2 ====================================================================
7 * Improved thread locking behaviour in blas_server and parallel getrf
8 * Imported bugfix 394 from LAPACK (spurious reference to "XERBL"
10 * Imported bugfix 403 from LAPACK (compile option "recursive" required
11 for correctness with Intel and PGI)
12 * Imported bugfix 408 from LAPACK (wrong scaling in ZHEEQUB)
13 * Imported bugfix 411 from LAPACK (infinite loop in LARGV/LARTG/LARTGP)
14 * Fixed mismatches between BUFFERSIZE and GEMM_UNROLL parameters that
15 could lead to crashes at large matrix sizes
16 * Restored internal soname in dynamic libraries on FreeBSD and Dragonfly
17 * Added API (openblas_setaffinity) to set the thread affinity on Linux
18 * Added initial infrastructure for half-precision floating point
19 (bfloat16) support with a generic implementation of SHGEMM
20 * Added CMAKE build system support for building the cblas_Xgemm3m
22 * Fixed CMAKE support for building in a path with embedded spaces
23 * Fixed CMAKE (non)handling of NO_EXPRECISION and MAX_STACK_ALLOC
24 * Fixed GCC version detection in the Makefiles
25 * Allowed overriding the names of AR, AS and LD in Makefile builds
28 * Fixed big-endian POWER8 ELFv2 builds on FreeBSD
29 * Fixed GCC version checks and DYNAMIC_ARCH builds on POWER9
30 * Fixed CMAKE build support for POWER9
31 * fixed a potential race condition in the thread buffer allocation
32 * Worked around LAPACK test failures on PPC G4
35 * Fixed a potential race condition in the thread buffer allocation
36 * Added support for MIPS 24K/24KE family based on P5600 kernels
39 * fixed a potential race condition in the thread buffer allocation
40 * Added TARGET=GENERIC
43 * Fixed a race condition in the thread buffer allocation
46 * Fixed a race condition in the thread buffer allocation
47 * Fixed zero initialisation in the assembly for SGEMM and DGEMM BETA
48 * Improved performance of the ThunderX2 DAXPY kernel
49 * Added an optimized SGEMM kernel for Cortex A53
50 * Fixed Makefile support for INTERFACE64 (8-byte integer)
53 * Fixed a syntax error in the CMAKE setup for SkylakeX
54 * Improved performance of STRSM on Haswell, SkylakeX and Ryzen
55 * Improved SGEMM performance on SGEMM for workloads with ldc a
57 * Improved DGEMM performance on Skylake X
58 * Fixed unwanted AVX512-dependency of SGEMM in DYNAMIC_ARCH
59 builds created on SkylakeX
60 * Removed data alignment requirement in the SSE2 copy kernels
61 that could cause spurious crashes
62 * Added a workaround for an optimizer bug in AppleClang 11.0.3
63 * Fixed LAPACK test failures due to wrong options for Intel Fortran
64 * Fixed compilation and LAPACK test results with recent Flang
66 * Fixed DYNAMIC_ARCH builds with CMAKE on OS X
67 * Fixed missing exports of cblas_i?amin, cblas_i?min, cblas_i?max,
68 cblas_?sum, cblas_?gemm3m in the shared library on OS
69 * Fixed reporting of cpu name in DYNAMIC_ARCH builds (would sometimes
70 show the name of an older generation chip supported by the same kernels)
73 * Improved performance of SGEMM/STRMM and DGEMM/DTRMM on Z14
75 ====================================================================
80 * Fixed a miscompilation of the GETRF functions with CMAKE
81 * Imported bugfix 390 from LAPACK (missing NaN propagation in xCOMBSSQ)
82 * The size of the memory buffer used for splitting GEMM tasks across
83 multiple threads can now be configured in the build system.
86 * Fixed several compilation problems related to endianness
87 and ELF version on POWER8 and POWER9
88 * Fixed use of the absolute value IAMIN/IAMAX instead of IMIN/IMAX
89 * Fixed a race condition in the level3 blas code
92 * Fixed use of the absoltute value IAMIN/IAMAX instead of IMIN/IMAX
95 * Fixed a race condition in the level3 blas code
96 * Fixed compilation on Android
98 * Added support for Ampere EMAG8180
99 * Added support for Neoverse N1
100 * Improved performance of the blas_lock function
101 * Fixed a race condition in the level3 blas code
102 * Fixed a performance regression on TSV110-based servers
105 * Fixed a long-standing error with undeclared register overwrites
106 in the DSCAL microkernel for HASWELL,SKYLAKEX and ZEN
107 * Fixed a long-standing bug in the SSE implementation of IAMAX
108 * Fixed a CMAKE build failure with DYNAMIC_ARCH
109 * Fixed cpu autodetection of Goldmont+, Cannon Lake and Ice Lake
110 * Fixed a compilation failure on OSX with compiler name containing dash
111 * Fixed compilation with MinGW on SkylakeX
112 * Improved speed of the AVX512 GEMM3M kernel on SkylakeX
113 * Added an AVX512 STRMM kernel for SkylakeX
114 * Improved GEMM performance on Haswell and Zen
117 * fixed compilation of the DYNAMIC_ARCH code
119 ====================================================================
124 ` * LAPACK has been updated to 3.9.0 (plus patches up to
126 * CMAKE support has been improved in several areas including
128 * a thread race condition in the GEMM3M kernels was resolved
129 * the "generic" (plain C) gemm beta kernel used by many targets
131 * an optimized version of the LAPACK trtrs functions has been added
132 * an incompatibilty between the LAPACK tests and the OpenBLAS
133 implementation of XERBLA was resolved, removing the numerous
134 warnings about wrong error exits in the former
135 * support for NetBSD has been added
