Chris@10: FFTW 3.3.3
Chris@10: 
Chris@10: * Fix deadlock bug in MPI transforms (thanks to Michael Pippig for the
Chris@10:   bug report and patch, and to Graham Dennis for the bug report).
Chris@10: 
Chris@10: * Use 128-bit ARM NEON instructions instead of 64-bits.  This change
Chris@10:   appears to speed up even ARM processors with a 64-bit NEON pipe.
Chris@10: 
Chris@10: * Speed improvements for single-precision AVX.
Chris@10: 
Chris@10: * Speed up planner on machines without "official" cycle counters, such as ARM.
Chris@10: 
Chris@10: FFTW 3.3.2
Chris@10: 
Chris@10: * Removed an archaic stack-alignment hack that was failing with
Chris@10:   gcc-4.7/i386.
Chris@10: 
Chris@10: * Added stack-alignment hack necessary for gcc on Windows/i386.  We
Chris@10:   will regret this in ten years (see previous change).
Chris@10:   
Chris@10: * Fix incompatibility with Intel icc which pretends to be gcc
Chris@10:   but does not support quad precision.
Chris@10: 
Chris@10: * make libfftw{threads,mpi} depend upon libfftw when using libtool;
Chris@10:   this is consistent with most other libraries and simplifies the life
Chris@10:   of various distributors of GNU/Linux.
Chris@10: 
Chris@10: FFTW 3.3.1
Chris@10: 
Chris@10: * Changes since 3.3.1-beta1:
Chris@10:   
Chris@10:   - Reduced planning time in estimate mode for sizes with large
Chris@10:     prime factors.
Chris@10: 
Chris@10:   - Added AVX autodetection under Visual Studio.  Thanks Carsten
Chris@10:     Steger for submitting the necessary code.
Chris@10: 
Chris@10:   - Modern Fortran interface now uses a separate fftw3l.f03 interface
Chris@10:     file for the long double interface, which is not supported by
Chris@10:     some Fortran compilers.  Provided new fftw3q.f03 interface file
Chris@10:     to access the quadruple-precision FFTW routines with recent
Chris@10:     versions of gcc/gfortran.
Chris@10: 
Chris@10: * Added support for the NEON extensions to the ARM ISA.  (Note to beta
Chris@10:   users: an ARM cycle counter is not yet implemented; please contact
Chris@10:   fftw@fftw.org if you know how to do it right.)
Chris@10: 
Chris@10: * MPI code now compiles even if mpicc is a C++ compiler; thanks to
Chris@10:   Kyle Spyksma for the bug report.
Chris@10: 
Chris@10: FFTW 3.3
Chris@10: 
Chris@10: * Changes since 3.3-beta1:
Chris@10: 
Chris@10:   - Compiling OpenMP support (--enable-openmp) now installs a
Chris@10:     fftw3_omp library, instead of fftw3_threads, so that OpenMP
Chris@10:     and POSIX threads (--enable-threads) libraries can be built
Chris@10:     and installed at the same time.
Chris@10: 
Chris@10:   - Various minor compilation fixes, corrections of manual typos, and
Chris@10:     improvements to the benchmark test program.
Chris@10: 
Chris@10: * Add support for the AVX extensions to x86 and x86-64.  The AVX code
Chris@10:   works with 16-byte alignment (as opposed to 32-byte alignment),
Chris@10:   so there is no ABI change compared to FFTW 3.2.2.
Chris@10: 
Chris@10: * Added Fortran 2003 interface, which should be usable on most modern
Chris@10:   Fortran compilers (e.g. gfortran) and provides type-checked access
Chris@10:   to the the C FFTW interface.  (The legacy Fortran-77 interface is
Chris@10:   still included also.)
Chris@10: 
Chris@10: * Added MPI distributed-memory transforms.  Compared to 3.3alpha,
Chris@10:   the major changes in the MPI transforms are:
Chris@10:     - Fixed some deadlock and crashing bugs.
Chris@10:     - Added Fortran 2003 interface.
Chris@10:     - Added new-array execute functions for MPI plans.
