2 ====================================================================
7 * fixes for regressions caused by the rewrite of the thread
8 initialization code in 0.3.1
11 * fixed cpu autodetection for the BSDs
14 * fixed utest errors in AXPY, DSDOT, ROT and SWAP
17 * added autodetection of AMD Ryzen 2
18 * fixed build with older versions of MSVC
20 ====================================================================
25 * rewritten thread initialization code with significantly reduced overhead
26 * added CBLAS interfaces to the IxAMIN BLAS extension functions
27 * fixed the lapack-test target
28 * CMAKE builds now create an OpenBLASConfig.cmake file
29 * ZAXPY now uses a single thread for small input sizes
30 * the LAPACK code was updated from Reference-LAPACK/lapack#253
31 (fixing LAPACKE interfaces to Aasen's functions)
34 * corrected CROT and ZROT behaviour with zero INC_X
37 * corrected xDOT behaviour with zero INC_X or INC_Y
40 * retired some older targets of DYNAMIC_ARCH builds to a new option DYNAMIC_OLDER,
41 this affects PENRYN,DUNNINGTON,OPTERON,OPTERON_SSE3,BOBCAT,ATOM and NANO
42 (which will still be supported via the slower PRESCOTT kernels when this option is not set)
43 * added an option DYNAMIC_LIST that (used in conjunction with DYNAMIC_ARCH) allows to
44 specify the list of x86_64 targets to include. Any target not on the list will be supported
45 by the Sandybridge or Nehalem kernels if available, or by Prescott.
46 * improved SWITCH_RATIO on Haswell for increased GEMM throughput
47 * added initial support for Intel Skylake X, including an AVX512 SGEMM kernel
48 * added autodetection of Intel Cannon Lake series as Skylake X
49 * added a default L2 cache size for hypervisors that return zero here (Chromebook)
50 * fixed a name clash with recent Windows10 headers that broke the build with (at least)
51 recent mingw from MSYS2
52 * fixed a link error in mixed clang/gfortran builds with OpenMP
53 * updated the OSX deployment target to 10.8
54 * switched on parallel make for builds on MS Windows by default
57 * fixed SSWAP and DSWAP behaviour with zero INC_X and INC_Y
59 ====================================================================
64 * fixed some more thread race and locking bugs
65 * added preliminary support for calling an OpenMP build of the library from multiple threads
66 * removed performance impact of thread locks added in 0.2.20 on OpenMP code
67 * general code cleanup
68 * optimized DSDOT implementation
69 * improved thread distribution for GEMM
70 * corrected IMATCOPY/OMATCOPY implementation
71 * fixed out-of-bounds accesses in the multithreaded xBMV/xPMV and SYMV implementations
72 * cmake build improvements
73 * pkgconfig file now contains build options
74 * openblas_get_config() now reports USE_OPENMP and NUM_THREADS settings used for the build
75 * corrections and improvements for systems with more than 64 cpus
76 * LAPACK code updated to 3.8.0 including later fixes
77 * added ReLAPACK, a recursive implementation of several LAPACK functions
78 * Rewrote ROTMG to handle cases that the netlib code failed to address
79 * Disabled (broken) multithreading code for xTRMV
80 * corrected prototypes of complex CBLAS functions to make our cblas.h match the generally accepted standard
81 * shared memory access failures on startup are now handled more gracefully
82 * restored utests from earlier releases (and made them pass on all affected systems)
85 * several fixes for cpu autodetection
88 * corrected vector register overwriting in several Power8 kernels
89 * optimized additional BLAS functions
92 * added support for CortexA53 and A72
93 * added autodetection for ThunderX2T99
94 * made most optimized kernels the default for generic ARMv8 targets
97 * parallelized DDOT kernel for Haswell
