2 ====================================================================
7 - when the build-time number of preconfigured threads is exceeded
8 at runtime (typically by an external program calling BLAS functions
9 from a larger number of threads in parallel), OpenBLAS will now
10 allocate an auxiliary control structure for up to 512 additional
11 threads instead of aborting
12 - added support for Loongson's LoongArch64 cpu architecture
13 - fixed building OpenBLAS with CMAKE and -DBUILD_BFLOAT16=ON
14 - added support for building OpenBLAS as a CMAKE subproject
15 - added support for building for Windows/ARM64 targets with clang
16 - improved support for building with the IBM xlf compiler
17 - imported Reference-LAPACK PR 625 (out-of-bounds reads in ?LARRV)
18 - imported Reference-LAPACK PR 597 for testsuite compatibility with
22 - added SkylakeX S/DGEMM kernels for small problem sizes (M*N*K<=1000000)
23 - added optimized SBGEMM for Intel Cooper Lake
24 - reinstated the performance patch for AVX512 SGEMV_T with a proper fix
25 - added a workaround for a gcc11 tree-vectorizer bug that caused spurious
26 failures in the test programs for complex BLAS3 when compiling at -O3
27 (the default for cmake "release" builds)
28 - added support for runtime cpu count detection under Haiku OS
29 - worked around a long-standing miscompilation issue of the Haswell DGEMV_T
30 kernel with gcc that could produce NaN output in some corner cases
33 - improved performance of DASUM on POWER10
36 - fixed crashes (use of reserved register x18) on Apple M1 under OSX
37 - fixed building with gcc releases earlier than 5.1
40 - fixed building under BSD
43 - fixed building under BSD
45 ====================================================================
50 - reverted the optimization of SGEMV_N/DGEMV_N for small input sizes
51 and consecutive arguments as it led to stack overflows on x86_64
52 with some operating systems (notably OSX and Windows)
55 - reverted the performance patch for SGEMV_T on AVX512 as it caused
56 wrong results in some applications
59 - fixed compilation with compilers other than gcc
60 ====================================================================
65 - drastically reduced the stack size requirements for running the LAPACK
66 testsuite (Reference-LAPACK PR 553)
67 - fixed spurious test failures in the LAPACK testsuite (Reference-LAPACK
69 - expressly setting DYNAMIC_ARCH=0 no longer enables dynamic_arch mode
70 - improved performance of xGER, xSPR, xSPR2, xSYR, xSYR2, xTRSV, SGEMV_N
71 and DGEMV_N, for small input sizes and consecutive arguments
72 - improved performance of xGETRF, xPORTF and xPOTRI for small input sizes
73 by disabling multithreading
74 - fixed installing with BSD versions of the "install" utility
77 - fixed the implementation of xIMIN
78 - improved the performance of DSDOT
79 - fixed linking of the tests on C910V with current vendor gcc
82 - fixed SBGEMM computation for some odd value inputs
83 - fixed compilation for PPCG4, PPC970, POWER3, POWER4 and POWER5
86 - improved performance of SGEMV_N and SGEMV_T for small N on AVX512-capable cpus
87 - worked around a miscompilation of ZGEMM/ZTRMM on Sandybridge with old gcc
89 - fixed compilation with MS Visual Studio versions older than 2017
90 - fixed macro name collision with winnt.h from the latest Win10 SDK
91 - added cpu type autodetection for Intel Ice Lake SP
92 - fixed cpu type autodetection for Intel Tiger Lake
93 - added cpu type autodetection for recent Centaur/Zhaoxin models
94 - fixed compilation with musl libc
97 - fixed compilation with gcc/gfortran on the Apple M1
98 - fixed linking of the tests on FreeBSD
99 - fixed missing restore of a register in the recently rewritten DNRM2 kernel
100 for ThunderX2 and Neoverse N1 that could cause spurious failures in e.g.
102 - added compiler optimization flags for the EMAG8180
103 - added initial support for Cortex A55
106 - fixed linking of the tests on FreeBSD
108 ====================================================================
113 - imported improvements and bugfixes from Reference-LAPACK 3.9.1
114 - imported LAPACKE interface fixes from Reference-LAPACK PRs 534 + 537
115 - fixed a problem in the cpu detection of 0.3.14 that prevented cross-compilation
116 - fixed a sequence problem in the generation of softlinks to the library in GMAKE
119 - fixed compilation on RISCV (missing entry in getarch)
120 - fixed a potential division by zero in CROTG and ZROTG
123 - fixed LAPACK testsuite failures seen with the NVIDIA HPC compiler
124 - improved CGEMM, DGEMM and ZGEMM performance on POWER10
125 - added an optimized ZGEMV kernel for POWER10
126 - fixed a potential division by zero in CROTG and ZROTG
129 - added support for Intel Control-flow Enforcement Technology (CET)
130 - reverted the DOMATCOPY_RT code to the generic C version
131 - fixed a bug in the AVX512 SGEMM kernel introduced in 0.3.14
132 - fixed misapplication of -msse flag to non-SSE cpus in DYNAMIC_ARCH
133 - added support for compilation of the benchmarks on older OSX versions
134 - fix propagation of the NO_AVX512 option in CMAKE builds
135 - fix compilation of the AVX512 SGEMM kernel with clang-cl on Windows
136 - fixed compilation of the CTESTs with INTERFACE64=1 (random faults on OSX)
137 - corrected the Haswell DROT kernel to require AVX2/FMA3 rather than AVX512
140 - fixed a potential division by zero in CROTG and ZROTG
141 - fixed a potential overflow in IMATCOPY/ZIMATCOPY and the CTESTs
144 - fixed spurious reads outside the array in the SGEMM tcopy macro
145 - fixed a potential division by zero in CROTG and ZROTG
146 - fixed a segmentation fault in DYNAMIC_ARCH builds (reappeared in 0.3.14)
149 - fixed a potential division by zero in CROTG and ZROTG
150 - fixed a potential overflow in IMATCOPY/ZIMATCOPY and the CTESTs
153 - fixed a potential division by zero in CROTG and ZROTG
156 - fixed a potential division by zero in CROTG and ZROTG
158 ====================================================================
163 * Fixed a race condition on thread shutdown in non-OpenMP builds
164 * Fixed custom BUFFERSIZE option getting ignored in gmake builds
165 * Fixed CMAKE compilation of the TRMM kernels for GENERIC platforms
166 * Added CBLAS interfaces for CROTG, ZROTG, CSROT and ZDROT
167 * Improved performance of OMATCOPY_RT across all platforms
168 * Changed perl scripts to use env instead of a hardcoded /usr/bin/perl
169 * Fixed potential misreading of the GCC compiler version in the build scripts
170 * Fixed convergence problems in LAPACK complex GGEV/GGES (Reference-LAPACK #477)
171 * Reduced the stacksize requirements for running the LAPACK testsuite (Reference-LAPACK #335)
174 * Fixed compilation on RISCV (missing entry in getarch)
177 * Fixed compilation for DYNAMIC_ARCH with clang and with old gcc versions
178 * Added support for compilation on FreeBSD/ppc64le
179 * Added optimized POWER10 kernels for SSCAL, DSCAL, CSCAL, ZSCAL
180 * Added optimized POWER10 kernels for SROT, DROT, CDOT, SASUM, DASUM
181 * Improved SSWAP, DSWAP, CSWAP, ZSWAP performance on POWER10
182 * Improved SCOPY and CCOPY performance on POWER10
183 * Improved SGEMM and DGEMM performance on POWER10
184 * Added support for compilation with the NVIDIA HPC compiler