136 * support for compilation with g95 and non-GNU versions of ld
138 * support for compilation with (upcoming) gcc 10 has been added
141 * worked around miscompilation of several POWER8 and POWER9
142 kernels by older versions of gcc
143 * added support for big-endian POWER8 and for compilation on AIX
144 * corrected bugs in the big-endian support for PPC440 and PPC970
145 * DYNAMIC_ARCH support is now available in CMAKE builds as well
148 * performance of DGEMM_BETA and SGEMM_NCOPY has been improved
149 * compilation for 32bit works again
150 * performance of the RPCC function has been improved
151 * improved performance on small systems
152 * DYNAMIC_ARCH support is now available in CMAKE builds as well
153 * cross-compilation from OSX to IOS was simplified
156 * a new AVX512 DGEMM kernel was added and the AVX512 SGEMM kernel
157 was significantly improved
158 * optimized AVX512 kernels for CGEMM and ZGEMM have been added
159 * AVX2 kernels for STRMM, SGEMM, and CGEMM have been significantly
160 sped up and optimized CGEMM3M and ZGEMM3M kernels have been added
161 * added support for QEMU virtual cpus
162 * a compilation problem with PGI and SUN compilers was fixed
163 * Intel "Goldmont plus" is now autodetected
164 * a potential crash on program exit on MS Windows has been fixed
167 * an unwanted case sensitivity in the implementation of LSAME
168 on older 32bit AMD cpus was fixed
171 * Z15 is now supported as Z14
172 * DYNAMIC_ARCH is now available on ZARCH as well
174 ====================================================================
179 * having the gmake special variables TARGET_ARCH or TARGET_MACH
180 defined no longer causes build failures in ctest or utest
181 * defining NO_AFFINITY or USE_TLS to 0 in gmake builds no longer
182 has the same effect as setting them to 1
183 * a new test program was added to allow checking the library for
185 * a new option USE_LOCKING was added to ensure thread safety when
186 OpenBLAS itself is built without multithreading but will be
187 called from multiple threads.
188 * a build failure on Linux with glibc versions earlier than 2.5
190 * a runtime error with CPU enumeration (and NO_AFFINITY not set)
191 on glibc 2.6 was fixed
192 * NO_AFFINITY was added to the CMAKE options (and defaults to being
193 active on Linux, as in the gmake builds)
196 * the build-time logic for detection of AVX512 availability in
197 the processor and compiler was fixed
198 * gmake builds on OSX now set the internal name of the library to
199 libopenblas.0.dylib (consistent with CMAKE)
200 * the Haswell DGEMM kernel received a significant speedup through
201 improved prefetch and load instructions
202 * performance of DGEMM, DTRMM, DTRSM and ZDOT on Zen/Zen2 was markedly
203 increased by avoiding vpermpd instructions
204 * the SKYLAKEX (AVX512) DGEMM helper functions have now been disabled
205 to fix remaining errors in DGEMM, DSYMM and DTRMM
208 * added support for building on FreeBSD/powerpc64 and FreeBSD/ppc970
209 * added optimized kernels for POWER9 SGEMM and STRMM
212 * fixed the softfp implementations of xAMAX and IxAMAX
213 * removed the predefined -march= flags on both ARMV5 and ARMV6 as
214 they were appropriate for only a subset of platforms
216 ====================================================================
221 * the build tools now check that a given cpu TARGET is actually valid
222 * the build-time check of system features (c_check) has been made
223 less dependent on particular perl features (this should mainly
224 benefit building on Windows)
225 * several problem with the ReLAPACK integration were fixed,
226 including INTERFACE64 support and building a shared library
227 * building with CMAKE on BSD systems was improved
228 * a non-absolute SUM function was added based on the
229 existing optimized code for ASUM
230 * CBLAS interfaces to the IxMIN and IxMAX functions were added
231 * a name clash between LAPACKE and BOOST headers was resolved
232 * CMAKE builds with OpenMP failed to include the appropriate getrf_parallel
234 * a crash on thread (key) deletion with the USE_TLS=1 memory management
236 * restored several earlier fixes, in particular for OpenMP performance,
237 building on BSD, and calling fork on CYGWIN, which had inadvertently
238 been dropped in the 0.3.3 rewrite of the memory management code.
241 * the AVX512 DGEMM kernel has been disabled again due to unsolved problems
242 * building with old versions of MSVC was fixed
243 * it is now possible to build a static library on Windows with CMAKE
244 * accessing environment variables on CYGWIN at run time was fixed
245 * the CMAKE build system now recognizes 32bit userspace on 64bit hardware
246 * Intel "Denverton" atom and Hygon "Dhyana" zen CPUs are now autodetected
247 * building for DYNAMIC_ARCH with a DYNAMIC_LIST of targets is now supported
249 * building for DYNAMIC_ARCH with GENERIC as the default target is now supported
250 * a buffer overflow in the SSE GEMM kernel for Intel Nano targets was fixed
251 * assembly bugs involving undeclared modification of input operands were fixed
252 in the AXPY, DOT, GEMV, GER, SCAL, SYMV and TRSM microkernels for Nehalem,
253 Sandybridge, Haswell, Bulldozer and Piledriver. These would typically cause
254 test failures or segfaults when compiled with recent versions of gcc from 8 onward.
