; + ch_ref(i, 2) = di2; + VCPLXMUL(dr3, di3, LD_PS1(wr2), LD_PS1(wi2)); + ch_ref(i - 1, 3) = dr3; + ch_ref(i, 3) = di3; + VCPLXMUL(dr4, di4, LD_PS1(wr3), LD_PS1(wi3)); + ch_ref(i - 1, 4) = dr4; + ch_ref(i, 4) = di4; + VCPLXMUL(dr5, di5, LD_PS1(wr4), LD_PS1(wi4)); + ch_ref(i - 1, 5) = dr5; + ch_ref(i, 5) = di5; + } + } +#undef ch_ref +#undef cc_ref +} +#endif + +static NEVER_INLINE(void) radf2_ps(int ido, int l1, const v4sf * RESTRICT cc, v4sf * RESTRICT ch, const float *wa1) { + static const float minus_one = -1.f; + int i, k, l1ido = l1*ido; + for (k=0; k < l1ido; k += ido) { + v4sf a = cc[k], b = cc[k + l1ido]; + ch[2*k] = VADD(a, b); + ch[2*(k+ido)-1] = VSUB(a, b); + } + if (ido < 2) return; + if (ido != 2) { + for (k=0; k < l1ido; k += ido) { + for (i=2; i 5) { + wa[i1-1] = wa[i-1]; + wa[i1] = wa[i]; + } + } + l1 = l2; + } +} /* cffti1 */ + + +static +v4sf *cfftf1_ps(int n, const v4sf *input_readonly, v4sf *work1, v4sf *work2, const float *wa, const int *ifac, int isign) { + v4sf *in = (v4sf*)input_readonly; + v4sf *out = (in == work2 ? work1 : work2); + int nf = ifac[1], k1; + int l1 = 1; + int iw = 0; + assert(in != out && work1 != work2); + for (k1=2; k1<=nf+1; k1++) { + int ip = ifac[k1]; + int l2 = ip*l1; + int ido = n / l2; + int idot = ido + ido; + switch (ip) { +#if 0 + case 5: { + int ix2 = iw + idot; + int ix3 = ix2 + idot; + int ix4 = ix3 + idot; + passf5_ps(idot, l1, in, out, &wa[iw], &wa[ix2], &wa[ix3], &wa[ix4], (float)isign); + } break; +#endif + case 4: { + int ix2 = iw + idot; + int ix3 = ix2 + idot; + passf4_ps(idot, l1, in, out, &wa[iw], &wa[ix2], &wa[ix3], (float)isign); + } break; + case 2: { + passf2_ps(idot, l1, in, out, &wa[iw], (float)isign); + } break; +#if 0 + case 3: { + int ix2 = iw + idot; + passf3_ps(idot, l1, in, out, &wa[iw], &wa[ix2], (float)isign); + } break; +#endif + default: + assert(0); + } + l1 = l2; + iw += (ip - 1)*idot; + if (out == work2) { + out = work1; in = work2; + } else { + out = work2; in = work1; + } + } + + return in; /* this is in fact the output .. */ +} + + +struct PFFFT_Setup { + int N; + int Ncvec; /* nb of complex simd vectors (N/4 if PFFFT_COMPLEX, N/8 if PFFFT_REAL) */ + int ifac[15]; + pffft_transform_t transform; + v4sf *data; /* allocated room for twiddle coefs */ + float *e; /* points into 'data' , N/4*3 elements */ + float *twiddle; /* points into 'data', N/4 elements */ +}; + +static +PFFFT_Setup *pffft_new_setup(int N, pffft_transform_t transform) { + PFFFT_Setup *s = (PFFFT_Setup*)malloc(sizeof(PFFFT_Setup)); + int k, m; + if (!s) return s; + /* unfortunately, the fft size must be a multiple of 16 for complex FFTs + and 32 for real FFTs -- a lot of stuff would need to be rewritten to + handle other cases (or maybe just switch to a scalar fft, I don't know..) */ + if (transform == PFFFT_REAL) { assert((N%(2*SIMD_SZ*SIMD_SZ))==0 && N>0); } + if (transform == PFFFT_COMPLEX) { assert((N%(SIMD_SZ*SIMD_SZ))==0 && N>0); } + /*assert((N % 32) == 0); */ + s->N = N; + s->transform = transform; + /* nb of complex simd vectors */ + s->Ncvec = (transform == PFFFT_REAL ? N/2 : N)/SIMD_SZ; + s->data = (v4sf*)pffft_aligned_malloc(2*(size_t)s->Ncvec * sizeof(v4sf)); + if (!s->data) {free(s); return 0;} + s->e = (float*)s->data; + s->twiddle = (float*)(s->data + (2*s->Ncvec*(SIMD_SZ-1))/SIMD_SZ); + + if (transform == PFFFT_REAL) { + for (k=0; k < s->Ncvec; ++k) { + int i = k/SIMD_SZ; + int j = k%SIMD_SZ; + for (m=0; m < SIMD_SZ-1; ++m) { + float A = (float)(-2*M_PI*(m+1)*k / N); + s->e[(2*(i*3 + m) + 0) * SIMD_SZ + j] = cos(A); + s->e[(2*(i*3 + m) + 1) * SIMD_SZ + j] = sin(A); + } + } + rffti1_ps(N/SIMD_SZ, s->twiddle, s->ifac); + } else { + for (k=0; k < s->Ncvec; ++k) { + int i = k/SIMD_SZ; + int j = k%SIMD_SZ; + for (m=0; m < SIMD_SZ-1; ++m) { + float A = (float)(-2*M_PI*(m+1)*k / N); + s->e[(2*(i*3 + m) + 0)*SIMD_SZ + j] = cos(A); + s->e[(2*(i*3 + m) + 1)*SIMD_SZ + j] = sin(A); + } + } + cffti1_ps(N/SIMD_SZ, s->twiddle, s->ifac); + } + + /* check that N is decomposable with allowed prime factors */ + for (k=0, m=1; k < s->ifac[1]; ++k) { m *= s->ifac[2+k]; } + if (m != N/SIMD_SZ) { + pffft_destroy_setup(s); s = 0; + } + + return s; +} + + +static +void pffft_destroy_setup(PFFFT_Setup *s) { + if (!s) return; + pffft_aligned_free(s->data); + free(s); +} + +#if !defined(PFFFT_SIMD_DISABLE) + +/* [0 0 1 2 3 4 5 6 7 8] -> [0 8 7 6 5 4 3 2 1] */ +static void reversed_copy(int N, const v4sf *in, int in_stride, v4sf *out) { + v4sf g0, g1; + int k; + INTERLEAVE2(in[0], in[1], g0, g1); in += in_stride; + + *--out = VSWAPHL(g0, g1); /* [g0l, g0h], [g1l g1h] -> [g1l, g0h] */ + for (k=1; k < N; ++k) { + v4sf h0, h1; + INTERLEAVE2(in[0], in[1], h0, h1); in += in_stride; + *--out = VSWAPHL(g1, h0); + *--out = VSWAPHL(h0, h1); + g1 = h1; + } + *--out = VSWAPHL(g1, g0); +} + +static void unreversed_copy(int N, const v4sf *in, v4sf *out, int out_stride) { + v4sf g0, g1, h0, h1; + int k; + g0 = g1 = in[0]; ++in; + for (k=1; k < N; ++k) { + h0 = *in++; h1 = *in++; + g1 = VSWAPHL(g1, h0); + h0 = VSWAPHL(h0, h1); + UNINTERLEAVE2(h0, g1, out[0], out[1]); out += out_stride; + g1 = h1; + } + h0 = *in++; h1 = g0; + g1 = VSWAPHL(g1, h0); + h0 = VSWAPHL(h0, h1); + UNINTERLEAVE2(h0, g1, out[0], out[1]); +} + +static +void pffft_zreorder(PFFFT_Setup *setup, const float *in, float *out, pffft_direction_t direction) { + int k, N = setup->N, Ncvec = setup->Ncvec; + const v4sf *vin = (const v4sf*)in; + v4sf *vout = (v4sf*)out; + assert(in != out); + if (setup->transform == PFFFT_REAL) { + int k, dk = N/32; + if (direction == PFFFT_FORWARD) { + for (k=0; k < dk; ++k) { + INTERLEAVE2(vin[k*8 + 0], vin[k*8 + 1], vout[2*(0*dk + k) + 0], vout[2*(0*dk + k) + 1]); + INTERLEAVE2(vin[k*8 + 4], vin[k*8 + 5], vout[2*(2*dk + k) + 0], vout[2*(2*dk + k) + 1]); + } + reversed_copy(dk, vin+2, 8, (v4sf*)(out + N/2)); + reversed_copy(dk, vin+6, 8, (v4sf*)(out + N)); + } else { + for (k=0; k < dk; ++k) { + UNINTERLEAVE2(vin[2*(0*dk + k) + 0], vin[2*(0*dk + k) + 1], vout[k*8 + 0], vout[k*8 + 1]); + UNINTERLEAVE2(vin[2*(2*dk + k) + 0], vin[2*(2*dk + k) + 1], vout[k*8 + 4], vout[k*8 + 5]); + } + unreversed_copy(dk, (v4sf*)(in + N/4), (v4sf*)(out + N - 6*SIMD_SZ), -8); + unreversed_copy(dk, (v4sf*)(in + 3*N/4), (v4sf*)(out + N - 2*SIMD_SZ), -8); + } + } else { + if (direction == PFFFT_FORWARD) { + for (k=0; k < Ncvec; ++k) { + int kk = (k/4) + (k%4)*(Ncvec/4); + INTERLEAVE2(vin[k*2], vin[k*2+1], vout[kk*2], vout[kk*2+1]); + } + } else { + for (k=0; k < Ncvec; ++k) { + int kk = (k/4) + (k%4)*(Ncvec/4); + UNINTERLEAVE2(vin[kk*2], vin[kk*2+1], vout[k*2], vout[k*2+1]); + } + } + } +} + +static +void pffft_cplx_finalize(int Ncvec, const v4sf *in, v4sf *out, const v4sf *e) { + int k, dk = Ncvec/SIMD_SZ; /* number of 4x4 matrix blocks */ + v4sf r0, i0, r1, i1, r2, i2, r3, i3; + v4sf sr0, dr0, sr1, dr1, si0, di0, si1, di1; + assert(in != out); + for (k=0; k < dk; ++k) { + r0 = in[8*k+0]; i0 = in[8*k+1]; + r1 = in[8*k+2]; i1 = in[8*k+3]; + r2 = in[8*k+4]; i2 = in[8*k+5]; + r3 = in[8*k+6]; i3 = in[8*k+7]; + VTRANSPOSE4(r0,r1,r2,r3); + VTRANSPOSE4(i0,i1,i2,i3); + VCPLXMUL(r1,i1,e[k*6+0],e[k*6+1]); + VCPLXMUL(r2,i2,e[k*6+2],e[k*6+3]); + VCPLXMUL(r3,i3,e[k*6+4],e[k*6+5]); + + sr0 = VADD(r0,r2); dr0 = VSUB(r0, r2); + sr1 = VADD(r1,r3); dr1 = VSUB(r1, r3); + si0 = VADD(i0,i2); di0 = VSUB(i0, i2); + si1 = VADD(i1,i3); di1 = VSUB(i1, i3); + + /* + transformation for each column is: + + [1 1 1 1 0 0 0 0] [r0] + [1 0 -1 0 0 -1 0 1] [r1] + [1 -1 1 -1 0 0 0 0] [r2] + [1 0 -1 0 0 1 0 -1] [r3] + [0 0 0 0 1 1 1 1] * [i0] + [0 1 0 -1 1 0 -1 0] [i1] + [0 0 0 0 1 -1 1 -1] [i2] + [0 -1 0 1 1 0 -1 0] [i3] + */ + + r0 = VADD(sr0, sr1); i0 = VADD(si0, si1); + r1 = VADD(dr0, di1); i1 = VSUB(di0, dr1); + r2 = VSUB(sr0, sr1); i2 = VSUB(si0, si1); + r3 = VSUB(dr0, di1); i3 = VADD(di0, dr1); + + *out++ = r0; *out++ = i0; *out++ = r1; *out++ = i1; + *out++ = r2; *out++ = i2; *out++ = r3; *out++ = i3; + } +} + +static +void pffft_cplx_preprocess(int Ncvec, const v4sf *in, v4sf *out, const v4sf *e) { + int k, dk = Ncvec/SIMD_SZ; /* number of 4x4 matrix blocks */ + v4sf r0, i0, r1, i1, r2, i2, r3, i3; + v4sf sr0, dr0, sr1, dr1, si0, di0, si1, di1; + assert(in != out); + for (k=0; k < dk; ++k) { + r0 = in[8*k+0]; i0 = in[8*k+1]; + r1 = in[8*k+2]; i1 = in[8*k+3]; + r2 = in[8*k+4]; i2 = in[8*k+5]; + r3 = in[8*k+6]; i3 = in[8*k+7]; + + sr0 = VADD(r0,r2); dr0 = VSUB(r0, r2); + sr1 = VADD(r1,r3); dr1 = VSUB(r1, r3); + si0 = VADD(i0,i2); di0 = VSUB(i0, i2); + si1 = VADD(i1,i3); di1 = VSUB(i1, i3); + + r0 = VADD(sr0, sr1); i0 = VADD(si0, si1); + r1 = VSUB(dr0, di1); i1 = VADD(di0, dr1); + r2 = VSUB(sr0, sr1); i2 = VSUB(si0, si1); + r3 = VADD(dr0, di1); i3 = VSUB(di0, dr1); + + VCPLXMULCONJ(r1,i1,e[k*6+0],e[k*6+1]); + VCPLXMULCONJ(r2,i2,e[k*6+2],e[k*6+3]); + VCPLXMULCONJ(r3,i3,e[k*6+4],e[k*6+5]); + + VTRANSPOSE4(r0,r1,r2,r3); + VTRANSPOSE4(i0,i1,i2,i3); + + *out++ = r0; *out++ = i0; *out++ = r1; *out++ = i1; + *out++ = r2; *out++ = i2; *out++ = r3; *out++ = i3; + } +} + + +static ALWAYS_INLINE(void) pffft_real_finalize_4x4(const v4sf *in0, const v4sf *in1, const v4sf *in, + const v4sf *e, v4sf *out) { + v4sf r0, i0, r1, i1, r2, i2, r3, i3; + v4sf sr0, dr0, sr1, dr1, si0, di0, si1, di1; + r0 = *in0; i0 = *in1; + r1 = *in++; i1 = *in++; r2 = *in++; i2 = *in++; r3 = *in++; i3 = *in++; + VTRANSPOSE4(r0,r1,r2,r3); + VTRANSPOSE4(i0,i1,i2,i3); + + /* + transformation for each column is: + + [1 1 1 1 0 0 0 0] [r0] + [1 0 -1 0 0 -1 0 1] [r1] + [1 0 -1 0 0 1 0 -1] [r2] + [1 -1 1 -1 0 0 0 0] [r3] + [0 0 0 0 1 1 1 1] * [i0] + [0 -1 0 1 -1 0 1 0] [i1] + [0 -1 0 1 1 0 -1 0] [i2] + [0 0 0 0 -1 1 -1 1] [i3] + */ + + /*cerr << "matrix initial, before e , REAL:\n 1: " << r0 << "\n 1: " << r1 << "\n 1: " << r2 << "\n 1: " << r3 << "\n"; */ + /*cerr << "matrix initial, before e, IMAG :\n 1: " << i0 << "\n 1: " << i1 << "\n 1: " << i2 << "\n 1: " << i3 << "\n"; */ + + VCPLXMUL(r1,i1,e[0],e[1]); + VCPLXMUL(r2,i2,e[2],e[3]); + VCPLXMUL(r3,i3,e[4],e[5]); + + /*cerr << "matrix initial, real part:\n 1: " << r0 << "\n 1: " << r1 << "\n 1: " << r2 << "\n 1: " << r3 << "\n"; */ + /*cerr << "matrix initial, imag part:\n 1: " << i0 << "\n 1: " << i1 << "\n 1: " << i2 << "\n 1: " << i3 << "\n"; */ + + sr0 = VADD(r0,r2); dr0 = VSUB(r0,r2); + sr1 = VADD(r1,r3); dr1 = VSUB(r3,r1); + si0 = VADD(i0,i2); di0 = VSUB(i0,i2); + si1 = VADD(i1,i3); di1 = VSUB(i3,i1); + + r0 = VADD(sr0, sr1); + r3 = VSUB(sr0, sr1); + i0 = VADD(si0, si1); + i3 = VSUB(si1, si0); + r1 = VADD(dr0, di1); + r2 = VSUB(dr0, di1); + i1 = VSUB(dr1, di0); + i2 = VADD(dr1, di0); + + *out++ = r0; + *out++ = i0; + *out++ = r1; + *out++ = i1; + *out++ = r2; + *out++ = i2; + *out++ = r3; + *out++ = i3; + +} + +static NEVER_INLINE(void) pffft_real_finalize(int Ncvec, const v4sf *in, v4sf *out, const v4sf *e) { + int k, dk = Ncvec/SIMD_SZ; /* number of 4x4 matrix blocks */ + /* fftpack order is f0r f1r f1i f2r f2i ... f(n-1)r f(n-1)i f(n)r */ + + v4sf_union cr, ci, *uout = (v4sf_union*)out; + v4sf save = in[7], zero=VZERO(); + float xr0, xi0, xr1, xi1, xr2, xi2, xr3, xi3; + static const float s = (float)(M_SQRT2/2); + + cr.v = in[0]; ci.v = in[Ncvec*2-1]; + assert(in != out); + pffft_real_finalize_4x4(&zero, &zero, in+1, e, out); + + /* + [cr0 cr1 cr2 cr3 ci0 ci1 ci2 ci3] + + [Xr(1)] ] [1 1 1 1 0 0 0 0] + [Xr(N/4) ] [0 0 0 0 1 s 0 -s] + [Xr(N/2) ] [1 0 -1 0 0 0 0 0] + [Xr(3N/4)] [0 0 0 0 1 -s 0 s] + [Xi(1) ] [1 -1 1 -1 0 0 0 0] + [Xi(N/4) ] [0 0 0 0 0 -s -1 -s] + [Xi(N/2) ] [0 -1 0 1 0 0 0 0] + [Xi(3N/4)] [0 0 0 0 0 -s 1 -s] + */ + + xr0=(cr.f[0]+cr.f[2]) + (cr.f[1]+cr.f[3]); uout[0].f[0] = xr0; + xi0=(cr.f[0]+cr.f[2]) - (cr.f[1]+cr.f[3]); uout[1].f[0] = xi0; + xr2=(cr.f[0]-cr.f[2]); uout[4].f[0] = xr2; + xi2=(cr.f[3]-cr.f[1]); uout[5].f[0] = xi2; + xr1= ci.f[0] + s*(ci.f[1]-ci.f[3]); uout[2].f[0] = xr1; + xi1=-ci.f[2] - s*(ci.f[1]+ci.f[3]); uout[3].f[0] = xi1; + xr3= ci.f[0] - s*(ci.f[1]-ci.f[3]); uout[6].f[0] = xr3; + xi3= ci.f[2] - s*(ci.f[1]+ci.f[3]); uout[7].f[0] = xi3; + + for (k=1; k < dk; ++k) { + v4sf save_next = in[8*k+7]; + pffft_real_finalize_4x4(&save, &in[8*k+0], in + 8*k+1, + e + k*6, out + k*8); + save = save_next; + } + +} + +static ALWAYS_INLINE(void) pffft_real_preprocess_4x4(const v4sf *in, + const v4sf *e, v4sf *out, int first) { + v4sf r0=in[0], i0=in[1], r1=in[2], i1=in[3], r2=in[4], i2=in[5], r3=in[6], i3=in[7]; + /* + transformation for each column is: + + [1 1 1 1 0 0 0 0] [r0] + [1 0 0 -1 0 -1 -1 0] [r1] + [1 -1 -1 1 0 0 0 0] [r2] + [1 0 0 -1 0 1 1 0] [r3] + [0 0 0 0 1 -1 1 -1] * [i0] + [0 -1 1 0 1 0 0 1] [i1] + [0 0 0 0 1 1 -1 -1] [i2] + [0 1 -1 0 1 0 0 1] [i3] + */ + + v4sf sr0 = VADD(r0,r3), dr0 = VSUB(r0,r3); + v4sf sr1 = VADD(r1,r2), dr1 = VSUB(r1,r2); + v4sf si0 = VADD(i0,i3), di0 = VSUB(i0,i3); + v4sf si1 = VADD(i1,i2), di1 = VSUB(i1,i2); + + r0 = VADD(sr0, sr1); + r2 = VSUB(sr0, sr1); + r1 = VSUB(dr0, si1); + r3 = VADD(dr0, si1); + i0 = VSUB(di0, di1); + i2 = VADD(di0, di1); + i1 = VSUB(si0, dr1); + i3 = VADD(si0, dr1); + + VCPLXMULCONJ(r1,i1,e[0],e[1]); + VCPLXMULCONJ(r2,i2,e[2],e[3]); + VCPLXMULCONJ(r3,i3,e[4],e[5]); + + VTRANSPOSE4(r0,r1,r2,r3); + VTRANSPOSE4(i0,i1,i2,i3); + + if (!first) { + *out++ = r0; + *out++ = i0; + } + *out++ = r1; + *out++ = i1; + *out++ = r2; + *out++ = i2; + *out++ = r3; + *out++ = i3; +} + +static NEVER_INLINE(void) pffft_real_preprocess(int Ncvec, const v4sf *in, v4sf *out, const v4sf *e) { + int k, dk = Ncvec/SIMD_SZ; /* number of 4x4 matrix blocks */ + /* fftpack order is f0r f1r f1i f2r f2i ... f(n-1)r f(n-1)i f(n)r */ + + v4sf_union Xr, Xi, *uout = (v4sf_union*)out; + float cr0, ci0, cr1, ci1, cr2, ci2, cr3, ci3; + static const float s = (float)M_SQRT2; + assert(in != out); + for (k=0; k < 4; ++k) { + Xr.f[k] = ((float*)in)[8*k]; + Xi.f[k] = ((float*)in)[8*k+4]; + } + + pffft_real_preprocess_4x4(in, e, out+1, 1); /* will write only 6 values */ + + /* + [Xr0 Xr1 Xr2 Xr3 Xi0 Xi1 Xi2 Xi3] + + [cr0] [1 0 2 0 1 0 0 0] + [cr1] [1 0 0 0 -1 0 -2 0] + [cr2] [1 0 -2 0 1 0 0 0] + [cr3] [1 0 0 0 -1 0 2 0] + [ci0] [0 2 0 2 0 0 0 0] + [ci1] [0 s 0 -s 0 -s 0 -s] + [ci2] [0 0 0 0 0 -2 0 2] + [ci3] [0 -s 0 s 0 -s 0 -s] + */ + for (k=1; k < dk; ++k) { + pffft_real_preprocess_4x4(in+8*k, e + k*6, out-1+k*8, 0); + } + + cr0=(Xr.f[0]+Xi.f[0]) + 2*Xr.f[2]; uout[0].f[0] = cr0; + cr1=(Xr.f[0]-Xi.f[0]) - 2*Xi.f[2]; uout[0].f[1] = cr1; + cr2=(Xr.f[0]+Xi.f[0]) - 2*Xr.f[2]; uout[0].f[2] = cr2; + cr3=(Xr.f[0]-Xi.f[0]) + 2*Xi.f[2]; uout[0].f[3] = cr3; + ci0= 2*(Xr.f[1]+Xr.f[3]); uout[2*Ncvec-1].f[0] = ci0; + ci1= s*(Xr.f[1]-Xr.f[3]) - s*(Xi.f[1]+Xi.f[3]); uout[2*Ncvec-1].f[1] = ci1; + ci2= 2*(Xi.f[3]-Xi.f[1]); uout[2*Ncvec-1].f[2] = ci2; + ci3=-s*(Xr.f[1]-Xr.f[3]) - s*(Xi.f[1]+Xi.f[3]); uout[2*Ncvec-1].f[3] = ci3; +} + + +static +void pffft_transform_internal(PFFFT_Setup *setup, const float *finput, float *foutput, v4sf *scratch, + pffft_direction_t direction, int ordered) { + int k, Ncvec = setup->Ncvec; + int nf_odd = (setup->ifac[1] & 1); + +#if 0 + /* temporary buffer is allocated on the stack if the scratch pointer is NULL */ + int stack_allocate = (scratch == 0 ? Ncvec*2 : 1); + VLA_ARRAY_ON_STACK(v4sf, scratch_on_stack, stack_allocate); +#endif + + const v4sf *vinput = (const v4sf*)finput; + v4sf *voutput = (v4sf*)foutput; + v4sf *buff[2]; + int ib = (nf_odd ^ ordered ? 1 : 0); + buff[0] = voutput; buff[1] = scratch; + + assert(VALIGNED(finput) && VALIGNED(foutput)); + + /*assert(finput != foutput); */ + if (direction == PFFFT_FORWARD) { + ib = !ib; + if (setup->transform == PFFFT_REAL) { + ib = (rfftf1_ps(Ncvec*2, vinput, buff[ib], buff[!ib], + setup->twiddle, &setup->ifac[0]) == buff[0] ? 0 : 1); + pffft_real_finalize(Ncvec, buff[ib], buff[!ib], (v4sf*)setup->e); + } else { + v4sf *tmp = buff[ib]; + for (k=0; k < Ncvec; ++k) { + UNINTERLEAVE2(vinput[k*2], vinput[k*2+1], tmp[k*2], tmp[k*2+1]); + } + ib = (cfftf1_ps(Ncvec, buff[ib], buff[!ib], buff[ib], + setup->twiddle, &setup->ifac[0], -1) == buff[0] ? 0 : 1); + pffft_cplx_finalize(Ncvec, buff[ib], buff[!ib], (v4sf*)setup->e); + } + if (ordered) { + pffft_zreorder(setup, (float*)buff[!ib], (float*)buff[ib], PFFFT_FORWARD); + } else ib = !ib; + } else { + if (vinput == buff[ib]) { + ib = !ib; /* may happen when finput == foutput */ + } + if (ordered) { + pffft_zreorder(setup, (float*)vinput, (float*)buff[ib], PFFFT_BACKWARD); + vinput = buff[ib]; ib = !ib; + } + if (setup->transform == PFFFT_REAL) { + pffft_real_preprocess(Ncvec, vinput, buff[ib], (v4sf*)setup->e); + ib = (rfftb1_ps(Ncvec*2, buff[ib], buff[0], buff[1], + setup->twiddle, &setup->ifac[0]) == buff[0] ? 0 : 1); + } else { + pffft_cplx_preprocess(Ncvec, vinput, buff[ib], (v4sf*)setup->e); + ib = (cfftf1_ps(Ncvec, buff[ib], buff[0], buff[1], + setup->twiddle, &setup->ifac[0], +1) == buff[0] ? 