[Git][ghc/ghc][wip/T22115] nativeGen/AArch64: Emit Atomic{Read, Write} inline

[Ben Gamari (@bgamari)]

Ben Gamari pushed to branch wip/T22115 at Glasgow Haskell Compiler / GHC


Commits:
ba98bfe0 by Ben Gamari at 2023-01-29T10:09:45-05:00
nativeGen/AArch64: Emit Atomic{Read,Write} inline

Previously the AtomicRead and AtomicWrite operations were emitted as
out-of-line calls. However, these tend to be very important for
performance, especially the RELAXED case (which only exists for
ThreadSanitizer checking).

Fixes #22115.

- - - - -


3 changed files:

- compiler/GHC/CmmToAsm/AArch64/CodeGen.hs
- compiler/GHC/CmmToAsm/AArch64/Instr.hs
- compiler/GHC/CmmToAsm/AArch64/Ppr.hs


Changes:

=====================================
compiler/GHC/CmmToAsm/AArch64/CodeGen.hs
=====================================
@@ -1532,9 +1532,34 @@ genCCall target dest_regs arg_regs bid = do
         MO_BRev w           -> mkCCall (bRevLabel w)
 
         -- -- Atomic read-modify-write.
+        MO_AtomicRead w ord
+          | [p_reg] <- arg_regs
+          , [dst_reg] <- dest_regs -> do
+              (p, _fmt_p, code_p) <- getSomeReg p_reg
+              platform <- getPlatform
+              let instr = case ord of
+                      MemOrderRelaxed -> LDR
+                      _               -> LDAR
+                  dst = getRegisterReg platform (CmmLocal dst_reg)
+                  code =
+                    code_p `snocOL`
+                    instr (intFormat w) (OpReg w dst) (OpReg W64 p)
+              return (code, Nothing)
+          | otherwise -> panic "mal-formed AtomicRead"
+        MO_AtomicWrite w ord
+          | [p_reg, val_reg] <- arg_regs -> do
+              (p, _fmt_p, code_p) <- getSomeReg p_reg
+              (val, fmt_val, code_val) <- getSomeReg val_reg
+              let instr = case ord of
+                      MemOrderRelaxed -> STR
+                      _               -> STLR
+                  code =
+                    code_p `appOL`
+                    code_val `snocOL`
+                    instr fmt_val (OpReg w val) (OpReg W64 p)
+              return (code, Nothing)
+          | otherwise -> panic "mal-formed AtomicWrite"
         MO_AtomicRMW w amop -> mkCCall (atomicRMWLabel w amop)
-        MO_AtomicRead w _   -> mkCCall (atomicReadLabel w)
-        MO_AtomicWrite w _  -> mkCCall (atomicWriteLabel w)
         MO_Cmpxchg w        -> mkCCall (cmpxchgLabel w)
         -- -- Should be an AtomicRMW variant eventually.
         -- -- Sequential consistent.


=====================================
compiler/GHC/CmmToAsm/AArch64/Instr.hs
=====================================
@@ -122,7 +122,9 @@ regUsageOfInstr platform instr = case instr of
   CBNZ src _               -> usage (regOp src, [])
   -- 7. Load and Store Instructions --------------------------------------------
   STR _ src dst            -> usage (regOp src ++ regOp dst, [])
+  STLR _ src dst           -> usage (regOp src ++ regOp dst, [])
   LDR _ dst src            -> usage (regOp src, regOp dst)
+  LDAR _ dst src           -> usage (regOp src, regOp dst)
   -- TODO is this right? see STR, which I'm only partial about being right?
   STP _ src1 src2 dst      -> usage (regOp src1 ++ regOp src2 ++ regOp dst, [])
   LDP _ dst1 dst2 src      -> usage (regOp src, regOp dst1 ++ regOp dst2)
@@ -254,7 +256,9 @@ patchRegsOfInstr instr env = case instr of
     CBNZ o l       -> CBNZ (patchOp o) l
     -- 7. Load and Store Instructions ------------------------------------------
     STR f o1 o2    -> STR f (patchOp o1) (patchOp o2)
+    STLR f o1 o2   -> STLR f (patchOp o1) (patchOp o2)
     LDR f o1 o2    -> LDR f (patchOp o1) (patchOp o2)
+    LDAR f o1 o2   -> LDAR f (patchOp o1) (patchOp o2)
     STP f o1 o2 o3 -> STP f (patchOp o1) (patchOp o2) (patchOp o3)
     LDP f o1 o2 o3 -> LDP f (patchOp o1) (patchOp o2) (patchOp o3)
 
@@ -608,7 +612,9 @@ data Instr
     -- Load and stores.
     -- TODO STR/LDR might want to change to STP/LDP with XZR for the second register.
     | STR Format Operand Operand -- str Xn, address-mode // Xn -> *addr
+    | STLR Format Operand Operand -- stlr Xn, address-mode // Xn -> *addr
     | LDR Format Operand Operand -- ldr Xn, address-mode // Xn <- *addr
+    | LDAR Format Operand Operand -- ldar Xn, address-mode // Xn <- *addr
     | STP Format Operand Operand Operand -- stp Xn, Xm, address-mode // Xn -> *addr, Xm -> *(addr + 8)
     | LDP Format Operand Operand Operand -- stp Xn, Xm, address-mode // Xn <- *addr, Xm <- *(addr + 8)
 


=====================================
compiler/GHC/CmmToAsm/AArch64/Ppr.hs
=====================================
@@ -469,6 +469,7 @@ pprInstr platform instr = case instr of
   STR _f o1@(OpReg W16 (RegReal (RealRegSingle i))) o2 | i < 32 ->
     op2 (text "\tstrh") o1 o2
   STR _f o1 o2 -> op2 (text "\tstr") o1 o2
+  STLR _f o1 o2 -> op2 (text "\tstlr") o1 o2
 
 #if defined(darwin_HOST_OS)
   LDR _f o1 (OpImm (ImmIndex lbl' off)) | Just (_info, lbl) <- dynamicLinkerLabelInfo lbl' ->
@@ -533,6 +534,7 @@ pprInstr platform instr = case instr of
   LDR _f o1@(OpReg W16 (RegReal (RealRegSingle i))) o2 | i < 32 ->
     op2 (text "\tldrh") o1 o2
   LDR _f o1 o2 -> op2 (text "\tldr") o1 o2
+  LDAR _f o1 o2 -> op2 (text "\tldar") o1 o2
 
   STP _f o1 o2 o3 -> op3 (text "\tstp") o1 o2 o3
   LDP _f o1 o2 o3 -> op3 (text "\tldp") o1 o2 o3



View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/ba98bfe0eb5e71e9e04a2a85e455f9dee12406b9

-- 
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/ba98bfe0eb5e71e9e04a2a85e455f9dee12406b9
You're receiving this email because of your account on gitlab.haskell.org.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-commits/attachments/20230129/3974704d/attachment-0001.html>