pmap.c
ClosedPublic
Actions

Authored by jkim on May 24 2016, 7:43 PM.

Details

Reviewers

kib
alc
jhb

Commits

rS300700: Both Clang and GCC cannot generate efficient reserve_pv_entries().

Summary

As alc pointed out, Clang does not generate optimal popcnt_pc_map_elem_pq().

http://docs.freebsd.org/cgi/mid.cgi?552BFEB2.8040407

I tried many implementations but I was not able to generate optimal code for both Clang and GCC. This patch just implements it in inline assembly without loop or unrolled loop. Also, newly added bit_count() is used instead of bitcount64() for the non-POPCNT case.

Diff Detail

Repository

rS FreeBSD src repository - subversion

Lint

Lint Not Applicable

Unit

Tests Not Applicable

Event Timeline

jkim updated this revision to Diff 16815.May 24 2016, 7:43 PM

jkim retitled this revision from to Optimize reserve_pv_entries() in sys/amd64/amd64/pmap.c.

jkim updated this object.

jkim edited the test plan for this revision. (Show Details)

jkim added reviewers: alc, jhb, kib.

jkim set the repository for this revision to rS FreeBSD src repository - subversion.

Herald added a subscriber: imp. · View Herald TranscriptMay 24 2016, 7:43 PM

kib accepted this revision.May 24 2016, 8:09 PM

kib edited edge metadata.

This revision is now accepted and ready to land.May 24 2016, 8:09 PM

alc added inline comments.May 25 2016, 8:45 PM

sys/amd64/amd64/pmap.c
3150 ↗	(On Diff #16815)	Since this function assumes a certain map size, I'd suggest adding a CTASSERT to that effect, so that a map size change would be detected at compile time.