diff options
Diffstat (limited to 'src/ffts_real_nd.c')
-rw-r--r-- | src/ffts_real_nd.c | 336 |
1 files changed, 336 insertions, 0 deletions
diff --git a/src/ffts_real_nd.c b/src/ffts_real_nd.c new file mode 100644 index 0000000..05bcc9c --- /dev/null +++ b/src/ffts_real_nd.c @@ -0,0 +1,336 @@ +/* + + This file is part of FFTS -- The Fastest Fourier Transform in the South + + Copyright (c) 2012, Anthony M. Blake <amb@anthonix.com> + Copyright (c) 2012, The University of Waikato + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + * Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + * Neither the name of the organization nor the + names of its contributors may be used to endorse or promote products + derived from this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND + ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + DISCLAIMED. IN NO EVENT SHALL ANTHONY M. BLAKE BE LIABLE FOR ANY + DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +*/ + +#include "ffts_real_nd.h" +#include "ffts_real.h" + +#ifdef __ARM_NEON__ +#include "neon.h" +#endif + +#ifdef HAVE_NEON +#include <arm_neon.h> +#endif + +#ifdef HAVE_SSE +#include <xmmintrin.h> +#endif + +#include <stdio.h> + +static void ffts_free_nd_real(ffts_plan_t *p) +{ + if (p->plans) { + int i; + + for (i = 0; i < p->rank; i++) { + ffts_plan_t *plan = p->plans[i]; + + if (plan) { + int j; + + for (j = i + 1; j < p->rank; j++) { + if (plan == p->plans[j]) { + p->plans[j] = NULL; + } + } + + ffts_free(plan); + } + } + + free(p->plans); + } + + if (p->transpose_buf) { + ffts_aligned_free(p->transpose_buf); + } + + if (p->buf) { + ffts_aligned_free(p->buf); + } + + if (p->Ns) { + free(p->Ns); + } + + if (p->Ms) { + free(p->Ms); + } + + free(p); +} + +static void ffts_scalar_transpose(uint64_t *src, uint64_t *dst, int w, int h, uint64_t *buf) +{ + const int bw = 1; + const int bh = 8; + int i = 0, j = 0; + + for (; i <= h - bh; i += bh) { + for (j = 0; j <= w - bw; j += bw) { + uint64_t const *ib = &src[w*i + j]; + uint64_t *ob = &dst[h*j + i]; + + uint64_t s_0_0 = ib[0*w + 0]; + uint64_t s_1_0 = ib[1*w + 0]; + uint64_t s_2_0 = ib[2*w + 0]; + uint64_t s_3_0 = ib[3*w + 0]; + uint64_t s_4_0 = ib[4*w + 0]; + uint64_t s_5_0 = ib[5*w + 0]; + uint64_t s_6_0 = ib[6*w + 0]; + uint64_t s_7_0 = ib[7*w + 0]; + + ob[0*h + 0] = s_0_0; + ob[0*h + 1] = s_1_0; + ob[0*h + 2] = s_2_0; + ob[0*h + 3] = s_3_0; + ob[0*h + 4] = s_4_0; + ob[0*h + 5] = s_5_0; + ob[0*h + 6] = s_6_0; + ob[0*h + 7] = s_7_0; + } + } + + if (i < h) { + int i1; + + for (i1 = 0; i1 < w; i1++) { + for (j = i; j < h; j++) { + dst[i1*h + j] = src[j*w + i1]; + } + } + } + + if (j < w) { + int j1; + + for (i = j; i < w; i++) { + for (j1 = 0; j1 < h; j1++) { + dst[i*h + j1] = src[j1*w + i]; + } + } + } +} + +static void ffts_execute_nd_real(ffts_plan_t *p, const void *in, void *out) +{ + const size_t Ms0 = p->Ms[0]; + const size_t Ns0 = p->Ns[0]; + + uint32_t *din = (uint32_t*) in; + uint64_t *buf = p->buf; + uint64_t *dout = (uint64_t*) out; + uint64_t *transpose_buf = (uint64_t*) p->transpose_buf; + + ffts_plan_t *plan; + size_t i, j; + + plan = p->plans[0]; + for (i = 0; i < Ns0; i++) { + plan->transform(plan, din + (i * Ms0), buf + (i * (Ms0 / 2 + 1))); + } + + ffts_scalar_transpose(buf, dout, Ms0 / 2 + 1, Ns0, transpose_buf); + + for (i = 1; i < p->rank; i++) { + const size_t Ms = p->Ms[i]; + const size_t Ns = p->Ns[i]; + + plan = p->plans[i]; + + for (j = 0; j < Ns; j++) { + plan->transform(plan, dout + (j * Ms), buf + (j * Ms)); + } + + ffts_scalar_transpose(buf, dout, Ms, Ns, transpose_buf); + } +} + +static void ffts_execute_nd_real_inv(ffts_plan_t *p, const void *in, void *out) +{ + const size_t Ms0 = p->Ms[0]; + const size_t Ms1 = p->Ms[1]; + const size_t Ns0 = p->Ns[0]; + const size_t Ns1 = p->Ns[1]; + + uint64_t *din = (uint64_t*) in; + uint64_t *buf = p->buf; + uint64_t *buf2; + uint64_t *transpose_buf = (uint64_t*) p->transpose_buf; + float *doutr = (float*) out; + + ffts_plan_t *plan; + size_t vol; + + size_t i, j; + + vol = p->Ns[0]; + for (i = 1; i < p->rank; i++) { + vol *= p->Ns[i]; + } + + buf2 = buf + vol; + + ffts_scalar_transpose(din, buf, Ms0, Ns0, transpose_buf); + + plan = p->plans[0]; + for (i = 0; i < Ms0; i++) { + plan->transform(plan, buf + (i * Ns0), buf2 + (i * Ns0)); + } + + ffts_scalar_transpose(buf2, buf, Ns0, Ms0, transpose_buf); + + plan = p->plans[1]; + for (j = 0; j < Ms1; j++) { + plan->transform(plan, buf + (j * Ms0), &doutr[j * Ns1]); + } +} + +ffts_plan_t *ffts_init_nd_real(int rank, size_t *Ns, int sign) +{ + int i; + size_t vol = 1; + size_t bufsize; + ffts_plan_t *p; + + p = (ffts_plan_t*) calloc(1, sizeof(*p)); + if (!p) { + return NULL; + } + + if (sign < 0) { + p->transform = &ffts_execute_nd_real; + } else { + p->transform = &ffts_execute_nd_real_inv; + } + + p->destroy = &ffts_free_nd_real; + p->rank = rank; + + p->Ms = (size_t*) malloc(rank * sizeof(*p->Ms)); + if (!p->Ms) { + goto cleanup; + } + + p->Ns = (size_t*) malloc(rank * sizeof(*p->Ns)); + if (!p->Ns) { + goto cleanup; + } + + for (i = 0; i < rank; i++) { + p->Ns[i] = Ns[i]; + vol *= Ns[i]; + } + + /* there is probably a prettier way of doing this, but it works.. */ + if (sign < 0) { + bufsize = 2 * vol; + } else { + bufsize = 2 * (Ns[0] * ((vol / Ns[0]) / 2 + 1) + vol); + } + + p->buf = ffts_aligned_malloc(bufsize * sizeof(float)); + if (!p->buf) { + goto cleanup; + } + + p->transpose_buf = ffts_aligned_malloc(2 * 8 * 8 * sizeof(float)); + if (!p->transpose_buf) { + goto cleanup; + } + + p->plans = (ffts_plan_t**) calloc(rank, sizeof(*p->plans)); + if (!p->plans) { + goto cleanup; + } + + for (i = 0; i < rank; i++) { + int k; + + p->Ms[i] = vol / p->Ns[i]; + + if (sign < 0) { + if (!i) { + p->plans[i] = ffts_init_1d_real(p->Ms[i], sign); + } else { + for (k = 1; k < i; k++) { + if (p->Ms[k] == p->Ms[i]) { + p->plans[i] = p->plans[k]; + break; + } + } + + if (!p->plans[i]) { + p->plans[i] = ffts_init_1d(p->Ms[i], sign); + p->Ns[i] = p->Ns[i] / 2 + 1; + } + } + } else { + if (i == rank - 1) { + p->plans[i] = ffts_init_1d_real(p->Ns[i], sign); + } else { + for (k = 0; k < i; k++) { + if (p->Ns[k] == p->Ns[i]) { + p->plans[i] = p->plans[k]; + break; + } + } + + if (!p->plans[i]) { + p->plans[i] = ffts_init_1d(p->Ns[i], sign); + p->Ms[i] = p->Ms[i] / 2 + 1; + } + } + } + + if (!p->plans[i]) { + goto cleanup; + } + } + + return p; + +cleanup: + ffts_free_nd_real(p); + return NULL; +} + +ffts_plan_t *ffts_init_2d_real(size_t N1, size_t N2, int sign) +{ + size_t Ns[2]; + + Ns[0] = N1; + Ns[1] = N2; + return ffts_init_nd_real(2, Ns, sign); +} |