Add RAID-Z routines for SSE2 instruction set, in x86_64 mode. #4815
Conversation
|
Nice, this LGTM and tested the local overnight testing I threw at it. |
behlendorf
added
the
Type: Performance
|
I pushed some code cleanup. I'm happy with the state of the patch, provided tests come back negative. |
|
I'm running the version before your cleanup an hour ago, but all of the older systems are showing "sse2" as available and say they're using the fastest algorithm. I'm running some disk scrubs right now, but everything looks good. |
@angstymeat This is by design. We run a quick benchmark on module load to find the fastest supported methods. You can switch explicitly to 'sse2' but there's no need. If sse2 is listed, it's used. |
|
I understand that, I was just trying to get across that it looks like sse2 is available and that they are using it. |
|
LGTM. @ironMann why don't you rebase this against master and force update the PR for one last test run. Feel free to add another small patch to address the other issues noted in your 4328 comment |
The patch covers low-end and older x86 CPUs. Parity generation is equivalent to SSSE3 implementation, but reconstruction is somewhat slower. Previous 'sse' implementation is renamed to 'ssse3' to indicate highest instr. set used. Benchmark results: scalar_rec_p 4 720476442 scalar_rec_q 4 187462804 scalar_rec_r 4 138996096 scalar_rec_pq 4 140834951 scalar_rec_pr 4 129332035 scalar_rec_qr 4 81619194 scalar_rec_pqr 4 53376668 sse2_rec_p 4 2427757064 sse2_rec_q 4 747120861 sse2_rec_r 4 499871637 sse2_rec_pq 4 522403710 sse2_rec_pr 4 464632780 sse2_rec_qr 4 319124434 sse2_rec_pqr 4 205794190 ssse3_rec_p 4 2519939444 ssse3_rec_q 4 1003019289 ssse3_rec_r 4 616428767 ssse3_rec_pq 4 706326396 ssse3_rec_pr 4 570493618 ssse3_rec_qr 4 400185250 ssse3_rec_pqr 4 377541245 original_rec_p 4 691658568 original_rec_q 4 195510948 original_rec_r 4 26075538 original_rec_pq 4 103087368 original_rec_pr 4 15767058 original_rec_qr 4 15513175 original_rec_pqr 4 10746357 Issue openzfs#4783 Signed-off-by: Gvozden Neskovic <[email protected]>
|
@behlendorf Rebased, some issues with testers though... |
|
The testing failures were unrelated. Merged as: ae25d22 Add RAID-Z routines for SSE2 instruction set, in x86_64 mode. |
Benchmark results:
Issue #4783