Chris@10:     - Eliminated use of large MPI tags, since Cray MPI requires tags < 2^24;
Chris@10:       thanks to Jonathan Bentz for the bug report.
Chris@10:     - Expanded documentation.
Chris@10:     - 'make check' now runs MPI tests
Chris@10:     - Some ABI changes - not binary-compatible with 3.3alpha MPI.
Chris@10: 
Chris@10: * Add support for quad-precision __float128 in gcc 4.6 or later (on x86.
Chris@10:   x86-64, and Itanium).  The new routines use the fftwq_ prefix.
Chris@10: 
Chris@10: * Removed support for MIPS paired-single instructions due to lack of
Chris@10:   available hardware for testing.  Users who want this functionality
Chris@10:   should continue using FFTW 3.2.x.  (Note that FFTW 3.3 still works
Chris@10:   on MIPS; this only concerns special instructions available on some
Chris@10:   MIPS chips.)
Chris@10: 
Chris@10: * Removed support for the Cell Broadband Engine.  Cell users should
Chris@10:   use FFTW 3.2.x.
Chris@10: 
Chris@10: * New convenience functions fftw_alloc_real and fftw_alloc_complex
Chris@10:   to use fftw_malloc for real and complex arrays without typecasts
Chris@10:   or sizeof.
Chris@10: 
Chris@10: * New convenience functions fftw_export_wisdom_to_filename and
Chris@10:   fftw_import_wisdom_from_filename that export/import wisdom
Chris@10:   to a file, which don't require you to open/close the file yourself.
Chris@10: 
Chris@10: * New function fftw_cost to return FFTW's internal cost metric for 
Chris@10:   a given plan; thanks to Rhys Ulerich and Nathanael Schaeffer for the
Chris@10:   suggestion.
Chris@10: 
Chris@10: * The --enable-sse2 configure flag now works in both double and single
Chris@10:   precision (and is equivalent to --enable-sse in the latter case).
Chris@10: 
Chris@10: * Remove --enable-portable-binary flag: we new produce portable binaries
Chris@10:   by default.
Chris@10: 
Chris@10: * Remove the automatic detection of native architecture flag for gcc
Chris@10:   which was introduced in fftw-3.1, since new gcc supports -mtune=native.
Chris@10:   Remove the --with-gcc-arch flag; if you want to specify a particlar
Chris@10:   arch to configure, use ./configure CC="gcc -mtune=...".
Chris@10: 
Chris@10: * --with-our-malloc16 configure flag is now renamed --with-our-malloc.
Chris@10: 
Chris@10: * Fixed build problem failure when srand48 declaration is missing;
Chris@10:   thanks to Ralf Wildenhues for the bug report.
Chris@10: 
Chris@10: * Fixed bug in fftw_set_timelimit: ensure that a negative timelimit
Chris@10:   is equivalent to no timelimit in all cases.  Thanks to William Andrew
Chris@10:   Burnson for the bug report.
Chris@10: 
Chris@10: * Fixed stack-overflow problem on OpenBSD caused by using alloca with
Chris@10:   too large a buffer.
Chris@10: 
Chris@10: FFTW 3.2.2
Chris@10: 
Chris@10: * Improve performance of some copy operations of complex arrays on
Chris@10:   x86 machines.
Chris@10: 
Chris@10: * Add configure flag to disable alloca(), which is broken in mingw64.
Chris@10: 
Chris@10: * Planning in FFTW_ESTIMATE mode for r2r transforms became slower
Chris@10:   between fftw-3.1.3 and 3.2.  This regression has now been fixed.
Chris@10: 
Chris@10: FFTW 3.2.1
Chris@10: 
Chris@10: * Performance improvements for some multidimensional r2c/c2r transforms;
Chris@10:   thanks to Eugene Miloslavsky for his benchmark reports.
Chris@10: 
Chris@10: * Compile with icc on MacOS X, use better icc compiler flags.
Chris@10: 
Chris@10: * Compilation fixes for systems where snprintf is defined as a macro;
Chris@10:   thanks to Marcus Mae for the bug report.
