Add toggle for contextualization.

Author: maleadt

src/compiler/common.jl

@@ -7,6 +7,8 @@ Base.@kwdef struct CompilerJob
     cap::VersionNumber
     kernel::Bool
 
+    contextualize::Bool = true
+
     # optional properties
     minthreads::Union{Nothing,CuDim} = nothing
     maxthreads::Union{Nothing,CuDim} = nothing

src/compiler/driver.jl

@@ -62,7 +62,7 @@ function codegen(target::Symbol, job::CompilerJob;
     @timeit_debug to "validation" check_method(job)
 
     @timeit_debug to "Julia front-end" begin
-        f = contextualize(job.f)
+        f = job.contextualize ? contextualize(job.f) : job.f
 
         # get the method instance
         world = typemax(UInt)

src/execution.jl

@@ -9,7 +9,7 @@ export @cuda, cudaconvert, cufunction, dynamic_cufunction, nearest_warpsize
 # the code it generates, or the execution
 function split_kwargs(kwargs)
     macro_kws    = [:dynamic]
-    compiler_kws = [:minthreads, :maxthreads, :blocks_per_sm, :maxregs, :name]
+    compiler_kws = [:minthreads, :maxthreads, :blocks_per_sm, :maxregs, :name, :contextualize]
     call_kws     = [:cooperative, :blocks, :threads, :config, :shmem, :stream]
     macro_kwargs = []
     compiler_kwargs = []
@@ -351,6 +351,7 @@ The following keyword arguments are supported:
 - `maxregs`: the maximum number of registers to be allocated to a single thread (only
   supported on LLVM 4.0+)
 - `name`: override the name that the kernel will have in the generated code
+- `contextualize`: whether to contextualize functions using Cassette (default: true)
 
 The output of this function is automatically cached, i.e. you can simply call `cufunction`
 in a hot path without degrading performance. New code will be generated automatically, when