0 : 1); + for (k=0; k < Ncvec; ++k) { + INTERLEAVE2(buff[ib][k*2], buff[ib][k*2+1], buff[ib][k*2], buff[ib][k*2+1]); + } + } + } + + if (buff[ib] != voutput) { + /* extra copy required -- this situation should only happen when finput == foutput */ + assert(finput==foutput); + for (k=0; k < Ncvec; ++k) { + v4sf a = buff[ib][2*k], b = buff[ib][2*k+1]; + voutput[2*k] = a; voutput[2*k+1] = b; + } + ib = !ib; + } + assert(buff[ib] == voutput); +} + +#if 0 +void pffft_zconvolve_accumulate(PFFFT_Setup *s, const float *a, const float *b, float *ab, float scaling) { + int Ncvec = s->Ncvec; + const v4sf * RESTRICT va = (const v4sf*)a; + const v4sf * RESTRICT vb = (const v4sf*)b; + v4sf * RESTRICT vab = (v4sf*)ab; + +#ifdef __arm__ + __builtin_prefetch(va); + __builtin_prefetch(vb); + __builtin_prefetch(vab); + __builtin_prefetch(va+2); + __builtin_prefetch(vb+2); + __builtin_prefetch(vab+2); + __builtin_prefetch(va+4); + __builtin_prefetch(vb+4); + __builtin_prefetch(vab+4); + __builtin_prefetch(va+6); + __builtin_prefetch(vb+6); + __builtin_prefetch(vab+6); +# ifndef __clang__ +# define ZCONVOLVE_USING_INLINE_NEON_ASM +# endif +#endif + + float ar, ai, br, bi, abr, abi; +#ifndef ZCONVOLVE_USING_INLINE_ASM + v4sf vscal = LD_PS1(scaling); + int i; +#endif + + assert(VALIGNED(a) && VALIGNED(b) && VALIGNED(ab)); + ar = ((v4sf_union*)va)[0].f[0]; + ai = ((v4sf_union*)va)[1].f[0]; + br = ((v4sf_union*)vb)[0].f[0]; + bi = ((v4sf_union*)vb)[1].f[0]; + abr = ((v4sf_union*)vab)[0].f[0]; + abi = ((v4sf_union*)vab)[1].f[0]; + +#ifdef ZCONVOLVE_USING_INLINE_ASM /* inline asm version, unfortunately miscompiled by clang 3.2, at least on ubuntu.. so this will be restricted to gcc */ + const float *a_ = a, *b_ = b; float *ab_ = ab; + int N = Ncvec; + asm volatile("mov r8, %2 \n" + "vdup.f32 q15, %4 \n" + "1: \n" + "pld [%0,#64] \n" + "pld [%1,#64] \n" + "pld [%2,#64] \n" + "pld [%0,#96] \n" + "pld [%1,#96] \n" + "pld [%2,#96] \n" + "vld1.f32 {q0,q1}, [%0,:128]! \n" + "vld1.f32 {q4,q5}, [%1,:128]! \n" + "vld1.f32 {q2,q3}, [%0,:128]! \n" + "vld1.f32 {q6,q7}, [%1,:128]! \n" + "vld1.f32 {q8,q9}, [r8,:128]! \n" + + "vmul.f32 q10, q0, q4 \n" + "vmul.f32 q11, q0, q5 \n" + "vmul.f32 q12, q2, q6 \n" + "vmul.f32 q13, q2, q7 \n" + "vmls.f32 q10, q1, q5 \n" + "vmla.f32 q11, q1, q4 \n" + "vld1.f32 {q0,q1}, [r8,:128]! \n" + "vmls.f32 q12, q3, q7 \n" + "vmla.f32 q13, q3, q6 \n" + "vmla.f32 q8, q10, q15 \n" + "vmla.f32 q9, q11, q15 \n" + "vmla.f32 q0, q12, q15 \n" + "vmla.f32 q1, q13, q15 \n" + "vst1.f32 {q8,q9},[%2,:128]! \n" + "vst1.f32 {q0,q1},[%2,:128]! \n" + "subs %3, #2 \n" + "bne 1b \n" + : "+r"(a_), "+r"(b_), "+r"(ab_), "+r"(N) : "r"(scaling) : "r8", "q0","q1","q2","q3","q4","q5","q6","q7","q8","q9", "q10","q11","q12","q13","q15","memory"); +#else /* default routine, works fine for non-arm cpus with current compilers */ + for (i=0; i < Ncvec; i += 2) { + v4sf ar, ai, br, bi; + ar = va[2*i+0]; ai = va[2*i+1]; + br = vb[2*i+0]; bi = vb[2*i+1]; + VCPLXMUL(ar, ai, br, bi); + vab[2*i+0] = VMADD(ar, vscal, vab[2*i+0]); + vab[2*i+1] = VMADD(ai, vscal, vab[2*i+1]); + ar = va[2*i+2]; ai = va[2*i+3]; + br = vb[2*i+2]; bi = vb[2*i+3]; + VCPLXMUL(ar, ai, br, bi); + vab[2*i+2] = VMADD(ar, vscal, vab[2*i+2]); + vab[2*i+3] = VMADD(ai, vscal, vab[2*i+3]); + } +#endif + if (s->transform == PFFFT_REAL) { + ((v4sf_union*)vab)[0].f[0] = abr + ar*br*scaling; + ((v4sf_union*)vab)[1].f[0] = abi + ai*bi*scaling; + } +} +#endif + + +#else /* defined(PFFFT_SIMD_DISABLE) */ + +/* standard routine using scalar floats, without SIMD stuff. */ + +#define pffft_zreorder_nosimd pffft_zreorder +static +void pffft_zreorder_nosimd(PFFFT_Setup *setup, const float *in, float *out, pffft_direction_t direction) { + int k, N = setup->N; + if (setup->transform == PFFFT_COMPLEX) { + for (k=0; k < 2*N; ++k) out[k] = in[k]; + return; + } + else if (direction == PFFFT_FORWARD) { + float x_N = in[N-1]; + for (k=N-1; k > 1; --k) out[k] = in[k-1]; + out[0] = in[0]; + out[1] = x_N; + } else { + float x_N = in[1]; + for (k=1; k < N-1; ++k) out[k] = in[k+1]; + out[0] = in[0]; + out[N-1] = x_N; + } +} + +#define pffft_transform_internal_nosimd pffft_transform_internal +static +void pffft_transform_internal_nosimd(PFFFT_Setup *setup, const float *input, float *output, float *scratch, + pffft_direction_t direction, int ordered) { + int Ncvec = setup->Ncvec; + int nf_odd = (setup->ifac[1] & 1); + +#if 0 + /* temporary buffer is allocated on the stack if the scratch pointer is NULL */ + int stack_allocate = (scratch == 0 ? Ncvec*2 : 1); + VLA_ARRAY_ON_STACK(v4sf, scratch_on_stack, stack_allocate); +#endif + float *buff[2]; + int ib; + /* if (scratch == 0) scratch = scratch_on_stack; */ + buff[0] = output; buff[1] = scratch; + + if (setup->transform == PFFFT_COMPLEX) ordered = 0; /* it is always ordered. */ + ib = (nf_odd ^ ordered ? 1 : 0); + + if (direction == PFFFT_FORWARD) { + if (setup->transform == PFFFT_REAL) { + ib = (rfftf1_ps(Ncvec*2, input, buff[ib], buff[!ib], + setup->twiddle, &setup->ifac[0]) == buff[0] ? 0 : 1); + } else { + ib = (cfftf1_ps(Ncvec, input, buff[ib], buff[!ib], + setup->twiddle, &setup->ifac[0], -1) == buff[0] ? 0 : 1); + } + if (ordered) { + pffft_zreorder(setup, buff[ib], buff[!ib], PFFFT_FORWARD); ib = !ib; + } + } else { + if (input == buff[ib]) { + ib = !ib; /* may happen when finput == foutput */ + } + if (ordered) { + pffft_zreorder(setup, input, buff[!ib], PFFFT_BACKWARD); + input = buff[!ib]; + } + if (setup->transform == PFFFT_REAL) { + ib = (rfftb1_ps(Ncvec*2, input, buff[ib], buff[!ib], + setup->twiddle, &setup->ifac[0]) == buff[0] ? 0 : 1); + } else { + ib = (cfftf1_ps(Ncvec, input, buff[ib], buff[!ib], + setup->twiddle, &setup->ifac[0], +1) == buff[0] ? 0 : 1); + } + } + if (buff[ib] != output) { + int k; + /* extra copy required -- this situation should happens only when finput == foutput */ + assert(input==output); + for (k=0; k < Ncvec; ++k) { + float a = buff[ib][2*k], b = buff[ib][2*k+1]; + output[2*k] = a; output[2*k+1] = b; + } + ib = !ib; + } + assert(buff[ib] == output); +} + +#if 0 +#define pffft_zconvolve_accumulate_nosimd pffft_zconvolve_accumulate +void pffft_zconvolve_accumulate_nosimd(PFFFT_Setup *s, const float *a, const float *b, + float *ab, float scaling) { + int i, Ncvec = s->Ncvec; + + if (s->transform == PFFFT_REAL) { + /* take care of the fftpack ordering */ + ab[0] += a[0]*b[0]*scaling; + ab[2*Ncvec-1] += a[2*Ncvec-1]*b[2*Ncvec-1]*scaling; + ++ab; ++a; ++b; --Ncvec; + } + for (i=0; i < Ncvec; ++i) { + float ar, ai, br, bi; + ar = a[2*i+0]; ai = a[2*i+1]; + br = b[2*i+0]; bi = b[2*i+1]; + VCPLXMUL(ar, ai, br, bi); + ab[2*i+0] += ar*scaling; + ab[2*i+1] += ai*scaling; + } +} +#endif + +#endif /* defined(PFFFT_SIMD_DISABLE) */ + +static +void pffft_transform(PFFFT_Setup *setup, const float *input, float *output, float *work, pffft_direction_t direction) { + pffft_transform_internal(setup, input, output, (v4sf*)work, direction, 0); +} + +static +void pffft_transform_ordered(PFFFT_Setup *setup, const float *input, float *output, float *work, pffft_direction_t direction) { + pffft_transform_internal(setup, input, output, (v4sf*)work, direction, 1); +} + +#endif diff --git a/soxr-sys/src/pffft.h b/soxr-sys/src/pffft.h new file mode 100644 index 000000000..63522cacb --- /dev/null +++ b/soxr-sys/src/pffft.h @@ -0,0 +1,197 @@ +/* https://bitbucket.org/jpommier/pffft/raw/483453d8f7661058e74aa4e7cf5c27bcd7887e7a/pffft.h + * with minor changes for libsoxr. */ + +#if !defined PFFT_MACROS_ONLY + +/* Copyright (c) 2013 Julien Pommier ( pommier@modartt.com ) + + Based on original fortran 77 code from FFTPACKv4 from NETLIB, + authored by Dr Paul Swarztrauber of NCAR, in 1985. + + As confirmed by the NCAR fftpack software curators, the following + FFTPACKv5 license applies to FFTPACKv4 sources. My changes are + released under the same terms. + + FFTPACK license: + + http://www.cisl.ucar.edu/css/software/fftpack5/ftpk.html + + Copyright (c) 2004 the University Corporation for Atmospheric + Research ("UCAR"). All rights reserved. Developed by NCAR's + Computational and Information Systems Laboratory, UCAR, + www.cisl.ucar.edu. + + Redistribution and use of the Software in source and binary forms, + with or without modification, is permitted provided that the + following conditions are met: + + - Neither the names of NCAR's Computational and Information Systems + Laboratory, the University Corporation for Atmospheric Research, + nor the names of its sponsors or contributors may be used to + endorse or promote products derived from this Software without + specific prior written permission. + + - Redistributions of source code must retain the above copyright + notices, this list of conditions, and the disclaimer below. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions, and the disclaimer below in the + documentation and/or other materials provided with the + distribution. + + THIS 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 CONTRIBUTORS OR COPYRIGHT + HOLDERS BE LIABLE FOR ANY CLAIM, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL 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 WITH THE + SOFTWARE. +*/ + +/* + PFFFT : a Pretty Fast FFT. + + This is basically an adaptation of the single precision fftpack + (v4) as found on netlib taking advantage of SIMD instruction found + on cpus such as intel x86 (SSE1), powerpc (Altivec), and arm (NEON). + + For architectures where no SIMD instruction is available, the code + falls back to a scalar version. + + Restrictions: + + - 1D transforms only, with 32-bit single precision. + + - supports only transforms for inputs of length N of the form + N=(2^a)*(3^b)*(5^c), a >= 5, b >=0, c >= 0 (32, 48, 64, 96, 128, + 144, 160, etc are all acceptable lengths). Performance is best for + 128<=N<=8192. + + - all (float*) pointers in the functions below are expected to + have an "simd-compatible" alignment, that is 16 bytes on x86 and + powerpc CPUs. + + You can allocate such buffers with the functions + pffft_aligned_malloc / pffft_aligned_free (or with stuff like + posix_memalign..) + +*/ + +#ifndef PFFFT_H +#define PFFFT_H + +#include + +#ifdef __cplusplus +extern "C" { +#endif + +#if PFFFT_DOUBLE +#define float double +#endif + + /* opaque struct holding internal stuff (precomputed twiddle factors) + this struct can be shared by many threads as it contains only + read-only data. + */ + typedef struct PFFFT_Setup PFFFT_Setup; + + /* direction of the transform */ + typedef enum { PFFFT_FORWARD, PFFFT_BACKWARD } pffft_direction_t; + + /* type of transform */ + typedef enum { PFFFT_REAL, PFFFT_COMPLEX } pffft_transform_t; + + /* + prepare for performing transforms of size N -- the returned + PFFFT_Setup structure is read-only so it can safely be shared by + multiple concurrent threads. + */ + static + PFFFT_Setup *pffft_new_setup(int N, pffft_transform_t transform); + static + void pffft_destroy_setup(PFFFT_Setup *); + /* + Perform a Fourier transform , The z-domain data is stored in the + most efficient order for transforming it back, or using it for + convolution. If you need to have its content sorted in the + "usual" way, that is as an array of interleaved complex numbers, + either use pffft_transform_ordered , or call pffft_zreorder after + the forward fft, and before the backward fft. + + Transforms are not scaled: PFFFT_BACKWARD(PFFFT_FORWARD(x)) = N*x. + Typically you will want to scale the backward transform by 1/N. + + The 'work' pointer should point to an area of N (2*N for complex + fft) floats, properly aligned. If 'work' is NULL, then stack will + be used instead (this is probably the best strategy for small + FFTs, say for N < 16384). + + input and output may alias. + */ + static + void pffft_transform(PFFFT_Setup *setup, const float *input, float *output, float *work, pffft_direction_t direction); + + /* + Similar to pffft_transform, but makes sure that the output is + ordered as expected (interleaved complex numbers). This is + similar to calling pffft_transform and then pffft_zreorder. + + input and output may alias. + */ + static + void pffft_transform_ordered(PFFFT_Setup *setup, const float *input, float *output, float *work, pffft_direction_t direction); + + /* + call pffft_zreorder(.., PFFFT_FORWARD) after pffft_transform(..., + PFFFT_FORWARD) if you want to have the frequency components in + the correct "canonical" order, as interleaved complex numbers. + + (for real transforms, both 0-frequency and half frequency + components, which are real, are assembled in the first entry as + F(0)+i*F(n/2+1). Note that the original fftpack did place + F(n/2+1) at the end of the arrays). + + input and output should not alias. + */ + static + void pffft_zreorder(PFFFT_Setup *setup, const float *input, float *output, pffft_direction_t direction); + + /* + Perform a multiplication of the frequency components of dft_a and + dft_b and accumulate them into dft_ab. The arrays should have + been obtained with pffft_transform(.., PFFFT_FORWARD) and should + *not* have been reordered with pffft_zreorder (otherwise just + perform the operation yourself as the dft coefs are stored as + interleaved complex numbers). + + the operation performed is: dft_ab += (dft_a * fdt_b)*scaling + + The dft_a, dft_b and dft_ab pointers may alias. + */ + void pffft_zconvolve_accumulate(PFFFT_Setup *setup, const float *dft_a, const float *dft_b, float *dft_ab, float scaling); + + /* + the float buffers must have the correct alignment (16-byte boundary + on intel and powerpc). This function may be used to obtain such + correctly aligned buffers. + */ +#if 0 + void *pffft_aligned_malloc(size_t nb_bytes); + void pffft_aligned_free(void *); + + /* return 4 or 1 wether support SSE/Altivec instructions was enable when building pffft.c */ + int pffft_simd_size(); +#endif + +#undef float + +#ifdef __cplusplus +} +#endif + +#endif + +#endif diff --git a/soxr-sys/src/pffft32.c b/soxr-sys/src/pffft32.c new file mode 100644 index 000000000..f48080949 --- /dev/null +++ b/soxr-sys/src/pffft32.c @@ -0,0 +1,39 @@ +/* SoX Resampler Library Copyright (c) 2007-13 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#define SIMD_ALIGNED_FREE free +#define SIMD_ALIGNED_MALLOC malloc +#define PFFFT_SIMD_DISABLE +#define PFFFT_DOUBLE 0 +#include "pffft-wrap.c" + +#include "filter.h" +#include "rdft_t.h" + +static void * setup(int len) {return pffft_new_setup(len, PFFFT_REAL);} +static void delete_setup(void * setup) {pffft_destroy_setup(setup);} +static void forward (int length, void * setup, float * h, float * scratch) {pffft_transform (setup, h, h, scratch, PFFFT_FORWARD); (void)length;} +static void oforward (int length, void * setup, float * h, float * scratch) {pffft_transform_ordered(setup, h, h, scratch, PFFFT_FORWARD); (void)length;} +static void backward (int length, void * setup, float * H, float * scratch) {pffft_transform (setup, H, H, scratch, PFFFT_BACKWARD);(void)length;} +static void obackward(int length, void * setup, float * H, float * scratch) {pffft_transform_ordered(setup, H, H, scratch, PFFFT_BACKWARD);(void)length;} +static void convolve(int length, void * setup, float * H, float const * with) { pffft_zconvolve(setup, H, with, H); (void)length;} +static int multiplier(void) {return 1;} +static int flags(void) {return RDFT_NEEDS_SCRATCH;} + +fn_t _soxr_rdft32_cb[] = { + (fn_t)setup, + (fn_t)setup, + (fn_t)delete_setup, + (fn_t)forward, + (fn_t)oforward, + (fn_t)backward, + (fn_t)obackward, + (fn_t)convolve, + (fn_t)_soxr_ordered_partial_convolve_f, + (fn_t)multiplier, + (fn_t)pffft_reorder_back, + (fn_t)malloc, + (fn_t)calloc, + (fn_t)free, + (fn_t)flags, +}; diff --git a/soxr-sys/src/pffft32s.c b/soxr-sys/src/pffft32s.c new file mode 100644 index 000000000..7798a45c0 --- /dev/null +++ b/soxr-sys/src/pffft32s.c @@ -0,0 +1,34 @@ +/* SoX Resampler Library Copyright (c) 2007-13 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#define PFFFT_DOUBLE 0 +#include "pffft-wrap.c" + +#include "rdft_t.h" + +static void * setup(int len) {return pffft_new_setup(len, PFFFT_REAL);} +static void forward (int length, void * setup, float * h, float * scratch) {pffft_transform (setup, h, h, scratch, PFFFT_FORWARD); (void)length;} +static void oforward (int length, void * setup, float * h, float * scratch) {pffft_transform_ordered(setup, h, h, scratch, PFFFT_FORWARD); (void)length;} +static void backward (int length, void * setup, float * H, float * scratch) {pffft_transform (setup, H, H, scratch, PFFFT_BACKWARD);(void)length;} +static void obackward(int length, void * setup, float * H, float * scratch) {pffft_transform_ordered(setup, H, H, scratch, PFFFT_BACKWARD);(void)length;} +static void convolve(int length, void * setup, float * H, float const * with) {pffft_zconvolve(setup, H, with, H); (void)length;} +static int multiplier(void) {return 1;} +static int flags(void) {return RDFT_IS_SIMD | RDFT_NEEDS_SCRATCH;} + +fn_t _soxr_rdft32s_cb[] = { + (fn_t)setup, + (fn_t)setup, + (fn_t)pffft_destroy_setup, + (fn_t)forward, + (fn_t)oforward, + (fn_t)backward, + (fn_t)obackward, + (fn_t)convolve, + (fn_t)ORDERED_PARTIAL_CONVOLVE_SIMD, + (fn_t)multiplier, + (fn_t)pffft_reorder_back, + (fn_t)SIMD_ALIGNED_MALLOC, + (fn_t)SIMD_ALIGNED_CALLOC, + (fn_t)SIMD_ALIGNED_FREE, + (fn_t)flags, +}; diff --git a/soxr-sys/src/pffft64s.c b/soxr-sys/src/pffft64s.c new file mode 100644 index 000000000..7c37c9d4d --- /dev/null +++ b/soxr-sys/src/pffft64s.c @@ -0,0 +1,34 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#define PFFFT_DOUBLE 1 +#include "pffft-wrap.c" + +#include "rdft_t.h" + +static void * setup(int len) {return pffft_new_setup(len, PFFFT_REAL);} +static void forward (int length, void * setup, double * h, double * scratch) {pffft_transform (setup, h, h, scratch, PFFFT_FORWARD); (void)length;} +static void oforward (int length, void * setup, double * h, double * scratch) {pffft_transform_ordered(setup, h, h, scratch, PFFFT_FORWARD); (void)length;} +static void backward (int length, void * setup, double * H, double * scratch) {pffft_transform (setup, H, H, scratch, PFFFT_BACKWARD);(void)length;} +static void obackward(int length, void * setup, double * H, double * scratch) {pffft_transform_ordered(setup, H, H, scratch, PFFFT_BACKWARD);(void)length;} +static void convolve(int length, void * setup, double * H, double const * with) {pffft_zconvolve(setup, H, with, H); (void)length;} +static int multiplier(void) {return 1;} +static int flags(void) {return RDFT_IS_SIMD | RDFT_NEEDS_SCRATCH;} + +fn_t _soxr_rdft64s_cb[] = { + (fn_t)setup, + (fn_t)setup, + (fn_t)pffft_destroy_setup, + (fn_t)forward, + (fn_t)oforward, + (fn_t)backward, + (fn_t)obackward, + (fn_t)convolve, + (fn_t)ORDERED_PARTIAL_CONVOLVE_SIMD, + (fn_t)multiplier, + (fn_t)pffft_reorder_back, + (fn_t)SIMD_ALIGNED_MALLOC, + (fn_t)SIMD_ALIGNED_CALLOC, + (fn_t)SIMD_ALIGNED_FREE, + (fn_t)flags, +}; diff --git a/soxr-sys/src/poly-fir.h b/soxr-sys/src/poly-fir.h new file mode 100644 index 000000000..d138e030f --- /dev/null +++ b/soxr-sys/src/poly-fir.h @@ -0,0 +1,150 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +/* Resample using an interpolated poly-phase FIR with length LEN. */ +/* Input must be followed by FIR_LENGTH-1 samples. */ + +#if COEF_INTERP != 1 && COEF_INTERP != 2 && COEF_INTERP != 3 + #error COEF_INTERP +#endif + +#if SIMD_AVX || SIMD_SSE || SIMD_NEON + #define N (FIR_LENGTH>>2) + + #if COEF_INTERP == 1 + #define _ sum=vMac(vMac(b,X,a),vLdu(in+j*4),sum), ++j; + #elif COEF_INTERP == 2 + #define _ sum=vMac(vMac(vMac(c,X,b),X,a),vLdu(in+j*4),sum), ++j; + #else + #define _ sum=vMac(vMac(vMac(vMac(d,X,c),X,b),X,a),vLdu(in+j*4),sum), ++j; + #endif + + #define a coefs[(COEF_INTERP+1)*(N*phase+j)+(COEF_INTERP-0)] + #define b coefs[(COEF_INTERP+1)*(N*phase+j)+(COEF_INTERP-1)] + #define c coefs[(COEF_INTERP+1)*(N*phase+j)+(COEF_INTERP-2)] + #define d coefs[(COEF_INTERP+1)*(N*phase+j)+(COEF_INTERP-3)] + + #define BEGINNING v4_t X = vLds(x), sum = vZero(); \ + v4_t const * const __restrict coefs = (v4_t *)COEFS + #define END vStorSum(output+i, sum) + #define cc(n) case n: core(n); break + #define CORE(n) switch (n) {cc(2); cc(3); cc(4); cc(5); cc(6); default: core(n);} +#else + #define N FIR_LENGTH + + #if COEF_INTERP == 1 + #define _ sum += (b*x + a)*in[j], ++j; + #elif COEF_INTERP == 2 + #define _ sum += ((c*x + b)*x + a)*in[j], ++j; + #else + #define _ sum += (((d*x + c)*x + b)*x + a)*in[j], ++j; + #endif + + #define a (coef(COEFS, COEF_INTERP, N, phase, 0,j)) + #define b (coef(COEFS, COEF_INTERP, N, phase, 1,j)) + #define c (coef(COEFS, COEF_INTERP, N, phase, 2,j)) + #define d (coef(COEFS, COEF_INTERP, N, phase, 3,j)) + + #define BEGINNING sample_t sum = 0 + #define END output[i] = sum + #define CORE(n) core(n) +#endif + + + +#define floatPrecCore(n) { \ + float_step_t at = p->at.flt; \ + for (i = 0; (int)at < num_in; ++i, at += p->step.flt) { \ + sample_t const * const __restrict in = input + (int)at; \ + float_step_t frac = at - (int)at; \ + int phase = (int)(frac * (1 << PHASE_BITS)); \ + sample_t x = (sample_t)(frac * (1 << PHASE_BITS) - phase); \ + int j = 0; \ + BEGINNING; CONVOLVE(n); END; \ + } \ + fifo_read(&p->fifo, (int)at, NULL); \ + p->at.flt = at - (int)at; } /* Could round to 1 in some cirmcumstances. */ + + + +#define highPrecCore(n) { \ + step_t at; at.fix = p->at.fix; \ + for (i = 0; at.integer < num_in; ++i, \ + at.fix.ls.all += p->step.fix.ls.all, \ + at.whole += p->step.whole + (at.fix.ls.all < p->step.fix.ls.all)) { \ + sample_t const * const __restrict in = input + at.integer; \ + uint32_t frac = at.fraction; \ + int phase = (int)(frac >> (32 - PHASE_BITS)); /* High-order bits */ \ + /* Low-order bits, scaled to [0,1): */ \ + sample_t x = (sample_t)((frac << PHASE_BITS) * (1 / MULT32)); \ + int j = 0; \ + BEGINNING; CONVOLVE(n); END; \ + } \ + fifo_read(&p->fifo, at.integer, NULL); \ + p->at.whole = at.fraction; \ + p->at.fix.ls = at.fix.ls; } + + + +#define stdPrecCore(n) { \ + int64p_t at; at.all = p->at.whole; \ + for (i = 0; at.parts.ms < num_in; ++i, at.all += p->step.whole) { \ + sample_t const * const __restrict in = input + at.parts.ms; \ + uint32_t const frac = at.parts.ls; \ + int phase = (int)(frac >> (32 - PHASE_BITS)); /* high-order bits */ \ + /* Low-order bits, scaled to [0,1): */ \ + sample_t x = (sample_t)((frac << PHASE_BITS) * (1 / MULT32)); \ + int j = 0; \ + BEGINNING; CONVOLVE(n); END; \ + } \ + fifo_read(&p->fifo, at.parts.ms, NULL); \ + p->at.whole = at.parts.ls; } + + + +#if WITH_FLOAT_STD_PREC_CLOCK + #define SPCORE floatPrecCore +#else + #define SPCORE stdPrecCore +#endif + + + +#if WITH_HI_PREC_CLOCK + #define core(n) if (p->use_hi_prec_clock) highPrecCore(n) else SPCORE(n) +#else + #define core(n) SPCORE(n) +#endif + + + +static void FUNCTION(stage_t * p, fifo_t * output_fifo) +{ + sample_t const * input = stage_read_p(p); + int num_in = min(stage_occupancy(p), p->input_size); + int i, max_num_out = 1 + (int)(num_in * p->out_in_ratio); + sample_t * const __restrict output = fifo_reserve(output_fifo, max_num_out); + + CORE(N); + assert(max_num_out - i >= 0); + fifo_trim_by(output_fifo, max_num_out - i); +} + + + +#undef _ +#undef a +#undef b +#undef c +#undef d +#undef CORE +#undef cc +#undef core +#undef COEF_INTERP +#undef N +#undef BEGINNING +#undef END +#undef CONVOLVE +#undef FIR_LENGTH +#undef FUNCTION +#undef PHASE_BITS diff --git a/soxr-sys/src/poly-fir0.h b/soxr-sys/src/poly-fir0.h new file mode 100644 index 000000000..76fca2d6b --- /dev/null +++ b/soxr-sys/src/poly-fir0.h @@ -0,0 +1,56 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +/* Resample using a non-interpolated poly-phase FIR with length LEN. */ +/* Input must be followed by FIR_LENGTH-1 samples. */ + +#if SIMD_AVX || SIMD_SSE || SIMD_NEON + #define N (FIR_LENGTH>>2) + #define BEGINNING v4_t sum = vZero(); \ + v4_t const * const __restrict coefs = (v4_t *)COEFS + N * rem; + #define _ sum = vMac(vLdu(at+j*4), coefs[j], sum), ++j; + #define END vStorSum(output+i, sum) + #define cc(n) case n: core(n); break + #define CORE(n) switch (n) {cc(2); cc(3); cc(4); cc(5); cc(6); default: core(n);} +#else + #define N FIR_LENGTH + #define BEGINNING sample_t sum = 0; \ + sample_t const * const __restrict coefs = (sample_t *)COEFS + N * rem; + #define _ sum += coefs[j]*at[j], ++j; + #define END output[i] = sum + #define CORE(n) core(n) +#endif + +#define core(n) \ + for (i = 0; at < num_in * p->L; ++i, at += step) { \ + int const div = at / p->L, rem = at % p->L; \ + sample_t const * const __restrict at = input + div; \ + int j = 0; BEGINNING; CONVOLVE(n); END;} + +static void FUNCTION(stage_t * p, fifo_t * output_fifo) +{ + int num_in = min(stage_occupancy(p), p->input_size); + if (num_in) { + sample_t const * input = stage_read_p(p); + int at = p->at.integer, step = p->step.integer; + int i, num_out = (num_in * p->L - at + step - 1) / step; + sample_t * __restrict output = fifo_reserve(output_fifo, num_out); + + CORE(N); + assert(i == num_out); + fifo_read(&p->fifo, at / p->L, NULL); + p->at.integer = at % p->L; + } +} + +#undef _ +#undef CORE +#undef cc +#undef core +#undef N +#undef BEGINNING +#undef MIDDLE +#undef END +#undef CONVOLVE +#undef FIR_LENGTH +#undef FUNCTION diff --git a/soxr-sys/src/rdft.h b/soxr-sys/src/rdft.h new file mode 100644 index 000000000..59ba17417 --- /dev/null +++ b/soxr-sys/src/rdft.h @@ -0,0 +1,31 @@ +/* SoX Resampler Library Copyright (c) 2007-13 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +void ORDERED_CONVOLVE(int n, void * not_used, DFT_FLOAT * a, const DFT_FLOAT * b) +{ + int i; + a[0] *= b[0]; + a[1] *= b[1]; + for (i = 2; i < n; i += 2) { + DFT_FLOAT tmp = a[i]; + a[i ] = b[i ] * tmp - b[i+1] * a[i+1]; + a[i+1] = b[i+1] * tmp + b[i ] * a[i+1]; + } + (void)not_used; +} + +void ORDERED_PARTIAL_CONVOLVE(int n, DFT_FLOAT * a, const DFT_FLOAT * b) +{ + int i; + a[0] *= b[0]; + for (i = 2; i < n; i += 2) { + DFT_FLOAT tmp = a[i]; + a[i ] = b[i ] * tmp - b[i+1] * a[i+1]; + a[i+1] = b[i+1] * tmp + b[i ] * a[i+1]; + } + a[1] = b[i] * a[i] - b[i+1] * a[i+1]; +} + +#undef ORDERED_CONVOLVE +#undef ORDERED_PARTIAL_CONVOLVE +#undef DFT_FLOAT diff --git a/soxr-sys/src/rdft_t.h b/soxr-sys/src/rdft_t.h new file mode 100644 index 000000000..293d9c37b --- /dev/null +++ b/soxr-sys/src/rdft_t.h @@ -0,0 +1,24 @@ +/* SoX Resampler Library Copyright (c) 2007-13 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +typedef void (* fn_t)(void); + +#define rdft_forward_setup (*(void * (*)(int))RDFT_CB[0]) +#define rdft_backward_setup (*(void * (*)(int))RDFT_CB[1]) +#define rdft_delete_setup (*(void (*)(void *))RDFT_CB[2]) +#define rdft_forward (*(void (*)(int, void *, void *, void *))RDFT_CB[3]) +#define rdft_oforward (*(void (*)(int, void *, void *, void *))RDFT_CB[4]) +#define rdft_backward (*(void (*)(int, void *, void *, void *))RDFT_CB[5]) +#define rdft_obackward (*(void (*)(int, void *, void *, void *))RDFT_CB[6]) +#define rdft_convolve (*(void (*)(int, void *, void *, void const *))RDFT_CB[7]) +#define rdft_convolve_portion (*(void (*)(int, void *, void const *))RDFT_CB[8]) +#define rdft_multiplier (*(int (*)(void))RDFT_CB[9]) +#define rdft_reorder_back (*(void (*)(int, void *, void *, void *))RDFT_CB[10]) +#define rdft_malloc (*(void * (*)(size_t))RDFT_CB[11]) +#define rdft_calloc (*(void * (*)(size_t, size_t))RDFT_CB[12]) +#define rdft_free (*(void (*)(void *))RDFT_CB[13]) +#define rdft_flags (*(int (*)(void))RDFT_CB[14]) + +/* Flag templates: */ +#define RDFT_IS_SIMD 1 +#define RDFT_NEEDS_SCRATCH 2 diff --git a/soxr-sys/src/rint-clip.h b/soxr-sys/src/rint-clip.h new file mode 100644 index 000000000..bfb645847 --- /dev/null +++ b/soxr-sys/src/rint-clip.h @@ -0,0 +1,158 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#if defined DITHER + +#define DITHERING + (1./32)*(int)(((ran1>>=3)&31)-((ran2>>=3)&31)) +#define DITHER_RAND (seed = 1664525UL * seed + 1013904223UL) >> 3 +#define DITHER_VARS unsigned long ran1 = DITHER_RAND, ran2 = DITHER_RAND +#define SEED_ARG , unsigned long * seed0 +#define SAVE_SEED *seed0 = seed +#define COPY_SEED unsigned long seed = *seed0; +#define COPY_SEED1 unsigned long seed1 = seed +#define PASS_SEED1 , &seed1 +#define PASS_SEED , &seed +#define FLOATD double + +#else + +#define DITHERING +#define DITHER_VARS +#define SEED_ARG +#define SAVE_SEED +#define COPY_SEED +#define COPY_SEED1 +#define PASS_SEED1 +#define PASS_SEED +#define FLOATD FLOATX + +#endif + +#define DO_16 _;_;_;_;_;_;_;_;_;_;_;_;_;_;_;_ + + + +#if defined FE_INVALID && defined FPU_RINT +static void RINT_CLIP(RINT_T * const dest, FLOATX const * const src, + unsigned stride, size_t i, size_t const n, size_t * const clips SEED_ARG) +{ + COPY_SEED + DITHER_VARS; + for (; i < n; ++i) { + FLOATD const d = src[i] DITHERING; + RINT(dest[stride * i], d); + if (fe_test_invalid()) { + fe_clear_invalid(); + dest[stride * i] = d > 0? RINT_MAX : -RINT_MAX - 1; + ++*clips; + } + } + SAVE_SEED; +} +#endif + + + +static size_t LSX_RINT_CLIP(void * * const dest0, FLOATX const * const src, + size_t const n SEED_ARG) +{ + size_t i, clips = 0; + RINT_T * dest = *dest0; + COPY_SEED +#if defined FE_INVALID && defined FPU_RINT +#define _ RINT(dest[i], src[i] DITHERING); ++i + for (i = 0; i < (n & ~15u);) { + COPY_SEED1; + DITHER_VARS; + DO_16; + if (fe_test_invalid()) { + fe_clear_invalid(); + RINT_CLIP(dest, src, 1, i - 16, i, &clips PASS_SEED1); + } + } + RINT_CLIP(dest, src, 1, i, n, &clips PASS_SEED); +#else +#define _ d = src[i] DITHERING, dest[i++] = (RINT_T)(d > 0? \ + d+.5 >= N? ++clips, N-1 : d+.5 : d-.5 <= -N-1? ++clips, -N:d-.5) + const double N = 1. + RINT_MAX; + double d; + for (i = 0; i < (n & ~15u);) { + DITHER_VARS; + DO_16; + } + { + DITHER_VARS; + for (; i < n; _); + } +#endif + SAVE_SEED; + *dest0 = dest + n; + return clips; +} +#undef _ + + + +static size_t LSX_RINT_CLIP_2(void * * dest0, FLOATX const * const * srcs, + unsigned const stride, size_t const n SEED_ARG) +{ + unsigned j; + size_t i, clips = 0; + RINT_T * dest = *dest0; + COPY_SEED +#if defined FE_INVALID && defined FPU_RINT +#define _ RINT(dest[stride * i], src[i] DITHERING); ++i + for (j = 0; j < stride; ++j, ++dest) { + FLOATX const * const src = srcs[j]; + for (i = 0; i < (n & ~15u);) { + COPY_SEED1; + DITHER_VARS; + DO_16; + if (fe_test_invalid()) { + fe_clear_invalid(); + RINT_CLIP(dest, src, stride, i - 16, i, &clips PASS_SEED1); + } + } + RINT_CLIP(dest, src, stride, i, n, &clips PASS_SEED); + } +#else +#define _ d = src[i] DITHERING, dest[stride * i++] = (RINT_T)(d > 0? \ + d+.5 >= N? ++clips, N-1 : d+.5 : d-.5 <= -N-1? ++clips, -N:d-.5) + const double N = 1. + RINT_MAX; + double d; + for (j = 0; j < stride; ++j, ++dest) { + FLOATX const * const src = srcs[j]; + for (i = 0; i < (n & ~15u);) { + DITHER_VARS; + DO_16; + } + { + DITHER_VARS; + for (; i < n; _); + } + } +#endif + SAVE_SEED; + *dest0 = dest + stride * (n - 1); + return clips; +} +#undef _ + +#undef FLOATD +#undef PASS_SEED +#undef PASS_SEED1 +#undef COPY_SEED1 +#undef COPY_SEED +#undef SAVE_SEED +#undef SEED_ARG +#undef DITHER_VARS +#undef DITHERING +#undef DITHER + +#undef RINT_MAX +#undef RINT_T +#undef FPU_RINT +#undef RINT +#undef RINT_CLIP +#undef LSX_RINT_CLIP +#undef LSX_RINT_CLIP_2 diff --git a/soxr-sys/src/rint.h b/soxr-sys/src/rint.h new file mode 100644 index 000000000..2f1dfbed6 --- /dev/null +++ b/soxr-sys/src/rint.h @@ -0,0 +1,102 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#if !defined soxr_rint_included +#define soxr_rint_included + +#include "std-types.h" + +/* For x86, compiler-supplied versions of these functions (where available) + * can have poor performance (e.g. mingw32), so prefer these asm versions: */ + +#if defined __GNUC__ && (defined __i386__ || defined __x86_64__) + #define FPU_RINT32 + #define FPU_RINT16 + #define rint32D(a,b) __asm__ __volatile__("fistpl %0": "=m"(a): "t"(b): "st") + #define rint16D(a,b) __asm__ __volatile__("fistps %0": "=m"(a): "t"(b): "st") + #define rint32F rint32D + #define rint16F rint16D + #define FE_INVALID 1 + static __inline int fe_test_invalid(void) { + int status_word; + __asm__ __volatile__("fnstsw %%ax": "=a"(status_word)); + return status_word & FE_INVALID; + } + static __inline int fe_clear_invalid(void) { + int32_t status[7]; + __asm__ __volatile__("fnstenv %0": "=m"(status)); + status[1] &= ~FE_INVALID; + __asm__ __volatile__("fldenv %0": : "m"(*status)); + return 0; + } +#elif defined _MSC_VER && defined _M_IX86 + #define FPU_RINT32 + #define FPU_RINT16 + #define rint_fn(N,Y,X) \ + static __inline void N(Y *y, X x) {Y t; {__asm fld x __asm fistp t} *y=t;} + rint_fn(rint32d, int32_t, double) + rint_fn(rint32f, int32_t, float ) + rint_fn(rint16d, int16_t, double) + rint_fn(rint16f, int16_t, float ) + #define rint32D(y,x) rint32d(&(y),x) + #define rint32F(y,x) rint32f(&(y),x) + #define rint16D(y,x) rint16d(&(y),x) + #define rint16F(y,x) rint16f(&(y),x) + #define FE_INVALID 1 + static __inline int fe_test_invalid(void) { + short status_word; + __asm fnstsw status_word + return status_word & FE_INVALID; + } + static __inline int fe_clear_invalid(void) { + int32_t status[7]; + __asm fnstenv status + status[1] &= ~FE_INVALID; + __asm fldenv status + return 0; + } +#elif defined _MSC_VER && defined _M_X64 + #include + #include + #define FPU_RINT32 + #define FPU_RINT16 + static __inline void rint32d(int32_t *y, double x) { + *y = _mm_cvtsd_si32(_mm_load_sd(&x));} + static __inline void rint32f(int32_t *y, float x) { + *y = _mm_cvtss_si32(_mm_load_ss(&x));} + static __inline void