98 * changed alignment directives in assembly kernels to boost performance on OSX
99 * fixed register handling in the GEMV microkernels (bug exposed by gcc7)
100 * added support for building on OpenBSD and Dragonfly
101 * updated compiler options to work with Intel release 2018
102 * support fully optimized build with clang/flang on Microsoft Windows
103 * fixed building on AIX
106 * added optimized BLAS 1/2 functions
109 * fixed cpu autodetection helper code
110 * added mips32 1004K cpu (Mediatek MT7621 and similar SoC)
111 * added mips64 I6500 cpu
113 ====================================================================
118 * Improved CMake support
119 * Fixed several thread race and locking bugs
120 * Fixed default LAPACK optimization level
121 * Updated LAPACK to 3.7.0
122 * Added ReLAPACK (https://github.com/HPAC/ReLAPACK, make BUILD_RELAPACK=1)
125 * Optimizations for Power9
126 * Fixed several Power8 assembly bugs
129 * New optimized Vulcan and ThunderX2T99 targets
130 * Support for ARMV7 SOFT_FP ABI (make ARM_SOFTFP_ABI=1)
131 * Detect all cpu cores including offline ones
132 * Fix compilation with CLANG
133 * Support building a shared library for Android
136 * Fixed several threading issues
137 * Fix compilation with CLANG
140 * Detect Intel Bay Trail and Apollo Lake
141 * Detect Intel Sky Lake and Kaby Lake
142 * Detect Intel Knights Landing
143 * Detect AMD A8, A10, A12 and Ryzen
144 * Support 64bit builds with Visual Studio
145 * Fix building with Intel and PGI compilers
146 * Fix building with MINGW and TDM-GCC
147 * Fix cmake builds for Haswell and related cpus
148 * Fix building for Sandybridge with CLANG 3.9
149 * Add support for the FLANG compiler
152 * New target z13 with BLAS3 optimizations
154 ====================================================================
158 * Improved cross compiling.
159 * Fix the bug on musl libc.
162 * Optimize BLAS on Power8
163 * Fixed Julia+OpenBLAS bugs on Power8
166 * Optimize BLAS on MIPS P5600 and I6400 (Thanks, Shivraj Patil, Kaustubh Raste)
169 * Improved on ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
172 ====================================================================
176 * If you set MAKE_NB_JOBS flag less or equal than zero,
177 make will be without -j.
180 * Support building Visual Studio static library. (#813, Thanks, theoractice)
181 * Fix bugs to pass buidbot CI tests (http://build.openblas.net)
184 * Provide DGEMM 8x4 kernel for Cortex-A57 (Thanks, Ashwin Sekhar T K)
187 * Optimize S and C BLAS3 on Power8
188 * Optimize BLAS2/1 on Power8
190 ====================================================================
194 * Enable BUILD_LAPACK_DEPRECATED=1 by default.
196 ====================================================================
200 * Avoid potential getenv segfault. (#716)
201 * Import LAPACK svn bugfix #142-#147,#150-#155
204 * Optimize c/zgemv for AMD Bulldozer, Piledriver, Steamroller
205 * Fix bug with scipy linalg test.
208 * Improve DGEMM for ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
211 * Optimize D and Z BLAS3 functions for Power8.
213 ====================================================================
217 * Upgrade LAPACK to 3.6.0 version.
218 Add BUILD_LAPACK_DEPRECATED option in Makefile.rule to build
219 LAPACK deprecated functions.
220 * Add MAKE_NB_JOBS option in Makefile.
221 Force number of make jobs.This is particularly
222 useful when using distcc. (#735. Thanks, Jerome Robert.)
223 * Redesign unit test. Run unit/regression test at every build (Travis-CI and Appveyor).
224 * Disable multi-threading for small size swap and ger. (#744. Thanks, Jerome Robert)
225 * Improve small zger, zgemv, ztrmv using stack alloction (#727. Thanks, Jerome Robert)
226 * Let openblas_get_num_threads return the number of active threads.
