2 ====================================================================
7 ` * LAPACK has been updated to 3.9.0 (plus patches up to
9 * CMAKE support has been improved in several areas including
11 * a thread race condition in the GEMM3M kernels was resolved
12 * the "generic" (plain C) gemm beta kernel used by many targets
14 * an optimized version of the LAPACK trtrs functions has been added
15 * an incompatibilty between the LAPACK tests and the OpenBLAS
16 implementation of XERBLA was resolved, removing the numerous
17 warnings about wrong error exits in the former
18 * support for NetBSD has been added
19 * support for compilation with g95 and non-GNU versions of ld
21 * support for compilation with (upcoming) gcc 10 has been added
24 * worked around miscompilation of several POWER8 and POWER9
25 kernels by older versions of gcc
26 * added support for big-endian POWER8 and for compilation on AIX
27 * corrected bugs in the big-endian support for PPC440 and PPC970
28 * DYNAMIC_ARCH support is now available in CMAKE builds as well
31 * performance of DGEMM_BETA and SGEMM_NCOPY has been improved
32 * compilation for 32bit works again
33 * performance of the RPCC function has been improved
34 * improved performance on small systems
35 * DYNAMIC_ARCH support is now available in CMAKE builds as well
36 * cross-compilation from OSX to IOS was simplified
39 * a new AVX512 DGEMM kernel was added and the AVX512 SGEMM kernel
40 was significantly improved
41 * optimized AVX512 kernels for CGEMM and ZGEMM have been added
42 * AVX2 kernels for STRMM, SGEMM, and CGEMM have been significantly
43 sped up and optimized CGEMM3M and ZGEMM3M kernels have been added
44 * added support for QEMU virtual cpus
45 * a compilation problem with PGI and SUN compilers was fixed
46 * Intel "Goldmont plus" is now autodetected
47 * a potential crash on program exit on MS Windows has been fixed
50 * an unwanted case sensitivity in the implementation of LSAME
51 on older 32bit AMD cpus was fixed
54 * Z15 is now supported as Z14
55 * DYNAMIC_ARCH is now available on ZARCH as well
57 ====================================================================
62 * having the gmake special variables TARGET_ARCH or TARGET_MACH
63 defined no longer causes build failures in ctest or utest
64 * defining NO_AFFINITY or USE_TLS to 0 in gmake builds no longer
65 has the same effect as setting them to 1
66 * a new test program was added to allow checking the library for
68 * a new option USE_LOCKING was added to ensure thread safety when
69 OpenBLAS itself is built without multithreading but will be
70 called from multiple threads.
71 * a build failure on Linux with glibc versions earlier than 2.5
73 * a runtime error with CPU enumeration (and NO_AFFINITY not set)
74 on glibc 2.6 was fixed
75 * NO_AFFINITY was added to the CMAKE options (and defaults to being
76 active on Linux, as in the gmake builds)
79 * the build-time logic for detection of AVX512 availability in
80 the processor and compiler was fixed
81 * gmake builds on OSX now set the internal name of the library to
82 libopenblas.0.dylib (consistent with CMAKE)
83 * the Haswell DGEMM kernel received a significant speedup through
84 improved prefetch and load instructions
85 * performance of DGEMM, DTRMM, DTRSM and ZDOT on Zen/Zen2 was markedly
86 increased by avoiding vpermpd instructions
87 * the SKYLAKEX (AVX512) DGEMM helper functions have now been disabled
88 to fix remaining errors in DGEMM, DSYMM and DTRMM
91 * added support for building on FreeBSD/powerpc64 and FreeBSD/ppc970
92 * added optimized kernels for POWER9 SGEMM and STRMM
95 * fixed the softfp implementations of xAMAX and IxAMAX
96 * removed the predefined -march= flags on both ARMV5 and ARMV6 as
97 they were appropriate for only a subset of platforms
99 ====================================================================
104 * the build tools now check that a given cpu TARGET is actually valid
105 * the build-time check of system features (c_check) has been made
106 less dependent on particular perl features (this should mainly
107 benefit building on Windows)
108 * several problem with the ReLAPACK integration were fixed,
109 including INTERFACE64 support and building a shared library
110 * building with CMAKE on BSD systems was improved
111 * a non-absolute SUM function was added based on the
112 existing optimized code for ASUM
113 * CBLAS interfaces to the IxMIN and IxMAX functions were added
114 * a name clash between LAPACKE and BOOST headers was resolved
115 * CMAKE builds with OpenMP failed to include the appropriate getrf_parallel
117 * a crash on thread (key) deletion with the USE_TLS=1 memory management
119 * restored several earlier fixes, in particular for OpenMP performance,
120 building on BSD, and calling fork on CYGWIN, which had inadvertently
121 been dropped in the 0.3.3 rewrite of the memory management code.
