c/mirror_zfs
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_zfs.git synced 2026-05-22 02:27:36 +03:00
Code Issues Packages Projects Releases Wiki Activity

zcommon: add specialized versions of cityhash4

Browse Source 
Specializing cityhash4 on 32-bit architectures can reduce the size
of stack frames as well as instruction count. This is a tiny but
useful optimization, since some callers invoke it frequently.

When specializing into 1/2/3/4-arg versions, the stack usage
(in bytes) on some 32-bit arches are listed as follows:

- x86: 32, 32, 32, 40
- arm-v7a: 20, 20, 28, 36
- riscv: 0, 0, 0, 16
- power: 16, 16, 16, 32
- mipsel: 8, 8, 8, 24

And each actual argument (even if passing 0) contributes evenly
to the number of multiplication instructions generated:

- x86: 9, 12, 15 ,18
- arm-v7a: 6, 8, 10, 12
- riscv / power: 12, 18, 20, 24
- mipsel: 9, 12, 15, 19

On 64-bit architectures, the tendencies are similar. But both stack
sizes and instruction counts are significantly smaller thus negligible.

Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Tino Reichardt <milky-zfs@mcmilk.de>
Signed-off-by: Shengqi Chen <harry-chen@outlook.com>
Closes #16131
Closes #16483
This commit is contained in:
Shengqi Chen
2024-09-07 21:55:03 +08:00
committed by Brian Behlendorf
parent 1c35206124
commit 0ae4460c61
3 changed files with 56 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/cityhash.h
+7
View File
@@ -32,6 +32,13 @@
extern "C" {
#endif
/*
* Define 1/2/3-argument specialized versions of cityhash4, which can reduce
* instruction count (especially multiplication) on some 32-bit arches.
*/
_SYS_CITYHASH_H uint64_t cityhash1(uint64_t);
_SYS_CITYHASH_H uint64_t cityhash2(uint64_t, uint64_t);
_SYS_CITYHASH_H uint64_t cityhash3(uint64_t, uint64_t, uint64_t);
_SYS_CITYHASH_H uint64_t cityhash4(uint64_t, uint64_t, uint64_t, uint64_t);
#ifdef __cplusplus