{-# LANGUAGE CPP, MagicHash #-}

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

module GHC.CmmToAsm.Ppr (
        doubleToBytes,
        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.Data.FastString
import GHC.Utils.Outputable
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

-- ToDo: this code is currently shared between SPARC and LLVM.
--       Similar functions for (single precision) floats are
--       present in the SPARC backend only. We need to fix both
--       LLVM and SPARC.

castDoubleToWord8Array :: STUArray s Int Double -> ST s (STUArray s Int Word8)
castDoubleToWord8Array :: forall s. STUArray s Int Double -> ST s (STUArray s Int Word8)
castDoubleToWord8Array = forall s ix a b. STUArray s ix a -> ST s (STUArray s ix b)
U.castSTUArray

-- floatToBytes and doubleToBytes convert to the host's byte
-- order.  Providing that we're not cross-compiling for a
-- target with the opposite endianness, this should work ok
-- on all targets.

-- ToDo: this stuff is very similar to the shenanigans in PprAbs,
-- could they be merged?

doubleToBytes :: Double -> [Int]
doubleToBytes :: Double -> [Int]
doubleToBytes Double
d
   = forall a. (forall s. ST s a) -> a
runST (do
        STUArray s Int Double
arr <- forall (a :: * -> * -> *) e (m :: * -> *) i.
(MArray a e m, Ix i) =>
(i, i) -> m (a i e)
newArray_ ((Int
0::Int),Int
7)
        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
        STUArray s Int Word8
arr <- forall s. STUArray s Int Double -> ST s (STUArray s Int Word8)
castDoubleToWord8Array STUArray s Int Double
arr
        Word8
i0 <- 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 <- 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 <- 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 <- 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 <- 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 <- 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 <- 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 <- 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
        forall (m :: * -> *) a. Monad m => a -> m a
return (forall a b. (a -> b) -> [a] -> [b]
map forall a b. (Integral a, Num b) => a -> b
fromIntegral [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]
  = String -> SDoc
text forall a b. (a -> b) -> a -> b
$ forall a. (Word8 -> a -> a) -> a -> ByteString -> a
BS.foldr (\Word8
w String
s -> Word8 -> String
do1 Word8
w forall a. [a] -> [a] -> [a]
++ String
s) String
"" ByteString
str
    where
       do1 :: Word8 -> String
       do1 :: Word8 -> String
do1 Word8
w | Word8
0x09 forall a. Eq a => a -> a -> Bool
== Word8
w = String
"\\t"
             | Word8
0x0A forall a. Eq a => a -> a -> Bool
== Word8
w = String
"\\n"
             | Word8
0x22 forall a. Eq a => a -> a -> Bool
== Word8
w = String
"\\\""
             | Word8
0x5C forall a. Eq a => a -> a -> Bool
== Word8
w = String
"\\\\"
               -- ASCII printable characters range
             | Word8
w forall a. Ord a => a -> a -> Bool
>= Word8
0x20 Bool -> Bool -> Bool
&& Word8
w forall a. Ord a => a -> a -> Bool
<= Word8
0x7E = [Word8 -> Char
chr' Word8
w]
             | Bool
otherwise = Char
'\\' forall a. a -> [a] -> [a]
: Word8 -> String
octal Word8
w

       -- 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#)))

       octal :: Word8 -> String
       octal :: Word8 -> String
octal Word8
w = [ Word8 -> Char
chr' (Word8
ord0 forall a. Num a => a -> a -> a
+ (Word8
w forall a. Bits a => a -> Int -> a
`unsafeShiftR` Int
6) forall a. Bits a => a -> a -> a
.&. Word8
0x07)
                 , Word8 -> Char
chr' (Word8
ord0 forall a. Num a => a -> a -> a
+ (Word8
w forall a. Bits a => a -> Int -> a
`unsafeShiftR` Int
3) forall a. Bits a => a -> a -> a
.&. Word8
0x07)
                 , Word8 -> Char
chr' (Word8
ord0 forall a. Num a => a -> a -> a
+ Word8
w forall a. Bits a => a -> a -> a
.&. Word8
0x07)
                 ]
       ord0 :: Word8
ord0 = Word8
0x30 -- = ord '0'

