2 ====================================================================
7 - Updated the included LAPACK to Reference-LAPACK release 3.10.1
8 - when no Fortran compiler is available, OpenBLAS builds will now automatically
9 build LAPACK from an f2c-converted copy of LAPACK 3.9.0 unless the NO_LAPACK option
11 - similarly added C versions of the BLAS and CBLAS tests
12 - enabled building of the ReLAPACK GEMMT kernels when ReLAPACK is built
13 - function LAPACKE_lsame is now annotated with the GCC attribute "const" to aid static analyzers
14 - added USE_TLS to the list of options reported by the openblas_get_config() function
15 - CMAKE builds now support the BUILD_TESTING keyword (to disable the LAPACK testsuite) of Reference-LAPACK
16 - fixed CMAKE builds of the laswp_ncopy and neg_tcopy kernels
17 - removed the build system requirements for PERL (while keeping the original perl scripts as backup)
18 - handle building and running OpenBLAS on systems that report zero available cpu cores
19 - added SYMBOLPREFIX/SYMBOLSUFFIX handling for LAPACK 3.10.0 functions added in 0.3.20
20 - fixed linking of the utests on QNX
21 - Added support for compilation with the Intel ifx compiler
22 - Added support for compilation with the Fujitsu FCC compiler for Fugaku
23 - Added support for compilation with the Cray C and Fortran compilers
24 - reverted OpenMP threadpool behaviour in the exec_blas call to its state before 0.3.11, that is
25 the threadpool will no longer grow or shrink on demand as the overhead for this is too big at least with
26 GNU OpenMP. The adaptive behaviour introduced in 0.3.11 can still be requested at runtime by setting
27 the environment variable OMP_ADAPTIVE
28 - worked around spurious STFSM/CTFSM errors reported by the LAPACK testsuite
31 - fixed determination of compiler support for AVX512 and removed the 0.3.19
32 workaround for building SKYLAKEX kernels on Sandybridge hardware
33 - fixed compilation for the SKYLAKEX target with gcc 6
34 - fixed compilation of the CooperLake SBGEMM kernel with LLVM
35 - fixed compilation of the SkyLakeX small matrix GEMM kernels with LLVM or ICC
36 - fixed compilation of some BFLOAT16 kernels with CMAKE
37 - added support for the Zhaoxin/Centaur KH40000 cpu
38 - fixed a potential crash in the ZSYMV kernel used for all targets except generic
39 - fixed gmake compilation for DYNAMIC_ARCH with a DYNAMIC_LIST including ATOM
40 - fixed compilation of LAPACKE with the INTEGER64 option on Windows
41 - added support for cross-compiling to individual Intel or AMD targets using CMAKE
42 (previously only CORE2 supported, added targets are ATOM, PRESCOTT, NEHALEM, SANDYBRIDGE,
43 HASWELL,SKYLAKEX, COOPERLAKE, SAPPHIRERAPIDS, OPTERON, BARCELONA, BULLDOZER, PILEDRIVER,
44 STEAMROLLER,EXCAVATOR, ZEN)
47 - worked around an overflow error in the DNRM2 kernel
50 - worked around an overflow error in the POWER6 DNRM2 kernel
51 - fixed compilation on PPC440
52 - fixed a performance regression in the level1 BLAS on POWER10
53 - fixed the POWER10 ZGEMM kernel
54 - fixed singlethreaded builds for POWER10
55 - fixed compilation of the POWER10 DGEMV kernel with older gcc versions
56 - enabled compilation of the BFLOAT16 kernels by default
57 - enabled the small matrix kernels by default for DYNAMIC_ARCH builds
58 - added a workaround for a miscompilation of the CDOT and ZDOT kernels by GCC 12
61 - fixed cpu autodetection logic
64 - added an SBGEMM kernel for Neoverse N2
65 - worked around an overflow error in the DNRM2 kernel used on M1, NeoverseN1, ThunderX2T99
66 - added support for ARM64 systems running MS Windows
67 - added support for cross-compiling to the GENERIC ARMV8 target under CMAKE (Windows/MSVC)
68 - fixed a performance regression in the generic ARMV8 DGEMM kernel introduced in 0.3.19
69 - added initial support for the Apple M1 cpu under Linux
70 - added initial support for the Phytium FT2000 cpu
71 - added initial support for the Cortex A510, A710, X1 and X2 cpu
72 - fixed an accidental mixup of cpu identifiers in the autodetection code introduced in 0.3.20
73 - fixed linking of Apple M1 builds on macOS 12 and later with recent XCode
74 - made Neoverse N2 available in DYNAMIC_ARCH builds
77 - worked around an overflow error in the DNRM2 kernel
80 - worked around an overflow error in the DNRM2 kernel
81 - added preliminary support for the LOONGSON2K1000 cpu
82 - added DYNAMIC_ARCH support
84 ====================================================================
89 - some code cleanup, with added casts etc.
90 - fixed obtaining the cpu count with OpenMP and OMP_PROC_BIND unset
91 - fixed pivot index calculation by ?LASWP for negative increments other than one
92 - fixed input argument check in LAPACK ? GEQRT2
93 - improved the check for a Fortran compiler in CMAKE builds
94 - disabled building OpenBLAS' optimized versions of LAPACK complex SPMV,SPR,SYMV,SYR with NO_LAPACK=1
95 - fixed building of LAPACK on certain distributed filesystems with parallel gmake
96 - fixed building the shared library on MacOS with classic flang
99 - fixed cross-compilation with CMAKE for CORE2 target
100 - fixed miscompilation of AVX512 code in DYNAMIC_ARCH builds
101 - added support for the "incidental" AVX512 hardware in Alder Lake when enabled in BIOS
104 - add new architecture (Russian Elbrus E2000 family)
110 - added SVE-enabled CGEMM and ZGEMM kernels for ARMV8SVE and A64FX
111 - added support for Neoverse N2 and V1 cpus
114 - fixed autodetection of MSA capability
117 - added an optimized DGEMM kernel
119 ====================================================================
124 - reverted unsafe TRSV/ZRSV optimizations introduced in 0.3.16
125 - fixed a potential thread race in the thread buffer reallocation routines
126 that were introduced in 0.3.18
127 - fixed miscounting of thread pool size on Linux with OMP_PROC_BIND=TRUE
128 - fixed CBLAS interfaces for CSROT/ZSROT and CROTG/ZROTG
129 - made automatic library suffix for CMAKE builds with INTERFACE64 available
133 - DYNAMIC_ARCH builds now fall back to the cpu with most similar capabilities
134 when an unknown CPUID is encountered, instead of defaulting to Prescott
135 - added cpu detection for Intel Alder Lake
136 - added cpu detection for Intel Sapphire Rapids
137 - added an optimized SBGEMM kernel for Sapphire Rapids
138 - fixed DYNAMIC_ARCH builds on OSX with CMAKE
139 - worked around DYNAMIC_ARCH builds made on Sandybridge failing on SkylakeX
140 - fixed missing thread initialization for static builds on Windows/MSVC
141 - fixed an excessive read in ZSYMV
144 - added support for POWER10 in big-endian mode
145 - added support for building with CMAKE
146 - added optimized SGEMM and DGEMM kernels for small matrix sizes
149 - added basic support and cputype detection for Fujitsu A64FX
150 - added a generic ARMV8SVE target
151 - added SVE-enabled SGEMM and DGEMM kernels for ARMV8SVE and A64FX
152 - added optimized CGEMM and ZGEMM kernels for Cortex A53 and A55 cpus
153 - fixed cpuid detection for Apple M1 and improved performance
154 - improved compiler flag setting in CMAKE builds
157 - fixed improper initialization in CSCAL/ZSCAL for strided access patterns
160 - added a GENERIC target for MIPS32
161 - added support for cross-compiling to MIPS32 on x86_64 using CMAKE
164 - fixed misdetection of MSA capability
166 ====================================================================
171 - when the build-time number of preconfigured threads is exceeded
172 at runtime (typically by an external program calling BLAS functions
173 from a larger number of threads in parallel), OpenBLAS will now
174 allocate an auxiliary control structure for up to 512 additional
175 threads instead of aborting
176 - added support for Loongson's LoongArch64 cpu architecture
177 - fixed building OpenBLAS with CMAKE and -DBUILD_BFLOAT16=ON
178 - added support for building OpenBLAS as a CMAKE subproject
179 - added support for building for Windows/ARM64 targets with clang
180 - improved support for building with the IBM xlf compiler
181 - imported Reference-LAPACK PR 625 (out-of-bounds reads in ?LARRV)
182 - imported Reference-LAPACK PR 597 for testsuite compatibility with
186 - added SkylakeX S/DGEMM kernels for small problem sizes (M*N*K<=1000000)
187 - added optimized SBGEMM for Intel Cooper Lake
188 - reinstated the performance patch for AVX512 SGEMV_T with a proper fix
189 - added a workaround for a gcc11 tree-vectorizer bug that caused spurious