187 * Added an optimized bfloat16 GEMM kernel for Cooperlake
188 * Added CPUID autodetection for Intel Rocket Lake and Tiger Lake cpus
189 * Improved the performance of SASUM,DASUM,SROT,DROT on AMD Ryzen cpus
190 * Added support for compilation with the NAG Fortran compiler
191 * Fixed recognition of the AMD AOCC compiler
192 * Fixed compilation for DYNAMIC_ARCH with clang on Windows
193 * Added support for running the BLAS/CBLAS tests on Windows
194 * Fixed signatures of the tls callback functions for Windows x64
195 * Fixed various issues with fma intrinsics support handling
198 * Added support for embedded Cortex M targets via a new option EMBEDDED
201 * Fixed the THUNDERX2T99 and NEOVERSEN1 DNRM2/ZNRM2 kernels for inputs with Inf
202 * Added support for the DYNAMIC_LIST option
203 * Added support for compilation with the NVIDIA HPC compiler
204 * Added support for compiling with the NAG Fortran compiler
206 ====================================================================
211 * Added a generic bfloat16 SBGEMV kernel
212 * Fixed a potentially severe memory leak after fork in OpenMP builds
213 that was introduced in 0.3.12
214 * Added detection of the Fujitsu Fortran compiler
215 * Added detection of the (e)gfortran compiler on OpenBSD
216 * Added support for overriding the default name of the library independently
217 from symbol suffixing in the gmake builds (already supported in cmake)
220 * Added a RISC V port optimized for C910V
223 * Added optimized POWER10 kernels for SAXPY, CAXPY, SDOT, DDOT and DGEMV_N
224 * Improved DGEMM performance on POWER10
225 * Improved STRSM and DTRSM performance on POWER9 and POWER10
226 * Fixed segmemtation faults in DYNAMIC_ARCH builds
227 * Fixed compilation with the PGI compiler
230 * Fixed compilation of kernels that require SSE2 intrinsics since 0.3.12
233 * Added an optimized bfloat16 SBGEMV kernel for SkylakeX and Cooperlake
234 * Improved the performance of SASUM and DASUM kernels through parallelization
235 * Improved the performance of SROT and DROT kernels
236 * Improved the performance of multithreaded xSYRK
237 * Fixed OpenMP builds that use the LLVM Clang compiler together with GNU gfortran
238 (where linking of both the LLVM libomp and GNU libgomp could lead to lockups or
240 * Fixed miscompilations by old gcc 4.6
241 * Fixed misdetection of AVX2 capability in some Sandybridge cpus
242 * Fixed lockups in builds combining DYNAMIC_ARCH with TARGET=GENERIC on OpenBSD
245 * Fixed segmemtation faults in DYNAMIC_ARCH builds
248 * Improved kernels for Loongson 3R3 ("3A") and 3R4 ("3B") models, including MSA
249 * Fixed bugs in the MSA kernels for CGEMM, CTRMM, CGEMV and ZGEMV
250 * Added handling of zero increments in the MSA kernels for SSWAP and DSWAP
251 * Added DYNAMIC_ARCH support for MIPS64 (currently Loongson3R3/3R4 only)
254 * Fixed building 32 and 64 bit SPARC kernels with the SolarisStudio compilers
256 ====================================================================
261 * Fixed missing BLAS/LAPACK functions (inadvertently dropped during
262 the build system restructuring)
263 * Fixed argument conversion macro in LAPACKE_zgesvdq (LAPACK #458)
266 * Added optimized SCOPY/CCOPY kernels for POWER10
267 * Increased and unified the default size of the GEMM BUFFER
268 * Fixed building for POWER10 in DYNAMIC_ARCH mode
269 * POWER10 compatibility test now checks binutils version as well
270 * Cleaned up compiler warnings
273 * corrected compiler version checks for AVX2 compatibility
274 * added compiler option -mavx2 for building with flang
275 * fixed direct SGEMM pathway for small matrix sizes (broken by
276 the code refactoring in 0.3.11)
277 * fixed unhandled partial register clobbers in several kernels
278 for AXPY,DOT,GEMV_N and GEMV_T flagged by gcc10 tree-vectorizer
281 * improved Apple Vortex support to include cross-compiling
283 ====================================================================
289 the newly added BFLOAT16 functions were renamed to use the
290 letter "B" instead of "H" to avoid potential confusion with
291 the IEEE "half precision float" type, i.e. the 0.3.10
292 SHGEMM is now SBGEMM and the corresponding build option
293 was changed from "BUILD_HALF" to "BUILD_BFLOAT16".
294 * Reduced the default BLAS3_MEM_ALLOC_THRESHOLD (used as an upper
295 limit for placing temporary arrays on the stack) to be compatible
296 with a stack size of 1mb (as imposed by the JAVA runtime library)
297 * Added mixed-precision dot function SBDOT and utility functions
298 shstobf16, shdtobf16, sbf16tos and dbf16tod to convert between
299 single or double precision float arrays and bfloat16 arrays
300 * Fixed prototypes of LAPACK_?ggsvp and LAPACK_?ggsvd functions
302 * Fixed underflow and rounding errors in LAPACK SLANV2 and DLANV2
303 (causing miscalculations in e.g. SHSEQR/DHSEQR, LAPACK issue #263)
304 * Fixed workspace calculation in LAPACK ?GELQ (LAPACK issue #415)
305 * Fixed several bugs in the LAPACK testsuite
306 * Improved performance of TRMM and TRSM for certain problem sizes
307 * Fixed infinite recursions and workspace miscalculations in ReLAPACK
308 * CMAKE builds no longer require pkg-config for creating the .pc file
309 * Makefile builds no longer misread NO_CBLAS=0 or NO_LAPACK=0 as
310 enabling these options
311 * Fixed detection of gfortran when invoked through an mpi wrapper
312 * Improve thread reinitialization performance with OpenMP after a fork
313 * Added support for building only the subset of the library required
314 for a particular precision by specifying BUILD_SINGLE, BUILD_DOUBLE
315 * Optional function name prefixes and suffixes are now correctly
316 reflected in the generated cblas.h
317 * Added CMAKE build support for the LAPACK and multithreading tests
320 * Added optimized support for POWER10
321 * Added support for compiling for POWER8 in 32bit mode
322 * Added support for compilation with LLVM/clang
323 * Added support for compilation with NVIDIA/PGI compilers
324 * Fixed building on big-endian POWER8
325 * Fixed miscompilation of ZDOTC by gcc10
326 * Fixed alignment errors in the POWER8 SAXPY kernel
327 * Improved CPU detection on AIX
328 * Supported building with older compilers on POWER9
331 * Added support for Intel Cooperlake
332 * Added autodetection of AMD Renoir/Matisse/Zen3 cpus
333 * Added autodetection of Intel Comet Lake cpus
334 * Reimplemented ?sum, ?dot and daxpy using universal intrinsics
335 * Reset the fpu state before using the fpu on Windows as a workaround
336 for a problem introduced in Windows 10 build 19041 (a.k.a. SDK 2004)
337 * Fixed potentially undefined behaviour in the dot and gemv_t kernels
338 * Fixed a potential segmentation fault in DYNAMIC_ARCH builds
339 * Fixed building for ZEN with PGI/NVIDIA and AMD AOCC compilers
342 * Fixed cpu detection on BSD-like systems
345 * Added preliminary support for Apple Vortex cpus
346 * Added support for the Cavium ThunderX3T110 cpu
347 * Fixed cpu detection on BSD-like systems
348 * Fixed compilation in -std=C18 mode
351 * Added support for compiling with the clang compiler
352 * Improved GEMM performance on Z14
354 ====================================================================
359 * Improved thread locking behaviour in blas_server and parallel getrf
360 * Imported bugfix 394 from LAPACK (spurious reference to "XERBL"