-- | 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
OSMinGW32 -> NCGConfig -> SDoc -> SectionType -> CLabel -> SDoc
pprGNUSectionHeader NCGConfig
config (Char -> SDoc
char Char
'$') SectionType
t CLabel
suffix
   OS
_         -> NCGConfig -> SDoc -> SectionType -> CLabel -> SDoc
pprGNUSectionHeader NCGConfig
config (Char -> SDoc
char Char
'.') SectionType
t CLabel
suffix

pprGNUSectionHeader :: NCGConfig -> SDoc -> SectionType -> CLabel -> SDoc
pprGNUSectionHeader :: NCGConfig -> SDoc -> SectionType -> CLabel -> SDoc
pprGNUSectionHeader NCGConfig
config SDoc
sep SectionType
t CLabel
suffix =
  String -> SDoc
text String
".section " SDoc -> SDoc -> SDoc
<> PtrString -> SDoc
ptext PtrString
header SDoc -> SDoc -> SDoc
<> SDoc
subsection SDoc -> SDoc -> SDoc
<> SDoc
flags
  where
    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
<> forall env a. OutputableP env a => env -> a -> SDoc
pdoc Platform
platform CLabel
suffix
      | Bool
otherwise     = SDoc
empty
    header :: PtrString
header = case SectionType
t of
      SectionType
Text -> String -> PtrString
sLit String
".text"
      SectionType
Data -> String -> PtrString
sLit String
".data"
      SectionType
ReadOnlyData  | OS
OSMinGW32 <- Platform -> OS
platformOS Platform
platform
                                -> String -> PtrString
sLit String
".rdata"
                    | Bool
otherwise -> String -> PtrString
sLit 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 -> PtrString
sLit String
".rdata$rel.ro"
                              | Bool
otherwise -> String -> PtrString
sLit String
".data.rel.ro"
      SectionType
UninitialisedData -> String -> PtrString
sLit String
".bss"
      SectionType
ReadOnlyData16 | OS
OSMinGW32 <- Platform -> OS
platformOS Platform
platform
                                 -> String -> PtrString
sLit String
".rdata$cst16"
                     | Bool
otherwise -> String -> PtrString
sLit String
".rodata.cst16"
      SectionType
CString
        | OS
OSMinGW32 <- Platform -> OS
platformOS Platform
platform
                    -> String -> PtrString
sLit String
".rdata"
        | Bool
otherwise -> String -> PtrString
sLit String
".rodata.str"
      OtherSection String
_ ->
        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 = String -> SDoc
text forall a b. (a -> b) -> a -> b
$ case SectionType
t of
     SectionType
Text                    -> String
".csect .text[PR]"
     SectionType
Data                    -> String
".csect .data[RW]"
     SectionType
ReadOnlyData            -> String
".csect .text[PR] # ReadOnlyData"
     SectionType
RelocatableReadOnlyData -> String
".csect .text[PR] # RelocatableReadOnlyData"
     SectionType
ReadOnlyData16          -> String
".csect .text[PR] # ReadOnlyData16"
     SectionType
CString                 -> String
".csect .text[PR] # CString"
     SectionType
UninitialisedData       -> String
".csect .data[BS]"
     OtherSection String
_          ->
       forall a. String -> a
panic String
"PprBase.pprXcoffSectionHeader: unknown section type"

pprDarwinSectionHeader :: SectionType -> SDoc
pprDarwinSectionHeader :: SectionType -> SDoc
pprDarwinSectionHeader SectionType
t =
  PtrString -> SDoc
ptext forall a b. (a -> b) -> a -> b
$ case SectionType
t of
     SectionType
Text -> String -> PtrString
sLit String
".text"
     SectionType
Data -> String -> PtrString
sLit String
".data"
     SectionType
ReadOnlyData -> String -> PtrString
sLit String
".const"
     SectionType
RelocatableReadOnlyData -> String -> PtrString
sLit String
".const_data"
     SectionType
UninitialisedData -> String -> PtrString
sLit String
".data"
     SectionType
ReadOnlyData16 -> String -> PtrString
sLit String
".const"
     SectionType
CString -> String -> PtrString
sLit String
".section\t__TEXT,__cstring,cstring_literals"
     OtherSection String
_ ->
       forall a. String -> a
panic String
"PprBase.pprDarwinSectionHeader: unknown section type"