- ghc-commits - Haskell.org

[Git][ghc/ghc][master] ghci: fix bootstrapping with 9.12.3-rc1 and above
by Marge Bot (＠marge-bot) 24 Sep '25

24 Sep '25

Marge Bot pushed to branch master at Glasgow Haskell Compiler / GHC Commits: f8919262 by Cheng Shao at 2025-09-23T20:44:54-04:00 ghci: fix bootstrapping with 9.12.3-rc1 and above This patch fixes bootstrapping GHC with 9.12.3-rc1 and above. ghci defines `Binary` instance for `HalfWord` in `ghc-heap`, which is a proper `newtype` in 9.14 and starting from 9.12.3. Given we don't build `ghc-heap` in stage0, we need to fix this predicate so that it corresponds to the boot ghc versions that contain the right version of `ghc-heap`. - - - - - 1 changed file: - libraries/ghci/GHCi/Message.hs Changes: ===================================== libraries/ghci/GHCi/Message.hs ===================================== @@ -525,7 +525,7 @@ instance Binary (FunPtr a) where put = put . castFunPtrToPtr get = castPtrToFunPtr <$> get -#if MIN_VERSION_ghc_internal(9,1500,0) +#if MIN_VERSION_GLASGOW_HASKELL(9,12,2,20250919) instance Binary Heap.HalfWord where put x = put (fromIntegral x :: Word32) get = fromIntegral <$> (get :: Get Word32) View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/f8919262dc3034851fbefab86fe45dd… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/f8919262dc3034851fbefab86fe45dd… You're receiving this email because of your account on gitlab.haskell.org.

1 0

[Git][ghc/ghc][master] docs: Fix typo in scoped_type_variables.rst
by Marge Bot (＠marge-bot) 24 Sep '25

24 Sep '25

Marge Bot pushed to branch master at Glasgow Haskell Compiler / GHC Commits: 30ef0aac by Simon Hengel at 2025-09-23T20:44:12-04:00 docs: Fix typo in scoped_type_variables.rst - - - - - 1 changed file: - docs/users_guide/exts/scoped_type_variables.rst Changes: ===================================== docs/users_guide/exts/scoped_type_variables.rst ===================================== @@ -6,7 +6,7 @@ Lexically scoped type variables =============================== .. extension:: ScopedTypeVariables - :shortdesc: Lexically scope explicitly-introduced type variables. + :shortdesc: Lexically scoped explicitly-introduced type variables. :implies: :extension:`ExplicitForAll` View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/30ef0aac1f03acbef5856d16081e24f… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/30ef0aac1f03acbef5856d16081e24f… You're receiving this email because of your account on gitlab.haskell.org.

1 0

[Git][ghc/ghc][master] Allow Core plugins to access unoptimized Core (#23337)
by Marge Bot (＠marge-bot) 24 Sep '25

24 Sep '25

Marge Bot pushed to branch master at Glasgow Haskell Compiler / GHC Commits: 50f6be09 by Sylvain Henry at 2025-09-23T20:43:29-04:00 Allow Core plugins to access unoptimized Core (#23337) Make the first simple optimization pass after desugaring a real CoreToDo pass. This allows CorePlugins to decide whether they want to be executed before or after this pass. - - - - - 6 changed files: - compiler/GHC/Core/Opt/Pipeline.hs - compiler/GHC/Core/Opt/Pipeline/Types.hs - compiler/GHC/HsToCore.hs - testsuite/tests/plugins/annotation-plugin/SayAnnNames.hs - testsuite/tests/plugins/late-plugin/LatePlugin.hs - testsuite/tests/plugins/simple-plugin/Simple/ReplacePlugin.hs Changes: ===================================== compiler/GHC/Core/Opt/Pipeline.hs ===================================== @@ -13,6 +13,7 @@ import GHC.Prelude import GHC.Driver.DynFlags import GHC.Driver.Plugins ( withPlugins, installCoreToDos ) import GHC.Driver.Env +import GHC.Driver.Config (initSimpleOpts) import GHC.Driver.Config.Core.Lint ( endPass ) import GHC.Driver.Config.Core.Opt.LiberateCase ( initLiberateCaseOpts ) import GHC.Driver.Config.Core.Opt.Simplify ( initSimplifyOpts, initSimplMode, initGentleSimplMode ) @@ -21,9 +22,10 @@ import GHC.Driver.Config.Core.Rules ( initRuleOpts ) import GHC.Platform.Ways ( hasWay, Way(WayProf) ) import GHC.Core +import GHC.Core.SimpleOpt (simpleOptPgm) import GHC.Core.Opt.CSE ( cseProgram ) import GHC.Core.Rules ( RuleBase, ruleCheckProgram, getRules ) -import GHC.Core.Ppr ( pprCoreBindings ) +import GHC.Core.Ppr ( pprCoreBindings, pprRules ) import GHC.Core.Utils ( dumpIdInfoOfProgram ) import GHC.Core.Lint ( lintAnnots ) import GHC.Core.Lint.Interactive ( interactiveInScope ) @@ -202,10 +204,14 @@ getCoreToDo dflags hpt_rule_base extra_vars core_todo = [ - -- We want to do the static argument transform before full laziness as it - -- may expose extra opportunities to float things outwards. However, to fix - -- up the output of the transformation we need at do at least one simplify - -- after this before anything else + -- We always perform a run of the simple optimizer after desugaring to + -- remove really bad code + CoreDesugarOpt, + + -- We want to do the static argument transform before full laziness as it + -- may expose extra opportunities to float things outwards. However, to fix + -- up the output of the transformation we need at do at least one simplify + -- after this before anything else runWhen static_args (CoreDoPasses [ simpl_gently, CoreDoStaticArgs ]), -- initial simplify: mk specialiser happy: minimum effort please @@ -467,6 +473,7 @@ doCorePass pass guts = do let fam_envs = (p_fam_env, mg_fam_inst_env guts) let updateBinds f = return $ guts { mg_binds = f (mg_binds guts) } let updateBindsM f = f (mg_binds guts) >>= \b' -> return $ guts { mg_binds = b' } + let updateBindsAndRulesM f = f (mg_binds guts) (mg_rules guts) >>= \(b',r') -> return $ guts { mg_binds = b', mg_rules = r' } -- Important to force this now as name_ppr_ctx lives through an entire phase in -- the optimiser and if it's not forced then the entire previous `ModGuts` will -- be retained until the end of the phase. (See #24328 for more analysis) @@ -479,6 +486,9 @@ doCorePass pass guts = do case pass of + CoreDesugarOpt -> {-# SCC "DesugarOpt" #-} + updateBindsAndRulesM (desugarOpt dflags logger (mg_module guts)) + CoreDoSimplify opts -> {-# SCC "Simplify" #-} liftIOWithCount $ simplifyPgm logger (hsc_unit_env hsc_env) name_ppr_ctx opts guts @@ -537,7 +547,6 @@ doCorePass pass guts = do CoreDoPluginPass _ p -> {-# SCC "Plugin" #-} p guts