255 * a similar bug was fixed in the blas_quickdivide code used to split workloads
257 * a bug in the IxMIN implementation for the GENERIC target made it return the result of IxMAX
258 * fixed building on SkylakeX systems when either the compiler or the (emulated) operating
259 environment does not support AVX512
260 * improved GEMM performance on ZEN targets
263 * build failures caused by the recently added checks for AVX512 were fixed
264 * an inline assembly bug involving undeclared modification of an input argument was
265 fixed in the blas_quickdivide code used to split workloads in most functions
266 * a bug in the IMIN implementation for the GENERIC target made it return the result of IMAX
269 * a bug in the IMIN implementation made it return the result of IMAX
272 * single precision BLAS1/2 functions have received optimized POWER8 kernels
273 * POWER9 is now a separate target, with an optimized DGEMM/DTRMM kernel
274 * building on PPC970 systems under OSX Leopard or Tiger is now supported
275 * out-of-bounds memory accesses in the gemm_beta microkernels were fixed
276 * building a shared library on AIX is now supported for POWER6
277 * DYNAMIC_ARCH support has been added for POWER6 and newer
280 * corrected xDOT behaviour with zero INC_X or INC_Y
281 * a bug in the IMIN implementation made it return the result of IMAX
284 * added support for HiSilicon TSV110 cpus
285 * the CMAKE build system now recognizes 32bit userspace on 64bit hardware
286 * cross-compilation with CMAKE now works again
287 * a bug in the IMIN implementation made it return the result of IMAX
288 * ARMV8 builds with the BINARY=32 option are now automatically handled as ARMV7
291 * optimized microkernels for single precicion BLAS1/2 functions have been added
294 ====================================================================
299 * loop unrolling in TRMV has been enabled again.
300 * A domain error in the thread workload distribution for SYRK
302 * gmake builds will now automatically add -fPIC to the build
303 options if the platform requires it.
304 * a pthreads key leakage (and associate crash on dlclose) in
305 the USE_TLS codepath was fixed.
306 * building of the utest cases on systems that do not provide
307 an implementation of complex.h was fixed.
310 * the SkylakeX code was changed to compile on OSX.
311 * unwanted application of the -march=skylake-avx512 option
312 to the common code parts of a DYNAMIC_ARCH build was fixed.
313 * improved performance of SGEMM for small workloads on Skylake X.
314 * performance of SGEMM and DGEMM was improved on Haswell.
317 * a configuration error that broke the CNRM2 kernel was corrected.
318 * compilation of the GEMM kernels with CMAKE was fixed.
319 * DYNAMIC_ARCH builds are now available with CMAKE as well.
320 * using CMAKE for cross-compilation to the new cpu TARGETs
321 introduced in 0.3.4 now works.
324 * a problem in cpu autodetection for AIX has been corrected.
326 ====================================================================
331 * the new, experimental thread-local memory allocation had
332 inadvertently been left enabled for gmake builds in 0.3.3
333 despite the announcement. It is now disabled by default, and
334 single-threaded builds will keep using the old allocator even
335 if the USE_TLS option is turned on.
336 * OpenBLAS will now provide enough buffer space for at least 50
338 * The output of openblas_get_config() now contains the version
340 * A serious thread safety bug in GEMV operation with small M and
341 large N size has been fixed.
342 * The code will now automatically call blas_thread_init after a
343 fork if needed before handling a call to openblas_set_num_threads
344 * Accesses to parallelized level3 functions from multiple callers
345 are now serialized to avoid thread races (unless using OpenMP).
346 This should provide better performance than the known-threadsafe
347 (but non-default) USE_SIMPLE_THREADED_LEVEL3 option.
348 * When building LAPACK with gfortran, -frecursive is now (again)
349 enabled by default to ensure correct behaviour.
350 * The OpenBLAS version cblas.h now supports both CBLAS_ORDER and
351 CBLAS_LAYOUT as the name of the matrix row/column order option.
352 * Externally set LDFLAGS are now passed through to the final compile/link
353 steps to facilitate setting platform-specific linker flags.
354 * A potential race condition during the build of LAPACK (that would
355 usually manifest itself as a failure to build TESTING/MATGEN) has been
357 * xHEMV has been changed to stay single-threaded for small input sizes
358 where the overhead of multithreading exceeds any possible gains
359 * CSWAP and ZSWAP have been limited to a single thread except on ARMV8 or
360 ThunderX hardware with sizable input.
361 * Linker flags for the PGI compiler have been updated
362 * Behaviour of AXPY with zero increments is now handled in the C interface,
363 correcting the result on at least Intel Atom.
364 * The result matrix from calling SGELSS with an all-zero input matrix is
365 now zeroed completely.
368 * Autodetection of AMD Ryzen2 has been fixed (again).
369 * CMAKE builds now support labeling of an INTERFACE64=1 build of
370 the library with the _64 suffix.
371 * AVX512 version of DGEMM has been added and the AVX512 SGEMM kernel
372 has been sped up by rewriting with C intrinsics
373 * Fixed compilation on RHEL5/CENTOS5 (issue with typename __WAIT_STATUS)
376 * added support for building on AIX (with gcc and GNU tools from AIX Toolbox).
377 * CPU type detection has been implemented for AIX.
378 * CPU type detection has been fixed for NETBSD.
381 * AXPY on LOONGSON3A has been corrected to pass "zero increment" utest.
382 * DSDOT on LOONGSON3A has been fixed.
383 * the SGEMM microkernel has been hardened against potential data loss.
386 * DYNAMic_ARCH support is now available for 64bit ARM
387 * cross-compiling for ARMV8 under iOS now works.
388 * cpu-specific code has been rearranged to make better use of both
389 hardware commonalities and model-specific compiler optimizations.
390 * XGENE1 has been removed as a TARGET, superseded by the improved generic
394 * Older assembly mnemonics have been converted to UAL form to allow
395 building with clang 7.0
396 * Cross compiling LAPACKE for Android has been fixed again (broken by
397 update to LAPACK 3.7.0 some while ago).
