2 ====================================================================
6 * Improved cross compiling.
7 * Fix the bug on musl libc.
10 * Optimize BLAS on Power8
11 * Fixed Julia+OpenBLAS bugs on Power8
14 * Optimize BLAS on MIPS P5600 and I6400 (Thanks, Shivraj Patil, Kaustubh Raste)
17 * Improved on ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
20 ====================================================================
24 * If you set MAKE_NB_JOBS flag less or equal than zero,
25 make will be without -j.
28 * Support building Visual Studio static library. (#813, Thanks, theoractice)
29 * Fix bugs to pass buidbot CI tests (http://build.openblas.net)
32 * Provide DGEMM 8x4 kernel for Cortex-A57 (Thanks, Ashwin Sekhar T K)
35 * Optimize S and C BLAS3 on Power8
36 * Optimize BLAS2/1 on Power8
38 ====================================================================
42 * Enable BUILD_LAPACK_DEPRECATED=1 by default.
44 ====================================================================
48 * Avoid potential getenv segfault. (#716)
49 * Import LAPACK svn bugfix #142-#147,#150-#155
52 * Optimize c/zgemv for AMD Bulldozer, Piledriver, Steamroller
53 * Fix bug with scipy linalg test.
56 * Improve DGEMM for ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
59 * Optimize D and Z BLAS3 functions for Power8.
61 ====================================================================
65 * Upgrade LAPACK to 3.6.0 version.
66 Add BUILD_LAPACK_DEPRECATED option in Makefile.rule to build
67 LAPACK deprecated functions.
68 * Add MAKE_NB_JOBS option in Makefile.
69 Force number of make jobs.This is particularly
70 useful when using distcc. (#735. Thanks, Jerome Robert.)
71 * Redesign unit test. Run unit/regression test at every build (Travis-CI and Appveyor).
72 * Disable multi-threading for small size swap and ger. (#744. Thanks, Jerome Robert)
73 * Improve small zger, zgemv, ztrmv using stack alloction (#727. Thanks, Jerome Robert)
74 * Let openblas_get_num_threads return the number of active threads.
75 (#760. Thanks, Jerome Robert)
76 * Support illumos(OmniOS). (#749. Thanks, Lauri Tirkkonen)
77 * Fix LAPACK Dormbr, Dormlq bug. (#711, #713. Thanks, Brendan Tracey)
78 * Update scipy benchmark script. (#745. Thanks, John Kirkham)
81 * Optimize trsm kernels for AMD Bulldozer, Piledriver, Steamroller.
82 * Detect Intel Avoton.
83 * Detect AMD Trinity, Richland, E2-3200.
84 * Fix gemv performance bug on Mac OSX Intel Haswell.
85 * Fix some bugs with CMake and Visual Studio
88 * Support and optimize Cortex-A57 AArch64.
89 (#686. Thanks, Ashwin Sekhar TK)
90 * Fix Android build on ARMV7 (#778. Thanks, Paul Mustiere)
91 * Update ARMV6 kernels.
94 * Fix detection of POWER architecture
95 (#684. Thanks, Sebastien Villemot)
97 ====================================================================
101 * Support cmake on x86/x86-64. Natively compiling on MS Visual Studio.
102 (experimental. Thank Hank Anderson for the initial cmake porting work.)
104 On Linux and Mac OSX, OpenBLAS cmake supports assembly kernels.
109 On Windows MS Visual Studio, OpenBLAS cmake only support C kernels.
110 (OpenBLAS uses AT&T style assembly, which is not supported by MSVC.)
111 e.g. cmake -G "Visual Studio 12 Win64" .
112 Open OpenBLAS.sln and build.
114 * Enable MAX_STACK_ALLOC flags by default.
115 Improve ger and gemv for small matrices.
116 * Improve gemv parallel with small m and large n case.
117 * Improve ?imatcopy when lda==ldb (#633. Thanks, Martin Koehler)
118 * Add vecLib benchmarks (#565. Thanks, Andreas Noack.)
119 * Fix LAPACK lantr for row major matrices (#634. Thanks, Dan Kortschak)
120 * Fix LAPACKE lansy (#640. Thanks, Dan Kortschak)
121 * Import bug fixes for LAPACKE s/dormlq, c/zunmlq
122 * Raise the signal when pthread_create fails (#668. Thanks, James K. Lowden)
123 * Remove g77 from compiler list.
