changelog

FFTW, a collection of fast C routines to compute the Discrete Fourier Transform in one or more dime

	$<, since according to GNU Makefile conventions document, many
	non-GNU 'make' implementations only set $< for implicit rules.

Thu Jan 27 16:18:03 2000  fftw  <[e-mail]>

	* configure.in, gensrc/Makefile.fftw.am, gensrc/Makefile.rfftw.am,
	mpi/Makefile.am, threads/Makefile.am: Used $@, $< in Makefile
	rules for srfftw.h, etcetera.  This should fix things in VPATH
	builds; thanks to Michael Poole for the suggestion.

Wed Nov 10 20:01:40 1999  Steven G. Johnson  <[e-mail]>

	* threads/fftw_threads-int.h: Yikes!  Self-#include should be
	fftw_threads.h.

Mon Nov  8 00:53:35 1999  Steven G. Johnson  <[e-mail]>

	* NEWS: noted (expected) 2.1.3 release date

	* doc/texi2html, AUTHORS: Updated Matteo's email address.

	* README: Noted mailing list URL.

Sun Nov  7 19:15:52 1999  Steven G. Johnson  <[e-mail]>

	* doc/fftw.texi: Reference latest version of gcc bug report.

	* acinclude.m4: Updated stack alignment check to also check for
	misaligned main() (due to OS/libc/loader problems), and to
	reference latest version of gcc bug report.

Fri Nov  5 01:49:21 1999  Steven G. Johnson  <[e-mail]>

	* NEWS: Thanked Diab Jerius for CFLAGS patch.

	* NEWS: noted that configure no longer overrides CFLAGS.

	* acinclude.m4: Fixed code so that configure will not override the
	CFLAGS environment variable.

Tue Nov  2 23:52:02 1999  Steven G. Johnson  <[e-mail]>

	* rfftw/rfftwnd.c, doc/fftw.texi: Ugh, ugh, triple ugh!  Had to
	modify the rule for stride interpretation of in-place rfftwnd
	transforms to handle a pathological case.  Before, if you did
	howmany transforms of size 1 (stride=1, dist=1) both the r2c and
	c2r transforms would work okay, BUT: the input of the r2c
	transform would have the required "padding" elements to make room
	for the complex output, while the c2r transform would output real
	numbers *without* padding (because of the idist==1 -> odist_t=1
	rule).  Thus, we'd have the bizarre situation of r2c + c2r = input
	in different order (no padding).  This caused the rfftwnd_mpi
	tests to fail for Nx1 transforms (N > 2).  So, I've modified the
	rule slightly to idist==1 && idist < istride.  This should
	continue to make rfftwnd do what we want in all the ordinary
	cases, but fixes this case.

Mon Nov  1 22:31:14 1999  Steven G. Johnson  <[e-mail]>

	* NEWS: noted that gcc stack alignment bug was only on x86.

Sun Oct 31 23:17:12 1999  Steven G. Johnson  <[e-mail]>

	* mpi/test_transpose_mpi.c: got rid of unused variables.

	* mpi/test_sched.c: got rid of unused variable

	* mpi/rfftw_f77_mpi.c: fixed typo

	* configure.in: awk -F can't be followed by a space in older
	awk's.

	* mpi/Makefile.am: added README.f77 to distribution

	* mpi/README.f77: minor fixes

	* fortran/fftw_f77.i: changed "c" comments to "!" to accommodate
	f90 freeform mode.  Also added params for experimental MPI
	wrappers.

Fri Oct 29 21:41:59 1999  Steven G. Johnson  <[e-mail]>

	* mpi/README.f77, mpi/fftw_f77_mpi.h: minor additions to
	documentation

	* Makefile.am: disable autoheader here too (it was a problem once
	today; why not before?)

	* mpi/README.f77, NEWS: Added description of experimental MPI
	wrappers.

	* acinclude.m4, configure.in: Use AC_F77_LIBRARY_LDFLAGS when
	attempting to link Fortran with C code in ACX_F77_FUNC_MANGLE.

	* fftw/config.h.in, mpi/Makefile.am, mpi/fftw_f77_mpi.c,
	mpi/fftw_f77_mpi.h, mpi/rfftw_f77_mpi.c, configure.in: Added
	experimental Fortran-callable wrappers for the MPI transforms.

	* acinclude.m4: small fix

Thu Oct 28 19:53:33 1999  Steven G. Johnson  <[e-mail]>

	* doc/fftw.texi: Noted stack alignment bug in gcc 2.95.[012].