399 ====================================================================
404 * thread memory allocation has been switched back to the method
405 used before version 0.3.1 due to unexpected problems caused by
406 the new code under some circumstances. A new compile-time option
407 USE_TLS has been added to enable the new code, and it is hoped
408 that this can become the default again in the next version.
409 * LAPAck PR272 has been integrated, which fixes spurious errors
410 in DSYEVR and related functions caused by missing conversion
411 from ILAENV to ILAENV_2STAGE in several _2stage routines.
412 * the cmake-generated OpenBLASConfig.cmake now uses correct case
413 for the name of the library
414 * added support for Haiku OS
417 * added AVX512 implementations of SDOT, DDOT, SAXPY, DAXPY,
418 DSCAL, DGEMVN and DSYMVL
419 * added a workaround for a cygwin issue that prevented compilation
423 * added autodetection of Z14
424 * fixed TRMM errors in the generic target
426 ====================================================================
431 * fixes for regressions caused by the rewrite of the thread
432 initialization code in 0.3.1
435 * fixed cpu autodetection for the BSDs
438 * fixed utest errors in AXPY, DSDOT, ROT and SWAP
441 * added autodetection of AMD Ryzen 2
442 * fixed build with older versions of MSVC
444 ====================================================================
449 * rewritten thread initialization code with significantly reduced overhead
450 * added CBLAS interfaces to the IxAMIN BLAS extension functions
451 * fixed the lapack-test target
452 * CMAKE builds now create an OpenBLASConfig.cmake file
453 * ZAXPY now uses a single thread for small input sizes
454 * the LAPACK code was updated from Reference-LAPACK/lapack#253
455 (fixing LAPACKE interfaces to Aasen's functions)
458 * corrected CROT and ZROT behaviour with zero INC_X
461 * corrected xDOT behaviour with zero INC_X or INC_Y
464 * retired some older targets of DYNAMIC_ARCH builds to a new option DYNAMIC_OLDER,
465 this affects PENRYN,DUNNINGTON,OPTERON,OPTERON_SSE3,BOBCAT,ATOM and NANO
466 (which will still be supported via the slower PRESCOTT kernels when this option is not set)
467 * added an option DYNAMIC_LIST that (used in conjunction with DYNAMIC_ARCH) allows to
468 specify the list of x86_64 targets to include. Any target not on the list will be supported
469 by the Sandybridge or Nehalem kernels if available, or by Prescott.
470 * improved SWITCH_RATIO on Haswell for increased GEMM throughput
471 * added initial support for Intel Skylake X, including an AVX512 SGEMM kernel
472 * added autodetection of Intel Cannon Lake series as Skylake X
473 * added a default L2 cache size for hypervisors that return zero here (Chromebook)
474 * fixed a name clash with recent Windows10 headers that broke the build with (at least)
475 recent mingw from MSYS2
476 * fixed a link error in mixed clang/gfortran builds with OpenMP
477 * updated the OSX deployment target to 10.8
478 * switched on parallel make for builds on MS Windows by default
481 * fixed SSWAP and DSWAP behaviour with zero INC_X and INC_Y
483 ====================================================================
488 * fixed some more thread race and locking bugs
489 * added preliminary support for calling an OpenMP build of the library from multiple threads
490 * removed performance impact of thread locks added in 0.2.20 on OpenMP code
491 * general code cleanup
492 * optimized DSDOT implementation
493 * improved thread distribution for GEMM
494 * corrected IMATCOPY/OMATCOPY implementation
495 * fixed out-of-bounds accesses in the multithreaded xBMV/xPMV and SYMV implementations
496 * cmake build improvements
497 * pkgconfig file now contains build options
498 * openblas_get_config() now reports USE_OPENMP and NUM_THREADS settings used for the build
499 * corrections and improvements for systems with more than 64 cpus
500 * LAPACK code updated to 3.8.0 including later fixes
501 * added ReLAPACK, a recursive implementation of several LAPACK functions
502 * Rewrote ROTMG to handle cases that the netlib code failed to address
503 * Disabled (broken) multithreading code for xTRMV
504 * corrected prototypes of complex CBLAS functions to make our cblas.h match the generally accepted standard
505 * shared memory access failures on startup are now handled more gracefully
506 * restored utests from earlier releases (and made them pass on all affected systems)
509 * several fixes for cpu autodetection
512 * corrected vector register overwriting in several Power8 kernels
513 * optimized additional BLAS functions
516 * added support for CortexA53 and A72
517 * added autodetection for ThunderX2T99
518 * made most optimized kernels the default for generic ARMv8 targets
521 * parallelized DDOT kernel for Haswell
522 * changed alignment directives in assembly kernels to boost performance on OSX
523 * fixed register handling in the GEMV microkernels (bug exposed by gcc7)
524 * added support for building on OpenBSD and Dragonfly
525 * updated compiler options to work with Intel release 2018
526 * support fully optimized build with clang/flang on Microsoft Windows
527 * fixed building on AIX
530 * added optimized BLAS 1/2 functions
533 * fixed cpu autodetection helper code
534 * added mips32 1004K cpu (Mediatek MT7621 and similar SoC)
535 * added mips64 I6500 cpu
537 ====================================================================
542 * Improved CMake support
543 * Fixed several thread race and locking bugs
544 * Fixed default LAPACK optimization level
545 * Updated LAPACK to 3.7.0
546 * Added ReLAPACK (https://github.com/HPAC/ReLAPACK, make BUILD_RELAPACK=1)
549 * Optimizations for Power9
550 * Fixed several Power8 assembly bugs
553 * New optimized Vulcan and ThunderX2T99 targets
554 * Support for ARMV7 SOFT_FP ABI (make ARM_SOFTFP_ABI=1)
555 * Detect all cpu cores including offline ones
556 * Fix compilation with CLANG
557 * Support building a shared library for Android
560 * Fixed several threading issues
561 * Fix compilation with CLANG
564 * Detect Intel Bay Trail and Apollo Lake
565 * Detect Intel Sky Lake and Kaby Lake
566 * Detect Intel Knights Landing
567 * Detect AMD A8, A10, A12 and Ryzen
568 * Support 64bit builds with Visual Studio
569 * Fix building with Intel and PGI compilers
570 * Fix building with MINGW and TDM-GCC
571 * Fix cmake builds for Haswell and related cpus
572 * Fix building for Sandybridge with CLANG 3.9
573 * Add support for the FLANG compiler
576 * New target z13 with BLAS3 optimizations
578 ====================================================================
582 * Improved cross compiling.
