{-# LANGUAGE CPP #-}
{-# LANGUAGE MagicHash #-}

-----------------------------------------------------------------------------
--
-- Pretty-printing assembly language
--
-- (c) The University of Glasgow 1993-2005
--
-----------------------------------------------------------------------------

module GHC.CmmToAsm.Ppr (
        doubleToBytes,
        floatToBytes,
        pprASCII,
        pprString,
        pprFileEmbed,
        pprSectionHeader
)

where

import GHC.Prelude

import GHC.Utils.Asm
import GHC.Cmm.CLabel
import GHC.Cmm
import GHC.CmmToAsm.Config
import GHC.Utils.Outputable as SDoc
import qualified GHC.Utils.Ppr as Pretty
import GHC.Utils.Panic
import GHC.Platform

import qualified Data.Array.Unsafe as U ( castSTUArray )
import Data.Array.ST

import Control.Monad.ST
import Data.Word
import Data.ByteString (ByteString)
import qualified Data.ByteString as BS
import GHC.Exts
import GHC.Word

#if !MIN_VERSION_base(4,16,0)
word8ToWord# :: Word# -> Word#
word8ToWord# w = w
{-# INLINE word8ToWord# #-}
#endif

-- -----------------------------------------------------------------------------
-- Converting floating-point literals to integrals for printing

-- | Get bytes of a Float representation
floatToBytes :: Float -> [Word8]
floatToBytes :: Float -> [Word8]
floatToBytes Float
f = (forall s. ST s [Word8]) -> [Word8]
forall a. (forall s. ST s a) -> a
runST ((forall s. ST s [Word8]) -> [Word8])
-> (forall s. ST s [Word8]) -> [Word8]
forall a b. (a -> b) -> a -> b
$ do
  STUArray s Int Float
arr <- (Int, Int) -> ST s (STUArray s Int Float)
forall i. Ix i => (i, i) -> ST s (STUArray s i Float)
forall (a :: * -> * -> *) e (m :: * -> *) i.
(MArray a e m, Ix i) =>
(i, i) -> m (a i e)
newArray_ ((Int
0::Int),Int
3)
  STUArray s Int Float -> Int -> Float -> ST s ()
forall (a :: * -> * -> *) e (m :: * -> *) i.
(MArray a e m, Ix i) =>
a i e -> i -> e -> m ()
writeArray STUArray s Int Float
arr Int
0 Float
f
  let cast :: STUArray s Int Float -> ST s (STUArray s Int Word8)
      cast :: forall s. STUArray s Int Float -> ST s (STUArray s Int Word8)
cast = STUArray s Int Float -> ST s (STUArray s Int Word8)
forall s ix a b. STUArray s ix a -> ST s (STUArray s ix b)
U.castSTUArray
  STUArray s Int Word8
arr <- STUArray s Int Float -> ST s (STUArray s Int Word8)
forall s. STUArray s Int Float -> ST s (STUArray s Int Word8)
cast STUArray s Int Float
arr
  Word8
i0 <- STUArray s Int Word8 -> Int -> ST s Word8
forall (a :: * -> * -> *) e (m :: * -> *) i.
(MArray a e m, Ix i) =>
a i e -> i -> m e
readArray STUArray s Int Word8
arr Int
0
  Word8
i1 <- STUArray s Int Word8 -> Int -> ST s Word8
forall (a :: * -> * -> *) e (m :: * -> *) i.
(MArray a e m, Ix i) =>
a i e -> i -> m e
readArray STUArray s Int Word8
arr Int
1
  Word8
i2 <- STUArray s Int Word8 -> Int -> ST s Word8
forall (a :: * -> * -> *) e (m :: * -> *) i.
(MArray a e m, Ix i) =>
a i e -> i -> m e
readArray STUArray s Int Word8
arr Int
2
  Word8
i3 <- STUArray s Int Word8 -> Int -> ST s Word8
forall (a :: * -> * -> *) e (m :: * -> *) i.
(MArray a e m, Ix i) =>
a i e -> i -> m e
readArray STUArray s Int Word8
arr Int
3
  [Word8] -> ST s [Word8]
forall a. a -> ST s a
forall (m :: * -> *) a. Monad m => a -> m a
return [Word8
i0,Word8
i1,Word8
i2,Word8
i3]

