winio: Add Atomic Exchange PrimOp and implement Atomic Ptr exchanges.

The initial version was rewritten by Tamar Christina.
It was rewritten in large parts by Andreas Klebinger.

Co-authored-by: Andreas Klebinger <klebinger.andreas@gmx.at>

12 files changed, 82 insertions, 3 deletions

diff --git a/compiler/GHC/Builtin/primops.txt.pp b/compiler/GHC/Builtin/primops.txt.pp
index 859fb99ae7..ee7bdac29a 100644
--- a/compiler/GHC/Builtin/primops.txt.pp
+++ b/compiler/GHC/Builtin/primops.txt.pp
@@ -2473,6 +2473,18 @@ primop  WriteOffAddrOp_Word64 "writeWord64OffAddr#" GenPrimOp
    with has_side_effects = True
         can_fail         = True
 
+primop  InterlockedExchange_Addr "interlockedExchangeAddr#" GenPrimOp
+   Addr# -> Addr# -> State# s -> (# State# s, Addr# #)
+   {The atomic exchange operation. Atomically exchanges the value at the first address
+    with the Addr# given as second argument. Implies a read barrier.}
+   with has_side_effects = True
+
+primop  InterlockedExchange_Int "interlockedExchangeInt#" GenPrimOp
+   Addr# -> Int# -> State# s -> (# State# s, Int# #)
+   {The atomic exchange operation. Atomically exchanges the value at the address
+    with the given value. Returns the old value. Implies a read barrier.}
+   with has_side_effects = True
+
 ------------------------------------------------------------------------
 section "Mutable variables"
         {Operations on MutVar\#s.}
diff --git a/compiler/GHC/Cmm/MachOp.hs b/compiler/GHC/Cmm/MachOp.hs
index 1b3dd2a531..077f663161 100644
--- a/compiler/GHC/Cmm/MachOp.hs
+++ b/compiler/GHC/Cmm/MachOp.hs
@@ -632,6 +632,9 @@ data CallishMachOp
   | MO_AtomicRead Width
   | MO_AtomicWrite Width
   | MO_Cmpxchg Width
+  -- Should be an AtomicRMW variant eventually.
+  -- Sequential consistent.
+  | MO_Xchg Width
   deriving (Eq, Show)
 
 -- | The operation to perform atomically.
diff --git a/compiler/GHC/Cmm/Parser.y b/compiler/GHC/Cmm/Parser.y
index bb502f8cbe..41d5d3d6d6 100644
--- a/compiler/GHC/Cmm/Parser.y
+++ b/compiler/GHC/Cmm/Parser.y
@@ -1022,7 +1022,12 @@ callishMachOps = listToUFM $
         ( "cmpxchg8",  (MO_Cmpxchg W8,)),
         ( "cmpxchg16", (MO_Cmpxchg W16,)),
         ( "cmpxchg32", (MO_Cmpxchg W32,)),
-        ( "cmpxchg64", (MO_Cmpxchg W64,))
+        ( "cmpxchg64", (MO_Cmpxchg W64,)),
+
+        ( "xchg8",  (MO_Xchg W8,)),
+        ( "xchg16", (MO_Xchg W16,)),
+        ( "xchg32", (MO_Xchg W32,)),
+        ( "xchg64", (MO_Xchg W64,))
 
         -- ToDo: the rest, maybe
         -- edit: which rest?
diff --git a/compiler/GHC/CmmToAsm/CPrim.hs b/compiler/GHC/CmmToAsm/CPrim.hs
index fc2d06262b..826a6e2224 100644
--- a/compiler/GHC/CmmToAsm/CPrim.hs
+++ b/compiler/GHC/CmmToAsm/CPrim.hs
@@ -4,6 +4,7 @@ module GHC.CmmToAsm.CPrim
     , atomicWriteLabel
     , atomicRMWLabel
     , cmpxchgLabel
+    , xchgLabel
     , popCntLabel
     , pdepLabel
     , pextLabel
@@ -105,6 +106,15 @@ atomicRMWLabel w amop = "hs_atomic_" ++ pprFunName amop ++ pprWidth w
     pprFunName AMO_Or   = "or"
     pprFunName AMO_Xor  = "xor"
 