124 * Enable AppVeyor Windows CI.
127 * Support pure C generic kernels for x86/x86-64.
128 * Support Intel Boardwell and Skylake by Haswell kernels.
129 * Support AMD Excavator by Steamroller kernels.
130 * Optimize s/d/c/zdot for Intel SandyBridge and Haswell.
131 * Optimize s/d/c/zdot for AMD Piledriver and Steamroller.
132 * Optimize s/d/c/zapxy for Intel SandyBridge and Haswell.
133 * Optimize s/d/c/zapxy for AMD Piledriver and Steamroller.
134 * Optimize d/c/zscal for Intel Haswell, dscal for Intel SandyBridge.
135 * Optimize d/c/zscal for AMD Bulldozer, Piledriver and Steamroller.
136 * Optimize s/dger for Intel SandyBridge.
137 * Optimize s/dsymv for Intel SandyBridge.
138 * Optimize ssymv for Intel Haswell.
139 * Optimize dgemv for Intel Nehalem and Haswell.
140 * Optimize dtrmm for Intel Haswell.
143 * Support Android NDK armeabi-v7a-hard ABI (-mfloat-abi=hard)
144 e.g. make HOSTCC=gcc CC=arm-linux-androideabi-gcc NO_LAPACK=1 TARGET=ARMV7
145 * Fix lock, rpcc bugs (#616, #617. Thanks, Grazvydas Ignotas)
147 * Support ppc64le platform (ELF ABI v2. #612. Thanks, Matthew Brandyberry.)
148 * Support POWER7/8 by POWER6 kernels. (#612. Thanks, Fábio Perez.)
150 ====================================================================
154 * Improve OpenBLASConfig.cmake. (#474, #475. Thanks, xantares.)
155 * Improve ger and gemv for small matrices by stack allocation.
156 e.g. make -DMAX_STACK_ALLOC=2048 (#482. Thanks, Jerome Robert.)
157 * Introduce openblas_get_num_threads and openblas_get_num_procs.
158 (#497. Thanks, Erik Schnetter.)
159 * Add ATLAS-style ?geadd function. (#509. Thanks, Martin Köhler.)
160 * Fix c/zsyr bug with negative incx. (#492.)
161 * Fix race condition during shutdown causing a crash in
162 gotoblas_set_affinity(). (#508. Thanks, Ton van den Heuvel.)
165 * Support AMD Streamroller.
168 * Add Cortex-A9 and Cortex-A15 targets.
170 ====================================================================
174 * Add SYMBOLPREFIX and SYMBOLSUFFIX makefile options
175 for adding a prefix or suffix to all exported symbol names
176 in the shared library.(#459, Thanks Tony Kelman)
177 * Provide OpenBLASConfig.cmake at installation.
178 * Fix Fortran compiler detection on FreeBSD.
179 (#470, Thanks Mike Nolta)
183 * Add generic kernel files for x86-64. make TARGET=GENERIC
184 * Fix a bug of sgemm kernel on Intel Sandy Bridge.
185 * Fix c_check bug on some amd64 systems. (#471, Thanks Mike Nolta)
188 * Support APM's X-Gene 1 AArch64 processors.
189 Optimize trmm and sgemm. (#465, Thanks Dave Nuechterlein)
191 ====================================================================
195 * Added CBLAS interface for ?omatcopy and ?imatcopy.
196 * Enable ?gemm3m functions.
197 * Added benchmark for ?gemm3m.
198 * Optimized multithreading lower limits.
199 * Disabled SYMM3M and HEMM3M functions
200 because of segment violations.
203 * Improved axpy and symv performance on AMD Bulldozer.
204 * Improved gemv performance on modern Intel and AMD CPUs.
206 ====================================================================
210 * Added some benchmark codes.
211 * Fix link error on Linux/musl.(Thanks Isaac Dunham)
214 * Improved s/c/zgemm performance for Intel Haswell.
215 * Improved s/d/c/zgemv performance.
216 * Support the big numa machine.(EXPERIMENT)
219 * Fix detection when cpuinfo uses "Processor". (Thanks Isaiah)
221 ====================================================================
225 * Added BLAS extensions as following.
226 s/d/c/zaxpby, s/d/c/zimatcopy, s/d/c/zomatcopy.