Chris@10: 
Chris@10: * Fortran documentation now recommends not using dfftw_execute,
Chris@10:   because of reports of problems with various Fortran compilers;
Chris@10:   it is better to use dfftw_execute_dft etcetera.
Chris@10: 
Chris@10: * Some documentation clarifications, e.g. of fact that --enable-openmp
Chris@10:   and --enable-threads are mutually exclusive (thanks to Long To),
Chris@10:   and document slightly odd behavior of plan_guru_r2r in Fortran
Chris@10:   (thanks to Alexander Pozdneev).
Chris@10: 
Chris@10: * FAQ was accidentally omitted from 3.2 tarball.
Chris@10: 
Chris@10: * Remove some extraneous (harmless) files accidentally included in 
Chris@10:   a subdirectory of the 3.2 tarball.
Chris@10: 
Chris@10: FFTW 3.2
Chris@10: 
Chris@10: * Worked around apparent glibc bug that leads to rare hangs when freeing
Chris@10:   semaphores.
Chris@10: 
Chris@10: * Fixed segfault due to unaligned access in certain obscure problems
Chris@10:   that use SSE and multiple threads.
Chris@10: 
Chris@10: * MPI transforms not included, as they are still in alpha; the alpha
Chris@10:   versions of the MPI transforms have been moved to FFTW 3.3alpha1.
Chris@10: 
Chris@10: FFTW 3.2alpha3
Chris@10: 
Chris@10: * Performance improvements for sizes with factors of 5 and 10.
Chris@10: 
Chris@10: * Documented FFTW_WISDOM_ONLY flag, at the suggestion of Mario
Chris@10:   Emmenlauer and Phil Dumont.
Chris@10: 
Chris@10: * Port Cell code to SDK2.1 (libspe2), as opposed to the old libspe1 code.
Chris@10: 
Chris@10: * Performance improvements in Cell code for N < 32k, thanks to Jan Wagner
Chris@10:   for the suggestions.
Chris@10: 
Chris@10: * Cycle counter for Sun x86_64 compiler, and compilation fix in cycle
Chris@10:   counter for AIX/xlc (thanks to Jeff Haferman for the bug report).
Chris@10: 
Chris@10: * Fixed incorrect type prefix in MPI code that prevented wisdom routines
Chris@10:   from working in single precision (thanks to Eric A. Borisch for the report).
Chris@10: 
Chris@10: * Added 'make check' for MPI code (which still fails in a couple corner
Chris@10:   cases, but should be much better than in alpha2).
Chris@10: 
Chris@10: * Many other small fixes.
Chris@10: 
Chris@10: FFTW 3.2alpha2
Chris@10: 
Chris@10: * Support for the Cell processor, donated by IBM Research; see README.Cell
Chris@10:   and the Cell section of the manual.
Chris@10: 
Chris@10: * New 64-bit API: for every "plan_guru" function there is a new "plan_guru64"
Chris@10:   function with the same semantics, but which takes fftw_iodim64 instead of
Chris@10:   fftw_iodim.  fftw_iodim64 is the same as fftw_iodim, except that it takes
Chris@10:   ptrdiff_t integer types as parameters, which is a 64-bit type on
Chris@10:   64-bit machines.  This is only useful for specifying very large transforms
Chris@10:   on 64-bit machines.  (Internally, FFTW uses ptrdiff_t everywhere
Chris@10:   regardless of what API you choose.)
Chris@10: 
Chris@10: * Experimental MPI support.  Complex one- and multi-dimensional FFTs,
Chris@10:   multi-dimensional r2r, multi-dimensional r2c/c2r transforms, and
Chris@10:   distributed transpose operations, with 1d block distributions.
Chris@10:   (This is an alpha preview: routines have not been exhaustively
Chris@10:   tested, documentation is incomplete, and some functionality is
Chris@10:   missing, e.g. Fortran support.)  See mpi/README and also the MPI
Chris@10:   section of the manual.
Chris@10: 
Chris@10: * Significantly faster r2c/c2r transforms, especially on machines with SIMD.