	* NEWS, acinclude.m4, configure.in: Work around buggy stack
	alignment in gcc-2.95.x.

	* acinclude.m4, configure.in: Reorganized macros in acinclude.m4
	somewhat.  New checks for -fstrict-aliasing and try to guess cpu
	type on powerpc by looking at /proc/cpuinfo (using -mcpu=750 makes
	a ~10% difference on a PowerPC G3).

Tue Oct 26 21:45:06 1999  Steven G. Johnson  <[e-mail]>

	* tests/test_main.c: Added obligatory new joke for 2.1.3 release.

	* threads/executor_threads.c, threads/rexec_threads.c,
	rfftw/rexec.c, rfftw/rexec2.c, gensrc/to_c.ml, fftw/executor.c,
	fftw/fftw-int.h: Technically, macros with empty argument lists
	invoke undefined behavior in ANSI C, although they will be allowed
	in the C99 language revision.  Fix HACK_ALIGN_STACK_* macros to
	remove this usage.

	* fftw/config.h.in, fftw/fftw-int.h, doc/fftw.texi, configure.in:
	--enable-i386-hacks is no longer needed in gcc-2.95+
	(i.e. versions which have the -mpreferred-stack-boundary flag).

	* fftw/fftw-int.h: Indented alignment #preprocessor statements to
	make them more readable.  Also, made it so that you can use
	--debug-alignment even without --enable-i386-hacks.

Tue Oct 26 16:17:51 1999  Matteo Frigo  <[e-mail]>

	* ChangeLog: Updated ChangeLog

Tue Oct 26 05:47:25 1999  Steven G. Johnson  <[e-mail]>

	* acinclude.m4: No point in doing both -mcpu and -mtune (-mcpu
	includes -mtune).

	* doc/fftw.texi: Some fixes in Fortran docs.

	* rfftw/rexec.c, rfftw/rplanner.c, fftw/config.h.in,
	fftw/executor.c, fftw/planner.c, configure.in: Vector recursion is
	now enabled by a --enable-vec-recurse flag (disabled by default).

	* NEWS: Added version number.

Tue Sep 28 09:13:39 1999  Steven G. Johnson  <[e-mail]>

	* rfftw/rplanner.c, fftw/planner.c: Disable vector recursion for
	now.

	* fftw/rader.c, fftw/twiddle.c, fftw/config.h.in, fftw/fftw-int.h,
	FAQ/fftw-faq.bfnn, NEWS, configure.in: Fixed overflow for (x*y)%p
	in Rader routines (for complex transforms of prime sizes).  Thanks
	to Ezio Riva ([e-mail]) for the bug report.

Thu Aug 19 03:08:29 1999  Steven G. Johnson  <[e-mail]>

	* NEWS: Noted Matlab wrapper bug fix.

Tue Aug 17 18:46:54 1999  Steven G. Johnson  <[e-mail]>

	* matlab/fftw.c: Fixed bug (memory leak; a new plan is created on
	each call for the multi-dim.  transforms).  Thanks to Matthew
	Davis <[e-mail]> for the bug report.

Wed Jul 28 19:12:15 1999  Steven G. Johnson  <[e-mail]>

	* NEWS, configure.in: Fixed configure script problems with
	--enable-threads on Digital Unix.

Tue Jul 27 20:03:22 1999  Steven G. Johnson  <[e-mail]>

	* fftw.spec.in: Many changes, including several based on
	suggestions by Keith Amidon <[e-mail]>.  Now install
	into the "build root" at build time so that building the RPM does
	not affect existing FFTW installations (and doesn't require root
	privileges).  Compilation (and configuration) now only occurs
	during the build phase.  Post install and uninstall scripts run
	ldconfig to update the linker database, and warn the user if the
	install directory is not in /etc/ld.so.conf.  Finally, the
	double-precision libraries are installed into the standard [r]fftw
	names instead of under d[r]fftw.

	* tests/test_main.c: Minor cleanup.

Sat Jul 24 20:32:58 1999  Steven G. Johnson  <[e-mail]>

	* tests/test_main.c, tests/test_main.h: Added -1
	(--only-one-speed-test) option, as I'm tired of waiting for the 8
	different speed tests to run when I only want one.

