2 ====================================================================
7 - reverted the optimization of SGEMV_N/DGEMV_N for small input sizes
8 and consecutive arguments as it led to stack overflows on x86_64
9 with some operating systems (notably OSX and Windows)
12 - reverted the performance patch for SGEMV_T on AVX512 as it caused
13 wrong results in some applications
16 - fixed compilation with compilers other than gcc
17 ====================================================================
22 - drastically reduced the stack size requirements for running the LAPACK
23 testsuite (Reference-LAPACK PR 553)
24 - fixed spurious test failures in the LAPACK testsuite (Reference-LAPACK
26 - expressly setting DYNAMIC_ARCH=0 no longer enables dynamic_arch mode
27 - improved performance of xGER, xSPR, xSPR2, xSYR, xSYR2, xTRSV, SGEMV_N
28 and DGEMV_N, for small input sizes and consecutive arguments
29 - improved performance of xGETRF, xPORTF and xPOTRI for small input sizes
30 by disabling multithreading
31 - fixed installing with BSD versions of the "install" utility
34 - fixed the implementation of xIMIN
35 - improved the performance of DSDOT
36 - fixed linking of the tests on C910V with current vendor gcc
39 - fixed SBGEMM computation for some odd value inputs
40 - fixed compilation for PPCG4, PPC970, POWER3, POWER4 and POWER5
43 - improved performance of SGEMV_N and SGEMV_T for small N on AVX512-capable cpus
44 - worked around a miscompilation of ZGEMM/ZTRMM on Sandybridge with old gcc
46 - fixed compilation with MS Visual Studio versions older than 2017
47 - fixed macro name collision with winnt.h from the latest Win10 SDK
48 - added cpu type autodetection for Intel Ice Lake SP
49 - fixed cpu type autodetection for Intel Tiger Lake
50 - added cpu type autodetection for recent Centaur/Zhaoxin models
51 - fixed compilation with musl libc
54 - fixed compilation with gcc/gfortran on the Apple M1
55 - fixed linking of the tests on FreeBSD
56 - fixed missing restore of a register in the recently rewritten DNRM2 kernel
57 for ThunderX2 and Neoverse N1 that could cause spurious failures in e.g.
59 - added compiler optimization flags for the EMAG8180
60 - added initial support for Cortex A55
63 - fixed linking of the tests on FreeBSD
65 ====================================================================
70 - imported improvements and bugfixes from Reference-LAPACK 3.9.1
71 - imported LAPACKE interface fixes from Reference-LAPACK PRs 534 + 537
72 - fixed a problem in the cpu detection of 0.3.14 that prevented cross-compilation
73 - fixed a sequence problem in the generation of softlinks to the library in GMAKE
76 - fixed compilation on RISCV (missing entry in getarch)
77 - fixed a potential division by zero in CROTG and ZROTG
80 - fixed LAPACK testsuite failures seen with the NVIDIA HPC compiler
81 - improved CGEMM, DGEMM and ZGEMM performance on POWER10
82 - added an optimized ZGEMV kernel for POWER10
83 - fixed a potential division by zero in CROTG and ZROTG
86 - added support for Intel Control-flow Enforcement Technology (CET)
87 - reverted the DOMATCOPY_RT code to the generic C version
88 - fixed a bug in the AVX512 SGEMM kernel introduced in 0.3.14
89 - fixed misapplication of -msse flag to non-SSE cpus in DYNAMIC_ARCH
90 - added support for compilation of the benchmarks on older OSX versions
91 - fix propagation of the NO_AVX512 option in CMAKE builds
92 - fix compilation of the AVX512 SGEMM kernel with clang-cl on Windows
93 - fixed compilation of the CTESTs with INTERFACE64=1 (random faults on OSX)
94 - corrected the Haswell DROT kernel to require AVX2/FMA3 rather than AVX512
97 - fixed a potential division by zero in CROTG and ZROTG
98 - fixed a potential overflow in IMATCOPY/ZIMATCOPY and the CTESTs
101 - fixed spurious reads outside the array in the SGEMM tcopy macro
102 - fixed a potential division by zero in CROTG and ZROTG
103 - fixed a segmentation fault in DYNAMIC_ARCH builds (reappeared in 0.3.14)
106 - fixed a potential division by zero in CROTG and ZROTG
107 - fixed a potential overflow in IMATCOPY/ZIMATCOPY and the CTESTs
110 - fixed a potential division by zero in CROTG and ZROTG
113 - fixed a potential division by zero in CROTG and ZROTG
115 ====================================================================
120 * Fixed a race condition on thread shutdown in non-OpenMP builds
121 * Fixed custom BUFFERSIZE option getting ignored in gmake builds
122 * Fixed CMAKE compilation of the TRMM kernels for GENERIC platforms
123 * Added CBLAS interfaces for CROTG, ZROTG, CSROT and ZDROT
124 * Improved performance of OMATCOPY_RT across all platforms
125 * Changed perl scripts to use env instead of a hardcoded /usr/bin/perl
126 * Fixed potential misreading of the GCC compiler version in the build scripts
127 * Fixed convergence problems in LAPACK complex GGEV/GGES (Reference-LAPACK #477)
128 * Reduced the stacksize requirements for running the LAPACK testsuite (Reference-LAPACK #335)
131 * Fixed compilation on RISCV (missing entry in getarch)
134 * Fixed compilation for DYNAMIC_ARCH with clang and with old gcc versions
135 * Added support for compilation on FreeBSD/ppc64le
136 * Added optimized POWER10 kernels for SSCAL, DSCAL, CSCAL, ZSCAL
137 * Added optimized POWER10 kernels for SROT, DROT, CDOT, SASUM, DASUM
138 * Improved SSWAP, DSWAP, CSWAP, ZSWAP performance on POWER10
139 * Improved SCOPY and CCOPY performance on POWER10
140 * Improved SGEMM and DGEMM performance on POWER10
141 * Added support for compilation with the NVIDIA HPC compiler
144 * Added an optimized bfloat16 GEMM kernel for Cooperlake
145 * Added CPUID autodetection for Intel Rocket Lake and Tiger Lake cpus
146 * Improved the performance of SASUM,DASUM,SROT,DROT on AMD Ryzen cpus
147 * Added support for compilation with the NAG Fortran compiler
148 * Fixed recognition of the AMD AOCC compiler
149 * Fixed compilation for DYNAMIC_ARCH with clang on Windows
150 * Added support for running the BLAS/CBLAS tests on Windows
151 * Fixed signatures of the tls callback functions for Windows x64
152 * Fixed various issues with fma intrinsics support handling
155 * Added support for embedded Cortex M targets via a new option EMBEDDED
158 * Fixed the THUNDERX2T99 and NEOVERSEN1 DNRM2/ZNRM2 kernels for inputs with Inf
159 * Added support for the DYNAMIC_LIST option
160 * Added support for compilation with the NVIDIA HPC compiler
161 * Added support for compiling with the NAG Fortran compiler
163 ====================================================================
168 * Added a generic bfloat16 SBGEMV kernel
169 * Fixed a potentially severe memory leak after fork in OpenMP builds
170 that was introduced in 0.3.12