Chris@10: 
Chris@10: * Rewritten multi-threaded support for better performance by
Chris@10:   re-using a fixed pool of threads rather than continually
Chris@10:   respawning and joining (which nowadays is much slower).
Chris@10: 
Chris@10: * Support for MIPS paired-single SIMD instructions, donated by
Chris@10:   Codesourcery.
Chris@10: 
Chris@10: * FFTW_WISDOM_ONLY planner flag, to create plan only if wisdom is
Chris@10:   available and return NULL otherwise.
Chris@10: 
Chris@10: * Removed k7 support, which only worked in 32-bit mode and is
Chris@10:   becoming obsolete.  Use --enable-sse instead.
Chris@10: 
Chris@10: * Added --with-g77-wrappers configure option to force inclusion
Chris@10:   of g77 wrappers, in addition to whatever is needed for the
Chris@10:   detected Fortran compilers.  This is mainly intended for GNU/Linux
Chris@10:   distros switching to gfortran that wish to include both
Chris@10:   gfortran and g77 support in FFTW.
Chris@10: 
Chris@10: * In manual, renamed "guru execute" functions to "new-array execute"
Chris@10:   functions, to reduce confusion with the guru planner interface.
Chris@10:   (The programming interface is unchanged.)
Chris@10: 
Chris@10: * Add missing __declspec attribute to threads API functions when compiling
Chris@10:   for Windows; thanks to Robert O. Morris for the bug report.
Chris@10: 
Chris@10: * Fixed missing return value from dfftw_init_threads in Fortran;
Chris@10:   thanks to Markus Wetzstein for the bug report.
Chris@10: 
Chris@10: FFTW 3.1.1
Chris@10: 
Chris@10: * Performance improvements for Intel EMT64.
Chris@10: 
Chris@10: * Performance improvements for large-size transforms with SIMD.
Chris@10: 
Chris@10: * Cycle counter support for Intel icc and Visual C++ on x86-64.
Chris@10: 
Chris@10: * In fftw-wisdom tool, replaced obsolete --impatient with --measure.
Chris@10: 
Chris@10: * Fixed compilation failure with AIX/xlc; thanks to Joseph Thomas.
Chris@10: 
Chris@10: * Windows DLL support for Fortran API (added missing __declspec(dllexport)).
Chris@10: 
Chris@10: * SSE/SSE2 code works properly (i.e. disables itself) on older 386 and 486
Chris@10:   CPUs lacking a CPUID instruction; thanks to Eric Korpela.
Chris@10: 
Chris@10: FFTW 3.1
Chris@10: 
Chris@10: * Faster FFTW_ESTIMATE planner.
Chris@10: 
Chris@10: * New (faster) algorithm for REDFT00/RODFT00 (type-I DCT/DST) of odd size.
Chris@10: 
Chris@10: * "4-step" algorithm for faster FFTs of very large sizes (> 2^18).
Chris@10: 
Chris@10: * Faster in-place real-data DFTs (for R2HC and HC2R r2r formats).
Chris@10: 
Chris@10: * Faster in-place non-square transpositions (FFTW uses these internally
Chris@10:   for in-place FFTs, and you can also perform them explicitly using
Chris@10:   the guru interface).
Chris@10: 
Chris@10: * Faster prime-size DFTs: implemented Bluestein's algorithm, as well
Chris@10:   as a zero-padded Rader variant to limit recursive use of Rader's algorithm.
Chris@10: 
Chris@10: * SIMD support for split complex arrays.
Chris@10: 
Chris@10: * Much faster Altivec/VMX performance.
Chris@10: 
Chris@10: * New fftw_set_timelimit function to specify a (rough) upper bound to the
Chris@10:   planning time (does not affect ESTIMATE mode).
Chris@10: 
Chris@10: * Removed --enable-3dnow support; use --enable-k7 instead.
Chris@10: 
Chris@10: * FMA (fused multiply-add) version is now included in "standard" FFTW,
Chris@10:   and is enabled with --enable-fma (the default on PowerPC and Itanium).