583 * Fix the bug on musl libc.
586 * Optimize BLAS on Power8
587 * Fixed Julia+OpenBLAS bugs on Power8
590 * Optimize BLAS on MIPS P5600 and I6400 (Thanks, Shivraj Patil, Kaustubh Raste)
593 * Improved on ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
596 ====================================================================
600 * If you set MAKE_NB_JOBS flag less or equal than zero,
601 make will be without -j.
604 * Support building Visual Studio static library. (#813, Thanks, theoractice)
605 * Fix bugs to pass buidbot CI tests (http://build.openblas.net)
608 * Provide DGEMM 8x4 kernel for Cortex-A57 (Thanks, Ashwin Sekhar T K)
611 * Optimize S and C BLAS3 on Power8
612 * Optimize BLAS2/1 on Power8
614 ====================================================================
618 * Enable BUILD_LAPACK_DEPRECATED=1 by default.
620 ====================================================================
624 * Avoid potential getenv segfault. (#716)
625 * Import LAPACK svn bugfix #142-#147,#150-#155
628 * Optimize c/zgemv for AMD Bulldozer, Piledriver, Steamroller
629 * Fix bug with scipy linalg test.
632 * Improve DGEMM for ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
635 * Optimize D and Z BLAS3 functions for Power8.
637 ====================================================================
641 * Upgrade LAPACK to 3.6.0 version.
642 Add BUILD_LAPACK_DEPRECATED option in Makefile.rule to build
643 LAPACK deprecated functions.
644 * Add MAKE_NB_JOBS option in Makefile.
645 Force number of make jobs.This is particularly
646 useful when using distcc. (#735. Thanks, Jerome Robert.)
647 * Redesign unit test. Run unit/regression test at every build (Travis-CI and Appveyor).
648 * Disable multi-threading for small size swap and ger. (#744. Thanks, Jerome Robert)
649 * Improve small zger, zgemv, ztrmv using stack alloction (#727. Thanks, Jerome Robert)
650 * Let openblas_get_num_threads return the number of active threads.
651 (#760. Thanks, Jerome Robert)
652 * Support illumos(OmniOS). (#749. Thanks, Lauri Tirkkonen)
653 * Fix LAPACK Dormbr, Dormlq bug. (#711, #713. Thanks, Brendan Tracey)
654 * Update scipy benchmark script. (#745. Thanks, John Kirkham)
657 * Optimize trsm kernels for AMD Bulldozer, Piledriver, Steamroller.
658 * Detect Intel Avoton.
659 * Detect AMD Trinity, Richland, E2-3200.
660 * Fix gemv performance bug on Mac OSX Intel Haswell.
661 * Fix some bugs with CMake and Visual Studio
664 * Support and optimize Cortex-A57 AArch64.
665 (#686. Thanks, Ashwin Sekhar TK)
666 * Fix Android build on ARMV7 (#778. Thanks, Paul Mustiere)
667 * Update ARMV6 kernels.
670 * Fix detection of POWER architecture
671 (#684. Thanks, Sebastien Villemot)
673 ====================================================================
677 * Support cmake on x86/x86-64. Natively compiling on MS Visual Studio.
678 (experimental. Thank Hank Anderson for the initial cmake porting work.)
680 On Linux and Mac OSX, OpenBLAS cmake supports assembly kernels.
685 On Windows MS Visual Studio, OpenBLAS cmake only support C kernels.
686 (OpenBLAS uses AT&T style assembly, which is not supported by MSVC.)
687 e.g. cmake -G "Visual Studio 12 Win64" .
688 Open OpenBLAS.sln and build.
690 * Enable MAX_STACK_ALLOC flags by default.
691 Improve ger and gemv for small matrices.
692 * Improve gemv parallel with small m and large n case.
693 * Improve ?imatcopy when lda==ldb (#633. Thanks, Martin Koehler)
694 * Add vecLib benchmarks (#565. Thanks, Andreas Noack.)
695 * Fix LAPACK lantr for row major matrices (#634. Thanks, Dan Kortschak)
696 * Fix LAPACKE lansy (#640. Thanks, Dan Kortschak)
697 * Import bug fixes for LAPACKE s/dormlq, c/zunmlq
698 * Raise the signal when pthread_create fails (#668. Thanks, James K. Lowden)
699 * Remove g77 from compiler list.
700 * Enable AppVeyor Windows CI.
703 * Support pure C generic kernels for x86/x86-64.
704 * Support Intel Boardwell and Skylake by Haswell kernels.
705 * Support AMD Excavator by Steamroller kernels.
706 * Optimize s/d/c/zdot for Intel SandyBridge and Haswell.
707 * Optimize s/d/c/zdot for AMD Piledriver and Steamroller.
708 * Optimize s/d/c/zapxy for Intel SandyBridge and Haswell.
709 * Optimize s/d/c/zapxy for AMD Piledriver and Steamroller.
710 * Optimize d/c/zscal for Intel Haswell, dscal for Intel SandyBridge.