171 * Added detection of the Fujitsu Fortran compiler
172 * Added detection of the (e)gfortran compiler on OpenBSD
173 * Added support for overriding the default name of the library independently
174 from symbol suffixing in the gmake builds (already supported in cmake)
177 * Added a RISC V port optimized for C910V
180 * Added optimized POWER10 kernels for SAXPY, CAXPY, SDOT, DDOT and DGEMV_N
181 * Improved DGEMM performance on POWER10
182 * Improved STRSM and DTRSM performance on POWER9 and POWER10
183 * Fixed segmemtation faults in DYNAMIC_ARCH builds
184 * Fixed compilation with the PGI compiler
187 * Fixed compilation of kernels that require SSE2 intrinsics since 0.3.12
190 * Added an optimized bfloat16 SBGEMV kernel for SkylakeX and Cooperlake
191 * Improved the performance of SASUM and DASUM kernels through parallelization
192 * Improved the performance of SROT and DROT kernels
193 * Improved the performance of multithreaded xSYRK
194 * Fixed OpenMP builds that use the LLVM Clang compiler together with GNU gfortran
195 (where linking of both the LLVM libomp and GNU libgomp could lead to lockups or
197 * Fixed miscompilations by old gcc 4.6
198 * Fixed misdetection of AVX2 capability in some Sandybridge cpus
199 * Fixed lockups in builds combining DYNAMIC_ARCH with TARGET=GENERIC on OpenBSD
202 * Fixed segmemtation faults in DYNAMIC_ARCH builds
205 * Improved kernels for Loongson 3R3 ("3A") and 3R4 ("3B") models, including MSA
206 * Fixed bugs in the MSA kernels for CGEMM, CTRMM, CGEMV and ZGEMV
207 * Added handling of zero increments in the MSA kernels for SSWAP and DSWAP
208 * Added DYNAMIC_ARCH support for MIPS64 (currently Loongson3R3/3R4 only)
211 * Fixed building 32 and 64 bit SPARC kernels with the SolarisStudio compilers
213 ====================================================================
218 * Fixed missing BLAS/LAPACK functions (inadvertently dropped during
219 the build system restructuring)
220 * Fixed argument conversion macro in LAPACKE_zgesvdq (LAPACK #458)
223 * Added optimized SCOPY/CCOPY kernels for POWER10
224 * Increased and unified the default size of the GEMM BUFFER
225 * Fixed building for POWER10 in DYNAMIC_ARCH mode
226 * POWER10 compatibility test now checks binutils version as well
227 * Cleaned up compiler warnings
230 * corrected compiler version checks for AVX2 compatibility
231 * added compiler option -mavx2 for building with flang
232 * fixed direct SGEMM pathway for small matrix sizes (broken by
233 the code refactoring in 0.3.11)
234 * fixed unhandled partial register clobbers in several kernels
235 for AXPY,DOT,GEMV_N and GEMV_T flagged by gcc10 tree-vectorizer
238 * improved Apple Vortex support to include cross-compiling
240 ====================================================================
246 the newly added BFLOAT16 functions were renamed to use the
247 letter "B" instead of "H" to avoid potential confusion with
248 the IEEE "half precision float" type, i.e. the 0.3.10
249 SHGEMM is now SBGEMM and the corresponding build option
250 was changed from "BUILD_HALF" to "BUILD_BFLOAT16".
251 * Reduced the default BLAS3_MEM_ALLOC_THRESHOLD (used as an upper
252 limit for placing temporary arrays on the stack) to be compatible
253 with a stack size of 1mb (as imposed by the JAVA runtime library)
254 * Added mixed-precision dot function SBDOT and utility functions
255 shstobf16, shdtobf16, sbf16tos and dbf16tod to convert between
256 single or double precision float arrays and bfloat16 arrays
257 * Fixed prototypes of LAPACK_?ggsvp and LAPACK_?ggsvd functions
259 * Fixed underflow and rounding errors in LAPACK SLANV2 and DLANV2
260 (causing miscalculations in e.g. SHSEQR/DHSEQR, LAPACK issue #263)
261 * Fixed workspace calculation in LAPACK ?GELQ (LAPACK issue #415)
262 * Fixed several bugs in the LAPACK testsuite
263 * Improved performance of TRMM and TRSM for certain problem sizes
264 * Fixed infinite recursions and workspace miscalculations in ReLAPACK
265 * CMAKE builds no longer require pkg-config for creating the .pc file
266 * Makefile builds no longer misread NO_CBLAS=0 or NO_LAPACK=0 as
267 enabling these options
268 * Fixed detection of gfortran when invoked through an mpi wrapper
269 * Improve thread reinitialization performance with OpenMP after a fork
270 * Added support for building only the subset of the library required
271 for a particular precision by specifying BUILD_SINGLE, BUILD_DOUBLE
272 * Optional function name prefixes and suffixes are now correctly
273 reflected in the generated cblas.h
274 * Added CMAKE build support for the LAPACK and multithreading tests
277 * Added optimized support for POWER10
278 * Added support for compiling for POWER8 in 32bit mode
279 * Added support for compilation with LLVM/clang
280 * Added support for compilation with NVIDIA/PGI compilers
281 * Fixed building on big-endian POWER8
282 * Fixed miscompilation of ZDOTC by gcc10
283 * Fixed alignment errors in the POWER8 SAXPY kernel
284 * Improved CPU detection on AIX
285 * Supported building with older compilers on POWER9
288 * Added support for Intel Cooperlake
289 * Added autodetection of AMD Renoir/Matisse/Zen3 cpus
290 * Added autodetection of Intel Comet Lake cpus
291 * Reimplemented ?sum, ?dot and daxpy using universal intrinsics
292 * Reset the fpu state before using the fpu on Windows as a workaround
293 for a problem introduced in Windows 10 build 19041 (a.k.a. SDK 2004)
294 * Fixed potentially undefined behaviour in the dot and gemv_t kernels
295 * Fixed a potential segmentation fault in DYNAMIC_ARCH builds
296 * Fixed building for ZEN with PGI/NVIDIA and AMD AOCC compilers
299 * Fixed cpu detection on BSD-like systems
302 * Added preliminary support for Apple Vortex cpus
303 * Added support for the Cavium ThunderX3T110 cpu
304 * Fixed cpu detection on BSD-like systems
305 * Fixed compilation in -std=C18 mode
308 * Added support for compiling with the clang compiler
309 * Improved GEMM performance on Z14
311 ====================================================================
316 * Improved thread locking behaviour in blas_server and parallel getrf
317 * Imported bugfix 394 from LAPACK (spurious reference to "XERBL"
318 due to overlong lines)
319 * Imported bugfix 403 from LAPACK (compile option "recursive" required
320 for correctness with Intel and PGI)
321 * Imported bugfix 408 from LAPACK (wrong scaling in ZHEEQUB)
322 * Imported bugfix 411 from LAPACK (infinite loop in LARGV/LARTG/LARTGP)
323 * Fixed mismatches between BUFFERSIZE and GEMM_UNROLL parameters that
324 could lead to crashes at large matrix sizes
325 * Restored internal soname in dynamic libraries on FreeBSD and Dragonfly
326 * Added API (openblas_setaffinity) to set the thread affinity on Linux
