rt/snapraid
1
0
Fork 0
mirror of https://github.com/amadvance/snapraid.git synced 2026-10-26 19:01:57 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

Preliminary support for AVX512GFNI

Browse source
This commit is contained in:
Andrea Mazzoleni 2026-05-11 11:07:01 +02:00
parent a043ab2257
commit ade2c6c96d
15 changed files with 958 additions and 80 deletions
Download patch file Download diff file Expand all files Collapse all files

42
raid/raid.h
View file

@ -4,6 +4,47 @@
#ifndef __RAID_H
#define __RAID_H
/*
* GF(2^8) reducing polynomial selection.
*
* RAID parity computation operates in GF(2^8), a finite field where
* arithmetic is performed modulo an irreducible polynomial of degree 8.
* The choice of polynomial affects both correctness and performance:
* all tables, coefficients, and parity data are tied to a specific
* polynomial and cannot be mixed.
*
* Two polynomials are supported:
*
* 0x1d (x^8 + x^4 + x^3 + x^2 + 1) -- Standard RAID polynomial
*
* The polynomial used by the original RAID-6 specification and the
* Linux kernel RAID implementation. Choosing this polynomial ensures
* on-disk compatibility with existing RAID arrays and interchange with
* other RAID implementations. Use this when reading or writing data
* that may be accessed by other software or hardware RAID controllers.
*
* 0x1b (x^8 + x^4 + x^3 + x + 1) -- AES polynomial (USE_RAID_AES)
*
* The polynomial used by the AES encryption standard. Choosing this
* polynomial enables the use of the Intel GFNI (Galois Field New
* Instructions) extension, specifically the vgf2p8mulb instruction,
* which performs GF(2^8) multiplication natively in hardware using
* exactly this polynomial. This eliminates the nibble-split pshufb
* multiply sequence (vpsrlw + vpandq + 2x vpshufb + vpxorq) and the
* conditional multiply-by-2 sequence (vpcmpgtb + vpmovm2b + vpaddb +
* vpandq + vpxorq) used in the AVX-512BW path, replacing each with a
* single vgf2p8mulb instruction. On CPUs supporting GFNI (Ice Lake and
* later), this yields a significant throughput improvement. The
* trade-off is that parity data is not compatible with arrays using
* the standard 0x1d polynomial, so this option requires that all
* data be generated and recovered using the same polynomial choice.
*/
#ifdef USE_RAID_AES
#define RAID_POLY 0x1b
#else
#define RAID_POLY 0x1d
#endif
/**
* RAID mode supporting up to 6 parities.
*
@ -215,4 +256,3 @@ int raid_check(int nr, int *ir, int nd, int np, size_t size, void **v);
int raid_scan(int *ir, int nd, int np, size_t size, void **v);
#endif