c/mirror_zfs
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_zfs.git synced 2026-05-22 18:40:43 +03:00
Code Issues Packages Projects Releases Wiki Activity

Fletcher4 algorithm implemented in pure NEON for Aarch64 / ARMv8 64 bits

Browse Source 
This is not useful on micro-architecture with a weak NEON
implementation (only 64 bits); the native version is slower &
the byteswap barely faster than scalar.  On A53 or A57, it's
a small improvement on scalar but OK for byteswap.

Results from an A53 system:
0 0 0x01 -1 0 1499068294333000 1499101101878000
implementation   native         byteswap       
scalar           1008227510     755880264      
aarch64_neon     1198098720     1044818671     
fastest          aarch64_neon   aarch64_neon 

Results from a A57 system:
0 0 0x01 -1 0 4407214734807033 4407233933777404
implementation   native         byteswap       
scalar           2302071241     1124873346     
aarch64_neon     2542214946     2245570352     
fastest          aarch64_neon   aarch64_neon 

Reviewed-by: Gvozden Neskovic <neskovic@gmail.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Romain Dolbeau <romain.dolbeau@atos.net>
Closes #5248
This commit is contained in:
Romain Dolbeau
2016-10-21 19:55:49 +02:00
committed by Brian Behlendorf
parent e4ffa98dca
commit 24cdeaf12e
6 changed files with 232 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/zfs_fletcher.h
+11
View File
@@ -77,6 +77,10 @@ typedef struct zfs_fletcher_avx512 {
uint64_t v[8] __attribute__((aligned(64)));
} zfs_fletcher_avx512_t;
typedef struct zfs_fletcher_aarch64_neon {
uint64_t v[2] __attribute__((aligned(16)));
} zfs_fletcher_aarch64_neon_t;
typedef union fletcher_4_ctx {
zio_cksum_t scalar;
@@ -90,6 +94,9 @@ typedef union fletcher_4_ctx {
#if defined(__x86_64) && defined(HAVE_AVX512F)
zfs_fletcher_avx512_t avx512[4];
#endif
#if defined(__aarch64__)
zfs_fletcher_aarch64_neon_t aarch64_neon[4];
#endif
} fletcher_4_ctx_t;
/*
@@ -128,6 +135,10 @@ extern const fletcher_4_ops_t fletcher_4_avx2_ops;
extern const fletcher_4_ops_t fletcher_4_avx512f_ops;
#endif
#if defined(__aarch64__)
extern const fletcher_4_ops_t fletcher_4_aarch64_neon_ops;
#endif
#ifdef __cplusplus
}
#endif