rt/snapraid
1
0
Fork 0
mirror of https://github.com/amadvance/snapraid.git synced 2026-10-25 18:40:19 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

Align integer gen and rec loops to 64 bytes
Some checks are pending
Build / build-mac (push) Waiting to run
Details
Build / build-linux (push) Waiting to run
Details
Build / build-arm32 (push) Waiting to run
Details
Build / build-win (push) Waiting to run
Details
Build / build-freebsd (push) Waiting to run
Details
Build / build-musl (push) Waiting to run
Details
CodeQL Advanced / Analyze (c-cpp) (push) Waiting to run
Details
CodeQL Advanced / Analyze (actions) (push) Waiting to run
Details

Browse source
This commit is contained in:
Andrea Mazzoleni 2026-10-05 13:46:22 +02:00
parent 143813a90e
commit 6ae014d6ed
3 changed files with 40 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

6
raid/NOTES
View file

@ -1055,6 +1055,12 @@ The implementation deliberately targets at most six parity levels.
Short list of ideas already tried or rejected
---------------------------------------------
* Loop alignment to 64 bytes:
improves scalar integer routines (raid/int.c) by +4% to +15% by avoiding fetch
and decode stalls on compact loops. Tried on vector SIMD routines (SSE, AVX,
NEON), with no improvement because wide unrolled loops are memory-bound and
cached in the uop cache. Kept only for integer routines.
* AVX512 128-byte generation processing:
tried, can be slower.

