Hello, Thanks for the patch.
On Wed, 4 Mar 2026 at 16:17, P1erreCashon <[email protected]> wrote: > > Implement EFD lookup internal function for RISC-V architecture > using RISC-V Vector Extension instruction set > > Performance results (EFD perf autotest) > These performance results are the average of five test runs. > Results (in CPU cycles/operation): > > Before optimization: > Keysize Add Lookup Lookup_bulk Delete > 4 136.0 5.6 5.0 25.6 > 8 149.4 6.0 6.0 31.4 > 16 156.8 8.0 7.0 36.8 > 32 179.2 13.0 14.0 45.4 > 48 205.8 13.0 15.0 45.6 > 64 214.4 16.0 18.4 46.6 > 9 152.8 6.2 6.0 30.0 > 13 160.6 7.2 7.0 35.0 > 37 191.2 12.0 13.0 44.2 > 40 195.0 12.0 13.6 43.2 > > After optimization: > Keysize Add Lookup Lookup_bulk Delete > 4 131.6 5.0 5.0 26.6 > 8 143.6 6.0 6.2 33.0 > 16 153.2 7.0 7.0 35.8 > 32 173.6 14.0 14.0 45.0 > 48 201.6 12.6 14.0 46.2 > 64 211.4 15.0 16.0 46.8 > 9 146.6 6.0 6.0 30.6 > 13 154.8 7.0 7.0 35.0 > 37 189.2 11.0 12.2 45.4 > 40 191.8 11.0 12.4 43.8 I am not sure what is significant in those numbers. For someone like me who does not run efd test, the difference seems quite small. Could you add more details? > > This patch is co-developed with Gong Xiaofei. > > Signed-off-by: gong-flying <[email protected]> > Signed-off-by: P1erreCashon <[email protected]> Please, use real names in the signatures, like for example: Signed-off-by: Gong Xiaofei <[email protected]> P1erre looks like an alias. -- David Marchand
