aboutsummaryrefslogtreecommitdiff
path: root/src/boot/be
Commit message (Collapse)AuthorAgeFilesLines
* When vec growth results in a newly allocated (extended) buffer, copy ↵Roy Frostig2010-09-031-2/+2
| | | | existing elements over via element-wise copy, not flat memcpy. Introduce new vec growth glue to achieve this.
* Whitespace churn.Graydon Hoare2010-08-311-1/+1
|
* Make vreg constrs per-quad, regfence on nontrivial constrs, back out ↵Graydon Hoare2010-08-261-46/+61
| | | | workaround to _uint, add regression test. Closes #152.
* Add check to catch bug underlying issue #152. Fix will take longer.Graydon Hoare2010-08-261-1/+14
|
* Add element to closure to hold captured tydesc (not body tydesc).Graydon Hoare2010-08-251-3/+4
|
* First pass of updating the in-memory layout of closures, for issue 81.Graydon Hoare2010-08-251-8/+9
|
* Janitorial work on obj box / body / field terminology, following froystig's ↵Graydon Hoare2010-08-252-4/+6
| | | | brave lead.
* Export all item code to stabs on Windows (including e.g. object methods)Patrick Walton2010-08-191-3/+5
|
* Export glue as stabs on Windows as wellPatrick Walton2010-08-191-5/+13
|
* Add some mangled stabs to help debugging on WindowsPatrick Walton2010-08-191-2/+18
|
* Zero LMA/VMA of debug sections in elf. Closes #148.Graydon Hoare2010-08-121-105/+42
|
* Some ELF correctness issues, but apparently none enough to placate gdb.Graydon Hoare2010-08-111-25/+50
|
* Put out burning tinderbox on OSX.Graydon Hoare2010-08-101-8/+19
|
* Always bounce mul/div/mod ops. Closes #131 harder.Graydon Hoare2010-08-101-23/+6
|
* Get object's captured typarams when calculating sizes in the backend's ↵Roy Frostig2010-08-093-8/+63
| | | | fn-prologue-generation for object methods.
* Changed seemingly incorrect stk_field_valgrind_id.Michael Bebenita2010-08-091-1/+1
|
* Degrade emitter size cache to just a flat hashtable with regular flushes ↵Graydon Hoare2010-08-052-8/+38
| | | | (sigh) and re-introduce horrible bounce-off-spill hack for DIV, MUL, etc.
* Move 'as' precedence up to just above relational; support indexing str and ↵Graydon Hoare2010-08-051-7/+12
| | | | vec by all integral types. Closes #94.
* Kill the preallocator, install a sane replacement. Closes #131. And probably ↵Graydon Hoare2010-08-043-261/+115
| | | | a lot of others.
* Factor append_quad out of IL.emit_full, for use elsewhere.Graydon Hoare2010-08-031-4/+12
|
* Remove dead implicit-destinations logic from IL.Graydon Hoare2010-08-035-20/+11
|
* Patch up another place where the newly-expanded 'closure-ptr' implicit arg ↵Graydon Hoare2010-07-301-1/+2
| | | | needs to be passed. Amazingly, it mostly-worked without this. On some platforms.
* Slight shaving on RA, no more optimizing today. It's fast enough for now.Graydon Hoare2010-07-261-8/+9
|
* Do some more iflog-guarding.Graydon Hoare2010-07-261-61/+43
|
* Don't write to NULL after calling C natives returning void.Roy Frostig2010-07-252-4/+7
|
* Notify copy glue of dst-initialization and fix _vec.alloc issues in lib and ↵Roy Frostig2010-07-221-2/+3
| | | | runtime. Closes #109.
* Added a message passing system based on lock free queues for inter-thread ↵Michael Bebenita2010-07-191-1/+1
| | | | communication. Channels now buffer on the sending side, and no longer require blocking when sending. Lots of other refactoring and bug fixes.
* Stub an interface to the (as-yet-nonexistent) structural comparison glue in ↵Patrick Walton2010-07-181-0/+2
| | | | trans
* More renaming of GEP offsets in abi/trans. This should settle it. Also, ↵Roy Frostig2010-07-161-5/+10
| | | | fix some overenthusiastic renaming in previous commit.
* Clean up nomenclature in the Abi constants brigade. Purge magic GEP numbers ↵Roy Frostig2010-07-162-17/+25
| | | | in trans. Adjust obj_closure_rty (and its consumers) to be more explicit about the fact that it is a box.
* Remove the __PAGEZERO segment from shared MachO libraries. This avoids aJeffrey Yasskin2010-07-161-10/+14
| | | | segfault in dlclose() and fixes the tests on OSX 10.5.8.
* In Mach-O shared libraries, avoid counting an extra load command. Closes #99.Jeffrey Yasskin2010-07-161-1/+7
|
* Define (in Abi) and use named constants for indices of elements of an obj IL ↵Roy Frostig2010-07-151-0/+4
| | | | referent type.
* Make the x86 backend push an extra word as closure/obj arg when emitting any ↵Roy Frostig2010-07-152-3/+5
| | | | code that amounts to a glue or native call.
* Begin moving closure-or-obj pointer out to front of call args by changing ↵Roy Frostig2010-07-151-5/+5
| | | | the order within the call-arg tup referent type. This breaks the world quite a bit, since it remains to fix custom call paths: glue, upcalls, natives.
* Delete some dead code in x86.Graydon Hoare2010-07-121-8/+0
|
* Hey, what's a coding sprint without an uncovered register allocation bug? No ↵Graydon Hoare2010-07-021-1/+2
| | | | fun, I say.
* Rename 'exterior' to 'box' and 'interior' to 'local' (at least wrt. slots; ↵Graydon Hoare2010-06-302-15/+15
| | | | keep MEM_interior for describing interior-parts-of-allocations)
* Some fixme-to-issue housekeeping.Graydon Hoare2010-06-281-2/+3
|
* Fix register-use bugs in sweep pass, factor sweep pass out and call twice ↵Graydon Hoare2010-06-251-57/+73
| | | | for separate sever/free phases.
* Add sever-glue, for missing first stage of sweep.Graydon Hoare2010-06-251-2/+3
|
* Pass type parameters to free function called from sweep loop.Graydon Hoare2010-06-251-9/+16
|
* Fix bad register use in GC sweep pass.Graydon Hoare2010-06-251-5/+9
|
* Remove obsolete comment.Graydon Hoare2010-06-251-1/+0
|
* I hear actually loading the next pointer into the right register helps when ↵Graydon Hoare2010-06-251-8/+7
| | | | walking a list.
* Debug GC glue enough that marking passes.Graydon Hoare2010-06-251-16/+33
|
* Update comment to reflect slightly-improved GC situation.Graydon Hoare2010-06-251-8/+2
|
* Emit gc glue and rearrange crate glue offsets slightly to have a regular order.Graydon Hoare2010-06-252-0/+2
|
* Return at end of GC glue.Graydon Hoare2010-06-251-0/+1
|
* Unlink (doubly) gc chain on free.Graydon Hoare2010-06-251-0/+1
|