|
Chris@82
|
1 FFTW 3.3.8:
|
|
Chris@82
|
2
|
|
Chris@82
|
3 * Fixed AVX, AVX2 for gcc-8.
|
|
Chris@82
|
4
|
|
Chris@82
|
5 By default, FFTW 3.3.7 was broken with gcc-8. AVX and AVX2 code
|
|
Chris@82
|
6 assumed that the compiler honors the distinction between +0 and -0,
|
|
Chris@82
|
7 but gcc-8 -ffast-math does not. The default CFLAGS included -ffast-math.
|
|
Chris@82
|
8 This release ensures that FFTW works with gcc-8 -ffast-math, and
|
|
Chris@82
|
9 removes -ffast-math from the default CFLAGS for good measure.
|
|
Chris@82
|
10
|
|
Chris@82
|
11 FFTW 3.3.7:
|
|
Chris@82
|
12
|
|
Chris@82
|
13 * Experimental support for CMake.
|
|
Chris@82
|
14
|
|
Chris@82
|
15 The primary build mechanism for FFTW remains GNU autoconf/automake.
|
|
Chris@82
|
16 CMake support is meant to offer an easy way to compile FFTW on
|
|
Chris@82
|
17 Windows, and as such it does not cover all the features of the
|
|
Chris@82
|
18 automake build system, such as exotic cycle counters,
|
|
Chris@82
|
19 cross-compiling, or build of binaries for a mixture of ISA's
|
|
Chris@82
|
20 (e.g., amd64 vs amd64+avx vs amd64+avx2). Patches are welcome.
|
|
Chris@82
|
21
|
|
Chris@82
|
22 * Fixes for armv7a cycle counter.
|
|
Chris@82
|
23 * Official support for aarch64, now that we have hardware to test it.
|
|
Chris@82
|
24 * Tweak usage of FMA instructions in a way that favors newer processors
|
|
Chris@82
|
25 (Skylake and Ryzen) over older processors (Haswell).
|
|
Chris@82
|
26 * tests/bench: use 64-bit precision to compute mflops.
|
|
Chris@82
|
27
|
|
Chris@82
|
28 FFTW 3.3.6-pl2:
|
|
Chris@82
|
29
|
|
Chris@82
|
30 * Bugfix: MPI Fortran-03 headers were missing in FFTW 3.3.6-pl1.
|
|
Chris@82
|
31
|
|
Chris@82
|
32 FFTW 3.3.6-pl1:
|
|
Chris@82
|
33
|
|
Chris@82
|
34 * Bugfix: FFTW 3.3.6 had the wrong libtool version number, and generated
|
|
Chris@82
|
35 shared libraries of the form libfftw3.so.2.6.6 instead of
|
|
Chris@82
|
36 libfftw3.so.3.*.
|
|
Chris@82
|
37
|
|
Chris@82
|
38 FFTW 3.3.6:
|
|
Chris@82
|
39
|
|
Chris@82
|
40 * The fftw_make_planner_thread_safe() API introduced in 3.3.5 didn't
|
|
Chris@82
|
41 work, and this 3.3.6 fixes it. Sorry about that.
|
|
Chris@82
|
42 * compilation fixes for IBM XLC
|
|
Chris@82
|
43 * compilation fixes for threads on Windows
|
|
Chris@82
|
44 * fix SIMD autodetection on amd64 when (_MSC_VER > 1500)
|
|
Chris@82
|
45
|
|
Chris@82
|
46 FFTW 3.3.5:
|
|
Chris@82
|
47
|
|
Chris@82
|
48 * New SIMD support:
|
|
Chris@82
|
49 - Power8 VSX instructions in single and double precision.
|
|
Chris@82
|
50 To use, add --enable-vsx to configure.
|
|
Chris@82
|
51 - Support for AVX2 (256-bit FMA instructions).
|
|
Chris@82
|
52 To use, add --enable-avx2 to configure.
|
|
Chris@82
|
53 - Experimental support for AVX512 and KCVI. (--enable-avx512, --enable-kcvi)
|
|
Chris@82
|
54 This code is expected to work but the FFTW maintainers do not have
|
|
Chris@82
|
55 hardware to test it.
|
|
Chris@82
|
56 - Support for AVX128/FMA (for some AMD machines) (--enable-avx128-fma)
|
|
Chris@82
|
57 - Double precision Neon SIMD for aarch64.
|
|
Chris@82
|
58 This code is expected to work but the FFTW maintainers do not have
|
|
Chris@82
|
59 hardware to test it.
|
|
Chris@82
|
60 - generic SIMD support using gcc vector intrinsics
|
|
Chris@82
|
61 * Add fftw_make_planner_thread_safe() API
|
|
Chris@82
|
62 * fix #18 (disable float128 for CUDACC)
|
|
Chris@82
|
63 * fix #19: missing Fortran interface for fftwq_alloc_real
|
|
Chris@82
|
64 * fix #21 (don't use float128 on Portland compilers, which pretend to be gcc)
|
|
Chris@82
|
65 * fix: Avoid segfaults due to double free in MPI transpose
|
|
Chris@82
|
66
|
|
Chris@82
|
67 * Special note for distribution maintainers: Although FFTW supports a
|
|
Chris@82
|
68 zillion SIMD instruction sets, enabling them all at the same time is
|
|
Chris@82
|
69 a bad idea, because it increases the planning time for minimal gain.
|
|
Chris@82
|
70 We recommend that general-purpose x86 distributions only enable SSE2
|
|
Chris@82
|
71 and perhaps AVX. Users who care about the last ounce of performance
|
|
Chris@82
|
72 should recompile FFTW themselves.