rint16d(int16_t *y, double x) { + x = x*65536+32738; *y = (int16_t)(_mm_cvtsd_si32(_mm_load_sd(&x)) >> 16);} + #define rint32D(y,x) rint32d(&(y),x) + #define rint32F(y,x) rint32f(&(y),x) + #define rint16D(y,x) rint16d(&(y),x) + #define rint16F(y,x) rint16d(&(y),(double)(x)) + #define FE_INVALID 1 + #define fe_test_invalid() (_statusfp() & _SW_INVALID) + #define fe_clear_invalid _clearfp /* Note: clears all. */ +#elif HAVE_LRINT && LONG_MAX == 2147483647L && HAVE_FENV_H + #include + #include + #define FPU_RINT32 + #define rint32D(y,x) ((y)=lrint(x)) + #define rint32F(y,x) ((y)=lrintf(x)) + #define fe_test_invalid() fetestexcept(FE_INVALID) + #define fe_clear_invalid() feclearexcept(FE_INVALID) +#endif + +#if !defined FPU_RINT32 + #define rint32D(y,x) ((y)=(int32_t)((x) < 0? x - .5 : x + .5)) + #define rint32F(y,x) rint32D(y,(double)(x)) +#endif + +#if !defined FPU_RINT16 + #define rint16D(y,x) ((y)=(int16_t)((x) < 0? x - .5 : x + .5)) + #define rint16F(y,x) rint16D(y,(double)(x)) +#endif + +static __inline int32_t rint32(double input) { + int32_t result; rint32D(result, input); return result;} + +static __inline int16_t rint16(double input) { + int16_t result; rint16D(result, input); return result;} + +#endif diff --git a/soxr-sys/src/samplerate.h b/soxr-sys/src/samplerate.h new file mode 100644 index 000000000..911cc5d0c --- /dev/null +++ b/soxr-sys/src/samplerate.h @@ -0,0 +1 @@ +#include "soxr-lsr.h" diff --git a/soxr-sys/src/soxr-config.h b/soxr-sys/src/soxr-config.h new file mode 100644 index 000000000..a559b5f10 --- /dev/null +++ b/soxr-sys/src/soxr-config.h @@ -0,0 +1,28 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + + +#if !defined soxr_config_included +#define soxr_config_included + +#define AVCODEC_FOUND 0 +#define AVUTIL_FOUND 0 +#define WITH_PFFFT 0 + +#define HAVE_FENV_H 1 +#define HAVE_STDBOOL_H 1 +#define HAVE_STDINT_H 1 +#define HAVE_LRINT 0 +#define HAVE_BIGENDIAN 0 + +#define WITH_CR32 1 +#define WITH_CR32S 0 +#define WITH_CR64 0 +#define WITH_CR64S 0 +#define WITH_VR32 1 + +#define WITH_HI_PREC_CLOCK 0 +#define WITH_FLOAT_STD_PREC_CLOCK 0 +#define WITH_DEV_TRACE 0 + +#endif diff --git a/soxr-sys/src/soxr-lsr.c b/soxr-sys/src/soxr-lsr.c new file mode 100644 index 000000000..58ab50a21 --- /dev/null +++ b/soxr-sys/src/soxr-lsr.c @@ -0,0 +1,198 @@ +/* SoX Resampler Library Copyright (c) 2007-18 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +/* Wrapper mostly compatible with `libsamplerate'. */ + +#include +#include +#include "soxr.h" +#include "soxr-lsr.h" +#include "rint.h" + + + +SRC_STATE *src_new(SRC_SRCTYPE id, int channels, SRC_ERROR * error) +{ + return src_callback_new(0, id, channels, error, 0); +} + + + +SRC_ERROR src_process(SRC_STATE *p, SRC_DATA * io) +{ + size_t idone , odone; + + if (!p || !io) return -1; + + soxr_set_error( + p, soxr_set_io_ratio(p, 1/io->src_ratio, (size_t)io->output_frames)); + + soxr_process(p, io->data_in, /* hack: */ + (size_t)(io->end_of_input? ~io->input_frames : io->input_frames), + &idone, io->data_out, (size_t)io->output_frames, &odone); + + io->input_frames_used = (long)idone, io->output_frames_gen = (long)odone; + return -!!soxr_error(p); +} + + + +SRC_ERROR src_set_ratio(SRC_STATE * p, double oi_ratio) +{ + return -!!soxr_set_io_ratio(p, 1/oi_ratio, 0); +} + + + +SRC_ERROR src_reset(SRC_STATE * p) +{ + return -!!soxr_clear(p); +} + + + +SRC_ERROR src_error(SRC_STATE * p) +{ + return -!!soxr_error(p); +} + + + +SRC_STATE * src_delete(SRC_STATE * p) +{ + soxr_delete(p); + return 0; +} + + + +SRC_STATE *src_callback_new(src_callback_t fn, + SRC_SRCTYPE id, int channels, SRC_ERROR * error0, void * p) +{ + soxr_quality_spec_t q_spec = soxr_quality_spec(SOXR_LSR0Q + (unsigned)id, 0); + char const * e = getenv("SOXR_LSR_NUM_THREADS"); + soxr_runtime_spec_t r_spec = soxr_runtime_spec(!(e && atoi(e) != 1)); + soxr_error_t error; + soxr_t soxr = 0; + + assert (channels > 0); + soxr = soxr_create(0, 0, (unsigned)channels, &error, 0, &q_spec, &r_spec); + + if (soxr) + error = soxr_set_input_fn(soxr, (soxr_input_fn_t)fn, p, 0); + + if (error0) + *error0 = -!!error; + + return soxr; +} + + + +long src_callback_read(SRC_STATE *p, double oi_ratio, long olen, float * obuf) +{ + if (!p || olen < 0) return -1; + + soxr_set_error(p, soxr_set_io_ratio(p, 1/oi_ratio, (size_t)olen)); + return (long)soxr_output(p, obuf, (size_t)olen); +} + + + +SRC_ERROR src_simple(SRC_DATA * io, SRC_SRCTYPE id, int channels) +{ + size_t idone, odone; + soxr_error_t error; + soxr_quality_spec_t q_spec = soxr_quality_spec(SOXR_LSR0Q + (unsigned)id, 0); + char const * e = getenv("SOXR_LSR_NUM_THREADS"); + soxr_runtime_spec_t r_spec = soxr_runtime_spec(!(e && atoi(e) != 1)); + + if (!io || channels<=0 || io->input_frames<0 || io->output_frames<0) return-1; + + error = soxr_oneshot(1, io->src_ratio, (unsigned)channels, io->data_in, + (size_t)io->input_frames, &idone, io->data_out, (size_t)io->output_frames, + &odone, 0, &q_spec, &r_spec); + + io->input_frames_used = (long)idone, io->output_frames_gen = (long)odone; + + return -!!error; +} + + + +char const * src_get_name(SRC_SRCTYPE id) +{ + static char const * const names[] = { + "LSR best sinc", "LSR medium sinc", "LSR fastest sinc", + "LSR ZOH", "LSR linear", "SoX VHQ"}; + + return (unsigned)id < 5u + !getenv("SOXR_LSR_STRICT")? names[id] : 0; +} + + + +char const * src_get_description(SRC_SRCTYPE id) +{ + return src_get_name(id); +} + + + +char const * src_get_version(void) +{ + return soxr_version(); +} + + + +char const * src_strerror(SRC_ERROR error) +{ + return error == 1? "Placeholder." : error ? "soxr error" : soxr_strerror(0); +} + + + +int src_is_valid_ratio(double oi_ratio) +{ + return getenv("SOXR_LSR_STRICT")? + oi_ratio >= 1./256 && oi_ratio <= 256 : oi_ratio > 0; +} + + + +void src_short_to_float_array(short const * src, float * dest, int len) +{ + assert (src && dest); + + while (len--) dest[len] = (float)(src[len] * (1 / (1. + SHRT_MAX))); +} + + + +void src_float_to_short_array(float const * src, short * dest, int len) +{ + double d, N = 1. + SHRT_MAX; + assert (src && dest); + + while (len--) d = src[len] * N, dest[len] = + (short)(d > N - 1? (short)(N - 1) : d < -N? (short)-N : rint16(d)); +} + + + +void src_int_to_float_array(int const * src, float * dest, int len) +{ + assert (src && dest); + while (len--) dest[len] = (float)(src[len] * (1 / (32768. * 65536.))); +} + + + +void src_float_to_int_array(float const * src, int * dest, int len) +{ + double d, N = 32768. * 65536.; /* N.B. int32, not int! (Also above fn.) */ + assert (src && dest); + + while (len--) d = src[len] * N, dest[len] = + d >= N - 1? (int)(N - 1) : d < -N? (int)(-N) : rint32(d); +} diff --git a/soxr-sys/src/soxr-lsr.h b/soxr-sys/src/soxr-lsr.h new file mode 100644 index 000000000..b1cc24706 --- /dev/null +++ b/soxr-sys/src/soxr-lsr.h @@ -0,0 +1,78 @@ +/* SoX Resampler Library Copyright (c) 2007-18 robs@users.sourceforge.net + * + * This library is free software; you can redistribute it and/or modify it + * under the terms of the GNU Lesser General Public License as published by + * the Free Software Foundation; either version 2.1 of the License, or (at + * your option) any later version. + * + * This library 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 Lesser + * General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public License + * along with this library; if not, write to the Free Software Foundation, + * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/* Wrapper compatible with `libsamplerate' (constant-rate). + * (Libsoxr's native API can be found in soxr.h). */ + +#if !defined SAMPLERATE_H +#define SAMPLERATE_H +#if defined __cplusplus + extern "C" { +#endif + +#if defined SOXR_DLL + #if defined soxr_lsr_EXPORTS + #define SOXR __declspec(dllexport) + #else + #define SOXR __declspec(dllimport) + #endif +#elif defined SOXR_VISIBILITY && defined __GNUC__ && (__GNUC__ > 4 || __GNUC__ == 4 && __GNUC_MINOR__ >= 1) + #define SOXR __attribute__ ((visibility("default"))) +#else + #define SOXR +#endif + +typedef float SRC_SAMPLE; +enum SRC_SRCTYPE_e {SRC_SINC_BEST_QUALITY, SRC_SINC_MEDIUM_QUALITY, + SRC_SINC_FASTEST, SRC_ZERO_ORDER_HOLD, SRC_LINEAR}; +typedef int SRC_SRCTYPE; +typedef int SRC_ERROR; +typedef long (* src_callback_t)(void *, SRC_SAMPLE * *); +typedef struct soxr SRC_STATE; +typedef struct SRC_DATA { + SRC_SAMPLE * data_in, * data_out; + long input_frames, output_frames; + long input_frames_used, output_frames_gen; + int end_of_input; + double src_ratio; +} SRC_DATA; +SOXR SRC_STATE * src_new(SRC_SRCTYPE, int num_channels, SRC_ERROR *); +SOXR SRC_ERROR src_process (SRC_STATE *, SRC_DATA *); +SOXR SRC_ERROR src_set_ratio(SRC_STATE *, double); +SOXR SRC_ERROR src_reset (SRC_STATE *); +SOXR SRC_ERROR src_error (SRC_STATE *); +SOXR SRC_STATE * src_delete (SRC_STATE *); +SOXR SRC_STATE * src_callback_new( + src_callback_t, SRC_SRCTYPE, int, SRC_ERROR *, void *); +SOXR long src_callback_read( + SRC_STATE *, double src_ratio, long, SRC_SAMPLE *); +SOXR SRC_ERROR src_simple(SRC_DATA *, SRC_SRCTYPE, int); +SOXR char const * src_get_name(SRC_SRCTYPE); +SOXR char const * src_get_description(SRC_SRCTYPE); +SOXR char const * src_get_version(void); +SOXR char const * src_strerror(SRC_ERROR); +SOXR int src_is_valid_ratio(double); +SOXR void src_short_to_float_array(short const *, float *, int); +SOXR void src_float_to_short_array(float const *, short *, int); +SOXR void src_int_to_float_array(int const *, float *, int); +SOXR void src_float_to_int_array(float const *, int *, int); + +#undef SOXR +#if defined __cplusplus + } +#endif +#endif diff --git a/soxr-sys/src/soxr.c b/soxr-sys/src/soxr.c new file mode 100644 index 000000000..c2861ac7c --- /dev/null +++ b/soxr-sys/src/soxr.c @@ -0,0 +1,842 @@ +/* SoX Resampler Library Copyright (c) 2007-18 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#include +#include +#include +#include + +#include "soxr.h" +#include "data-io.h" +#include "internal.h" + +#if AVUTIL_FOUND + #include +#endif + + + +#if WITH_DEV_TRACE + +#include +#include + +int _soxr_trace_level; + +void _soxr_trace(char const * fmt, ...) +{ + va_list args; + va_start(args, fmt); + vfprintf(stderr, fmt, args); + fputc('\n', stderr); + va_end(args); +} + +#endif + + + +char const * soxr_version(void) +{ + return "libsoxr-" SOXR_THIS_VERSION_STR; +} + + + + +typedef void sample_t; /* float or double */ +typedef void (* fn_t)(void); +typedef fn_t control_block_t[10]; + +#define resampler_input (*(sample_t * (*)(void *, sample_t * samples, size_t n))p->control_block[0]) +#define resampler_process (*(void (*)(void *, size_t))p->control_block[1]) +#define resampler_output (*(sample_t const * (*)(void *, sample_t * samples, size_t * n))p->control_block[2]) +#define resampler_flush (*(void (*)(void *))p->control_block[3]) +#define resampler_close (*(void (*)(void *))p->control_block[4]) +#define resampler_delay (*(double (*)(void *))p->control_block[5]) +#define resampler_sizes (*(void (*)(size_t * shared, size_t * channel))p->control_block[6]) +#define resampler_create (*(char const * (*)(void * channel, void * shared, double io_ratio, soxr_quality_spec_t * q_spec, soxr_runtime_spec_t * r_spec, double scale))p->control_block[7]) +#define resampler_set_io_ratio (*(void (*)(void *, double io_ratio, size_t len))p->control_block[8]) +#define resampler_id (*(char const * (*)(void))p->control_block[9]) + +typedef void * resampler_t; /* For one channel. */ +typedef void * resampler_shared_t; /* Between channels. */ +typedef void (* deinterleave_t)(sample_t * * dest, + soxr_datatype_t data_type, void const * * src0, size_t n, unsigned ch); +typedef size_t (* interleave_t)(soxr_datatype_t data_type, void * * dest, + sample_t const * const * src, size_t, unsigned, unsigned long *); + +struct soxr { + unsigned num_channels; + double io_ratio; + soxr_error_t error; + soxr_quality_spec_t q_spec; + soxr_io_spec_t io_spec; + soxr_runtime_spec_t runtime_spec; + + void * input_fn_state; + soxr_input_fn_t input_fn; + size_t max_ilen; + + resampler_shared_t shared; + resampler_t * resamplers; + control_block_t control_block; + deinterleave_t deinterleave; + interleave_t interleave; + + void * * channel_ptrs; + size_t clips; + unsigned long seed; + int flushing; +}; + + + +#if WITH_CR32 || WITH_CR32S || WITH_CR64 || WITH_CR64S + #include "filter.h" +#else + #define lsx_to_3dB(x) ((x)/(x)) +#endif + + + +soxr_quality_spec_t soxr_quality_spec(unsigned long recipe, unsigned long flags) +{ + soxr_quality_spec_t spec, * p = &spec; + unsigned q = recipe & 0xf; /* TODO: move to soxr-lsr.c: */ + unsigned quality = q > SOXR_LSR2Q+2? SOXR_VHQ : q > SOXR_LSR2Q? SOXR_QQ : q; + double rej; + memset(p, 0, sizeof(*p)); + if (quality > SOXR_PRECISIONQ) { + p->e = "invalid quality type"; + return spec; + } + flags |= quality < SOXR_LSR0Q ? RESET_ON_CLEAR : 0; + p->phase_response = "\62\31\144"[(recipe & 0x30)>>4]; + p->stopband_begin = 1; + p->precision = + quality == SOXR_QQ ? 0 : + quality <= SOXR_16_BITQ ? 16 : + quality <= SOXR_32_BITQ ? 4 + quality * 4 : + quality <= SOXR_LSR2Q ? 55 - quality * 4 : /* TODO: move to soxr-lsr.c */ + 0; + rej = p->precision * linear_to_dB(2.); + p->flags = flags; + if (quality <= SOXR_32_BITQ || quality == SOXR_PRECISIONQ) { + #define LOW_Q_BW0 (1385 / 2048.) /* 0.67625 rounded to be a FP exact. */ + p->passband_end = quality == 1? LOW_Q_BW0 : 1 - .05 / lsx_to_3dB(rej); + if (quality <= 2) + p->flags &= ~SOXR_ROLLOFF_NONE, p->flags |= SOXR_ROLLOFF_MEDIUM; + } + else { /* TODO: move to soxr-lsr.c */ + static float const bw[] = {.931f, .832f, .663f}; + p->passband_end = bw[quality - SOXR_LSR0Q]; + if (quality == SOXR_LSR2Q) { + p->flags &= ~SOXR_ROLLOFF_NONE; + p->flags |= SOXR_ROLLOFF_LSR2Q | SOXR_PROMOTE_TO_LQ; + } + } + if (recipe & SOXR_STEEP_FILTER) + p->passband_end = 1 - .01 / lsx_to_3dB(rej); + return spec; +} + + + +char const * soxr_engine(soxr_t p) +{ + return resampler_id(); +} + + + +size_t * soxr_num_clips(soxr_t p) +{ + return &p->clips; +} + + + +soxr_error_t soxr_error(soxr_t p) +{ + return p->error; +} + + + +soxr_runtime_spec_t soxr_runtime_spec(unsigned num_threads) +{ + soxr_runtime_spec_t spec, * p = &spec; + memset(p, 0, sizeof(*p)); + p->log2_min_dft_size = 10; + p->log2_large_dft_size = 17; + p->coef_size_kbytes = 400; + p->num_threads = num_threads; + return spec; +} + + + +soxr_io_spec_t soxr_io_spec( + soxr_datatype_t itype, + soxr_datatype_t otype) +{ + soxr_io_spec_t spec, * p = &spec; + memset(p, 0, sizeof(*p)); + if ((itype | otype) >= SOXR_SPLIT * 2) + p->e = "invalid io datatype(s)"; + else { + p->itype = itype; + p->otype = otype; + p->scale = 1; + } + return spec; +} + + + +#if (WITH_CR32S && WITH_CR32) || (WITH_CR64S && WITH_CR64) + #if defined __GNUC__ && defined __x86_64__ + #define CPUID(type, eax_, ebx_, ecx_, edx_) \ + __asm__ __volatile__ ( \ + "cpuid \n\t" \ + : "=a" (eax_), "=b" (ebx_), "=c" (ecx_), "=d" (edx_) \ + : "a" (type), "c" (0)); + #elif defined __GNUC__ && defined __i386__ + #define CPUID(type, eax_, ebx_, ecx_, edx_) \ + __asm__ __volatile__ ( \ + "mov %%ebx, %%edi \n\t" \ + "cpuid \n\t" \ + "xchg %%edi, %%ebx \n\t" \ + : "=a" (eax_), "=D" (ebx_), "=c" (ecx_), "=d" (edx_) \ + : "a" (type), "c" (0)); + #elif defined _M_X64 && defined _MSC_VER && _MSC_VER > 1500 + void __cpuidex(int CPUInfo[4], int info_type, int ecxvalue); + #pragma intrinsic(__cpuidex) + #define CPUID(type, eax_, ebx_, ecx_, edx_) do { \ + int regs[4]; \ + __cpuidex(regs, type, 0); \ + eax_ = regs[0], ebx_ = regs[1], ecx_ = regs[2], edx_ = regs[3]; \ + } while(0) + #elif defined _M_X64 && defined _MSC_VER + void __cpuidex(int CPUInfo[4], int info_type); + #pragma intrinsic(__cpuidex) + #define CPUID(type, eax_, ebx_, ecx_, edx_) do { \ + int regs[4]; \ + __cpuidex(regs, type); \ + eax_ = regs[0], ebx_ = regs[1], ecx_ = regs[2], edx_ = regs[3]; \ + } while(0) + #elif defined _M_IX86 && defined _MSC_VER + #define CPUID(type, eax_, ebx_, ecx_, edx_) \ + __asm pushad \ + __asm mov eax, type \ + __asm xor ecx, ecx \ + __asm cpuid \ + __asm mov eax_, eax \ + __asm mov ebx_, ebx \ + __asm mov ecx_, ecx \ + __asm mov edx_, edx \ + __asm popad + #endif +#endif + + + +#if WITH_CR32S && WITH_CR32 + static bool cpu_has_simd32(void) + { + #if defined __x86_64__ || defined _M_X64 + return true; + #elif defined __i386__ || defined _M_IX86 + enum {SSE = 1 << 25, SSE2 = 1 << 26}; + unsigned eax_, ebx_, ecx_, edx_; + CPUID(1, eax_, ebx_, ecx_, edx_); + return (edx_ & (SSE|SSE2)) != 0; + #elif defined AV_CPU_FLAG_NEON + return !!