227 (#760. Thanks, Jerome Robert)
228 * Support illumos(OmniOS). (#749. Thanks, Lauri Tirkkonen)
229 * Fix LAPACK Dormbr, Dormlq bug. (#711, #713. Thanks, Brendan Tracey)
230 * Update scipy benchmark script. (#745. Thanks, John Kirkham)
233 * Optimize trsm kernels for AMD Bulldozer, Piledriver, Steamroller.
234 * Detect Intel Avoton.
235 * Detect AMD Trinity, Richland, E2-3200.
236 * Fix gemv performance bug on Mac OSX Intel Haswell.
237 * Fix some bugs with CMake and Visual Studio
240 * Support and optimize Cortex-A57 AArch64.
241 (#686. Thanks, Ashwin Sekhar TK)
242 * Fix Android build on ARMV7 (#778. Thanks, Paul Mustiere)
243 * Update ARMV6 kernels.
246 * Fix detection of POWER architecture
247 (#684. Thanks, Sebastien Villemot)
249 ====================================================================
253 * Support cmake on x86/x86-64. Natively compiling on MS Visual Studio.
254 (experimental. Thank Hank Anderson for the initial cmake porting work.)
256 On Linux and Mac OSX, OpenBLAS cmake supports assembly kernels.
261 On Windows MS Visual Studio, OpenBLAS cmake only support C kernels.
262 (OpenBLAS uses AT&T style assembly, which is not supported by MSVC.)
263 e.g. cmake -G "Visual Studio 12 Win64" .
264 Open OpenBLAS.sln and build.
266 * Enable MAX_STACK_ALLOC flags by default.
267 Improve ger and gemv for small matrices.
268 * Improve gemv parallel with small m and large n case.
269 * Improve ?imatcopy when lda==ldb (#633. Thanks, Martin Koehler)
270 * Add vecLib benchmarks (#565. Thanks, Andreas Noack.)
271 * Fix LAPACK lantr for row major matrices (#634. Thanks, Dan Kortschak)
272 * Fix LAPACKE lansy (#640. Thanks, Dan Kortschak)
273 * Import bug fixes for LAPACKE s/dormlq, c/zunmlq
274 * Raise the signal when pthread_create fails (#668. Thanks, James K. Lowden)
275 * Remove g77 from compiler list.
276 * Enable AppVeyor Windows CI.
279 * Support pure C generic kernels for x86/x86-64.
280 * Support Intel Boardwell and Skylake by Haswell kernels.
281 * Support AMD Excavator by Steamroller kernels.
282 * Optimize s/d/c/zdot for Intel SandyBridge and Haswell.
283 * Optimize s/d/c/zdot for AMD Piledriver and Steamroller.
284 * Optimize s/d/c/zapxy for Intel SandyBridge and Haswell.
285 * Optimize s/d/c/zapxy for AMD Piledriver and Steamroller.
286 * Optimize d/c/zscal for Intel Haswell, dscal for Intel SandyBridge.
287 * Optimize d/c/zscal for AMD Bulldozer, Piledriver and Steamroller.
288 * Optimize s/dger for Intel SandyBridge.
289 * Optimize s/dsymv for Intel SandyBridge.
290 * Optimize ssymv for Intel Haswell.
291 * Optimize dgemv for Intel Nehalem and Haswell.
292 * Optimize dtrmm for Intel Haswell.
295 * Support Android NDK armeabi-v7a-hard ABI (-mfloat-abi=hard)
296 e.g. make HOSTCC=gcc CC=arm-linux-androideabi-gcc NO_LAPACK=1 TARGET=ARMV7
297 * Fix lock, rpcc bugs (#616, #617. Thanks, Grazvydas Ignotas)
299 * Support ppc64le platform (ELF ABI v2. #612. Thanks, Matthew Brandyberry.)
300 * Support POWER7/8 by POWER6 kernels. (#612. Thanks, Fábio Perez.)
302 ====================================================================
306 * Improve OpenBLASConfig.cmake. (#474, #475. Thanks, xantares.)
