Refactor IPE initialization

Here we refactor the representation of info table provenance information
in object code to significantly reduce its size and link-time impact.
Specifically, we deduplicate strings and represent them as 32-bit
offsets into a common string table.

In addition, we rework the registration logic to eliminate allocation
from the registration path, which is run from a static initializer where
things like allocation are technically undefined behavior (although it
did previously seem to work). For similar reasons we eliminate lock
usage from registration path, instead relying on atomic CAS.

Closes #22077.

2 files changed, 13 insertions, 5 deletions

diff --git a/compiler/GHC/Cmm/CLabel.hs b/compiler/GHC/Cmm/CLabel.hs
index 5f2b0eca59..9edccdccf5 100644
--- a/compiler/GHC/Cmm/CLabel.hs
+++ b/compiler/GHC/Cmm/CLabel.hs
@@ -302,6 +302,7 @@ data ModuleLabelKind
     | MLK_InitializerArray
     | MLK_Finalizer String
     | MLK_FinalizerArray
+    | MLK_IPEBuffer
     deriving (Eq, Ord)
 
 instance Outputable ModuleLabelKind where
@@ -309,6 +310,7 @@ instance Outputable ModuleLabelKind where
     ppr (MLK_Initializer s)  = text ("init__" ++ s)
     ppr MLK_FinalizerArray   = text "fini_arr"
     ppr (MLK_Finalizer s)    = text ("fini__" ++ s)
+    ppr MLK_IPEBuffer        = text "ipe_buf"
 
 isIdLabel :: CLabel -> Bool
 isIdLabel IdLabel{} = True
@@ -839,10 +841,10 @@ instance OutputableP Platform InfoProvEnt where
 -- Constructing Cost Center Labels
 mkCCLabel  :: CostCentre      -> CLabel
 mkCCSLabel :: CostCentreStack -> CLabel
-mkIPELabel :: InfoProvEnt -> CLabel
+mkIPELabel :: Module          -> CLabel
 mkCCLabel           cc          = CC_Label cc
 mkCCSLabel          ccs         = CCS_Label ccs
-mkIPELabel          ipe         = IPE_Label ipe
+mkIPELabel          mod         = ModuleLabel mod MLK_IPEBuffer
 
 mkRtsApFastLabel :: FastString -> CLabel
 mkRtsApFastLabel str = RtsLabel (RtsApFast (NonDetFastString str))
@@ -1020,6 +1022,7 @@ modLabelNeedsCDecl :: ModuleLabelKind -> Bool
 -- Code for finalizers and initializers are emitted in stub objects
 modLabelNeedsCDecl (MLK_Initializer _)  = True
 modLabelNeedsCDecl (MLK_Finalizer   _)  = True
+modLabelNeedsCDecl MLK_IPEBuffer        = True
 -- The finalizer and initializer arrays are emitted in the code of the module
 modLabelNeedsCDecl MLK_InitializerArray = False
 modLabelNeedsCDecl MLK_FinalizerArray   = False
@@ -1217,6 +1220,7 @@ moduleLabelKindType kind =
     MLK_InitializerArray -> DataLabel
     MLK_Finalizer _      -> CodeLabel
     MLK_FinalizerArray   -> DataLabel
+    MLK_IPEBuffer        -> DataLabel
 
 idInfoLabelType :: IdLabelInfo -> CLabelType
 idInfoLabelType info =
diff --git a/compiler/GHC/Cmm/Parser.y b/compiler/GHC/Cmm/Parser.y
index ed47fa7a7f..ae6e126b68 100644
--- a/compiler/GHC/Cmm/Parser.y
+++ b/compiler/GHC/Cmm/Parser.y
@@ -224,6 +224,7 @@ import GHC.StgToCmm.Layout     hiding (ArgRep(..))
 import GHC.StgToCmm.Ticky
 import GHC.StgToCmm.Prof
 import GHC.StgToCmm.Bind  ( emitBlackHoleCode, emitUpdateFrame )
+import GHC.StgToCmm.InfoTableProv
 
 import GHC.Cmm.Opt
 import GHC.Cmm.Graph
@@ -1517,9 +1518,12 @@ parseCmmFile cmmpConfig this_mod home_unit filename = do
         let fcode = do
               ((), cmm) <- getCmm $ unEC code "global" (initEnv (pdProfile pdConfig)) [] >> return ()
               -- See Note [Mapping Info Tables to Source Positions] (IPE Maps)
-              let used_info = map (cmmInfoTableToInfoProvEnt this_mod)
-                                              (mapMaybe topInfoTable cmm)
-              ((), cmm2) <- getCmm $ mapM_ emitInfoTableProv used_info
+              let used_info
+                    | do_ipe    = map (cmmInfoTableToInfoProvEnt this_mod) (mapMaybe topInfoTable cmm)
+                    | otherwise = []
+                    where
+                      do_ipe = stgToCmmInfoTableMap $ cmmpStgToCmmConfig cmmpConfig
+              ((), cmm2) <- getCmm $ emitIpeBufferListNode this_mod used_info
               return (cmm ++ cmm2, used_info)
             (cmm, _) = runC (cmmpStgToCmmConfig cmmpConfig) fstate st fcode
             (warnings,errors) = getPsMessages pst