Add support for LLVM address spaces#280
Add support for LLVM address spaces#280peterohanley wants to merge 6 commits intoGaloisInc:masterfrom
Conversation
This has only been tested by pretty-printing CHERI bc files, not yet other directions.
| , decAttrs = [] | ||
| , decComdat = protoComdat fp | ||
| -- TODO what if this isn't equal to adr? | ||
| , decAddrSpace = protoAddrSpace fp |
There was a problem hiding this comment.
I think this would still be a sensible construct if they were not equal, but it hasn't been used.
There was a problem hiding this comment.
Hm, that is tricky. Do you know what LLVM does here? If nothing else, it might be worth checking for an (opt-in) assert.
|
The formatting changes to match llvm-dis break golden tests. They're meant for helping compare output, so I'll remove them for now and just leave the semantic changes. |
| , decAttrs = [] | ||
| , decComdat = protoComdat fp | ||
| -- TODO what if this isn't equal to adr? | ||
| , decAddrSpace = protoAddrSpace fp |
There was a problem hiding this comment.
Hm, that is tricky. Do you know what LLVM does here? If nothing else, it might be worth checking for an (opt-in) assert.
| vis <- if length (recordFields r) > (6 + offset) && not (link `elem` [Internal, Private]) | ||
| then field 6 visibility | ||
| else pure DefaultVisibility | ||
| tl <- if length (recordFields r) > (7 + offset) -- && not (link `elem` [Internal, Private]) |
There was a problem hiding this comment.
Can this comment be deleted? I think thread locality is parsed independently of the linkage, based on my reading of the LLVM implementation here.
| else case ptrty of | ||
| PtrTo as _ -> return as | ||
| PtrOpaque as -> return as | ||
| _ -> fail $ "Invalid type for value: " ++ show ptrty |
There was a problem hiding this comment.
Can this case expression be simplified by defining it in terms of ptrAddrSpace from GaloisInc/llvm-pretty#148?
| if len == 0 | ||
| then do | ||
| n <- show <$> nextSymbolId | ||
| return (Symbol n, 2) |
There was a problem hiding this comment.
Can you leave some comments explaining what is going on in the len == 0 case? From this commit, I gather that it related to unnamed symbols in some way, but the details aren't clear to me.
| then | ||
| field 1 numeric | ||
| else | ||
| return 0 |
There was a problem hiding this comment.
Another candidate for defaultAddrSpace, alhtough the AddrSpace constructor will need to move to the then.
| setDataLayout :: DataLayout -> Parse () | ||
| setDataLayout v = Parse $ do | ||
| ps <- get | ||
| put $! ps { psDataLayout = Just v } |
There was a problem hiding this comment.
| put $! ps { psDataLayout = Just v } | |
| setDataLayout v = Parse $ modify $ \ps -> ps { psDataLayout = Just v } |
4 participants
Address spaces default to 0 and are currently mostly used by CHERI for capabilities (address space 200).
The thread_local property is properly read and reported.
Various formatting changes are implemented to more closely match llvm-dis output.