diff --git a/compiler/GHC/Cmm/Dataflow/Block.hs b/compiler/GHC/Cmm/Dataflow/Block.hs
--- a/compiler/GHC/Cmm/Dataflow/Block.hs
+++ b/compiler/GHC/Cmm/Dataflow/Block.hs
@@ -14,6 +14,7 @@
     , IndexedCO
     , Block(..)
     , blockAppend
+    , blockConcat
     , blockCons
     , blockFromList
     , blockJoin
@@ -136,6 +137,8 @@
 blockAppend :: Block n e O -> Block n O x -> Block n e x
 blockAppend = cat
 
+blockConcat :: [Block n O O] -> Block n O O
+blockConcat = foldr blockAppend emptyBlock
 
 -- Taking apart
 
diff --git a/compiler/GHC/Cmm/MachOp.hs b/compiler/GHC/Cmm/MachOp.hs
--- a/compiler/GHC/Cmm/MachOp.hs
+++ b/compiler/GHC/Cmm/MachOp.hs
@@ -24,6 +24,7 @@
     , machOpMemcpyishAlign
 
     -- Atomic read-modify-write
+    , MemoryOrdering(..)
     , AtomicMachOp(..)
    )
 where
@@ -514,8 +515,11 @@
     MO_FS_Conv from _   -> [from]
     MO_FF_Conv from _   -> [from]
 
-    MO_V_Insert  l r    -> [typeWidth (vec l (cmmBits r)),r,wordWidth platform]
-    MO_V_Extract l r    -> [typeWidth (vec l (cmmBits r)),wordWidth platform]
+    MO_V_Insert   l r   -> [typeWidth (vec l (cmmBits r)),r, W32]
+    MO_V_Extract  l r   -> [typeWidth (vec l (cmmBits r)), W32]
+    MO_VF_Insert  l r   -> [typeWidth (vec l (cmmFloat r)),r,W32]
+    MO_VF_Extract l r   -> [typeWidth (vec l (cmmFloat r)),W32]
+      -- SIMD vector indices are always 32 bit
 
     MO_V_Add _ r        -> [r,r]
     MO_V_Sub _ r        -> [r,r]
@@ -528,9 +532,6 @@
     MO_VU_Quot _ r      -> [r,r]
     MO_VU_Rem  _ r      -> [r,r]
 
-    MO_VF_Insert  l r   -> [typeWidth (vec l (cmmFloat r)),r,wordWidth platform]
-    MO_VF_Extract l r   -> [typeWidth (vec l (cmmFloat r)),wordWidth platform]
-
     MO_VF_Add  _ r      -> [r,r]
     MO_VF_Sub  _ r      -> [r,r]
     MO_VF_Mul  _ r      -> [r,r]
@@ -666,10 +667,12 @@
   | MO_BSwap Width
   | MO_BRev Width
 
-  -- Atomic read-modify-write.
+  -- | Atomic read-modify-write. Arguments are @[dest, n]@.
   | MO_AtomicRMW Width AtomicMachOp
-  | MO_AtomicRead Width
-  | MO_AtomicWrite Width
+  -- | Atomic read. Arguments are @[addr]@.
+  | MO_AtomicRead Width MemoryOrdering
+  -- | Atomic write. Arguments are @[addr, value]@.
+  | MO_AtomicWrite Width MemoryOrdering
   -- | Atomic compare-and-swap. Arguments are @[dest, expected, new]@.
   -- Sequentially consistent.
   -- Possible future refactoring: should this be an'MO_AtomicRMW' variant?
@@ -683,6 +686,14 @@
   | MO_SuspendThread
   | MO_ResumeThread
   deriving (Eq, Show)
+
+-- | C11 memory ordering semantics.
+data MemoryOrdering
+  = MemOrderRelaxed  -- ^ relaxed ordering
+  | MemOrderAcquire  -- ^ acquire ordering
+  | MemOrderRelease  -- ^ release ordering
+  | MemOrderSeqCst   -- ^ sequentially consistent
+  deriving (Eq, Ord, Show)
 
 -- | The operation to perform atomically.
 data AtomicMachOp =
diff --git a/compiler/GHC/Core/DataCon.hs b/compiler/GHC/Core/DataCon.hs
--- a/compiler/GHC/Core/DataCon.hs
+++ b/compiler/GHC/Core/DataCon.hs
@@ -795,20 +795,37 @@
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 Notice that we do *not* say the worker Id is strict even if the data
 constructor is declared strict
-     e.g.    data T = MkT !(Int,Int)
-Why?  Because the *wrapper* $WMkT is strict (and its unfolding has case
-expressions that do the evals) but the *worker* MkT itself is not. If we
-pretend it is strict then when we see
-     case x of y -> MkT y
-the simplifier thinks that y is "sure to be evaluated" (because the worker MkT
-is strict) and drops the case.  No, the workerId MkT is not strict.
+     e.g.    data T = MkT ![Int] Bool
+Even though most often the evals are done by the *wrapper* $WMkT, there are
+situations in which tag inference will re-insert evals around the worker.
+So for all intents and purposes the *worker* MkT is strict, too!
 
-However, the worker does have StrictnessMarks.  When the simplifier sees a
-pattern
-     case e of MkT x -> ...
-it uses the dataConRepStrictness of MkT to mark x as evaluated; but that's
-fine... dataConRepStrictness comes from the data con not from the worker Id.
+Unfortunately, if we exposed accurate strictness of DataCon workers, we'd
+see the following transformation:
 
+  f xs = case xs of xs' { __DEFAULT -> ... case MkT xs b of x { __DEFAULT -> [x] } } -- DmdAnal: Strict in xs
+  ==> { drop-seq, binder swap on xs' }
+  f xs = case MkT xs b of x { __DEFAULT -> [x] } -- DmdAnal: Still strict in xs
+  ==> { case-to-let }
+  f xs = let x = MkT xs' b in [x] -- DmdAnal: No longer strict in xs!
+
+I.e., we are ironically losing strictness in `xs` by dropping the eval on `xs`
+and then doing case-to-let. The issue is that `exprIsHNF` currently says that
+every DataCon worker app is a value. The implicit assumption is that surrounding
+evals will have evaluated strict fields like `xs` before! But now that we had
+just dropped the eval on `xs`, that assumption is no longer valid.
+
+Long story short: By keeping the demand signature lazy, the Simplifier will not
+drop the eval on `xs` and using `exprIsHNF` to decide case-to-let and others
+remains sound.
+
+Similarly, during demand analysis in dmdTransformDataConSig, we bump up the
+field demand with `C_01`, *not* `C_11`, because the latter exposes too much
+strictness that will drop the eval on `xs` above.
+
+This issue is discussed at length in
+"Failed idea: no wrappers for strict data constructors" in #21497 and #22475.
+
 Note [Bangs on data constructor arguments]
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 Consider
@@ -1456,14 +1473,18 @@
     res_ty
 