227 * Added OPENBLAS_CORETYPE environment for dynamic_arch. (a86d34)
228 * Added NO_AVX2 flag for old binutils. (#401)
229 * Support outputing the CPU corename on runtime.(#407)
230 * Patched LAPACK to fix bug 114, 117, 118.
231 (http://www.netlib.org/lapack/bug_list.html)
232 * Disabled ?gemm3m for a work-around fix. (#400)
234 * Fixed lots of bugs for optimized kernels on sandybridge,Haswell,
235 bulldozer, and piledriver.
236 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
239 * Improved LAPACK testing.
241 ====================================================================
245 * Improved the result for LAPACK testing. (#372)
246 * Installed DLL to prefix/bin instead of prefix/lib. (#366)
247 * Build import library on Windows.(#374)
249 * To improve LAPACK testing, we fallback some kernels. (#372)
250 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
252 ====================================================================
256 * Added OPENBLAS_VERBOSE environment variable.(#338)
257 * Make OpenBLAS thread-pool resilient to fork via pthread_atfork.
258 (#294, Thank Olivier Grisel)
262 * Detect Intel Haswell for new Macbook.
264 ====================================================================
268 * Update LAPACK to 3.5.0 version
269 * Fixed compatiable issues with Clang and Pathscale compilers.
272 * Optimization on Intel Haswell.
273 * Enable optimization kernels on AMD Bulldozer and Piledriver.
276 * Support ARMv6 and ARMv7 ISA.
277 * Optimization on ARM Cortex-A9.
279 ====================================================================
283 * Support Open64 5.0. (#266)
284 * Add executable stack markings. (#262, Thank Sébastien Fabbro)
285 * Respect user's LDFLAGS (Thank Sébastien Fabbro)
288 * Rollback bulldozer and piledriver kernels to barcelona kernels (#263)
289 We will fix the compuational error bug in bulldozer and piledriver kernels.
291 ====================================================================
295 * Support LSB (Linux Standard Base) 4.1.
297 * Include LAPACK 3.4.2 source codes to the repo.
298 Avoid downloading at compile time.
299 * Add NO_PARALLEL_MAKE flag to disable parallel make.
300 * Create openblas_get_parallel to retrieve information which
301 parallelization model is used by OpenBLAS. (Thank grisuthedragon)
302 * Detect LLVM/Clang compiler. The default compiler is Clang on Mac OS X.
303 * Change LIBSUFFIX from .lib to .a on windows.
304 * A work-around for dtrti_U single thread bug. Replace it with LAPACK codes. (#191)
307 * Optimize c/zgemm, trsm, dgemv_n, ddot, daxpy, dcopy on
308 AMD Bulldozer. (Thank Werner Saar)
309 * Add Intel Haswell support (using Sandybridge optimizations).
311 * Add AMD Piledriver support (using Bulldozer optimizations).
312 * Fix the computational error in zgemm avx kernel on
314 * Fix the overflow bug in gemv.
315 * Fix the overflow bug in multi-threaded BLAS3, getrf when NUM_THREADS
316 is very large.(#214, #221, #246).
318 * Support loongcc (Open64 based) compiler for ICT Loongson 3A/B.
321 * Support Power7 by old Power6 kernels. (#220)
323 ====================================================================
327 * Improved OpenMP performance slightly. (d744c9)
328 * Improved cblas.h compatibility with Intel MKL.(#185)
329 * Fixed the overflowing bug in single thread cholesky factorization.
330 * Fixed the overflowing buffer bug of multithreading hbmv and sbmv.(#174)
333 * Added AMD Bulldozer x86-64 S/DGEMM AVX kernels. (Thank Werner Saar)
334 We will tune the performance in future.
335 * Auto-detect Intel Xeon E7540.
336 * Fixed the overflowing buffer bug of gemv. (#173)
337 * Fixed the bug of s/cdot about invalid reading NAN on x86_64. (#189)
341 ====================================================================
345 * Added NO_SHARED flag to disable generating the shared library.