190 failures in the test programs for complex BLAS3 when compiling at -O3
191 (the default for cmake "release" builds)
192 - added support for runtime cpu count detection under Haiku OS
193 - worked around a long-standing miscompilation issue of the Haswell DGEMV_T
194 kernel with gcc that could produce NaN output in some corner cases
197 - improved performance of DASUM on POWER10
200 - fixed crashes (use of reserved register x18) on Apple M1 under OSX
201 - fixed building with gcc releases earlier than 5.1
204 - fixed building under BSD
207 - fixed building under BSD
209 ====================================================================
214 - reverted the optimization of SGEMV_N/DGEMV_N for small input sizes
215 and consecutive arguments as it led to stack overflows on x86_64
216 with some operating systems (notably OSX and Windows)
219 - reverted the performance patch for SGEMV_T on AVX512 as it caused
220 wrong results in some applications
223 - fixed compilation with compilers other than gcc
224 ====================================================================
229 - drastically reduced the stack size requirements for running the LAPACK
230 testsuite (Reference-LAPACK PR 553)
231 - fixed spurious test failures in the LAPACK testsuite (Reference-LAPACK
233 - expressly setting DYNAMIC_ARCH=0 no longer enables dynamic_arch mode
234 - improved performance of xGER, xSPR, xSPR2, xSYR, xSYR2, xTRSV, SGEMV_N
235 and DGEMV_N, for small input sizes and consecutive arguments
236 - improved performance of xGETRF, xPORTF and xPOTRI for small input sizes
237 by disabling multithreading
238 - fixed installing with BSD versions of the "install" utility
241 - fixed the implementation of xIMIN
242 - improved the performance of DSDOT
243 - fixed linking of the tests on C910V with current vendor gcc
246 - fixed SBGEMM computation for some odd value inputs
247 - fixed compilation for PPCG4, PPC970, POWER3, POWER4 and POWER5
250 - improved performance of SGEMV_N and SGEMV_T for small N on AVX512-capable cpus
251 - worked around a miscompilation of ZGEMM/ZTRMM on Sandybridge with old gcc
253 - fixed compilation with MS Visual Studio versions older than 2017
254 - fixed macro name collision with winnt.h from the latest Win10 SDK
255 - added cpu type autodetection for Intel Ice Lake SP
256 - fixed cpu type autodetection for Intel Tiger Lake
257 - added cpu type autodetection for recent Centaur/Zhaoxin models
258 - fixed compilation with musl libc
261 - fixed compilation with gcc/gfortran on the Apple M1
262 - fixed linking of the tests on FreeBSD
263 - fixed missing restore of a register in the recently rewritten DNRM2 kernel
264 for ThunderX2 and Neoverse N1 that could cause spurious failures in e.g.
266 - added compiler optimization flags for the EMAG8180
267 - added initial support for Cortex A55
270 - fixed linking of the tests on FreeBSD
272 ====================================================================
277 - imported improvements and bugfixes from Reference-LAPACK 3.9.1
278 - imported LAPACKE interface fixes from Reference-LAPACK PRs 534 + 537
279 - fixed a problem in the cpu detection of 0.3.14 that prevented cross-compilation
280 - fixed a sequence problem in the generation of softlinks to the library in GMAKE
283 - fixed compilation on RISCV (missing entry in getarch)
284 - fixed a potential division by zero in CROTG and ZROTG
287 - fixed LAPACK testsuite failures seen with the NVIDIA HPC compiler
288 - improved CGEMM, DGEMM and ZGEMM performance on POWER10
289 - added an optimized ZGEMV kernel for POWER10
290 - fixed a potential division by zero in CROTG and ZROTG
293 - added support for Intel Control-flow Enforcement Technology (CET)
294 - reverted the DOMATCOPY_RT code to the generic C version
295 - fixed a bug in the AVX512 SGEMM kernel introduced in 0.3.14
296 - fixed misapplication of -msse flag to non-SSE cpus in DYNAMIC_ARCH
297 - added support for compilation of the benchmarks on older OSX versions
298 - fix propagation of the NO_AVX512 option in CMAKE builds
299 - fix compilation of the AVX512 SGEMM kernel with clang-cl on Windows
300 - fixed compilation of the CTESTs with INTERFACE64=1 (random faults on OSX)
301 - corrected the Haswell DROT kernel to require AVX2/FMA3 rather than AVX512
304 - fixed a potential division by zero in CROTG and ZROTG
305 - fixed a potential overflow in IMATCOPY/ZIMATCOPY and the CTESTs
308 - fixed spurious reads outside the array in the SGEMM tcopy macro
309 - fixed a potential division by zero in CROTG and ZROTG
310 - fixed a segmentation fault in DYNAMIC_ARCH builds (reappeared in 0.3.14)
313 - fixed a potential division by zero in CROTG and ZROTG
314 - fixed a potential overflow in IMATCOPY/ZIMATCOPY and the CTESTs
317 - fixed a potential division by zero in CROTG and ZROTG
320 - fixed a potential division by zero in CROTG and ZROTG
322 ====================================================================
327 * Fixed a race condition on thread shutdown in non-OpenMP builds
328 * Fixed custom BUFFERSIZE option getting ignored in gmake builds
329 * Fixed CMAKE compilation of the TRMM kernels for GENERIC platforms
330 * Added CBLAS interfaces for CROTG, ZROTG, CSROT and ZDROT
331 * Improved performance of OMATCOPY_RT across all platforms
332 * Changed perl scripts to use env instead of a hardcoded /usr/bin/perl
333 * Fixed potential misreading of the GCC compiler version in the build scripts
334 * Fixed convergence problems in LAPACK complex GGEV/GGES (Reference-LAPACK #477)
335 * Reduced the stacksize requirements for running the LAPACK testsuite (Reference-LAPACK #335)
338 * Fixed compilation on RISCV (missing entry in getarch)
341 * Fixed compilation for DYNAMIC_ARCH with clang and with old gcc versions
342 * Added support for compilation on FreeBSD/ppc64le
343 * Added optimized POWER10 kernels for SSCAL, DSCAL, CSCAL, ZSCAL
344 * Added optimized POWER10 kernels for SROT, DROT, CDOT, SASUM, DASUM
345 * Improved SSWAP, DSWAP, CSWAP, ZSWAP performance on POWER10
346 * Improved SCOPY and CCOPY performance on POWER10
347 * Improved SGEMM and DGEMM performance on POWER10
348 * Added support for compilation with the NVIDIA HPC compiler
351 * Added an optimized bfloat16 GEMM kernel for Cooperlake
352 * Added CPUID autodetection for Intel Rocket Lake and Tiger Lake cpus
353 * Improved the performance of SASUM,DASUM,SROT,DROT on AMD Ryzen cpus
354 * Added support for compilation with the NAG Fortran compiler
355 * Fixed recognition of the AMD AOCC compiler
356 * Fixed compilation for DYNAMIC_ARCH with clang on Windows
357 * Added support for running the BLAS/CBLAS tests on Windows
358 * Fixed signatures of the tls callback functions for Windows x64
359 * Fixed various issues with fma intrinsics support handling
362 * Added support for embedded Cortex M targets via a new option EMBEDDED
365 * Fixed the THUNDERX2T99 and NEOVERSEN1 DNRM2/ZNRM2 kernels for inputs with Inf
366 * Added support for the DYNAMIC_LIST option
367 * Added support for compilation with the NVIDIA HPC compiler
368 * Added support for compiling with the NAG Fortran compiler
370 ====================================================================
375 * Added a generic bfloat16 SBGEMV kernel
376 * Fixed a potentially severe memory leak after fork in OpenMP builds
377 that was introduced in 0.3.12
378 * Added detection of the Fujitsu Fortran compiler
379 * Added detection of the (e)gfortran compiler on OpenBSD
380 * Added support for overriding the default name of the library independently
381 from symbol suffixing in the gmake builds (already supported in cmake)
384 * Added a RISC V port optimized for C910V
387 * Added optimized POWER10 kernels for SAXPY, CAXPY, SDOT, DDOT and DGEMV_N
388 * Improved DGEMM performance on POWER10
389 * Improved STRSM and DTRSM performance on POWER9 and POWER10
390 * Fixed segmemtation faults in DYNAMIC_ARCH builds
391 * Fixed compilation with the PGI compiler
394 * Fixed compilation of kernels that require SSE2 intrinsics since 0.3.12
397 * Added an optimized bfloat16 SBGEMV kernel for SkylakeX and Cooperlake
398 * Improved the performance of SASUM and DASUM kernels through parallelization
399 * Improved the performance of SROT and DROT kernels
400 * Improved the performance of multithreaded xSYRK
401 * Fixed OpenMP builds that use the LLVM Clang compiler together with GNU gfortran