 dataConNonlinearType :: DataCon -> Type
+-- Just like dataConWrapperType, but with the
+-- linearity on the arguments all zapped to Many
 dataConNonlinearType (MkData { dcUserTyVarBinders = user_tvbs,
                                dcOtherTheta = theta, dcOrigArgTys = arg_tys,
-                               dcOrigResTy = res_ty })
-  = let arg_tys' = map (\(Scaled w t) -> Scaled (case w of One -> Many; _ -> w) t) arg_tys
-    in mkInvisForAllTys user_tvbs $
-       mkInvisFunTysMany theta $
-       mkVisFunTys arg_tys' $
-       res_ty
+                               dcOrigResTy = res_ty,
+                               dcStupidTheta = stupid_theta })
+  = mkInvisForAllTys user_tvbs $
+    mkInvisFunTysMany (stupid_theta ++ theta) $
+    mkVisFunTys arg_tys' $
+    res_ty
+  where
+    arg_tys' = map (\(Scaled w t) -> Scaled (case w of One -> Many; _ -> w) t) arg_tys
 
 dataConDisplayType :: Bool -> DataCon -> Type
 dataConDisplayType show_linear_types dc
diff --git a/compiler/GHC/Core/Opt/Arity.hs b/compiler/GHC/Core/Opt/Arity.hs
--- a/compiler/GHC/Core/Opt/Arity.hs
+++ b/compiler/GHC/Core/Opt/Arity.hs
@@ -853,7 +853,7 @@
 andArityType env at1 (AT [] div2) = andWithTail env div2 at1
 
 andWithTail :: ArityEnv -> Divergence -> ArityType -> ArityType
-andWithTail env div1 at2@(AT lams2 _)
+andWithTail env div1 at2
   | isDeadEndDiv div1     -- case x of { T -> error; F -> \y.e }
   = at2        -- Note [ABot branches: max arity wins]
 
@@ -861,7 +861,7 @@
   = AT [] topDiv
 
   | otherwise  -- case x of { T -> plusInt <expensive>; F -> \y.e }
-  = AT lams2 topDiv    -- We know div1 = topDiv
+  = takeWhileOneShot at2    -- We know div1 = topDiv
     -- See Note [Combining case branches: andWithTail]
 
 
@@ -2148,9 +2148,12 @@
       | Just (tv, res_ty) <- splitForAllTyCoVar_maybe ty
       , let (subst', tv') = substVarBndr subst tv
       = go (n-1) res_ty subst' (tv' : rev_bs) (e `App` varToCoreExpr tv')
+        -- The varToCoreExpr is important: `tv` might be a coercion variable
       | Just (mult, arg_ty, res_ty) <- splitFunTy_maybe ty
       , let (subst', b) = freshEtaId n subst (Scaled mult arg_ty)
-      = go (n-1) res_ty subst' (b : rev_bs) (e `App` Var b)
+      = go (n-1) res_ty subst' (b : rev_bs) (e `App` varToCoreExpr b)
+        -- The varToCoreExpr is important: `b` might be a coercion variable
+
       | otherwise
       = pprPanic "etaBodyForJoinPoint" $ int need_args $$
                                          ppr body $$ ppr (exprType body)
diff --git a/compiler/GHC/Core/Ppr.hs b/compiler/GHC/Core/Ppr.hs
--- a/compiler/GHC/Core/Ppr.hs
+++ b/compiler/GHC/Core/Ppr.hs
@@ -172,8 +172,12 @@
 pprOptCo :: Coercion -> SDoc
 -- Print a coercion optionally; i.e. honouring -dsuppress-coercions
 pprOptCo co = sdocOption sdocSuppressCoercions $ \case
-              True  -> angleBrackets (text "Co:" <> int (coercionSize co)) <+> dcolon <+> ppr (coercionType co)
-              False -> parens $ sep [ppr co, dcolon <+> ppr (coercionType co)]
+              True  -> angleBrackets (text "Co:" <> int (coercionSize co)) <+> dcolon <+> co_type
+              False -> parens $ sep [ppr co, dcolon <+> co_type]
+    where
+      co_type = sdocOption sdocSuppressCoercionTypes $ \case
+          True -> text "..."
+          False -> ppr (coercionType co)
 
 ppr_id_occ :: (SDoc -> SDoc) -> Id -> SDoc
 ppr_id_occ add_par id
diff --git a/compiler/GHC/Core/TyCon.hs b/compiler/GHC/Core/TyCon.hs
--- a/compiler/GHC/Core/TyCon.hs
+++ b/compiler/GHC/Core/TyCon.hs
@@ -129,6 +129,8 @@
         primRepIsFloat,
         primRepsCompatible,
         primRepCompatible,
+        primRepIsWord,
+        primRepIsInt,
 
 ) where
 
@@ -1782,6 +1784,24 @@
 primRepIsFloat  DoubleRep    = Just True
 primRepIsFloat  (VecRep _ _) = Nothing
 primRepIsFloat  _            = Just False
+
+-- Rep is one of the word reps.
+primRepIsWord :: PrimRep -> Bool
+primRepIsWord WordRep = True
+primRepIsWord (Word8Rep) = True
+primRepIsWord (Word16Rep) = True
+primRepIsWord (Word32Rep) = True
+primRepIsWord (Word64Rep) = True
+primRepIsWord _ = False
+
+-- Rep is one of the int reps.
+primRepIsInt :: PrimRep -> Bool
+primRepIsInt (IntRep) = True
+primRepIsInt (Int8Rep) = True
+primRepIsInt (Int16Rep) = True
+primRepIsInt (Int32Rep) = True
+primRepIsInt (Int64Rep) = True
+primRepIsInt _ = False
 
 {-
 ************************************************************************
diff --git a/compiler/GHC/Core/Type.hs b/compiler/GHC/Core/Type.hs
--- a/compiler/GHC/Core/Type.hs
+++ b/compiler/GHC/Core/Type.hs
@@ -1327,7 +1327,7 @@
     split args orig_ty ty | Just ty' <- coreView ty = split args orig_ty ty'
     split args orig_ty _                   = (reverse args, orig_ty)
 
-funResultTy :: Type -> Type
+funResultTy :: HasDebugCallStack => Type -> Type
 -- ^ Extract the function result type and panic if that is not possible
 funResultTy ty
   | FunTy { ft_res = res } <- coreFullView ty = res
diff --git a/compiler/GHC/Driver/Flags.hs b/compiler/GHC/Driver/Flags.hs
--- a/compiler/GHC/Driver/Flags.hs
+++ b/compiler/GHC/Driver/Flags.hs
@@ -353,8 +353,10 @@
    | Opt_ShowLoadedModules
    | Opt_HexWordLiterals -- See Note [Print Hexadecimal Literals]
 