(av_get_cpu_flags() & AV_CPU_FLAG_NEON); + #else + return false; + #endif + } + + static bool should_use_simd32(void) + { + char const * e; + return ((e = getenv("SOXR_USE_SIMD" )))? !!atoi(e) : + ((e = getenv("SOXR_USE_SIMD32")))? !!atoi(e) : cpu_has_simd32(); + } +#else + #define should_use_simd32() true +#endif + + + +#if WITH_CR64S && WITH_CR64 + #if defined __GNUC__ + #define XGETBV(type, eax_, edx_) \ + __asm__ __volatile__ ( \ + ".byte 0x0f, 0x01, 0xd0\n" \ + : "=a"(eax_), "=d"(edx_) : "c" (type)); + #elif defined _M_X64 && defined _MSC_FULL_VER && _MSC_FULL_VER >= 160040219 + #include + #define XGETBV(type, eax_, edx_) do { \ + union {uint64_t x; uint32_t y[2];} a = {_xgetbv(0)}; \ + eax_ = a.y[0], edx_ = a.y[1]; \ + } while(0) + #elif defined _M_IX86 && defined _MSC_VER + #define XGETBV(type, eax_, edx_) \ + __asm pushad \ + __asm mov ecx, type \ + __asm _emit 0x0f \ + __asm _emit 0x01 \ + __asm _emit 0xd0 \ + __asm mov eax_, eax \ + __asm mov edx_, edx \ + __asm popad + #else + #define XGETBV(type, eax_, edx_) eax_ = edx_ = 0 + #endif + + static bool cpu_has_simd64(void) + { + enum {OSXSAVE = 1 << 27, AVX = 1 << 28}; + unsigned eax_, ebx_, ecx_, edx_; + CPUID(1, eax_, ebx_, ecx_, edx_); + if ((ecx_ & (OSXSAVE|AVX)) == (OSXSAVE|AVX)) { + XGETBV(0, eax_, edx_); + return (eax_ & 6) == 6; + } + return false; + } + + static bool should_use_simd64(void) + { + char const * e; + return ((e = getenv("SOXR_USE_SIMD" )))? !!atoi(e) : + ((e = getenv("SOXR_USE_SIMD64")))? !!atoi(e) : cpu_has_simd64(); + } +#else + #define should_use_simd64() true +#endif + + + +extern control_block_t + _soxr_rate32_cb, + _soxr_rate32s_cb, + _soxr_rate64_cb, + _soxr_rate64s_cb, + _soxr_vr32_cb; + + + +static void runtime_num(char const * env_name, + int min, int max, unsigned * field) +{ + char const * e = getenv(env_name); + if (e) { + int i = atoi(e); + if (i >= min && i <= max) + *field = (unsigned)i; + } +} + + + +static void runtime_flag(char const * env_name, + unsigned n_bits, unsigned n_shift, unsigned long * flags) +{ + char const * e = getenv(env_name); + if (e) { + int i = atoi(e); + unsigned long mask = (1UL << n_bits) - 1; + if (i >= 0 && i <= (int)mask) + *flags &= ~(mask << n_shift), *flags |= ((unsigned long)i << n_shift); + } +} + + + +soxr_t soxr_create( + double input_rate, double output_rate, + unsigned num_channels, + soxr_error_t * error0, + soxr_io_spec_t const * io_spec, + soxr_quality_spec_t const * q_spec, + soxr_runtime_spec_t const * runtime_spec) +{ + double io_ratio = output_rate!=0? input_rate!=0? + input_rate / output_rate : -1 : input_rate!=0? -1 : 0; + static const float datatype_full_scale[] = {1, 1, 65536.*32768, 32768}; + soxr_t p = 0; + soxr_error_t error = 0; + +#if WITH_DEV_TRACE +#define _(x) (char)(sizeof(x)>=10? 'a'+(char)(sizeof(x)-10):'0'+(char)sizeof(x)) + char const * e = getenv("SOXR_TRACE"); + _soxr_trace_level = e? atoi(e) : 0; + { + static char const arch[] = {_(char), _(short), _(int), _(long), _(long long) + , ' ', _(float), _(double), _(long double) + , ' ', _(int *), _(int (*)(int)) + , ' ', HAVE_BIGENDIAN ? 'B' : 'L' +#if defined _OPENMP + , ' ', 'O', 'M', 'P' +#endif + , 0}; +#undef _ + lsx_debug("arch: %s", arch); + } +#endif + + if (q_spec && q_spec->e) error = q_spec->e; + else if (io_spec && (io_spec->itype | io_spec->otype) >= SOXR_SPLIT * 2) + error = "invalid io datatype(s)"; + + if (!error && !(p = calloc(sizeof(*p), 1))) error = "malloc failed"; + + if (p) { + control_block_t * control_block; + + p->q_spec = q_spec? *q_spec : soxr_quality_spec(SOXR_HQ, 0); + + if (q_spec) { /* Backwards compatibility with original API: */ + if (p->q_spec.passband_end > 2) + p->q_spec.passband_end /= 100; + if (p->q_spec.stopband_begin > 2) + p->q_spec.stopband_begin = 2 - p->q_spec.stopband_begin / 100; + } + + p->io_ratio = io_ratio; + p->num_channels = num_channels; + if (io_spec) + p->io_spec = *io_spec; + else + p->io_spec.scale = 1; + + p->runtime_spec = runtime_spec? *runtime_spec : soxr_runtime_spec(1); + + runtime_num("SOXR_MIN_DFT_SIZE", 8, 15, &p->runtime_spec.log2_min_dft_size); + runtime_num("SOXR_LARGE_DFT_SIZE", 8, 20, &p->runtime_spec.log2_large_dft_size); + runtime_num("SOXR_COEFS_SIZE", 100, 800, &p->runtime_spec.coef_size_kbytes); + runtime_num("SOXR_NUM_THREADS", 0, 64, &p->runtime_spec.num_threads); + runtime_flag("SOXR_COEF_INTERP", 2, 0, &p->runtime_spec.flags); + + runtime_flag("SOXR_STRICT_BUF", 1, 2, &p->runtime_spec.flags); + runtime_flag("SOXR_NOSMALLINTOPT", 1, 3, &p->runtime_spec.flags); + + p->io_spec.scale *= datatype_full_scale[p->io_spec.otype & 3] / + datatype_full_scale[p->io_spec.itype & 3]; + + p->seed = (unsigned long)time(0) ^ (unsigned long)(size_t)p; + +#if WITH_CR32 || WITH_CR32S || WITH_VR32 + if (0 +#if WITH_VR32 + || ((!WITH_CR32 && !WITH_CR32S) || (p->q_spec.flags & SOXR_VR)) +#endif +#if WITH_CR32 || WITH_CR32S + || !(WITH_CR64 || WITH_CR64S) || (p->q_spec.precision <= 20 && !(p->q_spec.flags & SOXR_DOUBLE_PRECISION)) +#endif + ) { + p->deinterleave = (deinterleave_t)_soxr_deinterleave_f; + p->interleave = (interleave_t)_soxr_interleave_f; + control_block = +#if WITH_VR32 + ((!WITH_CR32 && !WITH_CR32S) || (p->q_spec.flags & SOXR_VR))? &_soxr_vr32_cb : +#endif +#if WITH_CR32S + !WITH_CR32 || should_use_simd32()? &_soxr_rate32s_cb : +#endif + &_soxr_rate32_cb; + } +#if WITH_CR64 || WITH_CR64S + else +#endif +#endif +#if WITH_CR64 || WITH_CR64S + { + p->deinterleave = (deinterleave_t)_soxr_deinterleave; + p->interleave = (interleave_t)_soxr_interleave; + control_block = +#if WITH_CR64S + !WITH_CR64 || should_use_simd64()? &_soxr_rate64s_cb : +#endif + &_soxr_rate64_cb; + } +#endif + memcpy(&p->control_block, control_block, sizeof(p->control_block)); + + if (p->num_channels && io_ratio!=0) + error = soxr_set_io_ratio(p, io_ratio, 0); + } + if (error) + soxr_delete(p), p = 0; + if (error0) + *error0 = error; + return p; +} + + + +soxr_error_t soxr_set_input_fn(soxr_t p, + soxr_input_fn_t input_fn, void * input_fn_state, size_t max_ilen) +{ + p->input_fn_state = input_fn_state; + p->input_fn = input_fn; + p->max_ilen = max_ilen? max_ilen : (size_t)-1; + return 0; +} + + + +static void soxr_delete0(soxr_t p) +{ + unsigned i; + + if (p->resamplers) for (i = 0; i < p->num_channels; ++i) { + if (p->resamplers[i]) + resampler_close(p->resamplers[i]); + free(p->resamplers[i]); + } + free(p->resamplers); + free(p->channel_ptrs); + free(p->shared); + + memset(p, 0, sizeof(*p)); +} + + + +double soxr_delay(soxr_t p) +{ + return + (p && !p->error && p->resamplers)? resampler_delay(p->resamplers[0]) : 0; +} + + + +static soxr_error_t fatal_error(soxr_t p, soxr_error_t error) +{ + soxr_delete0(p); + return p->error = error; +} + + + +static soxr_error_t initialise(soxr_t p) +{ + unsigned i; + size_t shared_size, channel_size; + + resampler_sizes(&shared_size, &channel_size); + p->channel_ptrs = calloc(sizeof(*p->channel_ptrs), p->num_channels); + p->shared = calloc(shared_size, 1); + p->resamplers = calloc(sizeof(*p->resamplers), p->num_channels); + if (!p->shared || !p->channel_ptrs || !p->resamplers) + return fatal_error(p, "malloc failed"); + + for (i = 0; i < p->num_channels; ++i) { + soxr_error_t error; + if (!(p->resamplers[i] = calloc(channel_size, 1))) + return fatal_error(p, "malloc failed"); + error = resampler_create( + p->resamplers[i], + p->shared, + p->io_ratio, + &p->q_spec, + &p->runtime_spec, + p->io_spec.scale); + if (error) + return fatal_error(p, error); + } + return 0; +} + + + +soxr_error_t soxr_set_num_channels(soxr_t p, unsigned num_channels) +{ + if (!p) return "invalid soxr_t pointer"; + if (num_channels == p->num_channels) return p->error; + if (!num_channels) return "invalid # of channels"; + if (p->resamplers) return "# of channels can't be changed"; + p->num_channels = num_channels; + return soxr_set_io_ratio(p, p->io_ratio, 0); +} + + + +soxr_error_t soxr_set_io_ratio(soxr_t p, double io_ratio, size_t slew_len) +{ + unsigned i; + soxr_error_t error; + if (!p) return "invalid soxr_t pointer"; + if ((error = p->error)) return error; + if (!p->num_channels) return "must set # channels before O/I ratio"; + if (io_ratio <= 0) return "I/O ratio out-of-range"; + if (!p->channel_ptrs) { + p->io_ratio = io_ratio; + return initialise(p); + } + if (p->control_block[8]) { + for (i = 0; !error && i < p->num_channels; ++i) + resampler_set_io_ratio(p->resamplers[i], io_ratio, slew_len); + return error; + } + return fabs(p->io_ratio - io_ratio) < 1e-15? 0 : + "varying O/I ratio is not supported with this quality level"; +} + + + +void soxr_delete(soxr_t p) +{ + if (p) + soxr_delete0(p), free(p); +} + + + +soxr_error_t soxr_clear(soxr_t p) /* TODO: this, properly. */ +{ + if (p) { + struct soxr tmp = *p; + soxr_delete0(p); + memset(p, 0, sizeof(*p)); + p->input_fn = tmp.input_fn; + p->runtime_spec = tmp.runtime_spec; + p->q_spec = tmp.q_spec; + p->io_spec = tmp.io_spec; + p->num_channels = tmp.num_channels; + p->input_fn_state = tmp.input_fn_state; + memcpy(p->control_block, tmp.control_block, sizeof(p->control_block)); + p->deinterleave = tmp.deinterleave; + p->interleave = tmp.interleave; + return (p->q_spec.flags & RESET_ON_CLEAR)? + soxr_set_io_ratio(p, tmp.io_ratio, 0) : 0; + } + return "invalid soxr_t pointer"; +} + + + +static void soxr_input_1ch(soxr_t p, unsigned i, soxr_cbuf_t src, size_t len) +{ + sample_t * dest = resampler_input(p->resamplers[i], NULL, len); + (*p->deinterleave)(&dest, p->io_spec.itype, &src, len, 1); +} + + + +static size_t soxr_input(soxr_t p, void const * in, size_t len) +{ + bool separated = !!(p->io_spec.itype & SOXR_SPLIT); + unsigned i; + if (!p || p->error) return 0; + if (!in && len) {p->error = "null input buffer pointer"; return 0;} + if (!len) { + p->flushing = true; + return 0; + } + if (separated) + for (i = 0; i < p->num_channels; ++i) + soxr_input_1ch(p, i, ((soxr_cbufs_t)in)[i], len); + else { + for (i = 0; i < p->num_channels; ++i) + p->channel_ptrs[i] = resampler_input(p->resamplers[i], NULL, len); + (*p->deinterleave)( + (sample_t **)p->channel_ptrs, p->io_spec.itype, &in, len, p->num_channels); + } + return len; +} + + + +static size_t soxr_output_1ch(soxr_t p, unsigned i, soxr_buf_t dest, size_t len, bool separated) +{ + sample_t const * src; + if (p->flushing) + resampler_flush(p->resamplers[i]); + resampler_process(p->resamplers[i], len); + src = resampler_output(p->resamplers[i], NULL, &len); + if (separated) + p->clips += (p->interleave)(p->io_spec.otype, &dest, &src, + len, 1, (p->io_spec.flags & SOXR_NO_DITHER)? 0 : &p->seed); + else p->channel_ptrs[i] = (void /* const */ *)src; + return len; +} + + + +static size_t soxr_output_no_callback(soxr_t p, soxr_buf_t out, size_t len) +{ + unsigned u; + size_t done = 0; + bool separated = !!(p->io_spec.otype & SOXR_SPLIT); +#if defined _OPENMP + int i; + if (!p->runtime_spec.num_threads && p->num_channels > 1) +#pragma omp parallel for + for (i = 0; i < (int)p->num_channels; ++i) { + size_t done1; + done1 = soxr_output_1ch(p, (unsigned)i, ((soxr_bufs_t)out)[i], len, separated); + if (!i) + done = done1; + } else +#endif + for (u = 0; u < p->num_channels; ++u) + done = soxr_output_1ch(p, u, ((soxr_bufs_t)out)[u], len, separated); + + if (!separated) + p->clips += (p->interleave)(p->io_spec.otype, &out, (sample_t const * const *)p->channel_ptrs, + done, p->num_channels, (p->io_spec.flags & SOXR_NO_DITHER)? 0 : &p->seed); + return done; +} + + + +size_t soxr_output(soxr_t p, void * out, size_t len0) +{ + size_t odone, odone0 = 0, olen = len0, osize, idone; + size_t ilen = min(p->max_ilen, (size_t)ceil((double)olen *p->io_ratio)); + void const * in = out; /* Set to !=0, so that caller may leave unset. */ + bool was_flushing; + + if (!p || p->error) return 0; + if (!out && len0) {p->error = "null output buffer pointer"; return 0;} + + do { + odone = soxr_output_no_callback(p, out, olen); + odone0 += odone; + if (odone0 == len0 || !p->input_fn || p->flushing) + break; + + osize = soxr_datatype_size(p->io_spec.otype) * p->num_channels; + out = (char *)out + osize * odone; + olen -= odone; + idone = p->input_fn(p->input_fn_state, &in, ilen); + was_flushing = p->flushing; + if (!in) + p->error = "input function reported failure"; + else soxr_input(p, in, idone); + } while (odone || idone || (!was_flushing && p->flushing)); + return odone0; +} + + + +static size_t soxr_i_for_o(soxr_t p, size_t olen, size_t ilen) +{ + size_t result; +#if 0 + if (p->runtime_spec.flags & SOXR_STRICT_BUFFERING) + result = rate_i_for_o(p->resamplers[0], olen); + else +#endif + result = (size_t)ceil((double)olen * p->io_ratio); + return min(result, ilen); +} + + + +#if 0 +static size_t soxr_o_for_i(soxr_t p, size_t ilen, size_t olen) +{ + size_t result = (size_t)ceil((double)ilen / p->io_ratio); + return min(result, olen); +} +#endif + + + +soxr_error_t soxr_process(soxr_t p, + void const * in , size_t ilen0, size_t * idone0, + void * out, size_t olen , size_t * odone0) +{ + size_t ilen, idone, odone = 0; + unsigned u; + bool flush_requested = false; + + if (!p) return "null pointer"; + + if (!in) + flush_requested = true, ilen = ilen0 = 0; + else { + if ((ptrdiff_t)ilen0 < 0) + flush_requested = true, ilen0 = ~ilen0; + if (idone0 && (1 || flush_requested)) + ilen = soxr_i_for_o(p, olen, ilen0); + else + ilen = ilen0/*, olen = soxr_o_for_i(p, ilen, olen)*/; + } + p->flushing |= ilen == ilen0 && flush_requested; + + if (!out && !in) + idone = ilen; + else if (p->io_spec.itype & p->io_spec.otype & SOXR_SPLIT) { /* Both i & o */ +#if defined _OPENMP + int i; + if (!p->runtime_spec.num_threads && p->num_channels > 1) +#pragma omp parallel for + for (i = 0; i < (int)p->num_channels; ++i) { + size_t done; + if (in) + soxr_input_1ch(p, (unsigned)i, ((soxr_cbufs_t)in)[i], ilen); + done = soxr_output_1ch(p, (unsigned)i, ((soxr_bufs_t)out)[i], olen, true); + if (!i) + odone = done; + } else +#endif + for (u = 0; u < p->num_channels; ++u) { + if (in) + soxr_input_1ch(p, u, ((soxr_cbufs_t)in)[u], ilen); + odone = soxr_output_1ch(p, u, ((soxr_bufs_t)out)[u], olen, true); + } + idone = ilen; + } + else { + idone = ilen? soxr_input (p, in , ilen) : 0; + odone = soxr_output(p, out, olen); + } + if (idone0) *idone0 = idone; + if (odone0) *odone0 = odone; + return p->error; +} + + + +soxr_error_t soxr_oneshot( + double irate, double orate, + unsigned num_channels, + void const * in , size_t ilen, size_t * idone, + void * out, size_t olen, size_t * odone, + soxr_io_spec_t const * io_spec, + soxr_quality_spec_t const * q_spec, + soxr_runtime_spec_t const * runtime_spec) +{ + soxr_t resampler; + soxr_error_t error = q_spec? q_spec->e : 0; + if (!error) { + soxr_quality_spec_t q_spec1; + if (!q_spec) + q_spec1 = soxr_quality_spec(SOXR_LQ, 0), q_spec = &q_spec1; + resampler = soxr_create(irate, orate, num_channels, + &error, io_spec, q_spec, runtime_spec); + } + if (!error) { + error = soxr_process(resampler, in, ~ilen, idone, out, olen, odone); + soxr_delete(resampler); + } + return error; +} + + + +soxr_error_t soxr_set_error(soxr_t p, soxr_error_t error) +{ + if (!p) return "null pointer"; + if (!p->error && p->error != error) return p->error; + p->error = error; + return 0; +} diff --git a/soxr-sys/src/soxr.h b/soxr-sys/src/soxr.h new file mode 100644 index 000000000..09ec7c466 --- /dev/null +++ b/soxr-sys/src/soxr.h @@ -0,0 +1,344 @@ +/* SoX Resampler Library Copyright (c) 2007-18 robs@users.sourceforge.net + * + * This library is free software; you can redistribute it and/or modify it + * under the terms of the GNU Lesser General Public License as published by + * the Free Software Foundation; either version 2.1 of the License, or (at + * your option) any later version. + * + * This library 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 Lesser + * General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public License + * along with this library; if not, write to the Free Software Foundation, + * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + + + +/* -------------------------------- Gubbins --------------------------------- */ + +#if !defined soxr_included +#define soxr_included + + +#if defined __cplusplus + #include + extern "C" { +#else + #include +#endif + +#if defined SOXR_DLL + #if defined soxr_EXPORTS + #define SOXR __declspec(dllexport) + #else + #define SOXR __declspec(dllimport) + #endif +#elif defined SOXR_VISIBILITY && defined __GNUC__ && (__GNUC__ > 4 || __GNUC__ == 4 && __GNUC_MINOR__ >= 1) + #define SOXR __attribute__ ((visibility("default"))) +#else + #define SOXR +#endif + +typedef struct soxr_io_spec soxr_io_spec_t; +typedef struct soxr_quality_spec soxr_quality_spec_t; +typedef struct soxr_runtime_spec soxr_runtime_spec_t; + + + +/* ---------------------------- API conventions -------------------------------- + +Buffer lengths (and occupancies) are expressed as the number of contained +samples per channel. + +Parameter names for buffer lengths have the suffix `len'. + +A single-character `i' or 'o' is often used in names to give context as +input or output (e.g. ilen, olen). */ + + + +/* --------------------------- Version management --------------------------- */ + +/* E.g. #if SOXR_THIS_VERSION >= SOXR_VERSION(0,1,1) ... */ + +#define SOXR_VERSION(x,y,z) (((x)<<16)|((y)<<8)|(z)) +#define SOXR_THIS_VERSION SOXR_VERSION(0,1,3) +#define SOXR_THIS_VERSION_STR "0.1.3" + + + +/* --------------------------- Type declarations ---------------------------- */ + +typedef struct soxr * soxr_t; /* A resampler for 1 or more channels. */ +typedef char const * soxr_error_t; /* 0:no-error; non-0:error. */ + +typedef void * soxr_buf_t; /* 1 buffer of channel-interleaved samples. */ +typedef void const * soxr_cbuf_t; /* Ditto; read-only. */ + +typedef soxr_buf_t const * soxr_bufs_t;/* Or, a separate buffer for each ch. */ +typedef soxr_cbuf_t const * soxr_cbufs_t; /* Ditto; read-only. */ + +typedef void const * soxr_in_t; /* Either a soxr_cbuf_t or soxr_cbufs_t, + depending on itype in soxr_io_spec_t. */ +typedef void * soxr_out_t; /* Either a soxr_buf_t or soxr_bufs_t, + depending on otype in soxr_io_spec_t. */ + + + +/* --------------------------- API main functions --------------------------- */ + +SOXR char const * soxr_version(void); /* Query library version: "libsoxr-x.y.z" */ + +#define soxr_strerror(e) /* Soxr counterpart to strerror. */ \ + ((e)?