Chris@69: /* Copyright (c) 2015 Xiph.Org Foundation
Chris@69:    Written by Viswanath Puttagunta */
Chris@69: /**
Chris@69:    @file celt_fft_ne10.c
Chris@69:    @brief ARM Neon optimizations for fft using NE10 library
Chris@69:  */
Chris@69: 
Chris@69: /*
Chris@69:    Redistribution and use in source and binary forms, with or without
Chris@69:    modification, are permitted provided that the following conditions
Chris@69:    are met:
Chris@69: 
Chris@69:    - Redistributions of source code must retain the above copyright
Chris@69:    notice, this list of conditions and the following disclaimer.
Chris@69: 
Chris@69:    - Redistributions in binary form must reproduce the above copyright
Chris@69:    notice, this list of conditions and the following disclaimer in the
Chris@69:    documentation and/or other materials provided with the distribution.
Chris@69: 
Chris@69:    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
Chris@69:    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
Chris@69:    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
Chris@69:    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
Chris@69:    OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
Chris@69:    EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
Chris@69:    PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
Chris@69:    PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
Chris@69:    LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
Chris@69:    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
Chris@69:    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
Chris@69: */
Chris@69: 
Chris@69: #ifndef SKIP_CONFIG_H
Chris@69: #ifdef HAVE_CONFIG_H
Chris@69: #include "config.h"
Chris@69: #endif
Chris@69: #endif
Chris@69: 
Chris@69: #include <NE10_dsp.h>
Chris@69: #include "os_support.h"
Chris@69: #include "kiss_fft.h"
Chris@69: #include "stack_alloc.h"
Chris@69: 
Chris@69: #if !defined(FIXED_POINT)
Chris@69: # define NE10_FFT_ALLOC_C2C_TYPE_NEON ne10_fft_alloc_c2c_float32_neon
Chris@69: # define NE10_FFT_CFG_TYPE_T ne10_fft_cfg_float32_t
Chris@69: # define NE10_FFT_STATE_TYPE_T ne10_fft_state_float32_t
Chris@69: # define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_float32
Chris@69: # define NE10_FFT_CPX_TYPE_T ne10_fft_cpx_float32_t
Chris@69: # define NE10_FFT_C2C_1D_TYPE_NEON ne10_fft_c2c_1d_float32_neon
Chris@69: #else
Chris@69: # define NE10_FFT_ALLOC_C2C_TYPE_NEON(nfft) ne10_fft_alloc_c2c_int32_neon(nfft)
Chris@69: # define NE10_FFT_CFG_TYPE_T ne10_fft_cfg_int32_t
Chris@69: # define NE10_FFT_STATE_TYPE_T ne10_fft_state_int32_t
Chris@69: # define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_int32
Chris@69: # define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_int32
Chris@69: # define NE10_FFT_CPX_TYPE_T ne10_fft_cpx_int32_t
Chris@69: # define NE10_FFT_C2C_1D_TYPE_NEON ne10_fft_c2c_1d_int32_neon
Chris@69: #endif
Chris@69: 
Chris@69: #if defined(CUSTOM_MODES)
Chris@69: 
Chris@69: /* nfft lengths in NE10 that support scaled fft */
Chris@69: # define NE10_FFTSCALED_SUPPORT_MAX 4
Chris@69: static const int ne10_fft_scaled_support[NE10_FFTSCALED_SUPPORT_MAX] = {
Chris@69:    480, 240, 120, 60
Chris@69: };
Chris@69: 
Chris@69: int opus_fft_alloc_arm_neon(kiss_fft_state *st)
Chris@69: {
Chris@69:    int i;
Chris@69:    size_t memneeded = sizeof(struct arch_fft_state);
Chris@69: 
Chris@69:    st->arch_fft = (arch_fft_state *)opus_alloc(memneeded);
Chris@69:    if (!st->arch_fft)
Chris@69:       return -1;
Chris@69: 
Chris@69:    for (i = 0; i < NE10_FFTSCALED_SUPPORT_MAX; i++) {
Chris@69:       if(st->nfft == ne10_fft_scaled_support[i])