124 * the AVX512 DGEMM kernel has been disabled again due to unsolved problems
125 * building with old versions of MSVC was fixed
126 * it is now possible to build a static library on Windows with CMAKE
127 * accessing environment variables on CYGWIN at run time was fixed
128 * the CMAKE build system now recognizes 32bit userspace on 64bit hardware
129 * Intel "Denverton" atom and Hygon "Dhyana" zen CPUs are now autodetected
130 * building for DYNAMIC_ARCH with a DYNAMIC_LIST of targets is now supported
132 * building for DYNAMIC_ARCH with GENERIC as the default target is now supported
133 * a buffer overflow in the SSE GEMM kernel for Intel Nano targets was fixed
134 * assembly bugs involving undeclared modification of input operands were fixed
135 in the AXPY, DOT, GEMV, GER, SCAL, SYMV and TRSM microkernels for Nehalem,
136 Sandybridge, Haswell, Bulldozer and Piledriver. These would typically cause
137 test failures or segfaults when compiled with recent versions of gcc from 8 onward.
138 * a similar bug was fixed in the blas_quickdivide code used to split workloads
140 * a bug in the IxMIN implementation for the GENERIC target made it return the result of IxMAX
141 * fixed building on SkylakeX systems when either the compiler or the (emulated) operating
142 environment does not support AVX512
143 * improved GEMM performance on ZEN targets
146 * build failures caused by the recently added checks for AVX512 were fixed
147 * an inline assembly bug involving undeclared modification of an input argument was
148 fixed in the blas_quickdivide code used to split workloads in most functions
149 * a bug in the IMIN implementation for the GENERIC target made it return the result of IMAX
152 * a bug in the IMIN implementation made it return the result of IMAX
155 * single precision BLAS1/2 functions have received optimized POWER8 kernels
156 * POWER9 is now a separate target, with an optimized DGEMM/DTRMM kernel
157 * building on PPC970 systems under OSX Leopard or Tiger is now supported
158 * out-of-bounds memory accesses in the gemm_beta microkernels were fixed
159 * building a shared library on AIX is now supported for POWER6
160 * DYNAMIC_ARCH support has been added for POWER6 and newer
163 * corrected xDOT behaviour with zero INC_X or INC_Y
164 * a bug in the IMIN implementation made it return the result of IMAX
167 * added support for HiSilicon TSV110 cpus
168 * the CMAKE build system now recognizes 32bit userspace on 64bit hardware
169 * cross-compilation with CMAKE now works again
170 * a bug in the IMIN implementation made it return the result of IMAX
171 * ARMV8 builds with the BINARY=32 option are now automatically handled as ARMV7
174 * optimized microkernels for single precicion BLAS1/2 functions have been added
177 ====================================================================
182 * loop unrolling in TRMV has been enabled again.
183 * A domain error in the thread workload distribution for SYRK
185 * gmake builds will now automatically add -fPIC to the build
186 options if the platform requires it.
187 * a pthreads key leakage (and associate crash on dlclose) in
188 the USE_TLS codepath was fixed.
189 * building of the utest cases on systems that do not provide
190 an implementation of complex.h was fixed.
193 * the SkylakeX code was changed to compile on OSX.
194 * unwanted application of the -march=skylake-avx512 option
195 to the common code parts of a DYNAMIC_ARCH build was fixed.
196 * improved performance of SGEMM for small workloads on Skylake X.
197 * performance of SGEMM and DGEMM was improved on Haswell.
200 * a configuration error that broke the CNRM2 kernel was corrected.
201 * compilation of the GEMM kernels with CMAKE was fixed.
202 * DYNAMIC_ARCH builds are now available with CMAKE as well.
203 * using CMAKE for cross-compilation to the new cpu TARGETs
204 introduced in 0.3.4 now works.
207 * a problem in cpu autodetection for AIX has been corrected.
209 ====================================================================
214 * the new, experimental thread-local memory allocation had
215 inadvertently been left enabled for gmake builds in 0.3.3
216 despite the announcement. It is now disabled by default, and
217 single-threaded builds will keep using the old allocator even
218 if the USE_TLS option is turned on.
219 * OpenBLAS will now provide enough buffer space for at least 50
221 * The output of openblas_get_config() now contains the version
223 * A serious thread safety bug in GEMV operation with small M and
224 large N size has been fixed.
225 * The code will now automatically call blas_thread_init after a
226 fork if needed before handling a call to openblas_set_num_threads
227 * Accesses to parallelized level3 functions from multiple callers
228 are now serialized to avoid thread races (unless using OpenMP).
229 This should provide better performance than the known-threadsafe
230 (but non-default) USE_SIMPLE_THREADED_LEVEL3 option.
231 * When building LAPACK with gfortran, -frecursive is now (again)
232 enabled by default to ensure correct behaviour.
233 * The OpenBLAS version cblas.h now supports both CBLAS_ORDER and
234 CBLAS_LAYOUT as the name of the matrix row/column order option.
235 * Externally set LDFLAGS are now passed through to the final compile/link
236 steps to facilitate setting platform-specific linker flags.
