2 ====================================================================
7 * Fixed a race condition on thread shutdown in non-OpenMP builds
8 * Fixed custom BUFFERSIZE option getting ignored in gmake builds
9 * Fixed CMAKE compilation of the TRMM kernels for GENERIC platforms
10 * Added CBLAS interfaces for CROTG, ZROTG, CSROT and ZDROT
11 * Improved performance of OMATCOPY_RT across all platforms
12 * Changed perl scripts to use env instead of a hardcoded /usr/bin/perl
13 * Fixed potential misreading of the GCC compiler version in the build scripts
14 * Fixed convergence problems in LAPACK complex GGEV/GGES (Reference-LAPACK #477)
15 * Reduced the stacksize requirements for running the LAPACK testsuite (Reference-LAPACK #335)
18 * Fixed compilation on RISCV (missing entry in getarch)
21 * Fixed compilation for DYNAMIC_ARCH with clang and with old gcc versions
22 * Added support for compilation on FreeBSD/ppc64le
23 * Added optimized POWER10 kernels for SSCAL, DSCAL, CSCAL, ZSCAL
24 * Added optimized POWER10 kernels for SROT, DROT, CDOT, SASUM, DASUM
25 * Improved SSWAP, DSWAP, CSWAP, ZSWAP performance on POWER10
26 * Improved SCOPY and CCOPY performance on POWER10
27 * Improved SGEMM and DGEMM performance on POWER10
28 * Added support for compilation with the NVIDIA HPC compiler
31 * Added an optimized bfloat16 GEMM kernel for Cooperlake
32 * Added CPUID autodetection for Intel Rocket Lake and Tiger Lake cpus
33 * Improved the performance of SASUM,DASUM,SROT,DROT on AMD Ryzen cpus
34 * Added support for compilation with the NAG Fortran compiler
35 * Fixed recognition of the AMD AOCC compiler
36 * Fixed compilation for DYNAMIC_ARCH with clang on Windows
37 * Added support for running the BLAS/CBLAS tests on Windows
38 * Fixed signatures of the tls callback functions for Windows x64
39 * Fixed various issues with fma intrinsics support handling
42 * Added support for embedded Cortex M targets via a new option EMBEDDED
45 * Fixed the THUNDERX2T99 and NEOVERSEN1 DNRM2/ZNRM2 kernels for inputs with Inf
46 * Added support for the DYNAMIC_LIST option
47 * Added support for compilation with the NVIDIA HPC compiler
48 * Added support for compiling with the NAG Fortran compiler
50 ====================================================================
55 * Added a generic bfloat16 SBGEMV kernel
56 * Fixed a potentially severe memory leak after fork in OpenMP builds
57 that was introduced in 0.3.12
58 * Added detection of the Fujitsu Fortran compiler
59 * Added detection of the (e)gfortran compiler on OpenBSD
60 * Added support for overriding the default name of the library independently
61 from symbol suffixing in the gmake builds (already supported in cmake)
64 * Added a RISC V port optimized for C910V
67 * Added optimized POWER10 kernels for SAXPY, CAXPY, SDOT, DDOT and DGEMV_N
68 * Improved DGEMM performance on POWER10
69 * Improved STRSM and DTRSM performance on POWER9 and POWER10
70 * Fixed segmemtation faults in DYNAMIC_ARCH builds
71 * Fixed compilation with the PGI compiler
74 * Fixed compilation of kernels that require SSE2 intrinsics since 0.3.12
77 * Added an optimized bfloat16 SBGEMV kernel for SkylakeX and Cooperlake
78 * Improved the performance of SASUM and DASUM kernels through parallelization
79 * Improved the performance of SROT and DROT kernels
80 * Improved the performance of multithreaded xSYRK
81 * Fixed OpenMP builds that use the LLVM Clang compiler together with GNU gfortran
82 (where linking of both the LLVM libomp and GNU libgomp could lead to lockups or
84 * Fixed miscompilations by old gcc 4.6
85 * Fixed misdetection of AVX2 capability in some Sandybridge cpus
86 * Fixed lockups in builds combining DYNAMIC_ARCH with TARGET=GENERIC on OpenBSD
89 * Fixed segmemtation faults in DYNAMIC_ARCH builds
92 * Improved kernels for Loongson 3R3 ("3A") and 3R4 ("3B") models, including MSA
93 * Fixed bugs in the MSA kernels for CGEMM, CTRMM, CGEMV and ZGEMV
94 * Added handling of zero increments in the MSA kernels for SSWAP and DSWAP
95 * Added DYNAMIC_ARCH support for MIPS64 (currently Loongson3R3/3R4 only)
98 * Fixed building 32 and 64 bit SPARC kernels with the SolarisStudio compilers
100 ====================================================================
105 * Fixed missing BLAS/LAPACK functions (inadvertently dropped during
106 the build system restructuring)
107 * Fixed argument conversion macro in LAPACKE_zgesvdq (LAPACK #458)
110 * Added optimized SCOPY/CCOPY kernels for POWER10
111 * Increased and unified the default size of the GEMM BUFFER
112 * Fixed building for POWER10 in DYNAMIC_ARCH mode
113 * POWER10 compatibility test now checks binutils version as well
114 * Cleaned up compiler warnings
117 * corrected compiler version checks for AVX2 compatibility
118 * added compiler option -mavx2 for building with flang
119 * fixed direct SGEMM pathway for small matrix sizes (broken by
120 the code refactoring in 0.3.11)
121 * fixed unhandled partial register clobbers in several kernels
122 for AXPY,DOT,GEMV_N and GEMV_T flagged by gcc10 tree-vectorizer
125 * improved Apple Vortex support to include cross-compiling
127 ====================================================================
133 the newly added BFLOAT16 functions were renamed to use the
134 letter "B" instead of "H" to avoid potential confusion with
135 the IEEE "half precision float" type, i.e. the 0.3.10
136 SHGEMM is now SBGEMM and the corresponding build option
137 was changed from "BUILD_HALF" to "BUILD_BFLOAT16".