|
|
Chris@82
|
73
|
|
Chris@82
|
74 FFTW 3.3.4
|
|
Chris@82
|
75
|
|
Chris@82
|
76 * New functions fftw_alignment_of (to check whether two arrays are
|
|
Chris@82
|
77 equally aligned for the purposes of applying a plan) and fftw_sprint_plan
|
|
Chris@82
|
78 (to output a description of plan to a string).
|
|
Chris@82
|
79
|
|
Chris@82
|
80 * Bugfix in fftw-wisdom-to-conf; thanks to Florian Oppermann for the
|
|
Chris@82
|
81 bug report.
|
|
Chris@82
|
82
|
|
Chris@82
|
83 * Fixed manual to work with texinfo-5.
|
|
Chris@82
|
84
|
|
Chris@82
|
85 * Increased timing interval on x86_64 to reduce timing errors.
|
|
Chris@82
|
86
|
|
Chris@82
|
87 * Default to Win32 threads, not pthreads, if both are present.
|
|
Chris@82
|
88
|
|
Chris@82
|
89 * Various build-script fixes.
|
|
Chris@82
|
90
|
|
Chris@82
|
91 FFTW 3.3.3
|
|
Chris@82
|
92
|
|
Chris@82
|
93 * Fix deadlock bug in MPI transforms (thanks to Michael Pippig for the
|
|
Chris@82
|
94 bug report and patch, and to Graham Dennis for the bug report).
|
|
Chris@82
|
95
|
|
Chris@82
|
96 * Use 128-bit ARM NEON instructions instead of 64-bits. This change
|
|
Chris@82
|
97 appears to speed up even ARM processors with a 64-bit NEON pipe.
|
|
Chris@82
|
98
|
|
Chris@82
|
99 * Speed improvements for single-precision AVX.
|
|
Chris@82
|
100
|
|
Chris@82
|
101 * Speed up planner on machines without "official" cycle counters, such as ARM.
|
|
Chris@82
|
102
|
|
Chris@82
|
103 FFTW 3.3.2
|
|
Chris@82
|
104
|
|
Chris@82
|
105 * Removed an archaic stack-alignment hack that was failing with
|
|
Chris@82
|
106 gcc-4.7/i386.
|
|
Chris@82
|
107
|
|
Chris@82
|
108 * Added stack-alignment hack necessary for gcc on Windows/i386. We
|
|
Chris@82
|
109 will regret this in ten years (see previous change).
|
|
Chris@82
|
110
|
|
Chris@82
|
111 * Fix incompatibility with Intel icc which pretends to be gcc
|
|
Chris@82
|
112 but does not support quad precision.
|
|
Chris@82
|
113
|
|
Chris@82
|
114 * make libfftw{threads,mpi} depend upon libfftw when using libtool;
|
|
Chris@82
|
115 this is consistent with most other libraries and simplifies the life
|
|
Chris@82
|
116 of various distributors of GNU/Linux.
|
|
Chris@82
|
117
|
|
Chris@82
|
118 FFTW 3.3.1
|
|
Chris@82
|
119
|
|
Chris@82
|
120 * Changes since 3.3.1-beta1:
|
|
Chris@82
|
121
|
|
Chris@82
|
122 - Reduced planning time in estimate mode for sizes with large
|
|
Chris@82
|
123 prime factors.
|
|
Chris@82
|
124
|
|
Chris@82
|
125 - Added AVX autodetection under Visual Studio. Thanks Carsten
|
|
Chris@82
|
126 Steger for submitting the necessary code.
|
|
Chris@82
|
127
|
|
Chris@82
|
128 - Modern Fortran interface now uses a separate fftw3l.f03 interface
|
|
Chris@82
|
129 file for the long double interface, which is not supported by
|
|
Chris@82
|
130 some Fortran compilers. Provided new fftw3q.f03 interface file
|
|
Chris@82
|
131 to access the quadruple-precision FFTW routines with recent
|
|
Chris@82
|
132 versions of gcc/gfortran.
|
|
Chris@82
|
133
|
|
Chris@82
|
134 * Added support for the NEON extensions to the ARM ISA. (Note to beta
|
|
Chris@82
|
135 users: an ARM cycle counter is not yet implemented; please contact
|
|
Chris@82
|
136 fftw@fftw.org if you know how to do it right.)
|
|
Chris@82
|
137
|
|
Chris@82
|
138 * MPI code now compiles even if mpicc is a C++ compiler; thanks to
|
|
Chris@82
|
139 Kyle Spyksma for the bug report.
|
|
Chris@82
|
140
|
|
Chris@82
|
141 FFTW 3.3
|
|
Chris@82
|
142
|
|
Chris@82
|
143 * Changes since 3.3-beta1:
|
|
Chris@82
|
144
|
|
Chris@82
|
145 - Compiling OpenMP support (--enable-openmp) now installs a
|
|
Chris@82
|
146 fftw3_omp library, instead of fftw3_threads, so that OpenMP
|
|
Chris@82
|
147 and POSIX threads (--enable-threads) libraries can be built
|
|
Chris@82
|
148 and installed at the same time.
|
|
Chris@82
|
149
|
|
Chris@82
|
150 - Various minor compilation fixes, corrections of manual typos, and
|
|
Chris@82
|
151 improvements to the benchmark test program.
|
|
Chris@82
|
152
|
|
Chris@82
|
153 * Add support for the AVX extensions to x86 and x86-64. The AVX code
|
|
Chris@82
|
154 works with 16-byte alignment (as opposed to 32-byte alignment),
|
|
Chris@82
|
155 so there is no ABI change compared to FFTW 3.2.2.
|
|
Chris@82
|
156
|
|
Chris@82
|
157 * Added Fortran 2003 interface, which should be usable on most modern
|
|
Chris@82
|
158 Fortran compilers (e.g. gfortran) and provides type-checked access
|
|
Chris@82
|
159 to the the C FFTW interface. (The legacy Fortran-77 interface is
|
|
Chris@82
|
160 still included also.)
