2 ====================================================================
7 - some code cleanup, with added casts etc.
8 - fixed obtaining the cpu count with OpenMP and OMP_PROC_BIND unset
9 - fixed pivot index calculation by ?LASWP for negative increments other than one
10 - fixed input argument check in LAPACK ? GEQRT2
11 - improved the check for a Fortran compiler in CMAKE builds
12 - disabled building OpenBLAS' optimized versions of LAPACK complex SPMV,SPR,SYMV,SYR with NO_LAPACK=1
13 - fixed building of LAPACK on certain distributed filesystems with parallel gmake
14 - fixed building the shared library on MacOS with classic flang
17 - fixed cross-compilation with CMAKE for CORE2 target
18 - fixed miscompilation of AVX512 code in DYNAMIC_ARCH builds
19 - added support for the "incidental" AVX512 hardware in Alder Lake when enabled in BIOS
22 - add new architecture (Russian Elbrus E2000 family)
28 - added SVE-enabled CGEMM and ZGEMM kernels for ARMV8SVE and A64FX
29 - added support for Neoverse N2 and V1 cpus
32 - fixed autodetection of MSA capability
35 - added an optimized DGEMM kernel
37 ====================================================================
42 - reverted unsafe TRSV/ZRSV optimizations introduced in 0.3.16
43 - fixed a potential thread race in the thread buffer reallocation routines
44 that were introduced in 0.3.18
45 - fixed miscounting of thread pool size on Linux with OMP_PROC_BIND=TRUE
46 - fixed CBLAS interfaces for CSROT/ZSROT and CROTG/ZROTG
47 - made automatic library suffix for CMAKE builds with INTERFACE64 available
51 - DYNAMIC_ARCH builds now fall back to the cpu with most similar capabilities
52 when an unknown CPUID is encountered, instead of defaulting to Prescott
53 - added cpu detection for Intel Alder Lake
54 - added cpu detection for Intel Sapphire Rapids
55 - added an optimized SBGEMM kernel for Sapphire Rapids
56 - fixed DYNAMIC_ARCH builds on OSX with CMAKE
57 - worked around DYNAMIC_ARCH builds made on Sandybridge failing on SkylakeX
58 - fixed missing thread initialization for static builds on Windows/MSVC
59 - fixed an excessive read in ZSYMV
62 - added support for POWER10 in big-endian mode
63 - added support for building with CMAKE
64 - added optimized SGEMM and DGEMM kernels for small matrix sizes
67 - added basic support and cputype detection for Fujitsu A64FX
68 - added a generic ARMV8SVE target
69 - added SVE-enabled SGEMM and DGEMM kernels for ARMV8SVE and A64FX
70 - added optimized CGEMM and ZGEMM kernels for Cortex A53 and A55 cpus
71 - fixed cpuid detection for Apple M1 and improved performance
72 - improved compiler flag setting in CMAKE builds
75 - fixed improper initialization in CSCAL/ZSCAL for strided access patterns
78 - added a GENERIC target for MIPS32
79 - added support for cross-compiling to MIPS32 on x86_64 using CMAKE
82 - fixed misdetection of MSA capability
84 ====================================================================
89 - when the build-time number of preconfigured threads is exceeded
90 at runtime (typically by an external program calling BLAS functions
91 from a larger number of threads in parallel), OpenBLAS will now
92 allocate an auxiliary control structure for up to 512 additional
93 threads instead of aborting
94 - added support for Loongson's LoongArch64 cpu architecture
95 - fixed building OpenBLAS with CMAKE and -DBUILD_BFLOAT16=ON
96 - added support for building OpenBLAS as a CMAKE subproject
97 - added support for building for Windows/ARM64 targets with clang
98 - improved support for building with the IBM xlf compiler
99 - imported Reference-LAPACK PR 625 (out-of-bounds reads in ?LARRV)
100 - imported Reference-LAPACK PR 597 for testsuite compatibility with
104 - added SkylakeX S/DGEMM kernels for small problem sizes (M*N*K<=1000000)
105 - added optimized SBGEMM for Intel Cooper Lake
106 - reinstated the performance patch for AVX512 SGEMV_T with a proper fix
107 - added a workaround for a gcc11 tree-vectorizer bug that caused spurious
108 failures in the test programs for complex BLAS3 when compiling at -O3
109 (the default for cmake "release" builds)
110 - added support for runtime cpu count detection under Haiku OS
111 - worked around a long-standing miscompilation issue of the Haswell DGEMV_T
112 kernel with gcc that could produce NaN output in some corner cases
115 - improved performance of DASUM on POWER10
118 - fixed crashes (use of reserved register x18) on Apple M1 under OSX
119 - fixed building with gcc releases earlier than 5.1
122 - fixed building under BSD
125 - fixed building under BSD
127 ====================================================================
132 - reverted the optimization of SGEMV_N/DGEMV_N for small input sizes
133 and consecutive arguments as it led to stack overflows on x86_64
134 with some operating systems (notably OSX and Windows)
137 - reverted the performance patch for SGEMV_T on AVX512 as it caused
138 wrong results in some applications
141 - fixed compilation with compilers other than gcc
142 ====================================================================
147 - drastically reduced the stack size requirements for running the LAPACK
148 testsuite (Reference-LAPACK PR 553)
149 - fixed spurious test failures in the LAPACK testsuite (Reference-LAPACK
151 - expressly setting DYNAMIC_ARCH=0 no longer enables dynamic_arch mode
152 - improved performance of xGER, xSPR, xSPR2, xSYR, xSYR2, xTRSV, SGEMV_N
153 and DGEMV_N, for small input sizes and consecutive arguments
154 - improved performance of xGETRF, xPORTF and xPOTRI for small input sizes
155 by disabling multithreading
156 - fixed installing with BSD versions of the "install" utility
159 - fixed the implementation of xIMIN
160 - improved the performance of DSDOT
161 - fixed linking of the tests on C910V with current vendor gcc
164 - fixed SBGEMM computation for some odd value inputs
165 - fixed compilation for PPCG4, PPC970, POWER3, POWER4 and POWER5
168 - improved performance of SGEMV_N and SGEMV_T for small N on AVX512-capable cpus
169 - worked around a miscompilation of ZGEMM/ZTRMM on Sandybridge with old gcc
171 - fixed compilation with MS Visual Studio versions older than 2017
172 - fixed macro name collision with winnt.h from the latest Win10 SDK
173 - added cpu type autodetection for Intel Ice Lake SP
174 - fixed cpu type autodetection for Intel Tiger Lake
175 - added cpu type autodetection for recent Centaur/Zhaoxin models
176 - fixed compilation with musl libc
179 - fixed compilation with gcc/gfortran on the Apple M1
180 - fixed linking of the tests on FreeBSD
181 - fixed missing restore of a register in the recently rewritten DNRM2 kernel
182 for ThunderX2 and Neoverse N1 that could cause spurious failures in e.g.
