]> Git Repo - linux.git/commit
Enumerate AVX Vector Neural Network instructions
authorKyung Min Park <[email protected]>
Tue, 5 Jan 2021 00:49:08 +0000 (08:49 +0800)
committerPaolo Bonzini <[email protected]>
Thu, 4 Feb 2021 10:27:16 +0000 (05:27 -0500)
commitb85a0425d8056f3bd8d0a94ecdddf2a39d32a801
tree4d150e857b24b4700b293da0c444bbdbd8af49a2
parent7ca7f3b944929c99637522d849138ba15f97e3fe
Enumerate AVX Vector Neural Network instructions

Add AVX version of the Vector Neural Network (VNNI) Instructions.

A processor supports AVX VNNI instructions if CPUID.0x07.0x1:EAX[4] is
present. The following instructions are available when this feature is
present.
  1. VPDPBUS: Multiply and Add Unsigned and Signed Bytes
  2. VPDPBUSDS: Multiply and Add Unsigned and Signed Bytes with Saturation
  3. VPDPWSSD: Multiply and Add Signed Word Integers
  4. VPDPWSSDS: Multiply and Add Signed Integers with Saturation

The only in-kernel usage of this is kvm passthrough. The CPU feature
flag is shown as "avx_vnni" in /proc/cpuinfo.

This instruction is currently documented in the latest "extensions"
manual (ISE). It will appear in the "main" manual (SDM) in the future.

Signed-off-by: Kyung Min Park <[email protected]>
Signed-off-by: Yang Zhong <[email protected]>
Reviewed-by: Tony Luck <[email protected]>
Message-Id: <20210105004909[email protected]>
Acked-by: Borislav Petkov <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
arch/x86/include/asm/cpufeatures.h
This page took 0.088596 seconds and 4 git commands to generate.