|
|
Chris@82
|
161
|
|
Chris@82
|
162 * Added MPI distributed-memory transforms. Compared to 3.3alpha,
|
|
Chris@82
|
163 the major changes in the MPI transforms are:
|
|
Chris@82
|
164 - Fixed some deadlock and crashing bugs.
|
|
Chris@82
|
165 - Added Fortran 2003 interface.
|
|
Chris@82
|
166 - Added new-array execute functions for MPI plans.
|
|
Chris@82
|
167 - Eliminated use of large MPI tags, since Cray MPI requires tags < 2^24;
|
|
Chris@82
|
168 thanks to Jonathan Bentz for the bug report.
|
|
Chris@82
|
169 - Expanded documentation.
|
|
Chris@82
|
170 - 'make check' now runs MPI tests
|
|
Chris@82
|
171 - Some ABI changes - not binary-compatible with 3.3alpha MPI.
|
|
Chris@82
|
172
|
|
Chris@82
|
173 * Add support for quad-precision __float128 in gcc 4.6 or later (on x86.
|
|
Chris@82
|
174 x86-64, and Itanium). The new routines use the fftwq_ prefix.
|
|
Chris@82
|
175
|
|
Chris@82
|
176 * Removed support for MIPS paired-single instructions due to lack of
|
|
Chris@82
|
177 available hardware for testing. Users who want this functionality
|
|
Chris@82
|
178 should continue using FFTW 3.2.x. (Note that FFTW 3.3 still works
|
|
Chris@82
|
179 on MIPS; this only concerns special instructions available on some
|
|
Chris@82
|
180 MIPS chips.)
|
|
Chris@82
|
181
|
|
Chris@82
|
182 * Removed support for the Cell Broadband Engine. Cell users should
|
|
Chris@82
|
183 use FFTW 3.2.x.
|
|
Chris@82
|
184
|
|
Chris@82
|
185 * New convenience functions fftw_alloc_real and fftw_alloc_complex
|
|
Chris@82
|
186 to use fftw_malloc for real and complex arrays without typecasts
|
|
Chris@82
|
187 or sizeof.
|
|
Chris@82
|
188
|
|
Chris@82
|
189 * New convenience functions fftw_export_wisdom_to_filename and
|
|
Chris@82
|
190 fftw_import_wisdom_from_filename that export/import wisdom
|
|
Chris@82
|
191 to a file, which don't require you to open/close the file yourself.
|
|
Chris@82
|
192
|
|
Chris@82
|
193 * New function fftw_cost to return FFTW's internal cost metric for
|
|
Chris@82
|
194 a given plan; thanks to Rhys Ulerich and Nathanael Schaeffer for the
|
|
Chris@82
|
195 suggestion.
|
|
Chris@82
|
196
|
|
Chris@82
|
197 * The --enable-sse2 configure flag now works in both double and single
|
|
Chris@82
|
198 precision (and is equivalent to --enable-sse in the latter case).
|
|
Chris@82
|
199
|
|
Chris@82
|
200 * Remove --enable-portable-binary flag: we new produce portable binaries
|
|
Chris@82
|
201 by default.
|
|
Chris@82
|
202
|
|
Chris@82
|
203 * Remove the automatic detection of native architecture flag for gcc
|
|
Chris@82
|
204 which was introduced in fftw-3.1, since new gcc supports -mtune=native.
|
|
Chris@82
|
205 Remove the --with-gcc-arch flag; if you want to specify a particlar
|
|
Chris@82
|
206 arch to configure, use ./configure CC="gcc -mtune=...".
|
|
Chris@82
|
207
|
|
Chris@82
|
208 * --with-our-malloc16 configure flag is now renamed --with-our-malloc.
|
|
Chris@82
|
209
|
|
Chris@82
|
210 * Fixed build problem failure when srand48 declaration is missing;
|
|
Chris@82
|
211 thanks to Ralf Wildenhues for the bug report.
|
|
Chris@82
|
212
|
|
Chris@82
|
213 * Fixed bug in fftw_set_timelimit: ensure that a negative timelimit
|
|
Chris@82
|
214 is equivalent to no timelimit in all cases. Thanks to William Andrew
|
|
Chris@82
|
215 Burnson for the bug report.
|
|
Chris@82
|
216
|
|
Chris@82
|
217 * Fixed stack-overflow problem on OpenBSD caused by using alloca with
|
|
Chris@82
|
218 too large a buffer.
|
|
Chris@82
|
219
|
|
Chris@82
|
220 FFTW 3.2.2
|
|
Chris@82
|
221
|
|
Chris@82
|
222 * Improve performance of some copy operations of complex arrays on
|
|
Chris@82
|
223 x86 machines.
|
|
Chris@82
|
224
|
|
Chris@82
|
225 * Add configure flag to disable alloca(), which is broken in mingw64.
|
|
Chris@82
|
226
|
|
Chris@82
|
227 * Planning in FFTW_ESTIMATE mode for r2r transforms became slower
|
|
Chris@82
|
228 between fftw-3.1.3 and 3.2. This regression has now been fixed.
|
|
Chris@82
|
229
|
|
Chris@82
|
230 FFTW 3.2.1
|
|
Chris@82
|
231
|
|
Chris@82
|
232 * Performance improvements for some multidimensional r2c/c2r transforms;
|
|
Chris@82
|
233 thanks to Eugene Miloslavsky for his benchmark reports.
|
|
Chris@82
|
234
|
|
Chris@82
|
235 * Compile with icc on MacOS X, use better icc compiler flags.
