TArray Indexing Performance (#83)

* improve TArray benchmarks

* update benchmarks

* 1. Base.@propagate_inbounds, 2. use function in benchmarks

* optimize indexing for Matrix{Float64}

* indent code

* move TArray data into TArray

* update benchmarks

* update comments and fix other issues

* more performant way to perdicate if a weakref is empty

* underlying array type as a parametric type of TArray

* minor update

* update according to code review

* fix broadcasting

* Update Project.toml

Co-authored-by: Hong Ge <[email protected]>

Author: KDr2

Diff for: Project.toml

@@ -3,7 +3,7 @@ uuid = "6f1fad26-d15e-5dc8-ae53-837a1d7b8c9f"
 license = "MIT"
 desc = "C shim for task copying in Turing"
 repo = "https://github.com/TuringLang/Libtask.jl.git"
-version = "0.5.1"
+version = "0.5.2"
 
 [deps]
 Libtask_jll = "3ae2931a-708c-5973-9c38-ccf7496fb450"

Diff for: src/ctask.jl

@@ -7,7 +7,10 @@ struct CTask
     task::Task
 
     function CTask(task::Task)
-        new(enable_stack_copying(task))
+        ret = new(enable_stack_copying(task))
+        task.storage === nothing && (task.storage = IdDict())
+        task.storage[:ctask] = ret
+        ret
     end
 end
 
@@ -24,12 +27,13 @@ end
 
 function Base.showerror(io::IO, ex::CTaskException)
     println(io, "CTaskException:")
+    ct = ex.task
     bt = @static if VERSION < v"1.6.0-DEV.1145"
         ct.backtrace
     else
         ct.storage[:_libtask_bt]
     end
-    showerror(io, ex.task.exception, bt)
+    showerror(io, ct.exception, bt)
 end
 
 # Utility function for self-copying mechanism
@@ -107,6 +111,7 @@ function Base.copy(ctask::CTask)
     setstate!(newtask, getstate(task))
     newtask.result = task.result
 
+    copy_tarrays(task, newtask)
     return CTask(newtask)
 end