Remove target dependent CPP for Word64/Int64 (#11470)

Primops types were dependent on the target word-size at *compiler*
compilation time. It's an issue for multi-target as GHC may not have the
correct primops types for the target.

This patch fixes some primops types: if they take or return fixed 64-bit
values they now always use `Int64#/Word64#`, even on 64-bit
architectures (where they used `Int#/Word#` before). Users of these
primops may now need to convert from Int64#/Word64# to Int#/Word# (a
no-op at runtime).

This is a stripped down version of !3658 which goes the all way of
changing the underlying primitive types of Word64/Int64. This is left
for future work.

T12545 allocations increase ~4% on some CI platforms and decrease ~3% on
AArch64.

Metric Increase:
    T12545

Metric Decrease:
    T12545

6 files changed, 53 insertions, 0 deletions

diff --git a/testsuite/tests/codeGen/should_run/T9340.hs b/testsuite/tests/codeGen/should_run/T9340.hs
index 45f791ba73..22f5824115 100644
--- a/testsuite/tests/codeGen/should_run/T9340.hs
+++ b/testsuite/tests/codeGen/should_run/T9340.hs
@@ -1,4 +1,5 @@
 {-# LANGUAGE MagicHash #-}
+{-# LANGUAGE CPP #-}
 
 import Control.Monad
 import Data.Bits
@@ -6,6 +7,8 @@ import GHC.Exts
 import GHC.Word
 import Numeric (showHex)
 
+#include "MachDeps.h"
+
 -- Reference Implementation
 
 -- count trailing zeros
@@ -58,8 +61,13 @@ ctzIUT32 (W# x#) = W# (ctz32# x#)
 clzIUT32 (W# x#) = W# (clz32# x#)
 
 ctzIUT64, clzIUT64 :: Word64 -> Word
+#if WORD_SIZE_IN_BITS < 64
 ctzIUT64 (W64# x#) = W# (ctz64# x#)
 clzIUT64 (W64# x#) = W# (clz64# x#)
+#else
+ctzIUT64 (W64# x#) = W# (ctz64# (wordToWord64# x#))
+clzIUT64 (W64# x#) = W# (clz64# (wordToWord64# x#))
+#endif
 
 main :: IO ()
 main = do
diff --git a/testsuite/tests/codeGen/should_run/cgrun072.hs b/testsuite/tests/codeGen/should_run/cgrun072.hs
index b97ce56d01..729564b631 100644
--- a/testsuite/tests/codeGen/should_run/cgrun072.hs
+++ b/testsuite/tests/codeGen/should_run/cgrun072.hs
@@ -37,7 +37,11 @@ bswap32 :: Word32 -> Word32
 bswap32 (W32# w#) = W32# (wordToWord32# (byteSwap32# (word32ToWord# w#)))
 
 bswap64 :: Word64 -> Word64
+#if WORD_SIZE_IN_BITS < 64
 bswap64 (W64# w#) = W64# (byteSwap64# w#)
+#else
+bswap64 (W64# w#) = W64# (word64ToWord# (byteSwap64# (wordToWord64# w#)))
+#endif
 
 slowBswap64 :: Word64 -> Word64
 slowBswap64 w =
diff --git a/testsuite/tests/codeGen/should_run/cgrun075.hs b/testsuite/tests/codeGen/should_run/cgrun075.hs
index 5babde1254..1cac98b2dd 100644
--- a/testsuite/tests/codeGen/should_run/cgrun075.hs
+++ b/testsuite/tests/codeGen/should_run/cgrun075.hs
@@ -36,7 +36,11 @@ instance Pdep Word32 where
   pdep (W32# src#) (W32# mask#) = W32# (wordToWord32# (pdep32# (word32ToWord# src#) (word32ToWord# mask#)))
 
 instance Pdep Word64 where
+#if WORD_SIZE_IN_BITS < 64
   pdep (W64# src#) (W64# mask#) = W64# (pdep64# src# mask#)
+#else
+  pdep (W64# src#) (W64# mask#) = W64# (word64ToWord# (pdep64# (wordToWord64# src#) (wordToWord64# mask#)))
+#endif
 
 class SlowPdep a where
   slowPdep :: a -> a -> a
diff --git a/testsuite/tests/codeGen/should_run/cgrun076.hs b/testsuite/tests/codeGen/should_run/cgrun076.hs
index 4779b5beb8..ce26e375d0 100644
--- a/testsuite/tests/codeGen/should_run/cgrun076.hs
+++ b/testsuite/tests/codeGen/should_run/cgrun076.hs
@@ -36,7 +36,11 @@ instance Pext Word32 where
   pext (W32# src#) (W32# mask#) = W32# (wordToWord32# (pext32# (word32ToWord# src#) (word32ToWord# mask#)))
 
 instance Pext Word64 where
+#if WORD_SIZE_IN_BITS < 64
   pext (W64# src#) (W64# mask#) = W64# (pext64# src# mask#)
+#else
+  pext (W64# src#) (W64# mask#) = W64# (word64ToWord# (pext64# (wordToWord64# src#) (wordToWord64# mask#)))
+#endif
 
 class SlowPext a where
   slowPext :: a -> a -> a
diff --git a/testsuite/tests/concurrent/should_run/AtomicPrimops.hs b/testsuite/tests/concurrent/should_run/AtomicPrimops.hs
index b8adb3c621..05ac80dcad 100644
--- a/testsuite/tests/concurrent/should_run/AtomicPrimops.hs
+++ b/testsuite/tests/concurrent/should_run/AtomicPrimops.hs
@@ -1,3 +1,4 @@
+{-# LANGUAGE CPP #-}
 {-# LANGUAGE MagicHash #-}
 {-# LANGUAGE UnboxedTuples #-}
 