184 - added compiler optimization flags for the EMAG8180
185 - added initial support for Cortex A55
188 - fixed linking of the tests on FreeBSD
190 ====================================================================
195 - imported improvements and bugfixes from Reference-LAPACK 3.9.1
196 - imported LAPACKE interface fixes from Reference-LAPACK PRs 534 + 537
197 - fixed a problem in the cpu detection of 0.3.14 that prevented cross-compilation
198 - fixed a sequence problem in the generation of softlinks to the library in GMAKE
201 - fixed compilation on RISCV (missing entry in getarch)
202 - fixed a potential division by zero in CROTG and ZROTG
205 - fixed LAPACK testsuite failures seen with the NVIDIA HPC compiler
206 - improved CGEMM, DGEMM and ZGEMM performance on POWER10
207 - added an optimized ZGEMV kernel for POWER10
208 - fixed a potential division by zero in CROTG and ZROTG
211 - added support for Intel Control-flow Enforcement Technology (CET)
212 - reverted the DOMATCOPY_RT code to the generic C version
213 - fixed a bug in the AVX512 SGEMM kernel introduced in 0.3.14
214 - fixed misapplication of -msse flag to non-SSE cpus in DYNAMIC_ARCH
215 - added support for compilation of the benchmarks on older OSX versions
216 - fix propagation of the NO_AVX512 option in CMAKE builds
217 - fix compilation of the AVX512 SGEMM kernel with clang-cl on Windows
218 - fixed compilation of the CTESTs with INTERFACE64=1 (random faults on OSX)
219 - corrected the Haswell DROT kernel to require AVX2/FMA3 rather than AVX512
222 - fixed a potential division by zero in CROTG and ZROTG
223 - fixed a potential overflow in IMATCOPY/ZIMATCOPY and the CTESTs
226 - fixed spurious reads outside the array in the SGEMM tcopy macro
227 - fixed a potential division by zero in CROTG and ZROTG
228 - fixed a segmentation fault in DYNAMIC_ARCH builds (reappeared in 0.3.14)
231 - fixed a potential division by zero in CROTG and ZROTG
232 - fixed a potential overflow in IMATCOPY/ZIMATCOPY and the CTESTs
235 - fixed a potential division by zero in CROTG and ZROTG
238 - fixed a potential division by zero in CROTG and ZROTG
240 ====================================================================
245 * Fixed a race condition on thread shutdown in non-OpenMP builds
246 * Fixed custom BUFFERSIZE option getting ignored in gmake builds
247 * Fixed CMAKE compilation of the TRMM kernels for GENERIC platforms
248 * Added CBLAS interfaces for CROTG, ZROTG, CSROT and ZDROT
249 * Improved performance of OMATCOPY_RT across all platforms
250 * Changed perl scripts to use env instead of a hardcoded /usr/bin/perl
251 * Fixed potential misreading of the GCC compiler version in the build scripts
252 * Fixed convergence problems in LAPACK complex GGEV/GGES (Reference-LAPACK #477)
253 * Reduced the stacksize requirements for running the LAPACK testsuite (Reference-LAPACK #335)
256 * Fixed compilation on RISCV (missing entry in getarch)
259 * Fixed compilation for DYNAMIC_ARCH with clang and with old gcc versions
260 * Added support for compilation on FreeBSD/ppc64le
261 * Added optimized POWER10 kernels for SSCAL, DSCAL, CSCAL, ZSCAL
262 * Added optimized POWER10 kernels for SROT, DROT, CDOT, SASUM, DASUM
263 * Improved SSWAP, DSWAP, CSWAP, ZSWAP performance on POWER10
264 * Improved SCOPY and CCOPY performance on POWER10
265 * Improved SGEMM and DGEMM performance on POWER10
266 * Added support for compilation with the NVIDIA HPC compiler
269 * Added an optimized bfloat16 GEMM kernel for Cooperlake
270 * Added CPUID autodetection for Intel Rocket Lake and Tiger Lake cpus
271 * Improved the performance of SASUM,DASUM,SROT,DROT on AMD Ryzen cpus
272 * Added support for compilation with the NAG Fortran compiler
273 * Fixed recognition of the AMD AOCC compiler
274 * Fixed compilation for DYNAMIC_ARCH with clang on Windows
275 * Added support for running the BLAS/CBLAS tests on Windows
276 * Fixed signatures of the tls callback functions for Windows x64
277 * Fixed various issues with fma intrinsics support handling
280 * Added support for embedded Cortex M targets via a new option EMBEDDED
283 * Fixed the THUNDERX2T99 and NEOVERSEN1 DNRM2/ZNRM2 kernels for inputs with Inf
284 * Added support for the DYNAMIC_LIST option
285 * Added support for compilation with the NVIDIA HPC compiler
286 * Added support for compiling with the NAG Fortran compiler
288 ====================================================================
293 * Added a generic bfloat16 SBGEMV kernel
294 * Fixed a potentially severe memory leak after fork in OpenMP builds
295 that was introduced in 0.3.12
296 * Added detection of the Fujitsu Fortran compiler
297 * Added detection of the (e)gfortran compiler on OpenBSD
298 * Added support for overriding the default name of the library independently
299 from symbol suffixing in the gmake builds (already supported in cmake)
302 * Added a RISC V port optimized for C910V
305 * Added optimized POWER10 kernels for SAXPY, CAXPY, SDOT, DDOT and DGEMV_N
306 * Improved DGEMM performance on POWER10
307 * Improved STRSM and DTRSM performance on POWER9 and POWER10
308 * Fixed segmemtation faults in DYNAMIC_ARCH builds
309 * Fixed compilation with the PGI compiler
312 * Fixed compilation of kernels that require SSE2 intrinsics since 0.3.12
315 * Added an optimized bfloat16 SBGEMV kernel for SkylakeX and Cooperlake
316 * Improved the performance of SASUM and DASUM kernels through parallelization
317 * Improved the performance of SROT and DROT kernels
318 * Improved the performance of multithreaded xSYRK
319 * Fixed OpenMP builds that use the LLVM Clang compiler together with GNU gfortran
320 (where linking of both the LLVM libomp and GNU libgomp could lead to lockups or
322 * Fixed miscompilations by old gcc 4.6
323 * Fixed misdetection of AVX2 capability in some Sandybridge cpus
324 * Fixed lockups in builds combining DYNAMIC_ARCH with TARGET=GENERIC on OpenBSD
327 * Fixed segmemtation faults in DYNAMIC_ARCH builds
330 * Improved kernels for Loongson 3R3 ("3A") and 3R4 ("3B") models, including MSA
331 * Fixed bugs in the MSA kernels for CGEMM, CTRMM, CGEMV and ZGEMV
332 * Added handling of zero increments in the MSA kernels for SSWAP and DSWAP
333 * Added DYNAMIC_ARCH support for MIPS64 (currently Loongson3R3/3R4 only)
336 * Fixed building 32 and 64 bit SPARC kernels with the SolarisStudio compilers
338 ====================================================================
343 * Fixed missing BLAS/LAPACK functions (inadvertently dropped during
344 the build system restructuring)
345 * Fixed argument conversion macro in LAPACKE_zgesvdq (LAPACK #458)
348 * Added optimized SCOPY/CCOPY kernels for POWER10
349 * Increased and unified the default size of the GEMM BUFFER
350 * Fixed building for POWER10 in DYNAMIC_ARCH mode
351 * POWER10 compatibility test now checks binutils version as well
352 * Cleaned up compiler warnings
355 * corrected compiler version checks for AVX2 compatibility
356 * added compiler option -mavx2 for building with flang
357 * fixed direct SGEMM pathway for small matrix sizes (broken by
358 the code refactoring in 0.3.11)
359 * fixed unhandled partial register clobbers in several kernels
360 for AXPY,DOT,GEMV_N and GEMV_T flagged by gcc10 tree-vectorizer
363 * improved Apple Vortex support to include cross-compiling
365 ====================================================================
371 the newly added BFLOAT16 functions were renamed to use the
372 letter "B" instead of "H" to avoid potential confusion with
373 the IEEE "half precision float" type, i.e. the 0.3.10
374 SHGEMM is now SBGEMM and the corresponding build option
375 was changed from "BUILD_HALF" to "BUILD_BFLOAT16".