|
|
Chris@82
|
236
|
|
Chris@82
|
237 * Compilation fixes for systems where snprintf is defined as a macro;
|
|
Chris@82
|
238 thanks to Marcus Mae for the bug report.
|
|
Chris@82
|
239
|
|
Chris@82
|
240 * Fortran documentation now recommends not using dfftw_execute,
|
|
Chris@82
|
241 because of reports of problems with various Fortran compilers;
|
|
Chris@82
|
242 it is better to use dfftw_execute_dft etcetera.
|
|
Chris@82
|
243
|
|
Chris@82
|
244 * Some documentation clarifications, e.g. of fact that --enable-openmp
|
|
Chris@82
|
245 and --enable-threads are mutually exclusive (thanks to Long To),
|
|
Chris@82
|
246 and document slightly odd behavior of plan_guru_r2r in Fortran
|
|
Chris@82
|
247 (thanks to Alexander Pozdneev).
|
|
Chris@82
|
248
|
|
Chris@82
|
249 * FAQ was accidentally omitted from 3.2 tarball.
|
|
Chris@82
|
250
|
|
Chris@82
|
251 * Remove some extraneous (harmless) files accidentally included in
|
|
Chris@82
|
252 a subdirectory of the 3.2 tarball.
|
|
Chris@82
|
253
|
|
Chris@82
|
254 FFTW 3.2
|
|
Chris@82
|
255
|
|
Chris@82
|
256 * Worked around apparent glibc bug that leads to rare hangs when freeing
|
|
Chris@82
|
257 semaphores.
|
|
Chris@82
|
258
|
|
Chris@82
|
259 * Fixed segfault due to unaligned access in certain obscure problems
|
|
Chris@82
|
260 that use SSE and multiple threads.
|
|
Chris@82
|
261
|
|
Chris@82
|
262 * MPI transforms not included, as they are still in alpha; the alpha
|
|
Chris@82
|
263 versions of the MPI transforms have been moved to FFTW 3.3alpha1.
|
|
Chris@82
|
264
|
|
Chris@82
|
265 FFTW 3.2alpha3
|
|
Chris@82
|
266
|
|
Chris@82
|
267 * Performance improvements for sizes with factors of 5 and 10.
|
|
Chris@82
|
268
|
|
Chris@82
|
269 * Documented FFTW_WISDOM_ONLY flag, at the suggestion of Mario
|
|
Chris@82
|
270 Emmenlauer and Phil Dumont.
|
|
Chris@82
|
271
|
|
Chris@82
|
272 * Port Cell code to SDK2.1 (libspe2), as opposed to the old libspe1 code.
|
|
Chris@82
|
273
|
|
Chris@82
|
274 * Performance improvements in Cell code for N < 32k, thanks to Jan Wagner
|
|
Chris@82
|
275 for the suggestions.
|
|
Chris@82
|
276
|
|
Chris@82
|
277 * Cycle counter for Sun x86_64 compiler, and compilation fix in cycle
|
|
Chris@82
|
278 counter for AIX/xlc (thanks to Jeff Haferman for the bug report).
|
|
Chris@82
|
279
|
|
Chris@82
|
280 * Fixed incorrect type prefix in MPI code that prevented wisdom routines
|
|
Chris@82
|
281 from working in single precision (thanks to Eric A. Borisch for the report).
|
|
Chris@82
|
282
|
|
Chris@82
|
283 * Added 'make check' for MPI code (which still fails in a couple corner
|
|
Chris@82
|
284 cases, but should be much better than in alpha2).
|
|
Chris@82
|
285
|
|
Chris@82
|
286 * Many other small fixes.
|
|
Chris@82
|
287
|
|
Chris@82
|
288 FFTW 3.2alpha2
|
|
Chris@82
|
289
|
|
Chris@82
|
290 * Support for the Cell processor, donated by IBM Research; see README.Cell
|
|
Chris@82
|
291 and the Cell section of the manual.
|
|
Chris@82
|
292
|
|
Chris@82
|
293 * New 64-bit API: for every "plan_guru" function there is a new "plan_guru64"
|
|
Chris@82
|
294 function with the same semantics, but which takes fftw_iodim64 instead of
|
|
Chris@82
|
295 fftw_iodim. fftw_iodim64 is the same as fftw_iodim, except that it takes
|
|
Chris@82
|
296 ptrdiff_t integer types as parameters, which is a 64-bit type on
|
|
Chris@82
|
297 64-bit machines. This is only useful for specifying very large transforms
|
|
Chris@82
|
298 on 64-bit machines. (Internally, FFTW uses ptrdiff_t everywhere
|
|
Chris@82
|
299 regardless of what API you choose.)
|
|
Chris@82
|
300
|
|
Chris@82
|
301 * Experimental MPI support. Complex one- and multi-dimensional FFTs,
|
|
Chris@82
|
302 multi-dimensional r2r, multi-dimensional r2c/c2r transforms, and
|
|
Chris@82
|
303 distributed transpose operations, with 1d block distributions.
|
|
Chris@82
|
304 (This is an alpha preview: routines have not been exhaustively
|
|
Chris@82
|
305 tested, documentation is incomplete, and some functionality is
|
|
Chris@82
|
306 missing, e.g. Fortran support.) See mpi/README and also the MPI
|
|
Chris@82
|
307 section of the manual.
|
|
Chris@82
|
308
|
|
Chris@82
|
309 * Significantly faster r2c/c2r transforms, especially on machines with SIMD.
|
|
Chris@82
|
310
|
|
Chris@82
|
311 * Rewritten multi-threaded support for better performance by
|
|
Chris@82
|
312 re-using a fixed pool of threads rather than continually
|
|
Chris@82
|
313 respawning and joining (which nowadays is much slower).