307 * Improve ger and gemv for small matrices by stack allocation.
308 e.g. make -DMAX_STACK_ALLOC=2048 (#482. Thanks, Jerome Robert.)
309 * Introduce openblas_get_num_threads and openblas_get_num_procs.
310 (#497. Thanks, Erik Schnetter.)
311 * Add ATLAS-style ?geadd function. (#509. Thanks, Martin Köhler.)
312 * Fix c/zsyr bug with negative incx. (#492.)
313 * Fix race condition during shutdown causing a crash in
314 gotoblas_set_affinity(). (#508. Thanks, Ton van den Heuvel.)
317 * Support AMD Streamroller.
320 * Add Cortex-A9 and Cortex-A15 targets.
322 ====================================================================
326 * Add SYMBOLPREFIX and SYMBOLSUFFIX makefile options
327 for adding a prefix or suffix to all exported symbol names
328 in the shared library.(#459, Thanks Tony Kelman)
329 * Provide OpenBLASConfig.cmake at installation.
330 * Fix Fortran compiler detection on FreeBSD.
331 (#470, Thanks Mike Nolta)
335 * Add generic kernel files for x86-64. make TARGET=GENERIC
336 * Fix a bug of sgemm kernel on Intel Sandy Bridge.
337 * Fix c_check bug on some amd64 systems. (#471, Thanks Mike Nolta)
340 * Support APM's X-Gene 1 AArch64 processors.
341 Optimize trmm and sgemm. (#465, Thanks Dave Nuechterlein)
343 ====================================================================
347 * Added CBLAS interface for ?omatcopy and ?imatcopy.
348 * Enable ?gemm3m functions.
349 * Added benchmark for ?gemm3m.
350 * Optimized multithreading lower limits.
351 * Disabled SYMM3M and HEMM3M functions
352 because of segment violations.
355 * Improved axpy and symv performance on AMD Bulldozer.
356 * Improved gemv performance on modern Intel and AMD CPUs.
358 ====================================================================
362 * Added some benchmark codes.
363 * Fix link error on Linux/musl.(Thanks Isaac Dunham)
366 * Improved s/c/zgemm performance for Intel Haswell.
367 * Improved s/d/c/zgemv performance.
368 * Support the big numa machine.(EXPERIMENT)
371 * Fix detection when cpuinfo uses "Processor". (Thanks Isaiah)
373 ====================================================================
377 * Added BLAS extensions as following.
378 s/d/c/zaxpby, s/d/c/zimatcopy, s/d/c/zomatcopy.
379 * Added OPENBLAS_CORETYPE environment for dynamic_arch. (a86d34)
380 * Added NO_AVX2 flag for old binutils. (#401)
381 * Support outputing the CPU corename on runtime.(#407)
382 * Patched LAPACK to fix bug 114, 117, 118.
383 (http://www.netlib.org/lapack/bug_list.html)
384 * Disabled ?gemm3m for a work-around fix. (#400)
386 * Fixed lots of bugs for optimized kernels on sandybridge,Haswell,
387 bulldozer, and piledriver.
388 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
391 * Improved LAPACK testing.
393 ====================================================================
397 * Improved the result for LAPACK testing. (#372)
398 * Installed DLL to prefix/bin instead of prefix/lib. (#366)
399 * Build import library on Windows.(#374)
401 * To improve LAPACK testing, we fallback some kernels. (#372)
402 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
404 ====================================================================
408 * Added OPENBLAS_VERBOSE environment variable.(#338)
409 * Make OpenBLAS thread-pool resilient to fork via pthread_atfork.
410 (#294, Thank Olivier Grisel)
414 * Detect Intel Haswell for new Macbook.
416 ====================================================================
420 * Update LAPACK to 3.5.0 version
421 * Fixed compatiable issues with Clang and Pathscale compilers.
424 * Optimization on Intel Haswell.