402 (where linking of both the LLVM libomp and GNU libgomp could lead to lockups or
404 * Fixed miscompilations by old gcc 4.6
405 * Fixed misdetection of AVX2 capability in some Sandybridge cpus
406 * Fixed lockups in builds combining DYNAMIC_ARCH with TARGET=GENERIC on OpenBSD
409 * Fixed segmemtation faults in DYNAMIC_ARCH builds
412 * Improved kernels for Loongson 3R3 ("3A") and 3R4 ("3B") models, including MSA
413 * Fixed bugs in the MSA kernels for CGEMM, CTRMM, CGEMV and ZGEMV
414 * Added handling of zero increments in the MSA kernels for SSWAP and DSWAP
415 * Added DYNAMIC_ARCH support for MIPS64 (currently Loongson3R3/3R4 only)
418 * Fixed building 32 and 64 bit SPARC kernels with the SolarisStudio compilers
420 ====================================================================
425 * Fixed missing BLAS/LAPACK functions (inadvertently dropped during
426 the build system restructuring)
427 * Fixed argument conversion macro in LAPACKE_zgesvdq (LAPACK #458)
430 * Added optimized SCOPY/CCOPY kernels for POWER10
431 * Increased and unified the default size of the GEMM BUFFER
432 * Fixed building for POWER10 in DYNAMIC_ARCH mode
433 * POWER10 compatibility test now checks binutils version as well
434 * Cleaned up compiler warnings
437 * corrected compiler version checks for AVX2 compatibility
438 * added compiler option -mavx2 for building with flang
439 * fixed direct SGEMM pathway for small matrix sizes (broken by
440 the code refactoring in 0.3.11)
441 * fixed unhandled partial register clobbers in several kernels
442 for AXPY,DOT,GEMV_N and GEMV_T flagged by gcc10 tree-vectorizer
445 * improved Apple Vortex support to include cross-compiling
447 ====================================================================
453 the newly added BFLOAT16 functions were renamed to use the
454 letter "B" instead of "H" to avoid potential confusion with
455 the IEEE "half precision float" type, i.e. the 0.3.10
456 SHGEMM is now SBGEMM and the corresponding build option
457 was changed from "BUILD_HALF" to "BUILD_BFLOAT16".
458 * Reduced the default BLAS3_MEM_ALLOC_THRESHOLD (used as an upper
459 limit for placing temporary arrays on the stack) to be compatible
460 with a stack size of 1mb (as imposed by the JAVA runtime library)
461 * Added mixed-precision dot function SBDOT and utility functions
462 shstobf16, shdtobf16, sbf16tos and dbf16tod to convert between
463 single or double precision float arrays and bfloat16 arrays
464 * Fixed prototypes of LAPACK_?ggsvp and LAPACK_?ggsvd functions
466 * Fixed underflow and rounding errors in LAPACK SLANV2 and DLANV2
467 (causing miscalculations in e.g. SHSEQR/DHSEQR, LAPACK issue #263)
468 * Fixed workspace calculation in LAPACK ?GELQ (LAPACK issue #415)
469 * Fixed several bugs in the LAPACK testsuite
470 * Improved performance of TRMM and TRSM for certain problem sizes
471 * Fixed infinite recursions and workspace miscalculations in ReLAPACK
472 * CMAKE builds no longer require pkg-config for creating the .pc file
473 * Makefile builds no longer misread NO_CBLAS=0 or NO_LAPACK=0 as
474 enabling these options
475 * Fixed detection of gfortran when invoked through an mpi wrapper
476 * Improve thread reinitialization performance with OpenMP after a fork
477 * Added support for building only the subset of the library required
478 for a particular precision by specifying BUILD_SINGLE, BUILD_DOUBLE
479 * Optional function name prefixes and suffixes are now correctly
480 reflected in the generated cblas.h
481 * Added CMAKE build support for the LAPACK and multithreading tests
484 * Added optimized support for POWER10
485 * Added support for compiling for POWER8 in 32bit mode
486 * Added support for compilation with LLVM/clang
487 * Added support for compilation with NVIDIA/PGI compilers
488 * Fixed building on big-endian POWER8
489 * Fixed miscompilation of ZDOTC by gcc10
490 * Fixed alignment errors in the POWER8 SAXPY kernel
491 * Improved CPU detection on AIX
492 * Supported building with older compilers on POWER9
495 * Added support for Intel Cooperlake
496 * Added autodetection of AMD Renoir/Matisse/Zen3 cpus
497 * Added autodetection of Intel Comet Lake cpus
498 * Reimplemented ?sum, ?dot and daxpy using universal intrinsics
499 * Reset the fpu state before using the fpu on Windows as a workaround
500 for a problem introduced in Windows 10 build 19041 (a.k.a. SDK 2004)
501 * Fixed potentially undefined behaviour in the dot and gemv_t kernels
502 * Fixed a potential segmentation fault in DYNAMIC_ARCH builds
503 * Fixed building for ZEN with PGI/NVIDIA and AMD AOCC compilers
506 * Fixed cpu detection on BSD-like systems
509 * Added preliminary support for Apple Vortex cpus
510 * Added support for the Cavium ThunderX3T110 cpu
511 * Fixed cpu detection on BSD-like systems
512 * Fixed compilation in -std=C18 mode
515 * Added support for compiling with the clang compiler
516 * Improved GEMM performance on Z14
518 ====================================================================
523 * Improved thread locking behaviour in blas_server and parallel getrf
524 * Imported bugfix 394 from LAPACK (spurious reference to "XERBL"
525 due to overlong lines)
526 * Imported bugfix 403 from LAPACK (compile option "recursive" required
527 for correctness with Intel and PGI)
528 * Imported bugfix 408 from LAPACK (wrong scaling in ZHEEQUB)
529 * Imported bugfix 411 from LAPACK (infinite loop in LARGV/LARTG/LARTGP)
530 * Fixed mismatches between BUFFERSIZE and GEMM_UNROLL parameters that
531 could lead to crashes at large matrix sizes
532 * Restored internal soname in dynamic libraries on FreeBSD and Dragonfly
533 * Added API (openblas_setaffinity) to set the thread affinity on Linux
534 * Added initial infrastructure for half-precision floating point
535 (bfloat16) support with a generic implementation of SHGEMM
536 * Added CMAKE build system support for building the cblas_Xgemm3m
538 * Fixed CMAKE support for building in a path with embedded spaces
539 * Fixed CMAKE (non)handling of NO_EXPRECISION and MAX_STACK_ALLOC
540 * Fixed GCC version detection in the Makefiles
541 * Allowed overriding the names of AR, AS and LD in Makefile builds
544 * Fixed big-endian POWER8 ELFv2 builds on FreeBSD
545 * Fixed GCC version checks and DYNAMIC_ARCH builds on POWER9
546 * Fixed CMAKE build support for POWER9
547 * fixed a potential race condition in the thread buffer allocation
548 * Worked around LAPACK test failures on PPC G4
551 * Fixed a potential race condition in the thread buffer allocation
552 * Added support for MIPS 24K/24KE family based on P5600 kernels
555 * fixed a potential race condition in the thread buffer allocation
556 * Added TARGET=GENERIC
559 * Fixed a race condition in the thread buffer allocation
562 * Fixed a race condition in the thread buffer allocation
563 * Fixed zero initialisation in the assembly for SGEMM and DGEMM BETA
564 * Improved performance of the ThunderX2 DAXPY kernel
565 * Added an optimized SGEMM kernel for Cortex A53
566 * Fixed Makefile support for INTERFACE64 (8-byte integer)
569 * Fixed a syntax error in the CMAKE setup for SkylakeX
570 * Improved performance of STRSM on Haswell, SkylakeX and Ryzen
571 * Improved SGEMM performance on SGEMM for workloads with ldc a
573 * Improved DGEMM performance on Skylake X
574 * Fixed unwanted AVX512-dependency of SGEMM in DYNAMIC_ARCH
575 builds created on SkylakeX
576 * Removed data alignment requirement in the SSE2 copy kernels
577 that could cause spurious crashes
578 * Added a workaround for an optimizer bug in AppleClang 11.0.3
579 * Fixed LAPACK test failures due to wrong options for Intel Fortran
580 * Fixed compilation and LAPACK test results with recent Flang
582 * Fixed DYNAMIC_ARCH builds with CMAKE on OS X
583 * Fixed missing exports of cblas_i?amin, cblas_i?min, cblas_i?max,
584 cblas_?sum, cblas_?gemm3m in the shared library on OS
585 * Fixed reporting of cpu name in DYNAMIC_ARCH builds (would sometimes
586 show the name of an older generation chip supported by the same kernels)
589 * Improved performance of SGEMM/STRMM and DGEMM/DTRMM on Z14
591 ====================================================================
596 * Fixed a miscompilation of the GETRF functions with CMAKE
597 * Imported bugfix 390 from LAPACK (missing NaN propagation in xCOMBSSQ)
598 * The size of the memory buffer used for splitting GEMM tasks across
599 multiple threads can now be configured in the build system.