CoreDesugar -> pprPanic "doCorePass" (ppr pass) - CoreDesugarOpt -> pprPanic "doCorePass" (ppr pass) CoreTidy -> pprPanic "doCorePass" (ppr pass) CorePrep -> pprPanic "doCorePass" (ppr pass) @@ -580,3 +589,22 @@ dmdAnal logger before_ww dflags fam_envs rules binds = do dumpIdInfoOfProgram (hasPprDebug dflags) (ppr . zapDmdEnvSig . dmdSigInfo) binds_plus_dmds -- See Note [Stamp out space leaks in demand analysis] in GHC.Core.Opt.DmdAnal seqBinds binds_plus_dmds `seq` return binds_plus_dmds + + +-- | Simple optimization after desugaring. +-- +-- This performs some quick basic optimizations even with -O0. +-- See Note [The simple optimiser] for details. +-- +-- We could call it directly in the desugarer but we implement it as the first +-- Core-to-Core pass to accomodate Core plugins that want to see Core even +-- before the first (simple) optimization took place. See #23337 +desugarOpt :: DynFlags -> Logger -> Module -> CoreProgram -> [CoreRule] -> CoreM (CoreProgram,[CoreRule]) +desugarOpt dflags logger mod binds rules = liftIO $ do + let simpl_opts = initSimpleOpts dflags + let !(ds_binds, ds_rules_for_imps, occ_anald_binds) = simpleOptPgm simpl_opts mod binds rules + + putDumpFileMaybe logger Opt_D_dump_occur_anal "Occurrence analysis" + FormatCore (pprCoreBindings occ_anald_binds $$ pprRules ds_rules_for_imps ) + + pure (ds_binds, ds_rules_for_imps) ===================================== compiler/GHC/Core/Opt/Pipeline/Types.hs ===================================== @@ -58,8 +58,7 @@ data CoreToDo -- These are diff core-to-core passes, | CoreDoPasses [CoreToDo] -- lists of these things | CoreDesugar -- Right after desugaring, no simple optimisation yet! - | CoreDesugarOpt -- CoreDesugarXXX: Not strictly a core-to-core pass, but produces - -- Core output, and hence useful to pass to endPass + | CoreDesugarOpt -- Simple optimisation after desugaring | CoreTidy | CorePrep ===================================== compiler/GHC/HsToCore.hs ===================================== @@ -48,7 +48,7 @@ import GHC.Core.TyCo.Compare( eqType ) import GHC.Core.TyCon ( tyConDataCons ) import GHC.Core import GHC.Core.FVs ( exprsSomeFreeVarsList, exprFreeVars ) -import GHC.Core.SimpleOpt ( simpleOptPgm, simpleOptExpr ) +import GHC.Core.SimpleOpt ( simpleOptExpr ) import GHC.Core.Utils import GHC.Core.Unfold.Make import GHC.Core.Coercion @@ -200,27 +200,18 @@ deSugar hsc_env do { -- Add export flags to bindings keep_alive <- readIORef keep_var - ; let (rules_for_locals, rules_for_imps) = partition isLocalRule all_rules + ; let (rules_for_locals, ds_rules_for_imps) = partition isLocalRule all_rules final_prs = addExportFlagsAndRules bcknd export_set keep_alive rules_for_locals (fromOL all_prs) - final_pgm = combineEvBinds ds_ev_binds final_prs + ds_binds = combineEvBinds ds_ev_binds final_prs -- Notice that we put the whole lot in a big Rec, even the foreign binds -- When compiling PrelFloat, which defines data Float = F# Float# -- we want F# to be in scope in the foreign marshalling code! -- You might think it doesn't matter, but the simplifier brings all top-level -- things into the in-scope set before simplifying; so we get no unfolding for F#! - ; endPassHscEnvIO hsc_env name_ppr_ctx CoreDesugar final_pgm rules_for_imps - ; let simpl_opts = initSimpleOpts dflags - ; let (ds_binds, ds_rules_for_imps, occ_anald_binds) - = simpleOptPgm simpl_opts mod final_pgm rules_for_imps - -- The simpleOptPgm gets rid of type - -- bindings plus any stupid dead code - ; putDumpFileMaybe logger Opt_D_dump_occur_anal "Occurrence analysis" - FormatCore (pprCoreBindings occ_anald_binds $$ pprRules ds_rules_for_imps ) - - ; endPassHscEnvIO hsc_env name_ppr_ctx CoreDesugarOpt ds_binds ds_rules_for_imps + ; endPassHscEnvIO hsc_env name_ppr_ctx CoreDesugar ds_binds ds_rules_for_imps ; let pluginModules = map lpModule (loadedPlugins (hsc_plugins hsc_env)) home_unit = hsc_home_unit hsc_env ===================================== testsuite/tests/plugins/annotation-plugin/SayAnnNames.hs ===================================== @@ -19,13 +19,14 @@ pass :: ModGuts -> CoreM ModGuts pass g = do dflags <- getDynFlags mapM_ (printAnn dflags g) (mg_binds g) >> return g - where printAnn :: DynFlags -> ModGuts -> CoreBind -> CoreM CoreBind - printAnn dflags guts bndr@(NonRec b _) = do + where printAnn :: DynFlags -> ModGuts -> CoreBind -> CoreM () + printAnn dflags guts (NonRec b _) = lookupAnn dflags guts b + printAnn dflags guts (Rec ps) = mapM_ (lookupAnn dflags guts . fst) ps + + lookupAnn dflags guts b = do anns <- annotationsOn guts b :: CoreM [SomeAnn] unless (null anns) $ putMsgS $ "Annotated binding found: " ++ showSDoc dflags (ppr b) - return bndr - printAnn _ _ bndr = return bndr annotationsOn :: Data a => ModGuts -> CoreBndr -> CoreM [a] annotationsOn guts bndr = do ===================================== testsuite/tests/plugins/late-plugin/LatePlugin.hs ===================================== @@ -43,8 +43,13 @@ editCoreBinding early modName pgm = do pure $ go pgm where go :: [CoreBind] -> [CoreBind] - go (b@(NonRec v e) : bs) - | occNameString (getOccName v) == "testBinding" && exprType e `eqType` intTy = - NonRec v (mkUncheckedIntExpr $ bool 222222 111111 early) : bs - go (b:bs) = b : go bs + go (Rec ps : bs) = Rec (map (uncurry (go_bind (,))) ps) : go bs + go (NonRec v e : bs) = go_bind NonRec v e : go bs go [] = [] + + go_bind c v e + | occNameString (getOccName v) == "testBinding" + , exprType e `eqType` intTy + = c v (mkUncheckedIntExpr $ bool 222222 111111 early) + | otherwise + = c v e ===================================== testsuite/tests/plugins/simple-plugin/Simple/ReplacePlugin.hs ===================================== @@ -51,5 +51,6 @@ fixGuts rep guts = pure $ guts { mg_binds = fmap fix_bind (mg_binds guts) } Tick t e -> Tick t (fix_expr e) Type t -> Type t Coercion c -> Coercion c + Let b body -> Let (fix_bind b) (fix_expr body) fix_alt (Alt c bs e) = Alt c bs (fix_expr e) View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/50f6be09df4822b4729bc8ceb2de8ff… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/50f6be09df4822b4729bc8ceb2de8ff… You're receiving this email because of your account on gitlab.haskell.org.