425 * Enable optimization kernels on AMD Bulldozer and Piledriver.
428 * Support ARMv6 and ARMv7 ISA.
429 * Optimization on ARM Cortex-A9.
431 ====================================================================
435 * Support Open64 5.0. (#266)
436 * Add executable stack markings. (#262, Thank Sébastien Fabbro)
437 * Respect user's LDFLAGS (Thank Sébastien Fabbro)
440 * Rollback bulldozer and piledriver kernels to barcelona kernels (#263)
441 We will fix the compuational error bug in bulldozer and piledriver kernels.
443 ====================================================================
447 * Support LSB (Linux Standard Base) 4.1.
449 * Include LAPACK 3.4.2 source codes to the repo.
450 Avoid downloading at compile time.
451 * Add NO_PARALLEL_MAKE flag to disable parallel make.
452 * Create openblas_get_parallel to retrieve information which
453 parallelization model is used by OpenBLAS. (Thank grisuthedragon)
454 * Detect LLVM/Clang compiler. The default compiler is Clang on Mac OS X.
455 * Change LIBSUFFIX from .lib to .a on windows.
456 * A work-around for dtrti_U single thread bug. Replace it with LAPACK codes. (#191)
459 * Optimize c/zgemm, trsm, dgemv_n, ddot, daxpy, dcopy on
460 AMD Bulldozer. (Thank Werner Saar)
461 * Add Intel Haswell support (using Sandybridge optimizations).
463 * Add AMD Piledriver support (using Bulldozer optimizations).
464 * Fix the computational error in zgemm avx kernel on
466 * Fix the overflow bug in gemv.
467 * Fix the overflow bug in multi-threaded BLAS3, getrf when NUM_THREADS
468 is very large.(#214, #221, #246).
470 * Support loongcc (Open64 based) compiler for ICT Loongson 3A/B.
473 * Support Power7 by old Power6 kernels. (#220)
475 ====================================================================
479 * Improved OpenMP performance slightly. (d744c9)
480 * Improved cblas.h compatibility with Intel MKL.(#185)
481 * Fixed the overflowing bug in single thread cholesky factorization.
482 * Fixed the overflowing buffer bug of multithreading hbmv and sbmv.(#174)
485 * Added AMD Bulldozer x86-64 S/DGEMM AVX kernels. (Thank Werner Saar)
486 We will tune the performance in future.
487 * Auto-detect Intel Xeon E7540.
488 * Fixed the overflowing buffer bug of gemv. (#173)
489 * Fixed the bug of s/cdot about invalid reading NAN on x86_64. (#189)
493 ====================================================================
497 * Added NO_SHARED flag to disable generating the shared library.
498 * Compile LAPACKE with ILP64 modle when INTERFACE64=1 (#158)
499 * Export LAPACK 3.4.2 symbols in shared library. (#147)
500 * Only detect the number of physical CPU cores on Mac OSX. (#157)
501 * Fixed NetBSD build. (#155)
502 * Fixed compilation with TARGET=GENERIC. (#160)
504 * Restore the original CPU affinity when calling
505 openblas_set_num_threads(1) (#153)
506 * Fixed a SEGFAULT bug in dgemv_t when m is very large.(#154)
509 ====================================================================
513 * Upgraded LAPACK to 3.4.2 version. (#145)
514 * Provided support for passing CFLAGS, FFLAGS, PFLAGS,
515 FPFLAGS to make. (#137)
516 * f77blas.h:compatibility for compilers without C99 complex
517 number support. (#141)
519 * Added NO_AVX flag. Check OS supporting AVX on runtime. (#139)
520 * Fixed zdot incompatibility ABI issue with GCC 4.7 on
521 Windows 32-bit. (#140)
523 * Fixed the generation of shared library bug.
524 * Fixed the detection bug on the Loongson 3A server.