361 due to overlong lines)
362 * Imported bugfix 403 from LAPACK (compile option "recursive" required
363 for correctness with Intel and PGI)
364 * Imported bugfix 408 from LAPACK (wrong scaling in ZHEEQUB)
365 * Imported bugfix 411 from LAPACK (infinite loop in LARGV/LARTG/LARTGP)
366 * Fixed mismatches between BUFFERSIZE and GEMM_UNROLL parameters that
367 could lead to crashes at large matrix sizes
368 * Restored internal soname in dynamic libraries on FreeBSD and Dragonfly
369 * Added API (openblas_setaffinity) to set the thread affinity on Linux
370 * Added initial infrastructure for half-precision floating point
371 (bfloat16) support with a generic implementation of SHGEMM
372 * Added CMAKE build system support for building the cblas_Xgemm3m
374 * Fixed CMAKE support for building in a path with embedded spaces
375 * Fixed CMAKE (non)handling of NO_EXPRECISION and MAX_STACK_ALLOC
376 * Fixed GCC version detection in the Makefiles
377 * Allowed overriding the names of AR, AS and LD in Makefile builds
380 * Fixed big-endian POWER8 ELFv2 builds on FreeBSD
381 * Fixed GCC version checks and DYNAMIC_ARCH builds on POWER9
382 * Fixed CMAKE build support for POWER9
383 * fixed a potential race condition in the thread buffer allocation
384 * Worked around LAPACK test failures on PPC G4
387 * Fixed a potential race condition in the thread buffer allocation
388 * Added support for MIPS 24K/24KE family based on P5600 kernels
391 * fixed a potential race condition in the thread buffer allocation
392 * Added TARGET=GENERIC
395 * Fixed a race condition in the thread buffer allocation
398 * Fixed a race condition in the thread buffer allocation
399 * Fixed zero initialisation in the assembly for SGEMM and DGEMM BETA
400 * Improved performance of the ThunderX2 DAXPY kernel
401 * Added an optimized SGEMM kernel for Cortex A53
402 * Fixed Makefile support for INTERFACE64 (8-byte integer)
405 * Fixed a syntax error in the CMAKE setup for SkylakeX
406 * Improved performance of STRSM on Haswell, SkylakeX and Ryzen
407 * Improved SGEMM performance on SGEMM for workloads with ldc a
409 * Improved DGEMM performance on Skylake X
410 * Fixed unwanted AVX512-dependency of SGEMM in DYNAMIC_ARCH
411 builds created on SkylakeX
412 * Removed data alignment requirement in the SSE2 copy kernels
413 that could cause spurious crashes
414 * Added a workaround for an optimizer bug in AppleClang 11.0.3
415 * Fixed LAPACK test failures due to wrong options for Intel Fortran
416 * Fixed compilation and LAPACK test results with recent Flang
418 * Fixed DYNAMIC_ARCH builds with CMAKE on OS X
419 * Fixed missing exports of cblas_i?amin, cblas_i?min, cblas_i?max,
420 cblas_?sum, cblas_?gemm3m in the shared library on OS
421 * Fixed reporting of cpu name in DYNAMIC_ARCH builds (would sometimes
422 show the name of an older generation chip supported by the same kernels)
425 * Improved performance of SGEMM/STRMM and DGEMM/DTRMM on Z14
427 ====================================================================
432 * Fixed a miscompilation of the GETRF functions with CMAKE
433 * Imported bugfix 390 from LAPACK (missing NaN propagation in xCOMBSSQ)
434 * The size of the memory buffer used for splitting GEMM tasks across
435 multiple threads can now be configured in the build system.
438 * Fixed several compilation problems related to endianness
439 and ELF version on POWER8 and POWER9
440 * Fixed use of the absolute value IAMIN/IAMAX instead of IMIN/IMAX
441 * Fixed a race condition in the level3 blas code
444 * Fixed use of the absoltute value IAMIN/IAMAX instead of IMIN/IMAX
447 * Fixed a race condition in the level3 blas code
448 * Fixed compilation on Android
450 * Added support for Ampere EMAG8180
451 * Added support for Neoverse N1
452 * Improved performance of the blas_lock function
453 * Fixed a race condition in the level3 blas code
454 * Fixed a performance regression on TSV110-based servers
457 * Fixed a long-standing error with undeclared register overwrites
458 in the DSCAL microkernel for HASWELL,SKYLAKEX and ZEN
459 * Fixed a long-standing bug in the SSE implementation of IAMAX
460 * Fixed a CMAKE build failure with DYNAMIC_ARCH
461 * Fixed cpu autodetection of Goldmont+, Cannon Lake and Ice Lake
462 * Fixed a compilation failure on OSX with compiler name containing dash
463 * Fixed compilation with MinGW on SkylakeX
464 * Improved speed of the AVX512 GEMM3M kernel on SkylakeX
465 * Added an AVX512 STRMM kernel for SkylakeX
466 * Improved GEMM performance on Haswell and Zen
469 * fixed compilation of the DYNAMIC_ARCH code
471 ====================================================================
476 ` * LAPACK has been updated to 3.9.0 (plus patches up to
478 * CMAKE support has been improved in several areas including
480 * a thread race condition in the GEMM3M kernels was resolved
481 * the "generic" (plain C) gemm beta kernel used by many targets
483 * an optimized version of the LAPACK trtrs functions has been added
484 * an incompatibilty between the LAPACK tests and the OpenBLAS
485 implementation of XERBLA was resolved, removing the numerous
486 warnings about wrong error exits in the former
487 * support for NetBSD has been added
488 * support for compilation with g95 and non-GNU versions of ld
490 * support for compilation with (upcoming) gcc 10 has been added
493 * worked around miscompilation of several POWER8 and POWER9
494 kernels by older versions of gcc
495 * added support for big-endian POWER8 and for compilation on AIX
496 * corrected bugs in the big-endian support for PPC440 and PPC970
497 * DYNAMIC_ARCH support is now available in CMAKE builds as well
500 * performance of DGEMM_BETA and SGEMM_NCOPY has been improved
501 * compilation for 32bit works again
502 * performance of the RPCC function has been improved
503 * improved performance on small systems
504 * DYNAMIC_ARCH support is now available in CMAKE builds as well
505 * cross-compilation from OSX to IOS was simplified
508 * a new AVX512 DGEMM kernel was added and the AVX512 SGEMM kernel
509 was significantly improved
510 * optimized AVX512 kernels for CGEMM and ZGEMM have been added
511 * AVX2 kernels for STRMM, SGEMM, and CGEMM have been significantly
512 sped up and optimized CGEMM3M and ZGEMM3M kernels have been added
513 * added support for QEMU virtual cpus
514 * a compilation problem with PGI and SUN compilers was fixed
515 * Intel "Goldmont plus" is now autodetected
516 * a potential crash on program exit on MS Windows has been fixed
519 * an unwanted case sensitivity in the implementation of LSAME
520 on older 32bit AMD cpus was fixed
523 * Z15 is now supported as Z14
524 * DYNAMIC_ARCH is now available on ZARCH as well
526 ====================================================================
531 * having the gmake special variables TARGET_ARCH or TARGET_MACH
532 defined no longer causes build failures in ctest or utest
533 * defining NO_AFFINITY or USE_TLS to 0 in gmake builds no longer
534 has the same effect as setting them to 1
535 * a new test program was added to allow checking the library for
537 * a new option USE_LOCKING was added to ensure thread safety when
538 OpenBLAS itself is built without multithreading but will be
539 called from multiple threads.