-- | Get bytes of a Double representation
doubleToBytes :: Double -> [Word8]
doubleToBytes :: Double -> [Word8]
doubleToBytes Double
d = (forall s. ST s [Word8]) -> [Word8]
forall a. (forall s. ST s a) -> a
runST ((forall s. ST s [Word8]) -> [Word8])
-> (forall s. ST s [Word8]) -> [Word8]
forall a b. (a -> b) -> a -> b
$ do
  STUArray s Int Double
arr <- (Int, Int) -> ST s (STUArray s Int Double)
forall i. Ix i => (i, i) -> ST s (STUArray s i Double)
forall (a :: * -> * -> *) e (m :: * -> *) i.
(MArray a e m, Ix i) =>
(i, i) -> m (a i e)
newArray_ ((Int
0::Int),Int
7)
  STUArray s Int Double -> Int -> Double -> ST s ()
forall (a :: * -> * -> *) e (m :: * -> *) i.
(MArray a e m, Ix i) =>
a i e -> i -> e -> m ()
writeArray STUArray s Int Double
arr Int
0 Double
d
  let cast :: STUArray s Int Double -> ST s (STUArray s Int Word8)
      cast :: forall s. STUArray s Int Double -> ST s (STUArray s Int Word8)
cast = STUArray s Int Double -> ST s (STUArray s Int Word8)
forall s ix a b. STUArray s ix a -> ST s (STUArray s ix b)
U.castSTUArray
  STUArray s Int Word8
arr <- STUArray s Int Double -> ST s (STUArray s Int Word8)
forall s. STUArray s Int Double -> ST s (STUArray s Int Word8)
cast STUArray s Int Double
arr
  Word8
i0 <- STUArray s Int Word8 -> Int -> ST s Word8
forall (a :: * -> * -> *) e (m :: * -> *) i.
(MArray a e m, Ix i) =>
a i e -> i -> m e
readArray STUArray s Int Word8
arr Int
0
  Word8
i1 <- STUArray s Int Word8 -> Int -> ST s Word8
forall (a :: * -> * -> *) e (m :: * -> *) i.
(MArray a e m, Ix i) =>
a i e -> i -> m e
readArray STUArray s Int Word8
arr Int
1
  Word8
i2 <- STUArray s Int Word8 -> Int -> ST s Word8
forall (a :: * -> * -> *) e (m :: * -> *) i.
(MArray a e m, Ix i) =>
a i e -> i -> m e
readArray STUArray s Int Word8
arr Int
2
  Word8
i3 <- STUArray s Int Word8 -> Int -> ST s Word8
forall (a :: * -> * -> *) e (m :: * -> *) i.
(MArray a e m, Ix i) =>
a i e -> i -> m e
readArray STUArray s Int Word8
arr Int
3
  Word8
i4 <- STUArray s Int Word8 -> Int -> ST s Word8
forall (a :: * -> * -> *) e (m :: * -> *) i.
(MArray a e m, Ix i) =>
a i e -> i -> m e
readArray STUArray s Int Word8
arr Int
4
  Word8
i5 <- STUArray s Int Word8 -> Int -> ST s Word8
forall (a :: * -> * -> *) e (m :: * -> *) i.
(MArray a e m, Ix i) =>
a i e -> i -> m e
readArray STUArray s Int Word8
arr Int
5
  Word8
i6 <- STUArray s Int Word8 -> Int -> ST s Word8
forall (a :: * -> * -> *) e (m :: * -> *) i.
(MArray a e m, Ix i) =>
a i e -> i -> m e
readArray STUArray s Int Word8
arr Int
6
  Word8
i7 <- STUArray s Int Word8 -> Int -> ST s Word8
forall (a :: * -> * -> *) e (m :: * -> *) i.
(MArray a e m, Ix i) =>
a i e -> i -> m e
readArray STUArray s Int Word8
arr Int
7
  [Word8] -> ST s [Word8]
forall a. a -> ST s a
forall (m :: * -> *) a. Monad m => a -> m a
return [Word8
i0,Word8
i1,Word8
i2,Word8
i3,Word8
i4,Word8
i5,Word8
i6,Word8
i7]