-   -- Suppress all coercions, them replacing with '...'
+   -- Suppress a coercions inner structure, replacing it with '...'
    | Opt_SuppressCoercions
+   -- Suppress the type of a coercion as well
+   | Opt_SuppressCoercionTypes
    | Opt_SuppressVarKinds
    -- Suppress module id prefixes on variables.
    | Opt_SuppressModulePrefixes
@@ -372,6 +374,7 @@
    -- variables that have otherwise identical names.
    | Opt_SuppressUniques
    | Opt_SuppressStgExts
+   | Opt_SuppressStgReps
    | Opt_SuppressTicks     -- Replaces Opt_PprShowTicks
    | Opt_SuppressTimestamps -- ^ Suppress timestamps in dumps
    | Opt_SuppressCoreSizes  -- ^ Suppress per binding Core size stats in dumps
diff --git a/compiler/GHC/Driver/Session.hs b/compiler/GHC/Driver/Session.hs
--- a/compiler/GHC/Driver/Session.hs
+++ b/compiler/GHC/Driver/Session.hs
@@ -2428,12 +2428,14 @@
   -- have otherwise identical names.
   , make_ord_flag defGhcFlag "dsuppress-all"
       (NoArg $ do setGeneralFlag Opt_SuppressCoercions
+                  setGeneralFlag Opt_SuppressCoercionTypes
                   setGeneralFlag Opt_SuppressVarKinds
                   setGeneralFlag Opt_SuppressModulePrefixes
                   setGeneralFlag Opt_SuppressTypeApplications
                   setGeneralFlag Opt_SuppressIdInfo
                   setGeneralFlag Opt_SuppressTicks
                   setGeneralFlag Opt_SuppressStgExts
+                  setGeneralFlag Opt_SuppressStgReps
                   setGeneralFlag Opt_SuppressTypeSignatures
                   setGeneralFlag Opt_SuppressCoreSizes
                   setGeneralFlag Opt_SuppressTimestamps)
@@ -3372,7 +3374,9 @@
   depFlagSpec' "suppress-stg-free-vars" Opt_SuppressStgExts
      (useInstead "-d" "suppress-stg-exts"),
   flagSpec "suppress-stg-exts"          Opt_SuppressStgExts,
+  flagSpec "suppress-stg-reps"          Opt_SuppressStgReps,
   flagSpec "suppress-coercions"         Opt_SuppressCoercions,
+  flagSpec "suppress-coercion-types"    Opt_SuppressCoercionTypes,
   flagSpec "suppress-idinfo"            Opt_SuppressIdInfo,
   flagSpec "suppress-unfoldings"        Opt_SuppressUnfoldings,
   flagSpec "suppress-module-prefixes"   Opt_SuppressModulePrefixes,
@@ -3816,7 +3820,8 @@
       Opt_SimplPreInlining,
       Opt_VersionMacros,
       Opt_RPath,
-      Opt_CompactUnwind
+      Opt_CompactUnwind,
+      Opt_SuppressStgReps
     ]
 
     ++ [f | (ns,f) <- optLevelFlags, 0 `elem` ns]
@@ -5024,11 +5029,13 @@
   , sdocSuppressTypeApplications    = gopt Opt_SuppressTypeApplications dflags
   , sdocSuppressIdInfo              = gopt Opt_SuppressIdInfo dflags
   , sdocSuppressCoercions           = gopt Opt_SuppressCoercions dflags
+  , sdocSuppressCoercionTypes       = gopt Opt_SuppressCoercionTypes dflags
   , sdocSuppressUnfoldings          = gopt Opt_SuppressUnfoldings dflags
   , sdocSuppressVarKinds            = gopt Opt_SuppressVarKinds dflags
   , sdocSuppressUniques             = gopt Opt_SuppressUniques dflags
   , sdocSuppressModulePrefixes      = gopt Opt_SuppressModulePrefixes dflags
   , sdocSuppressStgExts             = gopt Opt_SuppressStgExts dflags
+  , sdocSuppressStgReps             = gopt Opt_SuppressStgReps dflags
   , sdocErrorSpans                  = gopt Opt_ErrorSpans dflags
   , sdocStarIsType                  = xopt LangExt.StarIsType dflags
   , sdocImpredicativeTypes          = xopt LangExt.ImpredicativeTypes dflags
diff --git a/compiler/GHC/Stg/Syntax.hs b/compiler/GHC/Stg/Syntax.hs
--- a/compiler/GHC/Stg/Syntax.hs
+++ b/compiler/GHC/Stg/Syntax.hs
@@ -87,7 +87,7 @@
 import GHC.Builtin.PrimOps ( PrimOp, PrimCall )
 import GHC.Core.TyCon    ( PrimRep(..), TyCon )
 import GHC.Core.Type     ( Type )
-import GHC.Types.RepType ( typePrimRep1 )
+import GHC.Types.RepType ( typePrimRep1, typePrimRep )
 import GHC.Utils.Panic.Plain
 
 {-
@@ -740,12 +740,23 @@
 pprStgTopBindings :: OutputablePass pass => StgPprOpts -> [GenStgTopBinding pass] -> SDoc
 pprStgTopBindings = pprGenStgTopBindings
 
+pprIdWithRep :: Id -> SDoc
+pprIdWithRep v = ppr v <> pprTypeRep (idType v)
+
+pprTypeRep :: Type -> SDoc
+pprTypeRep ty =
+    ppUnlessOption sdocSuppressStgReps $
+    char ':' <> case typePrimRep ty of
+                  [r] -> ppr r
+                  r -> ppr r
+
+
 instance Outputable StgArg where
   ppr = pprStgArg
 
 pprStgArg :: StgArg -> SDoc
-pprStgArg (StgVarArg var) = ppr var
-pprStgArg (StgLitArg con) = ppr con
+pprStgArg (StgVarArg var) = pprIdWithRep var
+pprStgArg (StgLitArg con) = ppr con <> pprTypeRep (literalType con)
 
 instance OutputablePass pass => Outputable  (GenStgExpr pass) where
   ppr = pprStgExpr panicStgPprOpts
diff --git a/compiler/GHC/Types/Demand.hs b/compiler/GHC/Types/Demand.hs
--- a/compiler/GHC/Types/Demand.hs
+++ b/compiler/GHC/Types/Demand.hs
@@ -94,7 +94,7 @@
 
 import GHC.Core.Type    ( Type )
 import GHC.Core.TyCon   ( isNewTyCon, isClassTyCon )
-import GHC.Core.DataCon ( splitDataProductType_maybe )
+import GHC.Core.DataCon ( splitDataProductType_maybe, StrictnessMark, isMarkedStrict )
 import GHC.Core.Multiplicity    ( scaledThing )
 
 import GHC.Utils.Binary
@@ -1020,11 +1020,13 @@
 -- whether it was unsaturated in the form of a 'Card'inality, denoting
 -- how many times the lambda body was entered.
 -- See Note [Demands from unsaturated function calls].
-peelManyCalls :: Int -> SubDemand -> Card
-peelManyCalls 0 _                          = C_11
--- See Note [Call demands are relative]
-peelManyCalls n (viewCall -> Just (m, sd)) = m `multCard` peelManyCalls (n-1) sd
-peelManyCalls _ _                          = C_0N
+peelManyCalls :: Arity -> SubDemand -> (Card, SubDemand)
+peelManyCalls k sd = go k C_11 sd
+  where
+    go 0 !n !sd                        = (n, sd)
+    go k !n (viewCall -> Just (m, sd)) = go (k-1) (n `multCard` m) sd
+    go _ _  _                          = (topCard, topSubDmd)
+{-# INLINE peelManyCalls #-} -- so that the pair cancels away in a `fst _` context
 
 -- See Note [Demand on the worker] in GHC.Core.Opt.WorkWrap
 mkWorkerDemand :: Int -> Demand
@@ -1068,7 +1070,7 @@
 saturatedByOneShots :: Int -> Demand -> Bool
 saturatedByOneShots _ AbsDmd    = True
 saturatedByOneShots _ BotDmd    = True
-saturatedByOneShots n (_ :* sd) = isUsedOnce (peelManyCalls n sd)
+saturatedByOneShots n (_ :* sd) = isUsedOnce $ fst $ peelManyCalls n sd
 