540 * a build failure on Linux with glibc versions earlier than 2.5
542 * a runtime error with CPU enumeration (and NO_AFFINITY not set)
543 on glibc 2.6 was fixed
544 * NO_AFFINITY was added to the CMAKE options (and defaults to being
545 active on Linux, as in the gmake builds)
548 * the build-time logic for detection of AVX512 availability in
549 the processor and compiler was fixed
550 * gmake builds on OSX now set the internal name of the library to
551 libopenblas.0.dylib (consistent with CMAKE)
552 * the Haswell DGEMM kernel received a significant speedup through
553 improved prefetch and load instructions
554 * performance of DGEMM, DTRMM, DTRSM and ZDOT on Zen/Zen2 was markedly
555 increased by avoiding vpermpd instructions
556 * the SKYLAKEX (AVX512) DGEMM helper functions have now been disabled
557 to fix remaining errors in DGEMM, DSYMM and DTRMM
560 * added support for building on FreeBSD/powerpc64 and FreeBSD/ppc970
561 * added optimized kernels for POWER9 SGEMM and STRMM
564 * fixed the softfp implementations of xAMAX and IxAMAX
565 * removed the predefined -march= flags on both ARMV5 and ARMV6 as
566 they were appropriate for only a subset of platforms
568 ====================================================================
573 * the build tools now check that a given cpu TARGET is actually valid
574 * the build-time check of system features (c_check) has been made
575 less dependent on particular perl features (this should mainly
576 benefit building on Windows)
577 * several problem with the ReLAPACK integration were fixed,
578 including INTERFACE64 support and building a shared library
579 * building with CMAKE on BSD systems was improved
580 * a non-absolute SUM function was added based on the
581 existing optimized code for ASUM
582 * CBLAS interfaces to the IxMIN and IxMAX functions were added
583 * a name clash between LAPACKE and BOOST headers was resolved
584 * CMAKE builds with OpenMP failed to include the appropriate getrf_parallel
586 * a crash on thread (key) deletion with the USE_TLS=1 memory management
588 * restored several earlier fixes, in particular for OpenMP performance,
589 building on BSD, and calling fork on CYGWIN, which had inadvertently
590 been dropped in the 0.3.3 rewrite of the memory management code.
593 * the AVX512 DGEMM kernel has been disabled again due to unsolved problems
594 * building with old versions of MSVC was fixed
595 * it is now possible to build a static library on Windows with CMAKE
596 * accessing environment variables on CYGWIN at run time was fixed
597 * the CMAKE build system now recognizes 32bit userspace on 64bit hardware
598 * Intel "Denverton" atom and Hygon "Dhyana" zen CPUs are now autodetected
599 * building for DYNAMIC_ARCH with a DYNAMIC_LIST of targets is now supported
601 * building for DYNAMIC_ARCH with GENERIC as the default target is now supported
602 * a buffer overflow in the SSE GEMM kernel for Intel Nano targets was fixed
603 * assembly bugs involving undeclared modification of input operands were fixed
604 in the AXPY, DOT, GEMV, GER, SCAL, SYMV and TRSM microkernels for Nehalem,
605 Sandybridge, Haswell, Bulldozer and Piledriver. These would typically cause
606 test failures or segfaults when compiled with recent versions of gcc from 8 onward.
607 * a similar bug was fixed in the blas_quickdivide code used to split workloads
609 * a bug in the IxMIN implementation for the GENERIC target made it return the result of IxMAX
610 * fixed building on SkylakeX systems when either the compiler or the (emulated) operating
611 environment does not support AVX512
612 * improved GEMM performance on ZEN targets
615 * build failures caused by the recently added checks for AVX512 were fixed
616 * an inline assembly bug involving undeclared modification of an input argument was
617 fixed in the blas_quickdivide code used to split workloads in most functions
618 * a bug in the IMIN implementation for the GENERIC target made it return the result of IMAX
621 * a bug in the IMIN implementation made it return the result of IMAX
624 * single precision BLAS1/2 functions have received optimized POWER8 kernels
625 * POWER9 is now a separate target, with an optimized DGEMM/DTRMM kernel
626 * building on PPC970 systems under OSX Leopard or Tiger is now supported
627 * out-of-bounds memory accesses in the gemm_beta microkernels were fixed
628 * building a shared library on AIX is now supported for POWER6
629 * DYNAMIC_ARCH support has been added for POWER6 and newer
632 * corrected xDOT behaviour with zero INC_X or INC_Y
633 * a bug in the IMIN implementation made it return the result of IMAX
636 * added support for HiSilicon TSV110 cpus
637 * the CMAKE build system now recognizes 32bit userspace on 64bit hardware
638 * cross-compilation with CMAKE now works again
639 * a bug in the IMIN implementation made it return the result of IMAX
640 * ARMV8 builds with the BINARY=32 option are now automatically handled as ARMV7
643 * optimized microkernels for single precicion BLAS1/2 functions have been added
646 ====================================================================
651 * loop unrolling in TRMV has been enabled again.
652 * A domain error in the thread workload distribution for SYRK
654 * gmake builds will now automatically add -fPIC to the build
655 options if the platform requires it.
656 * a pthreads key leakage (and associate crash on dlclose) in
657 the USE_TLS codepath was fixed.
658 * building of the utest cases on systems that do not provide
659 an implementation of complex.h was fixed.
662 * the SkylakeX code was changed to compile on OSX.
663 * unwanted application of the -march=skylake-avx512 option
664 to the common code parts of a DYNAMIC_ARCH build was fixed.
665 * improved performance of SGEMM for small workloads on Skylake X.
666 * performance of SGEMM and DGEMM was improved on Haswell.
669 * a configuration error that broke the CNRM2 kernel was corrected.
670 * compilation of the GEMM kernels with CMAKE was fixed.
671 * DYNAMIC_ARCH builds are now available with CMAKE as well.
672 * using CMAKE for cross-compilation to the new cpu TARGETs
673 introduced in 0.3.4 now works.
676 * a problem in cpu autodetection for AIX has been corrected.
678 ====================================================================
683 * the new, experimental thread-local memory allocation had
684 inadvertently been left enabled for gmake builds in 0.3.3
685 despite the announcement. It is now disabled by default, and
686 single-threaded builds will keep using the old allocator even
687 if the USE_TLS option is turned on.
688 * OpenBLAS will now provide enough buffer space for at least 50
690 * The output of openblas_get_config() now contains the version
692 * A serious thread safety bug in GEMV operation with small M and
693 large N size has been fixed.
694 * The code will now automatically call blas_thread_init after a
695 fork if needed before handling a call to openblas_set_num_threads
696 * Accesses to parallelized level3 functions from multiple callers
697 are now serialized to avoid thread races (unless using OpenMP).
698 This should provide better performance than the known-threadsafe
699 (but non-default) USE_SIMPLE_THREADED_LEVEL3 option.
700 * When building LAPACK with gfortran, -frecursive is now (again)
701 enabled by default to ensure correct behaviour.
702 * The OpenBLAS version cblas.h now supports both CBLAS_ORDER and
703 CBLAS_LAYOUT as the name of the matrix row/column order option.
704 * Externally set LDFLAGS are now passed through to the final compile/link
705 steps to facilitate setting platform-specific linker flags.
706 * A potential race condition during the build of LAPACK (that would
707 usually manifest itself as a failure to build TESTING/MATGEN) has been
709 * xHEMV has been changed to stay single-threaded for small input sizes
710 where the overhead of multithreading exceeds any possible gains
711 * CSWAP and ZSWAP have been limited to a single thread except on ARMV8 or
712 ThunderX hardware with sizable input.
713 * Linker flags for the PGI compiler have been updated
714 * Behaviour of AXPY with zero increments is now handled in the C interface,
715 correcting the result on at least Intel Atom.
