freenode/#clasp - IRC Chatlog

0:16:28 drmeister Hello everyone - I'm back

0:17:46 drmeister karlosz: Yes - you generate code that is not SSA and then run it through optimization passes. One in particular is called mem2reg - it converts alloca's to registers with PHI nodes when possible.

0:18:29 karlosz yeah

0:18:37 karlosz so in that sense llvm and hir are very compatible

0:18:54 drmeister Yes.

0:18:55 karlosz i need to wrangle HIR somehow into stack machine bytecode

0:19:01 karlosz that seems a little stupid

0:19:22 karlosz perhaps theres a way around that...

0:20:05 drmeister Isn't there some transformation for converting stack machine code into register machine code?

0:20:38 drmeister I seem to recall that from my old CS classes.

0:21:14 karlosz yeah, that's relatively straightforward

0:21:30 karlosz but going the other way around is a bit harder

0:22:01 karlosz or more like wasteful

0:43:17 karlosz_ ** NICK karlosz

0:45:30 drmeister Bike: I need some updated sicl?

0:45:40 Bike yeah, the catch-unwind branch.

0:46:20 drmeister Is that in our fork?

0:46:27 drmeister https://github.com/drmeister/SICL.git

0:46:29 Bike it's in beach's repo

0:47:12 drmeister Got it.

1:33:05 karlosz cleavir is translating 1 as a load time value

1:33:22 Bike yeah.

1:33:24 karlosz but i don't see why; it doesn't seem like the environment should affect that

1:33:28 karlosz oh

1:33:33 Bike it translates all constants as load time values

1:33:40 karlosz isnt that what constant/immediate nodes are for?

1:33:40 Bike unless you define a method on the uh

1:33:44 Bike something or another

1:33:50 Bike you have to tell it what can be an immediate

1:33:52 drmeister There is a generic function for converting immediates.

1:34:10 drmeister cleavir-generate-ast:convert-constant-to-immediate

1:34:28 Bike or cst-to-ast:convert-constant-to-immediate

1:34:35 karlosz isnt that supposed to be the distinction between constant and immediate?

1:34:51 karlosz cleavir-ir:constant-input cleavir-ir:immediate-input

1:34:54 Bike immediates are things that cn be put directly into the instruction stream

1:35:01 karlosz yeah

1:35:03 Bike what those are depends on the backend, e.g. the word length

1:35:05 karlosz like machine integers

1:35:12 Bike so you have to define methods to tell cleavir what can be immediate

1:35:12 karlosz but then what is the constant input

1:35:22 drmeister Right - listen to Bike - I gave you the old-timey function

1:35:48 drmeister Does clisp use tagged pointers?

1:35:59 karlosz yeah, it does tagging

1:36:11 Bike i don't remember what uses constant input

1:36:18 karlosz i know ill have to tell it the object layout at some point

1:36:25 Bike you have to convert constants like load time values anyway

1:36:31 karlosz yeah

1:36:52 drmeister http://lars.nocrew.org/lisptags.html

1:37:29 drmeister That doesn't say much for clisp

1:37:49 karlosz oh sweet

1:37:55 karlosz i dont have to worry about tags

1:38:02 karlosz thits abstracted away by the bytecode vm

1:38:35 drmeister Well - if you ever do this is what clasp does: https://github.com/drmeister/clasp/blob/cst-catch/src/lisp/kernel/cleavir/setup.lisp#L53

1:39:01 Bike that's not totally helpful since it just calls another function...

1:39:31 karlosz thanks -- but it seems like the bytecodes are high level enought hat i can just deal with constants as a uniform blob

1:39:36 Bike also the immediate-literal thing is kind of weird. our load time value machinery is still sorta weird

1:44:26 karlosz also, what is that extra gensym output to enter-instruction doing there? the docs dont mention anything about it

1:44:55 Bike it's the closure.

1:45:01 Bike do they not? i should fix that

1:45:36 karlosz no, it only mentions the outputs associated with parameters

1:46:01 Bike it represents the closed over variables. the only thing it can be used for is an input to a fetch-instruction. cleavir-hir-transformations:process-captured-variables takes care of inserting the instructions

1:46:12 Bike you'll need to implement translations for fetch, write-cell, read-cell, and create-cell

1:46:26 Bike where a cell is just a thing you can read and write a value to, like an ML ref.

1:46:45 karlosz yeah, im familiar with how closures are usually implemented

1:46:55 Bike i have no idea how clisp does it, natch

1:47:19 karlosz they just have a vector bundled up with the function object

1:47:26 Bike ok, that's perfect then.

1:47:29 Bike fetch is an aref, etc

1:47:32 karlosz yep

1:48:02 karlosz the stack machine part complicates things

1:48:06 Bike i was thinking of changing the handling so that the enter would just output cells directly, but i've been doin other stuff

1:48:21 Bike a transform to use values instead of cells when the cells are read-only is on my list to

1:49:19 karlosz yeah, theres a name for that

1:49:26 karlosz when you represent everything initially boxed

1:49:43 karlosz but do some analysis to determine which can be rread-only values

5:58:11 d4ryus1 ** NICK d4ryus