138 * Reduced the default BLAS3_MEM_ALLOC_THRESHOLD (used as an upper
139 limit for placing temporary arrays on the stack) to be compatible
140 with a stack size of 1mb (as imposed by the JAVA runtime library)
141 * Added mixed-precision dot function SBDOT and utility functions
142 shstobf16, shdtobf16, sbf16tos and dbf16tod to convert between
143 single or double precision float arrays and bfloat16 arrays
144 * Fixed prototypes of LAPACK_?ggsvp and LAPACK_?ggsvd functions
146 * Fixed underflow and rounding errors in LAPACK SLANV2 and DLANV2
147 (causing miscalculations in e.g. SHSEQR/DHSEQR, LAPACK issue #263)
148 * Fixed workspace calculation in LAPACK ?GELQ (LAPACK issue #415)
149 * Fixed several bugs in the LAPACK testsuite
150 * Improved performance of TRMM and TRSM for certain problem sizes
151 * Fixed infinite recursions and workspace miscalculations in ReLAPACK
152 * CMAKE builds no longer require pkg-config for creating the .pc file
153 * Makefile builds no longer misread NO_CBLAS=0 or NO_LAPACK=0 as
154 enabling these options
155 * Fixed detection of gfortran when invoked through an mpi wrapper
156 * Improve thread reinitialization performance with OpenMP after a fork
157 * Added support for building only the subset of the library required
158 for a particular precision by specifying BUILD_SINGLE, BUILD_DOUBLE
159 * Optional function name prefixes and suffixes are now correctly
160 reflected in the generated cblas.h
161 * Added CMAKE build support for the LAPACK and multithreading tests
164 * Added optimized support for POWER10
165 * Added support for compiling for POWER8 in 32bit mode
166 * Added support for compilation with LLVM/clang
167 * Added support for compilation with NVIDIA/PGI compilers
168 * Fixed building on big-endian POWER8
169 * Fixed miscompilation of ZDOTC by gcc10
170 * Fixed alignment errors in the POWER8 SAXPY kernel
171 * Improved CPU detection on AIX
172 * Supported building with older compilers on POWER9
175 * Added support for Intel Cooperlake
176 * Added autodetection of AMD Renoir/Matisse/Zen3 cpus
177 * Added autodetection of Intel Comet Lake cpus
178 * Reimplemented ?sum, ?dot and daxpy using universal intrinsics
179 * Reset the fpu state before using the fpu on Windows as a workaround
180 for a problem introduced in Windows 10 build 19041 (a.k.a. SDK 2004)
181 * Fixed potentially undefined behaviour in the dot and gemv_t kernels
182 * Fixed a potential segmentation fault in DYNAMIC_ARCH builds
183 * Fixed building for ZEN with PGI/NVIDIA and AMD AOCC compilers
186 * Fixed cpu detection on BSD-like systems
189 * Added preliminary support for Apple Vortex cpus
190 * Added support for the Cavium ThunderX3T110 cpu
191 * Fixed cpu detection on BSD-like systems
192 * Fixed compilation in -std=C18 mode
195 * Added support for compiling with the clang compiler
196 * Improved GEMM performance on Z14
198 ====================================================================
203 * Improved thread locking behaviour in blas_server and parallel getrf
204 * Imported bugfix 394 from LAPACK (spurious reference to "XERBL"
205 due to overlong lines)
206 * Imported bugfix 403 from LAPACK (compile option "recursive" required
207 for correctness with Intel and PGI)
208 * Imported bugfix 408 from LAPACK (wrong scaling in ZHEEQUB)
209 * Imported bugfix 411 from LAPACK (infinite loop in LARGV/LARTG/LARTGP)
210 * Fixed mismatches between BUFFERSIZE and GEMM_UNROLL parameters that
211 could lead to crashes at large matrix sizes
212 * Restored internal soname in dynamic libraries on FreeBSD and Dragonfly
213 * Added API (openblas_setaffinity) to set the thread affinity on Linux
214 * Added initial infrastructure for half-precision floating point
215 (bfloat16) support with a generic implementation of SHGEMM
216 * Added CMAKE build system support for building the cblas_Xgemm3m
218 * Fixed CMAKE support for building in a path with embedded spaces
219 * Fixed CMAKE (non)handling of NO_EXPRECISION and MAX_STACK_ALLOC
220 * Fixed GCC version detection in the Makefiles
221 * Allowed overriding the names of AR, AS and LD in Makefile builds
224 * Fixed big-endian POWER8 ELFv2 builds on FreeBSD
225 * Fixed GCC version checks and DYNAMIC_ARCH builds on POWER9
226 * Fixed CMAKE build support for POWER9
227 * fixed a potential race condition in the thread buffer allocation
228 * Worked around LAPACK test failures on PPC G4
231 * Fixed a potential race condition in the thread buffer allocation
232 * Added support for MIPS 24K/24KE family based on P5600 kernels
235 * fixed a potential race condition in the thread buffer allocation
236 * Added TARGET=GENERIC
239 * Fixed a race condition in the thread buffer allocation
242 * Fixed a race condition in the thread buffer allocation
243 * Fixed zero initialisation in the assembly for SGEMM and DGEMM BETA
244 * Improved performance of the ThunderX2 DAXPY kernel
245 * Added an optimized SGEMM kernel for Cortex A53
246 * Fixed Makefile support for INTERFACE64 (8-byte integer)
249 * Fixed a syntax error in the CMAKE setup for SkylakeX
250 * Improved performance of STRSM on Haswell, SkylakeX and Ryzen
251 * Improved SGEMM performance on SGEMM for workloads with ldc a
253 * Improved DGEMM performance on Skylake X
254 * Fixed unwanted AVX512-dependency of SGEMM in DYNAMIC_ARCH
255 builds created on SkylakeX
256 * Removed data alignment requirement in the SSE2 copy kernels
257 that could cause spurious crashes
258 * Added a workaround for an optimizer bug in AppleClang 11.0.3
259 * Fixed LAPACK test failures due to wrong options for Intel Fortran
260 * Fixed compilation and LAPACK test results with recent Flang
262 * Fixed DYNAMIC_ARCH builds with CMAKE on OS X
263 * Fixed missing exports of cblas_i?amin, cblas_i?min, cblas_i?max,
264 cblas_?sum, cblas_?gemm3m in the shared library on OS
265 * Fixed reporting of cpu name in DYNAMIC_ARCH builds (would sometimes
266 show the name of an older generation chip supported by the same kernels)
269 * Improved performance of SGEMM/STRMM and DGEMM/DTRMM on Z14
271 ====================================================================
276 * Fixed a miscompilation of the GETRF functions with CMAKE
277 * Imported bugfix 390 from LAPACK (missing NaN propagation in xCOMBSSQ)
278 * The size of the memory buffer used for splitting GEMM tasks across
279 multiple threads can now be configured in the build system.