 {- Note [Strict demands]
 ~~~~~~~~~~~~~~~~~~~~~~~~
@@ -1188,6 +1190,46 @@
 the boxity info has to be stored in the *sub-demand* `sd`! There's no demand
 to store the boxity in. So we bit the bullet and now we store Boxity in
 'SubDemand', both in 'Prod' *and* 'Poly'. See also Note [Boxity in Poly].
+
+Note [Demand transformer for data constructors]
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+Consider the expression (x,y) with sub-demand P(SL,A).  What is the demand on
+x,y?  Obviously `x` is used strictly, and `y` not at all. So we want to
+decompose a product demand, and feed its components demands into the
+arguments.  That is the job of dmdTransformDataConSig.  More precisely,
+
+ * it gets the demand on the data constructor itself;
+   in the above example that is C(1,C(1,P(SL,A)))
+ * it returns the demands on the arguments;
+   in the above example that is [SL, A]
+
+Nasty wrinkle. Consider this code (#22475 has more realistic examples but
+assume this is what the demand analyser sees)
+
+   data T = MkT !Int Bool
+   get :: T -> Bool
+   get (MkT _ b) = b
+
+   foo = let v::Int = I# 7
+             t::T   = MkT v True
+         in get t
+
+Now `v` is unused by `get`, /but/ we can't give `v` an Absent demand,
+else we'll drop the binding and replace it with an error thunk.
+Then the code generator (more specifically GHC.Stg.InferTags.Rewrite)
+will add an extra eval of MkT's argument to give
+   foo = let v::Int = error "absent"
+             t::T   = case v of v' -> MkT v' True
+         in get t
+
+Boo!  Because of this extra eval (added in STG-land), the truth is that `MkT`
+may (or may not) evaluate its arguments (as established in #21497). Hence the
+use of `bump` in dmdTransformDataConSig, which adds in a `C_01` eval. The
+`C_01` says "may or may not evaluate" which is absolutely faithful to what
+InferTags.Rewrite does.
+
+In particular it is very important /not/ to make that a `C_11` eval,
+see Note [Data-con worker strictness].
 -}
 
 {- *********************************************************************
@@ -2039,20 +2081,24 @@
 -- return how the function evaluates its free variables and arguments.
 dmdTransformSig :: DmdSig -> DmdTransformer
 dmdTransformSig (DmdSig dmd_ty@(DmdType _ arg_ds _)) sd
-  = multDmdType (peelManyCalls (length arg_ds) sd) dmd_ty
+  = multDmdType (fst $ peelManyCalls (length arg_ds) sd) dmd_ty
     -- see Note [Demands from unsaturated function calls]
     -- and Note [What are demand signatures?]
 
 -- | A special 'DmdTransformer' for data constructors that feeds product
 -- demands into the constructor arguments.
-dmdTransformDataConSig :: Arity -> DmdTransformer
-dmdTransformDataConSig arity sd = case go arity sd of
-  Just dmds -> DmdType emptyDmdEnv dmds topDiv
-  Nothing   -> nopDmdType -- Not saturated
+dmdTransformDataConSig :: [StrictnessMark] -> DmdTransformer
+-- See Note [Demand transformer for data constructors]
+dmdTransformDataConSig str_marks sd = case viewProd arity body_sd of
+  Just (_, dmds) -> mk_body_ty n dmds
+  Nothing        -> nopDmdType
   where
-    go 0 sd             = snd <$> viewProd arity sd
-    go n (Call C_11 sd) = go (n-1) sd  -- strict calls only!
-    go _ _              = Nothing
+    arity = length str_marks
+    (n, body_sd) = peelManyCalls arity sd
+    mk_body_ty n dmds = DmdType emptyDmdEnv (zipWith (bump n) str_marks dmds) topDiv
+    bump n str dmd | isMarkedStrict str = multDmd n (plusDmd str_field_dmd dmd)
+                   | otherwise          = multDmd n dmd
+    str_field_dmd = C_01 :* seqSubDmd -- Why not C_11? See Note [Data-con worker strictness]
 
 -- | A special 'DmdTransformer' for dictionary selectors that feeds the demand
 -- on the result into the indicated dictionary component (if saturated).
diff --git a/compiler/GHC/Types/Id/Make.hs b/compiler/GHC/Types/Id/Make.hs
--- a/compiler/GHC/Types/Id/Make.hs
+++ b/compiler/GHC/Types/Id/Make.hs
@@ -694,7 +694,7 @@
                              -- applications are treated as values
                          `setInlinePragInfo`    wrap_prag
                          `setUnfoldingInfo`     wrap_unf
-                         `setDmdSigInfo`    wrap_sig
+                         `setDmdSigInfo`        wrap_sig
                              -- We need to get the CAF info right here because GHC.Iface.Tidy
                              -- does not tidy the IdInfo of implicit bindings (like the wrapper)
                              -- so it not make sure that the CAF info is sane
diff --git a/compiler/GHC/Types/Literal.hs b/compiler/GHC/Types/Literal.hs
--- a/compiler/GHC/Types/Literal.hs
+++ b/compiler/GHC/Types/Literal.hs
@@ -67,6 +67,7 @@
 import GHC.Prelude
 
 import GHC.Builtin.Types.Prim
+import GHC.Core.TyCo.Rep ( RuntimeRepType )
 import GHC.Core.Type
 import GHC.Utils.Outputable
 import GHC.Data.FastString
@@ -131,7 +132,7 @@
                                 -- that can be represented as a Literal. Create
                                 -- with 'nullAddrLit'
 
-  | LitRubbish Type             -- ^ A nonsense value of the given
+  | LitRubbish RuntimeRepType   -- ^ A nonsense value of the given
                                 -- representation. See Note [Rubbish literals].
                                 --
                                 -- The Type argument, rr, is of kind RuntimeRep.
diff --git a/compiler/GHC/Types/RepType.hs b/compiler/GHC/Types/RepType.hs
--- a/compiler/GHC/Types/RepType.hs
+++ b/compiler/GHC/Types/RepType.hs
@@ -13,7 +13,7 @@
