Vector Class Discussion

New version 1.14 supports AVX512
Author: Agner Date: 2014-07-24 05:20
The new version 1.14 of the vector class library supports the forthcoming AVX512 instruction set, which is expected to be available in Intel's Knights Landing coprocessor in 2015 and in ordinary CPU's in 2016. This doubles the maximum vector size to 512 bits for both integer and floating point vectors. Boolean vectors for use with the new 512-bit vectors are implemented in a new type of mask registers which are introduced with AVX512.

No CPU with AVX512 is available yet, so it has been tested with Intel's emulator. Compiling for AVX512 works with the newest Gnu and Intel compilers, but not yet with Microsoft and Clang compilers.

Version 1.14 of the vector class library also adds hyperbolic functions and a few other new features. See changelog.txt and the manual for details.

thread New version 1.14 supports AVX512 - Agner - 2014-07-24
last replythread New version 1.14 supports AVX512 new - Peter Cordes - 2016-05-31
last reply New version 1.14 supports AVX512 new - Agner - 2016-05-31