327 * Added initial infrastructure for half-precision floating point
328 (bfloat16) support with a generic implementation of SHGEMM
329 * Added CMAKE build system support for building the cblas_Xgemm3m
331 * Fixed CMAKE support for building in a path with embedded spaces
332 * Fixed CMAKE (non)handling of NO_EXPRECISION and MAX_STACK_ALLOC
333 * Fixed GCC version detection in the Makefiles
334 * Allowed overriding the names of AR, AS and LD in Makefile builds
337 * Fixed big-endian POWER8 ELFv2 builds on FreeBSD
338 * Fixed GCC version checks and DYNAMIC_ARCH builds on POWER9
339 * Fixed CMAKE build support for POWER9
340 * fixed a potential race condition in the thread buffer allocation
341 * Worked around LAPACK test failures on PPC G4
344 * Fixed a potential race condition in the thread buffer allocation
345 * Added support for MIPS 24K/24KE family based on P5600 kernels
348 * fixed a potential race condition in the thread buffer allocation
349 * Added TARGET=GENERIC
352 * Fixed a race condition in the thread buffer allocation
355 * Fixed a race condition in the thread buffer allocation
356 * Fixed zero initialisation in the assembly for SGEMM and DGEMM BETA
357 * Improved performance of the ThunderX2 DAXPY kernel
358 * Added an optimized SGEMM kernel for Cortex A53
359 * Fixed Makefile support for INTERFACE64 (8-byte integer)
362 * Fixed a syntax error in the CMAKE setup for SkylakeX
363 * Improved performance of STRSM on Haswell, SkylakeX and Ryzen
364 * Improved SGEMM performance on SGEMM for workloads with ldc a
366 * Improved DGEMM performance on Skylake X
367 * Fixed unwanted AVX512-dependency of SGEMM in DYNAMIC_ARCH
368 builds created on SkylakeX
369 * Removed data alignment requirement in the SSE2 copy kernels
370 that could cause spurious crashes
371 * Added a workaround for an optimizer bug in AppleClang 11.0.3
372 * Fixed LAPACK test failures due to wrong options for Intel Fortran
373 * Fixed compilation and LAPACK test results with recent Flang
375 * Fixed DYNAMIC_ARCH builds with CMAKE on OS X
376 * Fixed missing exports of cblas_i?amin, cblas_i?min, cblas_i?max,
377 cblas_?sum, cblas_?gemm3m in the shared library on OS
378 * Fixed reporting of cpu name in DYNAMIC_ARCH builds (would sometimes
379 show the name of an older generation chip supported by the same kernels)
382 * Improved performance of SGEMM/STRMM and DGEMM/DTRMM on Z14
384 ====================================================================
389 * Fixed a miscompilation of the GETRF functions with CMAKE
390 * Imported bugfix 390 from LAPACK (missing NaN propagation in xCOMBSSQ)
391 * The size of the memory buffer used for splitting GEMM tasks across
392 multiple threads can now be configured in the build system.
395 * Fixed several compilation problems related to endianness
396 and ELF version on POWER8 and POWER9
397 * Fixed use of the absolute value IAMIN/IAMAX instead of IMIN/IMAX
398 * Fixed a race condition in the level3 blas code
401 * Fixed use of the absoltute value IAMIN/IAMAX instead of IMIN/IMAX
404 * Fixed a race condition in the level3 blas code
405 * Fixed compilation on Android
407 * Added support for Ampere EMAG8180
408 * Added support for Neoverse N1
409 * Improved performance of the blas_lock function
410 * Fixed a race condition in the level3 blas code
411 * Fixed a performance regression on TSV110-based servers
414 * Fixed a long-standing error with undeclared register overwrites
415 in the DSCAL microkernel for HASWELL,SKYLAKEX and ZEN
416 * Fixed a long-standing bug in the SSE implementation of IAMAX
417 * Fixed a CMAKE build failure with DYNAMIC_ARCH
418 * Fixed cpu autodetection of Goldmont+, Cannon Lake and Ice Lake
419 * Fixed a compilation failure on OSX with compiler name containing dash
420 * Fixed compilation with MinGW on SkylakeX
421 * Improved speed of the AVX512 GEMM3M kernel on SkylakeX
422 * Added an AVX512 STRMM kernel for SkylakeX
423 * Improved GEMM performance on Haswell and Zen
426 * fixed compilation of the DYNAMIC_ARCH code
428 ====================================================================
433 ` * LAPACK has been updated to 3.9.0 (plus patches up to
435 * CMAKE support has been improved in several areas including
437 * a thread race condition in the GEMM3M kernels was resolved
438 * the "generic" (plain C) gemm beta kernel used by many targets
440 * an optimized version of the LAPACK trtrs functions has been added
441 * an incompatibilty between the LAPACK tests and the OpenBLAS
442 implementation of XERBLA was resolved, removing the numerous
443 warnings about wrong error exits in the former
444 * support for NetBSD has been added
445 * support for compilation with g95 and non-GNU versions of ld
447 * support for compilation with (upcoming) gcc 10 has been added
450 * worked around miscompilation of several POWER8 and POWER9
451 kernels by older versions of gcc
452 * added support for big-endian POWER8 and for compilation on AIX
453 * corrected bugs in the big-endian support for PPC440 and PPC970
454 * DYNAMIC_ARCH support is now available in CMAKE builds as well
457 * performance of DGEMM_BETA and SGEMM_NCOPY has been improved
458 * compilation for 32bit works again
459 * performance of the RPCC function has been improved
460 * improved performance on small systems
461 * DYNAMIC_ARCH support is now available in CMAKE builds as well
462 * cross-compilation from OSX to IOS was simplified
465 * a new AVX512 DGEMM kernel was added and the AVX512 SGEMM kernel
466 was significantly improved
467 * optimized AVX512 kernels for CGEMM and ZGEMM have been added
468 * AVX2 kernels for STRMM, SGEMM, and CGEMM have been significantly
469 sped up and optimized CGEMM3M and ZGEMM3M kernels have been added
470 * added support for QEMU virtual cpus
471 * a compilation problem with PGI and SUN compilers was fixed
472 * Intel "Goldmont plus" is now autodetected
473 * a potential crash on program exit on MS Windows has been fixed
476 * an unwanted case sensitivity in the implementation of LSAME
477 on older 32bit AMD cpus was fixed
480 * Z15 is now supported as Z14
481 * DYNAMIC_ARCH is now available on ZARCH as well
483 ====================================================================
488 * having the gmake special variables TARGET_ARCH or TARGET_MACH
489 defined no longer causes build failures in ctest or utest
490 * defining NO_AFFINITY or USE_TLS to 0 in gmake builds no longer
491 has the same effect as setting them to 1
492 * a new test program was added to allow checking the library for
494 * a new option USE_LOCKING was added to ensure thread safety when
495 OpenBLAS itself is built without multithreading but will be
496 called from multiple threads.