     -- * Type representation for the code generator
     typePrimRep, typePrimRep1,
     runtimeRepPrimRep, typePrimRepArgs,
-    PrimRep(..), primRepToType,
+    PrimRep(..), primRepToType, primRepToRuntimeRep,
     countFunRepArgs, countConRepArgs, dataConRuntimeRepStrictness,
     tyConPrimRep, tyConPrimRep1,
     runtimeRepPrimRep_maybe, kindPrimRep_maybe, typePrimRep_maybe,
@@ -245,7 +245,8 @@
     in
       sumRep
 
-layoutUbxSum :: SortedSlotTys -- Layout of sum. Does not include tag.
+layoutUbxSum :: HasDebugCallStack
+             => SortedSlotTys -- Layout of sum. Does not include tag.
                               -- We assume that they are in increasing order
              -> [SlotTy]      -- Slot types of things we want to map to locations in the
                               -- sum layout
@@ -268,7 +269,8 @@
       | otherwise
       = findSlot arg (slot_idx + 1) slots useds
     findSlot _ _ [] _
-      = pprPanic "findSlot" (text "Can't find slot" $$ ppr sum_slots0 $$ ppr arg_slots0)
+      = pprPanic "findSlot" (text "Can't find slot" $$ text "sum_slots:" <> ppr sum_slots0
+                                                    $$ text "arg_slots:" <> ppr arg_slots0 )
 
 --------------------------------------------------------------------------------
 
@@ -286,7 +288,7 @@
 --
 -- TODO(michalt): We should probably introduce `SlotTy`s for 8-/16-/32-bit
 -- values, so that we can pack things more tightly.
-data SlotTy = PtrLiftedSlot | PtrUnliftedSlot | WordSlot | Word64Slot | FloatSlot | DoubleSlot
+data SlotTy = PtrLiftedSlot | PtrUnliftedSlot | WordSlot | Word64Slot | FloatSlot | DoubleSlot | VecSlot Int PrimElemRep
   deriving (Eq, Ord)
     -- Constructor order is important! If slot A could fit into slot B
     -- then slot A must occur first.  E.g.  FloatSlot before DoubleSlot
@@ -301,6 +303,7 @@
   ppr WordSlot        = text "WordSlot"
   ppr DoubleSlot      = text "DoubleSlot"
   ppr FloatSlot       = text "FloatSlot"
+  ppr (VecSlot n e)   = text "VecSlot" <+> ppr n <+> ppr e
 
 typeSlotTy :: UnaryType -> Maybe SlotTy
 typeSlotTy ty
@@ -326,7 +329,7 @@
 primRepSlot AddrRep     = WordSlot
 primRepSlot FloatRep    = FloatSlot
 primRepSlot DoubleRep   = DoubleSlot
-primRepSlot VecRep{}    = pprPanic "primRepSlot" (text "No slot for VecRep")
+primRepSlot (VecRep n e) = VecSlot n e
 
 slotPrimRep :: SlotTy -> PrimRep
 slotPrimRep PtrLiftedSlot   = LiftedRep
@@ -335,6 +338,7 @@
 slotPrimRep WordSlot        = WordRep
 slotPrimRep DoubleSlot      = DoubleRep
 slotPrimRep FloatSlot       = FloatRep
+slotPrimRep (VecSlot n e)   = VecRep n e
 
 -- | Returns the bigger type if one fits into the other. (commutative)
 --
@@ -347,18 +351,17 @@
   = Just ty1
   | isWordSlot ty1 && isWordSlot ty2
   = Just (max ty1 ty2)
-  | isFloatSlot ty1 && isFloatSlot ty2
-  = Just (max ty1 ty2)
   | otherwise
   = Nothing
+  -- We used to share slots between Float/Double but currently we can't easily
+  -- covert between float/double in a way that is both work free and safe.
+  -- So we put them in different slots.
+  -- See Note [Casting slot arguments]
   where
     isWordSlot Word64Slot = True
     isWordSlot WordSlot   = True
     isWordSlot _          = False
 
-    isFloatSlot DoubleSlot = True
-    isFloatSlot FloatSlot  = True
-    isFloatSlot _          = False
 
 
 {- **********************************************************************
diff --git a/compiler/GHC/Unit/State.hs b/compiler/GHC/Unit/State.hs
--- a/compiler/GHC/Unit/State.hs
+++ b/compiler/GHC/Unit/State.hs
@@ -571,12 +571,24 @@
   -- 1. Find all modules providing the ModuleName (this accounts for visibility/thinning etc)
   providers <- Map.filter originVisible <$> Map.lookup mn (moduleNameProvidersMap unit_st)
   -- 2. Get the UnitIds of the candidates
-  let candidates_uid = map (toUnitId . moduleUnit) $ Map.keys providers
+  let candidates_uid = concatMap to_uid $ Map.assocs providers
   -- 3. Get the package names of the candidates
   let candidates_units = map (\ui -> ((unitPackageName ui), unitId ui))
                               $ mapMaybe (\uid -> Map.lookup uid (unitInfoMap unit_st)) candidates_uid
   -- 4. Check to see if the PackageName helps us disambiguate any candidates.
   lookup pn candidates_units
+
+  where
+
+    -- Get the UnitId from which a visible identifier is from
+    to_uid :: (Module, ModuleOrigin) -> [UnitId]
+    to_uid (mod, ModOrigin mo re_exps _ _) =
+      case mo of
+        -- Available directly, but also potentially from re-exports
+        Just True ->  (toUnitId (moduleUnit mod)) : map unitId re_exps
+        -- Just available from these re-exports
+        _ -> map unitId re_exps
+    to_uid _ = []
 
 -- | Create a Map UnitId UnitInfo
 --
diff --git a/compiler/GHC/Utils/Outputable.hs b/compiler/GHC/Utils/Outputable.hs
--- a/compiler/GHC/Utils/Outputable.hs
+++ b/compiler/GHC/Utils/Outputable.hs
@@ -376,11 +376,13 @@
   , sdocSuppressTypeApplications    :: !Bool
   , sdocSuppressIdInfo              :: !Bool
   , sdocSuppressCoercions           :: !Bool
+  , sdocSuppressCoercionTypes       :: !Bool
   , sdocSuppressUnfoldings          :: !Bool
   , sdocSuppressVarKinds            :: !Bool
   , sdocSuppressUniques             :: !Bool
   , sdocSuppressModulePrefixes      :: !Bool
   , sdocSuppressStgExts             :: !Bool
+  , sdocSuppressStgReps             :: !Bool
   , sdocErrorSpans                  :: !Bool
   , sdocStarIsType                  :: !Bool
   , sdocLinearTypes                 :: !Bool
@@ -435,11 +437,13 @@
