diff options
author | David Blaikie <dblaikie@gmail.com> | 2017-06-29 02:51:58 +0000 |
---|---|---|
committer | David Blaikie <dblaikie@gmail.com> | 2017-06-29 02:51:58 +0000 |
commit | d16a61d2c84b73144f232dcd75b6c5e03f9f898c (patch) | |
tree | 852c3fa067fdb423874dcc8f6f9239c697a8a272 /llvm/lib/ToolDrivers | |
parent | 1631129834a06f5a1438288fd25fc984fc40862d (diff) | |
download | bcm5719-llvm-d16a61d2c84b73144f232dcd75b6c5e03f9f898c.tar.gz bcm5719-llvm-d16a61d2c84b73144f232dcd75b6c5e03f9f898c.zip |
llvm-profdata: Indirect infrequently used fields to reduce memory usage
Examining a large profile example, it seems relatively few records have
non-empty IndirectCall and MemOP data, so indirecting these through a
unique_ptr (non-null only when they are non-empty) Reduces memory usage
on this particular example from 14GB to 10GB according to valgrind's
massif.
I suspect it'd still be worth moving InstrProfWriter to its own data
structure that had Counts and the indirected IndirectCall+MemOP, and did
not include the Name, Hash, or Error fields. This would reduce the size
of this dominant data structure by half of this new, lower amount.
(Name(2), Hash(1), Error(1) ~= Counts(vector, 3), ValueProfData
(unique_ptr, 1))
-> From code review feedback, might actually refactor InstrProfRecord
itself to have a sub-struct with all the counts, and use that from
InstrProfWriter, rather than InstrProfWriter owning its own data
structure for this.
Reviewers: davidxl
Differential Revision: https://reviews.llvm.org/D34694
llvm-svn: 306631
Diffstat (limited to 'llvm/lib/ToolDrivers')
0 files changed, 0 insertions, 0 deletions