@@ -14,6 +15,8 @@ import GHC.Int
 import GHC.IO
 import GHC.Word
 
+#include "MachDeps.h"
+
 -- | Iterations per worker.
 iters :: Word
 iters = 1000000
@@ -550,7 +553,11 @@ readInt32Array (MBA mba#) (I# ix#) = IO $ \ s# ->
 readInt64Array :: MByteArray -> Int -> IO Int64
 readInt64Array (MBA mba#) (I# ix#) = IO $ \ s# ->
     case readInt64Array# mba# ix# s# of
+#if WORD_SIZE_IN_BITS < 64
         (# s2#, n# #) -> (# s2#, I64# n# #)
+#else
+        (# s2#, n# #) -> (# s2#, I64# (int64ToInt# n#) #)
+#endif
 
 atomicWriteIntArray :: MByteArray -> Int -> Int -> IO ()
 atomicWriteIntArray (MBA mba#) (I# ix#) (I# n#) = IO $ \ s# ->
@@ -584,8 +591,13 @@ casInt32Array (MBA mba#) (I# ix#) (I32# old#) (I32# new#) = IO $ \ s# ->
 
 casInt64Array :: MByteArray -> Int -> Int64 -> Int64 -> IO Int64
 casInt64Array (MBA mba#) (I# ix#) (I64# old#) (I64# new#) = IO $ \ s# ->
+#if WORD_SIZE_IN_BITS < 64
     case casInt64Array# mba# ix# old# new# s# of
         (# s2#, old2# #) -> (# s2#, I64# old2# #)
+#else
+    case casInt64Array# mba# ix# (intToInt64# old#) (intToInt64# new#) s# of
+        (# s2#, old2# #) -> (# s2#, I64# (int64ToInt# old2#) #)
+#endif
 
 ------------------------------------------------------------------------
 -- Wrappers around Addr#
@@ -651,5 +663,10 @@ atomicCasWord32Ptr (Ptr addr#) (W32# old#) (W32# new#) = IO $ \ s# ->
         (# s2#, old2# #) -> (# s2#, W32# old2# #)
 atomicCasWord64Ptr :: Ptr Word64 -> Word64 -> Word64 -> IO Word64
 atomicCasWord64Ptr (Ptr addr#) (W64# old#) (W64# new#) = IO $ \ s# ->
+#if WORD_SIZE_IN_BITS < 64
     case atomicCasWord64Addr# addr# old# new# s# of
         (# s2#, old2# #) -> (# s2#, W64# old2# #)
+#else
+    case atomicCasWord64Addr# addr# (wordToWord64# old#) (wordToWord64# new#) s# of
+        (# s2#, old2# #) -> (# s2#, W64# (word64ToWord# old2#) #)
+#endif
diff --git a/testsuite/tests/primops/should_run/T4442.hs b/testsuite/tests/primops/should_run/T4442.hs
index dfdf93cc4f..ad0c882533 100644
--- a/testsuite/tests/primops/should_run/T4442.hs
+++ b/testsuite/tests/primops/should_run/T4442.hs
@@ -222,11 +222,19 @@ main = do
     (\arr i s -> case readWord8ArrayAsInt32# arr i s of (# s', a #) -> (# s', I32# a #))
     (\arr i (I32# a) s -> writeWord8ArrayAsInt32# arr i a s)
     12345678 4
+#if WORD_SIZE_IN_BITS < 64
   testInt64Array "Int64#"
     (\arr i -> I64# (indexWord8ArrayAsInt64# arr i))
     (\arr i s -> case readWord8ArrayAsInt64# arr i s of (# s', a #) -> (# s', I64# a #))
     (\arr i (I64# a) s -> writeWord8ArrayAsInt64# arr i a s)
     1234567890123 8
+#else
+  testInt64Array "Int64#"
+    (\arr i -> I64# (int64ToInt# (indexWord8ArrayAsInt64# arr i)))
+    (\arr i s -> case readWord8ArrayAsInt64# arr i s of (# s', a #) -> (# s', I64# (int64ToInt# a) #))
+    (\arr i (I64# a) s -> writeWord8ArrayAsInt64# arr i (intToInt64# a) s)
+    1234567890123 8
+#endif
   testIntArray "Int#"
     (\arr i -> I# (indexWord8ArrayAsInt# arr i))
     (\arr i s -> case readWord8ArrayAsInt# arr i s of (# s', a #) -> (# s', I# a #))
@@ -248,11 +256,19 @@ main = do
     (\arr i s -> case readWord8ArrayAsWord32# arr i s of (# s', a #) -> (# s', W32# a #))
     (\arr i (W32# a) s -> writeWord8ArrayAsWord32# arr i a s)
     12345678 4
+#if WORD_SIZE_IN_BITS < 64
   testWord64Array "Word64#"
     (\arr i -> W64# (indexWord8ArrayAsWord64# arr i))
     (\arr i s -> case readWord8ArrayAsWord64# arr i s of (# s', a #) -> (# s', W64# a #))
     (\arr i (W64# a) s -> writeWord8ArrayAsWord64# arr i a s)
     1234567890123 8
+#else
+  testWord64Array "Word64#"
+    (\arr i -> W64# (word64ToWord# (indexWord8ArrayAsWord64# arr i)))
+    (\arr i s -> case readWord8ArrayAsWord64# arr i s of (# s', a #) -> (# s', W64# (word64ToWord# a) #))
+    (\arr i (W64# a) s -> writeWord8ArrayAsWord64# arr i (wordToWord64# a) s)
+    1234567890123 8
+#endif
   testWordArray "Word#"
     (\arr i -> W# (indexWord8ArrayAsWord# arr i))
     (\arr i s -> case readWord8ArrayAsWord# arr i s of (# s', a #) -> (# s', W# a #))