602 * Fixed several compilation problems related to endianness
603 and ELF version on POWER8 and POWER9
604 * Fixed use of the absolute value IAMIN/IAMAX instead of IMIN/IMAX
605 * Fixed a race condition in the level3 blas code
608 * Fixed use of the absoltute value IAMIN/IAMAX instead of IMIN/IMAX
611 * Fixed a race condition in the level3 blas code
612 * Fixed compilation on Android
614 * Added support for Ampere EMAG8180
615 * Added support for Neoverse N1
616 * Improved performance of the blas_lock function
617 * Fixed a race condition in the level3 blas code
618 * Fixed a performance regression on TSV110-based servers
621 * Fixed a long-standing error with undeclared register overwrites
622 in the DSCAL microkernel for HASWELL,SKYLAKEX and ZEN
623 * Fixed a long-standing bug in the SSE implementation of IAMAX
624 * Fixed a CMAKE build failure with DYNAMIC_ARCH
625 * Fixed cpu autodetection of Goldmont+, Cannon Lake and Ice Lake
626 * Fixed a compilation failure on OSX with compiler name containing dash
627 * Fixed compilation with MinGW on SkylakeX
628 * Improved speed of the AVX512 GEMM3M kernel on SkylakeX
629 * Added an AVX512 STRMM kernel for SkylakeX
630 * Improved GEMM performance on Haswell and Zen
633 * fixed compilation of the DYNAMIC_ARCH code
635 ====================================================================
640 ` * LAPACK has been updated to 3.9.0 (plus patches up to
642 * CMAKE support has been improved in several areas including
644 * a thread race condition in the GEMM3M kernels was resolved
645 * the "generic" (plain C) gemm beta kernel used by many targets
647 * an optimized version of the LAPACK trtrs functions has been added
648 * an incompatibilty between the LAPACK tests and the OpenBLAS
649 implementation of XERBLA was resolved, removing the numerous
650 warnings about wrong error exits in the former
651 * support for NetBSD has been added
652 * support for compilation with g95 and non-GNU versions of ld
654 * support for compilation with (upcoming) gcc 10 has been added
657 * worked around miscompilation of several POWER8 and POWER9
658 kernels by older versions of gcc
659 * added support for big-endian POWER8 and for compilation on AIX
660 * corrected bugs in the big-endian support for PPC440 and PPC970
661 * DYNAMIC_ARCH support is now available in CMAKE builds as well
664 * performance of DGEMM_BETA and SGEMM_NCOPY has been improved
665 * compilation for 32bit works again
666 * performance of the RPCC function has been improved
667 * improved performance on small systems
668 * DYNAMIC_ARCH support is now available in CMAKE builds as well
669 * cross-compilation from OSX to IOS was simplified
672 * a new AVX512 DGEMM kernel was added and the AVX512 SGEMM kernel
673 was significantly improved
674 * optimized AVX512 kernels for CGEMM and ZGEMM have been added
675 * AVX2 kernels for STRMM, SGEMM, and CGEMM have been significantly
676 sped up and optimized CGEMM3M and ZGEMM3M kernels have been added
677 * added support for QEMU virtual cpus
678 * a compilation problem with PGI and SUN compilers was fixed
679 * Intel "Goldmont plus" is now autodetected
680 * a potential crash on program exit on MS Windows has been fixed
683 * an unwanted case sensitivity in the implementation of LSAME
684 on older 32bit AMD cpus was fixed
687 * Z15 is now supported as Z14
688 * DYNAMIC_ARCH is now available on ZARCH as well
690 ====================================================================
695 * having the gmake special variables TARGET_ARCH or TARGET_MACH
696 defined no longer causes build failures in ctest or utest
697 * defining NO_AFFINITY or USE_TLS to 0 in gmake builds no longer
698 has the same effect as setting them to 1
699 * a new test program was added to allow checking the library for
701 * a new option USE_LOCKING was added to ensure thread safety when
702 OpenBLAS itself is built without multithreading but will be
703 called from multiple threads.
704 * a build failure on Linux with glibc versions earlier than 2.5
706 * a runtime error with CPU enumeration (and NO_AFFINITY not set)
707 on glibc 2.6 was fixed
708 * NO_AFFINITY was added to the CMAKE options (and defaults to being
709 active on Linux, as in the gmake builds)
712 * the build-time logic for detection of AVX512 availability in
713 the processor and compiler was fixed
714 * gmake builds on OSX now set the internal name of the library to
715 libopenblas.0.dylib (consistent with CMAKE)
716 * the Haswell DGEMM kernel received a significant speedup through
717 improved prefetch and load instructions
718 * performance of DGEMM, DTRMM, DTRSM and ZDOT on Zen/Zen2 was markedly
719 increased by avoiding vpermpd instructions
720 * the SKYLAKEX (AVX512) DGEMM helper functions have now been disabled
721 to fix remaining errors in DGEMM, DSYMM and DTRMM
724 * added support for building on FreeBSD/powerpc64 and FreeBSD/ppc970
725 * added optimized kernels for POWER9 SGEMM and STRMM
728 * fixed the softfp implementations of xAMAX and IxAMAX
729 * removed the predefined -march= flags on both ARMV5 and ARMV6 as
730 they were appropriate for only a subset of platforms
732 ====================================================================
737 * the build tools now check that a given cpu TARGET is actually valid
738 * the build-time check of system features (c_check) has been made
739 less dependent on particular perl features (this should mainly
740 benefit building on Windows)
741 * several problem with the ReLAPACK integration were fixed,
742 including INTERFACE64 support and building a shared library
743 * building with CMAKE on BSD systems was improved
744 * a non-absolute SUM function was added based on the
745 existing optimized code for ASUM
746 * CBLAS interfaces to the IxMIN and IxMAX functions were added
747 * a name clash between LAPACKE and BOOST headers was resolved
748 * CMAKE builds with OpenMP failed to include the appropriate getrf_parallel
750 * a crash on thread (key) deletion with the USE_TLS=1 memory management
752 * restored several earlier fixes, in particular for OpenMP performance,
753 building on BSD, and calling fork on CYGWIN, which had inadvertently
754 been dropped in the 0.3.3 rewrite of the memory management code.
757 * the AVX512 DGEMM kernel has been disabled again due to unsolved problems
758 * building with old versions of MSVC was fixed
759 * it is now possible to build a static library on Windows with CMAKE
760 * accessing environment variables on CYGWIN at run time was fixed
761 * the CMAKE build system now recognizes 32bit userspace on 64bit hardware
762 * Intel "Denverton" atom and Hygon "Dhyana" zen CPUs are now autodetected
763 * building for DYNAMIC_ARCH with a DYNAMIC_LIST of targets is now supported
765 * building for DYNAMIC_ARCH with GENERIC as the default target is now supported
766 * a buffer overflow in the SSE GEMM kernel for Intel Nano targets was fixed
767 * assembly bugs involving undeclared modification of input operands were fixed
768 in the AXPY, DOT, GEMV, GER, SCAL, SYMV and TRSM microkernels for Nehalem,
769 Sandybridge, Haswell, Bulldozer and Piledriver. These would typically cause
770 test failures or segfaults when compiled with recent versions of gcc from 8 onward.