	* threads/fftw_threads_test.c, threads/rfftw_threads_test.c,
	tests/fftw_test.c, tests/rfftw_test.c, tests/test_main.c,
	tests/test_main.h, rfftw/rplanner.c, mpi/fftw_mpi_test.c,
	mpi/rfftw_mpi_test.c, fftw/fftw-int.h, fftw/fftw.h.in,
	fftw/fftwnd.c, fftw/planner.c, fftw/putils.c, fftw/rader.c,
	fftw/wisdom.c, fftw/wisdomio.c: Implemented new
	FFTW_NO_VECTOR_RECURSE flag to inhibit use of vector recursion.
	This flag is used internally in the planner to prevent vector
	recursion at anything other than the top level of the plan
	(instead of the planner_depth stuff used previously), and insures
	that the correct wisdom is used.  (Note, however, that the wisdom
	is still not specific to the vector_size parameter; this needs to
	be fixed.)

Mon Jul 19 17:04:16 1999  Steven G. Johnson  <[e-mail]>

	* fftw/fftw-int.h: Added a comment about why we have macros for
	trig. functions.

Mon Jun 21 16:01:44 1999  fftw  <[e-mail]>

	* FAQ/fftw-faq.bfnn: Noted that SGI MipsPro bugs seem to have been
	fixed.  Also mentioned problems with egcs-1.0.2 on the PowerPC.

Sat Jun 19 17:47:28 1999  fftw  <[e-mail]>

	* FAQ/fftw-faq.bfnn: Noted incorrect code generation in Metrowerks
	CodeWarrior Pro 4 for the Macintosh.  (Maybe we should rename this
	FAQ question?)

Fri Jun 18 19:20:28 1999  fftw  <[e-mail]>

	* FAQ/m-html.pl: Updated FFTW Manual bookmark.

Sat Jun 12 05:03:31 1999  fftw  <[e-mail]>

	* doc/fftw.texi: Added clarifications to fftwnd reference, as
	suggested by John F. Gibson of Cornell.

Tue Jun  8 22:17:35 1999  fftw  <[e-mail]>

	* threads/fftw_threads-int.h, tests/test_main.c, matlab/fftw.m,
	gensrc/README, matlab/README, fftw/fftw-int.h, doc/fftw.texi,
	FAQ/html.refs, FAQ/m-html.pl, cilk/README, FAQ/fftw-faq.bfnn,
	TODO, fftw.spec.in, Makefile.am, README: Great URL change: web
	page -> www.fftw.org, ftp -> ftp.fftw.org, email -> @fftw.org.
	Note that currently, this means that some of links in the
	documentation are broken (links to sub-pages within the main FFTW
	site).  This will be fixed once we get real hosting for fftw.org.

Mon Jun  7 23:36:19 1999  fftw  <[e-mail]>

	* fftw/config.h.in, fftw/fftw-int.h: Added warnings to flaky MacOS
	nanosecond timer routines.  (The ordinary Mac timer routines, with
	microsecond accuracy (in theory) are okay.)

Tue Jun  1 21:52:50 1999  fftw  <[e-mail]>

	* fftw/fftw-int.h: Oops, I declared fftw_time twice.

	* fftw/fftw-int.h: Incorporated new timer code by Sampo Niskanen.

Mon May 31 17:45:34 1999  Matteo Frigo  <[e-mail]>

	* acinclude.m4: Added -arch host to osf-alpha CFLAGS.

Sat May 29 20:38:03 1999  fftw  <[e-mail]>

	* Makefile.am: slight reformatting.

	* Makefile.am: Make sure lynx doesn't attempt to reformat logo gif
	data.

	* NEWS: Noted the date of each release (plus or minus a day or
	two, in some cases).  Merged release notes for beta releases with
	those of final releases.

Sat May 29 04:28:29 1999  Steven G. Johnson  <[e-mail]>

	* ChangeLog: ChangeLog now reflects all revisions of FFTW (instead
	of starting at 6/11/98).

	* ChangeLog: Updated change log.

Sat May 29 03:25:59 1999  fftw  <[e-mail]>

	* doc/fftw.texi: Fixed another typo.

	* doc/fftw.texi: Typo fix.

Fri May 28 19:33:41 1999  fftw  <[e-mail]>

	* fftw.spec.in, Attic/fftw.spec, Makefile.am, configure.in: You
	can now build the RPM package by running 'make rpm' (as root)
	after 'make dist'.  fftw.spec is now kept in sync automatically
	with the version numbers in configure.in.