   , sdocSuppressTypeApplications    = False
   , sdocSuppressIdInfo              = False
   , sdocSuppressCoercions           = False
+  , sdocSuppressCoercionTypes       = False
   , sdocSuppressUnfoldings          = False
   , sdocSuppressVarKinds            = False
   , sdocSuppressUniques             = False
   , sdocSuppressModulePrefixes      = False
   , sdocSuppressStgExts             = False
+  , sdocSuppressStgReps             = True
   , sdocErrorSpans                  = False
   , sdocStarIsType                  = False
   , sdocImpredicativeTypes          = False
diff --git a/ghc-lib-parser.cabal b/ghc-lib-parser.cabal
--- a/ghc-lib-parser.cabal
+++ b/ghc-lib-parser.cabal
@@ -1,7 +1,7 @@
 cabal-version: 2.0
 build-type: Simple
 name: ghc-lib-parser
-version: 9.4.3.20221104
+version: 9.4.4.20221225
 license: BSD3
 license-file: LICENSE
 category: Development
diff --git a/ghc-lib/stage0/compiler/build/primop-primop-info.hs-incl b/ghc-lib/stage0/compiler/build/primop-primop-info.hs-incl
--- a/ghc-lib/stage0/compiler/build/primop-primop-info.hs-incl
+++ b/ghc-lib/stage0/compiler/build/primop-primop-info.hs-incl
@@ -656,16 +656,16 @@
 primOpInfo (VecBroadcastOp IntVec 32 W16) = mkGenPrimOp (fsLit "broadcastInt16X32#")  [] [int16PrimTy] (int16X32PrimTy)
 primOpInfo (VecBroadcastOp IntVec 16 W32) = mkGenPrimOp (fsLit "broadcastInt32X16#")  [] [int32PrimTy] (int32X16PrimTy)
 primOpInfo (VecBroadcastOp IntVec 8 W64) = mkGenPrimOp (fsLit "broadcastInt64X8#")  [] [int64PrimTy] (int64X8PrimTy)
-primOpInfo (VecBroadcastOp WordVec 16 W8) = mkGenPrimOp (fsLit "broadcastWord8X16#")  [] [wordPrimTy] (word8X16PrimTy)
-primOpInfo (VecBroadcastOp WordVec 8 W16) = mkGenPrimOp (fsLit "broadcastWord16X8#")  [] [wordPrimTy] (word16X8PrimTy)
+primOpInfo (VecBroadcastOp WordVec 16 W8) = mkGenPrimOp (fsLit "broadcastWord8X16#")  [] [word8PrimTy] (word8X16PrimTy)
+primOpInfo (VecBroadcastOp WordVec 8 W16) = mkGenPrimOp (fsLit "broadcastWord16X8#")  [] [word16PrimTy] (word16X8PrimTy)
 primOpInfo (VecBroadcastOp WordVec 4 W32) = mkGenPrimOp (fsLit "broadcastWord32X4#")  [] [word32PrimTy] (word32X4PrimTy)
 primOpInfo (VecBroadcastOp WordVec 2 W64) = mkGenPrimOp (fsLit "broadcastWord64X2#")  [] [word64PrimTy] (word64X2PrimTy)
-primOpInfo (VecBroadcastOp WordVec 32 W8) = mkGenPrimOp (fsLit "broadcastWord8X32#")  [] [wordPrimTy] (word8X32PrimTy)
-primOpInfo (VecBroadcastOp WordVec 16 W16) = mkGenPrimOp (fsLit "broadcastWord16X16#")  [] [wordPrimTy] (word16X16PrimTy)
+primOpInfo (VecBroadcastOp WordVec 32 W8) = mkGenPrimOp (fsLit "broadcastWord8X32#")  [] [word8PrimTy] (word8X32PrimTy)
+primOpInfo (VecBroadcastOp WordVec 16 W16) = mkGenPrimOp (fsLit "broadcastWord16X16#")  [] [word16PrimTy] (word16X16PrimTy)
 primOpInfo (VecBroadcastOp WordVec 8 W32) = mkGenPrimOp (fsLit "broadcastWord32X8#")  [] [word32PrimTy] (word32X8PrimTy)
 primOpInfo (VecBroadcastOp WordVec 4 W64) = mkGenPrimOp (fsLit "broadcastWord64X4#")  [] [word64PrimTy] (word64X4PrimTy)
-primOpInfo (VecBroadcastOp WordVec 64 W8) = mkGenPrimOp (fsLit "broadcastWord8X64#")  [] [wordPrimTy] (word8X64PrimTy)
-primOpInfo (VecBroadcastOp WordVec 32 W16) = mkGenPrimOp (fsLit "broadcastWord16X32#")  [] [wordPrimTy] (word16X32PrimTy)
+primOpInfo (VecBroadcastOp WordVec 64 W8) = mkGenPrimOp (fsLit "broadcastWord8X64#")  [] [word8PrimTy] (word8X64PrimTy)
+primOpInfo (VecBroadcastOp WordVec 32 W16) = mkGenPrimOp (fsLit "broadcastWord16X32#")  [] [word16PrimTy] (word16X32PrimTy)
 primOpInfo (VecBroadcastOp WordVec 16 W32) = mkGenPrimOp (fsLit "broadcastWord32X16#")  [] [word32PrimTy] (word32X16PrimTy)
 primOpInfo (VecBroadcastOp WordVec 8 W64) = mkGenPrimOp (fsLit "broadcastWord64X8#")  [] [word64PrimTy] (word64X8PrimTy)
 primOpInfo (VecBroadcastOp FloatVec 4 W32) = mkGenPrimOp (fsLit "broadcastFloatX4#")  [] [floatPrimTy] (floatX4PrimTy)
@@ -686,16 +686,16 @@
 primOpInfo (VecPackOp IntVec 32 W16) = mkGenPrimOp (fsLit "packInt16X32#")  [] [(mkTupleTy Unboxed [int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy])] (int16X32PrimTy)
 primOpInfo (VecPackOp IntVec 16 W32) = mkGenPrimOp (fsLit "packInt32X16#")  [] [(mkTupleTy Unboxed [int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy])] (int32X16PrimTy)
 primOpInfo (VecPackOp IntVec 8 W64) = mkGenPrimOp (fsLit "packInt64X8#")  [] [(mkTupleTy Unboxed [int64PrimTy, int64PrimTy, int64PrimTy, int64PrimTy, int64PrimTy, int64PrimTy, int64PrimTy, int64PrimTy])] (int64X8PrimTy)
-primOpInfo (VecPackOp WordVec 16 W8) = mkGenPrimOp (fsLit "packWord8X16#")  [] [(mkTupleTy Unboxed [wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy])] (word8X16PrimTy)
-primOpInfo (VecPackOp WordVec 8 W16) = mkGenPrimOp (fsLit "packWord16X8#")  [] [(mkTupleTy Unboxed [wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy])] (word16X8PrimTy)
+primOpInfo (VecPackOp WordVec 16 W8) = mkGenPrimOp (fsLit "packWord8X16#")  [] [(mkTupleTy Unboxed [word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy])] (word8X16PrimTy)
+primOpInfo (VecPackOp WordVec 8 W16) = mkGenPrimOp (fsLit "packWord16X8#")  [] [(mkTupleTy Unboxed [word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy])] (word16X8PrimTy)
 primOpInfo (VecPackOp WordVec 4 W32) = mkGenPrimOp (fsLit "packWord32X4#")  [] [(mkTupleTy Unboxed [word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy])] (word32X4PrimTy)
 primOpInfo (VecPackOp WordVec 2 W64) = mkGenPrimOp (fsLit "packWord64X2#")  [] [(mkTupleTy Unboxed [word64PrimTy, word64PrimTy])] (word64X2PrimTy)