771 * a similar bug was fixed in the blas_quickdivide code used to split workloads
773 * a bug in the IxMIN implementation for the GENERIC target made it return the result of IxMAX
774 * fixed building on SkylakeX systems when either the compiler or the (emulated) operating
775 environment does not support AVX512
776 * improved GEMM performance on ZEN targets
779 * build failures caused by the recently added checks for AVX512 were fixed
780 * an inline assembly bug involving undeclared modification of an input argument was
781 fixed in the blas_quickdivide code used to split workloads in most functions
782 * a bug in the IMIN implementation for the GENERIC target made it return the result of IMAX
785 * a bug in the IMIN implementation made it return the result of IMAX
788 * single precision BLAS1/2 functions have received optimized POWER8 kernels
789 * POWER9 is now a separate target, with an optimized DGEMM/DTRMM kernel
790 * building on PPC970 systems under OSX Leopard or Tiger is now supported
791 * out-of-bounds memory accesses in the gemm_beta microkernels were fixed
792 * building a shared library on AIX is now supported for POWER6
793 * DYNAMIC_ARCH support has been added for POWER6 and newer
796 * corrected xDOT behaviour with zero INC_X or INC_Y
797 * a bug in the IMIN implementation made it return the result of IMAX
800 * added support for HiSilicon TSV110 cpus
801 * the CMAKE build system now recognizes 32bit userspace on 64bit hardware
802 * cross-compilation with CMAKE now works again
803 * a bug in the IMIN implementation made it return the result of IMAX
804 * ARMV8 builds with the BINARY=32 option are now automatically handled as ARMV7
807 * optimized microkernels for single precicion BLAS1/2 functions have been added
810 ====================================================================
815 * loop unrolling in TRMV has been enabled again.
816 * A domain error in the thread workload distribution for SYRK
818 * gmake builds will now automatically add -fPIC to the build
819 options if the platform requires it.
820 * a pthreads key leakage (and associate crash on dlclose) in
821 the USE_TLS codepath was fixed.
822 * building of the utest cases on systems that do not provide
823 an implementation of complex.h was fixed.
826 * the SkylakeX code was changed to compile on OSX.
827 * unwanted application of the -march=skylake-avx512 option
828 to the common code parts of a DYNAMIC_ARCH build was fixed.
829 * improved performance of SGEMM for small workloads on Skylake X.
830 * performance of SGEMM and DGEMM was improved on Haswell.
833 * a configuration error that broke the CNRM2 kernel was corrected.
834 * compilation of the GEMM kernels with CMAKE was fixed.
835 * DYNAMIC_ARCH builds are now available with CMAKE as well.
836 * using CMAKE for cross-compilation to the new cpu TARGETs
837 introduced in 0.3.4 now works.
840 * a problem in cpu autodetection for AIX has been corrected.
842 ====================================================================
847 * the new, experimental thread-local memory allocation had
848 inadvertently been left enabled for gmake builds in 0.3.3
849 despite the announcement. It is now disabled by default, and
850 single-threaded builds will keep using the old allocator even
851 if the USE_TLS option is turned on.
852 * OpenBLAS will now provide enough buffer space for at least 50
854 * The output of openblas_get_config() now contains the version
856 * A serious thread safety bug in GEMV operation with small M and
857 large N size has been fixed.
858 * The code will now automatically call blas_thread_init after a
859 fork if needed before handling a call to openblas_set_num_threads
860 * Accesses to parallelized level3 functions from multiple callers
861 are now serialized to avoid thread races (unless using OpenMP).
862 This should provide better performance than the known-threadsafe
863 (but non-default) USE_SIMPLE_THREADED_LEVEL3 option.
864 * When building LAPACK with gfortran, -frecursive is now (again)
865 enabled by default to ensure correct behaviour.
866 * The OpenBLAS version cblas.h now supports both CBLAS_ORDER and
867 CBLAS_LAYOUT as the name of the matrix row/column order option.
868 * Externally set LDFLAGS are now passed through to the final compile/link
869 steps to facilitate setting platform-specific linker flags.
870 * A potential race condition during the build of LAPACK (that would
871 usually manifest itself as a failure to build TESTING/MATGEN) has been
873 * xHEMV has been changed to stay single-threaded for small input sizes
874 where the overhead of multithreading exceeds any possible gains
875 * CSWAP and ZSWAP have been limited to a single thread except on ARMV8 or
876 ThunderX hardware with sizable input.
877 * Linker flags for the PGI compiler have been updated
878 * Behaviour of AXPY with zero increments is now handled in the C interface,
879 correcting the result on at least Intel Atom.
880 * The result matrix from calling SGELSS with an all-zero input matrix is
881 now zeroed completely.
884 * Autodetection of AMD Ryzen2 has been fixed (again).
885 * CMAKE builds now support labeling of an INTERFACE64=1 build of
886 the library with the _64 suffix.
887 * AVX512 version of DGEMM has been added and the AVX512 SGEMM kernel
888 has been sped up by rewriting with C intrinsics
889 * Fixed compilation on RHEL5/CENTOS5 (issue with typename __WAIT_STATUS)
892 * added support for building on AIX (with gcc and GNU tools from AIX Toolbox).
893 * CPU type detection has been implemented for AIX.
894 * CPU type detection has been fixed for NETBSD.
897 * AXPY on LOONGSON3A has been corrected to pass "zero increment" utest.
898 * DSDOT on LOONGSON3A has been fixed.
899 * the SGEMM microkernel has been hardened against potential data loss.
902 * DYNAMic_ARCH support is now available for 64bit ARM
903 * cross-compiling for ARMV8 under iOS now works.
904 * cpu-specific code has been rearranged to make better use of both
905 hardware commonalities and model-specific compiler optimizations.
906 * XGENE1 has been removed as a TARGET, superseded by the improved generic
910 * Older assembly mnemonics have been converted to UAL form to allow
911 building with clang 7.0
912 * Cross compiling LAPACKE for Android has been fixed again (broken by
913 update to LAPACK 3.7.0 some while ago).
915 ====================================================================
920 * thread memory allocation has been switched back to the method
921 used before version 0.3.1 due to unexpected problems caused by
922 the new code under some circumstances. A new compile-time option
923 USE_TLS has been added to enable the new code, and it is hoped
924 that this can become the default again in the next version.
925 * LAPAck PR272 has been integrated, which fixes spurious errors
926 in DSYEVR and related functions caused by missing conversion
927 from ILAENV to ILAENV_2STAGE in several _2stage routines.
928 * the cmake-generated OpenBLASConfig.cmake now uses correct case
929 for the name of the library
930 * added support for Haiku OS
933 * added AVX512 implementations of SDOT, DDOT, SAXPY, DAXPY,
934 DSCAL, DGEMVN and DSYMVL
935 * added a workaround for a cygwin issue that prevented compilation
939 * added autodetection of Z14
940 * fixed TRMM errors in the generic target
942 ====================================================================
947 * fixes for regressions caused by the rewrite of the thread
948 initialization code in 0.3.1
951 * fixed cpu autodetection for the BSDs
954 * fixed utest errors in AXPY, DSDOT, ROT and SWAP
957 * added autodetection of AMD Ryzen 2
958 * fixed build with older versions of MSVC
960 ====================================================================
965 * rewritten thread initialization code with significantly reduced overhead
966 * added CBLAS interfaces to the IxAMIN BLAS extension functions
967 * fixed the lapack-test target
968 * CMAKE builds now create an OpenBLASConfig.cmake file
969 * ZAXPY now uses a single thread for small input sizes
970 * the LAPACK code was updated from Reference-LAPACK/lapack#253
971 (fixing LAPACKE interfaces to Aasen's functions)
974 * corrected CROT and ZROT behaviour with zero INC_X
977 * corrected xDOT behaviour with zero INC_X or INC_Y
980 * retired some older targets of DYNAMIC_ARCH builds to a new option DYNAMIC_OLDER,
981 this affects PENRYN,DUNNINGTON,OPTERON,OPTERON_SSE3,BOBCAT,ATOM and NANO
982 (which will still be supported via the slower PRESCOTT kernels when this option is not set)
983 * added an option DYNAMIC_LIST that (used in conjunction with DYNAMIC_ARCH) allows to
984 specify the list of x86_64 targets to include. Any target not on the list will be supported
985 by the Sandybridge or Nehalem kernels if available, or by Prescott.