282 * Fixed several compilation problems related to endianness
283 and ELF version on POWER8 and POWER9
284 * Fixed use of the absolute value IAMIN/IAMAX instead of IMIN/IMAX
285 * Fixed a race condition in the level3 blas code
288 * Fixed use of the absoltute value IAMIN/IAMAX instead of IMIN/IMAX
291 * Fixed a race condition in the level3 blas code
292 * Fixed compilation on Android
294 * Added support for Ampere EMAG8180
295 * Added support for Neoverse N1
296 * Improved performance of the blas_lock function
297 * Fixed a race condition in the level3 blas code
298 * Fixed a performance regression on TSV110-based servers
301 * Fixed a long-standing error with undeclared register overwrites
302 in the DSCAL microkernel for HASWELL,SKYLAKEX and ZEN
303 * Fixed a long-standing bug in the SSE implementation of IAMAX
304 * Fixed a CMAKE build failure with DYNAMIC_ARCH
305 * Fixed cpu autodetection of Goldmont+, Cannon Lake and Ice Lake
306 * Fixed a compilation failure on OSX with compiler name containing dash
307 * Fixed compilation with MinGW on SkylakeX
308 * Improved speed of the AVX512 GEMM3M kernel on SkylakeX
309 * Added an AVX512 STRMM kernel for SkylakeX
310 * Improved GEMM performance on Haswell and Zen
313 * fixed compilation of the DYNAMIC_ARCH code
315 ====================================================================
320 ` * LAPACK has been updated to 3.9.0 (plus patches up to
322 * CMAKE support has been improved in several areas including
324 * a thread race condition in the GEMM3M kernels was resolved
325 * the "generic" (plain C) gemm beta kernel used by many targets
327 * an optimized version of the LAPACK trtrs functions has been added
328 * an incompatibilty between the LAPACK tests and the OpenBLAS
329 implementation of XERBLA was resolved, removing the numerous
330 warnings about wrong error exits in the former
331 * support for NetBSD has been added
332 * support for compilation with g95 and non-GNU versions of ld
334 * support for compilation with (upcoming) gcc 10 has been added
337 * worked around miscompilation of several POWER8 and POWER9
338 kernels by older versions of gcc
339 * added support for big-endian POWER8 and for compilation on AIX
340 * corrected bugs in the big-endian support for PPC440 and PPC970
341 * DYNAMIC_ARCH support is now available in CMAKE builds as well
344 * performance of DGEMM_BETA and SGEMM_NCOPY has been improved
345 * compilation for 32bit works again
346 * performance of the RPCC function has been improved
347 * improved performance on small systems
348 * DYNAMIC_ARCH support is now available in CMAKE builds as well
349 * cross-compilation from OSX to IOS was simplified
352 * a new AVX512 DGEMM kernel was added and the AVX512 SGEMM kernel
353 was significantly improved
354 * optimized AVX512 kernels for CGEMM and ZGEMM have been added
355 * AVX2 kernels for STRMM, SGEMM, and CGEMM have been significantly
356 sped up and optimized CGEMM3M and ZGEMM3M kernels have been added
357 * added support for QEMU virtual cpus
358 * a compilation problem with PGI and SUN compilers was fixed
359 * Intel "Goldmont plus" is now autodetected
360 * a potential crash on program exit on MS Windows has been fixed
363 * an unwanted case sensitivity in the implementation of LSAME
364 on older 32bit AMD cpus was fixed
367 * Z15 is now supported as Z14
368 * DYNAMIC_ARCH is now available on ZARCH as well
370 ====================================================================
375 * having the gmake special variables TARGET_ARCH or TARGET_MACH
376 defined no longer causes build failures in ctest or utest
377 * defining NO_AFFINITY or USE_TLS to 0 in gmake builds no longer
378 has the same effect as setting them to 1
379 * a new test program was added to allow checking the library for
381 * a new option USE_LOCKING was added to ensure thread safety when
382 OpenBLAS itself is built without multithreading but will be
383 called from multiple threads.
384 * a build failure on Linux with glibc versions earlier than 2.5
386 * a runtime error with CPU enumeration (and NO_AFFINITY not set)
387 on glibc 2.6 was fixed
388 * NO_AFFINITY was added to the CMAKE options (and defaults to being
389 active on Linux, as in the gmake builds)
392 * the build-time logic for detection of AVX512 availability in
393 the processor and compiler was fixed
394 * gmake builds on OSX now set the internal name of the library to
395 libopenblas.0.dylib (consistent with CMAKE)
396 * the Haswell DGEMM kernel received a significant speedup through
397 improved prefetch and load instructions
398 * performance of DGEMM, DTRMM, DTRSM and ZDOT on Zen/Zen2 was markedly
399 increased by avoiding vpermpd instructions
400 * the SKYLAKEX (AVX512) DGEMM helper functions have now been disabled
401 to fix remaining errors in DGEMM, DSYMM and DTRMM
404 * added support for building on FreeBSD/powerpc64 and FreeBSD/ppc970
405 * added optimized kernels for POWER9 SGEMM and STRMM
408 * fixed the softfp implementations of xAMAX and IxAMAX
409 * removed the predefined -march= flags on both ARMV5 and ARMV6 as
410 they were appropriate for only a subset of platforms
412 ====================================================================
417 * the build tools now check that a given cpu TARGET is actually valid
418 * the build-time check of system features (c_check) has been made
419 less dependent on particular perl features (this should mainly
420 benefit building on Windows)
421 * several problem with the ReLAPACK integration were fixed,
422 including INTERFACE64 support and building a shared library
423 * building with CMAKE on BSD systems was improved
424 * a non-absolute SUM function was added based on the
425 existing optimized code for ASUM
426 * CBLAS interfaces to the IxMIN and IxMAX functions were added
427 * a name clash between LAPACKE and BOOST headers was resolved
428 * CMAKE builds with OpenMP failed to include the appropriate getrf_parallel
430 * a crash on thread (key) deletion with the USE_TLS=1 memory management
432 * restored several earlier fixes, in particular for OpenMP performance,
433 building on BSD, and calling fork on CYGWIN, which had inadvertently
434 been dropped in the 0.3.3 rewrite of the memory management code.
437 * the AVX512 DGEMM kernel has been disabled again due to unsolved problems
438 * building with old versions of MSVC was fixed
439 * it is now possible to build a static library on Windows with CMAKE
440 * accessing environment variables on CYGWIN at run time was fixed
441 * the CMAKE build system now recognizes 32bit userspace on 64bit hardware
442 * Intel "Denverton" atom and Hygon "Dhyana" zen CPUs are now autodetected
443 * building for DYNAMIC_ARCH with a DYNAMIC_LIST of targets is now supported
445 * building for DYNAMIC_ARCH with GENERIC as the default target is now supported
446 * a buffer overflow in the SSE GEMM kernel for Intel Nano targets was fixed
447 * assembly bugs involving undeclared modification of input operands were fixed
448 in the AXPY, DOT, GEMV, GER, SCAL, SYMV and TRSM microkernels for Nehalem,
449 Sandybridge, Haswell, Bulldozer and Piledriver. These would typically cause
450 test failures or segfaults when compiled with recent versions of gcc from 8 onward.
