freenode/#clasp - IRC Chatlog
Search
17:46:20
beach
Bike: Good question. I think the recover restart should be there. It is more general. But maybe we should add a consider-special restart too.
17:48:46
beach
But consider-special is more accurately named. The recover restart is for use by Second Climacs where it doesn't matter how the recovery is made.
17:49:37
Bike
So do you want me to put in another restart called consider-special that duplicates the action?
17:57:02
Bike
i guess we sort of could with restart-bind, which would be my first use of restart-bind
18:01:47
karlosz
beach: i wrote a new basic block representation represented as a class with additional slots holding successor and predecessor block information.i also added functions like map-basic-block-instructions and functions to compute an instructions containing/create a has table map of that information, since ive been using those in my compiler for a while.
18:02:35
karlosz
maybe about 3x (when i stored predecessor and successor information in the block itself)
21:07:42
drmeister
karlosz: How do basic blocks interact with instructions? Does inserting and removing instructions have to worry about consistency with basic block info?
21:08:31
karlosz
yes, jjust like the current basic block stuff in cleavir, thre are no facilities for automatically keeping the basic block stuff consistent
21:10:06
karlosz
since we don't want to actually have instructions hold a slot to the containing block i don't know of any better way
21:10:10
drmeister
In llvm everything is done with basic blocks in mind - when you want to add an instruction you have to choose which basic block you want to insert into.
21:10:25
karlosz
beach: i'll package it up and send in a PR to replace the existing basic block implementation
21:12:43
karlosz
i can calculate all the successor and predecessor block information in that single pass as well
21:15:23
karlosz
it could compile the actual translation stuff fine though, so i don't know whether im miscompiling or falling into some circularity trap
21:17:34
drmeister
Crap! I'm hitting that "error: can't create dynamic relocation R_X86_64_PC32 against symbol:" error on Linux that attila pointed out
21:30:28
drmeister
::notify attila_lendvai I ran into the problem you saw on linux "error: can't create dynamic relocation R_X86_64_PC32 against symbol:" - I'm trying to figure out a work around. Can stuff just keep working?
21:57:09
drmeister
::notify attila_lendvai It looks like a workaround is to not use the lld linker when building in object mode on linux (phew). I added support for lld for bitcode/lto mode - so it's no big deal to switch back to ld for object mode.
1:03:35
Bike
git fetch origin clasp-changes; git checkout origin/clasp-changes; git pull upstream master --rebase
1:05:17
drmeister
But I can't do that on AWS machines without a hassle. I'll just wait until you push the changes.