Put cpy2d_pair into its own file, so that I can experiment with

buffering of nontwiddle codelets.

dft/dftw-directbuf.c

@@ -18,7 +18,7 @@
  *
  */
 
-/* $Id: dftw-directbuf.c,v 1.4 2005-02-19 14:31:14 athena Exp $ */
+/* $Id: dftw-directbuf.c,v 1.5 2005-02-23 03:32:06 athena Exp $ */
 
 #include "ct.h"
 
@@ -41,37 +41,16 @@ typedef struct {
      const S *slv;
 } P;
 
-/*
-   Copy A -> B, where A and B are n0 x n1 complex matrices
-   such that the (i0, i1) element has index (i0 * s0 + i1 * s1). 
-*/
-static void cpy(int n0, int n1, 
-		const R *rA, const R *iA, int sa0, int sa1, 
-		R *rB, R *iB, int sb0, int sb1)
-{
-     int i0, i1;
-     ptrdiff_t ima = iA - rA, imb = iB - rB;
-
-     for (i0 = 0; i0 < n0; ++i0) {
-	  const R *pa; 
-	  R *pb;
-
-	  pa = rA; rA += sa0;
-	  pb = rB; rB += sb0;
-	  for (i1 = 0; i1 < n1; ++i1) {
-	       R xr = pa[0], xi = pa[ima];
-	       pb[0] = xr; pb[imb] = xi; 
-	       pa += sa1; pb += sb1;
-	  }
-     }
-}
-
 static const R *doit(kdftw k, R *rA, R *iA, const R *W, int ios, int dist, 
 		     int r, int batchsz, R *buf, stride bufstride)
 {
-     cpy(r, batchsz, rA, iA, ios, dist, buf, buf + 1, 2, 2 * r);
+     X(cpy2d_pair_ci)(rA, iA, buf, buf + 1, 
+		      r, ios, 2,
+		      batchsz, dist, 2 * r);
      W = k(buf, buf + 1, W, bufstride, batchsz, 2 * r);
-     cpy(r, batchsz, buf, buf + 1, 2, 2 * r, rA, iA, ios, dist);
+     X(cpy2d_pair_co)(buf, buf + 1, rA, iA,
+		      r, 2, ios,
+		      batchsz, 2 * r, dist);
      return W;
 }
 

kernel/Makefile.am

@@ -2,8 +2,9 @@ AM_CPPFLAGS = -I$(top_srcdir)/simd
 noinst_LTLIBRARIES = libkernel.la
 
 libkernel_la_SOURCES = align.c alloc.c assert.c awake.c buffered.c	\
-cpy1d.c cpy2d.c ct.c debug.c hash.c iabs.c kalloc.c md5-1.c md5.c	\
-minmax.c ops.c pickdim.c plan.c planner.c primes.c print.c problem.c	\
-rader.c scan.c solver.c solvtab.c stride.c tensor.c tensor1.c		\
-tensor2.c tensor4.c tensor5.c tensor7.c tensor8.c tensor9.c timer.c	\
-transpose-rec.c transpose.c trig.c trig1.c twiddle.c cycle.h ifftw.h
+cpy1d.c cpy2d-pair.c cpy2d.c ct.c debug.c hash.c iabs.c kalloc.c	\
+md5-1.c md5.c minmax.c ops.c pickdim.c plan.c planner.c primes.c	\
+print.c problem.c rader.c scan.c solver.c solvtab.c stride.c tensor.c	\
+tensor1.c tensor2.c tensor4.c tensor5.c tensor7.c tensor8.c tensor9.c	\
+timer.c transpose-rec.c transpose.c trig.c trig1.c twiddle.c cycle.h	\
+ifftw.h

kernel/cpy2d-pair.c

@@ -0,0 +1,61 @@
+/*
+ * Copyright (c) 2003 Matteo Frigo
+ * Copyright (c) 2003 Massachusetts Institute of Technology
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ *
+ */
+
+/* out of place copy routines for pairs of isomorphic 2D arrays */
+#include "ifftw.h"
+
+void X(cpy2d_pair)(R *I0, R *I1, R *O0, R *O1,
+		   int n0, int is0, int os0,
+		   int n1, int is1, int os1)
+{
+     int i0, i1;
+
+     for (i1 = 0; i1 < n1; ++i1)
+	  for (i0 = 0; i0 < n0; ++i0) {
+	       R x0 = I0[i0 * is0 + i1 * is1];
+	       R x1 = I1[i0 * is0 + i1 * is1];
+	       O0[i0 * os0 + i1 * os1] = x0;
+	       O1[i0 * os0 + i1 * os1] = x1;
+	  }
+}
+
+#define IABS(x) (((x) < 0) ? (-(x)) : (x))
+
+/* like cpy2d_pair, but read input contiguously if possible */
+void X(cpy2d_pair_ci)(R *I0, R *I1, R *O0, R *O1,
+		      int n0, int is0, int os0,
+		      int n1, int is1, int os1)
+{
+     if (IABS(is0) < IABS(is1))	/* inner loop is for n0 */
+	  X(cpy2d_pair) (I0, I1, O0, O1, n0, is0, os0, n1, is1, os1);
+     else
+	  X(cpy2d_pair) (I0, I1, O0, O1, n1, is1, os1, n0, is0, os0);
+}
+
+/* like cpy2d_pair, but write output contiguously if possible */
+void X(cpy2d_pair_co)(R *I0, R *I1, R *O0, R *O1,
+		      int n0, int is0, int os0,
+		      int n1, int is1, int os1)
+{
+     if (IABS(os0) < IABS(os1))	/* inner loop is for n0 */
+	  X(cpy2d_pair) (I0, I1, O0, O1, n0, is0, os0, n1, is1, os1);
+     else
+	  X(cpy2d_pair) (I0, I1, O0, O1, n1, is1, os1, n0, is0, os0);
+}

kernel/ifftw.h

@@ -18,7 +18,7 @@
  *
  */
 
-/* $Id: ifftw.h,v 1.247 2005-02-22 15:06:13 athena Exp $ */
+/* $Id: ifftw.h,v 1.248 2005-02-23 03:32:06 athena Exp $ */
 
 /* FFTW internal header file */
 #ifndef __IFFTW_H__
@@ -741,6 +741,16 @@ void X(cpy2d_recbuf)(R *I, R *O,
 		     int n0, int is0, int os0,
 		     int n1, int is1, int os1,
 		     int vl);
+void X(cpy2d_pair)(R *I0, R *I1, R *O0, R *O1,
+		   int n0, int is0, int os0,
+		   int n1, int is1, int os1);
+void X(cpy2d_pair_ci)(R *I0, R *I1, R *O0, R *O1,
+		      int n0, int is0, int os0,
+		      int n1, int is1, int os1);
+void X(cpy2d_pair_co)(R *I0, R *I1, R *O0, R *O1,
+		      int n0, int is0, int os0,
+		      int n1, int is1, int os1);
+
 void X(transpose)(R *I, int n, int s0, int s1, int vl);
 void X(transpose_rec)(R *I, int n, int s0, int s1, int vl) ;
 void X(transpose_recbuf)(R *I, int n, int s0, int s1, int vl) ;