986 * improved SWITCH_RATIO on Haswell for increased GEMM throughput
987 * added initial support for Intel Skylake X, including an AVX512 SGEMM kernel
988 * added autodetection of Intel Cannon Lake series as Skylake X
989 * added a default L2 cache size for hypervisors that return zero here (Chromebook)
990 * fixed a name clash with recent Windows10 headers that broke the build with (at least)
991 recent mingw from MSYS2
992 * fixed a link error in mixed clang/gfortran builds with OpenMP
993 * updated the OSX deployment target to 10.8
994 * switched on parallel make for builds on MS Windows by default
997 * fixed SSWAP and DSWAP behaviour with zero INC_X and INC_Y
999 ====================================================================
1004 * fixed some more thread race and locking bugs
1005 * added preliminary support for calling an OpenMP build of the library from multiple threads
1006 * removed performance impact of thread locks added in 0.2.20 on OpenMP code
1007 * general code cleanup
1008 * optimized DSDOT implementation
1009 * improved thread distribution for GEMM
1010 * corrected IMATCOPY/OMATCOPY implementation
1011 * fixed out-of-bounds accesses in the multithreaded xBMV/xPMV and SYMV implementations
1012 * cmake build improvements
1013 * pkgconfig file now contains build options
1014 * openblas_get_config() now reports USE_OPENMP and NUM_THREADS settings used for the build
1015 * corrections and improvements for systems with more than 64 cpus
1016 * LAPACK code updated to 3.8.0 including later fixes
1017 * added ReLAPACK, a recursive implementation of several LAPACK functions
1018 * Rewrote ROTMG to handle cases that the netlib code failed to address
1019 * Disabled (broken) multithreading code for xTRMV
1020 * corrected prototypes of complex CBLAS functions to make our cblas.h match the generally accepted standard
1021 * shared memory access failures on startup are now handled more gracefully
1022 * restored utests from earlier releases (and made them pass on all affected systems)
1025 * several fixes for cpu autodetection
1028 * corrected vector register overwriting in several Power8 kernels
1029 * optimized additional BLAS functions
1032 * added support for CortexA53 and A72
1033 * added autodetection for ThunderX2T99
1034 * made most optimized kernels the default for generic ARMv8 targets
1037 * parallelized DDOT kernel for Haswell
1038 * changed alignment directives in assembly kernels to boost performance on OSX
1039 * fixed register handling in the GEMV microkernels (bug exposed by gcc7)
1040 * added support for building on OpenBSD and Dragonfly
1041 * updated compiler options to work with Intel release 2018
1042 * support fully optimized build with clang/flang on Microsoft Windows
1043 * fixed building on AIX
1046 * added optimized BLAS 1/2 functions
1049 * fixed cpu autodetection helper code
1050 * added mips32 1004K cpu (Mediatek MT7621 and similar SoC)
1051 * added mips64 I6500 cpu
1053 ====================================================================
1058 * Improved CMake support
1059 * Fixed several thread race and locking bugs
1060 * Fixed default LAPACK optimization level
1061 * Updated LAPACK to 3.7.0
1062 * Added ReLAPACK (https://github.com/HPAC/ReLAPACK, make BUILD_RELAPACK=1)
1065 * Optimizations for Power9
1066 * Fixed several Power8 assembly bugs
1069 * New optimized Vulcan and ThunderX2T99 targets
1070 * Support for ARMV7 SOFT_FP ABI (make ARM_SOFTFP_ABI=1)
1071 * Detect all cpu cores including offline ones
1072 * Fix compilation with CLANG
1073 * Support building a shared library for Android
1076 * Fixed several threading issues
1077 * Fix compilation with CLANG
1080 * Detect Intel Bay Trail and Apollo Lake
1081 * Detect Intel Sky Lake and Kaby Lake
1082 * Detect Intel Knights Landing
1083 * Detect AMD A8, A10, A12 and Ryzen
1084 * Support 64bit builds with Visual Studio
1085 * Fix building with Intel and PGI compilers
1086 * Fix building with MINGW and TDM-GCC
1087 * Fix cmake builds for Haswell and related cpus
1088 * Fix building for Sandybridge with CLANG 3.9
1089 * Add support for the FLANG compiler
1092 * New target z13 with BLAS3 optimizations
1094 ====================================================================
1098 * Improved cross compiling.
1099 * Fix the bug on musl libc.
1102 * Optimize BLAS on Power8
1103 * Fixed Julia+OpenBLAS bugs on Power8
1106 * Optimize BLAS on MIPS P5600 and I6400 (Thanks, Shivraj Patil, Kaustubh Raste)
1109 * Improved on ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
1112 ====================================================================
1116 * If you set MAKE_NB_JOBS flag less or equal than zero,
1117 make will be without -j.
1120 * Support building Visual Studio static library. (#813, Thanks, theoractice)
1121 * Fix bugs to pass buidbot CI tests (http://build.openblas.net)
1124 * Provide DGEMM 8x4 kernel for Cortex-A57 (Thanks, Ashwin Sekhar T K)
1127 * Optimize S and C BLAS3 on Power8
1128 * Optimize BLAS2/1 on Power8
1130 ====================================================================
1134 * Enable BUILD_LAPACK_DEPRECATED=1 by default.
1136 ====================================================================
1140 * Avoid potential getenv segfault. (#716)
1141 * Import LAPACK svn bugfix #142-#147,#150-#155
1144 * Optimize c/zgemv for AMD Bulldozer, Piledriver, Steamroller
1145 * Fix bug with scipy linalg test.
1148 * Improve DGEMM for ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
1151 * Optimize D and Z BLAS3 functions for Power8.
1153 ====================================================================
1157 * Upgrade LAPACK to 3.6.0 version.
1158 Add BUILD_LAPACK_DEPRECATED option in Makefile.rule to build
1159 LAPACK deprecated functions.
1160 * Add MAKE_NB_JOBS option in Makefile.
1161 Force number of make jobs.This is particularly
1162 useful when using distcc. (#735. Thanks, Jerome Robert.)
1163 * Redesign unit test. Run unit/regression test at every build (Travis-CI and Appveyor).
1164 * Disable multi-threading for small size swap and ger. (#744. Thanks, Jerome Robert)
1165 * Improve small zger, zgemv, ztrmv using stack alloction (#727. Thanks, Jerome Robert)
1166 * Let openblas_get_num_threads return the number of active threads.
1167 (#760. Thanks, Jerome Robert)
1168 * Support illumos(OmniOS). (#749. Thanks, Lauri Tirkkonen)
1169 * Fix LAPACK Dormbr, Dormlq bug. (#711, #713. Thanks, Brendan Tracey)
1170 * Update scipy benchmark script. (#745. Thanks, John Kirkham)
1173 * Optimize trsm kernels for AMD Bulldozer, Piledriver, Steamroller.
1174 * Detect Intel Avoton.
1175 * Detect AMD Trinity, Richland, E2-3200.
1176 * Fix gemv performance bug on Mac OSX Intel Haswell.
1177 * Fix some bugs with CMake and Visual Studio
1180 * Support and optimize Cortex-A57 AArch64.
1181 (#686. Thanks, Ashwin Sekhar TK)
1182 * Fix Android build on ARMV7 (#778. Thanks, Paul Mustiere)
1183 * Update ARMV6 kernels.
1186 * Fix detection of POWER architecture
1187 (#684. Thanks, Sebastien Villemot)
1189 ====================================================================
1193 * Support cmake on x86/x86-64. Natively compiling on MS Visual Studio.
1194 (experimental. Thank Hank Anderson for the initial cmake porting work.)
1196 On Linux and Mac OSX, OpenBLAS cmake supports assembly kernels.
1199 make test (Optional)
1201 On Windows MS Visual Studio, OpenBLAS cmake only support C kernels.
1202 (OpenBLAS uses AT&T style assembly, which is not supported by MSVC.)
1203 e.g. cmake -G "Visual Studio 12 Win64" .
1204 Open OpenBLAS.sln and build.
1206 * Enable MAX_STACK_ALLOC flags by default.
1207 Improve ger and gemv for small matrices.
1208 * Improve gemv parallel with small m and large n case.
1209 * Improve ?imatcopy when lda==ldb (#633. Thanks, Martin Koehler)
1210 * Add vecLib benchmarks (#565. Thanks, Andreas Noack.)
1211 * Fix LAPACK lantr for row major matrices (#634. Thanks, Dan Kortschak)
1212 * Fix LAPACKE lansy (#640. Thanks, Dan Kortschak)
1213 * Import bug fixes for LAPACKE s/dormlq, c/zunmlq
1214 * Raise the signal when pthread_create fails (#668. Thanks, James K. Lowden)
1215 * Remove g77 from compiler list.
1216 * Enable AppVeyor Windows CI.
1219 * Support pure C generic kernels for x86/x86-64.
1220 * Support Intel Boardwell and Skylake by Haswell kernels.
1221 * Support AMD Excavator by Steamroller kernels.
1222 * Optimize s/d/c/zdot for Intel SandyBridge and Haswell.
1223 * Optimize s/d/c/zdot for AMD Piledriver and Steamroller.
1224 * Optimize s/d/c/zapxy for Intel SandyBridge and Haswell.
1225 * Optimize s/d/c/zapxy for AMD Piledriver and Steamroller.
1226 * Optimize d/c/zscal for Intel Haswell, dscal for Intel SandyBridge.
1227 * Optimize d/c/zscal for AMD Bulldozer, Piledriver and Steamroller.