237 * A potential race condition during the build of LAPACK (that would
238 usually manifest itself as a failure to build TESTING/MATGEN) has been
240 * xHEMV has been changed to stay single-threaded for small input sizes
241 where the overhead of multithreading exceeds any possible gains
242 * CSWAP and ZSWAP have been limited to a single thread except on ARMV8 or
243 ThunderX hardware with sizable input.
244 * Linker flags for the PGI compiler have been updated
245 * Behaviour of AXPY with zero increments is now handled in the C interface,
246 correcting the result on at least Intel Atom.
247 * The result matrix from calling SGELSS with an all-zero input matrix is
248 now zeroed completely.
251 * Autodetection of AMD Ryzen2 has been fixed (again).
252 * CMAKE builds now support labeling of an INTERFACE64=1 build of
253 the library with the _64 suffix.
254 * AVX512 version of DGEMM has been added and the AVX512 SGEMM kernel
255 has been sped up by rewriting with C intrinsics
256 * Fixed compilation on RHEL5/CENTOS5 (issue with typename __WAIT_STATUS)
259 * added support for building on AIX (with gcc and GNU tools from AIX Toolbox).
260 * CPU type detection has been implemented for AIX.
261 * CPU type detection has been fixed for NETBSD.
264 * AXPY on LOONGSON3A has been corrected to pass "zero increment" utest.
265 * DSDOT on LOONGSON3A has been fixed.
266 * the SGEMM microkernel has been hardened against potential data loss.
269 * DYNAMic_ARCH support is now available for 64bit ARM
270 * cross-compiling for ARMV8 under iOS now works.
271 * cpu-specific code has been rearranged to make better use of both
272 hardware commonalities and model-specific compiler optimizations.
273 * XGENE1 has been removed as a TARGET, superseded by the improved generic
277 * Older assembly mnemonics have been converted to UAL form to allow
278 building with clang 7.0
279 * Cross compiling LAPACKE for Android has been fixed again (broken by
280 update to LAPACK 3.7.0 some while ago).
282 ====================================================================
287 * thread memory allocation has been switched back to the method
288 used before version 0.3.1 due to unexpected problems caused by
289 the new code under some circumstances. A new compile-time option
290 USE_TLS has been added to enable the new code, and it is hoped
291 that this can become the default again in the next version.
292 * LAPAck PR272 has been integrated, which fixes spurious errors
293 in DSYEVR and related functions caused by missing conversion
294 from ILAENV to ILAENV_2STAGE in several _2stage routines.
295 * the cmake-generated OpenBLASConfig.cmake now uses correct case
296 for the name of the library
297 * added support for Haiku OS
300 * added AVX512 implementations of SDOT, DDOT, SAXPY, DAXPY,
301 DSCAL, DGEMVN and DSYMVL
302 * added a workaround for a cygwin issue that prevented compilation
306 * added autodetection of Z14
307 * fixed TRMM errors in the generic target
309 ====================================================================
314 * fixes for regressions caused by the rewrite of the thread
315 initialization code in 0.3.1
318 * fixed cpu autodetection for the BSDs
321 * fixed utest errors in AXPY, DSDOT, ROT and SWAP
324 * added autodetection of AMD Ryzen 2
325 * fixed build with older versions of MSVC
327 ====================================================================
332 * rewritten thread initialization code with significantly reduced overhead
333 * added CBLAS interfaces to the IxAMIN BLAS extension functions
334 * fixed the lapack-test target
335 * CMAKE builds now create an OpenBLASConfig.cmake file
336 * ZAXPY now uses a single thread for small input sizes
337 * the LAPACK code was updated from Reference-LAPACK/lapack#253
338 (fixing LAPACKE interfaces to Aasen's functions)
341 * corrected CROT and ZROT behaviour with zero INC_X
344 * corrected xDOT behaviour with zero INC_X or INC_Y
347 * retired some older targets of DYNAMIC_ARCH builds to a new option DYNAMIC_OLDER,
348 this affects PENRYN,DUNNINGTON,OPTERON,OPTERON_SSE3,BOBCAT,ATOM and NANO
349 (which will still be supported via the slower PRESCOTT kernels when this option is not set)
350 * added an option DYNAMIC_LIST that (used in conjunction with DYNAMIC_ARCH) allows to
351 specify the list of x86_64 targets to include. Any target not on the list will be supported
352 by the Sandybridge or Nehalem kernels if available, or by Prescott.