(e):"no error") + + +/* Create a stream resampler: */ + +SOXR soxr_t soxr_create( + double input_rate, /* Input sample-rate. */ + double output_rate, /* Output sample-rate. */ + unsigned num_channels, /* Number of channels to be used. */ + /* All following arguments are optional (may be set to NULL). */ + soxr_error_t *, /* To report any error during creation. */ + soxr_io_spec_t const *, /* To specify non-default I/O formats. */ + soxr_quality_spec_t const *, /* To specify non-default resampling quality.*/ + soxr_runtime_spec_t const *);/* To specify non-default runtime resources. + + Default io_spec is per soxr_io_spec(SOXR_FLOAT32_I, SOXR_FLOAT32_I) + Default quality_spec is per soxr_quality_spec(SOXR_HQ, 0) + Default runtime_spec is per soxr_runtime_spec(1) */ + + + +/* If not using an app-supplied input function, after creating a stream + * resampler, repeatedly call: */ + +SOXR soxr_error_t soxr_process( + soxr_t resampler, /* As returned by soxr_create. */ + /* Input (to be resampled): */ + soxr_in_t in, /* Input buffer(s); may be NULL (see below). */ + size_t ilen, /* Input buf. length (samples per channel). */ + size_t * idone, /* To return actual # samples used (<= ilen). */ + /* Output (resampled): */ + soxr_out_t out, /* Output buffer(s).*/ + size_t olen, /* Output buf. length (samples per channel). */ + size_t * odone); /* To return actual # samples out (<= olen). + + Note that no special meaning is associated with ilen or olen equal to + zero. End-of-input (i.e. no data is available nor shall be available) + may be indicated by seting `in' to NULL. */ + + + +/* If using an app-supplied input function, it must look and behave like this:*/ + +typedef size_t /* data_len */ + (* soxr_input_fn_t)( /* Supply data to be resampled. */ + void * input_fn_state, /* As given to soxr_set_input_fn (below). */ + soxr_in_t * data, /* Returned data; see below. N.B. ptr to ptr(s)*/ + size_t requested_len); /* Samples per channel, >= returned data_len. + + data_len *data Indicates Meaning + ------- ------- ------------ ------------------------- + !=0 !=0 Success *data contains data to be + input to the resampler. + 0 !=0 (or End-of-input No data is available nor + not set) shall be available. + 0 0 Failure An error occurred whilst trying to + source data to be input to the resampler. */ + +/* and be registered with a previously created stream resampler using: */ + +SOXR soxr_error_t soxr_set_input_fn(/* Set (or reset) an input function.*/ + soxr_t resampler, /* As returned by soxr_create. */ + soxr_input_fn_t, /* Function to supply data to be resampled.*/ + void * input_fn_state, /* If needed by the input function. */ + size_t max_ilen); /* Maximum value for input fn. requested_len.*/ + +/* then repeatedly call: */ + +SOXR size_t /*odone*/ soxr_output(/* Resample and output a block of data.*/ + soxr_t resampler, /* As returned by soxr_create. */ + soxr_out_t data, /* App-supplied buffer(s) for resampled data.*/ + size_t olen); /* Amount of data to output; >= odone. */ + + + +/* Common stream resampler operations: */ + +SOXR soxr_error_t soxr_error(soxr_t); /* Query error status. */ +SOXR size_t * soxr_num_clips(soxr_t); /* Query int. clip counter (for R/W). */ +SOXR double soxr_delay(soxr_t); /* Query current delay in output samples.*/ +SOXR char const * soxr_engine(soxr_t); /* Query resampling engine name. */ + +SOXR soxr_error_t soxr_clear(soxr_t); /* Ready for fresh signal, same config. */ +SOXR void soxr_delete(soxr_t); /* Free resources. */ + + + +/* `Short-cut', single call to resample a (probably short) signal held entirely + * in memory. See soxr_create and soxr_process above for parameter details. + * Note that unlike soxr_create however, the default quality spec. for + * soxr_oneshot is per soxr_quality_spec(SOXR_LQ, 0). */ + +SOXR soxr_error_t soxr_oneshot( + double input_rate, + double output_rate, + unsigned num_channels, + soxr_in_t in , size_t ilen, size_t * idone, + soxr_out_t out, size_t olen, size_t * odone, + soxr_io_spec_t const *, + soxr_quality_spec_t const *, + soxr_runtime_spec_t const *); + + + +/* For variable-rate resampling. See example # 5 for how to create a + * variable-rate resampler and how to use this function. */ + +SOXR soxr_error_t soxr_set_io_ratio(soxr_t, double io_ratio, size_t slew_len); + + + +/* -------------------------- API type definitions -------------------------- */ + +typedef enum { /* Datatypes supported for I/O to/from the resampler: */ + /* Internal; do not use: */ + SOXR_FLOAT32, SOXR_FLOAT64, SOXR_INT32, SOXR_INT16, SOXR_SPLIT = 4, + + /* Use for interleaved channels: */ + SOXR_FLOAT32_I = SOXR_FLOAT32, SOXR_FLOAT64_I, SOXR_INT32_I, SOXR_INT16_I, + + /* Use for split channels: */ + SOXR_FLOAT32_S = SOXR_SPLIT , SOXR_FLOAT64_S, SOXR_INT32_S, SOXR_INT16_S + +} soxr_datatype_t; + +#define soxr_datatype_size(x) /* Returns `sizeof' a soxr_datatype_t sample. */\ + ((unsigned char *)"\4\10\4\2")[(x)&3] + + + +struct soxr_io_spec { /* Typically */ + soxr_datatype_t itype; /* Input datatype. SOXR_FLOAT32_I */ + soxr_datatype_t otype; /* Output datatype. SOXR_FLOAT32_I */ + double scale; /* Linear gain to apply during resampling. 1 */ + void * e; /* Reserved for internal use 0 */ + unsigned long flags; /* Per the following #defines. 0 */ +}; + +#define SOXR_TPDF 0 /* Applicable only if otype is INT16. */ +#define SOXR_NO_DITHER 8u /* Disable the above. */ + + + +struct soxr_quality_spec { /* Typically */ + double precision; /* Conversion precision (in bits). 20 */ + double phase_response; /* 0=minimum, ... 50=linear, ... 100=maximum 50 */ + double passband_end; /* 0dB pt. bandwidth to preserve; nyquist=1 0.913*/ + double stopband_begin; /* Aliasing/imaging control; > passband_end 1 */ + void * e; /* Reserved for internal use. 0 */ + unsigned long flags; /* Per the following #defines. 0 */ +}; + +#define SOXR_ROLLOFF_SMALL 0u /* <= 0.01 dB */ +#define SOXR_ROLLOFF_MEDIUM 1u /* <= 0.35 dB */ +#define SOXR_ROLLOFF_NONE 2u /* For Chebyshev bandwidth. */ + +#define SOXR_HI_PREC_CLOCK 8u /* Increase `irrational' ratio accuracy. */ +#define SOXR_DOUBLE_PRECISION 16u /* Use D.P. calcs even if precision <= 20. */ +#define SOXR_VR 32u /* Variable-rate resampling. */ + + + +struct soxr_runtime_spec { /* Typically */ + unsigned log2_min_dft_size; /* For DFT efficiency. [8,15] 10 */ + unsigned log2_large_dft_size; /* For DFT efficiency. [8,20] 17 */ + unsigned coef_size_kbytes; /* For SOXR_COEF_INTERP_AUTO (below). 400 */ + unsigned num_threads; /* 0: per OMP_NUM_THREADS; 1: 1 thread. 1 */ + void * e; /* Reserved for internal use. 0 */ + unsigned long flags; /* Per the following #defines. 0 */ +}; + /* For `irrational' ratios only: */ +#define SOXR_COEF_INTERP_AUTO 0u /* Auto select coef. interpolation. */ +#define SOXR_COEF_INTERP_LOW 2u /* Man. select: less CPU, more memory. */ +#define SOXR_COEF_INTERP_HIGH 3u /* Man. select: more CPU, less memory. */ + + + +/* -------------------------- API type constructors ------------------------- */ + +/* These functions allow setting of the most commonly-used structure + * parameters, with other parameters being given default values. The default + * values may then be overridden, directly in the structure, if needed. */ + +SOXR soxr_quality_spec_t soxr_quality_spec( + unsigned long recipe, /* Per the #defines immediately below. */ + unsigned long flags); /* As soxr_quality_spec_t.flags. */ + + /* The 5 standard qualities found in SoX: */ +#define SOXR_QQ 0 /* 'Quick' cubic interpolation. */ +#define SOXR_LQ 1 /* 'Low' 16-bit with larger rolloff. */ +#define SOXR_MQ 2 /* 'Medium' 16-bit with medium rolloff. */ +#define SOXR_HQ SOXR_20_BITQ /* 'High quality'. */ +#define SOXR_VHQ SOXR_28_BITQ /* 'Very high quality'. */ + +#define SOXR_16_BITQ 3 +#define SOXR_20_BITQ 4 +#define SOXR_24_BITQ 5 +#define SOXR_28_BITQ 6 +#define SOXR_32_BITQ 7 + /* Reserved for internal use (to be removed): */ +#define SOXR_LSR0Q 8 /* 'Best sinc'. */ +#define SOXR_LSR1Q 9 /* 'Medium sinc'. */ +#define SOXR_LSR2Q 10 /* 'Fast sinc'. */ + +#define SOXR_LINEAR_PHASE 0x00 +#define SOXR_INTERMEDIATE_PHASE 0x10 +#define SOXR_MINIMUM_PHASE 0x30 + +#define SOXR_STEEP_FILTER 0x40 + + + +SOXR soxr_runtime_spec_t soxr_runtime_spec( + unsigned num_threads); + + + +SOXR soxr_io_spec_t soxr_io_spec( + soxr_datatype_t itype, + soxr_datatype_t otype); + + + +/* --------------------------- Advanced use only ---------------------------- */ + +/* For new designs, the following functions/usage will probably not be needed. + * They might be useful when adding soxr into an existing design where values + * for the resampling-rate and/or number-of-channels parameters to soxr_create + * are not available when that function will be called. In such cases, the + * relevant soxr_create parameter(s) can be given as 0, then one or both of the + * following (as appropriate) later invoked (but prior to calling soxr_process + * or soxr_output): + * + * soxr_set_error(soxr, soxr_set_io_ratio(soxr, io_ratio, 0)); + * soxr_set_error(soxr, soxr_set_num_channels(soxr, num_channels)); + */ + +SOXR soxr_error_t soxr_set_error(soxr_t, soxr_error_t); +SOXR soxr_error_t soxr_set_num_channels(soxr_t, unsigned); + + + +#undef SOXR + +#if defined __cplusplus +} +#endif + +#endif diff --git a/soxr-sys/src/soxr.rs b/soxr-sys/src/soxr.rs new file mode 100644 index 000000000..442dd03aa --- /dev/null +++ b/soxr-sys/src/soxr.rs @@ -0,0 +1,403 @@ +/* automatically generated by rust-bindgen 0.69.4 */ + +pub const SOXR_THIS_VERSION_STR: &[u8; 6] = b"0.1.3\0"; +pub const SOXR_TPDF: u32 = 0; +pub const SOXR_NO_DITHER: u32 = 8; +pub const SOXR_ROLLOFF_SMALL: u32 = 0; +pub const SOXR_ROLLOFF_MEDIUM: u32 = 1; +pub const SOXR_ROLLOFF_NONE: u32 = 2; +pub const SOXR_HI_PREC_CLOCK: u32 = 8; +pub const SOXR_DOUBLE_PRECISION: u32 = 16; +pub const SOXR_VR: u32 = 32; +pub const SOXR_COEF_INTERP_AUTO: u32 = 0; +pub const SOXR_COEF_INTERP_LOW: u32 = 2; +pub const SOXR_COEF_INTERP_HIGH: u32 = 3; +pub const SOXR_QQ: u32 = 0; +pub const SOXR_LQ: u32 = 1; +pub const SOXR_MQ: u32 = 2; +pub const SOXR_16_BITQ: u32 = 3; +pub const SOXR_20_BITQ: u32 = 4; +pub const SOXR_24_BITQ: u32 = 5; +pub const SOXR_28_BITQ: u32 = 6; +pub const SOXR_32_BITQ: u32 = 7; +pub const SOXR_LSR0Q: u32 = 8; +pub const SOXR_LSR1Q: u32 = 9; +pub const SOXR_LSR2Q: u32 = 10; +pub const SOXR_LINEAR_PHASE: u32 = 0; +pub const SOXR_INTERMEDIATE_PHASE: u32 = 16; +pub const SOXR_MINIMUM_PHASE: u32 = 48; +pub const SOXR_STEEP_FILTER: u32 = 64; +pub type wchar_t = ::std::os::raw::c_int; +pub type max_align_t = f64; +pub type soxr_io_spec_t = soxr_io_spec; +pub type soxr_quality_spec_t = soxr_quality_spec; +pub type soxr_runtime_spec_t = soxr_runtime_spec; +#[repr(C)] +#[derive(Debug, Copy, Clone)] +pub struct soxr { + _unused: [u8; 0], +} +pub type soxr_t = *mut soxr; +pub type soxr_error_t = *const ::std::os::raw::c_char; +pub type soxr_buf_t = *mut ::std::os::raw::c_void; +pub type soxr_cbuf_t = *const ::std::os::raw::c_void; +pub type soxr_bufs_t = *const soxr_buf_t; +pub type soxr_cbufs_t = *const soxr_cbuf_t; +pub type soxr_in_t = *const ::std::os::raw::c_void; +pub type soxr_out_t = *mut ::std::os::raw::c_void; +extern "C" { + pub fn soxr_version() -> *const ::std::os::raw::c_char; +} +extern "C" { + pub fn soxr_create( + input_rate: f64, + output_rate: f64, + num_channels: ::std::os::raw::c_uint, + arg1: *mut soxr_error_t, + arg2: *const soxr_io_spec_t, + arg3: *const soxr_quality_spec_t, + arg4: *const soxr_runtime_spec_t, + ) -> soxr_t; +} +extern "C" { + pub fn soxr_process( + resampler: soxr_t, + in_: soxr_in_t, + ilen: usize, + idone: *mut usize, + out: soxr_out_t, + olen: usize, + odone: *mut usize, + ) -> soxr_error_t; +} +pub type soxr_input_fn_t = ::std::option::Option< + unsafe extern "C" fn( + input_fn_state: *mut ::std::os::raw::c_void, + data: *mut soxr_in_t, + requested_len: usize, + ) -> usize, +>; +extern "C" { + pub fn soxr_set_input_fn( + resampler: soxr_t, + arg1: soxr_input_fn_t, + input_fn_state: *mut ::std::os::raw::c_void, + max_ilen: usize, + ) -> soxr_error_t; +} +extern "C" { + pub fn soxr_output(resampler: soxr_t, data: soxr_out_t, olen: usize) -> usize; +} +extern "C" { + pub fn soxr_error(arg1: soxr_t) -> soxr_error_t; +} +extern "C" { + pub fn soxr_num_clips(arg1: soxr_t) -> *mut usize; +} +extern "C" { + pub fn soxr_delay(arg1: soxr_t) -> f64; +} +extern "C" { + pub fn soxr_engine(arg1: soxr_t) -> *const ::std::os::raw::c_char; +} +extern "C" { + pub fn soxr_clear(arg1: soxr_t) -> soxr_error_t; +} +extern "C" { + pub fn soxr_delete(arg1: soxr_t); +} +extern "C" { + pub fn soxr_oneshot( + input_rate: f64, + output_rate: f64, + num_channels: ::std::os::raw::c_uint, + in_: soxr_in_t, + ilen: usize, + idone: *mut usize, + out: soxr_out_t, + olen: usize, + odone: *mut usize, + arg1: *const soxr_io_spec_t, + arg2: *const soxr_quality_spec_t, + arg3: *const soxr_runtime_spec_t, + ) -> soxr_error_t; +} +extern "C" { + pub fn soxr_set_io_ratio(arg1: soxr_t, io_ratio: f64, slew_len: usize) -> soxr_error_t; +} +pub const SOXR_FLOAT32: soxr_datatype_t = 0; +pub const SOXR_FLOAT64: soxr_datatype_t = 1; +pub const SOXR_INT32: soxr_datatype_t = 2; +pub const SOXR_INT16: soxr_datatype_t = 3; +pub const SOXR_SPLIT: soxr_datatype_t = 4; +pub const SOXR_FLOAT32_I: soxr_datatype_t = 0; +pub const SOXR_FLOAT64_I: soxr_datatype_t = 1; +pub const SOXR_INT32_I: soxr_datatype_t = 2; +pub const SOXR_INT16_I: soxr_datatype_t = 3; +pub const SOXR_FLOAT32_S: soxr_datatype_t = 4; +pub const SOXR_FLOAT64_S: soxr_datatype_t = 5; +pub const SOXR_INT32_S: soxr_datatype_t = 6; +pub const SOXR_INT16_S: soxr_datatype_t = 7; +pub type soxr_datatype_t = ::std::os::raw::c_uint; +#[repr(C)] +#[derive(Debug, Copy, Clone)] +pub struct soxr_io_spec { + pub itype: soxr_datatype_t, + pub otype: soxr_datatype_t, + pub scale: f64, + pub e: *mut ::std::os::raw::c_void, + pub flags: ::std::os::raw::c_ulong, +} +#[test] +fn bindgen_test_layout_soxr_io_spec() { + const UNINIT: ::std::mem::MaybeUninit = ::std::mem::MaybeUninit::uninit(); + let ptr = UNINIT.as_ptr(); + assert_eq!( + ::std::mem::size_of::(), + 32usize, + concat!("Size of: ", stringify!(soxr_io_spec)) + ); + assert_eq!( + ::std::mem::align_of::(), + 8usize, + concat!("Alignment of ", stringify!(soxr_io_spec)) + ); + assert_eq!( + unsafe { ::std::ptr::addr_of!((*ptr).itype) as usize - ptr as usize }, + 0usize, + concat!( + "Offset of field: ", + stringify!(soxr_io_spec), + "::", + stringify!(itype) + ) + ); + assert_eq!( + unsafe { ::std::ptr::addr_of!((*ptr).otype) as usize - ptr as usize }, + 4usize, + concat!( + "Offset of field: ", + stringify!(soxr_io_spec), + "::", + stringify!(otype) + ) + ); + assert_eq!( + unsafe { ::std::ptr::addr_of!((*ptr).scale) as usize - ptr as usize }, + 8usize, + concat!( + "Offset of field: ", + stringify!(soxr_io_spec), + "::", + stringify!(scale) + ) + ); + assert_eq!( + unsafe { ::std::ptr::addr_of!((*ptr).e) as usize - ptr as usize }, + 16usize, + concat!( + "Offset of field: ", + stringify!(soxr_io_spec), + "::", + stringify!(e) + ) + ); + assert_eq!( + unsafe { ::std::ptr::addr_of!((*ptr).flags) as usize - ptr as usize }, + 24usize, + concat!( + "Offset of field: ", + stringify!(soxr_io_spec), + "::", + stringify!