Use a PTX-specific runtime that doesn't get optimized away entirely.

Otherwise we can't properly test the kernel state handling.

Author: maleadt

test/definitions/ptx.jl

@@ -16,6 +16,30 @@ GPUCompiler.kernel_state_type(@nospecialize(job::PTXCompilerJob)) = PTXKernelSta
 ptx_kernel_state() =
     unsafe_load(convert(Ptr{PTXKernelState}, GPUCompiler.kernel_state_pointer()))
 
+# a version of the test runtime that has some side effects, loading the kernel state
+# (so that we can test if kernel state arguments are appropriately optimized away)
+module PTXTestRuntime
+    using ..GPUCompiler
+    import ..PTXKernelState
+
+    kernel_state() = unsafe_load(convert(Ptr{PTXKernelState}, GPUCompiler.kernel_state_pointer()))
+
+    function signal_exception()
+        kernel_state()
+        return
+    end
+
+    # dummy methods
+    # HACK: if malloc returns 0 or traps, all calling functions (like jl_box_*)
+    #       get reduced to a trap, which really messes with our test suite.
+    malloc(sz) = Ptr{Cvoid}(Int(0xDEADBEEF))
+    report_oom(sz) = return
+    report_exception(ex) = return
+    report_exception_name(ex) = return
+    report_exception_frame(idx, func, file, line) = return
+end
+GPUCompiler.runtime_module(::PTXCompilerJob) = PTXTestRuntime
+
 function ptx_job(@nospecialize(func), @nospecialize(types); kernel::Bool=false,
                  minthreads=nothing, maxthreads=nothing, blocks_per_sm=nothing,
                  maxregs=nothing, kwargs...)