freenode/#clasp - IRC Chatlog
Search
2:00:25
drmeister
I always ask myself "WDSD" (What Does Sbcl Do?) It may not be the best idea - but if all I had was that level of performance - I'd be quite happy.
2:02:49
Bike
i mean, basically the loss is the first call is slow (which is also like sbcl) and it adds a hash table lookup
2:07:02
drmeister
When ironclad compilation goes nuts and takes a long time - this message shows up:
2:47:07
drmeister
So - I delayed building the everything in the RUN-ALL until the end of COMPILE-FILE. Everything is pushed into *run-all-objects* and then reversed and used to generate code in RUN-ALL.
4:57:48
drmeister
::notify Bike Could you go into the cl-jupyter project on github and review Kevin's pull request and approve it? I can't do it because I initiated the pull request. I made you an Admin.
10:59:19
Colleen
Bike: drmeister said 6 hours, 1 minute ago: Could you go into the cl-jupyter project on github and review Kevin's pull request and approve it? I can't do it because I initiated the pull request. I made you an Admin.
11:06:26
Bike
i figured, but i'm not sure what fits the memory model better, because i don't know what the model is
11:06:47
Bike
the two obvious ways to do it would be to keep memref2 as it is and add an ldb instruction, or to have a memref2 that specifies a width
11:07:17
Bike
the former keeps all the memory involves as word based, but then it would be difficult to compile into a machine access if there are halfword loads and stuff
11:12:58
beach
Bike: MIR was invented before Cleavir was, so the memory model is that of SICL. MIR can not be used in its current form for implementation-independent stuff.