451 * a similar bug was fixed in the blas_quickdivide code used to split workloads
453 * a bug in the IxMIN implementation for the GENERIC target made it return the result of IxMAX
454 * fixed building on SkylakeX systems when either the compiler or the (emulated) operating
455 environment does not support AVX512
456 * improved GEMM performance on ZEN targets
459 * build failures caused by the recently added checks for AVX512 were fixed
460 * an inline assembly bug involving undeclared modification of an input argument was
461 fixed in the blas_quickdivide code used to split workloads in most functions
462 * a bug in the IMIN implementation for the GENERIC target made it return the result of IMAX
465 * a bug in the IMIN implementation made it return the result of IMAX
468 * single precision BLAS1/2 functions have received optimized POWER8 kernels
469 * POWER9 is now a separate target, with an optimized DGEMM/DTRMM kernel
470 * building on PPC970 systems under OSX Leopard or Tiger is now supported
471 * out-of-bounds memory accesses in the gemm_beta microkernels were fixed
472 * building a shared library on AIX is now supported for POWER6
473 * DYNAMIC_ARCH support has been added for POWER6 and newer
476 * corrected xDOT behaviour with zero INC_X or INC_Y
477 * a bug in the IMIN implementation made it return the result of IMAX
480 * added support for HiSilicon TSV110 cpus
481 * the CMAKE build system now recognizes 32bit userspace on 64bit hardware
482 * cross-compilation with CMAKE now works again
483 * a bug in the IMIN implementation made it return the result of IMAX
484 * ARMV8 builds with the BINARY=32 option are now automatically handled as ARMV7
487 * optimized microkernels for single precicion BLAS1/2 functions have been added
490 ====================================================================
495 * loop unrolling in TRMV has been enabled again.
496 * A domain error in the thread workload distribution for SYRK
498 * gmake builds will now automatically add -fPIC to the build
499 options if the platform requires it.
500 * a pthreads key leakage (and associate crash on dlclose) in
501 the USE_TLS codepath was fixed.
502 * building of the utest cases on systems that do not provide
503 an implementation of complex.h was fixed.
506 * the SkylakeX code was changed to compile on OSX.
507 * unwanted application of the -march=skylake-avx512 option
508 to the common code parts of a DYNAMIC_ARCH build was fixed.
509 * improved performance of SGEMM for small workloads on Skylake X.
510 * performance of SGEMM and DGEMM was improved on Haswell.
513 * a configuration error that broke the CNRM2 kernel was corrected.
514 * compilation of the GEMM kernels with CMAKE was fixed.
515 * DYNAMIC_ARCH builds are now available with CMAKE as well.
516 * using CMAKE for cross-compilation to the new cpu TARGETs
517 introduced in 0.3.4 now works.
520 * a problem in cpu autodetection for AIX has been corrected.
522 ====================================================================
527 * the new, experimental thread-local memory allocation had
528 inadvertently been left enabled for gmake builds in 0.3.3
529 despite the announcement. It is now disabled by default, and
530 single-threaded builds will keep using the old allocator even
531 if the USE_TLS option is turned on.
532 * OpenBLAS will now provide enough buffer space for at least 50
534 * The output of openblas_get_config() now contains the version
536 * A serious thread safety bug in GEMV operation with small M and
537 large N size has been fixed.
538 * The code will now automatically call blas_thread_init after a
539 fork if needed before handling a call to openblas_set_num_threads
540 * Accesses to parallelized level3 functions from multiple callers
541 are now serialized to avoid thread races (unless using OpenMP).
542 This should provide better performance than the known-threadsafe
543 (but non-default) USE_SIMPLE_THREADED_LEVEL3 option.
544 * When building LAPACK with gfortran, -frecursive is now (again)
545 enabled by default to ensure correct behaviour.
546 * The OpenBLAS version cblas.h now supports both CBLAS_ORDER and
547 CBLAS_LAYOUT as the name of the matrix row/column order option.
548 * Externally set LDFLAGS are now passed through to the final compile/link
549 steps to facilitate setting platform-specific linker flags.
550 * A potential race condition during the build of LAPACK (that would
551 usually manifest itself as a failure to build TESTING/MATGEN) has been
553 * xHEMV has been changed to stay single-threaded for small input sizes
554 where the overhead of multithreading exceeds any possible gains
555 * CSWAP and ZSWAP have been limited to a single thread except on ARMV8 or
556 ThunderX hardware with sizable input.
557 * Linker flags for the PGI compiler have been updated
558 * Behaviour of AXPY with zero increments is now handled in the C interface,
559 correcting the result on at least Intel Atom.
560 * The result matrix from calling SGELSS with an all-zero input matrix is
561 now zeroed completely.
564 * Autodetection of AMD Ryzen2 has been fixed (again).
565 * CMAKE builds now support labeling of an INTERFACE64=1 build of
566 the library with the _64 suffix.
567 * AVX512 version of DGEMM has been added and the AVX512 SGEMM kernel
568 has been sped up by rewriting with C intrinsics
569 * Fixed compilation on RHEL5/CENTOS5 (issue with typename __WAIT_STATUS)
572 * added support for building on AIX (with gcc and GNU tools from AIX Toolbox).
573 * CPU type detection has been implemented for AIX.
574 * CPU type detection has been fixed for NETBSD.
577 * AXPY on LOONGSON3A has been corrected to pass "zero increment" utest.
578 * DSDOT on LOONGSON3A has been fixed.
579 * the SGEMM microkernel has been hardened against potential data loss.
582 * DYNAMic_ARCH support is now available for 64bit ARM
583 * cross-compiling for ARMV8 under iOS now works.
584 * cpu-specific code has been rearranged to make better use of both
585 hardware commonalities and model-specific compiler optimizations.
586 * XGENE1 has been removed as a TARGET, superseded by the improved generic
590 * Older assembly mnemonics have been converted to UAL form to allow
591 building with clang 7.0
592 * Cross compiling LAPACKE for Android has been fixed again (broken by
593 update to LAPACK 3.7.0 some while ago).
595 ====================================================================
600 * thread memory allocation has been switched back to the method
601 used before version 0.3.1 due to unexpected problems caused by
602 the new code under some circumstances. A new compile-time option
603 USE_TLS has been added to enable the new code, and it is hoped
604 that this can become the default again in the next version.
605 * LAPAck PR272 has been integrated, which fixes spurious errors
606 in DSYEVR and related functions caused by missing conversion
607 from ILAENV to ILAENV_2STAGE in several _2stage routines.
