changelog

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

	now.  The simplifier really sucks for this, at the moment.  Why?

Sun May 16 23:59:33 1999  fftw  <[e-mail]>

	* gensrc/genfft.ml: Fixed real[even,odd]2 codelet node type
	output.

	* gensrc/variable.ml, gensrc/variable.mli, gensrc/symmetry.ml,
	gensrc/genfft.ml: Added generators realeven2 and realodd2 for the
	modified DCT and DST (i.e. transforms for real data that are
	even/odd about n=-1/2, not n=0).

	* gensrc/exprdag.ml: Memoized eval for speed reasons.

	* gensrc/util.ml, gensrc/exprdag.ml: Implemented better
	statistics.

	* gensrc/exprdag.ml: Improved complexity of network transposition
	from O(n^2)=O(slow) to O(n) [times O(polylog)]

	* TODO: Noted that the generator can now output efficient
	hard-coded DCT/DST routines of small sizes.

	* gensrc/variable.mli, gensrc/variable.ml, gensrc/symmetry.ml,
	gensrc/genfft.ml, gensrc/complex.mli, gensrc/complex.ml: Added
	realeven and realodd codelet generation options.  (They work.)

Mon May 10 02:48:14 1999  fftw  <[e-mail]>

	* doc/fftw.texi: Noted how to force compiler choice via CC
	env. variable.  Also added "compiler" index entry.

Fri May  7 20:12:11 1999  fftw  <[e-mail]>

	* TODO, NEWS: Updated TODO.

	* README: Fixed Cilk URL.

	* NEWS: Noted GNU-style long options in the test programs.

Thu May  6 22:35:05 1999  fftw  <[e-mail]>

	* tests/test_main.c: --help option no longer causes an exit with
	an error code.

	* tests/test_main.c, configure.in, fftw/config.h.in: Test programs
	now use GNU-ly correct long options if getopt_long is available.

	* FAQ/fftw-faq.bfnn, NEWS: MPICH bug fix is apparently not
	specific to Linux (it also fixed the same problem on some
	Ultrasparcs).

	* Attic/fftw.spec: Updated for 2.1.2.

	* mpi/fftw_mpi_test.c: --only-parallel argument hack should also
	only be accessed by process 0.

Wed May  5 23:02:12 1999  fftw  <[e-mail]>

	* NEWS: Noted addition of omitted fftw_f77_threads_init function.

	* FAQ/fftw-faq.bfnn: Bug fix (dates need to have two digits for
	the day).

	* threads/rfftw_threads_test.c, threads/fftw_threads_test.c,
	tests/test_main.c, tests/test_main.h, tests/rfftw_test.c,
	tests/fftw_test.c, mpi/rfftw_mpi_test.c, mpi/test_transpose_mpi.c,
	FAQ/fftw-faq.bfnn, mpi/fftw_mpi_test.c, NEWS, configure.in: Fixed
	bug when running test programs under MPICH; prepared for 2.1.2
	release.

Sun Apr 18 04:16:39 1999  Steven G. Johnson  <[e-mail]>

	* Attic/fftw.spec: Made package relocatable.

	* Attic/fftw.spec: Created spec file for generating RPM packages
	of FFTW.

	* FAQ/fftw-faq.bfnn: Revised linker FAQ entry.

Mon Apr 12 19:34:17 1999  fftw  <[e-mail]>

	* tests/test_main.c: Added joke.

Sat Apr 10 00:14:37 1999  fftw  <[e-mail]>

	* FAQ/fftw-faq.bfnn: Added linker FAQ entry.

Wed Apr  7 17:17:48 1999  Steven G. Johnson  <[e-mail]>

	* threads/fftw_f77_threads.c: Added missing fftw_f77_threads_init
	function (thanks to V. Sundararajan for pointing out the
	omission).

Tue Apr  6 19:29:13 1999  Steven G. Johnson  <[e-mail]>

	* configure.in: Check for hcc in addition to mpicc (hcc is used by
	the LAM MPI implementation).