|
|
Chris@82
|
314
|
|
Chris@82
|
315 * Support for MIPS paired-single SIMD instructions, donated by
|
|
Chris@82
|
316 Codesourcery.
|
|
Chris@82
|
317
|
|
Chris@82
|
318 * FFTW_WISDOM_ONLY planner flag, to create plan only if wisdom is
|
|
Chris@82
|
319 available and return NULL otherwise.
|
|
Chris@82
|
320
|
|
Chris@82
|
321 * Removed k7 support, which only worked in 32-bit mode and is
|
|
Chris@82
|
322 becoming obsolete. Use --enable-sse instead.
|
|
Chris@82
|
323
|
|
Chris@82
|
324 * Added --with-g77-wrappers configure option to force inclusion
|
|
Chris@82
|
325 of g77 wrappers, in addition to whatever is needed for the
|
|
Chris@82
|
326 detected Fortran compilers. This is mainly intended for GNU/Linux
|
|
Chris@82
|
327 distros switching to gfortran that wish to include both
|
|
Chris@82
|
328 gfortran and g77 support in FFTW.
|
|
Chris@82
|
329
|
|
Chris@82
|
330 * In manual, renamed "guru execute" functions to "new-array execute"
|
|
Chris@82
|
331 functions, to reduce confusion with the guru planner interface.
|
|
Chris@82
|
332 (The programming interface is unchanged.)
|
|
Chris@82
|
333
|
|
Chris@82
|
334 * Add missing __declspec attribute to threads API functions when compiling
|
|
Chris@82
|
335 for Windows; thanks to Robert O. Morris for the bug report.
|
|
Chris@82
|
336
|
|
Chris@82
|
337 * Fixed missing return value from dfftw_init_threads in Fortran;
|
|
Chris@82
|
338 thanks to Markus Wetzstein for the bug report.
|
|
Chris@82
|
339
|
|
Chris@82
|
340 FFTW 3.1.3
|
|
Chris@82
|
341
|
|
Chris@82
|
342 * Bug fix: FFTW computes incorrect results when the user plans both
|
|
Chris@82
|
343 REDFT11 and RODFT11 transforms of certain sizes. The bug is caused
|
|
Chris@82
|
344 by incorrect sharing of twiddle-factor tables between the two
|
|
Chris@82
|
345 transforms, and only occurs when both are used. Thanks to Paul
|
|
Chris@82
|
346 A. Valiant for the bug report.
|
|
Chris@82
|
347
|
|
Chris@82
|
348 FFTW 3.1.2
|
|
Chris@82
|
349
|
|
Chris@82
|
350 * Correct bug in configure script: --enable-portable-binary option was ignored!
|
|
Chris@82
|
351 Thanks to Andrew Salamon for the bug report.
|
|
Chris@82
|
352
|
|
Chris@82
|
353 * Threads compilation fix on AIX: prefer xlc_r to cc_r, and don't use
|
|
Chris@82
|
354 either if we are using gcc. Thanks to Guy Moebs for the bug report.
|
|
Chris@82
|
355
|
|
Chris@82
|
356 * Updated FAQ to note that Apple gcc 4.0.1 on MacOS/Intel is broken,
|
|
Chris@82
|
357 and suggest a workaround. configure script now detects Core/Duo arch.
|
|
Chris@82
|
358
|
|
Chris@82
|
359 * Use -maltivec when checking for altivec.h. Fixes Gentoo bug #129304,
|
|
Chris@82
|
360 thanks to Markus Dittrich.
|
|
Chris@82
|
361
|
|
Chris@82
|
362 FFTW 3.1.1
|
|
Chris@82
|
363
|
|
Chris@82
|
364 * Performance improvements for Intel EMT64.
|
|
Chris@82
|
365
|
|
Chris@82
|
366 * Performance improvements for large-size transforms with SIMD.
|
|
Chris@82
|
367
|
|
Chris@82
|
368 * Cycle counter support for Intel icc and Visual C++ on x86-64.
|
|
Chris@82
|
369
|
|
Chris@82
|
370 * In fftw-wisdom tool, replaced obsolete --impatient with --measure.
|
|
Chris@82
|
371
|
|
Chris@82
|
372 * Fixed compilation failure with AIX/xlc; thanks to Joseph Thomas.
|
|
Chris@82
|
373
|
|
Chris@82
|
374 * Windows DLL support for Fortran API (added missing __declspec(dllexport)).
|
|
Chris@82
|
375
|
|
Chris@82
|
376 * SSE/SSE2 code works properly (i.e. disables itself) on older 386 and 486
|
|
Chris@82
|
377 CPUs lacking a CPUID instruction; thanks to Eric Korpela.
|
|
Chris@82
|
378
|
|
Chris@82
|
379 FFTW 3.1
|
|
Chris@82
|
380
|
|
Chris@82
|
381 * Faster FFTW_ESTIMATE planner.
|
|
Chris@82
|
382
|
|
Chris@82
|
383 * New (faster) algorithm for REDFT00/RODFT00 (type-I DCT/DST) of odd size.
|
|
Chris@82
|
384
|
|
Chris@82
|
385 * "4-step" algorithm for faster FFTs of very large sizes (> 2^18).
|
|
Chris@82
|
386
|
|
Chris@82
|
387 * Faster in-place real-data DFTs (for R2HC and HC2R r2r formats).
|
|
Chris@82
|
388
|
|
Chris@82
|
389 * Faster in-place non-square transpositions (FFTW uses these internally
|
|
Chris@82
|
390 for in-place FFTs, and you can also perform them explicitly using
|
|
Chris@82
|
391 the guru interface).
|
|
Chris@82
|
392
|
|
Chris@82
|
393 * Faster prime-size DFTs: implemented Bluestein's algorithm, as well
|
|
Chris@82
|
394 as a zero-padded Rader variant to limit recursive use of Rader's algorithm.