+xchgLabel :: Width -> String
+xchgLabel w = "hs_xchg" ++ pprWidth w
+  where
+    pprWidth W8  = "8"
+    pprWidth W16 = "16"
+    pprWidth W32 = "32"
+    pprWidth W64 = "64"
+    pprWidth w   = pprPanic "xchgLabel: Unsupported word width " (ppr w)
+
 cmpxchgLabel :: Width -> String
 cmpxchgLabel w = "hs_cmpxchg" ++ pprWidth w
   where
diff --git a/compiler/GHC/CmmToAsm/PPC/CodeGen.hs b/compiler/GHC/CmmToAsm/PPC/CodeGen.hs
index 764945c2bc..367faa25bc 100644
--- a/compiler/GHC/CmmToAsm/PPC/CodeGen.hs
+++ b/compiler/GHC/CmmToAsm/PPC/CodeGen.hs
@@ -2024,6 +2024,7 @@ genCCall' config gcp target dest_regs args
                     MO_Ctz _     -> unsupported
                     MO_AtomicRMW {} -> unsupported
                     MO_Cmpxchg w -> (fsLit $ cmpxchgLabel w, False)
+                    MO_Xchg w    -> (fsLit $ xchgLabel w, False)
                     MO_AtomicRead _  -> unsupported
                     MO_AtomicWrite _ -> unsupported
 
diff --git a/compiler/GHC/CmmToAsm/SPARC/CodeGen.hs b/compiler/GHC/CmmToAsm/SPARC/CodeGen.hs
index 2112983e73..9ecb0fcc76 100644
--- a/compiler/GHC/CmmToAsm/SPARC/CodeGen.hs
+++ b/compiler/GHC/CmmToAsm/SPARC/CodeGen.hs
@@ -677,6 +677,7 @@ outOfLineMachOp_table mop
         MO_Ctz w     -> fsLit $ ctzLabel w
         MO_AtomicRMW w amop -> fsLit $ atomicRMWLabel w amop
         MO_Cmpxchg w -> fsLit $ cmpxchgLabel w
+        MO_Xchg w -> fsLit $ xchgLabel w
         MO_AtomicRead w -> fsLit $ atomicReadLabel w
         MO_AtomicWrite w -> fsLit $ atomicWriteLabel w
 
diff --git a/compiler/GHC/CmmToAsm/X86/CodeGen.hs b/compiler/GHC/CmmToAsm/X86/CodeGen.hs
index dab4c62122..d842bcc94a 100644
--- a/compiler/GHC/CmmToAsm/X86/CodeGen.hs
+++ b/compiler/GHC/CmmToAsm/X86/CodeGen.hs
@@ -2518,6 +2518,22 @@ genCCall' _ is32Bit (PrimTarget (MO_Cmpxchg width)) [dst] [addr, old, new] _ = d
   where
     format = intFormat width
 
+genCCall' config is32Bit (PrimTarget (MO_Xchg width)) [dst] [addr, value] _
+  | (is32Bit && width == W64) = panic "gencCall: 64bit atomic exchange not supported on 32bit platforms"
+  | otherwise = do
+    let dst_r = getRegisterReg platform (CmmLocal dst)
+    Amode amode addr_code <- getSimpleAmode is32Bit addr
+    (newval, newval_code) <- getSomeReg value
+    -- Copy the value into the target register, perform the exchange.
+    let code     = toOL
+                   [ MOV format (OpReg newval) (OpReg dst_r)
+                   , XCHG format (OpAddr amode) dst_r
+                   ]
+    return $ addr_code `appOL` newval_code `appOL` code
+  where
+    format = intFormat width
+    platform = ncgPlatform config
+
 genCCall' _ is32Bit target dest_regs args bid = do
   platform <- ncgPlatform <$> getConfig
   case (target, dest_regs) of
@@ -3213,6 +3229,7 @@ outOfLineCmmOp bid mop res args
               MO_AtomicRead _  -> fsLit "atomicread"
               MO_AtomicWrite _ -> fsLit "atomicwrite"
               MO_Cmpxchg _     -> fsLit "cmpxchg"
+              MO_Xchg _        -> should_be_inline
 