1 0

[Git][ghc/ghc][master] NCG/LA64: Implement MO_BSwap and MO_BRev with bit-manipulation Instructions
by Marge Bot (＠marge-bot) 24 Sep '25

24 Sep '25

Marge Bot pushed to branch master at Glasgow Haskell Compiler / GHC Commits: 89e8ff3d by Peng Fan at 2025-09-23T20:42:37-04:00 NCG/LA64: Implement MO_BSwap and MO_BRev with bit-manipulation Instructions - - - - - 3 changed files: - compiler/GHC/CmmToAsm/LA64/CodeGen.hs - compiler/GHC/CmmToAsm/LA64/Instr.hs - compiler/GHC/CmmToAsm/LA64/Ppr.hs Changes: ===================================== compiler/GHC/CmmToAsm/LA64/CodeGen.hs ===================================== @@ -1805,6 +1805,49 @@ genCCall target dest_regs arg_regs = do where shift = (widthToInt w) + PrimTarget (MO_BSwap w) + | w `elem` [W16, W32, W64], + [arg_reg] <- arg_regs, + [dest_reg] <- dest_regs -> do + platform <- getPlatform + (reg_x, _, code_x) <- getSomeReg arg_reg + let dst_reg = getRegisterReg platform (CmmLocal dest_reg) + case w of + W64 -> return ( code_x `appOL` toOL + [ + REVBD (OpReg w dst_reg) (OpReg w reg_x) + ]) + W32 -> return ( code_x `appOL` toOL + [ + REVB2W (OpReg w dst_reg) (OpReg w reg_x) + ]) + _ -> return ( code_x `appOL` toOL + [ + REVB2H (OpReg w dst_reg) (OpReg w reg_x) + ]) + | otherwise -> unsupported (MO_BSwap w) + + PrimTarget (MO_BRev w) + | w `elem` [W8, W16, W32, W64], + [arg_reg] <- arg_regs, + [dest_reg] <- dest_regs -> do + platform <- getPlatform + (reg_x, _, code_x) <- getSomeReg arg_reg + let dst_reg = getRegisterReg platform (CmmLocal dest_reg) + case w of + W8 -> return ( code_x `appOL` toOL + [ + BITREV4B (OpReg W32 reg_x) (OpReg W32 reg_x), + AND (OpReg W64 dst_reg) (OpReg W64 reg_x) (OpImm (ImmInt 255)) + ]) + W16 -> return ( code_x `appOL` toOL + [ + BITREV (OpReg W64 reg_x) (OpReg W64 reg_x), + SRL (OpReg W64 dst_reg) (OpReg W64 reg_x) (OpImm (ImmInt 48)) + ]) + _ -> return ( code_x `snocOL` BITREV (OpReg w dst_reg) (OpReg w reg_x)) + | otherwise -> unsupported (MO_BRev w) + -- mop :: CallishMachOp (see GHC.Cmm.MachOp) PrimTarget mop -> do -- We'll need config to construct forien targets @@ -1939,8 +1982,6 @@ genCCall target dest_regs arg_regs = do MO_PopCnt w -> mkCCall (popCntLabel w) MO_Pdep w -> mkCCall (pdepLabel w) MO_Pext w -> mkCCall (pextLabel w) - MO_BSwap w -> mkCCall (bSwapLabel w) - MO_BRev w -> mkCCall (bRevLabel w) -- or a possibly side-effecting machine operation mo@(MO_AtomicRead w ord) ===================================== compiler/GHC/CmmToAsm/LA64/Instr.hs ===================================== @@ -126,8 +126,7 @@ regUsageOfInstr platform instr = case instr of REVHD dst src1 -> usage (regOp src1, regOp dst) BITREV4B dst src1 -> usage (regOp src1, regOp dst) BITREV8B dst src1 -> usage (regOp src1, regOp dst) - BITREVW dst src1 -> usage (regOp src1, regOp dst) - BITREVD dst src1 -> usage (regOp src1, regOp dst) + BITREV dst src1 -> usage (regOp src1, regOp dst) BSTRINS _ dst src1 src2 src3 -> usage (regOp src1 ++ regOp src2 ++ regOp src3, regOp dst) BSTRPICK _ dst src1 src2 src3 -> usage (regOp src1 ++ regOp src2 ++ regOp src3, regOp dst) MASKEQZ dst src1 src2 -> usage (regOp src1 ++ regOp src2, regOp dst) @@ -309,8 +308,7 @@ patchRegsOfInstr instr env = case instr of REVHD o1 o2 -> REVHD (patchOp o1) (patchOp o2) BITREV4B o1 o2 -> BITREV4B (patchOp o1) (patchOp o2) BITREV8B o1 o2 -> BITREV8B (patchOp o1) (patchOp o2) - BITREVW o1 o2 -> BITREVW (patchOp o1) (patchOp o2) - BITREVD o1 o2 -> BITREVD (patchOp o1) (patchOp o2) + BITREV o1 o2 -> BITREV (patchOp o1) (patchOp o2) BSTRINS f o1 o2 o3 o4 -> BSTRINS f (patchOp o1) (patchOp o2) (patchOp o3) (patchOp o4) BSTRPICK f o1 o2 o3 o4 -> BSTRPICK f (patchOp o1) (patchOp o2) (patchOp o3) (patchOp o4) MASKEQZ o1 o2 o3 -> MASKEQZ (patchOp o1) (patchOp o2) (patchOp o3) @@ -700,8 +698,7 @@ data Instr | REVHD Operand Operand | BITREV4B Operand Operand | BITREV8B Operand Operand - | BITREVW Operand Operand - | BITREVD Operand Operand + | BITREV Operand Operand | BSTRINS Format Operand Operand Operand Operand | BSTRPICK Format Operand Operand Operand Operand | MASKEQZ Operand Operand Operand @@ -824,8 +821,7 @@ instrCon i = REVHD{} -> "REVHD" BITREV4B{} -> "BITREV4B" BITREV8B{} -> "BITREV8B" - BITREVW{} -> "BITREVW" - BITREVD{} -> "BITREVD" + BITREV{} -> "BITREV" BSTRINS{} -> "BSTRINS" BSTRPICK{} -> "BSTRPICK" MASKEQZ{} -> "MASKEQZ" ===================================== compiler/GHC/CmmToAsm/LA64/Ppr.hs ===================================== @@ -802,8 +802,9 @@ pprInstr platform instr = case instr of -- BITREV.{W/D} BITREV4B o1 o2 -> op2 (text "\tbitrev.4b") o1 o2 BITREV8B o1 o2 -> op2 (text "\tbitrev.8b") o1 o2 - BITREVW o1 o2 -> op2 (text "\tbitrev.w") o1 o2 - BITREVD o1 o2 -> op2 (text "\tbitrev.d") o1 o2 + BITREV o1 o2 + | OpReg W32 _ <- o2 -> op2 (text "\tbitrev.w") o1 o2 + | OpReg W64 _ <- o2 -> op2 (text "\tbitrev.d") o1 o2 -- BSTRINS.{W/D} BSTRINS II64 o1 o2 o3 o4 -> op4 (text "\tbstrins.d") o1 o2 o3 o4 BSTRINS II32 o1 o2 o3 o4 -> op4 (text "\tbstrins.w") o1 o2 o3 o4 View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/89e8ff3d12689d73481cdb68d7408fd… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/89e8ff3d12689d73481cdb68d7408fd… You're receiving this email because of your account on gitlab.haskell.org.

1 0

[Git][ghc/ghc][master] Correcting LLVM linking of Intel BMI intrinsics pdep{8,16} and pext{8,16}.
by Marge Bot (＠marge-bot) 24 Sep '25