|
|
Chris@82
|
395
|
|
Chris@82
|
396 * SIMD support for split complex arrays.
|
|
Chris@82
|
397
|
|
Chris@82
|
398 * Much faster Altivec/VMX performance.
|
|
Chris@82
|
399
|
|
Chris@82
|
400 * New fftw_set_timelimit function to specify a (rough) upper bound to the
|
|
Chris@82
|
401 planning time (does not affect ESTIMATE mode).
|
|
Chris@82
|
402
|
|
Chris@82
|
403 * Removed --enable-3dnow support; use --enable-k7 instead.
|
|
Chris@82
|
404
|
|
Chris@82
|
405 * FMA (fused multiply-add) version is now included in "standard" FFTW,
|
|
Chris@82
|
406 and is enabled with --enable-fma (the default on PowerPC and Itanium).
|
|
Chris@82
|
407
|
|
Chris@82
|
408 * Automatic detection of native architecture flag for gcc. New
|
|
Chris@82
|
409 configure options: --enable-portable-binary and --with-gcc-arch=<arch>,
|
|
Chris@82
|
410 for people distributing compiled binaries of FFTW (see manual).
|
|
Chris@82
|
411
|
|
Chris@82
|
412 * Automatic detection of Altivec under Linux with gcc 3.4 (so that
|
|
Chris@82
|
413 same binary should work on both Altivec and non-Altivec PowerPCs).
|
|
Chris@82
|
414
|
|
Chris@82
|
415 * Compiler-specific tweaks/flags/workarounds for gcc 3.4, xlc, HP/UX,
|
|
Chris@82
|
416 Solaris/Intel.
|
|
Chris@82
|
417
|
|
Chris@82
|
418 * Various documentation clarifications.
|
|
Chris@82
|
419
|
|
Chris@82
|
420 * 64-bit clean. (Fixes a bug affecting the split guru planner on
|
|
Chris@82
|
421 64-bit machines, reported by David Necas.)
|
|
Chris@82
|
422
|
|
Chris@82
|
423 * Fixed Debian bug #259612: inadvertent use of SSE instructions on
|
|
Chris@82
|
424 non-SSE machines (causing a crash) for --enable-sse binaries.
|
|
Chris@82
|
425
|
|
Chris@82
|
426 * Fixed bug that caused HC2R transforms to destroy the input in
|
|
Chris@82
|
427 certain cases, even if the user specified FFTW_PRESERVE_INPUT.
|
|
Chris@82
|
428
|
|
Chris@82
|
429 * Fixed bug where wisdom would be lost under rare circumstances,
|
|
Chris@82
|
430 causing excessive planning time.
|
|
Chris@82
|
431
|
|
Chris@82
|
432 * FAQ notes bug in gcc-3.4.[1-3] that causes FFTW to crash with SSE/SSE2.
|
|
Chris@82
|
433
|
|
Chris@82
|
434 * Fixed accidentally exported symbol that prohibited simultaneous
|
|
Chris@82
|
435 linking to double/single multithreaded FFTW (thanks to Alessio Massaro).
|
|
Chris@82
|
436
|
|
Chris@82
|
437 * Support Win32 threads under MinGW (thanks to Alessio Massaro).
|
|
Chris@82
|
438
|
|
Chris@82
|
439 * Fixed problem with building DLL under Cygwin; thanks to Stephane Fillod.
|
|
Chris@82
|
440
|
|
Chris@82
|
441 * Fix build failure if no Fortran compiler is found (thanks to Charles
|
|
Chris@82
|
442 Radley for the bug report).
|
|
Chris@82
|
443
|
|
Chris@82
|
444 * Fixed compilation failure with icc 8.0 and SSE/SSE2. Automatic
|
|
Chris@82
|
445 detection of icc architecture flag (e.g. -xW).
|
|
Chris@82
|
446
|
|
Chris@82
|
447 * Fixed compilation with OpenMP on AIX (thanks to Greg Bauer).
|
|
Chris@82
|
448
|
|
Chris@82
|
449 * Fixed compilation failure on x86-64 with gcc (thanks to Orion Poplawski).
|
|
Chris@82
|
450
|
|
Chris@82
|
451 * Incorporated patch from FreeBSD ports (FreeBSD does not have memalign,
|
|
Chris@82
|
452 but its malloc is 16-byte aligned).
|
|
Chris@82
|
453
|
|
Chris@82
|
454 * Cycle-counter compilation fixes for Itanium, Alpha, x86-64, Sparc,
|
|
Chris@82
|
455 MacOS (thanks to Matt Boman, John Bowman, and James A. Treacy for
|
|
Chris@82
|
456 reports/fixes). Added x86-64 cycle counter for PGI compilers,
|
|
Chris@82
|
457 courtesy Cristiano Calonaci.
|
|
Chris@82
|
458
|
|
Chris@82
|
459 * Fix compilation problem in test program due to C99 conflict.
|
|
Chris@82
|
460
|
|
Chris@82
|
461 * Portability fix for import_system_wisdom with djgpp (thanks to Juan
|
|
Chris@82
|
462 Manuel Guerrero).
|
|
Chris@82
|
463
|
|
Chris@82
|
464 * Fixed compilation failure on MacOS 10.3 due to getopt conflict.
|
|
Chris@82
|
465
|
|
Chris@82
|
466 * Work around Visual C++ (version 6/7) bug in SSE compilation;
|
|
Chris@82
|
467 thanks to Eddie Yee for his detailed report.
|
|
Chris@82
|
468
|
|
Chris@82
|
469 Changes from FFTW 3.1 beta 2:
|
|
Chris@82
|
470
|
|
Chris@82
|
471 * Several minor compilation fixes.