608 * the cmake-generated OpenBLASConfig.cmake now uses correct case
609 for the name of the library
610 * added support for Haiku OS
613 * added AVX512 implementations of SDOT, DDOT, SAXPY, DAXPY,
614 DSCAL, DGEMVN and DSYMVL
615 * added a workaround for a cygwin issue that prevented compilation
619 * added autodetection of Z14
620 * fixed TRMM errors in the generic target
622 ====================================================================
627 * fixes for regressions caused by the rewrite of the thread
628 initialization code in 0.3.1
631 * fixed cpu autodetection for the BSDs
634 * fixed utest errors in AXPY, DSDOT, ROT and SWAP
637 * added autodetection of AMD Ryzen 2
638 * fixed build with older versions of MSVC
640 ====================================================================
645 * rewritten thread initialization code with significantly reduced overhead
646 * added CBLAS interfaces to the IxAMIN BLAS extension functions
647 * fixed the lapack-test target
648 * CMAKE builds now create an OpenBLASConfig.cmake file
649 * ZAXPY now uses a single thread for small input sizes
650 * the LAPACK code was updated from Reference-LAPACK/lapack#253
651 (fixing LAPACKE interfaces to Aasen's functions)
654 * corrected CROT and ZROT behaviour with zero INC_X
657 * corrected xDOT behaviour with zero INC_X or INC_Y
660 * retired some older targets of DYNAMIC_ARCH builds to a new option DYNAMIC_OLDER,
661 this affects PENRYN,DUNNINGTON,OPTERON,OPTERON_SSE3,BOBCAT,ATOM and NANO
662 (which will still be supported via the slower PRESCOTT kernels when this option is not set)
663 * added an option DYNAMIC_LIST that (used in conjunction with DYNAMIC_ARCH) allows to
664 specify the list of x86_64 targets to include. Any target not on the list will be supported
665 by the Sandybridge or Nehalem kernels if available, or by Prescott.
666 * improved SWITCH_RATIO on Haswell for increased GEMM throughput
667 * added initial support for Intel Skylake X, including an AVX512 SGEMM kernel
668 * added autodetection of Intel Cannon Lake series as Skylake X
669 * added a default L2 cache size for hypervisors that return zero here (Chromebook)
670 * fixed a name clash with recent Windows10 headers that broke the build with (at least)
671 recent mingw from MSYS2
672 * fixed a link error in mixed clang/gfortran builds with OpenMP
673 * updated the OSX deployment target to 10.8
674 * switched on parallel make for builds on MS Windows by default
677 * fixed SSWAP and DSWAP behaviour with zero INC_X and INC_Y
679 ====================================================================
684 * fixed some more thread race and locking bugs
685 * added preliminary support for calling an OpenMP build of the library from multiple threads
686 * removed performance impact of thread locks added in 0.2.20 on OpenMP code
687 * general code cleanup
688 * optimized DSDOT implementation
689 * improved thread distribution for GEMM
690 * corrected IMATCOPY/OMATCOPY implementation
691 * fixed out-of-bounds accesses in the multithreaded xBMV/xPMV and SYMV implementations
692 * cmake build improvements
693 * pkgconfig file now contains build options
694 * openblas_get_config() now reports USE_OPENMP and NUM_THREADS settings used for the build
695 * corrections and improvements for systems with more than 64 cpus
696 * LAPACK code updated to 3.8.0 including later fixes
697 * added ReLAPACK, a recursive implementation of several LAPACK functions
698 * Rewrote ROTMG to handle cases that the netlib code failed to address
699 * Disabled (broken) multithreading code for xTRMV
700 * corrected prototypes of complex CBLAS functions to make our cblas.h match the generally accepted standard
701 * shared memory access failures on startup are now handled more gracefully
702 * restored utests from earlier releases (and made them pass on all affected systems)
705 * several fixes for cpu autodetection
708 * corrected vector register overwriting in several Power8 kernels
709 * optimized additional BLAS functions
712 * added support for CortexA53 and A72
713 * added autodetection for ThunderX2T99
714 * made most optimized kernels the default for generic ARMv8 targets
717 * parallelized DDOT kernel for Haswell
718 * changed alignment directives in assembly kernels to boost performance on OSX
719 * fixed register handling in the GEMV microkernels (bug exposed by gcc7)
720 * added support for building on OpenBSD and Dragonfly
721 * updated compiler options to work with Intel release 2018
722 * support fully optimized build with clang/flang on Microsoft Windows
723 * fixed building on AIX
726 * added optimized BLAS 1/2 functions
729 * fixed cpu autodetection helper code
730 * added mips32 1004K cpu (Mediatek MT7621 and similar SoC)
731 * added mips64 I6500 cpu
733 ====================================================================
738 * Improved CMake support
739 * Fixed several thread race and locking bugs
740 * Fixed default LAPACK optimization level
741 * Updated LAPACK to 3.7.0
742 * Added ReLAPACK (https://github.com/HPAC/ReLAPACK, make BUILD_RELAPACK=1)
745 * Optimizations for Power9
746 * Fixed several Power8 assembly bugs
749 * New optimized Vulcan and ThunderX2T99 targets
750 * Support for ARMV7 SOFT_FP ABI (make ARM_SOFTFP_ABI=1)
751 * Detect all cpu cores including offline ones
752 * Fix compilation with CLANG
753 * Support building a shared library for Android
756 * Fixed several threading issues
757 * Fix compilation with CLANG
760 * Detect Intel Bay Trail and Apollo Lake
761 * Detect Intel Sky Lake and Kaby Lake
762 * Detect Intel Knights Landing
763 * Detect AMD A8, A10, A12 and Ryzen
764 * Support 64bit builds with Visual Studio
765 * Fix building with Intel and PGI compilers
766 * Fix building with MINGW and TDM-GCC
767 * Fix cmake builds for Haswell and related cpus
768 * Fix building for Sandybridge with CLANG 3.9
769 * Add support for the FLANG compiler
772 * New target z13 with BLAS3 optimizations
774 ====================================================================
778 * Improved cross compiling.
779 * Fix the bug on musl libc.
782 * Optimize BLAS on Power8
783 * Fixed Julia+OpenBLAS bugs on Power8
786 * Optimize BLAS on MIPS P5600 and I6400 (Thanks, Shivraj Patil, Kaustubh Raste)
789 * Improved on ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
792 ====================================================================
796 * If you set MAKE_NB_JOBS flag less or equal than zero,
797 make will be without -j.
800 * Support building Visual Studio static library. (#813, Thanks, theoractice)
801 * Fix bugs to pass buidbot CI tests (http://build.openblas.net)
804 * Provide DGEMM 8x4 kernel for Cortex-A57 (Thanks, Ashwin Sekhar T K)
807 * Optimize S and C BLAS3 on Power8
808 * Optimize BLAS2/1 on Power8
810 ====================================================================
814 * Enable BUILD_LAPACK_DEPRECATED=1 by default.
816 ====================================================================
820 * Avoid potential getenv segfault. (#716)
821 * Import LAPACK svn bugfix #142-#147,#150-#155
824 * Optimize c/zgemv for AMD Bulldozer, Piledriver, Steamroller
825 * Fix bug with scipy linalg test.
828 * Improve DGEMM for ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
831 * Optimize D and Z BLAS3 functions for Power8.
833 ====================================================================
837 * Upgrade LAPACK to 3.6.0 version.
838 Add BUILD_LAPACK_DEPRECATED option in Makefile.rule to build
839 LAPACK deprecated functions.
840 * Add MAKE_NB_JOBS option in Makefile.
841 Force number of make jobs.This is particularly
842 useful when using distcc. (#735. Thanks, Jerome Robert.)