-- ---------------------------------------------------------------------------
-- Printing ASCII strings.
--
-- Print as a string and escape non-printable characters.
-- This is similar to charToC in GHC.Utils.Misc

pprASCII :: ByteString -> SDoc
pprASCII :: ByteString -> SDoc
pprASCII ByteString
str
  -- Transform this given literal bytestring to escaped string and construct
  -- the literal SDoc directly.
  -- See #14741
  -- and Note [Pretty print ASCII when AsmCodeGen]
  --
  -- We work with a `Doc` instead of an `SDoc` because there is no need to carry
  -- an `SDocContext` that we don't use. It leads to nicer (STG) code.
  = Doc -> SDoc
docToSDoc ((Word8 -> Doc -> Doc) -> Doc -> ByteString -> Doc
forall a. (Word8 -> a -> a) -> a -> ByteString -> a
BS.foldr Word8 -> Doc -> Doc
f Doc
Pretty.empty ByteString
str)
    where
       f :: Word8 -> Pretty.Doc -> Pretty.Doc
       f :: Word8 -> Doc -> Doc
f Word8
w Doc
s = Word8 -> Doc
do1 Word8
w Doc -> Doc -> Doc
Pretty.<> Doc
s

       do1 :: Word8 -> Pretty.Doc
       do1 :: Word8 -> Doc
do1 Word8
w | Word8
0x09 Word8 -> Word8 -> Bool
forall a. Eq a => a -> a -> Bool
== Word8
w = String -> Doc
Pretty.text String
"\\t"
             | Word8
0x0A Word8 -> Word8 -> Bool
forall a. Eq a => a -> a -> Bool
== Word8
w = String -> Doc
Pretty.text String
"\\n"
             | Word8
0x22 Word8 -> Word8 -> Bool
forall a. Eq a => a -> a -> Bool
== Word8
w = String -> Doc
Pretty.text String
"\\\""
             | Word8
0x5C Word8 -> Word8 -> Bool
forall a. Eq a => a -> a -> Bool
== Word8
w = String -> Doc
Pretty.text String
"\\\\"
               -- ASCII printable characters range
             | Word8
w Word8 -> Word8 -> Bool
forall a. Ord a => a -> a -> Bool
>= Word8
0x20 Bool -> Bool -> Bool
&& Word8
w Word8 -> Word8 -> Bool
forall a. Ord a => a -> a -> Bool
<= Word8
0x7E = Char -> Doc
Pretty.char (Word8 -> Char
chr' Word8
w)
             | Bool
otherwise = Int -> String -> Doc
Pretty.sizedText Int
4 String
xs
                where
                 !xs :: String
xs = [ Char
'\\', Char
x0, Char
x1, Char
x2] -- octal
                 !x0 :: Char
x0 = Word8 -> Char
chr' (Word8
ord0 Word8 -> Word8 -> Word8
forall a. Num a => a -> a -> a
+ (Word8
w Word8 -> Int -> Word8
forall a. Bits a => a -> Int -> a
`unsafeShiftR` Int
6) Word8 -> Word8 -> Word8
forall a. Bits a => a -> a -> a
.&. Word8
0x07)
                 !x1 :: Char
x1 = Word8 -> Char
chr' (Word8
ord0 Word8 -> Word8 -> Word8
forall a. Num a => a -> a -> a
+ (Word8
w Word8 -> Int -> Word8
forall a. Bits a => a -> Int -> a
`unsafeShiftR` Int
3) Word8 -> Word8 -> Word8
forall a. Bits a => a -> a -> a
.&. Word8
0x07)
                 !x2 :: Char
x2 = Word8 -> Char
chr' (Word8
ord0 Word8 -> Word8 -> Word8
forall a. Num a => a -> a -> a
+ Word8
w Word8 -> Word8 -> Word8
forall a. Bits a => a -> a -> a
.&. Word8
0x07)
                 !ord0 :: Word8
ord0 = Word8
0x30 -- = ord '0'

       -- we know that the Chars we create are in the ASCII range
       -- so we bypass the check in "chr"
       chr' :: Word8 -> Char
       chr' :: Word8 -> Char
chr' (W8# Word8#
w#) = Char# -> Char
C# (Int# -> Char#
chr# (Word# -> Int#
word2Int# (Word8# -> Word#
word8ToWord# Word8#
w#)))


-- | Emit a ".string" directive
pprString :: ByteString -> SDoc
pprString :: ByteString -> SDoc
pprString ByteString
bs = String -> SDoc
text String
"\t.string " SDoc -> SDoc -> SDoc
<> SDoc -> SDoc
doubleQuotes (ByteString -> SDoc
pprASCII ByteString
bs)

-- | Emit a ".incbin" directive
--
-- A NULL byte is added after the binary data.
pprFileEmbed :: FilePath -> SDoc
pprFileEmbed :: String -> SDoc
pprFileEmbed String
path
   = String -> SDoc
text String
"\t.incbin "
     SDoc -> SDoc -> SDoc
<> String -> SDoc
pprFilePathString String
path -- proper escape (see #16389)
     SDoc -> SDoc -> SDoc
<> String -> SDoc
text String
"\n\t.byte 0"

{-
Note [Embedding large binary blobs]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

To embed a blob of binary data (e.g. an UTF-8 encoded string) into the generated
code object, we have several options:

   1. Generate a ".byte" directive for each byte. This is what was done in the past
      (see Note [Pretty print ASCII when AsmCodeGen]).