-primOpInfo (VecPackOp WordVec 32 W8) = mkGenPrimOp (fsLit "packWord8X32#")  [] [(mkTupleTy Unboxed [wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy])] (word8X32PrimTy)
-primOpInfo (VecPackOp WordVec 16 W16) = mkGenPrimOp (fsLit "packWord16X16#")  [] [(mkTupleTy Unboxed [wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy])] (word16X16PrimTy)
+primOpInfo (VecPackOp WordVec 32 W8) = mkGenPrimOp (fsLit "packWord8X32#")  [] [(mkTupleTy Unboxed [word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy])] (word8X32PrimTy)
+primOpInfo (VecPackOp WordVec 16 W16) = mkGenPrimOp (fsLit "packWord16X16#")  [] [(mkTupleTy Unboxed [word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy])] (word16X16PrimTy)
 primOpInfo (VecPackOp WordVec 8 W32) = mkGenPrimOp (fsLit "packWord32X8#")  [] [(mkTupleTy Unboxed [word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy])] (word32X8PrimTy)
 primOpInfo (VecPackOp WordVec 4 W64) = mkGenPrimOp (fsLit "packWord64X4#")  [] [(mkTupleTy Unboxed [word64PrimTy, word64PrimTy, word64PrimTy, word64PrimTy])] (word64X4PrimTy)
-primOpInfo (VecPackOp WordVec 64 W8) = mkGenPrimOp (fsLit "packWord8X64#")  [] [(mkTupleTy Unboxed [wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy])] (word8X64PrimTy)
-primOpInfo (VecPackOp WordVec 32 W16) = mkGenPrimOp (fsLit "packWord16X32#")  [] [(mkTupleTy Unboxed [wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy])] (word16X32PrimTy)
+primOpInfo (VecPackOp WordVec 64 W8) = mkGenPrimOp (fsLit "packWord8X64#")  [] [(mkTupleTy Unboxed [word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy])] (word8X64PrimTy)
+primOpInfo (VecPackOp WordVec 32 W16) = mkGenPrimOp (fsLit "packWord16X32#")  [] [(mkTupleTy Unboxed [word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy])] (word16X32PrimTy)
 primOpInfo (VecPackOp WordVec 16 W32) = mkGenPrimOp (fsLit "packWord32X16#")  [] [(mkTupleTy Unboxed [word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy])] (word32X16PrimTy)
 primOpInfo (VecPackOp WordVec 8 W64) = mkGenPrimOp (fsLit "packWord64X8#")  [] [(mkTupleTy Unboxed [word64PrimTy, word64PrimTy, word64PrimTy, word64PrimTy, word64PrimTy, word64PrimTy, word64PrimTy, word64PrimTy])] (word64X8PrimTy)
 primOpInfo (VecPackOp FloatVec 4 W32) = mkGenPrimOp (fsLit "packFloatX4#")  [] [(mkTupleTy Unboxed [floatPrimTy, floatPrimTy, floatPrimTy, floatPrimTy])] (floatX4PrimTy)
@@ -716,16 +716,16 @@
 primOpInfo (VecUnpackOp IntVec 32 W16) = mkGenPrimOp (fsLit "unpackInt16X32#")  [] [int16X32PrimTy] ((mkTupleTy Unboxed [int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy, int16PrimTy]))
 primOpInfo (VecUnpackOp IntVec 16 W32) = mkGenPrimOp (fsLit "unpackInt32X16#")  [] [int32X16PrimTy] ((mkTupleTy Unboxed [int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy, int32PrimTy]))
 primOpInfo (VecUnpackOp IntVec 8 W64) = mkGenPrimOp (fsLit "unpackInt64X8#")  [] [int64X8PrimTy] ((mkTupleTy Unboxed [int64PrimTy, int64PrimTy, int64PrimTy, int64PrimTy, int64PrimTy, int64PrimTy, int64PrimTy, int64PrimTy]))
-primOpInfo (VecUnpackOp WordVec 16 W8) = mkGenPrimOp (fsLit "unpackWord8X16#")  [] [word8X16PrimTy] ((mkTupleTy Unboxed [wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy]))
-primOpInfo (VecUnpackOp WordVec 8 W16) = mkGenPrimOp (fsLit "unpackWord16X8#")  [] [word16X8PrimTy] ((mkTupleTy Unboxed [wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy]))
+primOpInfo (VecUnpackOp WordVec 16 W8) = mkGenPrimOp (fsLit "unpackWord8X16#")  [] [word8X16PrimTy] ((mkTupleTy Unboxed [word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy]))
+primOpInfo (VecUnpackOp WordVec 8 W16) = mkGenPrimOp (fsLit "unpackWord16X8#")  [] [word16X8PrimTy] ((mkTupleTy Unboxed [word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy]))
 primOpInfo (VecUnpackOp WordVec 4 W32) = mkGenPrimOp (fsLit "unpackWord32X4#")  [] [word32X4PrimTy] ((mkTupleTy Unboxed [word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy]))
 primOpInfo (VecUnpackOp WordVec 2 W64) = mkGenPrimOp (fsLit "unpackWord64X2#")  [] [word64X2PrimTy] ((mkTupleTy Unboxed [word64PrimTy, word64PrimTy]))
-primOpInfo (VecUnpackOp WordVec 32 W8) = mkGenPrimOp (fsLit "unpackWord8X32#")  [] [word8X32PrimTy] ((mkTupleTy Unboxed [wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy]))
-primOpInfo (VecUnpackOp WordVec 16 W16) = mkGenPrimOp (fsLit "unpackWord16X16#")  [] [word16X16PrimTy] ((mkTupleTy Unboxed [wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy]))
+primOpInfo (VecUnpackOp WordVec 32 W8) = mkGenPrimOp (fsLit "unpackWord8X32#")  [] [word8X32PrimTy] ((mkTupleTy Unboxed [word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy]))
+primOpInfo (VecUnpackOp WordVec 16 W16) = mkGenPrimOp (fsLit "unpackWord16X16#")  [] [word16X16PrimTy] ((mkTupleTy Unboxed [word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy]))
 primOpInfo (VecUnpackOp WordVec 8 W32) = mkGenPrimOp (fsLit "unpackWord32X8#")  [] [word32X8PrimTy] ((mkTupleTy Unboxed [word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy]))
 primOpInfo (VecUnpackOp WordVec 4 W64) = mkGenPrimOp (fsLit "unpackWord64X4#")  [] [word64X4PrimTy] ((mkTupleTy Unboxed [word64PrimTy, word64PrimTy, word64PrimTy, word64PrimTy]))
-primOpInfo (VecUnpackOp WordVec 64 W8) = mkGenPrimOp (fsLit "unpackWord8X64#")  [] [word8X64PrimTy] ((mkTupleTy Unboxed [wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy]))
-primOpInfo (VecUnpackOp WordVec 32 W16) = mkGenPrimOp (fsLit "unpackWord16X32#")  [] [word16X32PrimTy] ((mkTupleTy Unboxed [wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy, wordPrimTy]))