Thu May 27 05:48:57 1999  fftw  <[e-mail]>

	* threads/rexec_threads.c, rfftw/rplanner.c,
	threads/executor_threads.c, rfftw/rexec2.c, rfftw/rfftw.h,
	rfftw/rexec.c, fftw/rader.c, fftw/wisdom.c, fftw/putils.c,
	fftw/planner.c, fftw/fftw.h.in, fftw/fftw-int.h, fftw/executor.c:
	Initial version of "vector recursion" in executor.  No vector
	codelets yet.  The threads code works, but does not do vector
	recursion.  No use of vector recursion for in-place howmany loops
	yet.

Wed May 26 18:06:47 1999  fftw  <[e-mail]>

	* gensrc/complex.ml: In times_3_3 routine, got rid of infinite
	loop for multiplication by non-constants (was biting us during
	generation of twiddle codelets).

	* gensrc/number.ml, gensrc/magic.ml, gensrc/genfft.ml,
	gensrc/exprdag.ml, gensrc/complex.ml: Various cleanup

	* tests/test_main.c, fftw/config.h.in, fftw/fftw.h.in: Centralized
	tests for Windows (in config.h.in) and use HAVE_WIN32 elsewhere.

Tue May 25 23:59:14 1999  fftw  <[e-mail]>

	* NEWS: Removed extra period.

	* doc/fftw.texi: Noted possibility of "d" or "s" prefix in the
	tutorial (as requested by many users).

	* FAQ/fftw-faq.bfnn, Attic/fftw.spec, NEWS, configure.in: Bumped
	version number to 2.1.3, documented AIX threads fix.

	* threads/fftw_threads-int.h, threads/fftw_threads.c,
	fftw/fftw-int.h, configure.in, fftw/config.h.in, acinclude.m4:
	Fixed threads bug on AIX: pthread_create on AIX spawns detached
	(non-joinable) threads by default.  Also fixed much autoconf
	lossage on AIX.  Thanks to Jim Lindsay ([e-mail])
	for the bug report and for the use of Northwestern's SP2.

	* fftw/timer.c: Only use BSDgettimeofday if gettimeofday is not
	available (causes problems on some AIX systems).

	* tests/rfftw_test.c, tests/fftw_test.c, rfftw/rplanner.c,
	fftw/planner.c, fftw/fftw-int.h: Put (r)fftw_plan_hook in
	fftw-int.h, with a typedef, and used Andrew Sterian's DL_IMPORT
	macro to succor the lost souls using VC++.

Fri May 21 17:58:30 1999  fftw  <[e-mail]>

	* tests/test_main.c: Fixed typo in help string.

Tue May 18 23:39:07 1999  fftw  <[e-mail]>

	* threads/rfftw_threads_test.c, tests/test_main.h,
	threads/fftw_threads_test.c, tests/test_main.c,
	tests/rfftw_test.c, tests/fftw_test.c, mpi/rfftw_mpi_test.c,
	mpi/fftw_mpi_test.c, FAQ/fftw-faq.bfnn, NEWS: Whoops, found the
	real source of the MPI bug that was supposedly fixed in 2.1.2.
	MPI_Init can modify argv (and does so, for processes other than
	the first), but we passed the original argv to getopt.  This is
	now fixed.  (This is a bug in the MPI test programs of 2.1.2 also,
	but as it's only in the test programs and doesn't seem to bite us
	on any known MPI implementation, it can wait until the next FFTW
	release.)  Sigh.

Mon May 17 23:09:31 1999  fftw  <[e-mail]>

	* NEWS: Went into more detail regarding the generator changes.

Mon May 17 22:20:27 1999  Matteo Frigo  <[e-mail]>

	* NEWS: Note reduction in rfftw code size.

	* gensrc/symmetry.ml, gensrc/genfft.ml, fftw/twiddle.c: Reduced
	code size and # of operations of rfftw twiddle codelets.

Mon May 17 19:57:11 1999  fftw  <[e-mail]>

	* ChangeLog: Updated

	* NEWS: Noted codelet generator speed.

	* gensrc/symmetry.ml, gensrc/genfft.ml, gensrc/fft.ml: Improved
	real{even|odd}2 transforms

	* doc/fftw.texi: Fixed typo.

	* gensrc/exprdag.ml: Improved simplifier for DCT-type transforms.

	* TODO: Noted that the transpose routines for MPI could use
	improving.

	* gensrc/symmetry.ml: Fixed comment.

	* gensrc/symmetry.ml, gensrc/genfft.ml, gensrc/fft.ml: Fixed
	modified DCT/DST generation (realeven2/realodd2) so that it works