19
raid/int.c
View file

@ -9,6 +9,7 @@
*
* Uses 16-byte chunks across four 32-bit words.
*/
__aligned_loops(64)
void raid_gen1_int32(int nd, size_t size, void **vv, int streaming)
{
uint8_t **v = (uint8_t **)vv;
@ -49,6 +50,7 @@ void raid_gen1_int32(int nd, size_t size, void **vv, int streaming)
*
* Uses 32-byte chunks across four 64-bit words.
*/
__aligned_loops(64)
void raid_gen1_int64(int nd, size_t size, void **vv, int streaming)
{
uint8_t **v = (uint8_t **)vv;
@ -193,6 +195,7 @@ static __always_inline void raid_gen2_int64_gen(int nd, size_t size, void **vv,
*
* Uses Horner's method with 8-byte chunks across two 32-bit words.
*/
__aligned_loops(64)
void raid_genz_int32_raid(int nd, size_t size, void **vv, int streaming)
{
uint8_t **v = (uint8_t **)vv;
@ -250,6 +253,7 @@ void raid_genz_int32_raid(int nd, size_t size, void **vv, int streaming)
*
* Uses Horner's method with 16-byte chunks across two 64-bit words.
*/
__aligned_loops(64)
void raid_genz_int64_raid(int nd, size_t size, void **vv, int streaming)
{
uint8_t **v = (uint8_t **)vv;
@ -307,6 +311,7 @@ void raid_genz_int64_raid(int nd, size_t size, void **vv, int streaming)
*
* Precomputes per-disk multiplication tables and processes 2-byte chunks.
*/
__aligned_loops(64)
void raid_gen2_int8(int nd, size_t size, void **vv, int streaming)
{
uint8_t **v = (uint8_t **)vv;
@ -361,6 +366,7 @@ void raid_gen2_int8(int nd, size_t size, void **vv, int streaming)
*
* Precomputes per-disk multiplication tables and processes 2-byte chunks.
*/
__aligned_loops(64)
void raid_gen3_int8(int nd, size_t size, void **vv, int streaming)
{
uint8_t **v = (uint8_t **)vv;
@ -425,6 +431,7 @@ void raid_gen3_int8(int nd, size_t size, void **vv, int streaming)
*
* Precomputes per-disk multiplication tables and processes 2-byte chunks.
*/
__aligned_loops(64)
void raid_gen4_int8(int nd, size_t size, void **vv, int streaming)
{
uint8_t **v = (uint8_t **)vv;
@ -498,6 +505,7 @@ void raid_gen4_int8(int nd, size_t size, void **vv, int streaming)
*
* Precomputes per-disk multiplication tables and processes 2-byte chunks.
*/
__aligned_loops(64)
void raid_gen5_int8(int nd, size_t size, void **vv, int streaming)
{
uint8_t **v = (uint8_t **)vv;
@ -580,6 +588,7 @@ void raid_gen5_int8(int nd, size_t size, void **vv, int streaming)
*
* Precomputes per-disk multiplication tables and processes 2-byte chunks.
*/
__aligned_loops(64)
void raid_gen6_int8(int nd, size_t size, void **vv, int streaming)
{
uint8_t **v = (uint8_t **)vv;
@ -1008,30 +1017,35 @@ static __always_inline void raid_recX_int8(int nr, int has_p, int *id, int *ip,
}
}
__aligned_loops(64)
void raid_gen2_int32_raid(int nd, size_t size, void **vv, int streaming)
{
(void)streaming;
raid_gen2_int32_gen(nd, size, vv, 2);
}
__aligned_loops(64)
void raid_gen2_int32_aes(int nd, size_t size, void **vv, int streaming)
{
(void)streaming;
raid_gen2_int32_gen(nd, size, vv, 3);
}
__aligned_loops(64)
void raid_gen2_int64_raid(int nd, size_t size, void **vv, int streaming)
{
(void)streaming;
raid_gen2_int64_gen(nd, size, vv, 2);
}
__aligned_loops(64)
void raid_gen2_int64_aes(int nd, size_t size, void **vv, int streaming)
{
(void)streaming;
raid_gen2_int64_gen(nd, size, vv, 3);
}
__aligned_loops(64)
void raid_rec1_int8(int nr, int *id, int *ip, int nd, size_t size, void **vv)
{
BUG_ON(nr != 1);
@ -1042,6 +1056,7 @@ void raid_rec1_int8(int nr, int *id, int *ip, int nd, size_t size, void **vv)
raid_recX_int8(1, 0, id, ip, nd, size, vv);
}
__aligned_loops(64)
void raid_rec2_int8(int nr, int *id, int *ip, int nd, size_t size, void **vv)
{
BUG_ON(nr != 2);
@ -1052,6 +1067,7 @@ void raid_rec2_int8(int nr, int *id, int *ip, int nd, size_t size, void **vv)
raid_recX_int8(2, 0, id, ip, nd, size, vv);
}
__aligned_loops(64)
void raid_rec3_int8(int nr, int *id, int *ip, int nd, size_t size, void **vv)
{
BUG_ON(nr != 3);
@ -1061,6 +1077,7 @@ void raid_rec3_int8(int nr, int *id, int *ip, int nd, size_t size, void **vv)
raid_recX_int8(3, 0, id, ip, nd, size, vv);
}
__aligned_loops(64)
void raid_rec4_int8(int nr, int *id, int *ip, int nd, size_t size, void **vv)
{
BUG_ON(nr != 4);
@ -1070,6 +1087,7 @@ void raid_rec4_int8(int nr, int *id, int *ip, int nd, size_t size, void **vv)
raid_recX_int8(4, 0, id, ip, nd, size, vv);
}
__aligned_loops(64)
void raid_rec5_int8(int nr, int *id, int *ip, int nd, size_t size, void **vv)
{
BUG_ON(nr != 5);
@ -1079,6 +1097,7 @@ void raid_rec5_int8(int nr, int *id, int *ip, int nd, size_t size, void **vv)
raid_recX_int8(5, 0, id, ip, nd, size, vv);
}
__aligned_loops(64)
void raid_rec6_int8(int nr, int *id, int *ip, int nd, size_t size, void **vv)
{
BUG_ON(nr != 6);

15
raid/internal.h
View file

@ -61,7 +61,22 @@
* Forced alignment.
*/
#ifndef __aligned
#if defined(__GNUC__) || defined(__clang__)
#define __aligned(a) __attribute__((aligned(a)))
#else
#define __aligned(a)
#endif
#endif
/*
* Forced loops alignment.
*/
#ifndef __aligned_loops
#if defined(__GNUC__) && !defined(__clang__)
#define __aligned_loops(n) __attribute__((optimize("align-loops=" #n)))
#else
#define __aligned_loops(n)
#endif
#endif
/*