716 * The result matrix from calling SGELSS with an all-zero input matrix is
717 now zeroed completely.
720 * Autodetection of AMD Ryzen2 has been fixed (again).
721 * CMAKE builds now support labeling of an INTERFACE64=1 build of
722 the library with the _64 suffix.
723 * AVX512 version of DGEMM has been added and the AVX512 SGEMM kernel
724 has been sped up by rewriting with C intrinsics
725 * Fixed compilation on RHEL5/CENTOS5 (issue with typename __WAIT_STATUS)
728 * added support for building on AIX (with gcc and GNU tools from AIX Toolbox).
729 * CPU type detection has been implemented for AIX.
730 * CPU type detection has been fixed for NETBSD.
733 * AXPY on LOONGSON3A has been corrected to pass "zero increment" utest.
734 * DSDOT on LOONGSON3A has been fixed.
735 * the SGEMM microkernel has been hardened against potential data loss.
738 * DYNAMic_ARCH support is now available for 64bit ARM
739 * cross-compiling for ARMV8 under iOS now works.
740 * cpu-specific code has been rearranged to make better use of both
741 hardware commonalities and model-specific compiler optimizations.
742 * XGENE1 has been removed as a TARGET, superseded by the improved generic
746 * Older assembly mnemonics have been converted to UAL form to allow
747 building with clang 7.0
748 * Cross compiling LAPACKE for Android has been fixed again (broken by
749 update to LAPACK 3.7.0 some while ago).
751 ====================================================================
756 * thread memory allocation has been switched back to the method
757 used before version 0.3.1 due to unexpected problems caused by
758 the new code under some circumstances. A new compile-time option
759 USE_TLS has been added to enable the new code, and it is hoped
760 that this can become the default again in the next version.
761 * LAPAck PR272 has been integrated, which fixes spurious errors
762 in DSYEVR and related functions caused by missing conversion
763 from ILAENV to ILAENV_2STAGE in several _2stage routines.
764 * the cmake-generated OpenBLASConfig.cmake now uses correct case
765 for the name of the library
766 * added support for Haiku OS
769 * added AVX512 implementations of SDOT, DDOT, SAXPY, DAXPY,
770 DSCAL, DGEMVN and DSYMVL
771 * added a workaround for a cygwin issue that prevented compilation
775 * added autodetection of Z14
776 * fixed TRMM errors in the generic target
778 ====================================================================
783 * fixes for regressions caused by the rewrite of the thread
784 initialization code in 0.3.1
787 * fixed cpu autodetection for the BSDs
790 * fixed utest errors in AXPY, DSDOT, ROT and SWAP
793 * added autodetection of AMD Ryzen 2
794 * fixed build with older versions of MSVC
796 ====================================================================
801 * rewritten thread initialization code with significantly reduced overhead
802 * added CBLAS interfaces to the IxAMIN BLAS extension functions
803 * fixed the lapack-test target
804 * CMAKE builds now create an OpenBLASConfig.cmake file
805 * ZAXPY now uses a single thread for small input sizes
806 * the LAPACK code was updated from Reference-LAPACK/lapack#253
807 (fixing LAPACKE interfaces to Aasen's functions)
810 * corrected CROT and ZROT behaviour with zero INC_X
813 * corrected xDOT behaviour with zero INC_X or INC_Y
816 * retired some older targets of DYNAMIC_ARCH builds to a new option DYNAMIC_OLDER,
817 this affects PENRYN,DUNNINGTON,OPTERON,OPTERON_SSE3,BOBCAT,ATOM and NANO
818 (which will still be supported via the slower PRESCOTT kernels when this option is not set)
819 * added an option DYNAMIC_LIST that (used in conjunction with DYNAMIC_ARCH) allows to
820 specify the list of x86_64 targets to include. Any target not on the list will be supported
821 by the Sandybridge or Nehalem kernels if available, or by Prescott.
822 * improved SWITCH_RATIO on Haswell for increased GEMM throughput
823 * added initial support for Intel Skylake X, including an AVX512 SGEMM kernel
824 * added autodetection of Intel Cannon Lake series as Skylake X
825 * added a default L2 cache size for hypervisors that return zero here (Chromebook)
826 * fixed a name clash with recent Windows10 headers that broke the build with (at least)
827 recent mingw from MSYS2
828 * fixed a link error in mixed clang/gfortran builds with OpenMP
829 * updated the OSX deployment target to 10.8
830 * switched on parallel make for builds on MS Windows by default
833 * fixed SSWAP and DSWAP behaviour with zero INC_X and INC_Y
835 ====================================================================
840 * fixed some more thread race and locking bugs
841 * added preliminary support for calling an OpenMP build of the library from multiple threads
842 * removed performance impact of thread locks added in 0.2.20 on OpenMP code
843 * general code cleanup
844 * optimized DSDOT implementation
845 * improved thread distribution for GEMM
846 * corrected IMATCOPY/OMATCOPY implementation
847 * fixed out-of-bounds accesses in the multithreaded xBMV/xPMV and SYMV implementations
848 * cmake build improvements
849 * pkgconfig file now contains build options
850 * openblas_get_config() now reports USE_OPENMP and NUM_THREADS settings used for the build
851 * corrections and improvements for systems with more than 64 cpus
852 * LAPACK code updated to 3.8.0 including later fixes
853 * added ReLAPACK, a recursive implementation of several LAPACK functions
854 * Rewrote ROTMG to handle cases that the netlib code failed to address
855 * Disabled (broken) multithreading code for xTRMV
856 * corrected prototypes of complex CBLAS functions to make our cblas.h match the generally accepted standard
857 * shared memory access failures on startup are now handled more gracefully
858 * restored utests from earlier releases (and made them pass on all affected systems)
861 * several fixes for cpu autodetection
864 * corrected vector register overwriting in several Power8 kernels
865 * optimized additional BLAS functions
868 * added support for CortexA53 and A72
869 * added autodetection for ThunderX2T99
870 * made most optimized kernels the default for generic ARMv8 targets
873 * parallelized DDOT kernel for Haswell
874 * changed alignment directives in assembly kernels to boost performance on OSX
875 * fixed register handling in the GEMV microkernels (bug exposed by gcc7)
876 * added support for building on OpenBSD and Dragonfly
877 * updated compiler options to work with Intel release 2018
878 * support fully optimized build with clang/flang on Microsoft Windows
879 * fixed building on AIX
882 * added optimized BLAS 1/2 functions
885 * fixed cpu autodetection helper code
886 * added mips32 1004K cpu (Mediatek MT7621 and similar SoC)
887 * added mips64 I6500 cpu
889 ====================================================================
894 * Improved CMake support
895 * Fixed several thread race and locking bugs
896 * Fixed default LAPACK optimization level
897 * Updated LAPACK to 3.7.0
898 * Added ReLAPACK (https://github.com/HPAC/ReLAPACK, make BUILD_RELAPACK=1)
901 * Optimizations for Power9
902 * Fixed several Power8 assembly bugs
905 * New optimized Vulcan and ThunderX2T99 targets
906 * Support for ARMV7 SOFT_FP ABI (make ARM_SOFTFP_ABI=1)
907 * Detect all cpu cores including offline ones
908 * Fix compilation with CLANG
909 * Support building a shared library for Android
912 * Fixed several threading issues
913 * Fix compilation with CLANG
916 * Detect Intel Bay Trail and Apollo Lake
917 * Detect Intel Sky Lake and Kaby Lake
918 * Detect Intel Knights Landing
919 * Detect AMD A8, A10, A12 and Ryzen
920 * Support 64bit builds with Visual Studio
921 * Fix building with Intel and PGI compilers
922 * Fix building with MINGW and TDM-GCC
923 * Fix cmake builds for Haswell and related cpus
924 * Fix building for Sandybridge with CLANG 3.9
925 * Add support for the FLANG compiler
928 * New target z13 with BLAS3 optimizations
930 ====================================================================
934 * Improved cross compiling.