525 ====================================================================
529 * Fixed LAPACK unstable bug about ?laswp. (#130)
530 * Fixed the shared library bug about unloading the library on
532 * Fixed the compilation failure on BlueGene/P (TARGET=PPC440FP2)
533 Please use gcc and IBM xlf. (#134)
535 * Supported goto_set_num_threads and openblas_set_num_threads
536 APIs in Windows. They can set the number of threads on runtime.
538 ====================================================================
542 * Fixed exporting DLL functions bug on Windows/MingW
543 * Support GNU Hurd (Thank Sylvestre Ledru)
544 * Support kfreebsd kernel (Thank Sylvestre Ledru)
546 * Support Intel Sandy Bridge 22nm desktop/mobile CPU
548 * Improve the detection of SPARC (Thank Sylvestre Ledru)
550 ====================================================================
555 * Fixed the SEGFAULT bug about hyper-theading
556 * Support AMD Bulldozer by using GotoBLAS2 AMD Barcelona codes
558 ====================================================================
562 * Removed the limitation (64) of numbers of CPU cores.
563 Now, it supports 256 cores at max.
564 * Supported clang compiler.
565 * Fixed some build bugs on FreeBSD
567 * Optimized Level-3 BLAS on Intel Sandy Bridge x86-64 by AVX instructions.
568 Please use gcc >= 4.6 or clang >=3.1.
569 * Support AMD Bobcat by using GotoBLAS2 AMD Barcelona codes.
571 ====================================================================
575 * Upgraded LAPACK to 3.4.1 version. (Thank Zaheer Chothia)
576 * Supported LAPACKE, a C interface to LAPACKE. (Thank Zaheer Chothia)
577 * Fixed the build bug (MD5 and download) on Mac OSX.
578 * Auto download CUnit 2.1.2-2 from SF.net with UTEST_CHECK=1.
579 * Fxied the compatibility issue for compilers without C99 complex number
582 * Auto-detect Intel Sandy Bridge Core i7-3xxx & Xeon E7 Westmere-EX.
583 * Test alpha=Nan in dscale.
584 * Fixed a SEGFAULT bug in samax on x86 windows.
586 ====================================================================
590 * Set soname of shared library on Linux.
591 * Added LIBNAMESUFFIX flag in Makefile.rule. The user can use
592 this flag to control the library name, e.g. libopenblas.a,
593 libopenblas_ifort.a or libopenblas_omp.a.
594 * Added GEMM_MULTITHREAD_THRESHOLD flag in Makefile.rule.
595 The lib use single thread in GEMM function with small matrices.
597 * Used GEMV SSE/SSE2 kernels on x86 32-bit.
598 * Exported CBLAS functions in Windows DLL.
600 * Completed Level-3 BLAS optimization on Loongson 3A CPU.
601 * Improved GEMV performance on Loongson 3A CPU.
602 * Improved Level-3 BLAS performance on Loongson 3B CPU. (EXPERIMENT)
604 ====================================================================
608 * Fixed missing "#include <sched.h>" bug on Mac OS X.
609 Thank Mike Nolta for the patch.
610 * Upgraded LAPACK to 3.4.0 version
611 * Fixed a bug on Mac OS X. Don't require SystemStubs on OS X.
612 SystemStubs does not exist on Lion. Thank Stefan Karpinski.
613 * Improved README with using OpenMP. Check the internal threads
614 count less than or equal to omp_get_max_threads()
616 * Auto-detect Intel Core i6/i7 (Sandy Bridge) CPU with Nehalem assembly kernels
617 * Fixed some bugs on MingW 64-bit including zgemv, cdot, zdot.
619 ====================================================================
623 * Fixed a bug about installation. The header file "fblas77.h"
625 * Fixed #61 a building bug about setting TARGET and DYNAMIC_ARCH.
626 * Try to handle absolute path of shared library in OSX. (#57)
627 Thank Dr Kane O'Donnell.
628 * Changed the installation folder layout to $(PREFIX)/include and
632 * Fixed #58 zdot/xdot SEGFAULT bug with GCC-4.6 on x86. According
633 to i386 calling convention, The callee should remove the first
634 hidden parameter.Thank Mr. John for this patch.