   2. Generate a single ".string"/".asciz" directive for the whole sequence of
      bytes. Bytes in the ASCII printable range are rendered as characters and
      other values are escaped (e.g., "\t", "\077", etc.).

   3. Create a temporary file into which we dump the binary data and generate a
      single ".incbin" directive. The assembler will include the binary file for
      us in the generated output object.

Now the code generator uses either (2) or (3), depending on the binary blob
size.  Using (3) for small blobs adds too much overhead (see benchmark results
in #16190), so we only do it when the size is above a threshold (500K at the
time of writing).

The threshold is configurable via the `-fbinary-blob-threshold` flag.

-}


{-
Note [Pretty print ASCII when AsmCodeGen]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Previously, when generating assembly code, we created SDoc with
`(ptext . sLit)` for every bytes in literal bytestring, then
combine them using `hcat`.

When handling literal bytestrings with millions of bytes,
millions of SDoc would be created and to combine, leading to
high memory usage.

Now we escape the given bytestring to string directly and construct
SDoc only once. This improvement could dramatically decrease the
memory allocation from 4.7GB to 1.3GB when embedding a 3MB literal
string in source code. See #14741 for profiling results.
-}

-- ----------------------------------------------------------------------------
-- Printing section headers.
--
-- If -split-section was specified, include the suffix label, otherwise just
-- print the section type. For Darwin, where subsections-for-symbols are
-- used instead, only print section type.
--
-- For string literals, additional flags are specified to enable merging of
-- identical strings in the linker. With -split-sections each string also gets
-- a unique section to allow strings from unused code to be GC'd.

pprSectionHeader :: NCGConfig -> Section -> SDoc
pprSectionHeader :: NCGConfig -> Section -> SDoc
pprSectionHeader NCGConfig
config (Section SectionType
t CLabel
suffix) =
 case Platform -> OS
platformOS (NCGConfig -> Platform
ncgPlatform NCGConfig
config) of
   OS
OSAIX     -> SectionType -> SDoc
pprXcoffSectionHeader SectionType
t
   OS
OSDarwin  -> SectionType -> SDoc
pprDarwinSectionHeader SectionType
t
   OS
_         -> NCGConfig -> SectionType -> CLabel -> SDoc
pprGNUSectionHeader NCGConfig
config SectionType
t CLabel
suffix