935 * Fix the bug on musl libc.
938 * Optimize BLAS on Power8
939 * Fixed Julia+OpenBLAS bugs on Power8
942 * Optimize BLAS on MIPS P5600 and I6400 (Thanks, Shivraj Patil, Kaustubh Raste)
945 * Improved on ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
948 ====================================================================
952 * If you set MAKE_NB_JOBS flag less or equal than zero,
953 make will be without -j.
956 * Support building Visual Studio static library. (#813, Thanks, theoractice)
957 * Fix bugs to pass buidbot CI tests (http://build.openblas.net)
960 * Provide DGEMM 8x4 kernel for Cortex-A57 (Thanks, Ashwin Sekhar T K)
963 * Optimize S and C BLAS3 on Power8
964 * Optimize BLAS2/1 on Power8
966 ====================================================================
970 * Enable BUILD_LAPACK_DEPRECATED=1 by default.
972 ====================================================================
976 * Avoid potential getenv segfault. (#716)
977 * Import LAPACK svn bugfix #142-#147,#150-#155
980 * Optimize c/zgemv for AMD Bulldozer, Piledriver, Steamroller
981 * Fix bug with scipy linalg test.
984 * Improve DGEMM for ARM Cortex-A57. (Thanks, Ashwin Sekhar T K)
987 * Optimize D and Z BLAS3 functions for Power8.
989 ====================================================================
993 * Upgrade LAPACK to 3.6.0 version.
994 Add BUILD_LAPACK_DEPRECATED option in Makefile.rule to build
995 LAPACK deprecated functions.
996 * Add MAKE_NB_JOBS option in Makefile.
997 Force number of make jobs.This is particularly
998 useful when using distcc. (#735. Thanks, Jerome Robert.)
999 * Redesign unit test. Run unit/regression test at every build (Travis-CI and Appveyor).
1000 * Disable multi-threading for small size swap and ger. (#744. Thanks, Jerome Robert)
1001 * Improve small zger, zgemv, ztrmv using stack alloction (#727. Thanks, Jerome Robert)
1002 * Let openblas_get_num_threads return the number of active threads.
1003 (#760. Thanks, Jerome Robert)
1004 * Support illumos(OmniOS). (#749. Thanks, Lauri Tirkkonen)
1005 * Fix LAPACK Dormbr, Dormlq bug. (#711, #713. Thanks, Brendan Tracey)
1006 * Update scipy benchmark script. (#745. Thanks, John Kirkham)
1009 * Optimize trsm kernels for AMD Bulldozer, Piledriver, Steamroller.
1010 * Detect Intel Avoton.
1011 * Detect AMD Trinity, Richland, E2-3200.
1012 * Fix gemv performance bug on Mac OSX Intel Haswell.
1013 * Fix some bugs with CMake and Visual Studio
1016 * Support and optimize Cortex-A57 AArch64.
1017 (#686. Thanks, Ashwin Sekhar TK)
1018 * Fix Android build on ARMV7 (#778. Thanks, Paul Mustiere)
1019 * Update ARMV6 kernels.
1022 * Fix detection of POWER architecture
1023 (#684. Thanks, Sebastien Villemot)
1025 ====================================================================
1029 * Support cmake on x86/x86-64. Natively compiling on MS Visual Studio.
1030 (experimental. Thank Hank Anderson for the initial cmake porting work.)
1032 On Linux and Mac OSX, OpenBLAS cmake supports assembly kernels.
1035 make test (Optional)
1037 On Windows MS Visual Studio, OpenBLAS cmake only support C kernels.
1038 (OpenBLAS uses AT&T style assembly, which is not supported by MSVC.)
1039 e.g. cmake -G "Visual Studio 12 Win64" .
1040 Open OpenBLAS.sln and build.
1042 * Enable MAX_STACK_ALLOC flags by default.
1043 Improve ger and gemv for small matrices.
1044 * Improve gemv parallel with small m and large n case.
1045 * Improve ?imatcopy when lda==ldb (#633. Thanks, Martin Koehler)
1046 * Add vecLib benchmarks (#565. Thanks, Andreas Noack.)
1047 * Fix LAPACK lantr for row major matrices (#634. Thanks, Dan Kortschak)
1048 * Fix LAPACKE lansy (#640. Thanks, Dan Kortschak)
1049 * Import bug fixes for LAPACKE s/dormlq, c/zunmlq
1050 * Raise the signal when pthread_create fails (#668. Thanks, James K. Lowden)
1051 * Remove g77 from compiler list.
1052 * Enable AppVeyor Windows CI.
1055 * Support pure C generic kernels for x86/x86-64.
1056 * Support Intel Boardwell and Skylake by Haswell kernels.
1057 * Support AMD Excavator by Steamroller kernels.
1058 * Optimize s/d/c/zdot for Intel SandyBridge and Haswell.
1059 * Optimize s/d/c/zdot for AMD Piledriver and Steamroller.
1060 * Optimize s/d/c/zapxy for Intel SandyBridge and Haswell.
1061 * Optimize s/d/c/zapxy for AMD Piledriver and Steamroller.
1062 * Optimize d/c/zscal for Intel Haswell, dscal for Intel SandyBridge.
1063 * Optimize d/c/zscal for AMD Bulldozer, Piledriver and Steamroller.
1064 * Optimize s/dger for Intel SandyBridge.
1065 * Optimize s/dsymv for Intel SandyBridge.
1066 * Optimize ssymv for Intel Haswell.
1067 * Optimize dgemv for Intel Nehalem and Haswell.
1068 * Optimize dtrmm for Intel Haswell.
1071 * Support Android NDK armeabi-v7a-hard ABI (-mfloat-abi=hard)
1072 e.g. make HOSTCC=gcc CC=arm-linux-androideabi-gcc NO_LAPACK=1 TARGET=ARMV7
1073 * Fix lock, rpcc bugs (#616, #617. Thanks, Grazvydas Ignotas)
1075 * Support ppc64le platform (ELF ABI v2. #612. Thanks, Matthew Brandyberry.)
1076 * Support POWER7/8 by POWER6 kernels. (#612. Thanks, Fábio Perez.)
1078 ====================================================================
1082 * Improve OpenBLASConfig.cmake. (#474, #475. Thanks, xantares.)
1083 * Improve ger and gemv for small matrices by stack allocation.
1084 e.g. make -DMAX_STACK_ALLOC=2048 (#482. Thanks, Jerome Robert.)
1085 * Introduce openblas_get_num_threads and openblas_get_num_procs.
1086 (#497. Thanks, Erik Schnetter.)
1087 * Add ATLAS-style ?geadd function. (#509. Thanks, Martin Köhler.)
1088 * Fix c/zsyr bug with negative incx. (#492.)
1089 * Fix race condition during shutdown causing a crash in
1090 gotoblas_set_affinity(). (#508. Thanks, Ton van den Heuvel.)
1093 * Support AMD Streamroller.
1096 * Add Cortex-A9 and Cortex-A15 targets.
1098 ====================================================================
1102 * Add SYMBOLPREFIX and SYMBOLSUFFIX makefile options
1103 for adding a prefix or suffix to all exported symbol names
1104 in the shared library.(#459, Thanks Tony Kelman)
1105 * Provide OpenBLASConfig.cmake at installation.
1106 * Fix Fortran compiler detection on FreeBSD.