24 Sep '25

Marge Bot pushed to branch master at Glasgow Haskell Compiler / GHC Commits: 0a601c30 by Alex Washburn at 2025-09-23T20:41:41-04:00 Correcting LLVM linking of Intel BMI intrinsics pdep{8,16} and pext{8,16}. This patch fixes #26065. The LLVM interface does not expose bindings to: - llvm.x86.bmi.pdep.8 - llvm.x86.bmi.pdep.16 - llvm.x86.bmi.pext.8 - llvm.x86.bmi.pext.16 So calls are instead made to llvm.x86.bmi.{pdep,pext}.32 in these cases, with pre/post-operation truncation to constrain the logical value range. - - - - - 4 changed files: - compiler/GHC/CmmToLlvm/CodeGen.hs - + testsuite/tests/llvm/should_run/T26065.hs - + testsuite/tests/llvm/should_run/T26065.stdout - testsuite/tests/llvm/should_run/all.T Changes: ===================================== compiler/GHC/CmmToLlvm/CodeGen.hs ===================================== @@ -240,12 +240,25 @@ genCall (PrimTarget op@(MO_BRev w)) [dst] args = genCallSimpleCast w op dst args genCall (PrimTarget op@(MO_BSwap w)) [dst] args = genCallSimpleCast w op dst args -genCall (PrimTarget op@(MO_Pdep w)) [dst] args = - genCallSimpleCast w op dst args -genCall (PrimTarget op@(MO_Pext w)) [dst] args = - genCallSimpleCast w op dst args genCall (PrimTarget op@(MO_PopCnt w)) [dst] args = genCallSimpleCast w op dst args +{- Note [LLVM PDep/PExt intrinsics] +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +Since x86 PDep/PExt instructions only exist for 32/64 bit widths +we use the 32bit variant to compute the 8/16bit primops. +To do so we extend/truncate the argument/result around the +call. +-} +genCall (PrimTarget op@(MO_Pdep w)) [dst] args = do + cfg <- getConfig + if llvmCgBmiVersion cfg >= Just BMI2 + then genCallMinimumTruncationCast W32 w op dst args + else genCallSimpleCast w op dst args +genCall (PrimTarget op@(MO_Pext w)) [dst] args = do + cfg <- getConfig + if llvmCgBmiVersion cfg >= Just BMI2 + then genCallMinimumTruncationCast W32 w op dst args + else genCallSimpleCast w op dst args genCall (PrimTarget (MO_AtomicRMW width amop)) [dst] [addr, n] = runStmtsDecls $ do addrVar <- exprToVarW addr @@ -641,8 +654,15 @@ genCallExtract _ _ _ _ = -- from i32 to i8 explicitly as LLVM is strict about types. genCallSimpleCast :: Width -> CallishMachOp -> CmmFormal -> [CmmActual] -> LlvmM StmtData -genCallSimpleCast specW op dst args = do - let width = widthToLlvmInt specW +genCallSimpleCast w = genCallMinimumTruncationCast w w + +-- Given the minimum machine bit-width to use and the logical bit-width of the +-- value range, perform a type-cast truncation and extension before and after the +-- specified operation, respectively. +genCallMinimumTruncationCast :: Width -> Width -> CallishMachOp -> CmmFormal + -> [CmmActual] -> LlvmM StmtData +genCallMinimumTruncationCast minW specW op dst args = do + let width = widthToLlvmInt $ max minW specW argsW = const width <$> args dstType = cmmToLlvmType $ localRegType dst signage = cmmPrimOpRetValSignage op @@ -945,9 +965,10 @@ cmmPrimOpFunctions mop = do W256 -> fsLit "llvm.cttz.i256" W512 -> fsLit "llvm.cttz.i512" MO_Pdep w + -- See Note [LLVM PDep/PExt intrinsics] | isBmi2Enabled -> case w of - W8 -> fsLit "llvm.x86.bmi.pdep.8" - W16 -> fsLit "llvm.x86.bmi.pdep.16" + W8 -> fsLit "llvm.x86.bmi.pdep.32" + W16 -> fsLit "llvm.x86.bmi.pdep.32" W32 -> fsLit "llvm.x86.bmi.pdep.32" W64 -> fsLit "llvm.x86.bmi.pdep.64" W128 -> fsLit "llvm.x86.bmi.pdep.128" @@ -963,8 +984,9 @@ cmmPrimOpFunctions mop = do W512 -> fsLit "hs_pdep512" MO_Pext w | isBmi2Enabled -> case w of - W8 -> fsLit "llvm.x86.bmi.pext.8" - W16 -> fsLit "llvm.x86.bmi.pext.16" + -- See Note [LLVM PDep/PExt intrinsics] + W8 -> fsLit "llvm.x86.bmi.pext.32" + W16 -> fsLit "llvm.x86.bmi.pext.32" W32 -> fsLit "llvm.x86.bmi.pext.32" W64 -> fsLit "llvm.x86.bmi.pext.64" W128 -> fsLit "llvm.x86.bmi.pext.128" ===================================== testsuite/tests/llvm/should_run/T26065.hs ===================================== @@ -0,0 +1,68 @@ +{-# LANGUAGE MagicHash #-} +{-# LANGUAGE ScopedTypeVariables #-} +import Data.Char (toUpper) +import GHC.Exts +import GHC.Word +import Numeric (showHex) + +pdep8 :: Word8 -> Word8 -> Word8 +pdep8 (W8# a) (W8# b) = W8# (wordToWord8# (pdep8# (word8ToWord# a) (word8ToWord# b))) +{-# NOINLINE pdep8 #-} + +pdep16 :: Word16 -> Word16 -> Word16 +pdep16 (W16# a) (W16# b) = W16# (wordToWord16# (pdep16# (word16ToWord# a) (word16ToWord# b))) +{-# NOINLINE pdep16 #-} + +pdep32 :: Word32 -> Word32 -> Word32 +pdep32 (W32# a) (W32# b) = W32# (wordToWord32# (pdep32# (word32ToWord# a) (word32ToWord# b))) +{-# NOINLINE pdep32 #-} + +pdep64 :: Word64 -> Word64 -> Word64 +pdep64 (W64# a) (W64# b) = W64# (pdep64# a b) +{-# NOINLINE pdep64 #-} + +pext8 :: Word8 -> Word8 -> Word8 +pext8 (W8# a) (W8# b) = W8# (wordToWord8# (pext8# (word8ToWord# a) (word8ToWord# b))) +{-# NOINLINE pext8 #-} + +pext16 :: Word16 -> Word16 -> Word16 +pext16 (W16# a) (W16# b) = W16# (wordToWord16# (pext16# (word16ToWord# a) (word16ToWord# b))) +{-# NOINLINE pext16 #-} + +pext32 :: Word32 -> Word32 -> Word32 +pext32 (W32# a) (W32# b) = W32# (wordToWord32# (pext32# (word32ToWord# a) (word32ToWord# b))) +{-# NOINLINE pext32 #-} + +pext64 :: Word64 -> Word64 -> Word64 +pext64 (W64# a) (W64# b) = W64# (pext64# a b) +{-# NOINLINE pext64 #-} + +valueSource :: Integral i => i +valueSource = fromInteger 0xA7F7A7F7A7F7A7F7 + +valueMask :: Integral i => i +valueMask = fromInteger 0x5555555555555555 + +printIntrinsicCall :: forall i. Integral i => String -> (i -> i -> i) -> IO () +printIntrinsicCall label f = + let op1 = valueSource + op2 = valueMask + pad s = + let hex :: Integral a => a -> String + hex = flip showHex "" + str = toUpper <$> hex s + len = length $ hex (maxBound :: Word64) + n = length str + in "0x" <> replicate (len - n) '0' <> str + in putStrLn $ unwords [ label, pad op1, pad op2, "=", pad (f op1 op2) ] + +main :: IO () +main = do + printIntrinsicCall "pdep8 " pdep8 + printIntrinsicCall "pdep16" pdep16 + printIntrinsicCall "pdep32" pdep32 + printIntrinsicCall "pdep64" pdep64 + printIntrinsicCall "pext8 " pext8 + printIntrinsicCall "pext16" pext16 + printIntrinsicCall "pext32" pext32 + printIntrinsicCall "pext64" pext64 ===================================== testsuite/tests/llvm/should_run/T26065.stdout ===================================== @@ -0,0 +1,8 @@ +pdep8 