Chris@69:          break;
Chris@69:    }
Chris@69:    if (i == NE10_FFTSCALED_SUPPORT_MAX) {
Chris@69:       /* This nfft length (scaled fft) is not supported in NE10 */
Chris@69:       st->arch_fft->is_supported = 0;
Chris@69:       st->arch_fft->priv = NULL;
Chris@69:    }
Chris@69:    else {
Chris@69:       st->arch_fft->is_supported = 1;
Chris@69:       st->arch_fft->priv = (void *)NE10_FFT_ALLOC_C2C_TYPE_NEON(st->nfft);
Chris@69:       if (st->arch_fft->priv == NULL) {
Chris@69:          return -1;
Chris@69:       }
Chris@69:    }
Chris@69:    return 0;
Chris@69: }
Chris@69: 
Chris@69: void opus_fft_free_arm_neon(kiss_fft_state *st)
Chris@69: {
Chris@69:    NE10_FFT_CFG_TYPE_T cfg;
Chris@69: 
Chris@69:    if (!st->arch_fft)
Chris@69:       return;
Chris@69: 
Chris@69:    cfg = (NE10_FFT_CFG_TYPE_T)st->arch_fft->priv;
Chris@69:    if (cfg)
Chris@69:       NE10_FFT_DESTROY_C2C_TYPE(cfg);
Chris@69:    opus_free(st->arch_fft);
Chris@69: }
Chris@69: #endif
Chris@69: 
Chris@69: void opus_fft_neon(const kiss_fft_state *st,
Chris@69:                    const kiss_fft_cpx *fin,
Chris@69:                    kiss_fft_cpx *fout)
Chris@69: {
Chris@69:    NE10_FFT_STATE_TYPE_T state;
Chris@69:    NE10_FFT_CFG_TYPE_T cfg = &state;
Chris@69:    VARDECL(NE10_FFT_CPX_TYPE_T, buffer);
Chris@69:    SAVE_STACK;
Chris@69:    ALLOC(buffer, st->nfft, NE10_FFT_CPX_TYPE_T);
Chris@69: 
Chris@69:    if (!st->arch_fft->is_supported) {
Chris@69:       /* This nfft length (scaled fft) not supported in NE10 */
Chris@69:       opus_fft_c(st, fin, fout);
Chris@69:    }
Chris@69:    else {
Chris@69:       memcpy((void *)cfg, st->arch_fft->priv, sizeof(NE10_FFT_STATE_TYPE_T));
Chris@69:       state.buffer = (NE10_FFT_CPX_TYPE_T *)&buffer[0];
Chris@69: #if !defined(FIXED_POINT)
Chris@69:       state.is_forward_scaled = 1;
Chris@69: 
Chris@69:       NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout,
Chris@69:                                 (NE10_FFT_CPX_TYPE_T *)fin,
Chris@69:                                 cfg, 0);
Chris@69: #else
Chris@69:       NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout,
Chris@69:                                 (NE10_FFT_CPX_TYPE_T *)fin,
Chris@69:                                 cfg, 0, 1);
Chris@69: #endif
Chris@69:    }
Chris@69:    RESTORE_STACK;
Chris@69: }
Chris@69: 
Chris@69: void opus_ifft_neon(const kiss_fft_state *st,
Chris@69:                     const kiss_fft_cpx *fin,
Chris@69:                     kiss_fft_cpx *fout)
Chris@69: {
Chris@69:    NE10_FFT_STATE_TYPE_T state;
Chris@69:    NE10_FFT_CFG_TYPE_T cfg = &state;
Chris@69:    VARDECL(NE10_FFT_CPX_TYPE_T, buffer);
Chris@69:    SAVE_STACK;
Chris@69:    ALLOC(buffer, st->nfft, NE10_FFT_CPX_TYPE_T);
Chris@69: 
Chris@69:    if (!st->arch_fft->is_supported) {
Chris@69:       /* This nfft length (scaled fft) not supported in NE10 */
Chris@69:       opus_ifft_c(st, fin, fout);
Chris@69:    }
Chris@69:    else {
Chris@69:       memcpy((void *)cfg, st->arch_fft->priv, sizeof(NE10_FFT_STATE_TYPE_T));
Chris@69:       state.buffer = (NE10_FFT_CPX_TYPE_T *)&buffer[0];
Chris@69: #if !defined(FIXED_POINT)
Chris@69:       state.is_backward_scaled = 0;
Chris@69: 
Chris@69:       NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout,
Chris@69:                                 (NE10_FFT_CPX_TYPE_T *)fin,
Chris@69:                                 cfg, 1);
Chris@69: #else
Chris@69:       NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout,
Chris@69:                                 (NE10_FFT_CPX_TYPE_T *)fin,
Chris@69:                                 cfg, 1, 0);
Chris@69: #endif
Chris@69:    }
Chris@69:    RESTORE_STACK;
Chris@69: }