tock-mirror/backends/AnalyseAsm.hs
Neil Brown b803b1ec91 Reworked how stack sizes are calculated, and changed how we use them in the build process
Now that we support separate compilation, some of the stack sizes for PROCs depend on the stack sizes of other PROCs that were compiled in different files.  We can't just assume the default 512 bytes for these "foreign" PROCs, because that often won't be enough.  So instead, we must make the stack sizes for the current PROCs depend on (i.e. use in the calculation) the stack sizes of the foreign PROCs.

This dependence adds some issues though.  We cannot declare in one C file a const int that depends on the value of an extern const int from another C file (not valid C, it seems).  So instead, we move all the stack size declarations to header files, and use #includes and the preprocessor to make sure that the stack sizes are statically determined.

This in turn simplifies the build process in some ways.  These headers only need to be compiled by the .occ file that has the main process, by including them all into a C file and compiling that as before.  It means that each .occ file only has one .o file resulting (plus two C headers*, and a .inc file) so linking is a bit less confusing.

* I am keeping the two C headers for now, rather than appending the sizes one to the normal header, because I'm not entirely sure whether having one header that the C file depends on may trigger a recompilation that we don't want in some build systems.  I can always merge them later if that's not a valid worry.
2009-04-03 12:24:53 +00:00

296 lines
11 KiB
Haskell

{-
Tock: a compiler for parallel languages
Copyright (C) 2007, 2008 University of Kent
This program is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by the
Free Software Foundation, either version 2 of the License, or (at your
option) any later version.
This program is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program. If not, see <http://www.gnu.org/licenses/>.
-}
-- | Analyse the assembly output from the C compiler for stack size
-- information.
-- FIXME: This only works for x86 at the moment.
-- FIXME: This should have a "just use a huge fixed number" mode for debugging.
module AnalyseAsm (
AsmItem(..),
parseAsmLine, analyseAsm
) where
import Control.Monad.State
import Data.Char
import Data.Generics
import Data.List
import qualified Data.Map as Map
import Data.Maybe
import qualified Data.Set as Set
import Numeric (readDec)
import Text.Printf
import CompState
import Errors
import Pass
import PrettyShow
import Utils
-- | Interesting things that we might find in the assembly source.
data AsmItem =
AsmLabel String
| AsmStackInc Int
| AsmCall String
deriving (Show, Eq, Data, Typeable)
-- | Examine a line of the assembly source to see whether it's something we're
-- interested in.
parseAsmLine :: String -> Maybe AsmItem
parseAsmLine s
= case words s of
[] -> Nothing
-- The x86 stack goes downwards, so subl makes the stack deeper.
["subl", '$':arg, "%esp"] -> parseInc arg
-- ... but GCC will sometimes generate "addl \$-n" rather than "subl
-- \$n".
["addl", '$':'-':arg, "%esp"] -> parseInc arg
-- A plain push also makes the stack deeper.
("pushl":_) -> Just $ AsmStackInc 4
["call", arg] -> parseCall arg
-- GCC does tail-call optimisation, so we need to look for jmp as well
-- as call.
["jmp", arg] -> parseCall arg
[label] -> parseLabel label
_ -> Nothing
where
-- | Parse a label: a line ending in @:@, and not starting with @.@ or a
-- digit.
parseLabel :: String -> Maybe AsmItem
parseLabel s@(c:cs)
| isDigit c || '.' `elem` s = Nothing
| last cs == ':' = Just $ AsmLabel (liat s)