Chris@10: 
Chris@10: * Automatic detection of native architecture flag for gcc.  New
Chris@10:   configure options: --enable-portable-binary and --with-gcc-arch=<arch>,
Chris@10:   for people distributing compiled binaries of FFTW (see manual).
Chris@10: 
Chris@10: * Automatic detection of Altivec under Linux with gcc 3.4 (so that
Chris@10:   same binary should work on both Altivec and non-Altivec PowerPCs).
Chris@10: 
Chris@10: * Compiler-specific tweaks/flags/workarounds for gcc 3.4, xlc, HP/UX,
Chris@10:   Solaris/Intel.
Chris@10: 
Chris@10: * Various documentation clarifications.
Chris@10: 
Chris@10: * 64-bit clean.  (Fixes a bug affecting the split guru planner on 
Chris@10:   64-bit machines, reported by David Necas.)
Chris@10: 
Chris@10: * Fixed Debian bug #259612: inadvertent use of SSE instructions on
Chris@10:   non-SSE machines (causing a crash) for --enable-sse binaries.
Chris@10: 
Chris@10: * Fixed bug that caused HC2R transforms to destroy the input in
Chris@10:   certain cases, even if the user specified FFTW_PRESERVE_INPUT.
Chris@10: 
Chris@10: * Fixed bug where wisdom would be lost under rare circumstances,
Chris@10:   causing excessive planning time.
Chris@10: 
Chris@10: * FAQ notes bug in gcc-3.4.[1-3] that causes FFTW to crash with SSE/SSE2.
Chris@10: 
Chris@10: * Fixed accidentally exported symbol that prohibited simultaneous
Chris@10:   linking to double/single multithreaded FFTW (thanks to Alessio Massaro).
Chris@10: 
Chris@10: * Support Win32 threads under MinGW (thanks to Alessio Massaro).
Chris@10: 
Chris@10: * Fixed problem with building DLL under Cygwin; thanks to Stephane Fillod.
Chris@10: 
Chris@10: * Fix build failure if no Fortran compiler is found (thanks to Charles
Chris@10:   Radley for the bug report).
Chris@10: 
Chris@10: * Fixed compilation failure with icc 8.0 and SSE/SSE2.  Automatic
Chris@10:   detection of icc architecture flag (e.g. -xW).
Chris@10: 
Chris@10: * Fixed compilation with OpenMP on AIX (thanks to Greg Bauer).
Chris@10: 
Chris@10: * Fixed compilation failure on x86-64 with gcc (thanks to Orion Poplawski).
Chris@10: 
Chris@10: * Incorporated patch from FreeBSD ports (FreeBSD does not have memalign,
Chris@10:   but its malloc is 16-byte aligned).
Chris@10: 
Chris@10: * Cycle-counter compilation fixes for Itanium, Alpha, x86-64, Sparc,
Chris@10:   MacOS (thanks to Matt Boman, John Bowman, and James A. Treacy for
Chris@10:   reports/fixes).  Added x86-64 cycle counter for PGI compilers,
Chris@10:   courtesy Cristiano Calonaci.
Chris@10: 
Chris@10: * Fix compilation problem in test program due to C99 conflict.
Chris@10: 
Chris@10: * Portability fix for import_system_wisdom with djgpp (thanks to Juan
Chris@10:   Manuel Guerrero).
Chris@10: 
Chris@10: * Fixed compilation failure on MacOS 10.3 due to getopt conflict.
Chris@10: 
Chris@10: * Work around Visual C++ (version 6/7) bug in SSE compilation;
Chris@10:   thanks to Eddie Yee for his detailed report.
Chris@10: 
Chris@10: Changes from FFTW 3.1 beta 2:
Chris@10: 
Chris@10: * Several minor compilation fixes.
Chris@10: 
Chris@10: * Eliminate FFTW_TIMELIMIT flag and replace fftw_timelimit global with
Chris@10:   fftw_set_timelimit function.  Make wisdom work with time-limited plans.
Chris@10: 
Chris@10: Changes from FFTW 3.1 beta 1:
Chris@10: 
Chris@10: * Fixes for creating DLLs under Windows; thanks to John Pavel for his feedback.