376 * Reduced the default BLAS3_MEM_ALLOC_THRESHOLD (used as an upper
377 limit for placing temporary arrays on the stack) to be compatible
378 with a stack size of 1mb (as imposed by the JAVA runtime library)
379 * Added mixed-precision dot function SBDOT and utility functions
380 shstobf16, shdtobf16, sbf16tos and dbf16tod to convert between
381 single or double precision float arrays and bfloat16 arrays
382 * Fixed prototypes of LAPACK_?ggsvp and LAPACK_?ggsvd functions
384 * Fixed underflow and rounding errors in LAPACK SLANV2 and DLANV2
385 (causing miscalculations in e.g. SHSEQR/DHSEQR, LAPACK issue #263)
386 * Fixed workspace calculation in LAPACK ?GELQ (LAPACK issue #415)
387 * Fixed several bugs in the LAPACK testsuite
388 * Improved performance of TRMM and TRSM for certain problem sizes
389 * Fixed infinite recursions and workspace miscalculations in ReLAPACK
390 * CMAKE builds no longer require pkg-config for creating the .pc file
391 * Makefile builds no longer misread NO_CBLAS=0 or NO_LAPACK=0 as
392 enabling these options
393 * Fixed detection of gfortran when invoked through an mpi wrapper
394 * Improve thread reinitialization performance with OpenMP after a fork
395 * Added support for building only the subset of the library required
396 for a particular precision by specifying BUILD_SINGLE, BUILD_DOUBLE
397 * Optional function name prefixes and suffixes are now correctly
398 reflected in the generated cblas.h
399 * Added CMAKE build support for the LAPACK and multithreading tests
402 * Added optimized support for POWER10
403 * Added support for compiling for POWER8 in 32bit mode
404 * Added support for compilation with LLVM/clang
405 * Added support for compilation with NVIDIA/PGI compilers
406 * Fixed building on big-endian POWER8
407 * Fixed miscompilation of ZDOTC by gcc10
408 * Fixed alignment errors in the POWER8 SAXPY kernel
409 * Improved CPU detection on AIX
410 * Supported building with older compilers on POWER9
413 * Added support for Intel Cooperlake
414 * Added autodetection of AMD Renoir/Matisse/Zen3 cpus
415 * Added autodetection of Intel Comet Lake cpus
416 * Reimplemented ?sum, ?dot and daxpy using universal intrinsics
417 * Reset the fpu state before using the fpu on Windows as a workaround
418 for a problem introduced in Windows 10 build 19041 (a.k.a. SDK 2004)
419 * Fixed potentially undefined behaviour in the dot and gemv_t kernels
420 * Fixed a potential segmentation fault in DYNAMIC_ARCH builds
421 * Fixed building for ZEN with PGI/NVIDIA and AMD AOCC compilers
424 * Fixed cpu detection on BSD-like systems
427 * Added preliminary support for Apple Vortex cpus
428 * Added support for the Cavium ThunderX3T110 cpu
429 * Fixed cpu detection on BSD-like systems
430 * Fixed compilation in -std=C18 mode
433 * Added support for compiling with the clang compiler
434 * Improved GEMM performance on Z14
436 ====================================================================
441 * Improved thread locking behaviour in blas_server and parallel getrf
442 * Imported bugfix 394 from LAPACK (spurious reference to "XERBL"
443 due to overlong lines)
444 * Imported bugfix 403 from LAPACK (compile option "recursive" required
445 for correctness with Intel and PGI)
446 * Imported bugfix 408 from LAPACK (wrong scaling in ZHEEQUB)
447 * Imported bugfix 411 from LAPACK (infinite loop in LARGV/LARTG/LARTGP)
448 * Fixed mismatches between BUFFERSIZE and GEMM_UNROLL parameters that
449 could lead to crashes at large matrix sizes
450 * Restored internal soname in dynamic libraries on FreeBSD and Dragonfly
451 * Added API (openblas_setaffinity) to set the thread affinity on Linux
452 * Added initial infrastructure for half-precision floating point
453 (bfloat16) support with a generic implementation of SHGEMM
454 * Added CMAKE build system support for building the cblas_Xgemm3m
456 * Fixed CMAKE support for building in a path with embedded spaces
457 * Fixed CMAKE (non)handling of NO_EXPRECISION and MAX_STACK_ALLOC
458 * Fixed GCC version detection in the Makefiles
459 * Allowed overriding the names of AR, AS and LD in Makefile builds
462 * Fixed big-endian POWER8 ELFv2 builds on FreeBSD
463 * Fixed GCC version checks and DYNAMIC_ARCH builds on POWER9
464 * Fixed CMAKE build support for POWER9
465 * fixed a potential race condition in the thread buffer allocation
466 * Worked around LAPACK test failures on PPC G4
469 * Fixed a potential race condition in the thread buffer allocation
470 * Added support for MIPS 24K/24KE family based on P5600 kernels
473 * fixed a potential race condition in the thread buffer allocation
474 * Added TARGET=GENERIC
477 * Fixed a race condition in the thread buffer allocation
480 * Fixed a race condition in the thread buffer allocation
481 * Fixed zero initialisation in the assembly for SGEMM and DGEMM BETA
482 * Improved performance of the ThunderX2 DAXPY kernel
483 * Added an optimized SGEMM kernel for Cortex A53
484 * Fixed Makefile support for INTERFACE64 (8-byte integer)
487 * Fixed a syntax error in the CMAKE setup for SkylakeX
488 * Improved performance of STRSM on Haswell, SkylakeX and Ryzen
489 * Improved SGEMM performance on SGEMM for workloads with ldc a
491 * Improved DGEMM performance on Skylake X
492 * Fixed unwanted AVX512-dependency of SGEMM in DYNAMIC_ARCH
493 builds created on SkylakeX
494 * Removed data alignment requirement in the SSE2 copy kernels
495 that could cause spurious crashes
496 * Added a workaround for an optimizer bug in AppleClang 11.0.3
497 * Fixed LAPACK test failures due to wrong options for Intel Fortran
498 * Fixed compilation and LAPACK test results with recent Flang
500 * Fixed DYNAMIC_ARCH builds with CMAKE on OS X
501 * Fixed missing exports of cblas_i?amin, cblas_i?min, cblas_i?max,
502 cblas_?sum, cblas_?gemm3m in the shared library on OS
503 * Fixed reporting of cpu name in DYNAMIC_ARCH builds (would sometimes
504 show the name of an older generation chip supported by the same kernels)
507 * Improved performance of SGEMM/STRMM and DGEMM/DTRMM on Z14
509 ====================================================================
514 * Fixed a miscompilation of the GETRF functions with CMAKE
515 * Imported bugfix 390 from LAPACK (missing NaN propagation in xCOMBSSQ)
516 * The size of the memory buffer used for splitting GEMM tasks across
517 multiple threads can now be configured in the build system.
520 * Fixed several compilation problems related to endianness
521 and ELF version on POWER8 and POWER9
522 * Fixed use of the absolute value IAMIN/IAMAX instead of IMIN/IMAX
523 * Fixed a race condition in the level3 blas code
526 * Fixed use of the absoltute value IAMIN/IAMAX instead of IMIN/IMAX
529 * Fixed a race condition in the level3 blas code
530 * Fixed compilation on Android
532 * Added support for Ampere EMAG8180
533 * Added support for Neoverse N1
534 * Improved performance of the blas_lock function
535 * Fixed a race condition in the level3 blas code
536 * Fixed a performance regression on TSV110-based servers
539 * Fixed a long-standing error with undeclared register overwrites
540 in the DSCAL microkernel for HASWELL,SKYLAKEX and ZEN
541 * Fixed a long-standing bug in the SSE implementation of IAMAX
542 * Fixed a CMAKE build failure with DYNAMIC_ARCH
543 * Fixed cpu autodetection of Goldmont+, Cannon Lake and Ice Lake
544 * Fixed a compilation failure on OSX with compiler name containing dash
545 * Fixed compilation with MinGW on SkylakeX
546 * Improved speed of the AVX512 GEMM3M kernel on SkylakeX
547 * Added an AVX512 STRMM kernel for SkylakeX
548 * Improved GEMM performance on Haswell and Zen
551 * fixed compilation of the DYNAMIC_ARCH code
553 ====================================================================
558 ` * LAPACK has been updated to 3.9.0 (plus patches up to
560 * CMAKE support has been improved in several areas including
562 * a thread race condition in the GEMM3M kernels was resolved
563 * the "generic" (plain C) gemm beta kernel used by many targets
565 * an optimized version of the LAPACK trtrs functions has been added
566 * an incompatibilty between the LAPACK tests and the OpenBLAS
567 implementation of XERBLA was resolved, removing the numerous
568 warnings about wrong error exits in the former
569 * support for NetBSD has been added
570 * support for compilation with g95 and non-GNU versions of ld
572 * support for compilation with (upcoming) gcc 10 has been added
575 * worked around miscompilation of several POWER8 and POWER9
576 kernels by older versions of gcc
577 * added support for big-endian POWER8 and for compilation on AIX
578 * corrected bugs in the big-endian support for PPC440 and PPC970
579 * DYNAMIC_ARCH support is now available in CMAKE builds as well
582 * performance of DGEMM_BETA and SGEMM_NCOPY has been improved
583 * compilation for 32bit works again
584 * performance of the RPCC function has been improved
585 * improved performance on small systems
586 * DYNAMIC_ARCH support is now available in CMAKE builds as well
587 * cross-compilation from OSX to IOS was simplified
590 * a new AVX512 DGEMM kernel was added and the AVX512 SGEMM kernel
591 was significantly improved
592 * optimized AVX512 kernels for CGEMM and ZGEMM have been added
593 * AVX2 kernels for STRMM, SGEMM, and CGEMM have been significantly
594 sped up and optimized CGEMM3M and ZGEMM3M kernels have been added
595 * added support for QEMU virtual cpus
596 * a compilation problem with PGI and SUN compilers was fixed
597 * Intel "Goldmont plus" is now autodetected
598 * a potential crash on program exit on MS Windows has been fixed
601 * an unwanted case sensitivity in the implementation of LSAME
602 on older 32bit AMD cpus was fixed
605 * Z15 is now supported as Z14
606 * DYNAMIC_ARCH is now available on ZARCH as well
608 ====================================================================
613 * having the gmake special variables TARGET_ARCH or TARGET_MACH
614 defined no longer causes build failures in ctest or utest
615 * defining NO_AFFINITY or USE_TLS to 0 in gmake builds no longer
616 has the same effect as setting them to 1
617 * a new test program was added to allow checking the library for
619 * a new option USE_LOCKING was added to ensure thread safety when
620 OpenBLAS itself is built without multithreading but will be
621 called from multiple threads.
