GC support for stack allocated objects #12205
Conversation
|
oh and the way I get the current task stack's bounds is kinda gore, I think we have at least the bottom registered somewhere but I didn't feel like going through the whole init thing just yet. I had a look and saw that it looks like we support stacks growing up ? |
the code for that is not really functional, since there is no use case for it
take a look at gc_mark_task_stack, either on master or https://github.com/JuliaLang/julia/pull/9461/files#diff-31b93f4272eaeede3738531e90317e0cR1499 the stack is either |
vtjnash
force-pushed
the
jn/codegen_rewrite
branch
from
328a752 to
9f9ecf0
Compare
vtjnash
force-pushed
the
jn/codegen_rewrite
branch
6 times, most recently
from
22c4315 to
8da2872
Compare
vtjnash
force-pushed
the
jn/codegen_rewrite
branch
from
8da2872 to
83019e2
Compare
|
👍 (exciting this would help with #14955). Looks like this could use a rebase now that |
|
Would this help for #9174 too? |
The font used on ReadTheDocs made the octal examples appear to be `OO777` or `00777` rather than `0o777`.
Improve svdfact(A, B) docs
Add setindex! for SymTridiagonal, Tridiagonal, and Bidiagonal
hey, every bit helps.
Made octal examples easier to read on ReadTheDocs
* Remove the usage of ia_misc.h, (re)define rdtsc * Fix the usage of __builtin_assume_aligned * Replace tabs with spaces
RFC: Fast, lispy views
when the matrices have the same size we can get away with simple copying of the internal fields.
fixes `%A` formatting (#14409)
This takes advantage of the simplifications in the previous commit to reduce the complexity in this script. Outwardly the only difference should be the formatting of the summary that is displayed and speed (where timings drop from ~18 to ~9 seconds.)
This commit saves some of the values computed by `doc` and `summarise` into the generated `MD` object that is returned. These valus are the `Binding` and signature searched for, as well as the vector of `DocStr` objects that match the `Binding` and signature.
Removed mention of current working directory.
Check stride on preallocated output for matmul (fixes #15286)
Ellide immutable allocation in some simple cases
put only values in static parameter environments
… relative errors of each type.
Fix type instability in nextprod
part of jb/linear
Set `isapprox` default relative error to be the larger of the default relative errors of each type.
Fix check of inner dimension in matmul for 2x2 and 3x3 outer dimension
Couple ispointer fixes
Refactor docsystem internals.
Only works with copy_stacks for now. The code is not used anywhere in codegen right now, it can be tested by using (:stknew typ args...) instead of :new. It is essentially the GC side of #8134 but without the global alloca stack & unmarking. Stack objects are kept marked (+young) permanently and are only scanned when coming from the owning task.
Only works with copy_stacks for now. The code is not used anywhere in
codegen right now, it can be tested by using (:stknew typ args...)
instead of :new. It is essentially the GC side of #8134 but without the
global alloca stack & unmarking, so @JeffBezanson will probably be happier.
Stack objects are kept marked (+young) permanently and are only scanned
when coming from the owning task.
Note that this is on top of jn/codegen_rewrite under the "no codegen change policy" :-)
cc @vtjnash