497 * a build failure on Linux with glibc versions earlier than 2.5
499 * a runtime error with CPU enumeration (and NO_AFFINITY not set)
500 on glibc 2.6 was fixed
501 * NO_AFFINITY was added to the CMAKE options (and defaults to being
502 active on Linux, as in the gmake builds)
505 * the build-time logic for detection of AVX512 availability in
506 the processor and compiler was fixed
507 * gmake builds on OSX now set the internal name of the library to
508 libopenblas.0.dylib (consistent with CMAKE)
509 * the Haswell DGEMM kernel received a significant speedup through
510 improved prefetch and load instructions
511 * performance of DGEMM, DTRMM, DTRSM and ZDOT on Zen/Zen2 was markedly
512 increased by avoiding vpermpd instructions
513 * the SKYLAKEX (AVX512) DGEMM helper functions have now been disabled
514 to fix remaining errors in DGEMM, DSYMM and DTRMM
517 * added support for building on FreeBSD/powerpc64 and FreeBSD/ppc970
518 * added optimized kernels for POWER9 SGEMM and STRMM
521 * fixed the softfp implementations of xAMAX and IxAMAX
522 * removed the predefined -march= flags on both ARMV5 and ARMV6 as
523 they were appropriate for only a subset of platforms
525 ====================================================================
530 * the build tools now check that a given cpu TARGET is actually valid
531 * the build-time check of system features (c_check) has been made
532 less dependent on particular perl features (this should mainly
533 benefit building on Windows)
534 * several problem with the ReLAPACK integration were fixed,
535 including INTERFACE64 support and building a shared library
536 * building with CMAKE on BSD systems was improved
537 * a non-absolute SUM function was added based on the
538 existing optimized code for ASUM
539 * CBLAS interfaces to the IxMIN and IxMAX functions were added
540 * a name clash between LAPACKE and BOOST headers was resolved
541 * CMAKE builds with OpenMP failed to include the appropriate getrf_parallel
543 * a crash on thread (key) deletion with the USE_TLS=1 memory management
545 * restored several earlier fixes, in particular for OpenMP performance,
546 building on BSD, and calling fork on CYGWIN, which had inadvertently
547 been dropped in the 0.3.3 rewrite of the memory management code.
550 * the AVX512 DGEMM kernel has been disabled again due to unsolved problems
551 * building with old versions of MSVC was fixed
552 * it is now possible to build a static library on Windows with CMAKE
553 * accessing environment variables on CYGWIN at run time was fixed
554 * the CMAKE build system now recognizes 32bit userspace on 64bit hardware
555 * Intel "Denverton" atom and Hygon "Dhyana" zen CPUs are now autodetected
556 * building for DYNAMIC_ARCH with a DYNAMIC_LIST of targets is now supported
558 * building for DYNAMIC_ARCH with GENERIC as the default target is now supported
559 * a buffer overflow in the SSE GEMM kernel for Intel Nano targets was fixed
560 * assembly bugs involving undeclared modification of input operands were fixed
561 in the AXPY, DOT, GEMV, GER, SCAL, SYMV and TRSM microkernels for Nehalem,
562 Sandybridge, Haswell, Bulldozer and Piledriver. These would typically cause
563 test failures or segfaults when compiled with recent versions of gcc from 8 onward.
564 * a similar bug was fixed in the blas_quickdivide code used to split workloads
566 * a bug in the IxMIN implementation for the GENERIC target made it return the result of IxMAX
567 * fixed building on SkylakeX systems when either the compiler or the (emulated) operating
568 environment does not support AVX512
569 * improved GEMM performance on ZEN targets
572 * build failures caused by the recently added checks for AVX512 were fixed
573 * an inline assembly bug involving undeclared modification of an input argument was
574 fixed in the blas_quickdivide code used to split workloads in most functions
575 * a bug in the IMIN implementation for the GENERIC target made it return the result of IMAX
578 * a bug in the IMIN implementation made it return the result of IMAX
581 * single precision BLAS1/2 functions have received optimized POWER8 kernels
582 * POWER9 is now a separate target, with an optimized DGEMM/DTRMM kernel
583 * building on PPC970 systems under OSX Leopard or Tiger is now supported
584 * out-of-bounds memory accesses in the gemm_beta microkernels were fixed
585 * building a shared library on AIX is now supported for POWER6
586 * DYNAMIC_ARCH support has been added for POWER6 and newer
589 * corrected xDOT behaviour with zero INC_X or INC_Y
590 * a bug in the IMIN implementation made it return the result of IMAX
593 * added support for HiSilicon TSV110 cpus
594 * the CMAKE build system now recognizes 32bit userspace on 64bit hardware
595 * cross-compilation with CMAKE now works again
596 * a bug in the IMIN implementation made it return the result of IMAX
597 * ARMV8 builds with the BINARY=32 option are now automatically handled as ARMV7
600 * optimized microkernels for single precicion BLAS1/2 functions have been added
603 ====================================================================
608 * loop unrolling in TRMV has been enabled again.
609 * A domain error in the thread workload distribution for SYRK
611 * gmake builds will now automatically add -fPIC to the build
612 options if the platform requires it.
613 * a pthreads key leakage (and associate crash on dlclose) in
614 the USE_TLS codepath was fixed.
615 * building of the utest cases on systems that do not provide
616 an implementation of complex.h was fixed.
619 * the SkylakeX code was changed to compile on OSX.
620 * unwanted application of the -march=skylake-avx512 option
621 to the common code parts of a DYNAMIC_ARCH build was fixed.
622 * improved performance of SGEMM for small workloads on Skylake X.
623 * performance of SGEMM and DGEMM was improved on Haswell.
626 * a configuration error that broke the CNRM2 kernel was corrected.
627 * compilation of the GEMM kernels with CMAKE was fixed.
628 * DYNAMIC_ARCH builds are now available with CMAKE as well.
629 * using CMAKE for cross-compilation to the new cpu TARGETs
630 introduced in 0.3.4 now works.
633 * a problem in cpu autodetection for AIX has been corrected.
635 ====================================================================
640 * the new, experimental thread-local memory allocation had
641 inadvertently been left enabled for gmake builds in 0.3.3
642 despite the announcement. It is now disabled by default, and
643 single-threaded builds will keep using the old allocator even
644 if the USE_TLS option is turned on.
645 * OpenBLAS will now provide enough buffer space for at least 50
647 * The output of openblas_get_config() now contains the version
649 * A serious thread safety bug in GEMV operation with small M and
650 large N size has been fixed.
651 * The code will now automatically call blas_thread_init after a
652 fork if needed before handling a call to openblas_set_num_threads
653 * Accesses to parallelized level3 functions from multiple callers
654 are now serialized to avoid thread races (unless using OpenMP).
655 This should provide better performance than the known-threadsafe
656 (but non-default) USE_SIMPLE_THREADED_LEVEL3 option.
657 * When building LAPACK with gfortran, -frecursive is now (again)
658 enabled by default to ensure correct behaviour.
659 * The OpenBLAS version cblas.h now supports both CBLAS_ORDER and
660 CBLAS_LAYOUT as the name of the matrix row/column order option.
661 * Externally set LDFLAGS are now passed through to the final compile/link
662 steps to facilitate setting platform-specific linker flags.
663 * A potential race condition during the build of LAPACK (that would
664 usually manifest itself as a failure to build TESTING/MATGEN) has been
666 * xHEMV has been changed to stay single-threaded for small input sizes
667 where the overhead of multithreading exceeds any possible gains
668 * CSWAP and ZSWAP have been limited to a single thread except on ARMV8 or
669 ThunderX hardware with sizable input.
670 * Linker flags for the PGI compiler have been updated
671 * Behaviour of AXPY with zero increments is now handled in the C interface,
672 correcting the result on at least Intel Atom.
673 * The result matrix from calling SGELSS with an all-zero input matrix is
674 now zeroed completely.
677 * Autodetection of AMD Ryzen2 has been fixed (again).
678 * CMAKE builds now support labeling of an INTERFACE64=1 build of
679 the library with the _64 suffix.
680 * AVX512 version of DGEMM has been added and the AVX512 SGEMM kernel
681 has been sped up by rewriting with C intrinsics
682 * Fixed compilation on RHEL5/CENTOS5 (issue with typename __WAIT_STATUS)
685 * added support for building on AIX (with gcc and GNU tools from AIX Toolbox).
686 * CPU type detection has been implemented for AIX.