353 * improved SWITCH_RATIO on Haswell for increased GEMM throughput
354 * added initial support for Intel Skylake X, including an AVX512 SGEMM kernel
355 * added autodetection of Intel Cannon Lake series as Skylake X
356 * added a default L2 cache size for hypervisors that return zero here (Chromebook)
357 * fixed a name clash with recent Windows10 headers that broke the build with (at least)
358 recent mingw from MSYS2
359 * fixed a link error in mixed clang/gfortran builds with OpenMP
360 * updated the OSX deployment target to 10.8
361 * switched on parallel make for builds on MS Windows by default
364 * fixed SSWAP and DSWAP behaviour with zero INC_X and INC_Y
366 ====================================================================
371 * fixed some more thread race and locking bugs
372 * added preliminary support for calling an OpenMP build of the library from multiple threads
373 * removed performance impact of thread locks added in 0.2.20 on OpenMP code
374 * general code cleanup
375 * optimized DSDOT implementation
376 * improved thread distribution for GEMM
377 * corrected IMATCOPY/OMATCOPY implementation
378 * fixed out-of-bounds accesses in the multithreaded xBMV/xPMV and SYMV implementations
379 * cmake build improvements
380 * pkgconfig file now contains build options
381 * openblas_get_config() now reports USE_OPENMP and NUM_THREADS settings used for the build
382 * corrections and improvements for systems with more than 64 cpus
383 * LAPACK code updated to 3.8.0 including later fixes
384 * added ReLAPACK, a recursive implementation of several LAPACK functions
385 * Rewrote ROTMG to handle cases that the netlib code failed to address
386 * Disabled (broken) multithreading code for xTRMV
387 * corrected prototypes of complex CBLAS functions to make our cblas.h match the generally accepted standard
388 * shared memory access failures on startup are now handled more gracefully
389 * restored utests from earlier releases (and made them pass on all affected systems)
392 * several fixes for cpu autodetection
395 * corrected vector register overwriting in several Power8 kernels
396 * optimized additional BLAS functions
399 * added support for CortexA53 and A72
400 * added autodetection for ThunderX2T99
401 * made most optimized kernels the default for generic ARMv8 targets
404 * parallelized DDOT kernel for Haswell
405 * changed alignment directives in assembly kernels to boost performance on OSX
406 * fixed register handling in the GEMV microkernels (bug exposed by gcc7)
407 * added support for building on OpenBSD and Dragonfly
408 * updated compiler options to work with Intel release 2018
409 * support fully optimized build with clang/flang on Microsoft Windows
410 * fixed building on AIX
413 * added optimized BLAS 1/2 functions
416 * fixed cpu autodetection helper code
417 * added mips32 1004K cpu (Mediatek MT7621 and similar SoC)
418 * added mips64 I6500 cpu
420 ====================================================================
425 * Improved CMake support
426 * Fixed several thread race and locking bugs
427 * Fixed default LAPACK optimization level
428 * Updated LAPACK to 3.7.0
429 * Added ReLAPACK (https://github.com/HPAC/ReLAPACK, make BUILD_RELAPACK=1)
432 * Optimizations for Power9
433 * Fixed several Power8 assembly bugs
436 * New optimized Vulcan and ThunderX2T99 targets
437 * Support for ARMV7 SOFT_FP ABI (make ARM_SOFTFP_ABI=1)
438 * Detect all cpu cores including offline ones
439 * Fix compilation with CLANG
440 * Support building a shared library for Android
443 * Fixed several threading issues
444 * Fix compilation with CLANG
447 * Detect Intel Bay Trail and Apollo Lake
448 * Detect Intel Sky Lake and Kaby Lake
449 * Detect Intel Knights Landing
450 * Detect AMD A8, A10, A12 and Ryzen
451 * Support 64bit builds with Visual Studio
452 * Fix building with Intel and PGI compilers
453 * Fix building with MINGW and TDM-GCC
454 * Fix cmake builds for Haswell and related cpus
455 * Fix building for Sandybridge with CLANG 3.9
456 * Add support for the FLANG compiler
459 * New target z13 with BLAS3 optimizations
461 ====================================================================
465 * Improved cross compiling.
466 * Fix the bug on musl libc.
469 * Optimize BLAS on Power8
470 * Fixed Julia+OpenBLAS bugs on Power8
473 * Optimize BLAS on MIPS P5600 and I6400 (Thanks, Shivraj Patil, Kaustubh Raste)
476 * Improved on ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
479 ====================================================================
483 * If you set MAKE_NB_JOBS flag less or equal than zero,
484 make will be without -j.
487 * Support building Visual Studio static library. (#813, Thanks, theoractice)
488 * Fix bugs to pass buidbot CI tests (http://build.openblas.net)
491 * Provide DGEMM 8x4 kernel for Cortex-A57 (Thanks, Ashwin Sekhar T K)
494 * Optimize S and C BLAS3 on Power8
495 * Optimize BLAS2/1 on Power8
497 ====================================================================
501 * Enable BUILD_LAPACK_DEPRECATED=1 by default.
503 ====================================================================
507 * Avoid potential getenv segfault. (#716)
508 * Import LAPACK svn bugfix #142-#147,#150-#155
511 * Optimize c/zgemv for AMD Bulldozer, Piledriver, Steamroller
512 * Fix bug with scipy linalg test.
515 * Improve DGEMM for ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
518 * Optimize D and Z BLAS3 functions for Power8.
520 ====================================================================
524 * Upgrade LAPACK to 3.6.0 version.
525 Add BUILD_LAPACK_DEPRECATED option in Makefile.rule to build
526 LAPACK deprecated functions.
527 * Add MAKE_NB_JOBS option in Makefile.