346 * Compile LAPACKE with ILP64 modle when INTERFACE64=1 (#158)
347 * Export LAPACK 3.4.2 symbols in shared library. (#147)
348 * Only detect the number of physical CPU cores on Mac OSX. (#157)
349 * Fixed NetBSD build. (#155)
350 * Fixed compilation with TARGET=GENERIC. (#160)
352 * Restore the original CPU affinity when calling
353 openblas_set_num_threads(1) (#153)
354 * Fixed a SEGFAULT bug in dgemv_t when m is very large.(#154)
357 ====================================================================
361 * Upgraded LAPACK to 3.4.2 version. (#145)
362 * Provided support for passing CFLAGS, FFLAGS, PFLAGS,
363 FPFLAGS to make. (#137)
364 * f77blas.h:compatibility for compilers without C99 complex
365 number support. (#141)
367 * Added NO_AVX flag. Check OS supporting AVX on runtime. (#139)
368 * Fixed zdot incompatibility ABI issue with GCC 4.7 on
369 Windows 32-bit. (#140)
371 * Fixed the generation of shared library bug.
372 * Fixed the detection bug on the Loongson 3A server.
373 ====================================================================
377 * Fixed LAPACK unstable bug about ?laswp. (#130)
378 * Fixed the shared library bug about unloading the library on
380 * Fixed the compilation failure on BlueGene/P (TARGET=PPC440FP2)
381 Please use gcc and IBM xlf. (#134)
383 * Supported goto_set_num_threads and openblas_set_num_threads
384 APIs in Windows. They can set the number of threads on runtime.
386 ====================================================================
390 * Fixed exporting DLL functions bug on Windows/MingW
391 * Support GNU Hurd (Thank Sylvestre Ledru)
392 * Support kfreebsd kernel (Thank Sylvestre Ledru)
394 * Support Intel Sandy Bridge 22nm desktop/mobile CPU
396 * Improve the detection of SPARC (Thank Sylvestre Ledru)
398 ====================================================================
403 * Fixed the SEGFAULT bug about hyper-theading
404 * Support AMD Bulldozer by using GotoBLAS2 AMD Barcelona codes
406 ====================================================================
410 * Removed the limitation (64) of numbers of CPU cores.
411 Now, it supports 256 cores at max.
412 * Supported clang compiler.
413 * Fixed some build bugs on FreeBSD
415 * Optimized Level-3 BLAS on Intel Sandy Bridge x86-64 by AVX instructions.
416 Please use gcc >= 4.6 or clang >=3.1.
417 * Support AMD Bobcat by using GotoBLAS2 AMD Barcelona codes.
419 ====================================================================
423 * Upgraded LAPACK to 3.4.1 version. (Thank Zaheer Chothia)
424 * Supported LAPACKE, a C interface to LAPACKE. (Thank Zaheer Chothia)
425 * Fixed the build bug (MD5 and download) on Mac OSX.
426 * Auto download CUnit 2.1.2-2 from SF.net with UTEST_CHECK=1.
427 * Fxied the compatibility issue for compilers without C99 complex number
430 * Auto-detect Intel Sandy Bridge Core i7-3xxx & Xeon E7 Westmere-EX.
431 * Test alpha=Nan in dscale.
432 * Fixed a SEGFAULT bug in samax on x86 windows.
434 ====================================================================
438 * Set soname of shared library on Linux.
439 * Added LIBNAMESUFFIX flag in Makefile.rule. The user can use
440 this flag to control the library name, e.g. libopenblas.a,
441 libopenblas_ifort.a or libopenblas_omp.a.
442 * Added GEMM_MULTITHREAD_THRESHOLD flag in Makefile.rule.
443 The lib use single thread in GEMM function with small matrices.
445 * Used GEMV SSE/SSE2 kernels on x86 32-bit.
446 * Exported CBLAS functions in Windows DLL.
448 * Completed Level-3 BLAS optimization on Loongson 3A CPU.
449 * Improved GEMV performance on Loongson 3A CPU.
450 * Improved Level-3 BLAS performance on Loongson 3B CPU. (EXPERIMENT)
452 ====================================================================
456 * Fixed missing "#include <sched.h>" bug on Mac OS X.
457 Thank Mike Nolta for the patch.
458 * Upgraded LAPACK to 3.4.0 version
459 * Fixed a bug on Mac OS X. Don't require SystemStubs on OS X.
460 SystemStubs does not exist on Lion. Thank Stefan Karpinski.
461 * Improved README with using OpenMP. Check the internal threads
462 count less than or equal to omp_get_max_threads()
464 * Auto-detect Intel Core i6/i7 (Sandy Bridge) CPU with Nehalem assembly kernels
465 * Fixed some bugs on MingW 64-bit including zgemv, cdot, zdot.