622 * a build failure on Linux with glibc versions earlier than 2.5
624 * a runtime error with CPU enumeration (and NO_AFFINITY not set)
625 on glibc 2.6 was fixed
626 * NO_AFFINITY was added to the CMAKE options (and defaults to being
627 active on Linux, as in the gmake builds)
630 * the build-time logic for detection of AVX512 availability in
631 the processor and compiler was fixed
632 * gmake builds on OSX now set the internal name of the library to
633 libopenblas.0.dylib (consistent with CMAKE)
634 * the Haswell DGEMM kernel received a significant speedup through
635 improved prefetch and load instructions
636 * performance of DGEMM, DTRMM, DTRSM and ZDOT on Zen/Zen2 was markedly
637 increased by avoiding vpermpd instructions
638 * the SKYLAKEX (AVX512) DGEMM helper functions have now been disabled
639 to fix remaining errors in DGEMM, DSYMM and DTRMM
642 * added support for building on FreeBSD/powerpc64 and FreeBSD/ppc970
643 * added optimized kernels for POWER9 SGEMM and STRMM
646 * fixed the softfp implementations of xAMAX and IxAMAX
647 * removed the predefined -march= flags on both ARMV5 and ARMV6 as
648 they were appropriate for only a subset of platforms
650 ====================================================================
655 * the build tools now check that a given cpu TARGET is actually valid
656 * the build-time check of system features (c_check) has been made
657 less dependent on particular perl features (this should mainly
658 benefit building on Windows)
659 * several problem with the ReLAPACK integration were fixed,
660 including INTERFACE64 support and building a shared library
661 * building with CMAKE on BSD systems was improved
662 * a non-absolute SUM function was added based on the
663 existing optimized code for ASUM
664 * CBLAS interfaces to the IxMIN and IxMAX functions were added
665 * a name clash between LAPACKE and BOOST headers was resolved
666 * CMAKE builds with OpenMP failed to include the appropriate getrf_parallel
668 * a crash on thread (key) deletion with the USE_TLS=1 memory management
670 * restored several earlier fixes, in particular for OpenMP performance,
671 building on BSD, and calling fork on CYGWIN, which had inadvertently
672 been dropped in the 0.3.3 rewrite of the memory management code.
675 * the AVX512 DGEMM kernel has been disabled again due to unsolved problems
676 * building with old versions of MSVC was fixed
677 * it is now possible to build a static library on Windows with CMAKE
678 * accessing environment variables on CYGWIN at run time was fixed
679 * the CMAKE build system now recognizes 32bit userspace on 64bit hardware
680 * Intel "Denverton" atom and Hygon "Dhyana" zen CPUs are now autodetected
681 * building for DYNAMIC_ARCH with a DYNAMIC_LIST of targets is now supported
683 * building for DYNAMIC_ARCH with GENERIC as the default target is now supported
684 * a buffer overflow in the SSE GEMM kernel for Intel Nano targets was fixed
685 * assembly bugs involving undeclared modification of input operands were fixed
686 in the AXPY, DOT, GEMV, GER, SCAL, SYMV and TRSM microkernels for Nehalem,
687 Sandybridge, Haswell, Bulldozer and Piledriver. These would typically cause
688 test failures or segfaults when compiled with recent versions of gcc from 8 onward.
689 * a similar bug was fixed in the blas_quickdivide code used to split workloads
691 * a bug in the IxMIN implementation for the GENERIC target made it return the result of IxMAX
692 * fixed building on SkylakeX systems when either the compiler or the (emulated) operating
693 environment does not support AVX512
694 * improved GEMM performance on ZEN targets
697 * build failures caused by the recently added checks for AVX512 were fixed
698 * an inline assembly bug involving undeclared modification of an input argument was
699 fixed in the blas_quickdivide code used to split workloads in most functions
700 * a bug in the IMIN implementation for the GENERIC target made it return the result of IMAX
703 * a bug in the IMIN implementation made it return the result of IMAX
706 * single precision BLAS1/2 functions have received optimized POWER8 kernels
707 * POWER9 is now a separate target, with an optimized DGEMM/DTRMM kernel
708 * building on PPC970 systems under OSX Leopard or Tiger is now supported
709 * out-of-bounds memory accesses in the gemm_beta microkernels were fixed
710 * building a shared library on AIX is now supported for POWER6
711 * DYNAMIC_ARCH support has been added for POWER6 and newer
714 * corrected xDOT behaviour with zero INC_X or INC_Y
715 * a bug in the IMIN implementation made it return the result of IMAX
718 * added support for HiSilicon TSV110 cpus
719 * the CMAKE build system now recognizes 32bit userspace on 64bit hardware
720 * cross-compilation with CMAKE now works again
721 * a bug in the IMIN implementation made it return the result of IMAX
722 * ARMV8 builds with the BINARY=32 option are now automatically handled as ARMV7
725 * optimized microkernels for single precicion BLAS1/2 functions have been added
728 ====================================================================
733 * loop unrolling in TRMV has been enabled again.
734 * A domain error in the thread workload distribution for SYRK
736 * gmake builds will now automatically add -fPIC to the build
737 options if the platform requires it.
738 * a pthreads key leakage (and associate crash on dlclose) in
739 the USE_TLS codepath was fixed.
740 * building of the utest cases on systems that do not provide
741 an implementation of complex.h was fixed.
744 * the SkylakeX code was changed to compile on OSX.
745 * unwanted application of the -march=skylake-avx512 option
746 to the common code parts of a DYNAMIC_ARCH build was fixed.
747 * improved performance of SGEMM for small workloads on Skylake X.
748 * performance of SGEMM and DGEMM was improved on Haswell.
751 * a configuration error that broke the CNRM2 kernel was corrected.
752 * compilation of the GEMM kernels with CMAKE was fixed.
753 * DYNAMIC_ARCH builds are now available with CMAKE as well.
754 * using CMAKE for cross-compilation to the new cpu TARGETs
755 introduced in 0.3.4 now works.
758 * a problem in cpu autodetection for AIX has been corrected.
760 ====================================================================
765 * the new, experimental thread-local memory allocation had
766 inadvertently been left enabled for gmake builds in 0.3.3
767 despite the announcement. It is now disabled by default, and
768 single-threaded builds will keep using the old allocator even
769 if the USE_TLS option is turned on.