0x00000000000000F7 0x0000000000000055 = 0x0000000000000015 +pdep16 0x000000000000A7F7 0x0000000000005555 = 0x0000000000005515 +pdep32 0x00000000A7F7A7F7 0x0000000055555555 = 0x0000000044155515 +pdep64 0xA7F7A7F7A7F7A7F7 0x5555555555555555 = 0x4415551544155515 +pext8 0x00000000000000F7 0x0000000000000055 = 0x000000000000000F +pext16 0x000000000000A7F7 0x0000000000005555 = 0x000000000000003F +pext32 0x00000000A7F7A7F7 0x0000000055555555 = 0x0000000000003F3F +pext64 0xA7F7A7F7A7F7A7F7 0x5555555555555555 = 0x000000003F3F3F3F ===================================== testsuite/tests/llvm/should_run/all.T ===================================== @@ -18,3 +18,8 @@ test('T22033', [normal, normalise_errmsg_fun(ignore_llvm_and_vortex)], compile_a test('T25730', [req_c, unless(arch('x86_64'), skip), normalise_errmsg_fun(ignore_llvm_and_vortex)], compile_and_run, ['T25730C.c']) # T25730C.c contains Intel instrinsics, so only run this test on x86 test('T20645', [normal, normalise_errmsg_fun(ignore_llvm_and_vortex), when(have_llvm(), extra_ways(["optllvm"]))], compile_and_run, ['']) +# T26065.c tests LLVM linking of Intel instrinsics, so only run this test on x86 +test('T26065', [normal, normalise_errmsg_fun(ignore_llvm_and_vortex), when(have_llvm(), extra_ways(["optllvm"])), + unless((arch('x86_64') or arch('i386')) and have_cpu_feature('bmi2'),skip)], + compile_and_run, ['-mbmi2']) + View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/0a601c30d5e885d9b15b202f1fce55c… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/0a601c30d5e885d9b15b202f1fce55c… You're receiving this email because of your account on gitlab.haskell.org.

1 0

[Git][ghc/ghc][master] 2 commits: RV64: Fix: Add missing truncation to MO_S_Shr (#26248)
by Marge Bot (＠marge-bot) 24 Sep '25

24 Sep '25

Marge Bot pushed to branch master at Glasgow Haskell Compiler / GHC Commits: c4d32493 by Sven Tennie at 2025-09-23T20:40:57-04:00 RV64: Fix: Add missing truncation to MO_S_Shr (#26248) Sub-double word (<W64) registers need to be truncated after the operation. - - - - - 41dce477 by Sven Tennie at 2025-09-23T20:40:57-04:00 RV64: Cleanup shift emitting cases/code Remove overlapping cases to make the shift logic easier to understand. - - - - - 1 changed file: - compiler/GHC/CmmToAsm/RV64/CodeGen.hs Changes: ===================================== compiler/GHC/CmmToAsm/RV64/CodeGen.hs ===================================== @@ -874,46 +874,18 @@ getRegister' config plat expr = ) -- 2. Shifts. x << n, x >> n. - CmmMachOp (MO_Shl w) [x, CmmLit (CmmInt n _)] - | w == W32, - 0 <= n, - n < 32 -> do - (reg_x, _format_x, code_x) <- getSomeReg x - return - $ Any - (intFormat w) - ( \dst -> - code_x - `snocOL` annExpr expr (SLL (OpReg w dst) (OpReg w reg_x) (OpImm (ImmInteger n))) - `appOL` truncateReg w w dst - ) - CmmMachOp (MO_Shl w) [x, CmmLit (CmmInt n _)] - | w == W64, - 0 <= n, - n < 64 -> do - (reg_x, _format_x, code_x) <- getSomeReg x - return - $ Any - (intFormat w) - ( \dst -> - code_x - `snocOL` annExpr expr (SLL (OpReg w dst) (OpReg w reg_x) (OpImm (ImmInteger n))) - `appOL` truncateReg w w dst - ) - CmmMachOp (MO_S_Shr w) [x, CmmLit (CmmInt n _)] | fitsIn12bitImm n -> do - (reg_x, format_x, code_x) <- getSomeReg x - (reg_x', code_x') <- signExtendReg (formatToWidth format_x) w reg_x + CmmMachOp (MO_Shl w) [x, CmmLit (CmmInt n _)] | fitsIn12bitImm n -> do + (reg_x, _format_x, code_x) <- getSomeReg x return $ Any (intFormat w) ( \dst -> code_x - `appOL` code_x' - `snocOL` annExpr expr (SRA (OpReg w dst) (OpReg w reg_x') (OpImm (ImmInteger n))) + `snocOL` annExpr expr (SLL (OpReg w dst) (OpReg w reg_x) (OpImm (ImmInteger n))) + `appOL` truncateReg w w dst ) - CmmMachOp (MO_S_Shr w) [x, y] -> do + CmmMachOp (MO_S_Shr w) [x, CmmLit (CmmInt n _)] | fitsIn12bitImm n -> do (reg_x, format_x, code_x) <- getSomeReg x - (reg_y, _format_y, code_y) <- getSomeReg y (reg_x', code_x') <- signExtendReg (formatToWidth format_x) w reg_x return $ Any @@ -921,72 +893,20 @@ getRegister' config plat expr = ( \dst -> code_x `appOL` code_x' - `appOL` code_y - `snocOL` annExpr expr (SRA (OpReg w dst) (OpReg w reg_x') (OpReg w reg_y)) + `snocOL` annExpr expr (SRA (OpReg w dst) (OpReg w reg_x') (OpImm (ImmInteger n))) + `appOL` truncateReg w w dst ) - CmmMachOp (MO_U_Shr w) [x, CmmLit (CmmInt n _)] - | w == W8, - 0 <= n, - n < 8 -> do - (reg_x, format_x, code_x) <- getSomeReg x - return - $ Any - (intFormat w) - ( \dst -> - code_x - `appOL` truncateReg (formatToWidth format_x) w reg_x - `snocOL` annExpr expr (SRL (OpReg w dst) (OpReg w reg_x) (OpImm (ImmInteger n))) - ) - CmmMachOp (MO_U_Shr w) [x, CmmLit (CmmInt n _)] - | w == W16, - 0 <= n, - n < 16 -> do - (reg_x, format_x, code_x) <- getSomeReg x - return - $ Any - (intFormat w) - ( \dst -> - code_x - `appOL` truncateReg (formatToWidth format_x) w reg_x - `snocOL` annExpr expr (SRL (OpReg w dst) (OpReg w reg_x) (OpImm (ImmInteger n))) - ) - CmmMachOp (MO_U_Shr w) [x, y] | w == W8 || w == W16 -> do + CmmMachOp (MO_U_Shr w) [x, CmmLit (CmmInt n _)] | fitsIn12bitImm n -> do (reg_x, format_x, code_x) <- getSomeReg x - (reg_y, _format_y, code_y) <- getSomeReg y return $ Any (intFormat w) ( \dst -> code_x - `appOL` code_y `appOL` truncateReg (formatToWidth format_x) w reg_x - `snocOL` annExpr expr (SRL (OpReg w dst) (OpReg w reg_x) (OpReg w reg_y)) + `snocOL` annExpr expr (SRL (OpReg w dst) (OpReg w reg_x) (OpImm (ImmInteger n))) + `appOL` truncateReg w w dst ) - CmmMachOp (MO_U_Shr w) [x, CmmLit (CmmInt n _)] - | w == W32, - 0 <= n, - n < 32 -> do - (reg_x, _format_x, code_x) <- getSomeReg x - return - $ Any - (intFormat w) - ( \dst -> - code_x - `snocOL` annExpr expr (SRL (OpReg w dst) (OpReg w reg_x) (OpImm (ImmInteger n))) - ) - CmmMachOp (MO_U_Shr w) [x, CmmLit (CmmInt n _)] - | w == W64, - 0 <= n, - n < 64 -> do - (reg_x, _format_x, code_x) <- getSomeReg x - return - $ Any - (intFormat w) - ( \dst -> - code_x - `snocOL` annExpr expr (SRL (OpReg w dst) (OpReg w reg_x) (OpImm (ImmInteger n))) - ) - -- 3. Logic &&, || CmmMachOp (MO_And w) [CmmReg reg, CmmLit (CmmInt n _)] | fitsIn12bitImm n -> View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/67de53a6ced23caad640d2c7421089… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/67de53a6ced23caad640d2c7421089… You're receiving this email because of your account on gitlab.haskell.org.