(flags) + ) + ); +} +#[repr(C)] +#[derive(Debug, Copy, Clone)] +pub struct soxr_quality_spec { + pub precision: f64, + pub phase_response: f64, + pub passband_end: f64, + pub stopband_begin: f64, + pub e: *mut ::std::os::raw::c_void, + pub flags: ::std::os::raw::c_ulong, +} +#[test] +fn bindgen_test_layout_soxr_quality_spec() { + const UNINIT: ::std::mem::MaybeUninit = ::std::mem::MaybeUninit::uninit(); + let ptr = UNINIT.as_ptr(); + assert_eq!( + ::std::mem::size_of::(), + 48usize, + concat!("Size of: ", stringify!(soxr_quality_spec)) + ); + assert_eq!( + ::std::mem::align_of::(), + 8usize, + concat!("Alignment of ", stringify!(soxr_quality_spec)) + ); + assert_eq!( + unsafe { ::std::ptr::addr_of!((*ptr).precision) as usize - ptr as usize }, + 0usize, + concat!( + "Offset of field: ", + stringify!(soxr_quality_spec), + "::", + stringify!(precision) + ) + ); + assert_eq!( + unsafe { ::std::ptr::addr_of!((*ptr).phase_response) as usize - ptr as usize }, + 8usize, + concat!( + "Offset of field: ", + stringify!(soxr_quality_spec), + "::", + stringify!(phase_response) + ) + ); + assert_eq!( + unsafe { ::std::ptr::addr_of!((*ptr).passband_end) as usize - ptr as usize }, + 16usize, + concat!( + "Offset of field: ", + stringify!(soxr_quality_spec), + "::", + stringify!(passband_end) + ) + ); + assert_eq!( + unsafe { ::std::ptr::addr_of!((*ptr).stopband_begin) as usize - ptr as usize }, + 24usize, + concat!( + "Offset of field: ", + stringify!(soxr_quality_spec), + "::", + stringify!(stopband_begin) + ) + ); + assert_eq!( + unsafe { ::std::ptr::addr_of!((*ptr).e) as usize - ptr as usize }, + 32usize, + concat!( + "Offset of field: ", + stringify!(soxr_quality_spec), + "::", + stringify!(e) + ) + ); + assert_eq!( + unsafe { ::std::ptr::addr_of!((*ptr).flags) as usize - ptr as usize }, + 40usize, + concat!( + "Offset of field: ", + stringify!(soxr_quality_spec), + "::", + stringify!(flags) + ) + ); +} +#[repr(C)] +#[derive(Debug, Copy, Clone)] +pub struct soxr_runtime_spec { + pub log2_min_dft_size: ::std::os::raw::c_uint, + pub log2_large_dft_size: ::std::os::raw::c_uint, + pub coef_size_kbytes: ::std::os::raw::c_uint, + pub num_threads: ::std::os::raw::c_uint, + pub e: *mut ::std::os::raw::c_void, + pub flags: ::std::os::raw::c_ulong, +} +#[test] +fn bindgen_test_layout_soxr_runtime_spec() { + const UNINIT: ::std::mem::MaybeUninit = ::std::mem::MaybeUninit::uninit(); + let ptr = UNINIT.as_ptr(); + assert_eq!( + ::std::mem::size_of::(), + 32usize, + concat!("Size of: ", stringify!(soxr_runtime_spec)) + ); + assert_eq!( + ::std::mem::align_of::(), + 8usize, + concat!("Alignment of ", stringify!(soxr_runtime_spec)) + ); + assert_eq!( + unsafe { ::std::ptr::addr_of!((*ptr).log2_min_dft_size) as usize - ptr as usize }, + 0usize, + concat!( + "Offset of field: ", + stringify!(soxr_runtime_spec), + "::", + stringify!(log2_min_dft_size) + ) + ); + assert_eq!( + unsafe { ::std::ptr::addr_of!((*ptr).log2_large_dft_size) as usize - ptr as usize }, + 4usize, + concat!( + "Offset of field: ", + stringify!(soxr_runtime_spec), + "::", + stringify!(log2_large_dft_size) + ) + ); + assert_eq!( + unsafe { ::std::ptr::addr_of!((*ptr).coef_size_kbytes) as usize - ptr as usize }, + 8usize, + concat!( + "Offset of field: ", + stringify!(soxr_runtime_spec), + "::", + stringify!(coef_size_kbytes) + ) + ); + assert_eq!( + unsafe { ::std::ptr::addr_of!((*ptr).num_threads) as usize - ptr as usize }, + 12usize, + concat!( + "Offset of field: ", + stringify!(soxr_runtime_spec), + "::", + stringify!(num_threads) + ) + ); + assert_eq!( + unsafe { ::std::ptr::addr_of!((*ptr).e) as usize - ptr as usize }, + 16usize, + concat!( + "Offset of field: ", + stringify!(soxr_runtime_spec), + "::", + stringify!(e) + ) + ); + assert_eq!( + unsafe { ::std::ptr::addr_of!((*ptr).flags) as usize - ptr as usize }, + 24usize, + concat!( + "Offset of field: ", + stringify!(soxr_runtime_spec), + "::", + stringify!(flags) + ) + ); +} +extern "C" { + pub fn soxr_quality_spec( + recipe: ::std::os::raw::c_ulong, + flags: ::std::os::raw::c_ulong, + ) -> soxr_quality_spec_t; +} +extern "C" { + pub fn soxr_runtime_spec(num_threads: ::std::os::raw::c_uint) -> soxr_runtime_spec_t; +} +extern "C" { + pub fn soxr_io_spec(itype: soxr_datatype_t, otype: soxr_datatype_t) -> soxr_io_spec_t; +} +extern "C" { + pub fn soxr_set_error(arg1: soxr_t, arg2: soxr_error_t) -> soxr_error_t; +} +extern "C" { + pub fn soxr_set_num_channels(arg1: soxr_t, arg2: ::std::os::raw::c_uint) -> soxr_error_t; +} diff --git a/soxr-sys/src/std-types.h b/soxr-sys/src/std-types.h new file mode 100644 index 000000000..c5e8636ac --- /dev/null +++ b/soxr-sys/src/std-types.h @@ -0,0 +1,48 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#if !defined soxr_std_types_included +#define soxr_std_types_included + +#include "soxr-config.h" + +#include + +#if HAVE_STDBOOL_H + #include +#else + #undef bool + #undef false + #undef true + #define bool int + #define false 0 + #define true 1 +#endif + +#if HAVE_STDINT_H + #include +#else + #undef int16_t + #undef int32_t + #undef int64_t + #undef uint32_t + #undef uint64_t + #define int16_t short + #if LONG_MAX > 2147483647L + #define int32_t int + #define int64_t long + #elif LONG_MAX < 2147483647L + #error this library requires that 'long int' has at least 32-bits + #else + #define int32_t long + #if defined _MSC_VER + #define int64_t __int64 + #else + #define int64_t long long + #endif + #endif + #define uint32_t unsigned int32_t + #define uint64_t unsigned int64_t +#endif + +#endif diff --git a/soxr-sys/src/util-simd.c b/soxr-sys/src/util-simd.c new file mode 100644 index 000000000..ec548fdee --- /dev/null +++ b/soxr-sys/src/util-simd.c @@ -0,0 +1,89 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#include +#include +#include + +#include "soxr-config.h" + +#define SIMD_ALIGNMENT (sizeof(float) * (1 + (PFFFT_DOUBLE|AVCODEC_FOUND)) * 4) + +void * SIMD_ALIGNED_MALLOC(size_t size) +{ + char * p1 = 0, * p = malloc(size + SIMD_ALIGNMENT); + if (p) { + p1 = (char *)((size_t)(p + SIMD_ALIGNMENT) & ~(SIMD_ALIGNMENT - 1)); + *((void * *)p1 - 1) = p; + } + return p1; +} + + + +void * SIMD_ALIGNED_CALLOC(size_t nmemb, size_t size) +{ + void * p = SIMD_ALIGNED_MALLOC(nmemb * size); + if (p) + memset(p, 0, nmemb * size); + return p; +} + + + +void SIMD_ALIGNED_FREE(void * p1) +{ + if (p1) + free(*((void * *)p1 - 1)); +} + + + +#define PFFT_MACROS_ONLY +#include "pffft.c" + + + +void ORDERED_CONVOLVE_SIMD(int n, void * not_used, float * a, float const * b) +{ + int i; + float ab0, ab1; + v4sf * RESTRICT va = (v4sf *)a; + v4sf const * RESTRICT vb = (v4sf const *)b; + assert(VALIGNED(a) && VALIGNED(b)); + ab0 = a[0] * b[0], ab1 = a[1] * b[1]; + for (i = 0; i < n / 4; i += 2) { + v4sf a1r = va[i+0], a1i = va[i+1]; + v4sf b1r = vb[i+0], b1i = vb[i+1]; + UNINTERLEAVE2(a1r, a1i, a1r, a1i); + UNINTERLEAVE2(b1r, b1i, b1r, b1i); + VCPLXMUL(a1r, a1i, b1r, b1i); + INTERLEAVE2(a1r, a1i, a1r, a1i); + va[i+0] = a1r, va[i+1] = a1i; + } + a[0] = ab0, a[1] = ab1; + (void)not_used; +} + + + +void ORDERED_PARTIAL_CONVOLVE_SIMD(int n, float * a, float const * b) +{ + int i; + float ab0; + v4sf * RESTRICT va = (v4sf *)a; + v4sf const * RESTRICT vb = (v4sf const *)b; + assert(VALIGNED(a) && VALIGNED(b)); + ab0 = a[0] * b[0]; + for (i = 0; i < n / 4; i += 2) { + v4sf a1r = va[i+0], a1i = va[i+1]; + v4sf b1r = vb[i+0], b1i = vb[i+1]; + UNINTERLEAVE2(a1r, a1i, a1r, a1i); + UNINTERLEAVE2(b1r, b1i, b1r, b1i); + VCPLXMUL(a1r, a1i, b1r, b1i); + INTERLEAVE2(a1r, a1i, a1r, a1i); + va[i+0] = a1r, va[i+1] = a1i; + } + a[0] = ab0; + a[1] = b[n] * a[n] - b[n+1] * a[n+1]; +} diff --git a/soxr-sys/src/util32s.c b/soxr-sys/src/util32s.c new file mode 100644 index 000000000..b9c9e08bd --- /dev/null +++ b/soxr-sys/src/util32s.c @@ -0,0 +1,8 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#define PFFFT_DOUBLE 0 + +#include "util32s.h" + +#include "util-simd.c" diff --git a/soxr-sys/src/util32s.h b/soxr-sys/src/util32s.h new file mode 100644 index 000000000..12226e501 --- /dev/null +++ b/soxr-sys/src/util32s.h @@ -0,0 +1,23 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#if !defined soxr_util32s_included +#define soxr_util32s_included + +#include + +void * _soxr_simd32_aligned_malloc(size_t); +void * _soxr_simd32_aligned_calloc(size_t, size_t); +void _soxr_simd32_aligned_free(void *); + +#define SIMD_ALIGNED_MALLOC _soxr_simd32_aligned_malloc +#define SIMD_ALIGNED_CALLOC _soxr_simd32_aligned_calloc +#define SIMD_ALIGNED_FREE _soxr_simd32_aligned_free + +void _soxr_ordered_convolve_simd32(int n, void * not_used, float * a, float const * b); +void _soxr_ordered_partial_convolve_simd32(int n, float * a, float const * b); + +#define ORDERED_CONVOLVE_SIMD _soxr_ordered_convolve_simd32 +#define ORDERED_PARTIAL_CONVOLVE_SIMD _soxr_ordered_partial_convolve_simd32 + +#endif diff --git a/soxr-sys/src/util64s.c b/soxr-sys/src/util64s.c new file mode 100644 index 000000000..0faa9e9ef --- /dev/null +++ b/soxr-sys/src/util64s.c @@ -0,0 +1,8 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#define PFFFT_DOUBLE 1 + +#include "util64s.h" + +#include "util-simd.c" diff --git a/soxr-sys/src/util64s.h b/soxr-sys/src/util64s.h new file mode 100644 index 000000000..7beeb8991 --- /dev/null +++ b/soxr-sys/src/util64s.h @@ -0,0 +1,23 @@ +/* SoX Resampler Library Copyright (c) 2007-16 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +#if !defined soxr_util64s_included +#define soxr_util64s_included + +#include + +void * _soxr_simd64_aligned_malloc(size_t); +void * _soxr_simd64_aligned_calloc(size_t, size_t); +void _soxr_simd64_aligned_free(void *); + +#define SIMD_ALIGNED_MALLOC _soxr_simd64_aligned_malloc +#define SIMD_ALIGNED_CALLOC _soxr_simd64_aligned_calloc +#define SIMD_ALIGNED_FREE _soxr_simd64_aligned_free + +void _soxr_ordered_convolve_simd64(int n, void * not_used, double * a, double const * b); +void _soxr_ordered_partial_convolve_simd64(int n, double * a, double const * b); + +#define ORDERED_CONVOLVE_SIMD _soxr_ordered_convolve_simd64 +#define ORDERED_PARTIAL_CONVOLVE_SIMD _soxr_ordered_partial_convolve_simd64 + +#endif diff --git a/soxr-sys/src/vr-coefs.c b/soxr-sys/src/vr-coefs.c new file mode 100644 index 000000000..a57bec8c2 --- /dev/null +++ b/soxr-sys/src/vr-coefs.c @@ -0,0 +1,115 @@ +/* SoX Resampler Library Copyright (c) 2013 robs@users.sourceforge.net + * Licence for this file: LGPL v2.1 See LICENCE for details. */ + +/* Generate the filter coefficients for variable-rate resampling. */ + +#include +#include +#include +#define PI 3.14159265358979323846 /* Since M_PI can't be relied on */ + +static void print(double * h, int m, double l, char const * name) +{ /* Print out a filter: */ + int i, N = l? (int)(l*m)-(l>1) : m, R=(N+1)/2; + int a = !l||l>1? 0:N-R, b = l>1? R:N; + printf("static float const %s[] = {\n", name); + if (l>1) printf(" 0.f,"); else if (!l) l=1; + for (i=a; h && i 0 && x E[i] >= x E[i z 1]) +#define PEAK do {if (k0)-(E[i]<0);} while (0) + +typedef struct {double x, beta, gamma;} coef_t; + +static double amp_response(coef_t * coef, int R, double f, int i) +{ + double n = 0, d = 0, x = cos(PI*f), t; + for (; i < R; d += t = coef[i].beta / t, n += coef[i].gamma * t, ++i) + if (fabs(t = x - coef[i].x) < 1e-9) return coef[i].gamma; + return n/d; +} + +static void fir(int m, double l, double Fp0, double Fs0, + double weight0, int density, char const * name) +{ + double Fp=Fp0/l, Fs=Fs0/l, weight=1/weight0, inc[2], Ws=1-Fs; + int N = (int)(l*m)-(l>1), R=(N+1)/2, NP=R+1, grid_size=1+density*R+1, pass=0; + int n1 = Ws>=(2*R-1)*Fp? 1:(int)(R*Fp/(Fp+Ws)+.5), n2=NP-n1, _1, i, j, k; + int * peak = calloc(sizeof(*peak), (size_t)(NP+1)), * P=peak, end[2]; + coef_t * coef = calloc(sizeof(*coef), (size_t)(NP)); + float * E = calloc(sizeof(*E ), (size_t)(grid_size)); + double d, n, e, f, mult, delta, sum, hi, lo, * A = (double*)E, *h=0; + + if (!P || !coef || !E) goto END; + end[0] = n1 * density, end[1] = grid_size-1; /* Create prototype peaks: */ + inc[0] = Fp/end[0], inc[1] = n2==1? 0 : Ws / ((n2-1)*density); + for (i=0; iE[i+1]) || (EE(-,-) && E[i] 1) goto END; /* Too many/few? */ + P = peak + k * (fabs(E[peak[0]]) < fabs(E[peak[NP]])); /* rm 1st? */ + + for (lo = hi = fabs(E[P[0]]), i=1; ihi? e:hi; + } while ((hi-lo)/hi > .001 && ++pass < 20); + /* Create impulse response from final amp. resp. coefs: */ + if (!(h = malloc(sizeof(*h)*(size_t)N))) goto END; + for (i = 0; i < R; f = 2.*i/N, A[i++] = amp_response(coef,R,f,0)*even_adj(f)); + for (i = 0; i < R; h[N-1-i] = h[i] = sum/N, ++i) + for (sum=*A, j=1; j +#include "math-wrap.h" +#include +#include +#include "internal.h" +#define FIFO_SIZE_T int +#define FIFO_MIN 0x8000 +#include "fifo.h" +#include "vr-coefs.h" + +#define FADE_LEN_BITS 9 +#define PHASE_BITS_D 10 +#define PHASE_BITS_U 9 + +#define PHASES0_D 12 +#define POLY_FIR_LEN_D 20 +#define PHASES0_U 6 +#define POLY_FIR_LEN_U 12 + +#define MULT32 (65536. * 65536.) +#define PHASES_D (1 << PHASE_BITS_D) +#define PHASES_U (1 << PHASE_BITS_U) + +#define CONVOLVE \ + _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \ + _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \ + _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ + +#define HALF_FIR_LEN_2 (iAL(half_fir_coefs) - 1) +#define HALF_FIR_LEN_4 (HALF_FIR_LEN_2 / 2) + +#define _ sum += (input[-i] + input[i]) * half_fir_coefs[i], ++i; +static float half_fir(float const * input) +{ + long i = 1; + float sum = input[0] * half_fir_coefs[0]; + CONVOLVE CONVOLVE + assert(i == HALF_FIR_LEN_2 + 1); + return (float)sum; +} +#undef _ + +#define _ sum += (input[-i] + input[i]) * half_fir_coefs[2*i], ++i; +static float double_fir0(float const * input) +{ + int i = 1; + float sum = input[0] * half_fir_coefs[0]; + CONVOLVE + assert(i == HALF_FIR_LEN_4 + 1); + return (float)(sum * 2); +} +#undef _ + +#define _ sum += (input[-i] + input[1+i]) * half_fir_coefs[2*i+1], ++i; +static float double_fir1(float const * input) +{ + int i = 0; + float sum = 0; + CONVOLVE + assert(i == HALF_FIR_LEN_4 + 0); + return (float)(sum * 2); +} +#undef _ + +static float fast_half_fir(float const * input) +{ + int i = 0; + float sum = input[0] * .5f; +#define _ sum += (input[-(2*i+1)] + input[2*i+1]) * fast_half_fir_coefs[i], ++i; + _ _ _ _ _ _ +#undef _ + return (float)sum; +} + +#define IIR_FILTER _ _ _ _ _ _ _ +#define _ in1=(in1-p->y[i])*iir_coefs[i]+tmp1;tmp1=p->y[i],p->y[i]=in1;++i;\ + in0=(in0-p->y[i])*iir_coefs[i]+tmp0;tmp0=p->y[i],p->y[i]=in0;++i; + +typedef struct {float x[2], y[AL(iir_coefs)];} half_iir_t; + +static float half_iir1(half_iir_t * p, float in0, float in1) +{ + int i = 0; + float tmp0, tmp1; + tmp0 = p->x[0], p->x[0] = in0; + tmp1 = p->x[1], p->x[1] = in1; + IIR_FILTER + p->y[i] = in1 = (in1 - p->y[i]) * iir_coefs[i] + tmp1; + return in1 + in0; +} +#undef _ + +static void half_iir(half_iir_t * p, float * obuf, float const * ibuf, int olen) +{ + int i; + for (i=0; i < olen; obuf[i] = (float)half_iir1(p, ibuf[i*2], ibuf[i*2+1]),++i); +} + +static void half_phase(half_iir_t * p, float * buf, int len) +{ + float const small_normal = 1/MULT32/MULT32; /* To quash denormals on path 0.*/ + int i; + for (i = 0; i < len; buf[i] = (float)half_iir1(p, buf[i], 0), ++i); +#define _ p->y[i] += small_normal, i += 2; + i = 0, _ IIR_FILTER +#undef _ +#define _ p->y[i] -= small_normal, i += 2; + i = 0, _ IIR_FILTER +#undef _ +} + +#define coef(coef_p, interp_order, fir_len, phase_num, coef_interp_num, \ + fir_coef_num) coef_p[(fir_len) * ((interp_order) + 1) * (phase_num) + \ + ((interp_order) + 1) * (fir_coef_num) + (interp_order - coef_interp_num)] + +#define COEF(h,l,i) ((i)<0||(i)>=(l)?0:(h)[(i)>(l)/2?(l)-(i):(i)]) +static void prepare_coefs(float * coefs, int n, int phases0, int phases, + float const * coefs0, double multiplier) +{ + double k[6]; + int length0 = n * phases0, length = n * phases, K0 = iAL(k)/2 - 1, i, j, pos; + float * coefs1 = malloc(((size_t)length / 2 + 1) * sizeof(*coefs1)); + float * p = coefs1, f0, f1 = 0; + + for (j = 0; j < iAL(k); k[j] = COEF(coefs0, length0, j - K0), ++j); + for (pos = i = 0; i < length0 / 2; ++i) { + double b=(1/24.)*(k[0]+k[4]+6*k[2]-4*(k[1]+k[3])),d=.5*(k[1]+k[3])-k[2]-b; + double a=(1/120.)*(k[5]-k[2]-9*(9*b+d)+2.5*(k[3]-k[1])-2*(k[4]-k[0])); + double c=(1/12.)*(k[4]-k[0]-2*(k[3]-k[1])-60*a),e=.5*(k[3]-k[1])-a-c; + for (; pos / phases == i; pos += phases0) { + double x = (double)(pos % phases) / phases; + *p++ = (float)(k[K0] + ((((a*x + b)*x + c)*x + d)*x + e)*x); + } + for (j = 0; j < iAL(k) - 1; k[j] = k[j + 1], ++j); + k[j] = COEF(coefs0, length0, i + iAL(k) / 2 + 1); + } + if (!