projects / platform / upstream / openblas.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 544b069) | patch
Create a AVX512 enabled version of DGEMM
authorArjan van de Ven <arjan@linux.intel.com>
Wed, 3 Oct 2018 14:45:25 +0000 (14:45 +0000)
committerArjan van de Ven <arjan@linux.intel.com>
Wed, 3 Oct 2018 14:45:25 +0000 (14:45 +0000)
commit45fe8cb0c5d06f890913e86078cb48ac379c65dc
tree5670df28701a51eaae865bbbc5950eca295b6886tree | snapshot
parent544b069e85254d41699afde16e2e81c123cb5f28commit | diff
Create a AVX512 enabled version of DGEMM

This patch adds dgemm_kernel_4x8_skylakex.c which is
* dgemm_kernel_4x8_haswell.s converted to C + intrinsics
* 8x8 support added
* 8x8 kernel implemented using AVX512

Performance is a work in progress, but already shows a 10% - 20%
increase for a wide range of matrix sizes.
kernel/x86_64/KERNEL.SKYLAKEX diff | blob | history
kernel/x86_64/dgemm_kernel_4x8_skylakex.c [new file with mode: 0644] blob
Domain: Machine Learning / ML Framework;