1107 (#470, Thanks Mike Nolta)
1111 * Add generic kernel files for x86-64. make TARGET=GENERIC
1112 * Fix a bug of sgemm kernel on Intel Sandy Bridge.
1113 * Fix c_check bug on some amd64 systems. (#471, Thanks Mike Nolta)
1116 * Support APM's X-Gene 1 AArch64 processors.
1117 Optimize trmm and sgemm. (#465, Thanks Dave Nuechterlein)
1119 ====================================================================
1123 * Added CBLAS interface for ?omatcopy and ?imatcopy.
1124 * Enable ?gemm3m functions.
1125 * Added benchmark for ?gemm3m.
1126 * Optimized multithreading lower limits.
1127 * Disabled SYMM3M and HEMM3M functions
1128 because of segment violations.
1131 * Improved axpy and symv performance on AMD Bulldozer.
1132 * Improved gemv performance on modern Intel and AMD CPUs.
1134 ====================================================================
1138 * Added some benchmark codes.
1139 * Fix link error on Linux/musl.(Thanks Isaac Dunham)
1142 * Improved s/c/zgemm performance for Intel Haswell.
1143 * Improved s/d/c/zgemv performance.
1144 * Support the big numa machine.(EXPERIMENT)
1147 * Fix detection when cpuinfo uses "Processor". (Thanks Isaiah)
1149 ====================================================================
1153 * Added BLAS extensions as following.
1154 s/d/c/zaxpby, s/d/c/zimatcopy, s/d/c/zomatcopy.
1155 * Added OPENBLAS_CORETYPE environment for dynamic_arch. (a86d34)
1156 * Added NO_AVX2 flag for old binutils. (#401)
1157 * Support outputing the CPU corename on runtime.(#407)
1158 * Patched LAPACK to fix bug 114, 117, 118.
1159 (http://www.netlib.org/lapack/bug_list.html)
1160 * Disabled ?gemm3m for a work-around fix. (#400)
1162 * Fixed lots of bugs for optimized kernels on sandybridge,Haswell,
1163 bulldozer, and piledriver.
1164 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
1167 * Improved LAPACK testing.
1169 ====================================================================
1173 * Improved the result for LAPACK testing. (#372)
1174 * Installed DLL to prefix/bin instead of prefix/lib. (#366)
1175 * Build import library on Windows.(#374)
1177 * To improve LAPACK testing, we fallback some kernels. (#372)
1178 https://github.com/xianyi/OpenBLAS/wiki/Fixed-optimized-kernels-To-do-List
1180 ====================================================================
1184 * Added OPENBLAS_VERBOSE environment variable.(#338)
1185 * Make OpenBLAS thread-pool resilient to fork via pthread_atfork.
1186 (#294, Thank Olivier Grisel)
1190 * Detect Intel Haswell for new Macbook.
1192 ====================================================================
1196 * Update LAPACK to 3.5.0 version
1197 * Fixed compatiable issues with Clang and Pathscale compilers.
1200 * Optimization on Intel Haswell.
1201 * Enable optimization kernels on AMD Bulldozer and Piledriver.
1204 * Support ARMv6 and ARMv7 ISA.
1205 * Optimization on ARM Cortex-A9.
1207 ====================================================================
1211 * Support Open64 5.0. (#266)
1212 * Add executable stack markings. (#262, Thank Sébastien Fabbro)
1213 * Respect user's LDFLAGS (Thank Sébastien Fabbro)
1216 * Rollback bulldozer and piledriver kernels to barcelona kernels (#263)
1217 We will fix the compuational error bug in bulldozer and piledriver kernels.
1219 ====================================================================
1223 * Support LSB (Linux Standard Base) 4.1.
1225 * Include LAPACK 3.4.2 source codes to the repo.
1226 Avoid downloading at compile time.
1227 * Add NO_PARALLEL_MAKE flag to disable parallel make.
1228 * Create openblas_get_parallel to retrieve information which
1229 parallelization model is used by OpenBLAS. (Thank grisuthedragon)
1230 * Detect LLVM/Clang compiler. The default compiler is Clang on Mac OS X.
1231 * Change LIBSUFFIX from .lib to .a on windows.
1232 * A work-around for dtrti_U single thread bug. Replace it with LAPACK codes. (#191)
1235 * Optimize c/zgemm, trsm, dgemv_n, ddot, daxpy, dcopy on
1236 AMD Bulldozer. (Thank Werner Saar)
1237 * Add Intel Haswell support (using Sandybridge optimizations).
1239 * Add AMD Piledriver support (using Bulldozer optimizations).
1240 * Fix the computational error in zgemm avx kernel on
1242 * Fix the overflow bug in gemv.
1243 * Fix the overflow bug in multi-threaded BLAS3, getrf when NUM_THREADS
1244 is very large.(#214, #221, #246).
1246 * Support loongcc (Open64 based) compiler for ICT Loongson 3A/B.
1249 * Support Power7 by old Power6 kernels. (#220)
1251 ====================================================================
1255 * Improved OpenMP performance slightly. (d744c9)
1256 * Improved cblas.h compatibility with Intel MKL.(#185)
1257 * Fixed the overflowing bug in single thread cholesky factorization.
1258 * Fixed the overflowing buffer bug of multithreading hbmv and sbmv.(#174)
1261 * Added AMD Bulldozer x86-64 S/DGEMM AVX kernels. (Thank Werner Saar)
1262 We will tune the performance in future.
1263 * Auto-detect Intel Xeon E7540.
1264 * Fixed the overflowing buffer bug of gemv. (#173)
1265 * Fixed the bug of s/cdot about invalid reading NAN on x86_64. (#189)
1269 ====================================================================
1273 * Added NO_SHARED flag to disable generating the shared library.
1274 * Compile LAPACKE with ILP64 modle when INTERFACE64=1 (#158)
1275 * Export LAPACK 3.4.2 symbols in shared library. (#147)
1276 * Only detect the number of physical CPU cores on Mac OSX. (#157)
1277 * Fixed NetBSD build. (#155)
1278 * Fixed compilation with TARGET=GENERIC. (#160)
1280 * Restore the original CPU affinity when calling
1281 openblas_set_num_threads(1) (#153)
1282 * Fixed a SEGFAULT bug in dgemv_t when m is very large.(#154)
1285 ====================================================================
1289 * Upgraded LAPACK to 3.4.2 version. (#145)
1290 * Provided support for passing CFLAGS, FFLAGS, PFLAGS,
1291 FPFLAGS to make. (#137)
1292 * f77blas.h:compatibility for compilers without C99 complex
1293 number support. (#141)
1295 * Added NO_AVX flag. Check OS supporting AVX on runtime. (#139)
1296 * Fixed zdot incompatibility ABI issue with GCC 4.7 on
1297 Windows 32-bit. (#140)
1299 * Fixed the generation of shared library bug.
1300 * Fixed the detection bug on the Loongson 3A server.
1301 ====================================================================
1305 * Fixed LAPACK unstable bug about ?laswp. (#130)
1306 * Fixed the shared library bug about unloading the library on
1308 * Fixed the compilation failure on BlueGene/P (TARGET=PPC440FP2)
1309 Please use gcc and IBM xlf. (#134)
1311 * Supported goto_set_num_threads and openblas_set_num_threads
1312 APIs in Windows. They can set the number of threads on runtime.
1314 ====================================================================
1318 * Fixed exporting DLL functions bug on Windows/MingW
1319 * Support GNU Hurd (Thank Sylvestre Ledru)
1320 * Support kfreebsd kernel (Thank Sylvestre Ledru)
1322 * Support Intel Sandy Bridge 22nm desktop/mobile CPU
1324 * Improve the detection of SPARC (Thank Sylvestre Ledru)
1326 ====================================================================
1331 * Fixed the SEGFAULT bug about hyper-theading
1332 * Support AMD Bulldozer by using GotoBLAS2 AMD Barcelona codes
1334 ====================================================================
1338 * Removed the limitation (64) of numbers of CPU cores.
