From: WANG Xuerui <[email protected]>
Hi,
Seeing the LoongArch port recently (finally!) gained the ability to use
the vector units, I've subsequently ported the RAID5/6 math to LSX and
LASX (which are LoongArch's 128-bit and 256-bit SIMD extensions), with
nice speedups observed. They are reasonably straight-forward conversions
of existing code, and I hope the comments I put in there are helpful
enough for anyone not familiar with LoongArch assembly to get a rough
picture of how things work here. Performance numbers are included in
each commit's commit message.
This series needs [1] ("LoongArch: Allow usage of LSX/LASX in the
kernel") as a prerequisite, or the vector context would likely get
corrupted by the vector-unaware kernel_fpu_{begin,end} calls. I tested
the changes on top of next-20230731 with the raid6test build fixes [2]
applied, but the series should apply cleanly to v6.5-rc4 (or maybe any
other tag) too; it doesn't depend on the raid6test fixes.
The base-commit and prerequisite-patch-id info is available for
minimally recreating a working configuration for both the kernel and
raid6test tool.
[1]: https://lore.kernel.org/loongarch/[email protected]/
[2]: https://lore.kernel.org/linux-raid/[email protected]/
Changes in v2 (no functional change):
- minor commit message tweaks
- changed comment style from // to /* */ throughout
WANG Xuerui (3):
LoongArch: Add SIMD-optimized XOR routines
raid6: Add LoongArch SIMD syndrome calculation
raid6: Add LoongArch SIMD recovery implementation
arch/loongarch/include/asm/xor.h | 68 ++++
arch/loongarch/include/asm/xor_simd.h | 42 +++
arch/loongarch/lib/Makefile | 3 +
arch/loongarch/lib/xor_simd.c | 93 +++++
arch/loongarch/lib/xor_simd.h | 46 +++
arch/loongarch/lib/xor_simd_glue.c | 71 ++++
arch/loongarch/lib/xor_template.c | 110 ++++++
include/linux/raid/pq.h | 4 +
lib/raid6/Makefile | 1 +
lib/raid6/algos.c | 16 +
lib/raid6/loongarch.h | 38 ++
lib/raid6/loongarch_simd.c | 422 +++++++++++++++++++++
lib/raid6/recov_loongarch_simd.c | 515 ++++++++++++++++++++++++++
lib/raid6/test/Makefile | 12 +
14 files changed, 1441 insertions(+)
create mode 100644 arch/loongarch/include/asm/xor.h
create mode 100644 arch/loongarch/include/asm/xor_simd.h
create mode 100644 arch/loongarch/lib/xor_simd.c
create mode 100644 arch/loongarch/lib/xor_simd.h
create mode 100644 arch/loongarch/lib/xor_simd_glue.c
create mode 100644 arch/loongarch/lib/xor_template.c
create mode 100644 lib/raid6/loongarch.h
create mode 100644 lib/raid6/loongarch_simd.c
create mode 100644 lib/raid6/recov_loongarch_simd.c
base-commit: 5d0c230f1de8c7515b6567d9afba1f196fb4e2f4
prerequisite-patch-id: 85d08a9828893250ae78dbca9d6e6f8dac755f61
prerequisite-patch-id: fe0bba41e0bbc676454365ed16fb13fc0aac6ee0
prerequisite-patch-id: 84ef8212b74e696ce019255bbfd9679d7516f7f7
prerequisite-patch-id: b1f8fc4e4acdaff7f821a9fcbd063475178e037b
prerequisite-patch-id: 82aacbf27f249fdefe40dd6bcc712e5795256926
prerequisite-patch-id: ae4e026e18f92ffcc93f6b135a3bd48fbdded39a
--
2.40.0