1 0

[Git][ghc/ghc][wip/marge_bot_batch_merge_job] 12 commits: Improve pretty printer for HsExpr
by Marge Bot (＠marge-bot) 23 Sep '25

23 Sep '25

Marge Bot pushed to branch wip/marge_bot_batch_merge_job at Glasgow Haskell Compiler / GHC Commits: e7df6cc0 by Simon Peyton Jones at 2025-09-23T14:34:39-04:00 Improve pretty printer for HsExpr Given a very deeply-nested application, it just kept printing deeper and deeper. This small change makes it cut off. Test is in #26330, but we also get a dramatic decrease in compile time for perf/compiler/InstanceMatching: InstanceMatching 4,086,884,584 1,181,767,232 -71.1% GOOD Why? Because before we got a GIGANTIC error message that took ages to pretty-print; now we get this much more civilised message (I have removed some whitespace.) Match.hs:1007:1: error: • No instance for ‘Show (F001 a)’ arising from a use of ‘showsPrec’ • In the second argument of ‘showString’, namely ‘(showsPrec 11 b1 (GHC.Internal.Show.showSpace (showsPrec 11 b2 (GHC.Internal.Show.showSpace (showsPrec 11 b3 (GHC.Internal.Show.showSpace (showsPrec 11 b4 (GHC.Internal.Show.showSpace (showsPrec 11 b5 (GHC.Internal.Show.showSpace (showsPrec 11 b6 (GHC.Internal.Show.showSpace (showsPrec ...)))))))))))))’ ----------------------- The main payload is * At the start of `pprExpr` * In the defn of `pprApp` A little bit of refactoring: * It turned out that we were setting the default cut-off depth to a fixed value in two places, so changing one didn't change the other. See defaultSDocDepth and defaultSDocCols * I refactored `pprDeeperList` a bit so I could understand it better. Because the depth calculation has changed, there are lots of small error message wibbles. Metric Decrease: InstanceMatching - - - - - 209f0158 by Simon Peyton Jones at 2025-09-23T14:34:39-04:00 Use Outputable.ellipsis rather than text "..." - - - - - 64bb0e37 by Sylvain Henry at 2025-09-23T14:35:56-04:00 deriveConstants: automatically pass -fcommon CC flag (#26393) By mistake we tried to use deriveConstants without passing `--gcc-flag -fcommon` (which Hadrian does) and it failed. This patch: 1. adds parsing support for constants stored in the .bss section (i.e. when -fcommon isn't passed) 2. enables passing `-fcommon` automatically to the C compiler because Windows requires this for subtle reasons 3. Documents the subtle reasons (1) isn't strictly necessary because we always do (2) but it does no harm and it is still useful if the CC flags ever contain -fno-common - - - - - afcdf92f by Oleg Grenrus at 2025-09-23T14:36:41-04:00 Don't wrap spaces in <span>s Doing similar comparison as in 63189b2ceca07edf4e179f4180ca60d470c62cb3 With this change the gzipped documentation is now 2% smaller (previously 1%) 12_694_206 Agda-2.9.0-docs-orig.tar.gz 12_436_829 Agda-2.9.0-docs.tar.gz Unzipped docs are 5% smaller (previously 3%) 178M Agda-2.9.0-docs-orig 169M Agda-2.9.0-docs Individual hyperlinked sources are around 7-10% smaller (previously 5%) (`Parser` module is generated by happy and has relatively little whitespace) 14_230_117 Agda.Syntax.Parser.Parser.html 13_220_758 Agda.Syntax.Parser.Parser.html Agda's hyperlinked sources are 9% smaller now: 121M Agda-2.9.0-docs-orig/src 110M Agda-2.9.0-docs/src - - - - - 67de53a6 by Cheng Shao at 2025-09-23T14:37:31-04:00 rts: remove obsolete __GNUC__ related logic This patch removes obsolete `__GNUC__` related logic, given on any currently supported platform and toolchain, `__GNUC__ >= 4` is universally true. Also pulls some other weeds and most notably, use `__builtin___clear_cache` for clang as well, since clang has supported this gcc intrinsic since 2014, see https://github.com/llvm/llvm-project/commit/c491a8d4577052bc6b3b4c72a7db6a7…. - - - - - 39365c97 by Sven Tennie at 2025-09-23T15:10:07-04:00 RV64: Fix: Add missing truncation to MO_S_Shr (#26248) Sub-double word (<W64) registers need to be truncated after the operation. - - - - - 648eb2ec by Sven Tennie at 2025-09-23T15:10:07-04:00 RV64: Cleanup shift emitting cases/code Remove overlapping cases to make the shift logic easier to understand. - - - - - 1a4982bb by Alex Washburn at 2025-09-23T15:10:08-04:00 Correcting LLVM linking of Intel BMI intrinsics pdep{8,16} and pext{8,16}. This patch fixes #26065. The LLVM interface does not expose bindings to: - llvm.x86.bmi.pdep.8 - llvm.x86.bmi.pdep.16 - llvm.x86.bmi.pext.8 - llvm.x86.bmi.pext.16 So calls are instead made to llvm.x86.bmi.{pdep,pext}.32 in these cases, with pre/post-operation truncation to constrain the logical value range. - - - - - 19257e0d by Peng Fan at 2025-09-23T15:10:22-04:00 NCG/LA64: Implement MO_BSwap and MO_BRev with bit-manipulation Instructions - - - - - 05cdcc1e by Sylvain Henry at 2025-09-23T15:10:31-04:00 Allow Core plugins to access unoptimized Core (#23337) Make the first simple optimization pass after desugaring a real CoreToDo pass. This allows CorePlugins to decide whether they want to be executed before or after this pass. - - - - - 03e71c3f by Simon Hengel at 2025-09-23T15:10:32-04:00 docs: Fix typo in scoped_type_variables.rst - - - - - 7668b4a8 by Cheng Shao at 2025-09-23T15:10:32-04:00 ghci: fix bootstrapping with 9.12.3-rc1 and above This patch fixes bootstrapping GHC with 9.12.3-rc1 and above. ghci defines `Binary` instance for `HalfWord` in `ghc-heap`, which is a proper `newtype` in 9.14 and starting from 9.12.3. Given we don't build `ghc-heap` in stage0, we need to fix this predicate so that it corresponds to the boot ghc versions that contain the right version of `ghc-heap`. - - - - - 96 changed files: - compiler/GHC/CmmToAsm/LA64/CodeGen.hs - compiler/GHC/CmmToAsm/LA64/Instr.hs - compiler/GHC/CmmToAsm/LA64/Ppr.hs - compiler/GHC/CmmToAsm/RV64/CodeGen.hs - compiler/GHC/CmmToLlvm/CodeGen.hs - compiler/GHC/Core/Opt/Pipeline.hs - compiler/GHC/Core/Opt/Pipeline/Types.hs - compiler/GHC/Core/Opt/Simplify/Env.hs - compiler/GHC/Core/Ppr.hs - compiler/GHC/Driver/DynFlags.hs - compiler/GHC/Hs/Expr.hs - compiler/GHC/HsToCore.hs - compiler/GHC/HsToCore/Errors/Ppr.hs - compiler/GHC/HsToCore/Pmc/Ppr.hs - compiler/GHC/HsToCore/Pmc/Types.hs - compiler/GHC/Iface/Syntax.hs - compiler/GHC/Llvm/Ppr.hs - compiler/GHC/Llvm/Types.hs - compiler/GHC/Runtime/Heap/Inspect.hs - compiler/GHC/Runtime/Loader.hs - compiler/GHC/Tc/Errors/Ppr.hs - compiler/GHC/Types/Hint/Ppr.hs - compiler/GHC/Utils/Outputable.hs - compiler/GHC/Utils/Ppr.hs - docs/users_guide/exts/scoped_type_variables.rst - hadrian/src/Settings/Builders/DeriveConstants.hs - libraries/ghc-internal/cbits/atomic.c - libraries/ghc-internal/cbits/ctz.c - libraries/ghci/GHCi/Message.hs - rts/RtsStartup.c - rts/RtsSymbols.c - rts/include/Rts.h - rts/include/Stg.h - rts/include/rts/Types.h - rts/sm/BlockAlloc.c - rts/sm/Evac.h - rts/sm/Storage.c - testsuite/tests/arrows/gadt/T17423.stderr - testsuite/tests/indexed-types/should_compile/PushedInAsGivens.stderr - testsuite/tests/indexed-types/should_fail/T26176.stderr - testsuite/tests/indexed-types/should_fail/T2693.stderr - testsuite/tests/indexed-types/should_fail/T4093b.stderr - testsuite/tests/indexed-types/should_fail/T8518.stderr - testsuite/tests/indexed-types/should_fail/T9662.stderr - testsuite/tests/linear/should_fail/Linear17.stderr - testsuite/tests/linear/should_fail/LinearLet7.stderr - + testsuite/tests/llvm/should_run/T26065.hs - + testsuite/tests/llvm/should_run/T26065.stdout - testsuite/tests/llvm/should_run/all.T - testsuite/tests/parser/should_fail/RecordDotSyntaxFail10.stderr - testsuite/tests/parser/should_fail/RecordDotSyntaxFail11.stderr - testsuite/tests/parser/should_fail/RecordDotSyntaxFail13.stderr - testsuite/tests/parser/should_fail/RecordDotSyntaxFail8.stderr - testsuite/tests/partial-sigs/should_compile/T21719.stderr - testsuite/tests/plugins/annotation-plugin/SayAnnNames.hs - testsuite/tests/plugins/late-plugin/LatePlugin.hs - testsuite/tests/plugins/simple-plugin/Simple/ReplacePlugin.hs - testsuite/tests/polykinds/T13393.stderr - testsuite/tests/rep-poly/T12709.stderr - testsuite/tests/simplCore/should_compile/simpl017.stderr - testsuite/tests/th/T10945.stderr - testsuite/tests/th/TH_StaticPointers02.stderr - testsuite/tests/typecheck/should_compile/T11339.stderr - testsuite/tests/typecheck/should_fail/DoExpansion3.stderr - testsuite/tests/typecheck/should_fail/T12177.stderr - testsuite/tests/typecheck/should_fail/T22707.stderr - testsuite/tests/typecheck/should_fail/T24064.stderr - + testsuite/tests/typecheck/should_fail/T26330.hs - + testsuite/tests/typecheck/should_fail/T26330.stderr - testsuite/tests/typecheck/should_fail/T8142.stderr - testsuite/tests/typecheck/should_fail/T8603.stderr - testsuite/tests/typecheck/should_fail/all.T - testsuite/tests/typecheck/should_fail/tcfail128.stderr - testsuite/tests/typecheck/should_fail/tcfail153.stderr - testsuite/tests/typecheck/should_fail/tcfail168.stderr - testsuite/tests/typecheck/should_fail/tcfail177.stderr - testsuite/tests/typecheck/should_fail/tcfail185.stderr - testsuite/tests/typecheck/should_run/Typeable1.stderr - utils/deriveConstants/Main.hs - utils/haddock/haddock-api/src/Haddock/Backends/Hyperlinker/Renderer.hs - utils/haddock/hypsrc-test/ref/src/Bug1091.html - utils/haddock/hypsrc-test/ref/src/CPP.html - utils/haddock/hypsrc-test/ref/src/Classes.html - utils/haddock/hypsrc-test/ref/src/Constructors.html - utils/haddock/hypsrc-test/ref/src/Identifiers.html - utils/haddock/hypsrc-test/ref/src/LinkingIdentifiers.html - utils/haddock/hypsrc-test/ref/src/Literals.html - utils/haddock/hypsrc-test/ref/src/Operators.html - utils/haddock/hypsrc-test/ref/src/Polymorphism.html - utils/haddock/hypsrc-test/ref/src/PositionPragmas.html - utils/haddock/hypsrc-test/ref/src/Quasiquoter.html - utils/haddock/hypsrc-test/ref/src/Records.html - utils/haddock/hypsrc-test/ref/src/TemplateHaskellQuasiquotes.html - utils/haddock/hypsrc-test/ref/src/TemplateHaskellSplices.html - utils/haddock/hypsrc-test/ref/src/Types.html - utils/haddock/hypsrc-test/ref/src/UsingQuasiquotes.html The diff was not included because it is too large. View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/401871c2a024fb42677f5048d60fa1… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/401871c2a024fb42677f5048d60fa1… You're receiving this email because of your account on gitlab.haskell.org.

1 0

[Git][ghc/ghc][master] rts: remove obsolete __GNUC__ related logic
by Marge Bot (＠marge-bot) 23 Sep '25