843 * Redesign unit test. Run unit/regression test at every build (Travis-CI and Appveyor).
844 * Disable multi-threading for small size swap and ger. (#744. Thanks, Jerome Robert)
845 * Improve small zger, zgemv, ztrmv using stack alloction (#727. Thanks, Jerome Robert)
846 * Let openblas_get_num_threads return the number of active threads.
847 (#760. Thanks, Jerome Robert)
848 * Support illumos(OmniOS). (#749. Thanks, Lauri Tirkkonen)
849 * Fix LAPACK Dormbr, Dormlq bug. (#711, #713. Thanks, Brendan Tracey)
850 * Update scipy benchmark script. (#745. Thanks, John Kirkham)
853 * Optimize trsm kernels for AMD Bulldozer, Piledriver, Steamroller.
854 * Detect Intel Avoton.
855 * Detect AMD Trinity, Richland, E2-3200.
856 * Fix gemv performance bug on Mac OSX Intel Haswell.
857 * Fix some bugs with CMake and Visual Studio
860 * Support and optimize Cortex-A57 AArch64.
861 (#686. Thanks, Ashwin Sekhar TK)
862 * Fix Android build on ARMV7 (#778. Thanks, Paul Mustiere)
863 * Update ARMV6 kernels.
866 * Fix detection of POWER architecture
867 (#684. Thanks, Sebastien Villemot)
869 ====================================================================
873 * Support cmake on x86/x86-64. Natively compiling on MS Visual Studio.
874 (experimental. Thank Hank Anderson for the initial cmake porting work.)
876 On Linux and Mac OSX, OpenBLAS cmake supports assembly kernels.
881 On Windows MS Visual Studio, OpenBLAS cmake only support C kernels.
882 (OpenBLAS uses AT&T style assembly, which is not supported by MSVC.)
883 e.g. cmake -G "Visual Studio 12 Win64" .
884 Open OpenBLAS.sln and build.
886 * Enable MAX_STACK_ALLOC flags by default.
887 Improve ger and gemv for small matrices.
888 * Improve gemv parallel with small m and large n case.
889 * Improve ?imatcopy when lda==ldb (#633. Thanks, Martin Koehler)
890 * Add vecLib benchmarks (#565. Thanks, Andreas Noack.)
891 * Fix LAPACK lantr for row major matrices (#634. Thanks, Dan Kortschak)
892 * Fix LAPACKE lansy (#640. Thanks, Dan Kortschak)
893 * Import bug fixes for LAPACKE s/dormlq, c/zunmlq
894 * Raise the signal when pthread_create fails (#668. Thanks, James K. Lowden)
895 * Remove g77 from compiler list.
896 * Enable AppVeyor Windows CI.
899 * Support pure C generic kernels for x86/x86-64.
900 * Support Intel Boardwell and Skylake by Haswell kernels.
901 * Support AMD Excavator by Steamroller kernels.
902 * Optimize s/d/c/zdot for Intel SandyBridge and Haswell.
903 * Optimize s/d/c/zdot for AMD Piledriver and Steamroller.
904 * Optimize s/d/c/zapxy for Intel SandyBridge and Haswell.
905 * Optimize s/d/c/zapxy for AMD Piledriver and Steamroller.
906 * Optimize d/c/zscal for Intel Haswell, dscal for Intel SandyBridge.
907 * Optimize d/c/zscal for AMD Bulldozer, Piledriver and Steamroller.
908 * Optimize s/dger for Intel SandyBridge.
909 * Optimize s/dsymv for Intel SandyBridge.
910 * Optimize ssymv for Intel Haswell.
911 * Optimize dgemv for Intel Nehalem and Haswell.
912 * Optimize dtrmm for Intel Haswell.
915 * Support Android NDK armeabi-v7a-hard ABI (-mfloat-abi=hard)
916 e.g. make HOSTCC=gcc CC=arm-linux-androideabi-gcc NO_LAPACK=1 TARGET=ARMV7
917 * Fix lock, rpcc bugs (#616, #617. Thanks, Grazvydas Ignotas)
919 * Support ppc64le platform (ELF ABI v2. #612. Thanks, Matthew Brandyberry.)
920 * Support POWER7/8 by POWER6 kernels. (#612. Thanks, Fábio Perez.)
922 ====================================================================
926 * Improve OpenBLASConfig.cmake. (#474, #475. Thanks, xantares.)
927 * Improve ger and gemv for small matrices by stack allocation.
928 e.g. make -DMAX_STACK_ALLOC=2048 (#482. Thanks, Jerome Robert.)
929 * Introduce openblas_get_num_threads and openblas_get_num_procs.
930 (#497. Thanks, Erik Schnetter.)
931 * Add ATLAS-style ?geadd function. (#509. Thanks, Martin Köhler.)
932 * Fix c/zsyr bug with negative incx. (#492.)
933 * Fix race condition during shutdown causing a crash in
934 gotoblas_set_affinity(). (#508. Thanks, Ton van den Heuvel.)
937 * Support AMD Streamroller.
940 * Add Cortex-A9 and Cortex-A15 targets.
942 ====================================================================
946 * Add SYMBOLPREFIX and SYMBOLSUFFIX makefile options
947 for adding a prefix or suffix to all exported symbol names
948 in the shared library.(#459, Thanks Tony Kelman)
949 * Provide OpenBLASConfig.cmake at installation.
950 * Fix Fortran compiler detection on FreeBSD.
951 (#470, Thanks Mike Nolta)
955 * Add generic kernel files for x86-64. make TARGET=GENERIC
956 * Fix a bug of sgemm kernel on Intel Sandy Bridge.
957 * Fix c_check bug on some amd64 systems. (#471, Thanks Mike Nolta)
960 * Support APM's X-Gene 1 AArch64 processors.
961 Optimize trmm and sgemm. (#465, Thanks Dave Nuechterlein)
963 ====================================================================
967 * Added CBLAS interface for ?omatcopy and ?imatcopy.
968 * Enable ?gemm3m functions.
969 * Added benchmark for ?gemm3m.
970 * Optimized multithreading lower limits.
971 * Disabled SYMM3M and HEMM3M functions
972 because of segment violations.
975 * Improved axpy and symv performance on AMD Bulldozer.
976 * Improved gemv performance on modern Intel and AMD CPUs.
978 ====================================================================
982 * Added some benchmark codes.
983 * Fix link error on Linux/musl.(Thanks Isaac Dunham)
986 * Improved s/c/zgemm performance for Intel Haswell.
987 * Improved s/d/c/zgemv performance.
988 * Support the big numa machine.(EXPERIMENT)
991 * Fix detection when cpuinfo uses "Processor". (Thanks Isaiah)
993 ====================================================================
997 * Added BLAS extensions as following.
998 s/d/c/zaxpby, s/d/c/zimatcopy, s/d/c/zomatcopy.