| otherwise = Nothing
where
liat :: String -> String
liat = reverse . tail . reverse
-- | Parse a stack increase: just a number.
parseInc :: String -> Maybe AsmItem
parseInc s
= case readDec s of
[(v, ",")] -> Just $ AsmStackInc v
_ -> Nothing
-- | Parse a called label, which mustn't start with @.@ or @*@.
parseCall :: String -> Maybe AsmItem
parseCall ('.':_) = Nothing
parseCall ('*':_) = Nothing
parseCall s = Just $ AsmCall s
-- | Turn assembly source into a list of interesting things.
parseAsm :: String -> [AsmItem]
parseAsm asm
= catMaybes [parseAsmLine l | l <- lines asm]
data StackInfo
= Fixed Int
| Remote String
| Max [StackInfo]
| Plus StackInfo StackInfo
deriving (Show, Data, Typeable)
findAllDependencies :: StackInfo -> Set.Set String
findAllDependencies (Remote s) = Set.singleton s
findAllDependencies (Max as) = foldl Set.union Set.empty $ map findAllDependencies as
findAllDependencies (Plus a b) = findAllDependencies a `Set.union` findAllDependencies b
findAllDependencies _ = Set.empty
-- | Information about defined functions.
data FunctionInfo = FunctionInfo {
fiStack :: Int
, fiTotalStack :: Maybe StackInfo
, fiCalls :: Set.Set String
}
deriving (Show, Data, Typeable)
emptyFI :: FunctionInfo
emptyFI = FunctionInfo {
fiStack = 0
, fiTotalStack = Nothing
, fiCalls = Set.empty
}
-- | Monad for `AnalyseAsm` operations.
type AAM = StateT (Map.Map String FunctionInfo) PassM
instance CSMR AAM where
getCompState = lift getCompState
-- | Collect information about each function that's been defined.
collectInfo :: [AsmItem] -> AAM ()
collectInfo ais = collectInfo' ais ""
where
collectInfo' :: [AsmItem] -> String -> AAM ()
collectInfo' [] _ = return ()
collectInfo' (ai:ais) func
= do fmap <- get
let fi = Map.findWithDefault emptyFI func fmap
let (func', fi')
= case ai of
AsmLabel newFunc -> (newFunc, fi)
AsmStackInc v ->
-- This overestimates: it adds together all the stack
-- allocations it finds, rather than trying to figure
-- out whether any of them are optional or get undone
-- (e.g. push; pop; push will result in allocating
-- two slots).
(func, fi {
fiStack = v + fiStack fi
})
AsmCall callFunc ->
(func, fi {
fiCalls = Set.insert callFunc $ fiCalls fi
})
modify $ Map.insert func fi'
collectInfo' ais func'
-- | Additional stack size to give to all functions.
-- This is necessary because CCSP does odd things with the provided stack
-- size; it doesn't calculate the space that it needs for the arguments.
baseStackSize :: Int
baseStackSize = 32
-- | Add the stack sizes for called functions to their callers.
addCalls :: [String] -> Int -> AAM ()
addCalls knownProcs unknownSize
= do fmap <- get
sequence_ $ map (computeStack True) (Map.keys fmap)
where
computeStack :: Bool -> String -> AAM StackInfo
computeStack processUser func
= do fmap <- get
let fi = Map.findWithDefault emptyFI func fmap
let tstack = fiTotalStack fi
tstack' <- if Map.member func fmap && processUser
then (case tstack of
Nothing -> userFunc fi
Just x -> return x)
else systemFunc func
when processUser $ modify $ Map.insert func (fi { fiTotalStack = Just tstack' })
return tstack'
systemFunc :: String -> AAM StackInfo
systemFunc func
= do cs <- getCompState
fmap <- get
if func `elem` (map fst (csExternals cs) ++ knownProcs)
then do return $ Remote $ func
else do lift $ warnPlainP WarnInternal $ "Unknown function " ++ func
++ "; allocating " ++ show unknownSize ++ " bytes stack"
return $ Fixed unknownSize
userFunc :: FunctionInfo -> AAM StackInfo
userFunc fi
= do let localStack = fiStack fi + baseStackSize
calledStacks <- mapM (computeStack False) $ Set.toList $ fiCalls fi