528 Force number of make jobs.This is particularly
529 useful when using distcc. (#735. Thanks, Jerome Robert.)
530 * Redesign unit test. Run unit/regression test at every build (Travis-CI and Appveyor).
531 * Disable multi-threading for small size swap and ger. (#744. Thanks, Jerome Robert)
532 * Improve small zger, zgemv, ztrmv using stack alloction (#727. Thanks, Jerome Robert)
533 * Let openblas_get_num_threads return the number of active threads.
534 (#760. Thanks, Jerome Robert)
535 * Support illumos(OmniOS). (#749. Thanks, Lauri Tirkkonen)
536 * Fix LAPACK Dormbr, Dormlq bug. (#711, #713. Thanks, Brendan Tracey)
537 * Update scipy benchmark script. (#745. Thanks, John Kirkham)
540 * Optimize trsm kernels for AMD Bulldozer, Piledriver, Steamroller.
541 * Detect Intel Avoton.
542 * Detect AMD Trinity, Richland, E2-3200.
543 * Fix gemv performance bug on Mac OSX Intel Haswell.
544 * Fix some bugs with CMake and Visual Studio
547 * Support and optimize Cortex-A57 AArch64.
548 (#686. Thanks, Ashwin Sekhar TK)
549 * Fix Android build on ARMV7 (#778. Thanks, Paul Mustiere)
550 * Update ARMV6 kernels.
553 * Fix detection of POWER architecture
554 (#684. Thanks, Sebastien Villemot)
556 ====================================================================
560 * Support cmake on x86/x86-64. Natively compiling on MS Visual Studio.
561 (experimental. Thank Hank Anderson for the initial cmake porting work.)
563 On Linux and Mac OSX, OpenBLAS cmake supports assembly kernels.
568 On Windows MS Visual Studio, OpenBLAS cmake only support C kernels.
569 (OpenBLAS uses AT&T style assembly, which is not supported by MSVC.)
570 e.g. cmake -G "Visual Studio 12 Win64" .
571 Open OpenBLAS.sln and build.
573 * Enable MAX_STACK_ALLOC flags by default.
574 Improve ger and gemv for small matrices.
575 * Improve gemv parallel with small m and large n case.
576 * Improve ?imatcopy when lda==ldb (#633. Thanks, Martin Koehler)
577 * Add vecLib benchmarks (#565. Thanks, Andreas Noack.)
578 * Fix LAPACK lantr for row major matrices (#634. Thanks, Dan Kortschak)
579 * Fix LAPACKE lansy (#640. Thanks, Dan Kortschak)
580 * Import bug fixes for LAPACKE s/dormlq, c/zunmlq
581 * Raise the signal when pthread_create fails (#668. Thanks, James K. Lowden)
582 * Remove g77 from compiler list.
583 * Enable AppVeyor Windows CI.
586 * Support pure C generic kernels for x86/x86-64.
587 * Support Intel Boardwell and Skylake by Haswell kernels.
588 * Support AMD Excavator by Steamroller kernels.
589 * Optimize s/d/c/zdot for Intel SandyBridge and Haswell.
590 * Optimize s/d/c/zdot for AMD Piledriver and Steamroller.
591 * Optimize s/d/c/zapxy for Intel SandyBridge and Haswell.
592 * Optimize s/d/c/zapxy for AMD Piledriver and Steamroller.
593 * Optimize d/c/zscal for Intel Haswell, dscal for Intel SandyBridge.
594 * Optimize d/c/zscal for AMD Bulldozer, Piledriver and Steamroller.
595 * Optimize s/dger for Intel SandyBridge.
596 * Optimize s/dsymv for Intel SandyBridge.
597 * Optimize ssymv for Intel Haswell.
598 * Optimize dgemv for Intel Nehalem and Haswell.
599 * Optimize dtrmm for Intel Haswell.
602 * Support Android NDK armeabi-v7a-hard ABI (-mfloat-abi=hard)
603 e.g. make HOSTCC=gcc CC=arm-linux-androideabi-gcc NO_LAPACK=1 TARGET=ARMV7
604 * Fix lock, rpcc bugs (#616, #617. Thanks, Grazvydas Ignotas)
606 * Support ppc64le platform (ELF ABI v2. #612. Thanks, Matthew Brandyberry.)
607 * Support POWER7/8 by POWER6 kernels. (#612. Thanks, Fábio Perez.)
609 ====================================================================
613 * Improve OpenBLASConfig.cmake. (#474, #475. Thanks, xantares.)
614 * Improve ger and gemv for small matrices by stack allocation.
615 e.g. make -DMAX_STACK_ALLOC=2048 (#482. Thanks, Jerome Robert.)
616 * Introduce openblas_get_num_threads and openblas_get_num_procs.
617 (#497. Thanks, Erik Schnetter.)
618 * Add ATLAS-style ?geadd function. (#509. Thanks, Martin Köhler.)
619 * Fix c/zsyr bug with negative incx. (#492.)
620 * Fix race condition during shutdown causing a crash in
621 gotoblas_set_affinity(). (#508. Thanks, Ton van den Heuvel.)