687 * CPU type detection has been fixed for NETBSD.
690 * AXPY on LOONGSON3A has been corrected to pass "zero increment" utest.
691 * DSDOT on LOONGSON3A has been fixed.
692 * the SGEMM microkernel has been hardened against potential data loss.
695 * DYNAMic_ARCH support is now available for 64bit ARM
696 * cross-compiling for ARMV8 under iOS now works.
697 * cpu-specific code has been rearranged to make better use of both
698 hardware commonalities and model-specific compiler optimizations.
699 * XGENE1 has been removed as a TARGET, superseded by the improved generic
703 * Older assembly mnemonics have been converted to UAL form to allow
704 building with clang 7.0
705 * Cross compiling LAPACKE for Android has been fixed again (broken by
706 update to LAPACK 3.7.0 some while ago).
708 ====================================================================
713 * thread memory allocation has been switched back to the method
714 used before version 0.3.1 due to unexpected problems caused by
715 the new code under some circumstances. A new compile-time option
716 USE_TLS has been added to enable the new code, and it is hoped
717 that this can become the default again in the next version.
718 * LAPAck PR272 has been integrated, which fixes spurious errors
719 in DSYEVR and related functions caused by missing conversion
720 from ILAENV to ILAENV_2STAGE in several _2stage routines.
721 * the cmake-generated OpenBLASConfig.cmake now uses correct case
722 for the name of the library
723 * added support for Haiku OS
726 * added AVX512 implementations of SDOT, DDOT, SAXPY, DAXPY,
727 DSCAL, DGEMVN and DSYMVL
728 * added a workaround for a cygwin issue that prevented compilation
732 * added autodetection of Z14
733 * fixed TRMM errors in the generic target
735 ====================================================================
740 * fixes for regressions caused by the rewrite of the thread
741 initialization code in 0.3.1
744 * fixed cpu autodetection for the BSDs
747 * fixed utest errors in AXPY, DSDOT, ROT and SWAP
750 * added autodetection of AMD Ryzen 2
751 * fixed build with older versions of MSVC
753 ====================================================================
758 * rewritten thread initialization code with significantly reduced overhead
759 * added CBLAS interfaces to the IxAMIN BLAS extension functions
760 * fixed the lapack-test target
761 * CMAKE builds now create an OpenBLASConfig.cmake file
762 * ZAXPY now uses a single thread for small input sizes
763 * the LAPACK code was updated from Reference-LAPACK/lapack#253
764 (fixing LAPACKE interfaces to Aasen's functions)
767 * corrected CROT and ZROT behaviour with zero INC_X
770 * corrected xDOT behaviour with zero INC_X or INC_Y
773 * retired some older targets of DYNAMIC_ARCH builds to a new option DYNAMIC_OLDER,
774 this affects PENRYN,DUNNINGTON,OPTERON,OPTERON_SSE3,BOBCAT,ATOM and NANO
775 (which will still be supported via the slower PRESCOTT kernels when this option is not set)
776 * added an option DYNAMIC_LIST that (used in conjunction with DYNAMIC_ARCH) allows to
777 specify the list of x86_64 targets to include. Any target not on the list will be supported
778 by the Sandybridge or Nehalem kernels if available, or by Prescott.
779 * improved SWITCH_RATIO on Haswell for increased GEMM throughput
780 * added initial support for Intel Skylake X, including an AVX512 SGEMM kernel
781 * added autodetection of Intel Cannon Lake series as Skylake X
782 * added a default L2 cache size for hypervisors that return zero here (Chromebook)
783 * fixed a name clash with recent Windows10 headers that broke the build with (at least)
784 recent mingw from MSYS2
785 * fixed a link error in mixed clang/gfortran builds with OpenMP
786 * updated the OSX deployment target to 10.8
787 * switched on parallel make for builds on MS Windows by default
790 * fixed SSWAP and DSWAP behaviour with zero INC_X and INC_Y
792 ====================================================================
797 * fixed some more thread race and locking bugs
798 * added preliminary support for calling an OpenMP build of the library from multiple threads
799 * removed performance impact of thread locks added in 0.2.20 on OpenMP code
800 * general code cleanup
801 * optimized DSDOT implementation
802 * improved thread distribution for GEMM
803 * corrected IMATCOPY/OMATCOPY implementation
804 * fixed out-of-bounds accesses in the multithreaded xBMV/xPMV and SYMV implementations
805 * cmake build improvements
806 * pkgconfig file now contains build options
807 * openblas_get_config() now reports USE_OPENMP and NUM_THREADS settings used for the build
808 * corrections and improvements for systems with more than 64 cpus
809 * LAPACK code updated to 3.8.0 including later fixes
810 * added ReLAPACK, a recursive implementation of several LAPACK functions
811 * Rewrote ROTMG to handle cases that the netlib code failed to address
812 * Disabled (broken) multithreading code for xTRMV
813 * corrected prototypes of complex CBLAS functions to make our cblas.h match the generally accepted standard
814 * shared memory access failures on startup are now handled more gracefully
815 * restored utests from earlier releases (and made them pass on all affected systems)
818 * several fixes for cpu autodetection
821 * corrected vector register overwriting in several Power8 kernels
822 * optimized additional BLAS functions
825 * added support for CortexA53 and A72
826 * added autodetection for ThunderX2T99
827 * made most optimized kernels the default for generic ARMv8 targets
830 * parallelized DDOT kernel for Haswell
831 * changed alignment directives in assembly kernels to boost performance on OSX
832 * fixed register handling in the GEMV microkernels (bug exposed by gcc7)
833 * added support for building on OpenBSD and Dragonfly
834 * updated compiler options to work with Intel release 2018
835 * support fully optimized build with clang/flang on Microsoft Windows
836 * fixed building on AIX
839 * added optimized BLAS 1/2 functions
842 * fixed cpu autodetection helper code
843 * added mips32 1004K cpu (Mediatek MT7621 and similar SoC)
844 * added mips64 I6500 cpu
846 ====================================================================
851 * Improved CMake support
852 * Fixed several thread race and locking bugs
853 * Fixed default LAPACK optimization level
854 * Updated LAPACK to 3.7.0
855 * Added ReLAPACK (https://github.com/HPAC/ReLAPACK, make BUILD_RELAPACK=1)
858 * Optimizations for Power9
859 * Fixed several Power8 assembly bugs
862 * New optimized Vulcan and ThunderX2T99 targets
863 * Support for ARMV7 SOFT_FP ABI (make ARM_SOFTFP_ABI=1)
864 * Detect all cpu cores including offline ones
865 * Fix compilation with CLANG
866 * Support building a shared library for Android
869 * Fixed several threading issues
870 * Fix compilation with CLANG
873 * Detect Intel Bay Trail and Apollo Lake
874 * Detect Intel Sky Lake and Kaby Lake
875 * Detect Intel Knights Landing
876 * Detect AMD A8, A10, A12 and Ryzen
877 * Support 64bit builds with Visual Studio
878 * Fix building with Intel and PGI compilers
879 * Fix building with MINGW and TDM-GCC
880 * Fix cmake builds for Haswell and related cpus
881 * Fix building for Sandybridge with CLANG 3.9
882 * Add support for the FLANG compiler
885 * New target z13 with BLAS3 optimizations
887 ====================================================================
891 * Improved cross compiling.
892 * Fix the bug on musl libc.