1228 * Optimize s/dger for Intel SandyBridge.
1229 * Optimize s/dsymv for Intel SandyBridge.
1230 * Optimize ssymv for Intel Haswell.
1231 * Optimize dgemv for Intel Nehalem and Haswell.
1232 * Optimize dtrmm for Intel Haswell.
1235 * Support Android NDK armeabi-v7a-hard ABI (-mfloat-abi=hard)
1236 e.g. make HOSTCC=gcc CC=arm-linux-androideabi-gcc NO_LAPACK=1 TARGET=ARMV7
1237 * Fix lock, rpcc bugs (#616, #617. Thanks, Grazvydas Ignotas)
1239 * Support ppc64le platform (ELF ABI v2. #612. Thanks, Matthew Brandyberry.)
1240 * Support POWER7/8 by POWER6 kernels. (#612. Thanks, Fábio Perez.)
1242 ====================================================================
1246 * Improve OpenBLASConfig.cmake. (#474, #475. Thanks, xantares.)
1247 * Improve ger and gemv for small matrices by stack allocation.
1248 e.g. make -DMAX_STACK_ALLOC=2048 (#482. Thanks, Jerome Robert.)
1249 * Introduce openblas_get_num_threads and openblas_get_num_procs.
1250 (#497. Thanks, Erik Schnetter.)
1251 * Add ATLAS-style ?geadd function. (#509. Thanks, Martin Köhler.)
1252 * Fix c/zsyr bug with negative incx. (#492.)
1253 * Fix race condition during shutdown causing a crash in
1254 gotoblas_set_affinity(). (#508. Thanks, Ton van den Heuvel.)
1257 * Support AMD Streamroller.
1260 * Add Cortex-A9 and Cortex-A15 targets.
1262 ====================================================================
1266 * Add SYMBOLPREFIX and SYMBOLSUFFIX makefile options
1267 for adding a prefix or suffix to all exported symbol names
1268 in the shared library.(#459, Thanks Tony Kelman)
1269 * Provide OpenBLASConfig.cmake at installation.
1270 * Fix Fortran compiler detection on FreeBSD.
1271 (#470, Thanks Mike Nolta)
1275 * Add generic kernel files for x86-64. make TARGET=GENERIC
1276 * Fix a bug of sgemm kernel on Intel Sandy Bridge.
1277 * Fix c_check bug on some amd64 systems. (#471, Thanks Mike Nolta)
1280 * Support APM's X-Gene 1 AArch64 processors.
1281 Optimize trmm and sgemm. (#465, Thanks Dave Nuechterlein)
1283 ====================================================================
1287 * Added CBLAS interface for ?omatcopy and ?imatcopy.
1288 * Enable ?gemm3m functions.
1289 * Added benchmark for ?gemm3m.
1290 * Optimized multithreading lower limits.
1291 * Disabled SYMM3M and HEMM3M functions
1292 because of segment violations.
1295 * Improved axpy and symv performance on AMD Bulldozer.
1296 * Improved gemv performance on modern Intel and AMD CPUs.
1298 ====================================================================
1302 * Added some benchmark codes.
1303 * Fix link error on Linux/musl.(Thanks Isaac Dunham)
1306 * Improved s/c/zgemm performance for Intel Haswell.
1307 * Improved s/d/c/zgemv performance.
1308 * Support the big numa machine.(EXPERIMENT)
1311 * Fix detection when cpuinfo uses "Processor". (Thanks Isaiah)
1313 ====================================================================
1317 * Added BLAS extensions as following.
1318 s/d/c/zaxpby, s/d/c/zimatcopy, s/d/c/zomatcopy.
1319 * Added OPENBLAS_CORETYPE environment for dynamic_arch. (a86d34)
1320 * Added NO_AVX2 flag for old binutils. (#401)
1321 * Support outputing the CPU corename on runtime.(#407)
1322 * Patched LAPACK to fix bug 114, 117, 118.
1323 (http://www.netlib.org/lapack/bug_list.html)
1324 * Disabled ?gemm3m for a work-around fix. (#400)
1326 * Fixed lots of bugs for optimized kernels on sandybridge,Haswell,
1327 bulldozer, and piledriver.
1328 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
1331 * Improved LAPACK testing.
1333 ====================================================================
1337 * Improved the result for LAPACK testing. (#372)
1338 * Installed DLL to prefix/bin instead of prefix/lib. (#366)
1339 * Build import library on Windows.(#374)
1341 * To improve LAPACK testing, we fallback some kernels. (#372)
1342 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
1344 ====================================================================
1348 * Added OPENBLAS_VERBOSE environment variable.(#338)
1349 * Make OpenBLAS thread-pool resilient to fork via pthread_atfork.
1350 (#294, Thank Olivier Grisel)
1354 * Detect Intel Haswell for new Macbook.
1356 ====================================================================
1360 * Update LAPACK to 3.5.0 version
1361 * Fixed compatiable issues with Clang and Pathscale compilers.
1364 * Optimization on Intel Haswell.
1365 * Enable optimization kernels on AMD Bulldozer and Piledriver.
1368 * Support ARMv6 and ARMv7 ISA.
1369 * Optimization on ARM Cortex-A9.
1371 ====================================================================
1375 * Support Open64 5.0. (#266)
1376 * Add executable stack markings. (#262, Thank Sébastien Fabbro)
1377 * Respect user's LDFLAGS (Thank Sébastien Fabbro)
1380 * Rollback bulldozer and piledriver kernels to barcelona kernels (#263)
1381 We will fix the compuational error bug in bulldozer and piledriver kernels.
1383 ====================================================================
1387 * Support LSB (Linux Standard Base) 4.1.
1389 * Include LAPACK 3.4.2 source codes to the repo.
1390 Avoid downloading at compile time.
1391 * Add NO_PARALLEL_MAKE flag to disable parallel make.
1392 * Create openblas_get_parallel to retrieve information which
1393 parallelization model is used by OpenBLAS. (Thank grisuthedragon)
1394 * Detect LLVM/Clang compiler. The default compiler is Clang on Mac OS X.
1395 * Change LIBSUFFIX from .lib to .a on windows.
1396 * A work-around for dtrti_U single thread bug. Replace it with LAPACK codes. (#191)
1399 * Optimize c/zgemm, trsm, dgemv_n, ddot, daxpy, dcopy on
1400 AMD Bulldozer. (Thank Werner Saar)
1401 * Add Intel Haswell support (using Sandybridge optimizations).
1403 * Add AMD Piledriver support (using Bulldozer optimizations).
1404 * Fix the computational error in zgemm avx kernel on
1406 * Fix the overflow bug in gemv.
1407 * Fix the overflow bug in multi-threaded BLAS3, getrf when NUM_THREADS
1408 is very large.(#214, #221, #246).
1410 * Support loongcc (Open64 based) compiler for ICT Loongson 3A/B.
1413 * Support Power7 by old Power6 kernels. (#220)
1415 ====================================================================
1419 * Improved OpenMP performance slightly. (d744c9)
1420 * Improved cblas.h compatibility with Intel MKL.(#185)
1421 * Fixed the overflowing bug in single thread cholesky factorization.
1422 * Fixed the overflowing buffer bug of multithreading hbmv and sbmv.(#174)
1425 * Added AMD Bulldozer x86-64 S/DGEMM AVX kernels. (Thank Werner Saar)
1426 We will tune the performance in future.
1427 * Auto-detect Intel Xeon E7540.
1428 * Fixed the overflowing buffer bug of gemv. (#173)
1429 * Fixed the bug of s/cdot about invalid reading NAN on x86_64. (#189)
1433 ====================================================================
1437 * Added NO_SHARED flag to disable generating the shared library.
1438 * Compile LAPACKE with ILP64 modle when INTERFACE64=1 (#158)
1439 * Export LAPACK 3.4.2 symbols in shared library. (#147)
1440 * Only detect the number of physical CPU cores on Mac OSX. (#157)
1441 * Fixed NetBSD build. (#155)
1442 * Fixed compilation with TARGET=GENERIC. (#160)
1444 * Restore the original CPU affinity when calling
1445 openblas_set_num_threads(1) (#153)
1446 * Fixed a SEGFAULT bug in dgemv_t when m is very large.(#154)
1449 ====================================================================
1453 * Upgraded LAPACK to 3.4.2 version. (#145)
1454 * Provided support for passing CFLAGS, FFLAGS, PFLAGS,
1455 FPFLAGS to make. (#137)
1456 * f77blas.h:compatibility for compilers without C99 complex
1457 number support. (#141)
1459 * Added NO_AVX flag. Check OS supporting AVX on runtime. (#139)
1460 * Fixed zdot incompatibility ABI issue with GCC 4.7 on
1461 Windows 32-bit. (#140)
1463 * Fixed the generation of shared library bug.