770 * OpenBLAS will now provide enough buffer space for at least 50
772 * The output of openblas_get_config() now contains the version
774 * A serious thread safety bug in GEMV operation with small M and
775 large N size has been fixed.
776 * The code will now automatically call blas_thread_init after a
777 fork if needed before handling a call to openblas_set_num_threads
778 * Accesses to parallelized level3 functions from multiple callers
779 are now serialized to avoid thread races (unless using OpenMP).
780 This should provide better performance than the known-threadsafe
781 (but non-default) USE_SIMPLE_THREADED_LEVEL3 option.
782 * When building LAPACK with gfortran, -frecursive is now (again)
783 enabled by default to ensure correct behaviour.
784 * The OpenBLAS version cblas.h now supports both CBLAS_ORDER and
785 CBLAS_LAYOUT as the name of the matrix row/column order option.
786 * Externally set LDFLAGS are now passed through to the final compile/link
787 steps to facilitate setting platform-specific linker flags.
788 * A potential race condition during the build of LAPACK (that would
789 usually manifest itself as a failure to build TESTING/MATGEN) has been
791 * xHEMV has been changed to stay single-threaded for small input sizes
792 where the overhead of multithreading exceeds any possible gains
793 * CSWAP and ZSWAP have been limited to a single thread except on ARMV8 or
794 ThunderX hardware with sizable input.
795 * Linker flags for the PGI compiler have been updated
796 * Behaviour of AXPY with zero increments is now handled in the C interface,
797 correcting the result on at least Intel Atom.
798 * The result matrix from calling SGELSS with an all-zero input matrix is
799 now zeroed completely.
802 * Autodetection of AMD Ryzen2 has been fixed (again).
803 * CMAKE builds now support labeling of an INTERFACE64=1 build of
804 the library with the _64 suffix.
805 * AVX512 version of DGEMM has been added and the AVX512 SGEMM kernel
806 has been sped up by rewriting with C intrinsics
807 * Fixed compilation on RHEL5/CENTOS5 (issue with typename __WAIT_STATUS)
810 * added support for building on AIX (with gcc and GNU tools from AIX Toolbox).
811 * CPU type detection has been implemented for AIX.
812 * CPU type detection has been fixed for NETBSD.
815 * AXPY on LOONGSON3A has been corrected to pass "zero increment" utest.
816 * DSDOT on LOONGSON3A has been fixed.
817 * the SGEMM microkernel has been hardened against potential data loss.
820 * DYNAMic_ARCH support is now available for 64bit ARM
821 * cross-compiling for ARMV8 under iOS now works.
822 * cpu-specific code has been rearranged to make better use of both
823 hardware commonalities and model-specific compiler optimizations.
824 * XGENE1 has been removed as a TARGET, superseded by the improved generic
828 * Older assembly mnemonics have been converted to UAL form to allow
829 building with clang 7.0
830 * Cross compiling LAPACKE for Android has been fixed again (broken by
831 update to LAPACK 3.7.0 some while ago).
833 ====================================================================
838 * thread memory allocation has been switched back to the method
839 used before version 0.3.1 due to unexpected problems caused by
840 the new code under some circumstances. A new compile-time option
841 USE_TLS has been added to enable the new code, and it is hoped
842 that this can become the default again in the next version.
843 * LAPAck PR272 has been integrated, which fixes spurious errors
844 in DSYEVR and related functions caused by missing conversion
845 from ILAENV to ILAENV_2STAGE in several _2stage routines.
846 * the cmake-generated OpenBLASConfig.cmake now uses correct case
847 for the name of the library
848 * added support for Haiku OS
851 * added AVX512 implementations of SDOT, DDOT, SAXPY, DAXPY,
852 DSCAL, DGEMVN and DSYMVL
853 * added a workaround for a cygwin issue that prevented compilation
857 * added autodetection of Z14
858 * fixed TRMM errors in the generic target
860 ====================================================================
865 * fixes for regressions caused by the rewrite of the thread
866 initialization code in 0.3.1
869 * fixed cpu autodetection for the BSDs
872 * fixed utest errors in AXPY, DSDOT, ROT and SWAP
875 * added autodetection of AMD Ryzen 2
876 * fixed build with older versions of MSVC
878 ====================================================================
883 * rewritten thread initialization code with significantly reduced overhead
884 * added CBLAS interfaces to the IxAMIN BLAS extension functions
885 * fixed the lapack-test target
886 * CMAKE builds now create an OpenBLASConfig.cmake file
887 * ZAXPY now uses a single thread for small input sizes
888 * the LAPACK code was updated from Reference-LAPACK/lapack#253
889 (fixing LAPACKE interfaces to Aasen's functions)
892 * corrected CROT and ZROT behaviour with zero INC_X
895 * corrected xDOT behaviour with zero INC_X or INC_Y
898 * retired some older targets of DYNAMIC_ARCH builds to a new option DYNAMIC_OLDER,
899 this affects PENRYN,DUNNINGTON,OPTERON,OPTERON_SSE3,BOBCAT,ATOM and NANO
900 (which will still be supported via the slower PRESCOTT kernels when this option is not set)
901 * added an option DYNAMIC_LIST that (used in conjunction with DYNAMIC_ARCH) allows to
902 specify the list of x86_64 targets to include. Any target not on the list will be supported
903 by the Sandybridge or Nehalem kernels if available, or by Prescott.
904 * improved SWITCH_RATIO on Haswell for increased GEMM throughput
905 * added initial support for Intel Skylake X, including an AVX512 SGEMM kernel
906 * added autodetection of Intel Cannon Lake series as Skylake X
907 * added a default L2 cache size for hypervisors that return zero here (Chromebook)
908 * fixed a name clash with recent Windows10 headers that broke the build with (at least)
909 recent mingw from MSYS2
910 * fixed a link error in mixed clang/gfortran builds with OpenMP
911 * updated the OSX deployment target to 10.8
912 * switched on parallel make for builds on MS Windows by default
915 * fixed SSWAP and DSWAP behaviour with zero INC_X and INC_Y
917 ====================================================================
922 * fixed some more thread race and locking bugs
923 * added preliminary support for calling an OpenMP build of the library from multiple threads
924 * removed performance impact of thread locks added in 0.2.20 on OpenMP code
925 * general code cleanup
926 * optimized DSDOT implementation
927 * improved thread distribution for GEMM
928 * corrected IMATCOPY/OMATCOPY implementation
929 * fixed out-of-bounds accesses in the multithreaded xBMV/xPMV and SYMV implementations
930 * cmake build improvements
931 * pkgconfig file now contains build options
932 * openblas_get_config() now reports USE_OPENMP and NUM_THREADS settings used for the build
933 * corrections and improvements for systems with more than 64 cpus
934 * LAPACK code updated to 3.8.0 including later fixes
935 * added ReLAPACK, a recursive implementation of several LAPACK functions
936 * Rewrote ROTMG to handle cases that the netlib code failed to address
937 * Disabled (broken) multithreading code for xTRMV
938 * corrected prototypes of complex CBLAS functions to make our cblas.h match the generally accepted standard
939 * shared memory access failures on startup are now handled more gracefully
940 * restored utests from earlier releases (and made them pass on all affected systems)
943 * several fixes for cpu autodetection
946 * corrected vector register overwriting in several Power8 kernels
947 * optimized additional BLAS functions
950 * added support for CortexA53 and A72
951 * added autodetection for ThunderX2T99
952 * made most optimized kernels the default for generic ARMv8 targets
955 * parallelized DDOT kernel for Haswell
956 * changed alignment directives in assembly kernels to boost performance on OSX
957 * fixed register handling in the GEMV microkernels (bug exposed by gcc7)
958 * added support for building on OpenBSD and Dragonfly
959 * updated compiler options to work with Intel release 2018
960 * support fully optimized build with clang/flang on Microsoft Windows
961 * fixed building on AIX
964 * added optimized BLAS 1/2 functions
967 * fixed cpu autodetection helper code
968 * added mips32 1004K cpu (Mediatek MT7621 and similar SoC)
969 * added mips64 I6500 cpu
971 ====================================================================
976 * Improved CMake support
977 * Fixed several thread race and locking bugs
978 * Fixed default LAPACK optimization level
979 * Updated LAPACK to 3.7.0
980 * Added ReLAPACK (https://github.com/HPAC/ReLAPACK, make BUILD_RELAPACK=1)
983 * Optimizations for Power9
984 * Fixed several Power8 assembly bugs
987 * New optimized Vulcan and ThunderX2T99 targets
988 * Support for ARMV7 SOFT_FP ABI (make ARM_SOFTFP_ABI=1)
989 * Detect all cpu cores including offline ones
990 * Fix compilation with CLANG
991 * Support building a shared library for Android
994 * Fixed several threading issues
995 * Fix compilation with CLANG
998 * Detect Intel Bay Trail and Apollo Lake
999 * Detect Intel Sky Lake and Kaby Lake
1000 * Detect Intel Knights Landing
1001 * Detect AMD A8, A10, A12 and Ryzen
1002 * Support 64bit builds with Visual Studio
1003 * Fix building with Intel and PGI compilers
1004 * Fix building with MINGW and TDM-GCC
1005 * Fix cmake builds for Haswell and related cpus
1006 * Fix building for Sandybridge with CLANG 3.9
1007 * Add support for the FLANG compiler
1010 * New target z13 with BLAS3 optimizations
1012 ====================================================================
1016 * Improved cross compiling.
