cannam@154: /* Copyright (c) 2015 Xiph.Org Foundation cannam@154: Written by Viswanath Puttagunta */ cannam@154: /** cannam@154: @file celt_fft_ne10.c cannam@154: @brief ARM Neon optimizations for fft using NE10 library cannam@154: */ cannam@154: cannam@154: /* cannam@154: Redistribution and use in source and binary forms, with or without cannam@154: modification, are permitted provided that the following conditions cannam@154: are met: cannam@154: cannam@154: - Redistributions of source code must retain the above copyright cannam@154: notice, this list of conditions and the following disclaimer. cannam@154: cannam@154: - Redistributions in binary form must reproduce the above copyright cannam@154: notice, this list of conditions and the following disclaimer in the cannam@154: documentation and/or other materials provided with the distribution. cannam@154: cannam@154: THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS cannam@154: ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT cannam@154: LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR cannam@154: A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER cannam@154: OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, cannam@154: EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, cannam@154: PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR cannam@154: PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF cannam@154: LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING cannam@154: NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS cannam@154: SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. cannam@154: */ cannam@154: cannam@154: #ifndef SKIP_CONFIG_H cannam@154: #ifdef HAVE_CONFIG_H cannam@154: #include "config.h" cannam@154: #endif cannam@154: #endif cannam@154: cannam@154: #include cannam@154: #include "os_support.h" cannam@154: #include "kiss_fft.h" cannam@154: #include "stack_alloc.h" cannam@154: cannam@154: #if !defined(FIXED_POINT) cannam@154: # define NE10_FFT_ALLOC_C2C_TYPE_NEON ne10_fft_alloc_c2c_float32_neon cannam@154: # define NE10_FFT_CFG_TYPE_T ne10_fft_cfg_float32_t cannam@154: # define NE10_FFT_STATE_TYPE_T ne10_fft_state_float32_t cannam@154: # define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_float32 cannam@154: # define NE10_FFT_CPX_TYPE_T ne10_fft_cpx_float32_t cannam@154: # define NE10_FFT_C2C_1D_TYPE_NEON ne10_fft_c2c_1d_float32_neon cannam@154: #else cannam@154: # define NE10_FFT_ALLOC_C2C_TYPE_NEON(nfft) ne10_fft_alloc_c2c_int32_neon(nfft) cannam@154: # define NE10_FFT_CFG_TYPE_T ne10_fft_cfg_int32_t cannam@154: # define NE10_FFT_STATE_TYPE_T ne10_fft_state_int32_t cannam@154: # define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_int32 cannam@154: # define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_int32 cannam@154: # define NE10_FFT_CPX_TYPE_T ne10_fft_cpx_int32_t cannam@154: # define NE10_FFT_C2C_1D_TYPE_NEON ne10_fft_c2c_1d_int32_neon cannam@154: #endif cannam@154: cannam@154: #if defined(CUSTOM_MODES) cannam@154: cannam@154: /* nfft lengths in NE10 that support scaled fft */ cannam@154: # define NE10_FFTSCALED_SUPPORT_MAX 4 cannam@154: static const int ne10_fft_scaled_support[NE10_FFTSCALED_SUPPORT_MAX] = { cannam@154: 480, 240, 120, 60 cannam@154: }; cannam@154: cannam@154: int opus_fft_alloc_arm_neon(kiss_fft_state *st) cannam@154: { cannam@154: int i; cannam@154: size_t memneeded = sizeof(struct arch_fft_state); cannam@154: cannam@154: st->arch_fft = (arch_fft_state *)opus_alloc(memneeded); cannam@154: if (!st->arch_fft) cannam@154: return -1; cannam@154: cannam@154: for (i = 0; i < NE10_FFTSCALED_SUPPORT_MAX; i++) { cannam@154: if(st->nfft == ne10_fft_scaled_support[i]) cannam@154: break; cannam@154: } cannam@154: if (i == NE10_FFTSCALED_SUPPORT_MAX) { cannam@154: /* This nfft length (scaled fft) is not supported in NE10 */ cannam@154: st->arch_fft->is_supported = 0; cannam@154: st->arch_fft->priv = NULL; cannam@154: } cannam@154: else { cannam@154: st->arch_fft->is_supported = 1; cannam@154: st->arch_fft->priv = (void *)NE10_FFT_ALLOC_C2C_TYPE_NEON(st->nfft); cannam@154: if (st->arch_fft->priv == NULL) { cannam@154: return -1; cannam@154: } cannam@154: } cannam@154: return 0; cannam@154: } cannam@154: cannam@154: void opus_fft_free_arm_neon(kiss_fft_state *st) cannam@154: { cannam@154: NE10_FFT_CFG_TYPE_T cfg; cannam@154: cannam@154: if (!st->arch_fft) cannam@154: return; cannam@154: cannam@154: cfg = (NE10_FFT_CFG_TYPE_T)st->arch_fft->priv; cannam@154: if (cfg) cannam@154: NE10_FFT_DESTROY_C2C_TYPE(cfg); cannam@154: opus_free(st->arch_fft); cannam@154: } cannam@154: #endif cannam@154: cannam@154: void opus_fft_neon(const kiss_fft_state *st, cannam@154: const kiss_fft_cpx *fin, cannam@154: kiss_fft_cpx *fout) cannam@154: { cannam@154: NE10_FFT_STATE_TYPE_T state; cannam@154: NE10_FFT_CFG_TYPE_T cfg = &state; cannam@154: VARDECL(NE10_FFT_CPX_TYPE_T, buffer); cannam@154: SAVE_STACK; cannam@154: ALLOC(buffer, st->nfft, NE10_FFT_CPX_TYPE_T); cannam@154: cannam@154: if (!st->arch_fft->is_supported) { cannam@154: /* This nfft length (scaled fft) not supported in NE10 */ cannam@154: opus_fft_c(st, fin, fout); cannam@154: } cannam@154: else { cannam@154: memcpy((void *)cfg, st->arch_fft->priv, sizeof(NE10_FFT_STATE_TYPE_T)); cannam@154: state.buffer = (NE10_FFT_CPX_TYPE_T *)&buffer[0]; cannam@154: #if !defined(FIXED_POINT) cannam@154: state.is_forward_scaled = 1; cannam@154: cannam@154: NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout, cannam@154: (NE10_FFT_CPX_TYPE_T *)fin, cannam@154: cfg, 0); cannam@154: #else cannam@154: NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout, cannam@154: (NE10_FFT_CPX_TYPE_T *)fin, cannam@154: cfg, 0, 1); cannam@154: #endif cannam@154: } cannam@154: RESTORE_STACK; cannam@154: } cannam@154: cannam@154: void opus_ifft_neon(const kiss_fft_state *st, cannam@154: const kiss_fft_cpx *fin, cannam@154: kiss_fft_cpx *fout) cannam@154: { cannam@154: NE10_FFT_STATE_TYPE_T state; cannam@154: NE10_FFT_CFG_TYPE_T cfg = &state; cannam@154: VARDECL(NE10_FFT_CPX_TYPE_T, buffer); cannam@154: SAVE_STACK; cannam@154: ALLOC(buffer, st->nfft, NE10_FFT_CPX_TYPE_T); cannam@154: cannam@154: if (!st->arch_fft->is_supported) { cannam@154: /* This nfft length (scaled fft) not supported in NE10 */ cannam@154: opus_ifft_c(st, fin, fout); cannam@154: } cannam@154: else { cannam@154: memcpy((void *)cfg, st->arch_fft->priv, sizeof(NE10_FFT_STATE_TYPE_T)); cannam@154: state.buffer = (NE10_FFT_CPX_TYPE_T *)&buffer[0]; cannam@154: #if !defined(FIXED_POINT) cannam@154: state.is_backward_scaled = 0; cannam@154: cannam@154: NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout, cannam@154: (NE10_FFT_CPX_TYPE_T *)fin, cannam@154: cfg, 1); cannam@154: #else cannam@154: NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout, cannam@154: (NE10_FFT_CPX_TYPE_T *)fin, cannam@154: cfg, 1, 0); cannam@154: #endif cannam@154: } cannam@154: RESTORE_STACK; cannam@154: }