1339 Now, it supports 256 cores at max.
1340 * Supported clang compiler.
1341 * Fixed some build bugs on FreeBSD
1343 * Optimized Level-3 BLAS on Intel Sandy Bridge x86-64 by AVX instructions.
1344 Please use gcc >= 4.6 or clang >=3.1.
1345 * Support AMD Bobcat by using GotoBLAS2 AMD Barcelona codes.
1347 ====================================================================
1351 * Upgraded LAPACK to 3.4.1 version. (Thank Zaheer Chothia)
1352 * Supported LAPACKE, a C interface to LAPACKE. (Thank Zaheer Chothia)
1353 * Fixed the build bug (MD5 and download) on Mac OSX.
1354 * Auto download CUnit 2.1.2-2 from SF.net with UTEST_CHECK=1.
1355 * Fxied the compatibility issue for compilers without C99 complex number
1356 (e.g. Visual Studio)
1358 * Auto-detect Intel Sandy Bridge Core i7-3xxx & Xeon E7 Westmere-EX.
1359 * Test alpha=Nan in dscale.
1360 * Fixed a SEGFAULT bug in samax on x86 windows.
1362 ====================================================================
1366 * Set soname of shared library on Linux.
1367 * Added LIBNAMESUFFIX flag in Makefile.rule. The user can use
1368 this flag to control the library name, e.g. libopenblas.a,
1369 libopenblas_ifort.a or libopenblas_omp.a.
1370 * Added GEMM_MULTITHREAD_THRESHOLD flag in Makefile.rule.
1371 The lib use single thread in GEMM function with small matrices.
1373 * Used GEMV SSE/SSE2 kernels on x86 32-bit.
1374 * Exported CBLAS functions in Windows DLL.
1376 * Completed Level-3 BLAS optimization on Loongson 3A CPU.
1377 * Improved GEMV performance on Loongson 3A CPU.
1378 * Improved Level-3 BLAS performance on Loongson 3B CPU. (EXPERIMENT)
1380 ====================================================================
1381 Version 0.1 alpha2.5
1384 * Fixed missing "#include <sched.h>" bug on Mac OS X.
1385 Thank Mike Nolta for the patch.
1386 * Upgraded LAPACK to 3.4.0 version
1387 * Fixed a bug on Mac OS X. Don't require SystemStubs on OS X.
1388 SystemStubs does not exist on Lion. Thank Stefan Karpinski.
1389 * Improved README with using OpenMP. Check the internal threads
1390 count less than or equal to omp_get_max_threads()
1392 * Auto-detect Intel Core i6/i7 (Sandy Bridge) CPU with Nehalem assembly kernels
1393 * Fixed some bugs on MingW 64-bit including zgemv, cdot, zdot.
1395 ====================================================================
1396 Version 0.1 alpha2.4
1399 * Fixed a bug about installation. The header file "fblas77.h"
1401 * Fixed #61 a building bug about setting TARGET and DYNAMIC_ARCH.
1402 * Try to handle absolute path of shared library in OSX. (#57)
1403 Thank Dr Kane O'Donnell.
1404 * Changed the installation folder layout to $(PREFIX)/include and
1408 * Fixed #58 zdot/xdot SEGFAULT bug with GCC-4.6 on x86. According
1409 to i386 calling convention, The callee should remove the first
1410 hidden parameter.Thank Mr. John for this patch.
1412 ====================================================================
1413 Version 0.1 alpha2.3
1417 * Added DTB_ENTRIES into dynamic arch setting parameters. Now,
1418 it can read DTB_ENTRIES on runtime. (Refs issue #55 on github)
1420 ====================================================================
1421 Version 0.1 alpha2.2
1425 * Fixed a building bug when DYNAMIC_ARCH=1 & INTERFACE64=1.
1426 (Refs issue #44 on github)
1428 ====================================================================
1429 Version 0.1 alpha2.1
1433 * Stop the build and output the error message when detecting
1434 fortran compiler failed. (Refs issue #42 on github)
1436 ====================================================================
1441 * Fixed blasint undefined bug in <cblas.h> file. Other software
1442 could include this header successfully(Refs issue #13 on github)
1443 * Fixed the SEGFAULT bug on 64 cores. On SMP server, the number
1444 of CPUs or cores should be less than or equal to 64.(Refs issue #14
1446 * Support "void goto_set_num_threads(int num_threads)" and "void
1447 openblas_set_num_threads(int num_threads)" when USE_OPENMP=1
1448 * Added extern "C" to support C++. Thank Tasio for the patch(Refs
1449 issue #21 on github)
1450 * Provided an error message when the arch is not supported.(Refs
1451 issue #19 on github)
1452 * Fixed issue #23. Fixed a bug of f_check script about generating link flags.
1453 * Added openblas_set_num_threads for Fortran.
1454 * Fixed #25 a wrong result of rotmg.
1455 * Fixed a bug about detecting underscore prefix in c_check.
1456 * Print the wall time (cycles) with enabling FUNCTION_PROFILE
1457 * Fixed #35 a build bug with NO_LAPACK=1 & DYNAMIC_ARCH=1
1458 * Added install target. You can use "make install". (Refs #20)
1462 * Fixed #28 a wrong result of dsdot on x86_64.
1463 * Fixed #32 a SEGFAULT bug of zdotc with gcc-4.6.
1464 * Fixed #33 ztrmm bug on Nehalem.
1465 * Work-around #27 the low performance axpy issue with small imput size & multithreads.
1468 * Fixed #28 a wrong result of dsdot on Loongson3A/MIPS64.
1469 * Optimized single/double precision BLAS Level3 on Loongson3A/MIPS64. (Refs #2)
1470 * Optimized single/double precision axpy function on Loongson3A/MIPS64. (Refs #3)
1472 ====================================================================
1477 * Support "make NO_LAPACK=1" to build the library without
1479 * Fixed randomly SEGFAULT when nodemask==NULL with above Linux 2.6.34.
1480 Thank Mr.Ei-ji Nakama providing this patch. (Refs issue #12 on github)
1481 * Added DEBUG=1 rule in Makefile.rule to build debug version.
1482 * Disable compiling quad precision in reference BLAS library(netlib BLAS).
1483 * Added unit testcases in utest/ subdir. Used CUnit framework.
1484 * Supported OPENBLAS_* & GOTO_* environment variables (Pleas see README)
1485 * Imported GotoBLAS2 1.13 BSD version
1488 * On x86 32bits, fixed a bug in zdot_sse2.S line 191. This would casue
1489 zdotu & zdotc failures. Instead, work-around it. (Refs issue #8 #9 on github)
1490 * Modified ?axpy functions to return same netlib BLAS results
1491 when incx==0 or incy==0 (Refs issue #7 on github)
1492 * Modified ?swap functions to return same netlib BLAS results
1493 when incx==0 or incy==0 (Refs issue #6 on github)
1494 * Modified ?rot functions to return same netlib BLAS results
1495 when incx==0 or incy==0 (Refs issue #4 on github)
1496 * Detect Intel Westmere,Intel Clarkdale and Intel Arrandale
1497 to use Nehalem codes.
1498 * Fixed a typo bug about compiling dynamic ARCH library.
1500 * Improve daxpy performance on ICT Loongson 3A.
1501 * Supported ICT Loongson 3A CPU (Refs issue #1 on github)
1502 ====================================================================