624 * Support AMD Streamroller.
627 * Add Cortex-A9 and Cortex-A15 targets.
629 ====================================================================
633 * Add SYMBOLPREFIX and SYMBOLSUFFIX makefile options
634 for adding a prefix or suffix to all exported symbol names
635 in the shared library.(#459, Thanks Tony Kelman)
636 * Provide OpenBLASConfig.cmake at installation.
637 * Fix Fortran compiler detection on FreeBSD.
638 (#470, Thanks Mike Nolta)
642 * Add generic kernel files for x86-64. make TARGET=GENERIC
643 * Fix a bug of sgemm kernel on Intel Sandy Bridge.
644 * Fix c_check bug on some amd64 systems. (#471, Thanks Mike Nolta)
647 * Support APM's X-Gene 1 AArch64 processors.
648 Optimize trmm and sgemm. (#465, Thanks Dave Nuechterlein)
650 ====================================================================
654 * Added CBLAS interface for ?omatcopy and ?imatcopy.
655 * Enable ?gemm3m functions.
656 * Added benchmark for ?gemm3m.
657 * Optimized multithreading lower limits.
658 * Disabled SYMM3M and HEMM3M functions
659 because of segment violations.
662 * Improved axpy and symv performance on AMD Bulldozer.
663 * Improved gemv performance on modern Intel and AMD CPUs.
665 ====================================================================
669 * Added some benchmark codes.
670 * Fix link error on Linux/musl.(Thanks Isaac Dunham)
673 * Improved s/c/zgemm performance for Intel Haswell.
674 * Improved s/d/c/zgemv performance.
675 * Support the big numa machine.(EXPERIMENT)
678 * Fix detection when cpuinfo uses "Processor". (Thanks Isaiah)
680 ====================================================================
684 * Added BLAS extensions as following.
685 s/d/c/zaxpby, s/d/c/zimatcopy, s/d/c/zomatcopy.
686 * Added OPENBLAS_CORETYPE environment for dynamic_arch. (a86d34)
687 * Added NO_AVX2 flag for old binutils. (#401)
688 * Support outputing the CPU corename on runtime.(#407)
689 * Patched LAPACK to fix bug 114, 117, 118.
690 (http://www.netlib.org/lapack/bug_list.html)
691 * Disabled ?gemm3m for a work-around fix. (#400)
693 * Fixed lots of bugs for optimized kernels on sandybridge,Haswell,
694 bulldozer, and piledriver.
695 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
698 * Improved LAPACK testing.
700 ====================================================================
704 * Improved the result for LAPACK testing. (#372)
705 * Installed DLL to prefix/bin instead of prefix/lib. (#366)
706 * Build import library on Windows.(#374)
708 * To improve LAPACK testing, we fallback some kernels. (#372)
709 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
711 ====================================================================
715 * Added OPENBLAS_VERBOSE environment variable.(#338)
716 * Make OpenBLAS thread-pool resilient to fork via pthread_atfork.
717 (#294, Thank Olivier Grisel)
721 * Detect Intel Haswell for new Macbook.
723 ====================================================================
727 * Update LAPACK to 3.5.0 version
728 * Fixed compatiable issues with Clang and Pathscale compilers.
731 * Optimization on Intel Haswell.
732 * Enable optimization kernels on AMD Bulldozer and Piledriver.
735 * Support ARMv6 and ARMv7 ISA.
736 * Optimization on ARM Cortex-A9.
738 ====================================================================
742 * Support Open64 5.0. (#266)
743 * Add executable stack markings. (#262, Thank Sébastien Fabbro)
744 * Respect user's LDFLAGS (Thank Sébastien Fabbro)
747 * Rollback bulldozer and piledriver kernels to barcelona kernels (#263)
748 We will fix the compuational error bug in bulldozer and piledriver kernels.
750 ====================================================================
754 * Support LSB (Linux Standard Base) 4.1.
756 * Include LAPACK 3.4.2 source codes to the repo.
757 Avoid downloading at compile time.
758 * Add NO_PARALLEL_MAKE flag to disable parallel make.
759 * Create openblas_get_parallel to retrieve information which
760 parallelization model is used by OpenBLAS. (Thank grisuthedragon)
761 * Detect LLVM/Clang compiler. The default compiler is Clang on Mac OS X.
762 * Change LIBSUFFIX from .lib to .a on windows.
763 * A work-around for dtrti_U single thread bug. Replace it with LAPACK codes. (#191)
766 * Optimize c/zgemm, trsm, dgemv_n, ddot, daxpy, dcopy on
767 AMD Bulldozer. (Thank Werner Saar)
768 * Add Intel Haswell support (using Sandybridge optimizations).
770 * Add AMD Piledriver support (using Bulldozer optimizations).
771 * Fix the computational error in zgemm avx kernel on
773 * Fix the overflow bug in gemv.
774 * Fix the overflow bug in multi-threaded BLAS3, getrf when NUM_THREADS
775 is very large.(#214, #221, #246).