Chris@10: 
Chris@10: * Fixed more 64-bit problems, thanks to John Pavel for the bug report.
Chris@10: 
Chris@10: * Further speed improvements for Altivec/VMX.
Chris@10: 
Chris@10: * Further speed improvements for non-square transpositions.
Chris@10: 
Chris@10: * Many minor tweaks.
Chris@10: 
Chris@10: FFTW 3.0.1
Chris@10: 
Chris@10: * Some speed improvements in SIMD code.
Chris@10: 
Chris@10: * --without-cycle-counter option is removed.  If no cycle counter is found,
Chris@10:   then the estimator is always used.  A --with-slow-timer option is provided
Chris@10:   to force the use of lower-resolution timers.
Chris@10: 
Chris@10: * Several fixes for compilation under Visual C++, with help from Stefane Ruel.
Chris@10: 
Chris@10: * Added x86 cycle counter for Visual C++, with help from Morten Nissov.
Chris@10: 
Chris@10: * Added S390 cycle counter, courtesy of James Treacy.
Chris@10: 
Chris@10: * Added missing static keyword that prevented simultaneous linkage
Chris@10:   of different-precision versions; thanks to Rasmus Larsen for the bug report.
Chris@10: 
Chris@10: * Corrected accidental omission of f77_wisdom.f file; thanks to Alan Watson.
Chris@10: 
Chris@10: * Support -xopenmp flag for SunOS; thanks to John Lou for the bug report.
Chris@10: 
Chris@10: * Compilation with HP/UX cc requires -Wp,-H128000 flag to increase
Chris@10:   preprocessor limits; thanks to Peter Vouras for the bug report.
Chris@10: 
Chris@10: * Removed non-portable use of 'tempfile' in fftw-wisdom-to-conf script;
Chris@10:   thanks to Nicolas Decoster for the patch.
Chris@10: 
Chris@10: * Added 'make smallcheck' target in tests/ directory, at the request of
Chris@10:   James Treacy.
Chris@10: 
Chris@10: FFTW 3.0
Chris@10: 
Chris@10: Major goals of this release:
Chris@10: 
Chris@10: * Speed: often 20% or more faster than FFTW 2.x, even without SIMD (see below).
Chris@10: 
Chris@10: * Complete rewrite, to make it easier to add new algorithms and transforms.
Chris@10: 
Chris@10: * New API, to support more general semantics.
Chris@10: 
Chris@10: Other enhancements:
Chris@10: 
Chris@10: * SIMD acceleration on supporting CPUs (SSE, SSE2, 3DNow!, and AltiVec).
Chris@10:  (With special thanks to Franz Franchetti for many experimental prototypes
Chris@10:   and to Stefan Kral for the vectorizing generator from fftwgel.)
Chris@10: 
Chris@10: * True in-place 1d transforms of large sizes (as well as compressed
Chris@10:   twiddle tables for additional memory/cache savings).
Chris@10: 
Chris@10: * More arbitrary placement of real & imaginary data, e.g. including
Chris@10:   interleaved (as in FFTW 2.x) as well as separate real/imag arrays.
Chris@10: 
Chris@10: * Efficient prime-size transforms of real data.
Chris@10: 
Chris@10: * Multidimensional transforms can operate on a subset of a larger matrix,
Chris@10:   and/or transform selected dimensions of a multidimensional array.
Chris@10: 
Chris@10: * By popular demand, simultaneous linking to double precision (fftw),
Chris@10:   single precision (fftwf), and long-double precision (fftwl) versions
Chris@10:   of FFTW is now supported.
Chris@10: 
Chris@10: * Cycle counters (on all modern CPUs) are exploited to speed planning.
Chris@10: 
Chris@10: * Efficient transforms of real even/odd arrays, a.k.a. discrete
Chris@10:   cosine/sine transforms (types I-IV).  (Currently work via pre/post
Chris@10:   processing of real transforms, ala FFTPACK, so are not optimal.)
Chris@10: 
Chris@10: * DHTs (Discrete Hartley Transforms), again via post-processing
Chris@10:   of real transforms (and thus suboptimal, for now).