1017 * Fix the bug on musl libc.
1020 * Optimize BLAS on Power8
1021 * Fixed Julia+OpenBLAS bugs on Power8
1024 * Optimize BLAS on MIPS P5600 and I6400 (Thanks, Shivraj Patil, Kaustubh Raste)
1027 * Improved on ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
1030 ====================================================================
1034 * If you set MAKE_NB_JOBS flag less or equal than zero,
1035 make will be without -j.
1038 * Support building Visual Studio static library. (#813, Thanks, theoractice)
1039 * Fix bugs to pass buidbot CI tests (http://build.openblas.net)
1042 * Provide DGEMM 8x4 kernel for Cortex-A57 (Thanks, Ashwin Sekhar T K)
1045 * Optimize S and C BLAS3 on Power8
1046 * Optimize BLAS2/1 on Power8
1048 ====================================================================
1052 * Enable BUILD_LAPACK_DEPRECATED=1 by default.
1054 ====================================================================
1058 * Avoid potential getenv segfault. (#716)
1059 * Import LAPACK svn bugfix #142-#147,#150-#155
1062 * Optimize c/zgemv for AMD Bulldozer, Piledriver, Steamroller
1063 * Fix bug with scipy linalg test.
1066 * Improve DGEMM for ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
1069 * Optimize D and Z BLAS3 functions for Power8.
1071 ====================================================================
1075 * Upgrade LAPACK to 3.6.0 version.
1076 Add BUILD_LAPACK_DEPRECATED option in Makefile.rule to build
1077 LAPACK deprecated functions.
1078 * Add MAKE_NB_JOBS option in Makefile.
1079 Force number of make jobs.This is particularly
1080 useful when using distcc. (#735. Thanks, Jerome Robert.)
1081 * Redesign unit test. Run unit/regression test at every build (Travis-CI and Appveyor).
1082 * Disable multi-threading for small size swap and ger. (#744. Thanks, Jerome Robert)
1083 * Improve small zger, zgemv, ztrmv using stack alloction (#727. Thanks, Jerome Robert)
1084 * Let openblas_get_num_threads return the number of active threads.
1085 (#760. Thanks, Jerome Robert)
1086 * Support illumos(OmniOS). (#749. Thanks, Lauri Tirkkonen)
1087 * Fix LAPACK Dormbr, Dormlq bug. (#711, #713. Thanks, Brendan Tracey)
1088 * Update scipy benchmark script. (#745. Thanks, John Kirkham)
1091 * Optimize trsm kernels for AMD Bulldozer, Piledriver, Steamroller.
1092 * Detect Intel Avoton.
1093 * Detect AMD Trinity, Richland, E2-3200.
1094 * Fix gemv performance bug on Mac OSX Intel Haswell.
1095 * Fix some bugs with CMake and Visual Studio
1098 * Support and optimize Cortex-A57 AArch64.
1099 (#686. Thanks, Ashwin Sekhar TK)
1100 * Fix Android build on ARMV7 (#778. Thanks, Paul Mustiere)
1101 * Update ARMV6 kernels.
1104 * Fix detection of POWER architecture
1105 (#684. Thanks, Sebastien Villemot)
1107 ====================================================================
1111 * Support cmake on x86/x86-64. Natively compiling on MS Visual Studio.
1112 (experimental. Thank Hank Anderson for the initial cmake porting work.)
1114 On Linux and Mac OSX, OpenBLAS cmake supports assembly kernels.
1117 make test (Optional)
1119 On Windows MS Visual Studio, OpenBLAS cmake only support C kernels.
1120 (OpenBLAS uses AT&T style assembly, which is not supported by MSVC.)
1121 e.g. cmake -G "Visual Studio 12 Win64" .
1122 Open OpenBLAS.sln and build.
1124 * Enable MAX_STACK_ALLOC flags by default.
1125 Improve ger and gemv for small matrices.
1126 * Improve gemv parallel with small m and large n case.
1127 * Improve ?imatcopy when lda==ldb (#633. Thanks, Martin Koehler)
1128 * Add vecLib benchmarks (#565. Thanks, Andreas Noack.)
1129 * Fix LAPACK lantr for row major matrices (#634. Thanks, Dan Kortschak)
1130 * Fix LAPACKE lansy (#640. Thanks, Dan Kortschak)
1131 * Import bug fixes for LAPACKE s/dormlq, c/zunmlq
1132 * Raise the signal when pthread_create fails (#668. Thanks, James K. Lowden)
1133 * Remove g77 from compiler list.
1134 * Enable AppVeyor Windows CI.
1137 * Support pure C generic kernels for x86/x86-64.
1138 * Support Intel Boardwell and Skylake by Haswell kernels.
1139 * Support AMD Excavator by Steamroller kernels.
1140 * Optimize s/d/c/zdot for Intel SandyBridge and Haswell.
1141 * Optimize s/d/c/zdot for AMD Piledriver and Steamroller.
1142 * Optimize s/d/c/zapxy for Intel SandyBridge and Haswell.
1143 * Optimize s/d/c/zapxy for AMD Piledriver and Steamroller.
1144 * Optimize d/c/zscal for Intel Haswell, dscal for Intel SandyBridge.
1145 * Optimize d/c/zscal for AMD Bulldozer, Piledriver and Steamroller.
1146 * Optimize s/dger for Intel SandyBridge.
1147 * Optimize s/dsymv for Intel SandyBridge.
1148 * Optimize ssymv for Intel Haswell.
1149 * Optimize dgemv for Intel Nehalem and Haswell.
1150 * Optimize dtrmm for Intel Haswell.
1153 * Support Android NDK armeabi-v7a-hard ABI (-mfloat-abi=hard)
1154 e.g. make HOSTCC=gcc CC=arm-linux-androideabi-gcc NO_LAPACK=1 TARGET=ARMV7
1155 * Fix lock, rpcc bugs (#616, #617. Thanks, Grazvydas Ignotas)
1157 * Support ppc64le platform (ELF ABI v2. #612. Thanks, Matthew Brandyberry.)
1158 * Support POWER7/8 by POWER6 kernels. (#612. Thanks, Fábio Perez.)
1160 ====================================================================
1164 * Improve OpenBLASConfig.cmake. (#474, #475. Thanks, xantares.)
1165 * Improve ger and gemv for small matrices by stack allocation.
1166 e.g. make -DMAX_STACK_ALLOC=2048 (#482. Thanks, Jerome Robert.)
1167 * Introduce openblas_get_num_threads and openblas_get_num_procs.