23 Sep '25

Marge Bot pushed to branch master at Glasgow Haskell Compiler / GHC Commits: 67de53a6 by Cheng Shao at 2025-09-23T14:37:31-04:00 rts: remove obsolete __GNUC__ related logic This patch removes obsolete `__GNUC__` related logic, given on any currently supported platform and toolchain, `__GNUC__ >= 4` is universally true. Also pulls some other weeds and most notably, use `__builtin___clear_cache` for clang as well, since clang has supported this gcc intrinsic since 2014, see https://github.com/llvm/llvm-project/commit/c491a8d4577052bc6b3b4c72a7db6a7…. - - - - - 10 changed files: - libraries/ghc-internal/cbits/atomic.c - libraries/ghc-internal/cbits/ctz.c - rts/RtsStartup.c - rts/RtsSymbols.c - rts/include/Rts.h - rts/include/Stg.h - rts/include/rts/Types.h - rts/sm/BlockAlloc.c - rts/sm/Evac.h - rts/sm/Storage.c Changes: ===================================== libraries/ghc-internal/cbits/atomic.c ===================================== @@ -163,7 +163,7 @@ hs_atomic_and64(StgWord x, StgWord64 val) #pragma GCC diagnostic push #if defined(__clang__) #pragma GCC diagnostic ignored "-Wsync-fetch-and-nand-semantics-changed" -#elif defined(__GNUC__) +#else #pragma GCC diagnostic ignored "-Wsync-nand" #endif ===================================== libraries/ghc-internal/cbits/ctz.c ===================================== @@ -31,7 +31,7 @@ hs_ctz32(StgWord x) StgWord hs_ctz64(StgWord64 x) { -#if defined(__GNUC__) && (defined(i386_HOST_ARCH) || defined(powerpc_HOST_ARCH)) +#if defined(i386_HOST_ARCH) || defined(powerpc_HOST_ARCH) /* On Linux/i386, the 64bit `__builtin_ctzll()` intrinsic doesn't get inlined by GCC but rather a short `__ctzdi2` runtime function is inserted when needed into compiled object files. ===================================== rts/RtsStartup.c ===================================== @@ -125,13 +125,7 @@ void _fpreset(void) x86_init_fpu(); } -#if defined(__GNUC__) void __attribute__((alias("_fpreset"))) fpreset(void); -#else -void fpreset(void) { - _fpreset(); -} -#endif /* Set the console's CodePage to UTF-8 if using the new I/O manager and the CP is still the default one. */ ===================================== rts/RtsSymbols.c ===================================== @@ -963,7 +963,7 @@ extern char **environ; RTS_INTCHAR_SYMBOLS // 64-bit support functions in libgcc.a -#if defined(__GNUC__) && SIZEOF_VOID_P <= 4 && !defined(_ABIN32) +#if SIZEOF_VOID_P <= 4 && !defined(_ABIN32) #define RTS_LIBGCC_SYMBOLS \ SymI_NeedsProto(__divdi3) \ SymI_NeedsProto(__udivdi3) \ @@ -974,7 +974,7 @@ extern char **environ; SymI_NeedsProto(__ashrdi3) \ SymI_NeedsProto(__lshrdi3) \ SymI_NeedsProto(__fixunsdfdi) -#elif defined(__GNUC__) && SIZEOF_VOID_P == 8 +#elif SIZEOF_VOID_P == 8 #define RTS_LIBGCC_SYMBOLS \ SymI_NeedsProto(__udivti3) \ SymI_NeedsProto(__umodti3) ===================================== rts/include/Rts.h ===================================== @@ -54,35 +54,19 @@ extern "C" { #include "rts/Types.h" #include "rts/Time.h" -#if __GNUC__ >= 3 #define ATTRIBUTE_ALIGNED(n) __attribute__((aligned(n))) -#else -#define ATTRIBUTE_ALIGNED(n) /*nothing*/ -#endif // Symbols that are extern, but private to the RTS, are declared // with visibility "hidden" to hide them outside the RTS shared // library. #define RTS_PRIVATE GNUC3_ATTRIBUTE(visibility("hidden")) -#if __GNUC__ >= 4 #define RTS_UNLIKELY(p) __builtin_expect((p),0) -#else -#define RTS_UNLIKELY(p) (p) -#endif -#if __GNUC__ >= 4 #define RTS_LIKELY(p) __builtin_expect(!!(p), 1) -#else -#define RTS_LIKELY(p) (p) -#endif /* __builtin_unreachable is supported since GNU C 4.5 */ -#if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 5) #define RTS_UNREACHABLE __builtin_unreachable() -#else -#define RTS_UNREACHABLE abort() -#endif /* Prefetch primitives */ #define prefetchForRead(ptr) __builtin_prefetch(ptr, 0) @@ -374,17 +358,8 @@ TICK_VAR(2) Useful macros and inline functions -------------------------------------------------------------------------- */ -#if defined(__GNUC__) -#define SUPPORTS_TYPEOF -#endif - -#if defined(SUPPORTS_TYPEOF) #define stg_min(a,b) ({typeof(a) _a = (a), _b = (b); _a <= _b ? _a : _b; }) #define stg_max(a,b) ({typeof(a) _a = (a), _b = (b); _a <= _b ? _b : _a; }) -#else -#define stg_min(a,b) ((a) <= (b) ? (a) : (b)) -#define stg_max(a,b) ((a) <= (b) ? (b) : (a)) -#endif /* -------------------------------------------------------------------------- */ ===================================== rts/include/Stg.h ===================================== @@ -131,20 +131,12 @@ /* * GCC attributes */ -#if defined(__GNUC__) #define GNU_ATTRIBUTE(at) __attribute__((at)) -#else -#define GNU_ATTRIBUTE(at) -#endif -#if __GNUC__ >= 3 #define GNUC3_ATTRIBUTE(at) __attribute__((at)) -#else -#define GNUC3_ATTRIBUTE(at) -#endif /* Used to mark a switch case that falls-through */ -#if (defined(__GNUC__) && __GNUC__ >= 7) +#if __GNUC__ >= 7 // N.B. Don't enable fallthrough annotations when compiling with Clang. // Apparently clang doesn't enable implicitly fallthrough warnings by default // http://llvm.org/viewvc/llvm-project?revision=167655&view=revision @@ -154,7 +146,7 @@ #define FALLTHROUGH ((void)0) #endif /* __GNUC__ >= 7 */ -#if !defined(DEBUG) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3)) +#if !defined(DEBUG) #define GNUC_ATTR_HOT __attribute__((hot)) #else #define GNUC_ATTR_HOT /* nothing */ @@ -168,21 +160,13 @@ See Note [Windows Stack allocations] */ #if defined(__clang__) #define STG_NO_OPTIMIZE __attribute__((optnone)) -#elif defined(__GNUC__) || defined(__GNUG__) -#define STG_NO_OPTIMIZE __attribute__((optimize("O0"))) #else -#define STG_NO_OPTIMIZE /* nothing */ +#define STG_NO_OPTIMIZE __attribute__((optimize("O0"))) #endif // Mark a function as accepting a printf-like format string. -#if !defined(__GNUC__) && defined(mingw32_HOST_OS) -/* On Win64, if we say "printf" then gcc thinks we are going to use - MS format specifiers like %I64d rather than %llu */ -#define STG_PRINTF_ATTR(fmt_arg, rest) GNUC3_ATTRIBUTE(format(gnu_printf, fmt_arg, rest)) -#else /* However, on OS X, "gnu_printf" isn't recognised */ #define STG_PRINTF_ATTR(fmt_arg, rest) GNUC3_ATTRIBUTE(format(printf, fmt_arg, rest)) -#endif #define STG_RESTRICT __restrict__ @@ -204,13 +188,9 @@ # define stg__has_attribute(attr) (0) #endif -#ifdef __GNUC__ -# define STG_GNUC_GUARD_VERSION(major, minor) \ +#define STG_GNUC_GUARD_VERSION(major, minor) \ ((__GNUC__ > (major)) || \ ((__GNUC__ == (major)) && (__GNUC_MINOR__ >= (minor)))) -#else -# define STG_GNUC_GUARD_VERSION(major, minor) (0) -#endif /* * The versions of the `__malloc__` attribute which take arguments are only ===================================== rts/include/rts/Types.h ===================================== @@ -16,16 +16,6 @@ #include <stddef.h> #include <stdbool.h> -// Deprecated, use uint32_t instead. -typedef unsigned int nat __attribute__((deprecated)); /* uint32_t */ - -/* ullong (64|128-bit) type: only include if needed (not ANSI) */ -#if defined(__GNUC__) -#define LL(x) (x##LL) -#else -#define LL(x) (x##L) -#endif - typedef struct StgClosure_ StgClosure; typedef struct StgInfoTable_ StgInfoTable; typedef struct StgTSO_ StgTSO; ===================================== rts/sm/BlockAlloc.c ===================================== @@ -274,19 +274,9 @@ STATIC_INLINE uint32_t log_2(W_ n) { ASSERT(n > 0 && n < (1<<NUM_FREE_LISTS)); -#if defined(__GNUC__) return CLZW(n) ^ (sizeof(StgWord)*8 - 1); // generates good code on x86. __builtin_clz() compiles to bsr+xor, but // we want just bsr, so the xor here cancels out gcc's xor. -#else - W_ i, x; - x = n; - for (i=0; i < NUM_FREE_LISTS; i++) { - x = x >> 1; - if (x == 0) return i; - } - return NUM_FREE_LISTS; -#endif } // log base 2 (ceiling), needs to support up to (2^NUM_FREE_LISTS)-1 @@ -294,18 +284,8 @@ STATIC_INLINE uint32_t log_2_ceil(W_ n) { ASSERT(n > 0 && n < (1<<NUM_FREE_LISTS)); -#if defined(__GNUC__) uint32_t r = log_2(n); return (n & (n-1)) ? r+1 : r; -#else - W_ i, x; - x = 1; - for (i=0; i < MAX_FREE_LIST; i++) { - if (x >= n) return i; - x = x << 1; - } - return MAX_FREE_LIST; -#endif } STATIC_INLINE void ===================================== rts/sm/Evac.h ===================================== @@ -25,7 +25,7 @@ // registers EAX, EDX, and ECX instead of on the stack. Functions that // take a variable number of arguments will continue to be passed all of // their arguments on the stack. -#if __GNUC__ >= 2 && (defined(x86_64_HOST_ARCH) || defined(i386_HOST_ARCH)) +#if defined(x86_64_HOST_ARCH) || defined(i386_HOST_ARCH) #define REGPARM1 __attribute__((regparm(1))) #else #define REGPARM1 ===================================== rts/sm/Storage.c ===================================== @@ -1818,25 +1818,6 @@ StgWord calcTotalCompactW (void) #include <libkern/OSCacheControl.h> #endif -/* __builtin___clear_cache is supported since GNU C 4.3.6. - * We pick 4.4 to simplify condition a bit. - */ -#define GCC_HAS_BUILTIN_CLEAR_CACHE (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 4)) - -#if defined(__clang__) -/* clang defines __clear_cache as a builtin on some platforms. - * For example on armv7-linux-androideabi. The type slightly - * differs from gcc. - */ -extern void __clear_cache(void * begin, void * end); -#elif defined(__GNUC__) && !GCC_HAS_BUILTIN_CLEAR_CACHE -/* __clear_cache is a libgcc function. - * It existed before __builtin___clear_cache was introduced. - * See #8562. - */ -extern void __clear_cache(char * begin, char * end); -#endif /* __GNUC__ */ - /* On ARM and other platforms, we need to flush the cache after writing code into memory, so the processor reliably sees it. */ void flushExec (W_ len, AdjustorExecutable exec_addr) @@ -1849,26 +1830,11 @@ void flushExec (W_ len, AdjustorExecutable exec_addr) /* On iOS we need to use the special 'sys_icache_invalidate' call. */ sys_icache_invalidate(exec_addr, len); #elif defined(wasm32_HOST_ARCH) -#elif defined(__clang__) - unsigned char* begin = (unsigned char*)exec_addr; - unsigned char* end = begin + len; -# if __has_builtin(__builtin___clear_cache) - __builtin___clear_cache((void*)begin, (void*)end); -# else - __clear_cache((void*)begin, (void*)end); -# endif -#elif defined(__GNUC__) - /* For all other platforms, fall back to a libgcc builtin. */ +#else + /* For all other platforms, fall back to __builtin___clear_cache. */ unsigned char* begin = (unsigned char*)exec_addr; unsigned char* end = begin + len; -# if GCC_HAS_BUILTIN_CLEAR_CACHE __builtin___clear_cache((void*)begin, (void*)end); -# else - /* For all other platforms, fall back to a libgcc builtin. */ - __clear_cache((void*)begin, (void*)end); -# endif -#else -#error Missing support to flush the instruction cache #endif } View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/67de53a6ced23caad640d2c74210892… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/67de53a6ced23caad640d2c74210892… You're receiving this email because of your account on gitlab.haskell.org.

