summaryrefslogtreecommitdiffstats
path: root/src/ffts_real_nd.c
diff options
context:
space:
mode:
authorHaruki Hasegawa <h6a.h4i.0@gmail.com>2016-05-05 13:24:55 +0900
committerHaruki Hasegawa <h6a.h4i.0@gmail.com>2016-05-05 13:24:55 +0900
commitda3213cf045e0c7c4971d8b44272d1d86d689ceb (patch)
tree33e17e0a166f03307ebf11e8ab2891ae1ab90f61 /src/ffts_real_nd.c
parentfa1780c68593762b1e4bdbc46d83912db3eba27a (diff)
parent944d14c9151f6b20145de0cdae38e366e73c9432 (diff)
downloadffts-da3213cf045e0c7c4971d8b44272d1d86d689ceb.zip
ffts-da3213cf045e0c7c4971d8b44272d1d86d689ceb.tar.gz
Merge remote-tracking branch 'linkotec/master'
Diffstat (limited to 'src/ffts_real_nd.c')
-rw-r--r--src/ffts_real_nd.c410
1 files changed, 241 insertions, 169 deletions
diff --git a/src/ffts_real_nd.c b/src/ffts_real_nd.c
index fe9ef69..89ef7f7 100644
--- a/src/ffts_real_nd.c
+++ b/src/ffts_real_nd.c
@@ -1,197 +1,269 @@
/*
-
- 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.
+
+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"
+#include "ffts_internal.h"
+#include "ffts_transpose.h"
+
+static void
+ffts_free_nd_real(ffts_plan_t *p)
+{
+ if (p->plans) {
+ int i, j;
+
+ for (i = 0; i < p->rank; i++) {
+ ffts_plan_t *plan = p->plans[i];
+
+ if (plan) {
+ for (j = 0; j < i; j++) {
+ if (p->Ns[i] == p->Ns[j]) {
+ plan = NULL;
+ break;
+ }
+ }
+
+ if (plan) {
+ ffts_free(plan);
+ }
+ }
+ }
-#ifdef __ARM_NEON__
-#include "neon.h"
-#endif
-
-void ffts_free_nd_real(ffts_plan_t *p) {
-
- int i;
- for(i=0;i<p->rank;i++) {
-
- ffts_plan_t *x = p->plans[i];
-
- int k;
- for(k=i+1;k<p->rank;k++) {
- if(x == p->plans[k]) p->plans[k] = NULL;
- }
-
- if(x) ffts_free(x);
- }
-
- free(p->Ns);
- free(p->Ms);
- free(p->plans);
- free(p->buf);
- free(p->transpose_buf);
- free(p);
-}
+ free(p->plans);
+ }
+
+ if (p->buf) {
+ ffts_aligned_free(p->buf);
+ }
-void ffts_scalar_transpose(uint64_t *in, uint64_t *out, int w, int h, uint64_t *buf) {
+ if (p->Ns) {
+ free(p->Ns);
+ }
- size_t i,j;
- for(i=0;i<w;i+=1) {
- for(j=0;j<h;j+=1) {
- out[i*h + j] = in[j*w + i];
- }
- }
+ if (p->Ms) {
+ free(p->Ms);
+ }
+ free(p);
}
-void ffts_execute_nd_real(ffts_plan_t *p, const void * in, void * out) {
+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;
+
+ ffts_plan_t *plan;
+ int i;
+ size_t j;
+
+ plan = p->plans[0];
+ for (j = 0; j < Ns0; j++) {
+ plan->transform(plan, din + (j * Ms0), buf + (j * (Ms0 / 2 + 1)));
+ }
- uint32_t *din = (uint32_t *)in;
- uint64_t *buf = p->buf;
- uint64_t *dout = (uint64_t *)out;
+ ffts_transpose(buf, dout, Ms0 / 2 + 1, Ns0);
- size_t i,j;
- for(i=0;i<p->Ns[0];i++) {
- p->plans[0]->transform(p->plans[0], din + (i * p->Ms[0]), buf + (i * (p->Ms[0] / 2 + 1)));
- }
- ffts_scalar_transpose(buf, dout, p->Ms[0] / 2 + 1, p->Ns[0], p->transpose_buf);
+ for (i = 1; i < p->rank; i++) {
+ const size_t Ms = p->Ms[i];
+ const size_t Ns = p->Ns[i];
- for(i=1;i<p->rank;i++) {
- for(j=0;j<p->Ns[i];j++) {
- p->plans[i]->transform(p->plans[i], dout + (j * p->Ms[i]), buf + (j * p->Ms[i]));
- }