895 * Optimize BLAS on Power8
896 * Fixed Julia+OpenBLAS bugs on Power8
899 * Optimize BLAS on MIPS P5600 and I6400 (Thanks, Shivraj Patil, Kaustubh Raste)
902 * Improved on ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
905 ====================================================================
909 * If you set MAKE_NB_JOBS flag less or equal than zero,
910 make will be without -j.
913 * Support building Visual Studio static library. (#813, Thanks, theoractice)
914 * Fix bugs to pass buidbot CI tests (http://build.openblas.net)
917 * Provide DGEMM 8x4 kernel for Cortex-A57 (Thanks, Ashwin Sekhar T K)
920 * Optimize S and C BLAS3 on Power8
921 * Optimize BLAS2/1 on Power8
923 ====================================================================
927 * Enable BUILD_LAPACK_DEPRECATED=1 by default.
929 ====================================================================
933 * Avoid potential getenv segfault. (#716)
934 * Import LAPACK svn bugfix #142-#147,#150-#155
937 * Optimize c/zgemv for AMD Bulldozer, Piledriver, Steamroller
938 * Fix bug with scipy linalg test.
941 * Improve DGEMM for ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
944 * Optimize D and Z BLAS3 functions for Power8.
946 ====================================================================
950 * Upgrade LAPACK to 3.6.0 version.
951 Add BUILD_LAPACK_DEPRECATED option in Makefile.rule to build
952 LAPACK deprecated functions.
953 * Add MAKE_NB_JOBS option in Makefile.
954 Force number of make jobs.This is particularly
955 useful when using distcc. (#735. Thanks, Jerome Robert.)
956 * Redesign unit test. Run unit/regression test at every build (Travis-CI and Appveyor).
957 * Disable multi-threading for small size swap and ger. (#744. Thanks, Jerome Robert)
958 * Improve small zger, zgemv, ztrmv using stack alloction (#727. Thanks, Jerome Robert)
959 * Let openblas_get_num_threads return the number of active threads.
960 (#760. Thanks, Jerome Robert)
961 * Support illumos(OmniOS). (#749. Thanks, Lauri Tirkkonen)
962 * Fix LAPACK Dormbr, Dormlq bug. (#711, #713. Thanks, Brendan Tracey)
963 * Update scipy benchmark script. (#745. Thanks, John Kirkham)
966 * Optimize trsm kernels for AMD Bulldozer, Piledriver, Steamroller.
967 * Detect Intel Avoton.
968 * Detect AMD Trinity, Richland, E2-3200.
969 * Fix gemv performance bug on Mac OSX Intel Haswell.
970 * Fix some bugs with CMake and Visual Studio
973 * Support and optimize Cortex-A57 AArch64.
974 (#686. Thanks, Ashwin Sekhar TK)
975 * Fix Android build on ARMV7 (#778. Thanks, Paul Mustiere)
976 * Update ARMV6 kernels.
979 * Fix detection of POWER architecture
980 (#684. Thanks, Sebastien Villemot)
982 ====================================================================
986 * Support cmake on x86/x86-64. Natively compiling on MS Visual Studio.
987 (experimental. Thank Hank Anderson for the initial cmake porting work.)
989 On Linux and Mac OSX, OpenBLAS cmake supports assembly kernels.
994 On Windows MS Visual Studio, OpenBLAS cmake only support C kernels.
995 (OpenBLAS uses AT&T style assembly, which is not supported by MSVC.)
996 e.g. cmake -G "Visual Studio 12 Win64" .
997 Open OpenBLAS.sln and build.
999 * Enable MAX_STACK_ALLOC flags by default.
1000 Improve ger and gemv for small matrices.
1001 * Improve gemv parallel with small m and large n case.
1002 * Improve ?imatcopy when lda==ldb (#633. Thanks, Martin Koehler)
1003 * Add vecLib benchmarks (#565. Thanks, Andreas Noack.)
1004 * Fix LAPACK lantr for row major matrices (#634. Thanks, Dan Kortschak)
1005 * Fix LAPACKE lansy (#640. Thanks, Dan Kortschak)
1006 * Import bug fixes for LAPACKE s/dormlq, c/zunmlq
1007 * Raise the signal when pthread_create fails (#668. Thanks, James K. Lowden)
1008 * Remove g77 from compiler list.
1009 * Enable AppVeyor Windows CI.
1012 * Support pure C generic kernels for x86/x86-64.
1013 * Support Intel Boardwell and Skylake by Haswell kernels.
1014 * Support AMD Excavator by Steamroller kernels.
1015 * Optimize s/d/c/zdot for Intel SandyBridge and Haswell.
1016 * Optimize s/d/c/zdot for AMD Piledriver and Steamroller.
1017 * Optimize s/d/c/zapxy for Intel SandyBridge and Haswell.
1018 * Optimize s/d/c/zapxy for AMD Piledriver and Steamroller.
1019 * Optimize d/c/zscal for Intel Haswell, dscal for Intel SandyBridge.
1020 * Optimize d/c/zscal for AMD Bulldozer, Piledriver and Steamroller.
1021 * Optimize s/dger for Intel SandyBridge.
1022 * Optimize s/dsymv for Intel SandyBridge.
1023 * Optimize ssymv for Intel Haswell.
1024 * Optimize dgemv for Intel Nehalem and Haswell.
1025 * Optimize dtrmm for Intel Haswell.
1028 * Support Android NDK armeabi-v7a-hard ABI (-mfloat-abi=hard)
1029 e.g. make HOSTCC=gcc CC=arm-linux-androideabi-gcc NO_LAPACK=1 TARGET=ARMV7
1030 * Fix lock, rpcc bugs (#616, #617. Thanks, Grazvydas Ignotas)
1032 * Support ppc64le platform (ELF ABI v2. #612. Thanks, Matthew Brandyberry.)
1033 * Support POWER7/8 by POWER6 kernels. (#612. Thanks, Fábio Perez.)
1035 ====================================================================
1039 * Improve OpenBLASConfig.cmake. (#474, #475. Thanks, xantares.)
1040 * Improve ger and gemv for small matrices by stack allocation.
1041 e.g. make -DMAX_STACK_ALLOC=2048 (#482. Thanks, Jerome Robert.)
1042 * Introduce openblas_get_num_threads and openblas_get_num_procs.
1043 (#497. Thanks, Erik Schnetter.)
1044 * Add ATLAS-style ?geadd function. (#509. Thanks, Martin Köhler.)
1045 * Fix c/zsyr bug with negative incx. (#492.)
1046 * Fix race condition during shutdown causing a crash in
1047 gotoblas_set_affinity(). (#508. Thanks, Ton van den Heuvel.)
1050 * Support AMD Streamroller.
1053 * Add Cortex-A9 and Cortex-A15 targets.
1055 ====================================================================
1059 * Add SYMBOLPREFIX and SYMBOLSUFFIX makefile options
1060 for adding a prefix or suffix to all exported symbol names
1061 in the shared library.(#459, Thanks Tony Kelman)
1062 * Provide OpenBLASConfig.cmake at installation.
1063 * Fix Fortran compiler detection on FreeBSD.
1064 (#470, Thanks Mike Nolta)
1068 * Add generic kernel files for x86-64. make TARGET=GENERIC
1069 * Fix a bug of sgemm kernel on Intel Sandy Bridge.
1070 * Fix c_check bug on some amd64 systems. (#471, Thanks Mike Nolta)
1073 * Support APM's X-Gene 1 AArch64 processors.