1 0

[Git][ghc/ghc][master] Don't wrap spaces in <span>s
by Marge Bot (＠marge-bot) 23 Sep '25

23 Sep '25

Marge Bot pushed to branch master at Glasgow Haskell Compiler / GHC Commits: afcdf92f by Oleg Grenrus at 2025-09-23T14:36:41-04:00 Don't wrap spaces in <span>s Doing similar comparison as in 63189b2ceca07edf4e179f4180ca60d470c62cb3 With this change the gzipped documentation is now 2% smaller (previously 1%) 12_694_206 Agda-2.9.0-docs-orig.tar.gz 12_436_829 Agda-2.9.0-docs.tar.gz Unzipped docs are 5% smaller (previously 3%) 178M Agda-2.9.0-docs-orig 169M Agda-2.9.0-docs Individual hyperlinked sources are around 7-10% smaller (previously 5%) (`Parser` module is generated by happy and has relatively little whitespace) 14_230_117 Agda.Syntax.Parser.Parser.html 13_220_758 Agda.Syntax.Parser.Parser.html Agda's hyperlinked sources are 9% smaller now: 121M Agda-2.9.0-docs-orig/src 110M Agda-2.9.0-docs/src - - - - - 17 changed files: - utils/haddock/haddock-api/src/Haddock/Backends/Hyperlinker/Renderer.hs - utils/haddock/hypsrc-test/ref/src/Bug1091.html - utils/haddock/hypsrc-test/ref/src/CPP.html - utils/haddock/hypsrc-test/ref/src/Classes.html - utils/haddock/hypsrc-test/ref/src/Constructors.html - utils/haddock/hypsrc-test/ref/src/Identifiers.html - utils/haddock/hypsrc-test/ref/src/LinkingIdentifiers.html - utils/haddock/hypsrc-test/ref/src/Literals.html - utils/haddock/hypsrc-test/ref/src/Operators.html - utils/haddock/hypsrc-test/ref/src/Polymorphism.html - utils/haddock/hypsrc-test/ref/src/PositionPragmas.html - utils/haddock/hypsrc-test/ref/src/Quasiquoter.html - utils/haddock/hypsrc-test/ref/src/Records.html - utils/haddock/hypsrc-test/ref/src/TemplateHaskellQuasiquotes.html - utils/haddock/hypsrc-test/ref/src/TemplateHaskellSplices.html - utils/haddock/hypsrc-test/ref/src/Types.html - utils/haddock/hypsrc-test/ref/src/UsingQuasiquotes.html The diff was not included because it is too large. View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/afcdf92fef46cd73d85258732b8001e… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/afcdf92fef46cd73d85258732b8001e… You're receiving this email because of your account on gitlab.haskell.org.

1 0

[Git][ghc/ghc][master] deriveConstants: automatically pass -fcommon CC flag (#26393)
by Marge Bot (＠marge-bot) 23 Sep '25

23 Sep '25

Marge Bot pushed to branch master at Glasgow Haskell Compiler / GHC Commits: 64bb0e37 by Sylvain Henry at 2025-09-23T14:35:56-04:00 deriveConstants: automatically pass -fcommon CC flag (#26393) By mistake we tried to use deriveConstants without passing `--gcc-flag -fcommon` (which Hadrian does) and it failed. This patch: 1. adds parsing support for constants stored in the .bss section (i.e. when -fcommon isn't passed) 2. enables passing `-fcommon` automatically to the C compiler because Windows requires this for subtle reasons 3. Documents the subtle reasons (1) isn't strictly necessary because we always do (2) but it does no harm and it is still useful if the CC flags ever contain -fno-common - - - - - 2 changed files: - hadrian/src/Settings/Builders/DeriveConstants.hs - utils/deriveConstants/Main.hs Changes: ===================================== hadrian/src/Settings/Builders/DeriveConstants.hs ===================================== @@ -48,4 +48,4 @@ includeCcArgs = do , arg "-Irts/include" , arg $ "-I" ++ rtsPath </> "include" , notM targetSupportsSMP ? arg "-DNOSMP" - , arg "-fcommon" ] + ] ===================================== utils/deriveConstants/Main.hs ===================================== @@ -1,3 +1,5 @@ +{-# LANGUAGE LambdaCase #-} + {- ------------------------------------------------------------------------ (c) The GHC Team, 1992-2012 @@ -65,7 +67,17 @@ main = do opts <- parseArgs "mingw32" -> Windows _ -> DefaultOS verbose = o_verbose opts - gccFlags = o_gccFlags opts + gccFlags0 = o_gccFlags opts + -- nm applied to COFF files doesn't correctly report + -- the size of global variable symbols (it reports 0) + -- except for Common symbols (probably because the + -- merge algorithm of Common symbols takes symbol size + -- into account). So we force the use of `-fcommon` + -- here. + -- As far as we know, it is only required on Windows + -- but enabling it on other platforms does no harm, so + -- we enable it unconditionally. + gccFlags = "-fcommon" : gccFlags0 rs <- case os of JS -> getWantedJS _ -> getWanted verbose os tmpdir gccProg gccFlags nmProg @@ -751,7 +763,7 @@ getWanted verbose os tmpdir gccProgram gccFlags nmProgram mobjdumpProgram ++ "Workaround: You may want to pass\n" ++ " --with-nm=$(xcrun --find nm-classic)\n" ++ "to 'configure'.\n" - Just x -> die ("unexpected value round-tripped for CONTROL_GROUP_CONST_291: " ++ show x) + Just x -> die ("unexpected value round-tripped for CONTROL_GROUP_CONST_291: " ++ show x ++ "\n" ++ show m ++ "\n" ++ xs) mapM (lookupResult m) (wanteds (Just os)) where headers = ["#define IN_STG_CODE 0", @@ -841,14 +853,23 @@ getWanted verbose os tmpdir gccProgram gccFlags nmProgram mobjdumpProgram -- and returns ("MAX_Vanilla_REG", 11) parseNmLine line = case words line of - ('_' : n) : "C" : s : _ -> mkP n s - n : "C" : s : _ -> mkP n s - [n, "D", _, s] -> mkP n s - [s, "O", "*COM*", _, n] -> mkP n s + -- in common section + (n : "C" : s : _ ) -> mkP n s -- nm format (ELF, COFF) + [s, "O", "*COM*", _, n] -> mkP n s -- objdump format (ELF) + -- in .bss section + [n, "B", _, s ] -> mkP n s -- nm format (ELF) + [_, "g", "O", ".bss", s, n] -> mkP n s -- objdump format (ELF) + -- in data section + [n, "D", _, s] -> mkP n s -- nm format (ELF) + [_, "g", "O", ".data", s, n] -> mkP n s -- objdump format (ELF) _ -> Nothing - where mkP r s = case (stripPrefix prefix r, readHex s) of + where mkP r s = case (stripPrefix prefix (strip_ r), readHex s) of (Just name, [(size, "")]) -> Just (name, size) _ -> Nothing + -- strip leading underscore: some platforms (e.g. Darwin) add one + strip_ = \case + ('_':n) -> n + n -> n -- On AIX, `nm` isn't able to tell us the symbol size, so we -- need to use `objdump --syms`. However, unlike on OpenBSD, View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/64bb0e3738beba7f7f82a6885441704… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/64bb0e3738beba7f7f82a6885441704… You're receiving this email because of your account on gitlab.haskell.org.

1 0