- ffts_scalar_transpose(buf, dout, p->Ms[i], p->Ns[i], p->transpose_buf);
- }
+ plan = p->plans[i];
+
+ for (j = 0; j < Ns; j++) {
+ plan->transform(plan, dout + (j * Ms), buf + (j * Ms));
+ }
+
+ ffts_transpose(buf, dout, Ms, Ns);
+ }
}
-void ffts_execute_nd_real_inv(ffts_plan_t *p, const void * in, void * out) {
+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;
+ float *doutr = (float*) out;
- uint64_t *din = (uint64_t *)in;
- uint64_t *buf = p->buf;
- uint64_t *buf2;
- uint64_t *dout = (uint64_t *)out;
- size_t vol = 1;
-
- float *bufr = (float *)(p->buf);
- float *doutr = (float *)out;
+ ffts_plan_t *plan;
+ size_t vol;
- size_t i,j;
+ int i;
+ size_t j;
- for(i=0;i<p->rank;i++) {
- vol *= p->Ns[i];
- }
+ vol = p->Ns[0];
+ for (i = 1; i < p->rank; i++) {
+ vol *= p->Ns[i];
+ }
- buf2 = buf + vol;
+ buf2 = buf + vol;
- ffts_scalar_transpose(din, buf, p->Ms[0], p->Ns[0], p->transpose_buf);
+ ffts_transpose(din, buf, Ms0, Ns0);
- for(i=0;i<p->Ms[0];i++) {
- p->plans[0]->transform(p->plans[0], buf + (i * p->Ns[0]), buf2 + (i * p->Ns[0]));
- }
-
- ffts_scalar_transpose(buf2, buf, p->Ns[0], p->Ms[0], p->transpose_buf);
- for(j=0;j<p->Ms[1];j++) {
- p->plans[1]->transform(p->plans[1], buf + (j * (p->Ms[0])), &doutr[j * p->Ns[1]]);
- }
+ plan = p->plans[0];
+ for (j = 0; j < Ms0; j++) {
+ plan->transform(plan, buf + (j * Ns0), buf2 + (j * Ns0));
+ }
+
+ ffts_transpose(buf2, buf, Ns0, Ms0);
+
+ 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) {
- size_t vol = 1;
- size_t bufsize;
-
- ffts_plan_t *p = malloc(sizeof(ffts_plan_t));
-
- 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->Ns = malloc(sizeof(size_t) * rank);
- p->Ms = malloc(sizeof(size_t) * rank);
- p->plans = malloc(sizeof(ffts_plan_t **) * rank);
- int i;
- 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 = valloc(sizeof(float) * bufsize);
-
- for(i=0;i<rank;i++) {
- p->Ms[i] = vol / p->Ns[i];
-
- p->plans[i] = NULL;
- int k;
-
- if(sign < 0) {
- for(k=1;k<i;k++) {
- if(p->Ms[k] == p->Ms[i]) p->plans[i] = p->plans[k];
- }
- if(!i) p->plans[i] = ffts_init_1d_real(p->Ms[i], sign);
- else if(!p->plans[i]) p->plans[i] = ffts_init_1d(p->Ms[i], sign);
- }else{
- for(k=0;k<i;k++) {
- if(p->Ns[k] == p->Ns[i]) p->plans[i] = p->plans[k];
- }
- if(i==rank-1) p->plans[i] = ffts_init_1d_real(p->Ns[i], sign);
- else if(!p->plans[i]) p->plans[i] = ffts_init_1d(p->Ns[i], sign);
- }
- }
- if(sign < 0) {
- for(i=1;i<rank;i++) {
- p->Ns[i] = p->Ns[i] / 2 + 1;
- }
- }else{
- for(i=0;i<rank-1;i++) {
- p->Ms[i] = p->Ms[i] / 2 + 1;
- }
- }
-
- p->transpose_buf = valloc(sizeof(float) * 2 * 8 * 8);
- return p;
+FFTS_API 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->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_API ffts_plan_t*
+ffts_init_2d_real(size_t N1, size_t N2, int sign)
+{
+ size_t Ns[2];
-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);
+ Ns[0] = N1;
+ Ns[1] = N2;
+ return ffts_init_nd_real(2, Ns, sign);
}
-// vim: set autoindent noexpandtab tabstop=3 shiftwidth=3:
OpenPOWER on IntegriCloud