ClickHouse/libs/libvectorclass/changelog.txt
2014-12-02 20:25:10 +03:00

111 lines
4.8 KiB
Plaintext

change log for vectorclass.zip
------------------------------
2015-10-24 version 1.16
* workaround for problem in Clang compiler extended to version 3.09 because not fixed yet by Clang
(vectorf128.h line 134)
* recognize problem with Apple version of Clang reporting wrong version number
* remove various minor problems with Clang
* function pow(vector, int) modified to strengthen type checking and avoid compiler warnings
* manual discusses dynamic allocation of arrays of vectors
* various minor changes
2015-10-17 version 1.15
* added files ranvec1.h and ranvec1.cpp for random number generator
* constructors to make boolean vectors from their elements
* constructors and = operators to broadcast boolean scalar into boolean vectors
* various lookup functions improved
* operators &, |, ^, ~, etc. defined for various boolean vectors to avoid converson
to integer vectors
* nmul_add functions
* mul_add etc. moved to main header files
* explicit fused multiply-and-add used in math functions to improve performance
on compilers that don't automatically insert FMA
2014-07-24 version 1.14
* support for AVX-512f instruction set and 512-bit vectors:
Vec16i, Vec16ui, Vec8q, Vec8uq, Vec16f, Vec8d, and corresponding boolean vectors
* new define MAX_VECTOR_SIZE, valid values are 128, 256 and 512
* added hyperbolic functions sinh, cosh, tanh, asinh, acosh, atanh
* size() member function on all vector classes returns the number of elements
* functions for conversion between boolean vectors and integer bitfields
* extracting an element from a boolean vector now returns a bool, not an int
* improved precision in exp2 and exp10 functions
* various bug fixes
2014-05-11 version 1.13
* pow function improved
* mul_add, mul_sub, mul_sub_x functions
* propagation of error codes through nan_code function
* "denormal" renamed to "subnormal" everywhere, in accordance with IEEE 754-2008 standard
2014-04-20 version 1.12
* inline implementation of mathematical functions added (vectormath_exp.h vectormath_trig.h vectormath_common.h)
* vectormath.h renamed to vectormath_lib.h because a new alternative is added
* gather functions with constant indexes
* function sign_combine
* function pow_const(vector, const int)
* function pow_ratio(vector, const int, const int)
* functions horizontal_find_first, horizontal_count
* function recipr_sqrt removed
* functions round_to_int64_limited, truncate_to_int64_limited, to_double_limited
* function cubic_root renamed to cbrt
* function atan(vector,vector) renamed to atan2
* function if_mul
* function Vec4i round_to_int(Vec2d)
* operator & (float vector, boolean vector)
* operator &= (int vector, int vector)
* removed constructor Vec128b(int) and Vec256b(int) to avoid implicit conversion
* removed signalling nan function
* minor improvements in various blend and lookup functions
2014-03-01 version 1.11
* fixed missing unsigned operators >>= in vectori256.h
2013-10-04 version 1.10
* clear distinction between boolean vectors and integer vectors for the sake of
compatibility with mask registers in forthcoming AVX512 instruction set
* added function if_add
* tentative support for clang version 3.3 with workaround for bugs
* remove ambiguity for builtin m128i operator == in clang compiler.
* problems in clang compiler, bug reports filed at clang
(http://llvm.org/bugs/show_bug.cgi?id=17164, 17312)
* instrset.h fixes problem with macros named min and max in MS windows.h
* workaround problem in MS Visual Studio 11.0. Bug report 735861 and 804274
* minor bug fixes
2013-03-31 version 1.03 beta
* bug fix for Vec2d cos (Vec2d const & x), VECTORMATH = 1
2012-08-01 version 1.02 beta
* added file vector3d.h for 3-dimensional vectors
* added file complexvec.h for complex numbers and complex vectors
* added file quaternion.h for quaternions
* added function change_sign for floating point vectors
* added operators +, -, *, / between floating point vectors and scalars to remove overloading ambiguity
2012-07-08 version 1.01 beta
* added file decimal.h with Number <-> string conversion functions:
bin2bcd, bin2ascii, bin2hex_ascii, ascii2bin
* added andnot function for boolean vectors
* added functions shift_bytes_up and shift_bytes_down
* added operators for unsigned integer vector classes: >>=, &, &&, |, ||, ^, ~
* inteldispatchpatch.cpp removed. Use asmlib instead (www.agner.org/optimize/#asmlib)
* prefix ++ and -- operators now return a reference, postfix operators return a value
* various improvements in permute and blend functions
* minor improvement in abs function
* added version number to VECTORCLASS_H
2012-05-30 version 1.00 beta
* first public release