1168 (#497. Thanks, Erik Schnetter.)
1169 * Add ATLAS-style ?geadd function. (#509. Thanks, Martin Köhler.)
1170 * Fix c/zsyr bug with negative incx. (#492.)
1171 * Fix race condition during shutdown causing a crash in
1172 gotoblas_set_affinity(). (#508. Thanks, Ton van den Heuvel.)
1175 * Support AMD Streamroller.
1178 * Add Cortex-A9 and Cortex-A15 targets.
1180 ====================================================================
1184 * Add SYMBOLPREFIX and SYMBOLSUFFIX makefile options
1185 for adding a prefix or suffix to all exported symbol names
1186 in the shared library.(#459, Thanks Tony Kelman)
1187 * Provide OpenBLASConfig.cmake at installation.
1188 * Fix Fortran compiler detection on FreeBSD.
1189 (#470, Thanks Mike Nolta)
1193 * Add generic kernel files for x86-64. make TARGET=GENERIC
1194 * Fix a bug of sgemm kernel on Intel Sandy Bridge.
1195 * Fix c_check bug on some amd64 systems. (#471, Thanks Mike Nolta)
1198 * Support APM's X-Gene 1 AArch64 processors.
1199 Optimize trmm and sgemm. (#465, Thanks Dave Nuechterlein)
1201 ====================================================================
1205 * Added CBLAS interface for ?omatcopy and ?imatcopy.
1206 * Enable ?gemm3m functions.
1207 * Added benchmark for ?gemm3m.
1208 * Optimized multithreading lower limits.
1209 * Disabled SYMM3M and HEMM3M functions
1210 because of segment violations.
1213 * Improved axpy and symv performance on AMD Bulldozer.
1214 * Improved gemv performance on modern Intel and AMD CPUs.
1216 ====================================================================
1220 * Added some benchmark codes.
1221 * Fix link error on Linux/musl.(Thanks Isaac Dunham)
1224 * Improved s/c/zgemm performance for Intel Haswell.
1225 * Improved s/d/c/zgemv performance.
1226 * Support the big numa machine.(EXPERIMENT)
1229 * Fix detection when cpuinfo uses "Processor". (Thanks Isaiah)
1231 ====================================================================
1235 * Added BLAS extensions as following.
1236 s/d/c/zaxpby, s/d/c/zimatcopy, s/d/c/zomatcopy.
1237 * Added OPENBLAS_CORETYPE environment for dynamic_arch. (a86d34)
1238 * Added NO_AVX2 flag for old binutils. (#401)
1239 * Support outputing the CPU corename on runtime.(#407)
1240 * Patched LAPACK to fix bug 114, 117, 118.
1241 (http://www.netlib.org/lapack/bug_list.html)
1242 * Disabled ?gemm3m for a work-around fix. (#400)
1244 * Fixed lots of bugs for optimized kernels on sandybridge,Haswell,
1245 bulldozer, and piledriver.
1246 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
1249 * Improved LAPACK testing.
1251 ====================================================================
1255 * Improved the result for LAPACK testing. (#372)
1256 * Installed DLL to prefix/bin instead of prefix/lib. (#366)
1257 * Build import library on Windows.(#374)
1259 * To improve LAPACK testing, we fallback some kernels. (#372)
1260 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
1262 ====================================================================
1266 * Added OPENBLAS_VERBOSE environment variable.(#338)
1267 * Make OpenBLAS thread-pool resilient to fork via pthread_atfork.
1268 (#294, Thank Olivier Grisel)
1272 * Detect Intel Haswell for new Macbook.
1274 ====================================================================
1278 * Update LAPACK to 3.5.0 version
1279 * Fixed compatiable issues with Clang and Pathscale compilers.
1282 * Optimization on Intel Haswell.
1283 * Enable optimization kernels on AMD Bulldozer and Piledriver.
1286 * Support ARMv6 and ARMv7 ISA.
1287 * Optimization on ARM Cortex-A9.
1289 ====================================================================
1293 * Support Open64 5.0. (#266)
1294 * Add executable stack markings. (#262, Thank Sébastien Fabbro)
1295 * Respect user's LDFLAGS (Thank Sébastien Fabbro)
1298 * Rollback bulldozer and piledriver kernels to barcelona kernels (#263)
1299 We will fix the compuational error bug in bulldozer and piledriver kernels.
1301 ====================================================================
1305 * Support LSB (Linux Standard Base) 4.1.
1307 * Include LAPACK 3.4.2 source codes to the repo.
1308 Avoid downloading at compile time.
1309 * Add NO_PARALLEL_MAKE flag to disable parallel make.
1310 * Create openblas_get_parallel to retrieve information which
1311 parallelization model is used by OpenBLAS. (Thank grisuthedragon)
1312 * Detect LLVM/Clang compiler. The default compiler is Clang on Mac OS X.
1313 * Change LIBSUFFIX from .lib to .a on windows.
1314 * A work-around for dtrti_U single thread bug. Replace it with LAPACK codes. (#191)
1317 * Optimize c/zgemm, trsm, dgemv_n, ddot, daxpy, dcopy on
1318 AMD Bulldozer. (Thank Werner Saar)
1319 * Add Intel Haswell support (using Sandybridge optimizations).
1321 * Add AMD Piledriver support (using Bulldozer optimizations).
1322 * Fix the computational error in zgemm avx kernel on
1324 * Fix the overflow bug in gemv.
1325 * Fix the overflow bug in multi-threaded BLAS3, getrf when NUM_THREADS
1326 is very large.(#214, #221, #246).
1328 * Support loongcc (Open64 based) compiler for ICT Loongson 3A/B.
1331 * Support Power7 by old Power6 kernels. (#220)
1333 ====================================================================
1337 * Improved OpenMP performance slightly. (d744c9)
1338 * Improved cblas.h compatibility with Intel MKL.(#185)
1339 * Fixed the overflowing bug in single thread cholesky factorization.
1340 * Fixed the overflowing buffer bug of multithreading hbmv and sbmv.(#174)
1343 * Added AMD Bulldozer x86-64 S/DGEMM AVX kernels. (Thank Werner Saar)
1344 We will tune the performance in future.
1345 * Auto-detect Intel Xeon E7540.
1346 * Fixed the overflowing buffer bug of gemv. (#173)
1347 * Fixed the bug of s/cdot about invalid reading NAN on x86_64. (#189)
1351 ====================================================================
1355 * Added NO_SHARED flag to disable generating the shared library.
1356 * Compile LAPACKE with ILP64 modle when INTERFACE64=1 (#158)
1357 * Export LAPACK 3.4.2 symbols in shared library. (#147)
1358 * Only detect the number of physical CPU cores on Mac OSX. (#157)
1359 * Fixed NetBSD build. (#155)
1360 * Fixed compilation with TARGET=GENERIC. (#160)
1362 * Restore the original CPU affinity when calling
1363 openblas_set_num_threads(1) (#153)
1364 * Fixed a SEGFAULT bug in dgemv_t when m is very large.(#154)
1367 ====================================================================
1371 * Upgraded LAPACK to 3.4.2 version. (#145)
1372 * Provided support for passing CFLAGS, FFLAGS, PFLAGS,
1373 FPFLAGS to make. (#137)
1374 * f77blas.h:compatibility for compilers without C99 complex
1375 number support. (#141)
1377 * Added NO_AVX flag. Check OS supporting AVX on runtime. (#139)
1378 * Fixed zdot incompatibility ABI issue with GCC 4.7 on
1379 Windows 32-bit. (#140)
1381 * Fixed the generation of shared library bug.
1382 * Fixed the detection bug on the Loongson 3A server.