|
|
Chris@82
|
472
|
|
Chris@82
|
473 * Eliminate FFTW_TIMELIMIT flag and replace fftw_timelimit global with
|
|
Chris@82
|
474 fftw_set_timelimit function. Make wisdom work with time-limited plans.
|
|
Chris@82
|
475
|
|
Chris@82
|
476 Changes from FFTW 3.1 beta 1:
|
|
Chris@82
|
477
|
|
Chris@82
|
478 * Fixes for creating DLLs under Windows; thanks to John Pavel for his feedback.
|
|
Chris@82
|
479
|
|
Chris@82
|
480 * Fixed more 64-bit problems, thanks to John Pavel for the bug report.
|
|
Chris@82
|
481
|
|
Chris@82
|
482 * Further speed improvements for Altivec/VMX.
|
|
Chris@82
|
483
|
|
Chris@82
|
484 * Further speed improvements for non-square transpositions.
|
|
Chris@82
|
485
|
|
Chris@82
|
486 * Many minor tweaks.
|
|
Chris@82
|
487
|
|
Chris@82
|
488 FFTW 3.0.1
|
|
Chris@82
|
489
|
|
Chris@82
|
490 * Some speed improvements in SIMD code.
|
|
Chris@82
|
491
|
|
Chris@82
|
492 * --without-cycle-counter option is removed. If no cycle counter is found,
|
|
Chris@82
|
493 then the estimator is always used. A --with-slow-timer option is provided
|
|
Chris@82
|
494 to force the use of lower-resolution timers.
|
|
Chris@82
|
495
|
|
Chris@82
|
496 * Several fixes for compilation under Visual C++, with help from Stefane Ruel.
|
|
Chris@82
|
497
|
|
Chris@82
|
498 * Added x86 cycle counter for Visual C++, with help from Morten Nissov.
|
|
Chris@82
|
499
|
|
Chris@82
|
500 * Added S390 cycle counter, courtesy of James Treacy.
|
|
Chris@82
|
501
|
|
Chris@82
|
502 * Added missing static keyword that prevented simultaneous linkage
|
|
Chris@82
|
503 of different-precision versions; thanks to Rasmus Larsen for the bug report.
|
|
Chris@82
|
504
|
|
Chris@82
|
505 * Corrected accidental omission of f77_wisdom.f file; thanks to Alan Watson.
|
|
Chris@82
|
506
|
|
Chris@82
|
507 * Support -xopenmp flag for SunOS; thanks to John Lou for the bug report.
|
|
Chris@82
|
508
|
|
Chris@82
|
509 * Compilation with HP/UX cc requires -Wp,-H128000 flag to increase
|
|
Chris@82
|
510 preprocessor limits; thanks to Peter Vouras for the bug report.
|
|
Chris@82
|
511
|
|
Chris@82
|
512 * Removed non-portable use of 'tempfile' in fftw-wisdom-to-conf script;
|
|
Chris@82
|
513 thanks to Nicolas Decoster for the patch.
|
|
Chris@82
|
514
|
|
Chris@82
|
515 * Added 'make smallcheck' target in tests/ directory, at the request of
|
|
Chris@82
|
516 James Treacy.
|
|
Chris@82
|
517
|
|
Chris@82
|
518 FFTW 3.0
|
|
Chris@82
|
519
|
|
Chris@82
|
520 Major goals of this release:
|
|
Chris@82
|
521
|
|
Chris@82
|
522 * Speed: often 20% or more faster than FFTW 2.x, even without SIMD (see below).
|
|
Chris@82
|
523
|
|
Chris@82
|
524 * Complete rewrite, to make it easier to add new algorithms and transforms.
|
|
Chris@82
|
525
|
|
Chris@82
|
526 * New API, to support more general semantics.
|
|
Chris@82
|
527
|
|
Chris@82
|
528 Other enhancements:
|
|
Chris@82
|
529
|
|
Chris@82
|
530 * SIMD acceleration on supporting CPUs (SSE, SSE2, 3DNow!, and AltiVec).
|
|
Chris@82
|
531 (With special thanks to Franz Franchetti for many experimental prototypes
|
|
Chris@82
|
532 and to Stefan Kral for the vectorizing generator from fftwgel.)
|
|
Chris@82
|
533
|
|
Chris@82
|
534 * True in-place 1d transforms of large sizes (as well as compressed
|
|
Chris@82
|
535 twiddle tables for additional memory/cache savings).
|
|
Chris@82
|
536
|
|
Chris@82
|
537 * More arbitrary placement of real & imaginary data, e.g. including
|
|
Chris@82
|
538 interleaved (as in FFTW 2.x) as well as separate real/imag arrays.
|
|
Chris@82
|
539
|
|
Chris@82
|
540 * Efficient prime-size transforms of real data.
|
|
Chris@82
|
541
|
|
Chris@82
|
542 * Multidimensional transforms can operate on a subset of a larger matrix,
|
|
Chris@82
|
543 and/or transform selected dimensions of a multidimensional array.
|
|
Chris@82
|
544
|
|
Chris@82
|
545 * By popular demand, simultaneous linking to double precision (fftw),
|
|
Chris@82
|
546 single precision (fftwf), and long-double precision (fftwl) versions
|
|
Chris@82
|
547 of FFTW is now supported.
|
|
Chris@82
|
548
|
|
Chris@82
|
549 * Cycle counters (on all modern CPUs) are exploited to speed planning.
|
|
Chris@82
|
550
|
|
Chris@82
|
551 * Efficient transforms of real even/odd arrays, a.k.a. discrete
|
|
Chris@82
|
552 cosine/sine transforms (types I-IV). (Currently work via pre/post
|
|
Chris@82
|
553 processing of real transforms, ala FFTPACK, so are not optimal.)