777 * Support loongcc (Open64 based) compiler for ICT Loongson 3A/B.
780 * Support Power7 by old Power6 kernels. (#220)
782 ====================================================================
786 * Improved OpenMP performance slightly. (d744c9)
787 * Improved cblas.h compatibility with Intel MKL.(#185)
788 * Fixed the overflowing bug in single thread cholesky factorization.
789 * Fixed the overflowing buffer bug of multithreading hbmv and sbmv.(#174)
792 * Added AMD Bulldozer x86-64 S/DGEMM AVX kernels. (Thank Werner Saar)
793 We will tune the performance in future.
794 * Auto-detect Intel Xeon E7540.
795 * Fixed the overflowing buffer bug of gemv. (#173)
796 * Fixed the bug of s/cdot about invalid reading NAN on x86_64. (#189)
800 ====================================================================
804 * Added NO_SHARED flag to disable generating the shared library.
805 * Compile LAPACKE with ILP64 modle when INTERFACE64=1 (#158)
806 * Export LAPACK 3.4.2 symbols in shared library. (#147)
807 * Only detect the number of physical CPU cores on Mac OSX. (#157)
808 * Fixed NetBSD build. (#155)
809 * Fixed compilation with TARGET=GENERIC. (#160)
811 * Restore the original CPU affinity when calling
812 openblas_set_num_threads(1) (#153)
813 * Fixed a SEGFAULT bug in dgemv_t when m is very large.(#154)
816 ====================================================================
820 * Upgraded LAPACK to 3.4.2 version. (#145)
821 * Provided support for passing CFLAGS, FFLAGS, PFLAGS,
822 FPFLAGS to make. (#137)
823 * f77blas.h:compatibility for compilers without C99 complex
824 number support. (#141)
826 * Added NO_AVX flag. Check OS supporting AVX on runtime. (#139)
827 * Fixed zdot incompatibility ABI issue with GCC 4.7 on
828 Windows 32-bit. (#140)
830 * Fixed the generation of shared library bug.
831 * Fixed the detection bug on the Loongson 3A server.
832 ====================================================================
836 * Fixed LAPACK unstable bug about ?laswp. (#130)
837 * Fixed the shared library bug about unloading the library on
839 * Fixed the compilation failure on BlueGene/P (TARGET=PPC440FP2)
840 Please use gcc and IBM xlf. (#134)
842 * Supported goto_set_num_threads and openblas_set_num_threads
843 APIs in Windows. They can set the number of threads on runtime.
845 ====================================================================
849 * Fixed exporting DLL functions bug on Windows/MingW
850 * Support GNU Hurd (Thank Sylvestre Ledru)
851 * Support kfreebsd kernel (Thank Sylvestre Ledru)
853 * Support Intel Sandy Bridge 22nm desktop/mobile CPU
855 * Improve the detection of SPARC (Thank Sylvestre Ledru)
857 ====================================================================
862 * Fixed the SEGFAULT bug about hyper-theading
863 * Support AMD Bulldozer by using GotoBLAS2 AMD Barcelona codes
865 ====================================================================
869 * Removed the limitation (64) of numbers of CPU cores.
870 Now, it supports 256 cores at max.
871 * Supported clang compiler.
872 * Fixed some build bugs on FreeBSD
874 * Optimized Level-3 BLAS on Intel Sandy Bridge x86-64 by AVX instructions.
875 Please use gcc >= 4.6 or clang >=3.1.
876 * Support AMD Bobcat by using GotoBLAS2 AMD Barcelona codes.
878 ====================================================================
882 * Upgraded LAPACK to 3.4.1 version. (Thank Zaheer Chothia)
883 * Supported LAPACKE, a C interface to LAPACKE. (Thank Zaheer Chothia)
884 * Fixed the build bug (MD5 and download) on Mac OSX.
885 * Auto download CUnit 2.1.2-2 from SF.net with UTEST_CHECK=1.
886 * Fxied the compatibility issue for compilers without C99 complex number
889 * Auto-detect Intel Sandy Bridge Core i7-3xxx & Xeon E7 Westmere-EX.
890 * Test alpha=Nan in dscale.
891 * Fixed a SEGFAULT bug in samax on x86 windows.
893 ====================================================================
897 * Set soname of shared library on Linux.
898 * Added LIBNAMESUFFIX flag in Makefile.rule. The user can use
899 this flag to control the library name, e.g. libopenblas.a,
900 libopenblas_ifort.a or libopenblas_omp.a.
901 * Added GEMM_MULTITHREAD_THRESHOLD flag in Makefile.rule.
902 The lib use single thread in GEMM function with small matrices.
904 * Used GEMV SSE/SSE2 kernels on x86 32-bit.
905 * Exported CBLAS functions in Windows DLL.
907 * Completed Level-3 BLAS optimization on Loongson 3A CPU.
908 * Improved GEMV performance on Loongson 3A CPU.
909 * Improved Level-3 BLAS performance on Loongson 3B CPU. (EXPERIMENT)
911 ====================================================================
915 * Fixed missing "#include <sched.h>" bug on Mac OS X.