636 ====================================================================
641 * Added DTB_ENTRIES into dynamic arch setting parameters. Now,
642 it can read DTB_ENTRIES on runtime. (Refs issue #55 on github)
644 ====================================================================
649 * Fixed a building bug when DYNAMIC_ARCH=1 & INTERFACE64=1.
650 (Refs issue #44 on github)
652 ====================================================================
657 * Stop the build and output the error message when detecting
658 fortran compiler failed. (Refs issue #42 on github)
660 ====================================================================
665 * Fixed blasint undefined bug in <cblas.h> file. Other software
666 could include this header successfully(Refs issue #13 on github)
667 * Fixed the SEGFAULT bug on 64 cores. On SMP server, the number
668 of CPUs or cores should be less than or equal to 64.(Refs issue #14
670 * Support "void goto_set_num_threads(int num_threads)" and "void
671 openblas_set_num_threads(int num_threads)" when USE_OPENMP=1
672 * Added extern "C" to support C++. Thank Tasio for the patch(Refs
674 * Provided an error message when the arch is not supported.(Refs
676 * Fixed issue #23. Fixed a bug of f_check script about generating link flags.
677 * Added openblas_set_num_threads for Fortran.
678 * Fixed #25 a wrong result of rotmg.
679 * Fixed a bug about detecting underscore prefix in c_check.
680 * Print the wall time (cycles) with enabling FUNCTION_PROFILE
681 * Fixed #35 a build bug with NO_LAPACK=1 & DYNAMIC_ARCH=1
682 * Added install target. You can use "make install". (Refs #20)
686 * Fixed #28 a wrong result of dsdot on x86_64.
687 * Fixed #32 a SEGFAULT bug of zdotc with gcc-4.6.
688 * Fixed #33 ztrmm bug on Nehalem.
689 * Work-around #27 the low performance axpy issue with small imput size & multithreads.
692 * Fixed #28 a wrong result of dsdot on Loongson3A/MIPS64.
693 * Optimized single/double precision BLAS Level3 on Loongson3A/MIPS64. (Refs #2)
694 * Optimized single/double precision axpy function on Loongson3A/MIPS64. (Refs #3)
696 ====================================================================
701 * Support "make NO_LAPACK=1" to build the library without
703 * Fixed randomly SEGFAULT when nodemask==NULL with above Linux 2.6.34.
704 Thank Mr.Ei-ji Nakama providing this patch. (Refs issue #12 on github)
705 * Added DEBUG=1 rule in Makefile.rule to build debug version.
706 * Disable compiling quad precision in reference BLAS library(netlib BLAS).
707 * Added unit testcases in utest/ subdir. Used CUnit framework.
708 * Supported OPENBLAS_* & GOTO_* environment variables (Pleas see README)
709 * Imported GotoBLAS2 1.13 BSD version
712 * On x86 32bits, fixed a bug in zdot_sse2.S line 191. This would casue
713 zdotu & zdotc failures. Instead, work-around it. (Refs issue #8 #9 on github)
714 * Modified ?axpy functions to return same netlib BLAS results
715 when incx==0 or incy==0 (Refs issue #7 on github)
716 * Modified ?swap functions to return same netlib BLAS results
717 when incx==0 or incy==0 (Refs issue #6 on github)
718 * Modified ?rot functions to return same netlib BLAS results
719 when incx==0 or incy==0 (Refs issue #4 on github)
720 * Detect Intel Westmere,Intel Clarkdale and Intel Arrandale
721 to use Nehalem codes.
722 * Fixed a typo bug about compiling dynamic ARCH library.
724 * Improve daxpy performance on ICT Loongson 3A.
725 * Supported ICT Loongson 3A CPU (Refs issue #1 on github)
726 ====================================================================