|
|
Chris@82
|
554
|
|
Chris@82
|
555 * DHTs (Discrete Hartley Transforms), again via post-processing
|
|
Chris@82
|
556 of real transforms (and thus suboptimal, for now).
|
|
Chris@82
|
557
|
|
Chris@82
|
558 * Support for linking to just those parts of FFTW that you need,
|
|
Chris@82
|
559 greatly reducing the size of statically linked programs when
|
|
Chris@82
|
560 only a limited set of transform sizes/types are required.
|
|
Chris@82
|
561
|
|
Chris@82
|
562 * Canonical global wisdom file (/etc/fftw/wisdom) on Unix, along
|
|
Chris@82
|
563 with a command-line tool (fftw-wisdom) to generate/update it.
|
|
Chris@82
|
564
|
|
Chris@82
|
565 * Fortran API can be used with both g77 and non-g77 compilers
|
|
Chris@82
|
566 simultaneously.
|
|
Chris@82
|
567
|
|
Chris@82
|
568 * Multi-threaded version has optional OpenMP support.
|
|
Chris@82
|
569
|
|
Chris@82
|
570 * Authors' good looks have greatly improved with age.
|
|
Chris@82
|
571
|
|
Chris@82
|
572 Changes from 3.0beta3:
|
|
Chris@82
|
573
|
|
Chris@82
|
574 * Separate FMA distribution to better exploit fused multiply-add instructions
|
|
Chris@82
|
575 on PowerPC (and possibly other) architectures.
|
|
Chris@82
|
576
|
|
Chris@82
|
577 * Performance improvements via some inlining tweaks.
|
|
Chris@82
|
578
|
|
Chris@82
|
579 * fftw_flops now returns double arguments, not int, to avoid overflows
|
|
Chris@82
|
580 for large sizes.
|
|
Chris@82
|
581
|
|
Chris@82
|
582 * Workarounds for automake bugs.
|
|
Chris@82
|
583
|
|
Chris@82
|
584 Changes from 3.0beta2:
|
|
Chris@82
|
585
|
|
Chris@82
|
586 * The standard REDFT00/RODFT00 (DCT-I/DST-I) algorithm (used in
|
|
Chris@82
|
587 FFTPACK, NR, etcetera) turns out to have poor numerical accuracy, so
|
|
Chris@82
|
588 we replaced it with a slower routine that is more accurate.
|
|
Chris@82
|
589
|
|
Chris@82
|
590 * The guru planner and execute functions now have two variants, one that
|
|
Chris@82
|
591 takes complex arguments and one that takes separate real/imag pointers.
|
|
Chris@82
|
592
|
|
Chris@82
|
593 * Execute and planner routines now automatically align the stack on x86,
|
|
Chris@82
|
594 in case the calling program is misaligned.
|
|
Chris@82
|
595
|
|
Chris@82
|
596 * README file for test program.
|
|
Chris@82
|
597
|
|
Chris@82
|
598 * Fixed bugs in the combination of SIMD with multi-threaded transforms.
|
|
Chris@82
|
599
|
|
Chris@82
|
600 * Eliminated internal fftw_threads_init function, which some people were
|
|
Chris@82
|
601 calling accidentally instead of the fftw_init_threads API function.
|
|
Chris@82
|
602
|
|
Chris@82
|
603 * Check for -openmp flag (Intel C compiler) when --enable-openmp is used.
|
|
Chris@82
|
604
|
|
Chris@82
|
605 * Support AMD x86-64 SIMD and cycle counter.
|
|
Chris@82
|
606
|
|
Chris@82
|
607 * Support SSE2 intrinsics in forthcoming gcc 3.3.
|
|
Chris@82
|
608
|
|
Chris@82
|
609 Changes from 3.0beta1:
|
|
Chris@82
|
610
|
|
Chris@82
|
611 * Faster in-place 1d transforms of non-power-of-two sizes.
|
|
Chris@82
|
612
|
|
Chris@82
|
613 * SIMD improvements for in-place, multi-dimensional, and/or non-FFTW_PATIENT
|
|
Chris@82
|
614 transforms.
|
|
Chris@82
|
615
|
|
Chris@82
|
616 * Added support for hard-coded DCT/DST/DHT codelets of small sizes; the
|
|
Chris@82
|
617 default distribution only includes hard-coded size-8 DCT-II/III, however.
|
|
Chris@82
|
618
|
|
Chris@82
|
619 * Many minor improvements to the manual. Added section on using the
|
|
Chris@82
|
620 codelet generator to customize and enhance FFTW.
|
|
Chris@82
|
621
|
|
Chris@82
|
622 * The default 'make check' should now only take a few minutes; for more
|
|
Chris@82
|
623 strenuous tests (which may take a day or so), do 'cd tests; make bigcheck'.
|
|
Chris@82
|
624
|
|
Chris@82
|
625 * fftw_print_plan is split into fftw_fprint_plan and fftw_print_plan, where
|
|
Chris@82
|
626 the latter uses stdout.
|
|
Chris@82
|
627
|
|
Chris@82
|
628 * Fixed ability to compile with a C++ compiler.
|
|
Chris@82
|
629
|
|
Chris@82
|
630 * Fixed support for C99 complex type under glibc.
|
|
Chris@82
|
631
|
|
Chris@82
|
632 * Fixed problems with alloca under MinGW, AIX.
|
|
Chris@82
|
633
|
|
Chris@82
|
634 * Workaround for gcc/SPARC bug.
|
|
Chris@82
|
635
|
|
Chris@82
|
636 * Fixed multi-threaded initialization failure on IRIX due to lack of
|
|
Chris@82
|
637 user-accessible PTHREAD_SCOPE_SYSTEM there.
|
