Small fixes to enable LTO compilation
To do this in racket/src:
./configure CFLAGS="-O3 -march=native -flto"
LDFLAGS="-O3 -march=native -flto"
--prefix=...
One fix deals with variable `errnum` that might be used uninitialized.
The other is to ensure that `boot_file_data` is marked as used
otherwise, it is removed by LTO and ends up crashing build
in `embed-boot.rkt`.
Instead of having leftover Scheme threads swept by the main thread,
distribute Scheme threads among sweeper threads, and swap into each
thread's content to preserve its allocation ownership. This change
makes parallelism more consistent for different timings that end up
assigning sweepers differently.
Move per-thread allocation information to a separate object, so the
needs of the collector are better separated from the thread
representation. This refactoring sets up a change in the collector to
detangle sweeper threads from Scheme threads.
Change the internal parallelism strategy for the GC to record an owner
for each allocated segment of memory, and have the owner be solely
responsible for copying or marking objects of the segment. When
sweeping, a collecting thread handles references to objects that it
owns or that have been copied or marked already, and it asks another
collecting thread to resweep an object that refers to objects owned by
that that thread. At worst, an object ends up being swept by all
collecting threads, one at a time, but that's unlikely for a given
object.
The approach seems likely to scale better than a lock-based approach,
even the one that used a lightweight, CAS-based lock and retries on
lock failure.
That is, use pairs on the property in more places, as the pair
already was computed and the value-blame function already does
the needful when it sees a pair on the property.
When the GC needs to copy/mark a record, it previously forced a
copy/mark on the record's type descriptor, since the copy/mark needs
information from the descriptor. But the needed information is not in
danger of being overwritten for forwading (since it's after the first
two words of the type descriptor), so it's ok to use the old reference
as-is --- at least in non-counting mode. Simplifying record-type
handling and deferring the record-type update to the sweep phase, the
same as for other components of the record type, makes the GC slightly
faster.
Fall back to `current-directory` when `current-load-relative-directory`
is #f.
This change also affects `deserialize` --- not because byte code
loading uses it directly in this case, but because they share a helper
function, which exposes the issue. This implementation change is
worrying (even though it makes the implementation match the
documentation), but unless we discover that some use of serialization
needs absolute paths deseialized as relative, is seems better to be
consistent everywhere about falling back to `currenrt-directory`. This
aspect of the change can be reverted separately (by adding more code)
if needed.
Closesracket/drracket#421
All allocation is now thread-local, which recovers a small bit of
performance that was lost when adding thread-local allocation
alongside global allocation.
Parallelism uses thread contexts created by Chez Scheme threads (which
correspond to Racket places and future-running threads), but it
creates its own OS-level threads to perform collection. The number of
collection-helper threads is limited to the number of active Chez
Scheme threads. Only the main "sweep" pass runs in parallel --- that
is, after roots have been traversed, and before weak references and
finalization are handled --- but that's the bulk of collection work.
Also, memory-accounting collections always run as single-threaded.
Improved support for thread-location allocation (and using more
fine-grained locks in fasl reading) may provide a small direct
benefit, but the change is mainly intended as setup for more
parallelism in the collector.