pprGNUSectionHeader :: NCGConfig -> SectionType -> CLabel -> SDoc
pprGNUSectionHeader :: NCGConfig -> SectionType -> CLabel -> SDoc
pprGNUSectionHeader NCGConfig
config SectionType
t CLabel
suffix =
  [SDoc] -> SDoc
hcat [String -> SDoc
text String
".section ", SDoc
header, SDoc
subsection, SDoc
flags]
  where
    sep :: SDoc
sep
      | OS
OSMinGW32 <- Platform -> OS
platformOS Platform
platform = Char -> SDoc
char Char
'$'
      | Bool
otherwise                        = Char -> SDoc
char Char
'.'
    platform :: Platform
platform      = NCGConfig -> Platform
ncgPlatform NCGConfig
config
    splitSections :: Bool
splitSections = NCGConfig -> Bool
ncgSplitSections NCGConfig
config
    subsection :: SDoc
subsection
      | Bool
splitSections = SDoc
sep SDoc -> SDoc -> SDoc
<> Platform -> CLabel -> SDoc
forall env a. OutputableP env a => env -> a -> SDoc
pdoc Platform
platform CLabel
suffix
      | Bool
otherwise     = SDoc
empty
    header :: SDoc
header = case SectionType
t of
      SectionType
Text -> String -> SDoc
text String
".text"
      SectionType
Data -> String -> SDoc
text String
".data"
      SectionType
ReadOnlyData  | OS
OSMinGW32 <- Platform -> OS
platformOS Platform
platform
                                -> String -> SDoc
text String
".rdata"
                    | Bool
otherwise -> String -> SDoc
text String
".rodata"
      SectionType
RelocatableReadOnlyData | OS
OSMinGW32 <- Platform -> OS
platformOS Platform
platform
                                -- Concept does not exist on Windows,
                                -- So map these to R/O data.
                                          -> String -> SDoc
text String
".rdata$rel.ro"
                              | Bool
otherwise -> String -> SDoc
text String
".data.rel.ro"
      SectionType
UninitialisedData -> String -> SDoc
text String
".bss"
      SectionType
ReadOnlyData16 | OS
OSMinGW32 <- Platform -> OS
platformOS Platform
platform
                                 -> String -> SDoc
text String
".rdata$cst16"
                     | Bool
otherwise -> String -> SDoc
text String
".rodata.cst16"
      SectionType
InitArray
        | OS
OSMinGW32 <- Platform -> OS
platformOS Platform
platform
                    -> String -> SDoc
text String
".ctors"
        | Bool
otherwise -> String -> SDoc
text String
".init_array"
      SectionType
FiniArray
        | OS
OSMinGW32 <- Platform -> OS
platformOS Platform
platform
                    -> String -> SDoc
text String
".dtors"
        | Bool
otherwise -> String -> SDoc
text String
".fini_array"
      SectionType
CString
        | OS
OSMinGW32 <- Platform -> OS
platformOS Platform
platform
                    -> String -> SDoc
text String
".rdata"
        | Bool
otherwise -> String -> SDoc
text String
".rodata.str"
      OtherSection String
_ ->
        String -> SDoc
forall a. String -> a
panic String
"PprBase.pprGNUSectionHeader: unknown section type"
    flags :: SDoc
flags = case SectionType
t of
      SectionType
CString
        | OS
OSMinGW32 <- Platform -> OS
platformOS Platform
platform
                    -> SDoc
empty
        | Bool
otherwise -> String -> SDoc
text String
",\"aMS\"," SDoc -> SDoc -> SDoc
<> Platform -> String -> SDoc
sectionType Platform
platform String
"progbits" SDoc -> SDoc -> SDoc
<> String -> SDoc
text String
",1"
      SectionType
_ -> SDoc
empty

-- XCOFF doesn't support relocating label-differences, so we place all
-- RO sections into .text[PR] sections
pprXcoffSectionHeader :: SectionType -> SDoc
pprXcoffSectionHeader :: SectionType -> SDoc
pprXcoffSectionHeader SectionType
t = case SectionType
t of
  SectionType
Text                    -> String -> SDoc
text String
".csect .text[PR]"
  SectionType
Data                    -> String -> SDoc
text String
".csect .data[RW]"
  SectionType
ReadOnlyData            -> String -> SDoc
text String
".csect .text[PR] # ReadOnlyData"
  SectionType
RelocatableReadOnlyData -> String -> SDoc
text String
".csect .text[PR] # RelocatableReadOnlyData"
  SectionType
ReadOnlyData16          -> String -> SDoc
text String
".csect .text[PR] # ReadOnlyData16"
  SectionType
CString                 -> String -> SDoc
text String
".csect .text[PR] # CString"
  SectionType
UninitialisedData       -> String -> SDoc
text String
".csect .data[BS]"
  SectionType
_                       -> String -> SDoc
forall a. String -> a
panic String
"pprXcoffSectionHeader: unknown section type"

pprDarwinSectionHeader :: SectionType -> SDoc
pprDarwinSectionHeader :: SectionType -> SDoc
pprDarwinSectionHeader SectionType
t = case SectionType
t of
  SectionType
Text                    -> String -> SDoc
text String
".text"
  SectionType
Data                    -> String -> SDoc
text String
".data"
  SectionType
ReadOnlyData            -> String -> SDoc
text String
".const"
  SectionType
RelocatableReadOnlyData -> String -> SDoc
text String
".const_data"
  SectionType
UninitialisedData       -> String -> SDoc
text String
".data"
  SectionType
ReadOnlyData16          -> String -> SDoc
text String
".const"
  SectionType
InitArray               -> String -> SDoc
text String
".section\t__DATA,__mod_init_func,mod_init_funcs"
  SectionType
FiniArray               -> String -> SDoc
forall a. String -> a
panic String
"pprDarwinSectionHeader: fini not supported"
  SectionType
CString                 -> String -> SDoc
text String
".section\t__TEXT,__cstring,cstring_literals"
  OtherSection String
_          -> String -> SDoc
forall a. String -> a
panic String
"pprDarwinSectionHeader: unknown section type"