Adds AArch64 Native Code Generator

In which we add a new code generator to the Glasgow Haskell
Compiler. This codegen supports ELF and Mach-O targets, thus covering
Linux, macOS, and BSDs in principle.  It was tested only on macOS and
Linux.  The NCG follows a similar structure as the other native code
generators we already have, and should therfore be realtively easy to
follow.

It supports most of the features required for a proper native code
generator, but does not claim to be perfect or fully optimised.  There
are still opportunities for optimisations.

Metric Decrease:
    ManyAlternatives
    ManyConstructors
    MultiLayerModules
    PmSeriesG
    PmSeriesS
    PmSeriesT
    PmSeriesV
    T10421
    T10421a
    T10858
    T11195
    T11276
    T11303b
    T11374
    T11822
    T12227
    T12545
    T12707
    T13035
    T13253
    T13253-spj
    T13379
    T13701
    T13719
    T14683
    T14697
    T15164
    T15630
    T16577
    T17096
    T17516
    T17836
    T17836b
    T17977
    T17977b
    T18140
    T18282
    T18304
    T18478
    T18698a
    T18698b
    T18923
    T1969
    T3064
    T5030
    T5321FD
    T5321Fun
    T5631
    T5642
    T5837
    T783
    T9198
    T9233
    T9630
    T9872d
    T9961
    WWRec
Metric Increase:
    T4801

1 files changed, 34 insertions, 5 deletions

diff --git a/compiler/GHC/CmmToAsm/PIC.hs b/compiler/GHC/CmmToAsm/PIC.hs
index 7fe90c3ec6..81ce9d34a9 100644
--- a/compiler/GHC/CmmToAsm/PIC.hs
+++ b/compiler/GHC/CmmToAsm/PIC.hs
@@ -122,6 +122,15 @@ cmmMakeDynamicReference config referenceKind lbl
               addImport stub
               return $ CmmLit $ CmmLabel stub
 
+        -- GOT relative loads work differently on AArch64.  We don't do two
+        -- step loads. The got symbol is loaded directly, and not through an
+        -- additional load. Thus we do not need the CmmLoad decoration we have
+        -- on other platforms.
+        AccessViaSymbolPtr | ArchAArch64 <- platformArch platform -> do
+              let symbolPtr = mkDynamicLinkerLabel SymbolPtr lbl
+              addImport symbolPtr
+              return $ cmmMakePicReference config symbolPtr
+
         AccessViaSymbolPtr -> do
               let symbolPtr = mkDynamicLinkerLabel SymbolPtr lbl
               addImport symbolPtr
@@ -135,7 +144,6 @@ cmmMakeDynamicReference config referenceKind lbl
                 -- so just jump there if it's a call or a jump
               _ -> return $ CmmLit $ CmmLabel lbl
 
-
 -- -----------------------------------------------------------------------------
 -- Create a position independent reference to a label.
 -- (but do not bother with dynamic linking).
@@ -150,6 +158,11 @@ cmmMakePicReference config lbl
   | OSMinGW32 <- platformOS platform
   = CmmLit $ CmmLabel lbl
 
+  -- no pic base reg on AArch64, however indicate this symbol should go through
+  -- the global offset table (GOT).
+  | ArchAArch64 <- platformArch platform
+  = CmmLit $ CmmLabel lbl
+
   | OSAIX <- platformOS platform
   = CmmMachOp (MO_Add W32)
           [ CmmReg (CmmGlobal PicBaseReg)
@@ -241,6 +254,20 @@ howToAccessLabel config _arch OSMinGW32 _kind lbl
         | otherwise
         = AccessDirectly
 
+-- On AArch64, relocations for JUMP and CALL will be emitted with 26bits, this
+-- is enough for ~64MB of range. Anything else will need to go through a veneer,
+-- which is the job of the linker to build.  We might only want to lookup
+-- Data References through the GOT.
+howToAccessLabel config ArchAArch64 _os _kind lbl
+        | not (ncgExternalDynamicRefs config)
+        = AccessDirectly
+
+        | labelDynamic config lbl
+        = AccessViaSymbolPtr
+
+        | otherwise
+        = AccessDirectly
+
 
 -- Mach-O (Darwin, Mac OS X)
 --
@@ -275,7 +302,7 @@ howToAccessLabel config arch OSDarwin JumpReference lbl
         -- dyld code stubs don't work for tailcalls because the
         -- stack alignment is only right for regular calls.
         -- Therefore, we have to go via a symbol pointer:
-        | arch == ArchX86 || arch == ArchX86_64
+        | arch == ArchX86 || arch == ArchX86_64 || arch == ArchAArch64
         , labelDynamic config lbl
         = AccessViaSymbolPtr
 
@@ -283,15 +310,15 @@ howToAccessLabel config arch OSDarwin JumpReference lbl
 howToAccessLabel config arch OSDarwin _kind lbl
         -- Code stubs are the usual method of choice for imported code;
         -- not needed on x86_64 because Apple's new linker, ld64, generates
-        -- them automatically.
+        -- them automatically, neither on Aarch64 (arm64).
         | arch /= ArchX86_64
+        , arch /= ArchAArch64
         , labelDynamic config lbl
         = AccessViaStub
 
         | otherwise
         = AccessDirectly
 
-
 ----------------------------------------------------------------------------
 -- AIX
 
@@ -616,7 +643,9 @@ pprImportedSymbol config importedLbl = case (arch,os) of
         | otherwise
         -> empty
 
-   (_, OSDarwin) -> empty
+   (ArchAArch64, OSDarwin)
+        -> empty
+
 
 
    -- XCOFF / AIX