1074 Optimize trmm and sgemm. (#465, Thanks Dave Nuechterlein)
1076 ====================================================================
1080 * Added CBLAS interface for ?omatcopy and ?imatcopy.
1081 * Enable ?gemm3m functions.
1082 * Added benchmark for ?gemm3m.
1083 * Optimized multithreading lower limits.
1084 * Disabled SYMM3M and HEMM3M functions
1085 because of segment violations.
1088 * Improved axpy and symv performance on AMD Bulldozer.
1089 * Improved gemv performance on modern Intel and AMD CPUs.
1091 ====================================================================
1095 * Added some benchmark codes.
1096 * Fix link error on Linux/musl.(Thanks Isaac Dunham)
1099 * Improved s/c/zgemm performance for Intel Haswell.
1100 * Improved s/d/c/zgemv performance.
1101 * Support the big numa machine.(EXPERIMENT)
1104 * Fix detection when cpuinfo uses "Processor". (Thanks Isaiah)
1106 ====================================================================
1110 * Added BLAS extensions as following.
1111 s/d/c/zaxpby, s/d/c/zimatcopy, s/d/c/zomatcopy.
1112 * Added OPENBLAS_CORETYPE environment for dynamic_arch. (a86d34)
1113 * Added NO_AVX2 flag for old binutils. (#401)
1114 * Support outputing the CPU corename on runtime.(#407)
1115 * Patched LAPACK to fix bug 114, 117, 118.
1116 (http://www.netlib.org/lapack/bug_list.html)
1117 * Disabled ?gemm3m for a work-around fix. (#400)
1119 * Fixed lots of bugs for optimized kernels on sandybridge,Haswell,
1120 bulldozer, and piledriver.
1121 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
1124 * Improved LAPACK testing.
1126 ====================================================================
1130 * Improved the result for LAPACK testing. (#372)
1131 * Installed DLL to prefix/bin instead of prefix/lib. (#366)
1132 * Build import library on Windows.(#374)
1134 * To improve LAPACK testing, we fallback some kernels. (#372)
1135 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
1137 ====================================================================
1141 * Added OPENBLAS_VERBOSE environment variable.(#338)
1142 * Make OpenBLAS thread-pool resilient to fork via pthread_atfork.
1143 (#294, Thank Olivier Grisel)
1147 * Detect Intel Haswell for new Macbook.
1149 ====================================================================
1153 * Update LAPACK to 3.5.0 version
1154 * Fixed compatiable issues with Clang and Pathscale compilers.
1157 * Optimization on Intel Haswell.
1158 * Enable optimization kernels on AMD Bulldozer and Piledriver.
1161 * Support ARMv6 and ARMv7 ISA.
1162 * Optimization on ARM Cortex-A9.
1164 ====================================================================
1168 * Support Open64 5.0. (#266)
1169 * Add executable stack markings. (#262, Thank Sébastien Fabbro)
1170 * Respect user's LDFLAGS (Thank Sébastien Fabbro)
1173 * Rollback bulldozer and piledriver kernels to barcelona kernels (#263)
1174 We will fix the compuational error bug in bulldozer and piledriver kernels.
1176 ====================================================================
1180 * Support LSB (Linux Standard Base) 4.1.
1182 * Include LAPACK 3.4.2 source codes to the repo.
1183 Avoid downloading at compile time.
1184 * Add NO_PARALLEL_MAKE flag to disable parallel make.
1185 * Create openblas_get_parallel to retrieve information which
1186 parallelization model is used by OpenBLAS. (Thank grisuthedragon)
1187 * Detect LLVM/Clang compiler. The default compiler is Clang on Mac OS X.
1188 * Change LIBSUFFIX from .lib to .a on windows.
1189 * A work-around for dtrti_U single thread bug. Replace it with LAPACK codes. (#191)
1192 * Optimize c/zgemm, trsm, dgemv_n, ddot, daxpy, dcopy on
1193 AMD Bulldozer. (Thank Werner Saar)
1194 * Add Intel Haswell support (using Sandybridge optimizations).
1196 * Add AMD Piledriver support (using Bulldozer optimizations).
1197 * Fix the computational error in zgemm avx kernel on
1199 * Fix the overflow bug in gemv.
1200 * Fix the overflow bug in multi-threaded BLAS3, getrf when NUM_THREADS
1201 is very large.(#214, #221, #246).
1203 * Support loongcc (Open64 based) compiler for ICT Loongson 3A/B.
1206 * Support Power7 by old Power6 kernels. (#220)
1208 ====================================================================
1212 * Improved OpenMP performance slightly. (d744c9)
1213 * Improved cblas.h compatibility with Intel MKL.(#185)
1214 * Fixed the overflowing bug in single thread cholesky factorization.
1215 * Fixed the overflowing buffer bug of multithreading hbmv and sbmv.(#174)
1218 * Added AMD Bulldozer x86-64 S/DGEMM AVX kernels. (Thank Werner Saar)
1219 We will tune the performance in future.
1220 * Auto-detect Intel Xeon E7540.
1221 * Fixed the overflowing buffer bug of gemv. (#173)
1222 * Fixed the bug of s/cdot about invalid reading NAN on x86_64. (#189)
1226 ====================================================================
1230 * Added NO_SHARED flag to disable generating the shared library.
1231 * Compile LAPACKE with ILP64 modle when INTERFACE64=1 (#158)
1232 * Export LAPACK 3.4.2 symbols in shared library. (#147)
1233 * Only detect the number of physical CPU cores on Mac OSX. (#157)
1234 * Fixed NetBSD build. (#155)
1235 * Fixed compilation with TARGET=GENERIC. (#160)
1237 * Restore the original CPU affinity when calling
1238 openblas_set_num_threads(1) (#153)
1239 * Fixed a SEGFAULT bug in dgemv_t when m is very large.(#154)
1242 ====================================================================
1246 * Upgraded LAPACK to 3.4.2 version. (#145)
1247 * Provided support for passing CFLAGS, FFLAGS, PFLAGS,
1248 FPFLAGS to make. (#137)
1249 * f77blas.h:compatibility for compilers without C99 complex
1250 number support. (#141)
1252 * Added NO_AVX flag. Check OS supporting AVX on runtime. (#139)
1253 * Fixed zdot incompatibility ABI issue with GCC 4.7 on
1254 Windows 32-bit. (#140)
1256 * Fixed the generation of shared library bug.
1257 * Fixed the detection bug on the Loongson 3A server.
1258 ====================================================================
1262 * Fixed LAPACK unstable bug about ?laswp. (#130)
1263 * Fixed the shared library bug about unloading the library on
1265 * Fixed the compilation failure on BlueGene/P (TARGET=PPC440FP2)
1266 Please use gcc and IBM xlf. (#134)
1268 * Supported goto_set_num_threads and openblas_set_num_threads
1269 APIs in Windows. They can set the number of threads on runtime.
1271 ====================================================================
1275 * Fixed exporting DLL functions bug on Windows/MingW
1276 * Support GNU Hurd (Thank Sylvestre Ledru)
1277 * Support kfreebsd kernel (Thank Sylvestre Ledru)
1279 * Support Intel Sandy Bridge 22nm desktop/mobile CPU
1281 * Improve the detection of SPARC (Thank Sylvestre Ledru)
1283 ====================================================================
1288 * Fixed the SEGFAULT bug about hyper-theading
1289 * Support AMD Bulldozer by using GotoBLAS2 AMD Barcelona codes
1291 ====================================================================
1295 * Removed the limitation (64) of numbers of CPU cores.