916 Thank Mike Nolta for the patch.
917 * Upgraded LAPACK to 3.4.0 version
918 * Fixed a bug on Mac OS X. Don't require SystemStubs on OS X.
919 SystemStubs does not exist on Lion. Thank Stefan Karpinski.
920 * Improved README with using OpenMP. Check the internal threads
921 count less than or equal to omp_get_max_threads()
923 * Auto-detect Intel Core i6/i7 (Sandy Bridge) CPU with Nehalem assembly kernels
924 * Fixed some bugs on MingW 64-bit including zgemv, cdot, zdot.
926 ====================================================================
930 * Fixed a bug about installation. The header file "fblas77.h"
932 * Fixed #61 a building bug about setting TARGET and DYNAMIC_ARCH.
933 * Try to handle absolute path of shared library in OSX. (#57)
934 Thank Dr Kane O'Donnell.
935 * Changed the installation folder layout to $(PREFIX)/include and
939 * Fixed #58 zdot/xdot SEGFAULT bug with GCC-4.6 on x86. According
940 to i386 calling convention, The callee should remove the first
941 hidden parameter.Thank Mr. John for this patch.
943 ====================================================================
948 * Added DTB_ENTRIES into dynamic arch setting parameters. Now,
949 it can read DTB_ENTRIES on runtime. (Refs issue #55 on github)
951 ====================================================================
956 * Fixed a building bug when DYNAMIC_ARCH=1 & INTERFACE64=1.
957 (Refs issue #44 on github)
959 ====================================================================
964 * Stop the build and output the error message when detecting
965 fortran compiler failed. (Refs issue #42 on github)
967 ====================================================================
972 * Fixed blasint undefined bug in <cblas.h> file. Other software
973 could include this header successfully(Refs issue #13 on github)
974 * Fixed the SEGFAULT bug on 64 cores. On SMP server, the number
975 of CPUs or cores should be less than or equal to 64.(Refs issue #14
977 * Support "void goto_set_num_threads(int num_threads)" and "void
978 openblas_set_num_threads(int num_threads)" when USE_OPENMP=1
979 * Added extern "C" to support C++. Thank Tasio for the patch(Refs
981 * Provided an error message when the arch is not supported.(Refs
983 * Fixed issue #23. Fixed a bug of f_check script about generating link flags.
984 * Added openblas_set_num_threads for Fortran.
985 * Fixed #25 a wrong result of rotmg.
986 * Fixed a bug about detecting underscore prefix in c_check.
987 * Print the wall time (cycles) with enabling FUNCTION_PROFILE
988 * Fixed #35 a build bug with NO_LAPACK=1 & DYNAMIC_ARCH=1
989 * Added install target. You can use "make install". (Refs #20)
993 * Fixed #28 a wrong result of dsdot on x86_64.
994 * Fixed #32 a SEGFAULT bug of zdotc with gcc-4.6.
995 * Fixed #33 ztrmm bug on Nehalem.
996 * Work-around #27 the low performance axpy issue with small imput size & multithreads.
999 * Fixed #28 a wrong result of dsdot on Loongson3A/MIPS64.
1000 * Optimized single/double precision BLAS Level3 on Loongson3A/MIPS64. (Refs #2)
1001 * Optimized single/double precision axpy function on Loongson3A/MIPS64. (Refs #3)
1003 ====================================================================
1008 * Support "make NO_LAPACK=1" to build the library without
1010 * Fixed randomly SEGFAULT when nodemask==NULL with above Linux 2.6.34.
1011 Thank Mr.Ei-ji Nakama providing this patch. (Refs issue #12 on github)
1012 * Added DEBUG=1 rule in Makefile.rule to build debug version.
1013 * Disable compiling quad precision in reference BLAS library(netlib BLAS).
1014 * Added unit testcases in utest/ subdir. Used CUnit framework.
1015 * Supported OPENBLAS_* & GOTO_* environment variables (Pleas see README)
1016 * Imported GotoBLAS2 1.13 BSD version
1019 * On x86 32bits, fixed a bug in zdot_sse2.S line 191. This would casue
1020 zdotu & zdotc failures. Instead, work-around it. (Refs issue #8 #9 on github)
1021 * Modified ?axpy functions to return same netlib BLAS results
1022 when incx==0 or incy==0 (Refs issue #7 on github)
1023 * Modified ?swap functions to return same netlib BLAS results
1024 when incx==0 or incy==0 (Refs issue #6 on github)
1025 * Modified ?rot functions to return same netlib BLAS results
1026 when incx==0 or incy==0 (Refs issue #4 on github)
1027 * Detect Intel Westmere,Intel Clarkdale and Intel Arrandale
1028 to use Nehalem codes.
1029 * Fixed a typo bug about compiling dynamic ARCH library.
1031 * Improve daxpy performance on ICT Loongson 3A.
1032 * Supported ICT Loongson 3A CPU (Refs issue #1 on github)
1033 ====================================================================