1383 ====================================================================
1387 * Fixed LAPACK unstable bug about ?laswp. (#130)
1388 * Fixed the shared library bug about unloading the library on
1390 * Fixed the compilation failure on BlueGene/P (TARGET=PPC440FP2)
1391 Please use gcc and IBM xlf. (#134)
1393 * Supported goto_set_num_threads and openblas_set_num_threads
1394 APIs in Windows. They can set the number of threads on runtime.
1396 ====================================================================
1400 * Fixed exporting DLL functions bug on Windows/MingW
1401 * Support GNU Hurd (Thank Sylvestre Ledru)
1402 * Support kfreebsd kernel (Thank Sylvestre Ledru)
1404 * Support Intel Sandy Bridge 22nm desktop/mobile CPU
1406 * Improve the detection of SPARC (Thank Sylvestre Ledru)
1408 ====================================================================
1413 * Fixed the SEGFAULT bug about hyper-theading
1414 * Support AMD Bulldozer by using GotoBLAS2 AMD Barcelona codes
1416 ====================================================================
1420 * Removed the limitation (64) of numbers of CPU cores.
1421 Now, it supports 256 cores at max.
1422 * Supported clang compiler.
1423 * Fixed some build bugs on FreeBSD
1425 * Optimized Level-3 BLAS on Intel Sandy Bridge x86-64 by AVX instructions.
1426 Please use gcc >= 4.6 or clang >=3.1.
1427 * Support AMD Bobcat by using GotoBLAS2 AMD Barcelona codes.
1429 ====================================================================
1433 * Upgraded LAPACK to 3.4.1 version. (Thank Zaheer Chothia)
1434 * Supported LAPACKE, a C interface to LAPACKE. (Thank Zaheer Chothia)
1435 * Fixed the build bug (MD5 and download) on Mac OSX.
1436 * Auto download CUnit 2.1.2-2 from SF.net with UTEST_CHECK=1.
1437 * Fxied the compatibility issue for compilers without C99 complex number
1438 (e.g. Visual Studio)
1440 * Auto-detect Intel Sandy Bridge Core i7-3xxx & Xeon E7 Westmere-EX.
1441 * Test alpha=Nan in dscale.
1442 * Fixed a SEGFAULT bug in samax on x86 windows.
1444 ====================================================================
1448 * Set soname of shared library on Linux.
1449 * Added LIBNAMESUFFIX flag in Makefile.rule. The user can use
1450 this flag to control the library name, e.g. libopenblas.a,
1451 libopenblas_ifort.a or libopenblas_omp.a.
1452 * Added GEMM_MULTITHREAD_THRESHOLD flag in Makefile.rule.
1453 The lib use single thread in GEMM function with small matrices.
1455 * Used GEMV SSE/SSE2 kernels on x86 32-bit.
1456 * Exported CBLAS functions in Windows DLL.
1458 * Completed Level-3 BLAS optimization on Loongson 3A CPU.
1459 * Improved GEMV performance on Loongson 3A CPU.
1460 * Improved Level-3 BLAS performance on Loongson 3B CPU. (EXPERIMENT)
1462 ====================================================================
1463 Version 0.1 alpha2.5
1466 * Fixed missing "#include <sched.h>" bug on Mac OS X.
1467 Thank Mike Nolta for the patch.
1468 * Upgraded LAPACK to 3.4.0 version
1469 * Fixed a bug on Mac OS X. Don't require SystemStubs on OS X.
1470 SystemStubs does not exist on Lion. Thank Stefan Karpinski.
1471 * Improved README with using OpenMP. Check the internal threads
1472 count less than or equal to omp_get_max_threads()
1474 * Auto-detect Intel Core i6/i7 (Sandy Bridge) CPU with Nehalem assembly kernels
1475 * Fixed some bugs on MingW 64-bit including zgemv, cdot, zdot.
1477 ====================================================================
1478 Version 0.1 alpha2.4
1481 * Fixed a bug about installation. The header file "fblas77.h"
1483 * Fixed #61 a building bug about setting TARGET and DYNAMIC_ARCH.
1484 * Try to handle absolute path of shared library in OSX. (#57)
1485 Thank Dr Kane O'Donnell.
1486 * Changed the installation folder layout to $(PREFIX)/include and
1490 * Fixed #58 zdot/xdot SEGFAULT bug with GCC-4.6 on x86. According
1491 to i386 calling convention, The callee should remove the first
1492 hidden parameter.Thank Mr. John for this patch.
1494 ====================================================================
1495 Version 0.1 alpha2.3
1499 * Added DTB_ENTRIES into dynamic arch setting parameters. Now,
1500 it can read DTB_ENTRIES on runtime. (Refs issue #55 on github)
1502 ====================================================================
1503 Version 0.1 alpha2.2
1507 * Fixed a building bug when DYNAMIC_ARCH=1 & INTERFACE64=1.
1508 (Refs issue #44 on github)
1510 ====================================================================
1511 Version 0.1 alpha2.1
1515 * Stop the build and output the error message when detecting
1516 fortran compiler failed. (Refs issue #42 on github)
1518 ====================================================================
1523 * Fixed blasint undefined bug in <cblas.h> file. Other software
1524 could include this header successfully(Refs issue #13 on github)
1525 * Fixed the SEGFAULT bug on 64 cores. On SMP server, the number
1526 of CPUs or cores should be less than or equal to 64.(Refs issue #14
1528 * Support "void goto_set_num_threads(int num_threads)" and "void
1529 openblas_set_num_threads(int num_threads)" when USE_OPENMP=1
1530 * Added extern "C" to support C++. Thank Tasio for the patch(Refs
1531 issue #21 on github)
1532 * Provided an error message when the arch is not supported.(Refs
1533 issue #19 on github)
1534 * Fixed issue #23. Fixed a bug of f_check script about generating link flags.
1535 * Added openblas_set_num_threads for Fortran.
1536 * Fixed #25 a wrong result of rotmg.
1537 * Fixed a bug about detecting underscore prefix in c_check.
1538 * Print the wall time (cycles) with enabling FUNCTION_PROFILE
1539 * Fixed #35 a build bug with NO_LAPACK=1 & DYNAMIC_ARCH=1
1540 * Added install target. You can use "make install". (Refs #20)
1544 * Fixed #28 a wrong result of dsdot on x86_64.
1545 * Fixed #32 a SEGFAULT bug of zdotc with gcc-4.6.
1546 * Fixed #33 ztrmm bug on Nehalem.
1547 * Work-around #27 the low performance axpy issue with small imput size & multithreads.
1550 * Fixed #28 a wrong result of dsdot on Loongson3A/MIPS64.
1551 * Optimized single/double precision BLAS Level3 on Loongson3A/MIPS64. (Refs #2)
1552 * Optimized single/double precision axpy function on Loongson3A/MIPS64. (Refs #3)
1554 ====================================================================
1559 * Support "make NO_LAPACK=1" to build the library without
1561 * Fixed randomly SEGFAULT when nodemask==NULL with above Linux 2.6.34.
1562 Thank Mr.Ei-ji Nakama providing this patch. (Refs issue #12 on github)
1563 * Added DEBUG=1 rule in Makefile.rule to build debug version.
1564 * Disable compiling quad precision in reference BLAS library(netlib BLAS).
1565 * Added unit testcases in utest/ subdir. Used CUnit framework.
1566 * Supported OPENBLAS_* & GOTO_* environment variables (Pleas see README)
1567 * Imported GotoBLAS2 1.13 BSD version
1570 * On x86 32bits, fixed a bug in zdot_sse2.S line 191. This would casue
1571 zdotu & zdotc failures. Instead, work-around it. (Refs issue #8 #9 on github)
1572 * Modified ?axpy functions to return same netlib BLAS results
1573 when incx==0 or incy==0 (Refs issue #7 on github)
1574 * Modified ?swap functions to return same netlib BLAS results
1575 when incx==0 or incy==0 (Refs issue #6 on github)
1576 * Modified ?rot functions to return same netlib BLAS results
1577 when incx==0 or incy==0 (Refs issue #4 on github)
1578 * Detect Intel Westmere,Intel Clarkdale and Intel Arrandale
1579 to use Nehalem codes.
1580 * Fixed a typo bug about compiling dynamic ARCH library.
1582 * Improve daxpy performance on ICT Loongson 3A.
1583 * Supported ICT Loongson 3A CPU (Refs issue #1 on github)
1584 ====================================================================