+primOpInfo (VecUnpackOp WordVec 64 W8) = mkGenPrimOp (fsLit "unpackWord8X64#")  [] [word8X64PrimTy] ((mkTupleTy Unboxed [word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy, word8PrimTy]))
+primOpInfo (VecUnpackOp WordVec 32 W16) = mkGenPrimOp (fsLit "unpackWord16X32#")  [] [word16X32PrimTy] ((mkTupleTy Unboxed [word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy, word16PrimTy]))
 primOpInfo (VecUnpackOp WordVec 16 W32) = mkGenPrimOp (fsLit "unpackWord32X16#")  [] [word32X16PrimTy] ((mkTupleTy Unboxed [word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy, word32PrimTy]))
 primOpInfo (VecUnpackOp WordVec 8 W64) = mkGenPrimOp (fsLit "unpackWord64X8#")  [] [word64X8PrimTy] ((mkTupleTy Unboxed [word64PrimTy, word64PrimTy, word64PrimTy, word64PrimTy, word64PrimTy, word64PrimTy, word64PrimTy, word64PrimTy]))
 primOpInfo (VecUnpackOp FloatVec 4 W32) = mkGenPrimOp (fsLit "unpackFloatX4#")  [] [floatX4PrimTy] ((mkTupleTy Unboxed [floatPrimTy, floatPrimTy, floatPrimTy, floatPrimTy]))
@@ -746,16 +746,16 @@
 primOpInfo (VecInsertOp IntVec 32 W16) = mkGenPrimOp (fsLit "insertInt16X32#")  [] [int16X32PrimTy, int16PrimTy, intPrimTy] (int16X32PrimTy)
 primOpInfo (VecInsertOp IntVec 16 W32) = mkGenPrimOp (fsLit "insertInt32X16#")  [] [int32X16PrimTy, int32PrimTy, intPrimTy] (int32X16PrimTy)
 primOpInfo (VecInsertOp IntVec 8 W64) = mkGenPrimOp (fsLit "insertInt64X8#")  [] [int64X8PrimTy, int64PrimTy, intPrimTy] (int64X8PrimTy)
-primOpInfo (VecInsertOp WordVec 16 W8) = mkGenPrimOp (fsLit "insertWord8X16#")  [] [word8X16PrimTy, wordPrimTy, intPrimTy] (word8X16PrimTy)
-primOpInfo (VecInsertOp WordVec 8 W16) = mkGenPrimOp (fsLit "insertWord16X8#")  [] [word16X8PrimTy, wordPrimTy, intPrimTy] (word16X8PrimTy)
+primOpInfo (VecInsertOp WordVec 16 W8) = mkGenPrimOp (fsLit "insertWord8X16#")  [] [word8X16PrimTy, word8PrimTy, intPrimTy] (word8X16PrimTy)
+primOpInfo (VecInsertOp WordVec 8 W16) = mkGenPrimOp (fsLit "insertWord16X8#")  [] [word16X8PrimTy, word16PrimTy, intPrimTy] (word16X8PrimTy)
 primOpInfo (VecInsertOp WordVec 4 W32) = mkGenPrimOp (fsLit "insertWord32X4#")  [] [word32X4PrimTy, word32PrimTy, intPrimTy] (word32X4PrimTy)
 primOpInfo (VecInsertOp WordVec 2 W64) = mkGenPrimOp (fsLit "insertWord64X2#")  [] [word64X2PrimTy, word64PrimTy, intPrimTy] (word64X2PrimTy)
-primOpInfo (VecInsertOp WordVec 32 W8) = mkGenPrimOp (fsLit "insertWord8X32#")  [] [word8X32PrimTy, wordPrimTy, intPrimTy] (word8X32PrimTy)
-primOpInfo (VecInsertOp WordVec 16 W16) = mkGenPrimOp (fsLit "insertWord16X16#")  [] [word16X16PrimTy, wordPrimTy, intPrimTy] (word16X16PrimTy)
+primOpInfo (VecInsertOp WordVec 32 W8) = mkGenPrimOp (fsLit "insertWord8X32#")  [] [word8X32PrimTy, word8PrimTy, intPrimTy] (word8X32PrimTy)
+primOpInfo (VecInsertOp WordVec 16 W16) = mkGenPrimOp (fsLit "insertWord16X16#")  [] [word16X16PrimTy, word16PrimTy, intPrimTy] (word16X16PrimTy)
 primOpInfo (VecInsertOp WordVec 8 W32) = mkGenPrimOp (fsLit "insertWord32X8#")  [] [word32X8PrimTy, word32PrimTy, intPrimTy] (word32X8PrimTy)
 primOpInfo (VecInsertOp WordVec 4 W64) = mkGenPrimOp (fsLit "insertWord64X4#")  [] [word64X4PrimTy, word64PrimTy, intPrimTy] (word64X4PrimTy)
-primOpInfo (VecInsertOp WordVec 64 W8) = mkGenPrimOp (fsLit "insertWord8X64#")  [] [word8X64PrimTy, wordPrimTy, intPrimTy] (word8X64PrimTy)
-primOpInfo (VecInsertOp WordVec 32 W16) = mkGenPrimOp (fsLit "insertWord16X32#")  [] [word16X32PrimTy, wordPrimTy, intPrimTy] (word16X32PrimTy)
+primOpInfo (VecInsertOp WordVec 64 W8) = mkGenPrimOp (fsLit "insertWord8X64#")  [] [word8X64PrimTy, word8PrimTy, intPrimTy] (word8X64PrimTy)
+primOpInfo (VecInsertOp WordVec 32 W16) = mkGenPrimOp (fsLit "insertWord16X32#")  [] [word16X32PrimTy, word16PrimTy, intPrimTy] (word16X32PrimTy)
 primOpInfo (VecInsertOp WordVec 16 W32) = mkGenPrimOp (fsLit "insertWord32X16#")  [] [word32X16PrimTy, word32PrimTy, intPrimTy] (word32X16PrimTy)
 primOpInfo (VecInsertOp WordVec 8 W64) = mkGenPrimOp (fsLit "insertWord64X8#")  [] [word64X8PrimTy, word64PrimTy, intPrimTy] (word64X8PrimTy)
 primOpInfo (VecInsertOp FloatVec 4 W32) = mkGenPrimOp (fsLit "insertFloatX4#")  [] [floatX4PrimTy, floatPrimTy, intPrimTy] (floatX4PrimTy)
diff --git a/ghc-lib/stage0/libraries/ghc-boot/build/GHC/Version.hs b/ghc-lib/stage0/libraries/ghc-boot/build/GHC/Version.hs
--- a/ghc-lib/stage0/libraries/ghc-boot/build/GHC/Version.hs
+++ b/ghc-lib/stage0/libraries/ghc-boot/build/GHC/Version.hs
@@ -3,19 +3,19 @@
 import Prelude -- See Note [Why do we import Prelude here?]
 
 cProjectGitCommitId   :: String
-cProjectGitCommitId   = "8f8dba0190fe2a3a8b148fecf0dc83a725fb3fd2"
+cProjectGitCommitId   = "cafe75946c465dd20c324918807464e09f12ac2f"
 
 cProjectVersion       :: String
-cProjectVersion       = "9.4.3"
+cProjectVersion       = "9.4.4"
 
 cProjectVersionInt    :: String
 cProjectVersionInt    = "904"
 
 cProjectPatchLevel    :: String
-cProjectPatchLevel    = "3"
+cProjectPatchLevel    = "4"
 
 cProjectPatchLevel1   :: String
-cProjectPatchLevel1   = "3"
+cProjectPatchLevel1   = "4"
 
 cProjectPatchLevel2   :: String
 cProjectPatchLevel2   = "0"
diff --git a/libraries/ghci/GHCi/Message.hs b/libraries/ghci/GHCi/Message.hs
--- a/libraries/ghci/GHCi/Message.hs
+++ b/libraries/ghci/GHCi/Message.hs
@@ -465,7 +465,7 @@
 #define MIN_VERSION_ghc_heap(major1,major2,minor) (\
   (major1) <  9 || \
   (major1) == 9 && (major2) <  4 || \
-  (major1) == 9 && (major2) == 4 && (minor) <= 3)
+  (major1) == 9 && (major2) == 4 && (minor) <= 4)
 #endif /* MIN_VERSION_ghc_heap */
 #if MIN_VERSION_ghc_heap(8,11,0)
 instance Binary Heap.StgTSOProfInfo