1296 Now, it supports 256 cores at max.
1297 * Supported clang compiler.
1298 * Fixed some build bugs on FreeBSD
1300 * Optimized Level-3 BLAS on Intel Sandy Bridge x86-64 by AVX instructions.
1301 Please use gcc >= 4.6 or clang >=3.1.
1302 * Support AMD Bobcat by using GotoBLAS2 AMD Barcelona codes.
1304 ====================================================================
1308 * Upgraded LAPACK to 3.4.1 version. (Thank Zaheer Chothia)
1309 * Supported LAPACKE, a C interface to LAPACKE. (Thank Zaheer Chothia)
1310 * Fixed the build bug (MD5 and download) on Mac OSX.
1311 * Auto download CUnit 2.1.2-2 from SF.net with UTEST_CHECK=1.
1312 * Fxied the compatibility issue for compilers without C99 complex number
1313 (e.g. Visual Studio)
1315 * Auto-detect Intel Sandy Bridge Core i7-3xxx & Xeon E7 Westmere-EX.
1316 * Test alpha=Nan in dscale.
1317 * Fixed a SEGFAULT bug in samax on x86 windows.
1319 ====================================================================
1323 * Set soname of shared library on Linux.
1324 * Added LIBNAMESUFFIX flag in Makefile.rule. The user can use
1325 this flag to control the library name, e.g. libopenblas.a,
1326 libopenblas_ifort.a or libopenblas_omp.a.
1327 * Added GEMM_MULTITHREAD_THRESHOLD flag in Makefile.rule.
1328 The lib use single thread in GEMM function with small matrices.
1330 * Used GEMV SSE/SSE2 kernels on x86 32-bit.
1331 * Exported CBLAS functions in Windows DLL.
1333 * Completed Level-3 BLAS optimization on Loongson 3A CPU.
1334 * Improved GEMV performance on Loongson 3A CPU.
1335 * Improved Level-3 BLAS performance on Loongson 3B CPU. (EXPERIMENT)
1337 ====================================================================
1338 Version 0.1 alpha2.5
1341 * Fixed missing "#include <sched.h>" bug on Mac OS X.
1342 Thank Mike Nolta for the patch.
1343 * Upgraded LAPACK to 3.4.0 version
1344 * Fixed a bug on Mac OS X. Don't require SystemStubs on OS X.
1345 SystemStubs does not exist on Lion. Thank Stefan Karpinski.
1346 * Improved README with using OpenMP. Check the internal threads
1347 count less than or equal to omp_get_max_threads()
1349 * Auto-detect Intel Core i6/i7 (Sandy Bridge) CPU with Nehalem assembly kernels
1350 * Fixed some bugs on MingW 64-bit including zgemv, cdot, zdot.
1352 ====================================================================
1353 Version 0.1 alpha2.4
1356 * Fixed a bug about installation. The header file "fblas77.h"
1358 * Fixed #61 a building bug about setting TARGET and DYNAMIC_ARCH.
1359 * Try to handle absolute path of shared library in OSX. (#57)
1360 Thank Dr Kane O'Donnell.
1361 * Changed the installation folder layout to $(PREFIX)/include and
1365 * Fixed #58 zdot/xdot SEGFAULT bug with GCC-4.6 on x86. According
1366 to i386 calling convention, The callee should remove the first
1367 hidden parameter.Thank Mr. John for this patch.
1369 ====================================================================
1370 Version 0.1 alpha2.3
1374 * Added DTB_ENTRIES into dynamic arch setting parameters. Now,
1375 it can read DTB_ENTRIES on runtime. (Refs issue #55 on github)
1377 ====================================================================
1378 Version 0.1 alpha2.2
1382 * Fixed a building bug when DYNAMIC_ARCH=1 & INTERFACE64=1.
1383 (Refs issue #44 on github)
1385 ====================================================================
1386 Version 0.1 alpha2.1
1390 * Stop the build and output the error message when detecting
1391 fortran compiler failed. (Refs issue #42 on github)
1393 ====================================================================
1398 * Fixed blasint undefined bug in <cblas.h> file. Other software
1399 could include this header successfully(Refs issue #13 on github)
1400 * Fixed the SEGFAULT bug on 64 cores. On SMP server, the number
1401 of CPUs or cores should be less than or equal to 64.(Refs issue #14
1403 * Support "void goto_set_num_threads(int num_threads)" and "void
1404 openblas_set_num_threads(int num_threads)" when USE_OPENMP=1
1405 * Added extern "C" to support C++. Thank Tasio for the patch(Refs
1406 issue #21 on github)
1407 * Provided an error message when the arch is not supported.(Refs
1408 issue #19 on github)
1409 * Fixed issue #23. Fixed a bug of f_check script about generating link flags.
1410 * Added openblas_set_num_threads for Fortran.
1411 * Fixed #25 a wrong result of rotmg.
1412 * Fixed a bug about detecting underscore prefix in c_check.
1413 * Print the wall time (cycles) with enabling FUNCTION_PROFILE
1414 * Fixed #35 a build bug with NO_LAPACK=1 & DYNAMIC_ARCH=1
1415 * Added install target. You can use "make install". (Refs #20)
1419 * Fixed #28 a wrong result of dsdot on x86_64.
1420 * Fixed #32 a SEGFAULT bug of zdotc with gcc-4.6.
1421 * Fixed #33 ztrmm bug on Nehalem.
1422 * Work-around #27 the low performance axpy issue with small imput size & multithreads.
1425 * Fixed #28 a wrong result of dsdot on Loongson3A/MIPS64.
1426 * Optimized single/double precision BLAS Level3 on Loongson3A/MIPS64. (Refs #2)
1427 * Optimized single/double precision axpy function on Loongson3A/MIPS64. (Refs #3)
1429 ====================================================================
1434 * Support "make NO_LAPACK=1" to build the library without
1436 * Fixed randomly SEGFAULT when nodemask==NULL with above Linux 2.6.34.
1437 Thank Mr.Ei-ji Nakama providing this patch. (Refs issue #12 on github)
1438 * Added DEBUG=1 rule in Makefile.rule to build debug version.
1439 * Disable compiling quad precision in reference BLAS library(netlib BLAS).
1440 * Added unit testcases in utest/ subdir. Used CUnit framework.
1441 * Supported OPENBLAS_* & GOTO_* environment variables (Pleas see README)
1442 * Imported GotoBLAS2 1.13 BSD version
1445 * On x86 32bits, fixed a bug in zdot_sse2.S line 191. This would casue
1446 zdotu & zdotc failures. Instead, work-around it. (Refs issue #8 #9 on github)
1447 * Modified ?axpy functions to return same netlib BLAS results
1448 when incx==0 or incy==0 (Refs issue #7 on github)
1449 * Modified ?swap functions to return same netlib BLAS results
1450 when incx==0 or incy==0 (Refs issue #6 on github)
1451 * Modified ?rot functions to return same netlib BLAS results
1452 when incx==0 or incy==0 (Refs issue #4 on github)
1453 * Detect Intel Westmere,Intel Clarkdale and Intel Arrandale
1454 to use Nehalem codes.
1455 * Fixed a typo bug about compiling dynamic ARCH library.
1457 * Improve daxpy performance on ICT Loongson 3A.
1458 * Supported ICT Loongson 3A CPU (Refs issue #1 on github)
1459 ====================================================================