return $ Fixed localStack `Plus` Max (Fixed 0 : calledStacks)
-- I don't think we can use sortBy here because we only have a partial ordering,
-- not a total ordering (transitivity, for one, isn't automatic).
--
-- So our plan is as follows. We calculate all the dependencies for each item.
-- We put all the items with no dependents first, and then we recurse, removing
-- all the no-dependent items from the dependencies of the others.
dependenceSort :: Set.Set String -> [(String, FunctionInfo)] -> [(String, FunctionInfo)]
dependenceSort ofInterest origItems = map fst $ dependenceSort' itemsWithDependents
where
itemsWithDependents = [(item, ofInterest `Set.intersection`
(maybe Set.empty findAllDependencies $ fiTotalStack $ snd item)) | item <- origItems]
dependenceSort' :: [((String, FunctionInfo), Set.Set String)]
-> [((String, FunctionInfo), Set.Set String)]
dependenceSort' [] = []
dependenceSort' items
| null firstItems -- Infinite loop if we don't stop it:
= error $ "Cyclic dependencies in stack sizes: "
++ show [n ++ " depends on " ++ show deps | ((n, _), deps) <- rest]
| otherwise
= firstItems ++ dependenceSort' [(item, deps `Set.difference` ignore)
| (item, deps) <- rest]
where
(firstItems, rest) = partition (Set.null . snd) items
ignore = Set.fromList $ map (fst . fst) firstItems
-- | Analyse assembler and return C source defining sizes.
analyseAsm :: Maybe [String] -> String -> PassM String
analyseAsm mprocs asm
= do let stream = parseAsm asm
veryDebug $ pshow stream
cs <- getCompState
let unique = concat [if c `elem` (['A'..'Z'] ++ ['a'..'z'] ++ ['0'..'9'])
then [c]
else '_' : show (ord c)
| c <- csCurrentFile cs]
info <- execStateT (collectInfo stream >> addCalls (fromMaybe [] mprocs) (csUnknownStackSize cs)) Map.empty
debug $ "Analysed function information:"
-- debug $ concat [printf " %-40s %5d %5d %s\n"
-- func (fiStack fi) (fiTotalStack fi)
-- (concat $ intersperse " " $ Set.toList $ fiCalls fi)
-- | (func, fi) <- Map.toList $ filterNames info]
let lines = -- Can't remember if max is a standard function so let's make our own:
"#ifndef TOCK_MAX\n#define TOCK_MAX(x,y) ((x) > (y) ? (x) : (y))\n#endif\n" :
["#include \"" ++ f ++ ".tock_post.h\"\n"
| f <- Set.toList $ csUsedFiles cs] ++
["#define " ++ func ++ "_stack_size_CPP "
++ maybe "#error Unknown!" toC (fiTotalStack fi) ++ "\n"
++ "const int " ++ func ++ "_stack_size = " ++ func ++ "_stack_size_CPP;\n"
| (func, fi) <- dependenceSort (maybe Set.empty Set.fromList mprocs) $ Map.toList $ filterNames info]
return $ "#ifndef INCLUDED_" ++ unique ++ "\n#define INCLUDED_" ++ unique
++ "\n" ++ concat lines ++ "\n#endif\n"
where
filterNames = case mprocs of
Nothing -> id
Just m -> (`Map.intersection` (Map.fromList (zip m (repeat ()))))
findAllPlus :: StackInfo -> (Int, [StackInfo])
findAllPlus (Fixed n) = (n, [])
findAllPlus (Plus a b) = findAllPlus a `with` findAllPlus b
where
with (m, as) (n, bs) = (m + n, as ++ bs)
findAllPlus a = (0, [a])
-- Without the simplifications in this function, the nesting of TOCK_MAX (and
-- its exponentially-sized expansion) was blowing the mind of the C compiler,
-- and the memory of my machine.
toC :: StackInfo -> String
toC (Fixed n) = show n
toC (Remote s) = s ++ "_stack_size_CPP"
toC x@(Plus {}) = let (m, as) = findAllPlus x in
(if m == 0 then id else \x -> "(" ++ show m ++ "+" ++ x ++ ")") $
concat (intersperse "+" $ map toC as)
toC (Max as) = foldl makeMax (show fixed) (map toC other)
where
fixed = maximum [n | Fixed n <- as]
other = filter isNotFixed as
makeMax a b = "TOCK_MAX(" ++ a ++ "," ++ b ++ ")"
isNotFixed (Fixed {}) = False
isNotFixed _ = True