Wed Mar 31 00:54:20 1999  Steven G. Johnson  <[e-mail]>

	* NEWS, doc/fftw.texi, ChangeLog: Fixed credit for 2.1.1 bug fix
	(real credit goes to Ming-Chang Liu, according to Jeff
	Briedenbach, whose name was misspelled anyway).

Tue Mar 30 18:19:21 1999  Steven G. Johnson  <[e-mail]>

	* FAQ/fftw-faq.bfnn: Moderated language.

	* NEWS: This isn't LaTeX, Matteo; an en dash in a monospaced font
	is rendered the same as a hyphen (i.e. "-" not "--").

Mon Mar 29 19:23:24 1999  Matteo Frigo  <[e-mail]>

	* ChangeLog: Updated

	* NEWS: Minor change.

Sun Mar 28 20:24:30 1999  Matteo Frigo  <[e-mail]>

	* tests/rfftw_test.c, tests/fftw_test.c: Implemented paranoid
	check for real->complex and complex->real transforms

Sun Mar 28 00:11:47 1999  fftw  <[e-mail]>

	* NEWS: Moderated language.  I do not want to claim that the bug
	occurs in `rare' circumstances.  A bug is a bug, period.

	* tests/fftw_test.c, rfftw/rplanner.c: Implemented paranoid check
	for in-place complex planners.  I still don't know how to check
	real->complex plans, though (apart from rewriting test_ergun for
	real->complex and complex->real)

Fri Mar 26 22:58:57 1999  Steven G. Johnson  <[e-mail]>

	* doc/fftw.texi, FAQ/fftw-faq.bfnn, NEWS: Updated documentation
	for FFTW 2.1.1.

	* configure.in: Updated version number for 2.1.1 (as well as
	shared lib. version).

	* threads/Makefile.am, mpi/Makefile.am: Fixed typo in comment.

	* threads/fftw_threads_test.c, threads/rfftw_threads_test.c,
	mpi/rfftw_mpi_test.c, mpi/fftw_mpi_test.c: Added definition of
	enter_paranoid_mode() to threads & MPI test programs.

	* fftw/planner.c: Slight modification for future safety (not
	necessary now).

Thu Mar 25 02:29:08 1999  Matteo Frigo  <[e-mail]>

	* tests/test_main.c, tests/test_main.h, tests/rfftw_test.c,
	tests/fftw_test.c, fftw/planner.c: Added hooks for paranoid
	verification of every single plan.

Tue Mar 23 17:44:35 1999  Matteo Frigo  <[e-mail]>

	* fftw/generic.c: Array read out of bounds---fixed.  Thanks to
	Jeff Breiden.

Tue Mar  9 01:40:54 1999  Steven G. Johnson  <[e-mail]>

	* FAQ/fftw-faq.bfnn: Noted free-ness in "What is FFTW?"
	description.

Mon Mar  8 20:21:46 1999  Steven G. Johnson  <[e-mail]>

	* threads/fftw_threads.c: By default, don't even try to specify
	PTHREAD_SCOPE_SYSTEM, since it causes problems on @!#%$ IRIX 6.5
	(in which PTHREAD_SCOPE_SYSTEM is not supported, but
	pthread_attr_setscope doesn't return an error!!!).  Just use the
	default attributes (fftw_pthread_attributes_p == NULL).

	* threads/fftw_threads.c: Fixed typo in comment.

	* NEWS: Added 2.1 news.

Mon Mar  8 18:00:37 1999  fftw  <[e-mail]>

	* tests/rfftw_test.c, tests/fftw_test.c: Fixed pow(8192, x) -->
	pow(8192.0, x) .  This failed on Digital unix, I don't know why.

Mon Mar  8 15:16:51 1999  Matteo Frigo  <[e-mail]>

	* ChangeLog: Updated

	* configure.in: Updated version number

Sun Mar  7 19:47:54 1999  Steven G. Johnson  <[e-mail]>

	* doc/fftw.texi: a_slow_array --> a_bad_array

	* doc/fftw.texi: Removed all reference to performance from
	"Dynamic Arrays--The Wrong Way" section.