Chris@10: 
Chris@10: * Support for linking to just those parts of FFTW that you need,
Chris@10:   greatly reducing the size of statically linked programs when
Chris@10:   only a limited set of transform sizes/types are required.
Chris@10: 
Chris@10: * Canonical global wisdom file (/etc/fftw/wisdom) on Unix, along
Chris@10:   with a command-line tool (fftw-wisdom) to generate/update it.
Chris@10: 
Chris@10: * Fortran API can be used with both g77 and non-g77 compilers
Chris@10:   simultaneously.
Chris@10: 
Chris@10: * Multi-threaded version has optional OpenMP support.
Chris@10: 
Chris@10: * Authors' good looks have greatly improved with age.
Chris@10: 
Chris@10: Changes from 3.0beta3:
Chris@10: 
Chris@10: * Separate FMA distribution to better exploit fused multiply-add instructions
Chris@10:   on PowerPC (and possibly other) architectures.
Chris@10: 
Chris@10: * Performance improvements via some inlining tweaks.
Chris@10: 
Chris@10: * fftw_flops now returns double arguments, not int, to avoid overflows
Chris@10:   for large sizes.
Chris@10: 
Chris@10: * Workarounds for automake bugs.
Chris@10: 
Chris@10: Changes from 3.0beta2:
Chris@10: 
Chris@10: * The standard REDFT00/RODFT00 (DCT-I/DST-I) algorithm (used in
Chris@10:   FFTPACK, NR, etcetera) turns out to have poor numerical accuracy, so
Chris@10:   we replaced it with a slower routine that is more accurate.
Chris@10: 
Chris@10: * The guru planner and execute functions now have two variants, one that
Chris@10:   takes complex arguments and one that takes separate real/imag pointers.
Chris@10: 
Chris@10: * Execute and planner routines now automatically align the stack on x86,
Chris@10:   in case the calling program is misaligned.
Chris@10: 
Chris@10: * README file for test program.
Chris@10: 
Chris@10: * Fixed bugs in the combination of SIMD with multi-threaded transforms.
Chris@10: 
Chris@10: * Eliminated internal fftw_threads_init function, which some people were
Chris@10:   calling accidentally instead of the fftw_init_threads API function.
Chris@10: 
Chris@10: * Check for -openmp flag (Intel C compiler) when --enable-openmp is used.
Chris@10: 
Chris@10: * Support AMD x86-64 SIMD and cycle counter.
Chris@10: 
Chris@10: * Support SSE2 intrinsics in forthcoming gcc 3.3.
Chris@10: 
Chris@10: Changes from 3.0beta1:
Chris@10: 
Chris@10: * Faster in-place 1d transforms of non-power-of-two sizes.
Chris@10: 
Chris@10: * SIMD improvements for in-place, multi-dimensional, and/or non-FFTW_PATIENT
Chris@10:   transforms.
Chris@10: 
Chris@10: * Added support for hard-coded DCT/DST/DHT codelets of small sizes; the
Chris@10:   default distribution only includes hard-coded size-8 DCT-II/III, however.
Chris@10: 
Chris@10: * Many minor improvements to the manual.  Added section on using the
Chris@10:   codelet generator to customize and enhance FFTW.
Chris@10: 
Chris@10: * The default 'make check' should now only take a few minutes; for more
Chris@10:   strenuous tests (which may take a day or so), do 'cd tests; make bigcheck'.
Chris@10: 
Chris@10: * fftw_print_plan is split into fftw_fprint_plan and fftw_print_plan, where
Chris@10:   the latter uses stdout.
Chris@10: 
Chris@10: * Fixed ability to compile with a C++ compiler.
Chris@10: 
Chris@10: * Fixed support for C99 complex type under glibc.
Chris@10: 
Chris@10: * Fixed problems with alloca under MinGW, AIX.
Chris@10: 
Chris@10: * Workaround for gcc/SPARC bug.
Chris@10: 
Chris@10: * Fixed multi-threaded initialization failure on IRIX due to lack of
Chris@10:   user-accessible PTHREAD_SCOPE_SYSTEM there.