999 * Added OPENBLAS_CORETYPE environment for dynamic_arch. (a86d34)
1000 * Added NO_AVX2 flag for old binutils. (#401)
1001 * Support outputing the CPU corename on runtime.(#407)
1002 * Patched LAPACK to fix bug 114, 117, 118.
1003 (http://www.netlib.org/lapack/bug_list.html)
1004 * Disabled ?gemm3m for a work-around fix. (#400)
1006 * Fixed lots of bugs for optimized kernels on sandybridge,Haswell,
1007 bulldozer, and piledriver.
1008 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
1011 * Improved LAPACK testing.
1013 ====================================================================
1017 * Improved the result for LAPACK testing. (#372)
1018 * Installed DLL to prefix/bin instead of prefix/lib. (#366)
1019 * Build import library on Windows.(#374)
1021 * To improve LAPACK testing, we fallback some kernels. (#372)
1022 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
1024 ====================================================================
1028 * Added OPENBLAS_VERBOSE environment variable.(#338)
1029 * Make OpenBLAS thread-pool resilient to fork via pthread_atfork.
1030 (#294, Thank Olivier Grisel)
1034 * Detect Intel Haswell for new Macbook.
1036 ====================================================================
1040 * Update LAPACK to 3.5.0 version
1041 * Fixed compatiable issues with Clang and Pathscale compilers.
1044 * Optimization on Intel Haswell.
1045 * Enable optimization kernels on AMD Bulldozer and Piledriver.
1048 * Support ARMv6 and ARMv7 ISA.
1049 * Optimization on ARM Cortex-A9.
1051 ====================================================================
1055 * Support Open64 5.0. (#266)
1056 * Add executable stack markings. (#262, Thank Sébastien Fabbro)
1057 * Respect user's LDFLAGS (Thank Sébastien Fabbro)
1060 * Rollback bulldozer and piledriver kernels to barcelona kernels (#263)
1061 We will fix the compuational error bug in bulldozer and piledriver kernels.
1063 ====================================================================
1067 * Support LSB (Linux Standard Base) 4.1.
1069 * Include LAPACK 3.4.2 source codes to the repo.
1070 Avoid downloading at compile time.
1071 * Add NO_PARALLEL_MAKE flag to disable parallel make.
1072 * Create openblas_get_parallel to retrieve information which
1073 parallelization model is used by OpenBLAS. (Thank grisuthedragon)
1074 * Detect LLVM/Clang compiler. The default compiler is Clang on Mac OS X.
1075 * Change LIBSUFFIX from .lib to .a on windows.
1076 * A work-around for dtrti_U single thread bug. Replace it with LAPACK codes. (#191)
1079 * Optimize c/zgemm, trsm, dgemv_n, ddot, daxpy, dcopy on
1080 AMD Bulldozer. (Thank Werner Saar)
1081 * Add Intel Haswell support (using Sandybridge optimizations).
1083 * Add AMD Piledriver support (using Bulldozer optimizations).
1084 * Fix the computational error in zgemm avx kernel on
1086 * Fix the overflow bug in gemv.
1087 * Fix the overflow bug in multi-threaded BLAS3, getrf when NUM_THREADS
1088 is very large.(#214, #221, #246).
1090 * Support loongcc (Open64 based) compiler for ICT Loongson 3A/B.
1093 * Support Power7 by old Power6 kernels. (#220)
1095 ====================================================================
1099 * Improved OpenMP performance slightly. (d744c9)
1100 * Improved cblas.h compatibility with Intel MKL.(#185)
1101 * Fixed the overflowing bug in single thread cholesky factorization.
1102 * Fixed the overflowing buffer bug of multithreading hbmv and sbmv.(#174)
1105 * Added AMD Bulldozer x86-64 S/DGEMM AVX kernels. (Thank Werner Saar)
1106 We will tune the performance in future.
1107 * Auto-detect Intel Xeon E7540.
1108 * Fixed the overflowing buffer bug of gemv. (#173)
1109 * Fixed the bug of s/cdot about invalid reading NAN on x86_64. (#189)
1113 ====================================================================
1117 * Added NO_SHARED flag to disable generating the shared library.
1118 * Compile LAPACKE with ILP64 modle when INTERFACE64=1 (#158)
1119 * Export LAPACK 3.4.2 symbols in shared library. (#147)
1120 * Only detect the number of physical CPU cores on Mac OSX. (#157)
1121 * Fixed NetBSD build. (#155)
1122 * Fixed compilation with TARGET=GENERIC. (#160)
1124 * Restore the original CPU affinity when calling
1125 openblas_set_num_threads(1) (#153)
1126 * Fixed a SEGFAULT bug in dgemv_t when m is very large.(#154)
1129 ====================================================================
1133 * Upgraded LAPACK to 3.4.2 version. (#145)
1134 * Provided support for passing CFLAGS, FFLAGS, PFLAGS,
1135 FPFLAGS to make. (#137)
1136 * f77blas.h:compatibility for compilers without C99 complex
1137 number support. (#141)
1139 * Added NO_AVX flag. Check OS supporting AVX on runtime. (#139)
1140 * Fixed zdot incompatibility ABI issue with GCC 4.7 on
1141 Windows 32-bit. (#140)
1143 * Fixed the generation of shared library bug.
1144 * Fixed the detection bug on the Loongson 3A server.
1145 ====================================================================
1149 * Fixed LAPACK unstable bug about ?laswp. (#130)
1150 * Fixed the shared library bug about unloading the library on
1152 * Fixed the compilation failure on BlueGene/P (TARGET=PPC440FP2)
1153 Please use gcc and IBM xlf. (#134)
1155 * Supported goto_set_num_threads and openblas_set_num_threads
1156 APIs in Windows. They can set the number of threads on runtime.
1158 ====================================================================
1162 * Fixed exporting DLL functions bug on Windows/MingW
1163 * Support GNU Hurd (Thank Sylvestre Ledru)
1164 * Support kfreebsd kernel (Thank Sylvestre Ledru)
1166 * Support Intel Sandy Bridge 22nm desktop/mobile CPU
1168 * Improve the detection of SPARC (Thank Sylvestre Ledru)
1170 ====================================================================
1175 * Fixed the SEGFAULT bug about hyper-theading
1176 * Support AMD Bulldozer by using GotoBLAS2 AMD Barcelona codes
1178 ====================================================================
1182 * Removed the limitation (64) of numbers of CPU cores.
1183 Now, it supports 256 cores at max.
1184 * Supported clang compiler.
1185 * Fixed some build bugs on FreeBSD
1187 * Optimized Level-3 BLAS on Intel Sandy Bridge x86-64 by AVX instructions.
1188 Please use gcc >= 4.6 or clang >=3.1.
1189 * Support AMD Bobcat by using GotoBLAS2 AMD Barcelona codes.