	* doc/fftw.texi: Clarified output parameter of Fortran
	multi-dimensional in-place transforms.

	* NEWS, doc/fftw.texi: Thanks to Erik Scheirer ([e-mail])
	for testing the Mach C threads code.

Sat Mar  6 05:59:42 1999  Steven G. Johnson  <[e-mail]>

	* mpi/fftw_mpi_test.c: Added undocumented --only-parallel flag.

	* threads/rfftw_f77_threads.c, threads/fftw_f77_threads.c,
	threads/Makefile.am, doc/fftw.texi, NEWS: Added Fortran-callable
	wrappers for the multi-threaded routines.

Thu Feb 25 16:42:10 1999  Steven G. Johnson  <[e-mail]>

	* threads/fftw_threads.c, threads/fftw_threads-int.h: Made
	pthreads code more tolerant if user doesn't call
	fftw_threads_init.

	* tests/rfftw_test.c, tests/fftw_test.c, mpi/rfftw_mpi_test.c,
	mpi/fftw_mpi_test.c: Reduced array sizes somewhat for planner
	tests.

Wed Feb 24 21:43:26 1999  Steven G. Johnson  <[e-mail]>

	* doc/fftw.texi: Added user thank-you to acknowledgements.

	* doc/fftw.texi: Clarified output format of 1d MPI transforms.

	* FAQ/fftw-faq.bfnn: Small change.

Wed Feb 24 19:45:30 1999  fftw  <[e-mail]>

	* FAQ/fftw-faq.bfnn: Changed wording regarding VC++5.0.

	* doc/texi2html: Fixed bug in index generation

Wed Feb 24 04:16:19 1999  Steven G. Johnson  <[e-mail]>

	* doc/fftw.texi: moved configure to fpindex, and added more
	entries.

	* doc/fftw.texi: Added more padding index entries.

	* doc/fftw.texi: Objective Caml -> Caml in the index.

	* doc/fftw.texi: Added @cindex floating-point precision

	* doc/fftw.texi: Updated index.

	* doc/fftw.texi: typo fix: rfftwnd_threads_*_one ->
	rfftwnd_threads_one_*

	* doc/fftw.texi: Print table of contents at the end, as directed
	by the texinfo manual, so that pagination is not screwed up when
	the TOC has an odd number of pages.

Tue Feb 23 04:32:15 1999  Steven G. Johnson  <[e-mail]>

	* doc/fftw.texi: Fixed typo.

	* doc/fftw.texi: Indentation fix.

Mon Feb 22 19:20:41 1999  Matteo Frigo  <[e-mail]>

	* ChangeLog: Updated

Mon Feb 22 19:16:12 1999  Steven G. Johnson  <[e-mail]>

	* FAQ/fftw-faq.bfnn: Noted workaround for buggy SGI compilers.

	* ChangeLog: Line-wrapped (I wish you would remember to do this,
	Matteo).

	* gensrc/Makefile.rfftw.am, mpi/Makefile.am, threads/Makefile.am,
	gensrc/Makefile.fftw.am: Got rid of hackery in header creation,
	since circular dependencies are gone.  Also added prefixed headers
	to CLEANFILES so they get deleted by 'make clean'.

	* configure.in: Used xyz for prefix1 instead of just x, to make
	future conflicts less likely.

Mon Feb 22 18:55:16 1999  Matteo Frigo  <[e-mail]>

	* configure.in: Fix for irix didn't work, let's try this

	* ChangeLog: Fixed typo

	* ChangeLog: Updated

	* configure.in: Accounted for irix thread lossage

	* tests/test_main.c: usage() must normally exit, otherwise the
	test programs tries to read argv[] out of bounds.