(length & 1)) + *p++ = (float)k[K0]; + assert(p - coefs1 == length / 2 + 1); + + for (i = 0; i < n; ++i) for (j = phases - 1; j >= 0; --j, f1 = f0) { + pos = (n - 1 - i) * phases + j; + f0 = COEF(coefs1, length, pos) * (float)multiplier; + coef(coefs, 1, n, j, 0, i) = (float)f0; + coef(coefs, 1, n, j, 1, i) = (float)(f1 - f0); + } + free(coefs1); +} + +#define _ sum += (b *x + a)*input[i], ++i; +#define a (coef(poly_fir_coefs_d, 1, POLY_FIR_LEN_D, phase, 0,i)) +#define b (coef(poly_fir_coefs_d, 1, POLY_FIR_LEN_D, phase, 1,i)) +static float poly_fir_coefs_d[POLY_FIR_LEN_D * PHASES_D * 2]; + +static float poly_fir1_d(float const * input, uint32_t frac) +{ + int i = 0, phase = (int)(frac >> (32 - PHASE_BITS_D)); + float sum = 0, x = (float)(frac << PHASE_BITS_D) * (float)(1 / MULT32); + _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ + assert(i == POLY_FIR_LEN_D); + return (float)sum; +} +#undef a +#undef b +#define a (coef(poly_fir_coefs_u, 1, POLY_FIR_LEN_U, phase, 0,i)) +#define b (coef(poly_fir_coefs_u, 1, POLY_FIR_LEN_U, phase, 1,i)) +static float poly_fir_coefs_u[POLY_FIR_LEN_U * PHASES_U * 2]; + +static float poly_fir1_u(float const * input, uint32_t frac) +{ + int i = 0, phase = (int)(frac >> (32 - PHASE_BITS_U)); + float sum = 0, x = (float)(frac << PHASE_BITS_U) * (float)(1 / MULT32); + _ _ _ _ _ _ _ _ _ _ _ _ + assert(i == POLY_FIR_LEN_U); + return (float)sum; +} +#undef a +#undef b +#undef _ + +#define ADD_TO(x,y) x.all += y.all +#define SUBTRACT_FROM(x,y) x.all -= y.all +#define FRAC(x) x.part.frac +#define INT(x) x.part.integer + +typedef struct { + union { + int64_t all; +#if HAVE_BIGENDIAN + struct {int32_t integer; uint32_t frac;} part; +#else + struct {uint32_t frac; int32_t integer;} part; +#endif + } at, step, step_step; + float const * input; + int len, stage_num; + bool is_d; /* true: downsampling at x2 rate; false: upsampling at 1x rate. */ + double step_mult; +} stream_t; + +static int poly_fir_d(stream_t * s, float * output, int olen) +{ + int i; + float const * input = s->input - POLY_FIR_LEN_D / 2 + 1; + for (i = 0; i < olen && INT(s->at) < s->len; ++i) { + output[i] = poly_fir1_d(input + INT(s->at), FRAC(s->at)); + ADD_TO(s->at, s->step); + if (!(INT(s->at) < s->len)) { + SUBTRACT_FROM(s->at, s->step); + break; + } + output[++i] = poly_fir1_d(input + INT(s->at), FRAC(s->at)); + ADD_TO(s->at, s->step); + ADD_TO(s->step, s->step_step); + } + return i; +} + +static int poly_fir_fade_d( + stream_t * s, float const * vol, int step, float * output, int olen) +{ + int i; + float const * input = s->input - POLY_FIR_LEN_D / 2 + 1; + for (i = 0; i < olen && INT(s->at) < s->len; ++i, vol += step) { + output[i] += *vol * poly_fir1_d(input + INT(s->at), FRAC(s->at)); + ADD_TO(s->at, s->step); + if (!(INT(s->at) < s->len)) { + SUBTRACT_FROM(s->at, s->step); + break; + } + output[++i] += *(vol += step) * poly_fir1_d(input + INT(s->at),FRAC(s->at)); + ADD_TO(s->at, s->step); + ADD_TO(s->step, s->step_step); + } + return i; +} + +static int poly_fir_u(stream_t * s, float * output, int olen) +{ + int i; + float const * input = s->input - POLY_FIR_LEN_U / 2 + 1; + for (i = 0; i < olen && INT(s->at) < s->len; ++i) { + output[i] = poly_fir1_u(input + INT(s->at), FRAC(s->at)); + ADD_TO(s->at, s->step); + ADD_TO(s->step, s->step_step); + } + return i; +} + +static int poly_fir_fade_u( + stream_t * s, float const * vol, int step, float * output, int olen) +{ + int i; + float const * input = s->input - POLY_FIR_LEN_U / 2 + 1; + for (i = 0; i < olen && INT(s->at) < s->len; i += 2, vol += step) { + output[i] += *vol * poly_fir1_u(input + INT(s->at), FRAC(s->at)); + ADD_TO(s->at, s->step); + ADD_TO(s->step, s->step_step); + } + return i; +} + +#define shiftr(x,by) ((by) < 0? (x) << (-(by)) : (x) >> (by)) +#define shiftl(x,by) shiftr(x,-(by)) +#define stage_occupancy(s) (fifo_occupancy(&(s)->fifo) - 4*HALF_FIR_LEN_2) +#define stage_read_p(s) ((float *)fifo_read_ptr(&(s)->fifo) + 2*HALF_FIR_LEN_2) +#define stage_preload(s) memset(fifo_reserve(&(s)->fifo, (s)->preload), \ + 0, sizeof(float) * (size_t)(s)->preload); + +typedef struct { + fifo_t fifo; + double step_mult; + int is_fast, x_fade_len, preload; +} stage_t; + +typedef struct { + int num_stages0, num_stages, flushing; + int fade_len, slew_len, xfade, stage_inc, switch_stage_num; + double new_io_ratio, default_io_ratio; + stage_t * stages; + fifo_t output_fifo; + half_iir_t halfer; + stream_t current, fadeout; /* Current/fade-in, fadeout streams. */ +} rate_t; + +static float fade_coefs[(2 << FADE_LEN_BITS) + 1]; + +static void vr_init(rate_t * p, double default_io_ratio, int num_stages, double mult) +{ + int i; + assert(num_stages >= 0); + memset(p, 0, sizeof(*p)); + + p->num_stages0 = num_stages; + p->num_stages = num_stages = max(num_stages, 1); + p->stages = (stage_t *)calloc((unsigned)num_stages + 1, sizeof(*p->stages)) + 1; + for (i = -1; i < p->num_stages; ++i) { + stage_t * s = &p->stages[i]; + fifo_create(&s->fifo, sizeof(float)); + s->step_mult = 2 * MULT32 / shiftl(2, i); + s->preload = i < 0? 0 : i == 0? 2 * HALF_FIR_LEN_2 : 3 * HALF_FIR_LEN_2 / 2; + stage_preload(s); + s->is_fast = true; + lsx_debug("%-3i preload=%i", i, s->preload); + } + fifo_create(&p->output_fifo, sizeof(float)); + p->default_io_ratio = default_io_ratio; + if (fade_coefs[0]==0) { + for (i = 0; i < iAL(fade_coefs); ++i) + fade_coefs[i] = (float)(.5 * (1 + cos(M_PI * i / (AL(fade_coefs) - 1)))); + prepare_coefs(poly_fir_coefs_u, POLY_FIR_LEN_U, PHASES0_U, PHASES_U, coefs0_u, mult); + prepare_coefs(poly_fir_coefs_d, POLY_FIR_LEN_D, PHASES0_D, PHASES_D, coefs0_d, mult *.5); + } + assert(fade_coefs[0]); +} + +static void enter_new_stage(rate_t * p, int occupancy0) +{ + p->current.len = shiftr(occupancy0, p->current.stage_num); + p->current.input = stage_read_p(&p->stages[p->current.stage_num]); + + p->current.step_mult = p->stages[p->current.stage_num].step_mult; + p->current.is_d = p->current.stage_num >= 0; + if (p->current.is_d) + p->current.step_mult *= .5; +} + +static void set_step(stream_t * p, double io_ratio) +{ + p->step.all = (int64_t)(io_ratio * p->step_mult + .5); +} + +static bool set_step_step(stream_t * p, double io_ratio, int slew_len) +{ + int64_t dif; + int difi; + stream_t tmp = *p; + set_step(&tmp, io_ratio); + dif = tmp.step.all - p->step.all; + dif = dif < 0? dif - (slew_len >> 1) : dif + (slew_len >> 1); + difi = (int)dif; /* Try to avoid int64_t div. */ + p->step_step.all = difi == dif? difi / slew_len : dif / slew_len; + return p->step_step.all != 0; +} + +static void vr_set_io_ratio(rate_t * p, double io_ratio, size_t slew_len) +{ + assert(io_ratio > 0); + if (slew_len) { + if (!set_step_step(&p->current, io_ratio, p->slew_len = (int)slew_len)) + p->slew_len = 0, p->new_io_ratio = 0, p->fadeout.step_step.all = 0; + else { + p->new_io_ratio = io_ratio; + if (p->fade_len) + set_step_step(&p->fadeout, io_ratio, p->slew_len); + } + } + else { + if (p->default_io_ratio!=0) { /* Then this is the first call to this fn. */ + int octave = (int)floor(log(io_ratio) / M_LN2); + p->current.stage_num = octave < 0? -1 : min(octave, p->num_stages0-1); + enter_new_stage(p, 0); + } + else if (p->fade_len) + set_step(&p->fadeout, io_ratio); + set_step(&p->current, io_ratio); + if (p->default_io_ratio!=0) FRAC(p->current.at) = FRAC(p->current.step) >> 1; + p->default_io_ratio = 0; + } +} + +static bool do_input_stage(rate_t * p, int stage_num, int sign, int min_stage_num) +{ + int i = 0; + float * dest; + stage_t * s = &p->stages[stage_num]; + stage_t * s1 = &p->stages[stage_num - sign]; + float const * src = (float *)fifo_read_ptr(&s1->fifo) + HALF_FIR_LEN_2; + int len = shiftr(fifo_occupancy(&s1->fifo) - HALF_FIR_LEN_2 * 2, sign); + int already_done = fifo_occupancy(&s->fifo) - s->preload; + if ((len -= already_done) <= 0) + return false; + src += shiftl(already_done, sign); + + dest = fifo_reserve(&s->fifo, len); + if (stage_num < 0) for (; i < len; ++src) + dest[i++] = double_fir0(src), dest[i++] = double_fir1(src); + else { + bool should_be_fast = p->stage_inc; + if (!s->x_fade_len && stage_num == p->switch_stage_num) { + p->switch_stage_num = 0; + if (s->is_fast != should_be_fast) { + s->x_fade_len = 1 << FADE_LEN_BITS, s->is_fast = should_be_fast, ++p->xfade; + lsx_debug("xfade level %i, inc?=%i", stage_num, p->stage_inc); + } + } + if (s->x_fade_len) { + float const * vol1 = fade_coefs + (s->x_fade_len << 1); + float const * vol2 = fade_coefs + (((1 << FADE_LEN_BITS) - s->x_fade_len) << 1); + int n = min(len, s->x_fade_len); + /*lsx_debug("xfade level %i, inc?=%i len=%i n=%i", stage_num, p->stage_inc, s->x_fade_len, n);*/ + if (should_be_fast) + for (; i < n; vol2 += 2, vol1 -= 2, src += 2) + dest[i++] = *vol1 * fast_half_fir(src) + *vol2 * half_fir(src); + else for (; i < n; vol2 += 2, vol1 -= 2, src += 2) + dest[i++] = *vol2 * fast_half_fir(src) + *vol1 * half_fir(src); + s->x_fade_len -= n; + p->xfade -= !s->x_fade_len; + } + if (stage_num < min_stage_num) + for (; i < len; dest[i++] = fast_half_fir(src), src += 2); + else for (; i < len; dest[i++] = half_fir(src), src += 2); + } + if (p->flushing > 0) + stage_preload(s); + return true; +} + +static int vr_process(rate_t * p, int olen0) +{ + assert(p->num_stages > 0); + if (p->default_io_ratio!=0) + vr_set_io_ratio(p, p->default_io_ratio, 0); + { + float * output = fifo_reserve(&p->output_fifo, olen0); + int j, odone0 = 0, min_stage_num = p->current.stage_num; + int occupancy0, max_stage_num = min_stage_num; + if (p->fade_len) { + min_stage_num = min(min_stage_num, p->fadeout.stage_num); + max_stage_num = max(max_stage_num, p->fadeout.stage_num); + } + + for (j = min(min_stage_num, 0); j <= max_stage_num; ++j) + if (j && !do_input_stage(p, j, j < 0? -1 : 1, min_stage_num)) + break; + if (p->flushing > 0) + p->flushing = -1; + + occupancy0 = shiftl(max(0,stage_occupancy(&p->stages[max_stage_num])), max_stage_num); + p->current.len = shiftr(occupancy0, p->current.stage_num); + p->current.input = stage_read_p(&p->stages[p->current.stage_num]); + if (p->fade_len) { + p->fadeout.len = shiftr(occupancy0, p->fadeout.stage_num); + p->fadeout.input = stage_read_p(&p->stages[p->fadeout.stage_num]); + } + + while (odone0 < olen0) { + int odone, odone2, olen = olen0 - odone0, stage_dif = 0, shift; + float buf[64 << 1]; + + olen = min(olen, (int)(AL(buf) >> 1)); + if (p->slew_len) + olen = min(olen, p->slew_len); + else if (p->new_io_ratio!=0) { + set_step(&p->current, p->new_io_ratio); + set_step(&p->fadeout, p->new_io_ratio); + p->fadeout.step_step.all = p->current.step_step.all = 0; + p->new_io_ratio = 0; + } + if (!p->flushing && !p->fade_len && !p->xfade) { + if (p->current.is_d) { + if (INT(p->current.step) && FRAC(p->current.step)) + stage_dif = 1, ++max_stage_num; + else if (!INT(p->current.step) && FRAC(p->current.step) < (1u << 31)) + stage_dif = -1, --min_stage_num; + } else if (INT(p->current.step) > 1 && FRAC(p->current.step)) + stage_dif = 1, ++max_stage_num; + } + if (stage_dif) { + int n = p->current.stage_num + stage_dif; + if (n >= p->num_stages) + --max_stage_num; + else { + p->stage_inc = stage_dif > 0; + p->fadeout = p->current; + p->current.stage_num += stage_dif; + if (!p->stage_inc) + p->switch_stage_num = p->current.stage_num; + if ((p->current.stage_num < 0 && stage_dif < 0) || + (p->current.stage_num > 0 && stage_dif > 0)) { + stage_t * s = &p->stages[p->current.stage_num]; + fifo_clear(&s->fifo); + stage_preload(s); + s->is_fast = false; + do_input_stage(p, p->current.stage_num, stage_dif, p->current.stage_num); + } + if (p->current.stage_num > 0 && stage_dif < 0) { + int idone = INT(p->current.at); + stage_t * s = &p->stages[p->current.stage_num]; + fifo_trim_to(&s->fifo, 2 * HALF_FIR_LEN_2 + idone + (POLY_FIR_LEN_D >> 1)); + do_input_stage(p, p->current.stage_num, 1, p->current.stage_num); + } + enter_new_stage(p, occupancy0); + shift = -stage_dif; +#define lshift(x,by) (x)=(by)>0?(x)<<(by):(x)>>-(by) + lshift(p->current.at.all, shift); + shift += p->fadeout.is_d - p->current.is_d; + lshift(p->current.step.all, shift); + lshift(p->current.step_step.all, shift); + p->fade_len = AL(fade_coefs) - 1; + lsx_debug("switch from stage %i to %i, x2 from %i to %i", p->fadeout.stage_num, p->current.stage_num, p->fadeout.is_d, p->current.is_d); + } + } + + if (p->fade_len) { + float const * vol1 = fade_coefs + p->fade_len; + float const * vol2 = fade_coefs + (iAL(fade_coefs) - 1 - p->fade_len); + int olen2 = (olen = min(olen, p->fade_len >> 1)) << 1; + + /* x2 is more fine-grained so may fail to produce a pair of samples + * where x1 would not (the x1 second sample is a zero so is always + * available). So do x2 first, then feed odone to the second one. */ + memset(buf, 0, sizeof(*buf) * (size_t)olen2); + if (p->current.is_d && p->fadeout.is_d) { + odone = poly_fir_fade_d(&p->current, vol1,-1, buf, olen2); + odone2 = poly_fir_fade_d(&p->fadeout, vol2, 1, buf, odone); + } else if (p->current.is_d) { + odone = poly_fir_fade_d(&p->current, vol1,-1, buf, olen2); + odone2 = poly_fir_fade_u(&p->fadeout, vol2, 2, buf, odone); + } else { + assert(p->fadeout.is_d); + odone = poly_fir_fade_d(&p->fadeout, vol2, 1, buf, olen2); + odone2 = poly_fir_fade_u(&p->current, vol1,-2, buf, odone); + } + assert(odone == odone2); + (void)odone2; + p->fade_len -= odone; + if (!p->fade_len) { + if (p->stage_inc) + p->switch_stage_num = min_stage_num++; + else + --max_stage_num; + } + half_iir(&p->halfer, &output[odone0], buf, odone >>= 1); + } + else if (p->current.is_d) { + odone = poly_fir_d(&p->current, buf, olen << 1) >> 1; + half_iir(&p->halfer, &output[odone0], buf, odone); + } + else { + odone = poly_fir_u(&p->current, &output[odone0], olen); + if (p->num_stages0) + half_phase(&p->halfer, &output[odone0], odone); + } + odone0 += odone; + if (p->slew_len) + p->slew_len -= odone; + if (odone != olen) + break; /* Need more input. */ + } { + int from = max(0, max_stage_num), to = min(0, min_stage_num); + int i, idone = shiftr(INT(p->current.at), from - p->current.stage_num); + INT(p->current.at) -= shiftl(idone, from - p->current.stage_num); + if (p->fade_len) + INT(p->fadeout.at) -= shiftl(idone, from - p->fadeout.stage_num); + for (i = from; i >= to; --i, idone <<= 1) + fifo_read(&p->stages[i].fifo, idone, NULL); + } + fifo_trim_by(&p->output_fifo, olen0 - odone0); + return odone0; + } +} + +static float * vr_input(rate_t * p, float const * input, size_t n) +{ + return fifo_write(&p->stages[0].fifo, (int)n, input); +} + +static float const * vr_output(rate_t * p, float * output, size_t * n) +{ + fifo_t * fifo = &p->output_fifo; + if (1 || !p->num_stages0) + return fifo_read(fifo, (int)(*n = min(*n, (size_t)fifo_occupancy(fifo))), output); + else { /* Ignore this complication for now. */ + int const IIR_DELAY = 2; + float * ptr = fifo_read_ptr(fifo); + int olen = min((int)*n, max(0, fifo_occupancy(fifo) - IIR_DELAY)); + *n = (size_t)olen; + if (output) + memcpy(output, ptr + IIR_DELAY, *n * sizeof(*output)); + fifo_read(fifo, olen, NULL); + return ptr + IIR_DELAY; + } +} + +static void vr_flush(rate_t * p) +{ + if (!p->flushing) { + stage_preload(&p->stages[0]); + ++p->flushing; + } +} + +static void vr_close(rate_t * p) +{ + int i; + + fifo_delete(&p->output_fifo); + for (i = -1; i < p->num_stages; ++i) { + stage_t * s = &p->stages[i]; + fifo_delete(&s->fifo); + } + free(p->stages - 1); +} + +static double vr_delay(rate_t * p) +{ + return 100; /* TODO */ + (void)p; +} + +static void vr_sizes(size_t * shared, size_t * channel) +{ + *shared = 0; + *channel = sizeof(rate_t); +} + +static char const * vr_create(void * channel, void * shared,double max_io_ratio, + void * q_spec, void * r_spec, double scale) +{ + double x = max_io_ratio; + int n; + for (n = 0; x > 1; x *= .5, ++n); + vr_init(channel, max_io_ratio, n, scale); + return 0; + (void)shared, (void)q_spec, (void)r_spec; +} + +static char const * vr_id(void) +{ + return "vr32"; +} + +typedef void (* fn_t)(void); +fn_t _soxr_vr32_cb[] = { + (fn_t)vr_input, + (fn_t)vr_process, + (fn_t)vr_output, + (fn_t)vr_flush, + (fn_t)vr_close, + (fn_t)vr_delay, + (fn_t)vr_sizes, + (fn_t)vr_create, + (fn_t)vr_set_io_ratio, + (fn_t)vr_id, +}; From af8aef4929dc855b44db7779cff54117d9c00e1a Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?The=CC=81o=20Monnom?= Date: Sat, 21 Sep 2024 20:11:31 -0700 Subject: [PATCH 2/2] nampa --- .nanparc | 2 +- soxr-sys/.nanparc | 2 ++ 2 files changed, 3 insertions(+), 1 deletion(-) create mode 100644 soxr-sys/.nanparc diff --git a/.nanparc b/.nanparc index d7523fefa..748a8a360 100644 --- a/.nanparc +++ b/.nanparc @@ -1 +1 @@ -packages livekit livekit-ffi livekit-protocol livekit-runtime livekit-api libwebrtc webrtc-sys webrtc-sys/build +packages livekit livekit-ffi livekit-protocol livekit-runtime livekit-api libwebrtc webrtc-sys webrtc-sys/build soxr-sys diff --git a/soxr-sys/.nanparc b/soxr-sys/.nanparc new file mode 100644 index 000000000..2124a88b3 --- /dev/null +++ b/soxr-sys/.nanparc @@ -0,0 +1,2 @@ +version 0.1.0 +language rust