               MO_UF_Conv _ -> unsupported
 
@@ -3232,6 +3249,11 @@ outOfLineCmmOp bid mop res args
               (MO_Prefetch_Data _ ) -> unsupported
         unsupported = panic ("outOfLineCmmOp: " ++ show mop
                           ++ " not supported here")
+        -- If we generate a call for the given primop
+        -- something went wrong.
+        should_be_inline = panic ("outOfLineCmmOp: " ++ show mop
+                          ++ " should be handled inline")
+
 
 -- -----------------------------------------------------------------------------
 -- Generating a table-branch
diff --git a/compiler/GHC/CmmToAsm/X86/Instr.hs b/compiler/GHC/CmmToAsm/X86/Instr.hs
index 67a6ffb930..eafe7198e6 100644
--- a/compiler/GHC/CmmToAsm/X86/Instr.hs
+++ b/compiler/GHC/CmmToAsm/X86/Instr.hs
@@ -329,6 +329,7 @@ data Instr
         | LOCK        Instr -- lock prefix
         | XADD        Format Operand Operand -- src (r), dst (r/m)
         | CMPXCHG     Format Operand Operand -- src (r), dst (r/m), eax implicit
+        | XCHG        Format Operand Reg     -- src (r/m), dst (r/m)
         | MFENCE
 
 data PrefetchVariant = NTA | Lvl0 | Lvl1 | Lvl2
@@ -431,6 +432,7 @@ x86_regUsageOfInstr platform instr
     LOCK i              -> x86_regUsageOfInstr platform i
     XADD _ src dst      -> usageMM src dst
     CMPXCHG _ src dst   -> usageRMM src dst (OpReg eax)
+    XCHG _ src dst      -> usageMM src (OpReg dst)
     MFENCE -> noUsage
 
     _other              -> panic "regUsage: unrecognised instr"
@@ -460,6 +462,7 @@ x86_regUsageOfInstr platform instr
     usageMM :: Operand -> Operand -> RegUsage
     usageMM (OpReg src) (OpReg dst) = mkRU [src, dst] [src, dst]
     usageMM (OpReg src) (OpAddr ea) = mkRU (use_EA ea [src]) [src]
+    usageMM (OpAddr ea) (OpReg dst) = mkRU (use_EA ea [dst]) [dst]
     usageMM _ _                     = panic "X86.RegInfo.usageMM: no match"
 
     -- 3 operand form; first operand Read; second Modified; third Modified
@@ -589,6 +592,7 @@ x86_patchRegsOfInstr instr env
     LOCK i               -> LOCK (x86_patchRegsOfInstr i env)
     XADD fmt src dst     -> patch2 (XADD fmt) src dst
     CMPXCHG fmt src dst  -> patch2 (CMPXCHG fmt) src dst
+    XCHG fmt src dst     -> XCHG fmt (patchOp src) (env dst)
     MFENCE               -> instr
 
     _other              -> panic "patchRegs: unrecognised instr"
diff --git a/compiler/GHC/CmmToAsm/X86/Ppr.hs b/compiler/GHC/CmmToAsm/X86/Ppr.hs
index 6ed5842389..6be1c8ef4d 100644
--- a/compiler/GHC/CmmToAsm/X86/Ppr.hs
+++ b/compiler/GHC/CmmToAsm/X86/Ppr.hs
@@ -824,6 +824,9 @@ pprInstr platform i = case i of
    SETCC cond op
       -> pprCondInstr (sLit "set") cond (pprOperand platform II8 op)
 
+   XCHG format src val
+      -> pprFormatOpReg (sLit "xchg") format src val
+
    JXX cond blockid
       -> pprCondInstr (sLit "j") cond (ppr lab)
          where lab = blockLbl blockid
diff --git a/compiler/GHC/CmmToC.hs b/compiler/GHC/CmmToC.hs
index d7b3fb05eb..bcead719a9 100644
--- a/compiler/GHC/CmmToC.hs
+++ b/compiler/GHC/CmmToC.hs
@@ -835,6 +835,7 @@ pprCallishMachOp_for_C mop
         (MO_Ctz w)      -> ptext (sLit $ ctzLabel w)
         (MO_AtomicRMW w amop) -> ptext (sLit $ atomicRMWLabel w amop)
         (MO_Cmpxchg w)  -> ptext (sLit $ cmpxchgLabel w)
+        (MO_Xchg w)     -> ptext (sLit $ xchgLabel w)
         (MO_AtomicRead w)  -> ptext (sLit $ atomicReadLabel w)
         (MO_AtomicWrite w) -> ptext (sLit $ atomicWriteLabel w)
         (MO_UF_Conv w)  -> ptext (sLit $ word2FloatLabel w)
diff --git a/compiler/GHC/CmmToLlvm/CodeGen.hs b/compiler/GHC/CmmToLlvm/CodeGen.hs
index 672fc84e43..53f17f545c 100644
--- a/compiler/GHC/CmmToLlvm/CodeGen.hs
+++ b/compiler/GHC/CmmToLlvm/CodeGen.hs
@@ -281,6 +281,16 @@ genCall (PrimTarget (MO_Cmpxchg _width))
     retVar' <- doExprW targetTy $ ExtractV retVar 0
     statement $ Store retVar' dstVar
 
+genCall (PrimTarget (MO_Xchg _width)) [dst] [addr, val] = runStmtsDecls $ do
+    dstV <- getCmmRegW (CmmLocal dst) :: WriterT LlvmAccum LlvmM LlvmVar
+    addrVar <- exprToVarW addr
+    valVar <- exprToVarW val
+    let ptrTy = pLift $ getVarType valVar
+        ptrExpr = Cast LM_Inttoptr addrVar ptrTy
+    ptrVar <- doExprW ptrTy ptrExpr
+    resVar <- doExprW (getVarType valVar) (AtomicRMW LAO_Xchg ptrVar valVar SyncSeqCst)
+    statement $ Store resVar dstV
+
 genCall (PrimTarget (MO_AtomicWrite _width)) [] [addr, val] = runStmtsDecls $ do
     addrVar <- exprToVarW addr
     valVar <- exprToVarW val
@@ -856,6 +866,7 @@ cmmPrimOpFunctions mop = do
     MO_AtomicRMW _ _ -> unsupported
     MO_AtomicWrite _ -> unsupported
     MO_Cmpxchg _     -> unsupported
+    MO_Xchg _        -> unsupported
 
 -- | Tail function calls
 genJump :: CmmExpr -> [GlobalReg] -> LlvmM StmtData
@@ -1946,10 +1957,10 @@ toIWord platform = mkIntLit (llvmWord platform)
 
 
 -- | Error functions
-panic :: String -> a
+panic :: HasCallStack => String -> a
 panic s = Outputable.panic $ "GHC.CmmToLlvm.CodeGen." ++ s
 
-pprPanic :: String -> SDoc -> a
+pprPanic :: HasCallStack => String -> SDoc -> a
 pprPanic s d = Outputable.pprPanic ("GHC.CmmToLlvm.CodeGen." ++ s) d
 
 
diff --git a/compiler/GHC/StgToCmm/Prim.hs b/compiler/GHC/StgToCmm/Prim.hs
index e70f50ee84..fee96f31f8 100644
--- a/compiler/GHC/StgToCmm/Prim.hs
+++ b/compiler/GHC/StgToCmm/Prim.hs
@@ -856,6 +856,12 @@ emitPrimOp dflags = \case
   Word2DoubleOp -> \[w] -> opAllDone $ \[res] -> do
     emitPrimCall [res] (MO_UF_Conv W64) [w]
 
+-- Atomic operations
+  InterlockedExchange_Addr -> \[src, value] -> opAllDone $ \[res] ->
+    emitPrimCall [res] (MO_Xchg (wordWidth platform)) [src, value]
+  InterlockedExchange_Int -> \[src, value] -> opAllDone $ \[res] ->
+    emitPrimCall [res] (MO_Xchg (wordWidth platform)) [src, value]
+
 -- SIMD primops
   (VecBroadcastOp vcat n w) -> \[e] -> opAllDone $ \[res] -> do
     checkVecCompatibility dflags vcat n w