[Git][ghc/ghc][wip/andreask/arm_mem_model] NCG: AArch64 - Add -finter-module-far-jumps.
Andreas Klebinger (@AndreasK)
gitlab at gitlab.haskell.org
Fri Mar 7 16:16:14 UTC 2025
Andreas Klebinger pushed to branch wip/andreask/arm_mem_model at Glasgow Haskell Compiler / GHC
Commits:
7b13d63b by Andreas Klebinger at 2025-03-07T16:54:06+01:00
NCG: AArch64 - Add -finter-module-far-jumps.
When enabled the arm backend will assume jumps to targets outside of the
current module are further than 128MB away.
This will allow for code to work if:
* The current module results in less than 128MB of code.
* The whole program is loaded within a 4GB memory region.
We enable this by default on mac where the lack of split sections can
sometimes cause us to go over this limit - see #24648.
This works around #24648 for now.
-------------------------
Metric Increase:
T783
-------------------------
- - - - -
7 changed files:
- compiler/GHC/CmmToAsm/AArch64/CodeGen.hs
- compiler/GHC/CmmToAsm/Config.hs
- compiler/GHC/Driver/Config/CmmToAsm.hs
- compiler/GHC/Driver/DynFlags.hs
- compiler/GHC/Driver/Flags.hs
- compiler/GHC/Driver/Session.hs
- docs/users_guide/using-optimisation.rst
Changes:
=====================================
compiler/GHC/CmmToAsm/AArch64/CodeGen.hs
=====================================
@@ -24,7 +24,7 @@ import GHC.Cmm.DebugBlock
import GHC.CmmToAsm.Monad
( NatM, getNewRegNat
, getPicBaseMaybeNat, getPlatform, getConfig
- , getDebugBlock, getFileId, getNewLabelNat
+ , getDebugBlock, getFileId, getNewLabelNat, getThisModuleNat
)
-- import GHC.CmmToAsm.Instr
import GHC.CmmToAsm.PIC
@@ -1505,8 +1505,19 @@ assignReg_FltCode = assignReg_IntCode
-- Jumps
genJump :: CmmExpr{-the branch target-} -> NatM InstrBlock
-genJump expr@(CmmLit (CmmLabel lbl))
- = return $ unitOL (annExpr expr (J (TLabel lbl)))
+genJump expr@(CmmLit (CmmLabel lbl)) = do
+ cur_mod <- getThisModuleNat
+ !useFarJumps <- ncgEnableInterModuleFarJumps <$> getConfig
+ let is_local = isLocalCLabel cur_mod lbl
+
+ -- We prefer to generate a near jump using a simble `B` instruction
+ -- with a range (+/-128MB). But if the target is outside the current module
+ -- we might have to account for large code offsets. (#24648)
+ if not useFarJumps || is_local
+ then return $ unitOL (annExpr expr (J (TLabel lbl)))
+ else do
+ (target, _format, code) <- getSomeReg expr
+ return (code `appOL` unitOL (annExpr expr (J (TReg target))))
genJump expr = do
(target, _format, code) <- getSomeReg expr
=====================================
compiler/GHC/CmmToAsm/Config.hs
=====================================
@@ -47,6 +47,7 @@ data NCGConfig = NCGConfig
, ncgDwarfSourceNotes :: !Bool -- ^ Enable GHC-specific source note DIEs
, ncgCmmStaticPred :: !Bool -- ^ Enable static control-flow prediction
, ncgEnableShortcutting :: !Bool -- ^ Enable shortcutting (don't jump to blocks only containing a jump)
+ , ncgEnableInterModuleFarJumps:: !Bool -- ^ Use far-jumps for cross-module jumps.
, ncgComputeUnwinding :: !Bool -- ^ Compute block unwinding tables
, ncgEnableDeadCodeElimination :: !Bool -- ^ Whether to enable the dead-code elimination
}
=====================================
compiler/GHC/Driver/Config/CmmToAsm.hs
=====================================
@@ -70,6 +70,7 @@ initNCGConfig dflags this_mod = NCGConfig
, ncgExposeInternalSymbols = gopt Opt_ExposeInternalSymbols dflags
, ncgCmmStaticPred = gopt Opt_CmmStaticPred dflags
, ncgEnableShortcutting = gopt Opt_AsmShortcutting dflags
+ , ncgEnableInterModuleFarJumps = gopt Opt_InterModuleFarJumps dflags
, ncgComputeUnwinding = debugLevel dflags > 0
, ncgEnableDeadCodeElimination = not (gopt Opt_InfoTableMap dflags)
-- Disable when -finfo-table-map is on (#20428)
=====================================
compiler/GHC/Driver/DynFlags.hs
=====================================
@@ -1191,6 +1191,17 @@ defaultFlags settings
++ validHoleFitDefaults
+ -- Platform/OS specific stuff
+ ++ case platformOS platform of
+ -- On mac naturally the linker is broken for jumps with very large
+ -- offsets. (#24648) So we enable HugeTextSections by default to generate
+ -- far jumps when crossing module boundries instead. As these don't rely
+ -- on linker fixups.
+ OSDarwin
+ | platformArch platform == ArchAArch64
+ -> [Opt_InterModuleFarJumps]
+ _ -> []
+
where platform = sTargetPlatform settings
=====================================
compiler/GHC/Driver/Flags.hs
=====================================
@@ -657,6 +657,7 @@ data GeneralFlag
| Opt_CmmElimCommonBlocks
| Opt_CmmControlFlow
| Opt_AsmShortcutting
+ | Opt_InterModuleFarJumps
| Opt_OmitYields
| Opt_FunToThunk -- deprecated
| Opt_DictsStrict -- be strict in argument dictionaries
@@ -906,6 +907,7 @@ optimisationFlags = EnumSet.fromList
, Opt_CmmSink
, Opt_CmmElimCommonBlocks
, Opt_AsmShortcutting
+ , Opt_InterModuleFarJumps
, Opt_FunToThunk
, Opt_DmdTxDictSel
, Opt_Loopification
=====================================
compiler/GHC/Driver/Session.hs
=====================================
@@ -2462,6 +2462,7 @@ fFlagsDeps = [
flagSpec "gen-manifest" Opt_GenManifest,
flagSpec "ghci-history" Opt_GhciHistory,
flagSpec "ghci-leak-check" Opt_GhciLeakCheck,
+ flagSpec "inter-module-far-jumps" Opt_InterModuleFarJumps,
flagSpec "validate-ide-info" Opt_ValidateHie,
flagGhciSpec "local-ghci-history" Opt_LocalGhciHistory,
flagGhciSpec "no-it" Opt_NoIt,
=====================================
docs/users_guide/using-optimisation.rst
=====================================
@@ -687,6 +687,23 @@ as such you shouldn't need to set any of them explicitly. A flag
``-fno-full-laziness``. If that is inconvenient for you, please leave a
comment `on the issue tracker (#21204) <https://gitlab.haskell.org/ghc/ghc/-/issues/21204>`__.
+.. ghc-flag:: -finter-module-far-jumps
+ :shortdesc: Assume code sections can be very large.
+ :type: dynamic
+ :reverse: -fno-inter-module-far-jumps
+ :category:
+
+ :default: On for AArch64 MacOS
+
+ This flag forces GHC to use far jumps instead of near jumps for all jumps
+ which cross module boundries. This removes the need for jump islands/linker
+ jump fixups which some linkers struggle to deal with. (:ghc-ticket:`24648`)
+
+ This comes at a very modest code size (~2%) and runtime (~0.6%) overhead.
+
+ Note that this flag currently only affects the NCG AArch64 backend.
+
+
.. ghc-flag:: -fignore-asserts
:shortdesc: Ignore assertions in the source. Implied by :ghc-flag:`-O`.
:type: dynamic
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/7b13d63bd4e421f3daadcf8f20e634f5628543cd
--
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/7b13d63bd4e421f3daadcf8f20e634f5628543cd
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/20250307/ab62baa6/attachment-0001.html>
More information about the ghc-commits
mailing list