711 * Optimize d/c/zscal for AMD Bulldozer, Piledriver and Steamroller.
712 * Optimize s/dger for Intel SandyBridge.
713 * Optimize s/dsymv for Intel SandyBridge.
714 * Optimize ssymv for Intel Haswell.
715 * Optimize dgemv for Intel Nehalem and Haswell.
716 * Optimize dtrmm for Intel Haswell.
719 * Support Android NDK armeabi-v7a-hard ABI (-mfloat-abi=hard)
720 e.g. make HOSTCC=gcc CC=arm-linux-androideabi-gcc NO_LAPACK=1 TARGET=ARMV7
721 * Fix lock, rpcc bugs (#616, #617. Thanks, Grazvydas Ignotas)
723 * Support ppc64le platform (ELF ABI v2. #612. Thanks, Matthew Brandyberry.)
724 * Support POWER7/8 by POWER6 kernels. (#612. Thanks, Fábio Perez.)
726 ====================================================================
730 * Improve OpenBLASConfig.cmake. (#474, #475. Thanks, xantares.)
731 * Improve ger and gemv for small matrices by stack allocation.
732 e.g. make -DMAX_STACK_ALLOC=2048 (#482. Thanks, Jerome Robert.)
733 * Introduce openblas_get_num_threads and openblas_get_num_procs.
734 (#497. Thanks, Erik Schnetter.)
735 * Add ATLAS-style ?geadd function. (#509. Thanks, Martin Köhler.)
736 * Fix c/zsyr bug with negative incx. (#492.)
737 * Fix race condition during shutdown causing a crash in
738 gotoblas_set_affinity(). (#508. Thanks, Ton van den Heuvel.)
741 * Support AMD Streamroller.
744 * Add Cortex-A9 and Cortex-A15 targets.
746 ====================================================================
750 * Add SYMBOLPREFIX and SYMBOLSUFFIX makefile options
751 for adding a prefix or suffix to all exported symbol names
752 in the shared library.(#459, Thanks Tony Kelman)
753 * Provide OpenBLASConfig.cmake at installation.
754 * Fix Fortran compiler detection on FreeBSD.
755 (#470, Thanks Mike Nolta)
759 * Add generic kernel files for x86-64. make TARGET=GENERIC
760 * Fix a bug of sgemm kernel on Intel Sandy Bridge.
761 * Fix c_check bug on some amd64 systems. (#471, Thanks Mike Nolta)
764 * Support APM's X-Gene 1 AArch64 processors.
765 Optimize trmm and sgemm. (#465, Thanks Dave Nuechterlein)
767 ====================================================================
771 * Added CBLAS interface for ?omatcopy and ?imatcopy.
772 * Enable ?gemm3m functions.
773 * Added benchmark for ?gemm3m.
774 * Optimized multithreading lower limits.
775 * Disabled SYMM3M and HEMM3M functions
776 because of segment violations.
779 * Improved axpy and symv performance on AMD Bulldozer.
780 * Improved gemv performance on modern Intel and AMD CPUs.
782 ====================================================================
786 * Added some benchmark codes.
787 * Fix link error on Linux/musl.(Thanks Isaac Dunham)
790 * Improved s/c/zgemm performance for Intel Haswell.
791 * Improved s/d/c/zgemv performance.
792 * Support the big numa machine.(EXPERIMENT)
795 * Fix detection when cpuinfo uses "Processor". (Thanks Isaiah)
797 ====================================================================
801 * Added BLAS extensions as following.
802 s/d/c/zaxpby, s/d/c/zimatcopy, s/d/c/zomatcopy.
803 * Added OPENBLAS_CORETYPE environment for dynamic_arch. (a86d34)
804 * Added NO_AVX2 flag for old binutils. (#401)
805 * Support outputing the CPU corename on runtime.(#407)
806 * Patched LAPACK to fix bug 114, 117, 118.
807 (http://www.netlib.org/lapack/bug_list.html)
808 * Disabled ?gemm3m for a work-around fix. (#400)
810 * Fixed lots of bugs for optimized kernels on sandybridge,Haswell,
811 bulldozer, and piledriver.
812 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
815 * Improved LAPACK testing.
817 ====================================================================
821 * Improved the result for LAPACK testing. (#372)
822 * Installed DLL to prefix/bin instead of prefix/lib. (#366)
823 * Build import library on Windows.(#374)
825 * To improve LAPACK testing, we fallback some kernels. (#372)
826 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
828 ====================================================================
832 * Added OPENBLAS_VERBOSE environment variable.(#338)
833 * Make OpenBLAS thread-pool resilient to fork via pthread_atfork.
834 (#294, Thank Olivier Grisel)
838 * Detect Intel Haswell for new Macbook.
840 ====================================================================
844 * Update LAPACK to 3.5.0 version
845 * Fixed compatiable issues with Clang and Pathscale compilers.
848 * Optimization on Intel Haswell.
849 * Enable optimization kernels on AMD Bulldozer and Piledriver.
852 * Support ARMv6 and ARMv7 ISA.
853 * Optimization on ARM Cortex-A9.
855 ====================================================================
859 * Support Open64 5.0. (#266)
860 * Add executable stack markings. (#262, Thank Sébastien Fabbro)
861 * Respect user's LDFLAGS (Thank Sébastien Fabbro)
864 * Rollback bulldozer and piledriver kernels to barcelona kernels (#263)
865 We will fix the compuational error bug in bulldozer and piledriver kernels.
867 ====================================================================
871 * Support LSB (Linux Standard Base) 4.1.
873 * Include LAPACK 3.4.2 source codes to the repo.
874 Avoid downloading at compile time.
875 * Add NO_PARALLEL_MAKE flag to disable parallel make.