	* doc/Makefile.am: fftw.ps used compressed fonts

	* gensrc/Makefile.rfftw.am, gensrc/Makefile.fftw.am: rfftw.h
	didn't go into distribution.  Fixed

	* threads/Makefile.am, gensrc/Makefile.rfftw.am, mpi/Makefile.am,
	configure.in, gensrc/Makefile.fftw.am: Removed circular
	dependencies

	* configure.in: Fixed typo

Mon Feb 22 05:19:40 1999  Steven G. Johnson  <[e-mail]>

	* tests/test_main.c: Added some new jokes.  Also, print out
	floating-point precision for -v.

	* mpi/rfftw_mpi_test.c, threads/Makefile.am,
	gensrc/Makefile.rfftw.am, mpi/Makefile.am, mpi/fftw_mpi_test.c,
	gensrc/Makefile.fftw.am: Made sure header files are not deleted
	when no prefix is added.  (Also got rid of some extraneous
	newlines printed by mpi code.)

	* configure.in, NEWS: Renamed 2.1 -> 2.1-beta1.

	* doc/Makefile.am: make clean shouldn't remove rfftwnd.gif.

Sun Feb 21 20:33:51 1999  Steven G. Johnson  <[e-mail]>

	* NEWS: Feature set for 2.1 is now frozen (I think).

	* doc/fftw.texi: slight change

	* mpi/Makefile.am, threads/Makefile.am, gensrc/Makefile.fftw.am,
	gensrc/Makefile.rfftw.am: Whoops!  Don't include
	XXX_FFTW_PREFIX_XXX header files in CLEANFILES, because that will
	cause 'make clean' to delete irreplaceable files when no prefix is
	being used.

	* mpi/Makefile.am, threads/Makefile.am, gensrc/Makefile.fftw.am,
	gensrc/Makefile.rfftw.am: Fixed installed headers to use
	FFTW_PREFIX for their own header inclusions.

	* configure.in, doc/fftw.texi, NEWS: Documented
	--enable-type-prefix.

	* mpi/Makefile.am, tests/Makefile.am, threads/Makefile.am,
	gensrc/Makefile.fftw.am, gensrc/Makefile.rfftw.am, acinclude.m4,
	configure.in: Use FFTW_PREFIX for all installed files.

	* configure.in: Use g77 in preference to fort77.

Sun Feb 21 17:56:49 1999  Matteo Frigo  <[e-mail]>

	* tests/Makefile.am: Fixed forgotted XXX_FFTW_NAME_XXX

	* configure.in: Allowed user specification of [r]fftw-name

Sun Feb 21 17:35:28 1999  fftw  <[e-mail]>

	* tests/Makefile.am, configure.in, gensrc/Makefile.fftw.am,
	gensrc/Makefile.rfftw.am, acinclude.m4: (Preliminary) hack: allow
	user to change name of fftw library.

	* configure.in: Incremented shared library version number.

	* configure.in: Added rfftw_mpi.h to list of files to install for
	--enable-mpi.

Sat Feb 20 23:00:41 1999  Steven G. Johnson  <[e-mail]>

	* FAQ/fftw-faq.bfnn: The ascii and info (but not html) versions of
	the FAQ were chopping off the answer to the last question.  Rather
	than fix the bug, I just added a trailing comment like in the
	Linux FAQ, which seems to correct the problem.

	* FAQ/fftw-faq.bfnn: Minor fixes and updates for 2.1.

	* mpi/rfftw_mpi_test.c, mpi/fftw_mpi.h, mpi/fftw_mpi_test.c,
	mpi/fftw_mpi.c: Share plans between forward and backward fftw_mpi
	plans (and between other fftw_mpi plans of the same size).

	* tests/rfftw_test.c, tests/fftw_test.c: In -p test with rank > 1,
	reduce the size of the arrays for the plan.

	* mpi/transpose_mpi.c: Got rid of compiler warning (code was okay,
	though).

	* doc/fftw.texi: small fix.

Sat Feb 20 03:27:36 1999  fftw  <[e-mail]>

	* doc/fftw.texi: Minor tweaks

Sat Feb 20 02:57:37 1999  Steven G. Johnson  <[e-mail]>

	* doc/Makefile.am, doc/rfftwnd.gif: fig2dev gif output stinks;