1464 * Fixed the detection bug on the Loongson 3A server.
1465 ====================================================================
1469 * Fixed LAPACK unstable bug about ?laswp. (#130)
1470 * Fixed the shared library bug about unloading the library on
1472 * Fixed the compilation failure on BlueGene/P (TARGET=PPC440FP2)
1473 Please use gcc and IBM xlf. (#134)
1475 * Supported goto_set_num_threads and openblas_set_num_threads
1476 APIs in Windows. They can set the number of threads on runtime.
1478 ====================================================================
1482 * Fixed exporting DLL functions bug on Windows/MingW
1483 * Support GNU Hurd (Thank Sylvestre Ledru)
1484 * Support kfreebsd kernel (Thank Sylvestre Ledru)
1486 * Support Intel Sandy Bridge 22nm desktop/mobile CPU
1488 * Improve the detection of SPARC (Thank Sylvestre Ledru)
1490 ====================================================================
1495 * Fixed the SEGFAULT bug about hyper-theading
1496 * Support AMD Bulldozer by using GotoBLAS2 AMD Barcelona codes
1498 ====================================================================
1502 * Removed the limitation (64) of numbers of CPU cores.
1503 Now, it supports 256 cores at max.
1504 * Supported clang compiler.
1505 * Fixed some build bugs on FreeBSD
1507 * Optimized Level-3 BLAS on Intel Sandy Bridge x86-64 by AVX instructions.
1508 Please use gcc >= 4.6 or clang >=3.1.
1509 * Support AMD Bobcat by using GotoBLAS2 AMD Barcelona codes.
1511 ====================================================================
1515 * Upgraded LAPACK to 3.4.1 version. (Thank Zaheer Chothia)
1516 * Supported LAPACKE, a C interface to LAPACKE. (Thank Zaheer Chothia)
1517 * Fixed the build bug (MD5 and download) on Mac OSX.
1518 * Auto download CUnit 2.1.2-2 from SF.net with UTEST_CHECK=1.
1519 * Fxied the compatibility issue for compilers without C99 complex number
1520 (e.g. Visual Studio)
1522 * Auto-detect Intel Sandy Bridge Core i7-3xxx & Xeon E7 Westmere-EX.
1523 * Test alpha=Nan in dscale.
1524 * Fixed a SEGFAULT bug in samax on x86 windows.
1526 ====================================================================
1530 * Set soname of shared library on Linux.
1531 * Added LIBNAMESUFFIX flag in Makefile.rule. The user can use
1532 this flag to control the library name, e.g. libopenblas.a,
1533 libopenblas_ifort.a or libopenblas_omp.a.
1534 * Added GEMM_MULTITHREAD_THRESHOLD flag in Makefile.rule.
1535 The lib use single thread in GEMM function with small matrices.
1537 * Used GEMV SSE/SSE2 kernels on x86 32-bit.
1538 * Exported CBLAS functions in Windows DLL.
1540 * Completed Level-3 BLAS optimization on Loongson 3A CPU.
1541 * Improved GEMV performance on Loongson 3A CPU.
1542 * Improved Level-3 BLAS performance on Loongson 3B CPU. (EXPERIMENT)
1544 ====================================================================
1545 Version 0.1 alpha2.5
1548 * Fixed missing "#include <sched.h>" bug on Mac OS X.
1549 Thank Mike Nolta for the patch.
1550 * Upgraded LAPACK to 3.4.0 version
1551 * Fixed a bug on Mac OS X. Don't require SystemStubs on OS X.
1552 SystemStubs does not exist on Lion. Thank Stefan Karpinski.
1553 * Improved README with using OpenMP. Check the internal threads
1554 count less than or equal to omp_get_max_threads()
1556 * Auto-detect Intel Core i6/i7 (Sandy Bridge) CPU with Nehalem assembly kernels
1557 * Fixed some bugs on MingW 64-bit including zgemv, cdot, zdot.
1559 ====================================================================
1560 Version 0.1 alpha2.4
1563 * Fixed a bug about installation. The header file "fblas77.h"
1565 * Fixed #61 a building bug about setting TARGET and DYNAMIC_ARCH.
1566 * Try to handle absolute path of shared library in OSX. (#57)
1567 Thank Dr Kane O'Donnell.
1568 * Changed the installation folder layout to $(PREFIX)/include and
1572 * Fixed #58 zdot/xdot SEGFAULT bug with GCC-4.6 on x86. According
1573 to i386 calling convention, The callee should remove the first
1574 hidden parameter.Thank Mr. John for this patch.
1576 ====================================================================
1577 Version 0.1 alpha2.3
1581 * Added DTB_ENTRIES into dynamic arch setting parameters. Now,
1582 it can read DTB_ENTRIES on runtime. (Refs issue #55 on github)
1584 ====================================================================
1585 Version 0.1 alpha2.2
1589 * Fixed a building bug when DYNAMIC_ARCH=1 & INTERFACE64=1.
1590 (Refs issue #44 on github)
1592 ====================================================================
1593 Version 0.1 alpha2.1
1597 * Stop the build and output the error message when detecting
1598 fortran compiler failed. (Refs issue #42 on github)
1600 ====================================================================
1605 * Fixed blasint undefined bug in <cblas.h> file. Other software
1606 could include this header successfully(Refs issue #13 on github)
1607 * Fixed the SEGFAULT bug on 64 cores. On SMP server, the number
1608 of CPUs or cores should be less than or equal to 64.(Refs issue #14
1610 * Support "void goto_set_num_threads(int num_threads)" and "void
1611 openblas_set_num_threads(int num_threads)" when USE_OPENMP=1
1612 * Added extern "C" to support C++. Thank Tasio for the patch(Refs
1613 issue #21 on github)
1614 * Provided an error message when the arch is not supported.(Refs
1615 issue #19 on github)
1616 * Fixed issue #23. Fixed a bug of f_check script about generating link flags.
1617 * Added openblas_set_num_threads for Fortran.
1618 * Fixed #25 a wrong result of rotmg.
1619 * Fixed a bug about detecting underscore prefix in c_check.
1620 * Print the wall time (cycles) with enabling FUNCTION_PROFILE
1621 * Fixed #35 a build bug with NO_LAPACK=1 & DYNAMIC_ARCH=1
1622 * Added install target. You can use "make install". (Refs #20)
1626 * Fixed #28 a wrong result of dsdot on x86_64.
1627 * Fixed #32 a SEGFAULT bug of zdotc with gcc-4.6.
1628 * Fixed #33 ztrmm bug on Nehalem.
1629 * Work-around #27 the low performance axpy issue with small imput size & multithreads.
1632 * Fixed #28 a wrong result of dsdot on Loongson3A/MIPS64.
1633 * Optimized single/double precision BLAS Level3 on Loongson3A/MIPS64. (Refs #2)
1634 * Optimized single/double precision axpy function on Loongson3A/MIPS64. (Refs #3)
1636 ====================================================================
1641 * Support "make NO_LAPACK=1" to build the library without
1643 * Fixed randomly SEGFAULT when nodemask==NULL with above Linux 2.6.34.
1644 Thank Mr.Ei-ji Nakama providing this patch. (Refs issue #12 on github)
1645 * Added DEBUG=1 rule in Makefile.rule to build debug version.
1646 * Disable compiling quad precision in reference BLAS library(netlib BLAS).
1647 * Added unit testcases in utest/ subdir. Used CUnit framework.
1648 * Supported OPENBLAS_* & GOTO_* environment variables (Pleas see README)
1649 * Imported GotoBLAS2 1.13 BSD version
1652 * On x86 32bits, fixed a bug in zdot_sse2.S line 191. This would casue
1653 zdotu & zdotc failures. Instead, work-around it. (Refs issue #8 #9 on github)
1654 * Modified ?axpy functions to return same netlib BLAS results
1655 when incx==0 or incy==0 (Refs issue #7 on github)
1656 * Modified ?swap functions to return same netlib BLAS results
1657 when incx==0 or incy==0 (Refs issue #6 on github)
1658 * Modified ?rot functions to return same netlib BLAS results
1659 when incx==0 or incy==0 (Refs issue #4 on github)
1660 * Detect Intel Westmere,Intel Clarkdale and Intel Arrandale
1661 to use Nehalem codes.
1662 * Fixed a typo bug about compiling dynamic ARCH library.
1664 * Improve daxpy performance on ICT Loongson 3A.
1665 * Supported ICT Loongson 3A CPU (Refs issue #1 on github)
1666 ====================================================================