876 * Create openblas_get_parallel to retrieve information which
877 parallelization model is used by OpenBLAS. (Thank grisuthedragon)
878 * Detect LLVM/Clang compiler. The default compiler is Clang on Mac OS X.
879 * Change LIBSUFFIX from .lib to .a on windows.
880 * A work-around for dtrti_U single thread bug. Replace it with LAPACK codes. (#191)
883 * Optimize c/zgemm, trsm, dgemv_n, ddot, daxpy, dcopy on
884 AMD Bulldozer. (Thank Werner Saar)
885 * Add Intel Haswell support (using Sandybridge optimizations).
887 * Add AMD Piledriver support (using Bulldozer optimizations).
888 * Fix the computational error in zgemm avx kernel on
890 * Fix the overflow bug in gemv.
891 * Fix the overflow bug in multi-threaded BLAS3, getrf when NUM_THREADS
892 is very large.(#214, #221, #246).
894 * Support loongcc (Open64 based) compiler for ICT Loongson 3A/B.
897 * Support Power7 by old Power6 kernels. (#220)
899 ====================================================================
903 * Improved OpenMP performance slightly. (d744c9)
904 * Improved cblas.h compatibility with Intel MKL.(#185)
905 * Fixed the overflowing bug in single thread cholesky factorization.
906 * Fixed the overflowing buffer bug of multithreading hbmv and sbmv.(#174)
909 * Added AMD Bulldozer x86-64 S/DGEMM AVX kernels. (Thank Werner Saar)
910 We will tune the performance in future.
911 * Auto-detect Intel Xeon E7540.
912 * Fixed the overflowing buffer bug of gemv. (#173)
913 * Fixed the bug of s/cdot about invalid reading NAN on x86_64. (#189)
917 ====================================================================
921 * Added NO_SHARED flag to disable generating the shared library.
922 * Compile LAPACKE with ILP64 modle when INTERFACE64=1 (#158)
923 * Export LAPACK 3.4.2 symbols in shared library. (#147)
924 * Only detect the number of physical CPU cores on Mac OSX. (#157)
925 * Fixed NetBSD build. (#155)
926 * Fixed compilation with TARGET=GENERIC. (#160)
928 * Restore the original CPU affinity when calling
929 openblas_set_num_threads(1) (#153)
930 * Fixed a SEGFAULT bug in dgemv_t when m is very large.(#154)
933 ====================================================================
937 * Upgraded LAPACK to 3.4.2 version. (#145)
938 * Provided support for passing CFLAGS, FFLAGS, PFLAGS,
939 FPFLAGS to make. (#137)
940 * f77blas.h:compatibility for compilers without C99 complex
941 number support. (#141)
943 * Added NO_AVX flag. Check OS supporting AVX on runtime. (#139)
944 * Fixed zdot incompatibility ABI issue with GCC 4.7 on
945 Windows 32-bit. (#140)
947 * Fixed the generation of shared library bug.
948 * Fixed the detection bug on the Loongson 3A server.
949 ====================================================================
953 * Fixed LAPACK unstable bug about ?laswp. (#130)
954 * Fixed the shared library bug about unloading the library on
956 * Fixed the compilation failure on BlueGene/P (TARGET=PPC440FP2)
957 Please use gcc and IBM xlf. (#134)
959 * Supported goto_set_num_threads and openblas_set_num_threads
960 APIs in Windows. They can set the number of threads on runtime.
962 ====================================================================
966 * Fixed exporting DLL functions bug on Windows/MingW
967 * Support GNU Hurd (Thank Sylvestre Ledru)
968 * Support kfreebsd kernel (Thank Sylvestre Ledru)
970 * Support Intel Sandy Bridge 22nm desktop/mobile CPU
972 * Improve the detection of SPARC (Thank Sylvestre Ledru)
974 ====================================================================
979 * Fixed the SEGFAULT bug about hyper-theading
980 * Support AMD Bulldozer by using GotoBLAS2 AMD Barcelona codes
982 ====================================================================
986 * Removed the limitation (64) of numbers of CPU cores.
987 Now, it supports 256 cores at max.
988 * Supported clang compiler.
989 * Fixed some build bugs on FreeBSD
991 * Optimized Level-3 BLAS on Intel Sandy Bridge x86-64 by AVX instructions.
992 Please use gcc >= 4.6 or clang >=3.1.
993 * Support AMD Bobcat by using GotoBLAS2 AMD Barcelona codes.
995 ====================================================================
999 * Upgraded LAPACK to 3.4.1 version. (Thank Zaheer Chothia)
1000 * Supported LAPACKE, a C interface to LAPACKE. (Thank Zaheer Chothia)
1001 * Fixed the build bug (MD5 and download) on Mac OSX.
1002 * Auto download CUnit 2.1.2-2 from SF.net with UTEST_CHECK=1.
1003 * Fxied the compatibility issue for compilers without C99 complex number
1004 (e.g. Visual Studio)
1006 * Auto-detect Intel Sandy Bridge Core i7-3xxx & Xeon E7 Westmere-EX.
1007 * Test alpha=Nan in dscale.
1008 * Fixed a SEGFAULT bug in samax on x86 windows.
1010 ====================================================================
1014 * Set soname of shared library on Linux.
1015 * Added LIBNAMESUFFIX flag in Makefile.rule. The user can use
1016 this flag to control the library name, e.g. libopenblas.a,
1017 libopenblas_ifort.a or libopenblas_omp.a.
1018 * Added GEMM_MULTITHREAD_THRESHOLD flag in Makefile.rule.
1019 The lib use single thread in GEMM function with small matrices.
1021 * Used GEMV SSE/SSE2 kernels on x86 32-bit.