1191 ====================================================================
1195 * Upgraded LAPACK to 3.4.1 version. (Thank Zaheer Chothia)
1196 * Supported LAPACKE, a C interface to LAPACKE. (Thank Zaheer Chothia)
1197 * Fixed the build bug (MD5 and download) on Mac OSX.
1198 * Auto download CUnit 2.1.2-2 from SF.net with UTEST_CHECK=1.
1199 * Fxied the compatibility issue for compilers without C99 complex number
1200 (e.g. Visual Studio)
1202 * Auto-detect Intel Sandy Bridge Core i7-3xxx & Xeon E7 Westmere-EX.
1203 * Test alpha=Nan in dscale.
1204 * Fixed a SEGFAULT bug in samax on x86 windows.
1206 ====================================================================
1210 * Set soname of shared library on Linux.
1211 * Added LIBNAMESUFFIX flag in Makefile.rule. The user can use
1212 this flag to control the library name, e.g. libopenblas.a,
1213 libopenblas_ifort.a or libopenblas_omp.a.
1214 * Added GEMM_MULTITHREAD_THRESHOLD flag in Makefile.rule.
1215 The lib use single thread in GEMM function with small matrices.
1217 * Used GEMV SSE/SSE2 kernels on x86 32-bit.
1218 * Exported CBLAS functions in Windows DLL.
1220 * Completed Level-3 BLAS optimization on Loongson 3A CPU.
1221 * Improved GEMV performance on Loongson 3A CPU.
1222 * Improved Level-3 BLAS performance on Loongson 3B CPU. (EXPERIMENT)
1224 ====================================================================
1225 Version 0.1 alpha2.5
1228 * Fixed missing "#include <sched.h>" bug on Mac OS X.
1229 Thank Mike Nolta for the patch.
1230 * Upgraded LAPACK to 3.4.0 version
1231 * Fixed a bug on Mac OS X. Don't require SystemStubs on OS X.
1232 SystemStubs does not exist on Lion. Thank Stefan Karpinski.
1233 * Improved README with using OpenMP. Check the internal threads
1234 count less than or equal to omp_get_max_threads()
1236 * Auto-detect Intel Core i6/i7 (Sandy Bridge) CPU with Nehalem assembly kernels
1237 * Fixed some bugs on MingW 64-bit including zgemv, cdot, zdot.
1239 ====================================================================
1240 Version 0.1 alpha2.4
1243 * Fixed a bug about installation. The header file "fblas77.h"
1245 * Fixed #61 a building bug about setting TARGET and DYNAMIC_ARCH.
1246 * Try to handle absolute path of shared library in OSX. (#57)
1247 Thank Dr Kane O'Donnell.
1248 * Changed the installation folder layout to $(PREFIX)/include and
1252 * Fixed #58 zdot/xdot SEGFAULT bug with GCC-4.6 on x86. According
1253 to i386 calling convention, The callee should remove the first
1254 hidden parameter.Thank Mr. John for this patch.
1256 ====================================================================
1257 Version 0.1 alpha2.3
1261 * Added DTB_ENTRIES into dynamic arch setting parameters. Now,
1262 it can read DTB_ENTRIES on runtime. (Refs issue #55 on github)
1264 ====================================================================
1265 Version 0.1 alpha2.2
1269 * Fixed a building bug when DYNAMIC_ARCH=1 & INTERFACE64=1.
1270 (Refs issue #44 on github)
1272 ====================================================================
1273 Version 0.1 alpha2.1
1277 * Stop the build and output the error message when detecting
1278 fortran compiler failed. (Refs issue #42 on github)
1280 ====================================================================
1285 * Fixed blasint undefined bug in <cblas.h> file. Other software
1286 could include this header successfully(Refs issue #13 on github)
1287 * Fixed the SEGFAULT bug on 64 cores. On SMP server, the number
1288 of CPUs or cores should be less than or equal to 64.(Refs issue #14
1290 * Support "void goto_set_num_threads(int num_threads)" and "void
1291 openblas_set_num_threads(int num_threads)" when USE_OPENMP=1
1292 * Added extern "C" to support C++. Thank Tasio for the patch(Refs
1293 issue #21 on github)
1294 * Provided an error message when the arch is not supported.(Refs
1295 issue #19 on github)
1296 * Fixed issue #23. Fixed a bug of f_check script about generating link flags.
1297 * Added openblas_set_num_threads for Fortran.
1298 * Fixed #25 a wrong result of rotmg.
1299 * Fixed a bug about detecting underscore prefix in c_check.
1300 * Print the wall time (cycles) with enabling FUNCTION_PROFILE
1301 * Fixed #35 a build bug with NO_LAPACK=1 & DYNAMIC_ARCH=1
1302 * Added install target. You can use "make install". (Refs #20)
1306 * Fixed #28 a wrong result of dsdot on x86_64.
1307 * Fixed #32 a SEGFAULT bug of zdotc with gcc-4.6.
1308 * Fixed #33 ztrmm bug on Nehalem.
1309 * Work-around #27 the low performance axpy issue with small imput size & multithreads.
1312 * Fixed #28 a wrong result of dsdot on Loongson3A/MIPS64.
1313 * Optimized single/double precision BLAS Level3 on Loongson3A/MIPS64. (Refs #2)
1314 * Optimized single/double precision axpy function on Loongson3A/MIPS64. (Refs #3)
1316 ====================================================================
1321 * Support "make NO_LAPACK=1" to build the library without
1323 * Fixed randomly SEGFAULT when nodemask==NULL with above Linux 2.6.34.
1324 Thank Mr.Ei-ji Nakama providing this patch. (Refs issue #12 on github)
1325 * Added DEBUG=1 rule in Makefile.rule to build debug version.
1326 * Disable compiling quad precision in reference BLAS library(netlib BLAS).
1327 * Added unit testcases in utest/ subdir. Used CUnit framework.
1328 * Supported OPENBLAS_* & GOTO_* environment variables (Pleas see README)
1329 * Imported GotoBLAS2 1.13 BSD version
1332 * On x86 32bits, fixed a bug in zdot_sse2.S line 191. This would casue
1333 zdotu & zdotc failures. Instead, work-around it. (Refs issue #8 #9 on github)
1334 * Modified ?axpy functions to return same netlib BLAS results
1335 when incx==0 or incy==0 (Refs issue #7 on github)
1336 * Modified ?swap functions to return same netlib BLAS results
1337 when incx==0 or incy==0 (Refs issue #6 on github)
1338 * Modified ?rot functions to return same netlib BLAS results
1339 when incx==0 or incy==0 (Refs issue #4 on github)
1340 * Detect Intel Westmere,Intel Clarkdale and Intel Arrandale
1341 to use Nehalem codes.
1342 * Fixed a typo bug about compiling dynamic ARCH library.
1344 * Improve daxpy performance on ICT Loongson 3A.
1345 * Supported ICT Loongson 3A CPU (Refs issue #1 on github)
1346 ====================================================================