Add cumulative_compile_time_ns() to return the cumulative compile tim…

…e for a thread without stopping measurement.

```julia
julia> Base.cumulative_compile_time_ns_before()  # enable constantly measuring compilation time
0x0000000000000000

julia> @time @eval module M ; f(x) = 2+x; end
  0.006730 seconds (919 allocations: 55.212 KiB, 57.20% compilation time)
Main.M

julia> Base.cumulative_compile_time_ns()
0x00000000075246b3

julia> @time 2+2
  0.000000 seconds
4

julia> Base.cumulative_compile_time_ns()
0x0000000007fe4a46

julia> @time @eval M.f(2)
  0.003538 seconds (750 allocations: 46.247 KiB, 94.64% compilation time)
4

julia> Base.cumulative_compile_time_ns()
0x000000000831619e
```

base/timing.jl

@@ -58,6 +58,8 @@ end
 # cumulative total time spent on compilation
 cumulative_compile_time_ns_before() = ccall(:jl_cumulative_compile_time_ns_before, UInt64, ())
 cumulative_compile_time_ns_after() = ccall(:jl_cumulative_compile_time_ns_after, UInt64, ())
+# cumulative total time this thread has spent on compilation since process start.
+cumulative_compile_time_ns() = ccall(:jl_cumulative_compile_time_ns, UInt64, ())
 
 # total time spend in garbage collection, in nanoseconds
 gc_time_ns() = ccall(:jl_gc_total_hrtime, UInt64, ())

src/jitlayers.cpp

@@ -90,6 +90,12 @@ uint64_t jl_cumulative_compile_time_ns_after()
  return jl_cumulative_compile_time[tid];
 }
 
+extern "C" JL_DLLEXPORT
+uint64_t jl_cumulative_compile_time_ns() {
+ int tid = jl_threadid();
+ return jl_cumulative_compile_time[tid];
+}
+
 // this generates llvm code for the lambda info
 // and adds the result to the jitlayers
 // (and the shadow module),

test/misc.jl

@@ -263,6 +263,19 @@ function timev_macro_scope()
 end
 @test timev_macro_scope() == 1
 
+
+t1 = @async @time begin
+ sleep(2)
+ @eval module M ; f(x,y) = x+y ; end
+ @eval M.f(2,3)
+end
+
+t2 = @async begin
+ sleep(1)
+ @time 2 + 2
+end
+
+
 # interactive utilities
 
 struct ambigconvert; end # inject a problematic `convert` method to ensure it still works