1022 * Exported CBLAS functions in Windows DLL.
1024 * Completed Level-3 BLAS optimization on Loongson 3A CPU.
1025 * Improved GEMV performance on Loongson 3A CPU.
1026 * Improved Level-3 BLAS performance on Loongson 3B CPU. (EXPERIMENT)
1028 ====================================================================
1029 Version 0.1 alpha2.5
1032 * Fixed missing "#include <sched.h>" bug on Mac OS X.
1033 Thank Mike Nolta for the patch.
1034 * Upgraded LAPACK to 3.4.0 version
1035 * Fixed a bug on Mac OS X. Don't require SystemStubs on OS X.
1036 SystemStubs does not exist on Lion. Thank Stefan Karpinski.
1037 * Improved README with using OpenMP. Check the internal threads
1038 count less than or equal to omp_get_max_threads()
1040 * Auto-detect Intel Core i6/i7 (Sandy Bridge) CPU with Nehalem assembly kernels
1041 * Fixed some bugs on MingW 64-bit including zgemv, cdot, zdot.
1043 ====================================================================
1044 Version 0.1 alpha2.4
1047 * Fixed a bug about installation. The header file "fblas77.h"
1049 * Fixed #61 a building bug about setting TARGET and DYNAMIC_ARCH.
1050 * Try to handle absolute path of shared library in OSX. (#57)
1051 Thank Dr Kane O'Donnell.
1052 * Changed the installation folder layout to $(PREFIX)/include and
1056 * Fixed #58 zdot/xdot SEGFAULT bug with GCC-4.6 on x86. According
1057 to i386 calling convention, The callee should remove the first
1058 hidden parameter.Thank Mr. John for this patch.
1060 ====================================================================
1061 Version 0.1 alpha2.3
1065 * Added DTB_ENTRIES into dynamic arch setting parameters. Now,
1066 it can read DTB_ENTRIES on runtime. (Refs issue #55 on github)
1068 ====================================================================
1069 Version 0.1 alpha2.2
1073 * Fixed a building bug when DYNAMIC_ARCH=1 & INTERFACE64=1.
1074 (Refs issue #44 on github)
1076 ====================================================================
1077 Version 0.1 alpha2.1
1081 * Stop the build and output the error message when detecting
1082 fortran compiler failed. (Refs issue #42 on github)
1084 ====================================================================
1089 * Fixed blasint undefined bug in <cblas.h> file. Other software
1090 could include this header successfully(Refs issue #13 on github)
1091 * Fixed the SEGFAULT bug on 64 cores. On SMP server, the number
1092 of CPUs or cores should be less than or equal to 64.(Refs issue #14
1094 * Support "void goto_set_num_threads(int num_threads)" and "void
1095 openblas_set_num_threads(int num_threads)" when USE_OPENMP=1
1096 * Added extern "C" to support C++. Thank Tasio for the patch(Refs
1097 issue #21 on github)
1098 * Provided an error message when the arch is not supported.(Refs
1099 issue #19 on github)
1100 * Fixed issue #23. Fixed a bug of f_check script about generating link flags.
1101 * Added openblas_set_num_threads for Fortran.
1102 * Fixed #25 a wrong result of rotmg.
1103 * Fixed a bug about detecting underscore prefix in c_check.
1104 * Print the wall time (cycles) with enabling FUNCTION_PROFILE
1105 * Fixed #35 a build bug with NO_LAPACK=1 & DYNAMIC_ARCH=1
1106 * Added install target. You can use "make install". (Refs #20)
1110 * Fixed #28 a wrong result of dsdot on x86_64.
1111 * Fixed #32 a SEGFAULT bug of zdotc with gcc-4.6.
1112 * Fixed #33 ztrmm bug on Nehalem.
1113 * Work-around #27 the low performance axpy issue with small imput size & multithreads.
1116 * Fixed #28 a wrong result of dsdot on Loongson3A/MIPS64.
1117 * Optimized single/double precision BLAS Level3 on Loongson3A/MIPS64. (Refs #2)
1118 * Optimized single/double precision axpy function on Loongson3A/MIPS64. (Refs #3)
1120 ====================================================================
1125 * Support "make NO_LAPACK=1" to build the library without
1127 * Fixed randomly SEGFAULT when nodemask==NULL with above Linux 2.6.34.
1128 Thank Mr.Ei-ji Nakama providing this patch. (Refs issue #12 on github)
1129 * Added DEBUG=1 rule in Makefile.rule to build debug version.
1130 * Disable compiling quad precision in reference BLAS library(netlib BLAS).
1131 * Added unit testcases in utest/ subdir. Used CUnit framework.
1132 * Supported OPENBLAS_* & GOTO_* environment variables (Pleas see README)
1133 * Imported GotoBLAS2 1.13 BSD version
1136 * On x86 32bits, fixed a bug in zdot_sse2.S line 191. This would casue
1137 zdotu & zdotc failures. Instead, work-around it. (Refs issue #8 #9 on github)
1138 * Modified ?axpy functions to return same netlib BLAS results
1139 when incx==0 or incy==0 (Refs issue #7 on github)
1140 * Modified ?swap functions to return same netlib BLAS results
1141 when incx==0 or incy==0 (Refs issue #6 on github)
1142 * Modified ?rot functions to return same netlib BLAS results
1143 when incx==0 or incy==0 (Refs issue #4 on github)
1144 * Detect Intel Westmere,Intel Clarkdale and Intel Arrandale
1145 to use Nehalem codes.
1146 * Fixed a typo bug about compiling dynamic ARCH library.
1148 * Improve daxpy performance on ICT Loongson 3A.
1149 * Supported ICT Loongson 3A CPU (Refs issue #1 on github)
1150 ====================================================================