467 ====================================================================
471 * Fixed a bug about installation. The header file "fblas77.h"
473 * Fixed #61 a building bug about setting TARGET and DYNAMIC_ARCH.
474 * Try to handle absolute path of shared library in OSX. (#57)
475 Thank Dr Kane O'Donnell.
476 * Changed the installation folder layout to $(PREFIX)/include and
480 * Fixed #58 zdot/xdot SEGFAULT bug with GCC-4.6 on x86. According
481 to i386 calling convention, The callee should remove the first
482 hidden parameter.Thank Mr. John for this patch.
484 ====================================================================
489 * Added DTB_ENTRIES into dynamic arch setting parameters. Now,
490 it can read DTB_ENTRIES on runtime. (Refs issue #55 on github)
492 ====================================================================
497 * Fixed a building bug when DYNAMIC_ARCH=1 & INTERFACE64=1.
498 (Refs issue #44 on github)
500 ====================================================================
505 * Stop the build and output the error message when detecting
506 fortran compiler failed. (Refs issue #42 on github)
508 ====================================================================
513 * Fixed blasint undefined bug in <cblas.h> file. Other software
514 could include this header successfully(Refs issue #13 on github)
515 * Fixed the SEGFAULT bug on 64 cores. On SMP server, the number
516 of CPUs or cores should be less than or equal to 64.(Refs issue #14
518 * Support "void goto_set_num_threads(int num_threads)" and "void
519 openblas_set_num_threads(int num_threads)" when USE_OPENMP=1
520 * Added extern "C" to support C++. Thank Tasio for the patch(Refs
522 * Provided an error message when the arch is not supported.(Refs
524 * Fixed issue #23. Fixed a bug of f_check script about generating link flags.
525 * Added openblas_set_num_threads for Fortran.
526 * Fixed #25 a wrong result of rotmg.
527 * Fixed a bug about detecting underscore prefix in c_check.
528 * Print the wall time (cycles) with enabling FUNCTION_PROFILE
529 * Fixed #35 a build bug with NO_LAPACK=1 & DYNAMIC_ARCH=1
530 * Added install target. You can use "make install". (Refs #20)
534 * Fixed #28 a wrong result of dsdot on x86_64.
535 * Fixed #32 a SEGFAULT bug of zdotc with gcc-4.6.
536 * Fixed #33 ztrmm bug on Nehalem.
537 * Work-around #27 the low performance axpy issue with small imput size & multithreads.
540 * Fixed #28 a wrong result of dsdot on Loongson3A/MIPS64.
541 * Optimized single/double precision BLAS Level3 on Loongson3A/MIPS64. (Refs #2)
542 * Optimized single/double precision axpy function on Loongson3A/MIPS64. (Refs #3)
544 ====================================================================
549 * Support "make NO_LAPACK=1" to build the library without
551 * Fixed randomly SEGFAULT when nodemask==NULL with above Linux 2.6.34.
552 Thank Mr.Ei-ji Nakama providing this patch. (Refs issue #12 on github)
553 * Added DEBUG=1 rule in Makefile.rule to build debug version.
554 * Disable compiling quad precision in reference BLAS library(netlib BLAS).
555 * Added unit testcases in utest/ subdir. Used CUnit framework.
556 * Supported OPENBLAS_* & GOTO_* environment variables (Pleas see README)
557 * Imported GotoBLAS2 1.13 BSD version
560 * On x86 32bits, fixed a bug in zdot_sse2.S line 191. This would casue
561 zdotu & zdotc failures. Instead, work-around it. (Refs issue #8 #9 on github)
562 * Modified ?axpy functions to return same netlib BLAS results
563 when incx==0 or incy==0 (Refs issue #7 on github)
564 * Modified ?swap functions to return same netlib BLAS results
565 when incx==0 or incy==0 (Refs issue #6 on github)
566 * Modified ?rot functions to return same netlib BLAS results
567 when incx==0 or incy==0 (Refs issue #4 on github)
568 * Detect Intel Westmere,Intel Clarkdale and Intel Arrandale
569 to use Nehalem codes.
570 * Fixed a typo bug about compiling dynamic ARCH library.
572 * Improve daxpy performance on ICT Loongson 3A.
573 * Supported ICT Loongson 3A CPU (Refs issue #1 on github)
574 ====================================================================