inference: use the same method lookup cache across same inference trial

Previously the method lookup result was created per frame and so the
look cache hasn't been use that much. With this change the cache is
created per inference, and so the cached result will be used when we
already saw the same match in the same inference shot, and it may speed
up the lookup time a bit.

This commit also setups new `AbstractInterpreter` interface `get_method_lookup_cache`
which specifies what method lookup cache is used by each `AbstractInterpreter`.
`NativeInterpreter` creates a cache per inference, and so it is valid
since lookup is done in the same world age in the same inference shot.
External `AbstractInterpreter` doesn't opt into this cache by default,
and its behavior won't change in anyway.

Author: aviatesk

base/compiler/inferencestate.jl

@@ -141,7 +141,7 @@ mutable struct InferenceState
             cache === :global, false, false,
             Effects(consistent, ALWAYS_TRUE, ALWAYS_TRUE, ALWAYS_TRUE,
                    inbounds_taints_consistency),
-            CachedMethodTable(method_table(interp)),
+            CachedMethodTable(get_method_lookup_cache(interp), method_table(interp)),
             interp)
         result.result = frame
         cache !== :no && push!(get_inference_cache(interp), result)

base/compiler/methodtable.jl

@@ -2,22 +2,6 @@
 
 abstract type MethodTableView; end
 
-struct MethodLookupResult
-    # Really Vector{Core.MethodMatch}, but it's easier to represent this as
-    # and work with Vector{Any} on the C side.
-    matches::Vector{Any}
-    valid_worlds::WorldRange
-    ambig::Bool
-end
-length(result::MethodLookupResult) = length(result.matches)
-function iterate(result::MethodLookupResult, args...)
-    r = iterate(result.matches, args...)
-    r === nothing && return nothing
-    match, state = r
-    return (match::MethodMatch, state)
-end
-getindex(result::MethodLookupResult, idx::Int) = getindex(result.matches, idx)::MethodMatch
-
 """
     struct InternalMethodTable <: MethodTableView
 
@@ -46,12 +30,11 @@ Overlays another method table view with an additional local fast path cache that
 can respond to repeated, identical queries faster than the original method table.
 """
 struct CachedMethodTable{T} <: MethodTableView
-    cache::IdDict{Any, Union{Missing, MethodLookupResult}}
+    cache::MethodLookupCache
     table::T
+    CachedMethodTable(cache::MethodLookupCache, table::T) where T = new{T}(cache, table)
+    CachedMethodTable(::Nothing, table::T) where T = new{T}(MethodLookupCache(), table)
 end
-CachedMethodTable(table::T) where T =
-    CachedMethodTable{T}(IdDict{Any, Union{Missing, MethodLookupResult}}(),
-        table)
 
 """
     findall(sig::Type, view::MethodTableView; limit=typemax(Int))

base/compiler/types.jl

@@ -111,6 +111,23 @@ decode_effects_override(e::UInt8) =
         (e & 0x08) != 0x00,
         (e & 0x10) != 0x00)
 
+struct MethodLookupResult
+    # Really Vector{Core.MethodMatch}, but it's easier to represent this as
+    # and work with Vector{Any} on the C side.
+    matches::Vector{Any}
+    valid_worlds::WorldRange
+    ambig::Bool
+end
+length(result::MethodLookupResult) = length(result.matches)
+function iterate(result::MethodLookupResult, args...)
+    r = iterate(result.matches, args...)
+    r === nothing && return nothing
+    match, state = r
+    return (match::MethodMatch, state)
+end
+getindex(result::MethodLookupResult, idx::Int) = getindex(result.matches, idx)::MethodMatch
+const MethodLookupCache = IdDict{Any, Union{Missing, MethodLookupResult}}
+
 """
     InferenceResult
 
@@ -242,6 +259,8 @@ It contains many parameters used by the compilation pipeline.
 struct NativeInterpreter <: AbstractInterpreter
     # Cache of inference results for this particular interpreter
     cache::Vector{InferenceResult}
+    # cache of method lookup results
+    method_lookup_cache::MethodLookupCache
     # The world age we're working inside of
     world::UInt
 
@@ -263,10 +282,10 @@ struct NativeInterpreter <: AbstractInterpreter
         # incorrect, fail out loudly.
         @assert world <= get_world_counter()
 
-
         return new(
-            # Initially empty cache
+            # Initially empty caches
             Vector{InferenceResult}(),
+            MethodLookupCache(),
 
             # world age counter
             world,
@@ -316,6 +335,8 @@ may_discard_trees(::AbstractInterpreter) = true
 verbose_stmt_info(::AbstractInterpreter) = false
 
 method_table(interp::AbstractInterpreter) = InternalMethodTable(get_world_counter(interp))
+get_method_lookup_cache(ni::NativeInterpreter) = ni.method_lookup_cache
+get_method_lookup_cache(::AbstractInterpreter) = nothing
 
 """
 By default `AbstractInterpreter` implements the following inference bail out logic: