first steps towards ptx generation

package/src/Transpiler.jl

@@ -1,3 +1,6 @@
+module Transpiler
+using CUDA
+using ..ExpressionProcessing
 
 # culoadtest(N, rand(["add.f32", "sub.f32", "mul.f32", "div.approx.f32"]))
 function culoadtest(N::Int32, op = "add.f32")
@@ -73,4 +76,47 @@ function culoadtest(N::Int32, op = "add.f32")
 		blocksPerGrid  = (N + threadsPerBlock - 1) ÷ threadsPerBlock;
 	
 		@time CUDA.@sync cudacall(func, Tuple{CuPtr{Cfloat},CuPtr{Cfloat},CuPtr{Cfloat},Cint}, d_a, d_b, d_c, N; threads=threadsPerBlock, blocks=blocksPerGrid)
-	end
+end
+
+function transpile(expression::ExpressionProcessing.PostfixType)
+
+end
+
+# TODO: Make version, target and address_size configurable
+function get_cuda_header()::String
+	return "
+	.version 7.1
+	.target sm_52
+	.address_size 64
+	"
+end
+
+function get_kernel_signature(kernelName::String, parameters::Vector{Type})::String
+	signature = ".visible .entry " * kernelName
+
+	stringBuilder = IOBuffer()
+	print(stringBuilder, "(")
+
+	for i in eachindex(parameters)
+		type = type_to_cuda_type(parameters[i])
+		print(stringBuilder, 
+			  ".param ", type, " ", kernelName, "_param_", i, ",")
+	end
+
+	print(stringBuilder, ")")
+	return String(take!(stringBuilder))
+end
+
+function type_to_cuda_type(type::Type)::String
+	if type == Int64
+		return ".s64"
+	elseif type == Float64
+		return ".f64"
+	else
+		return ".b64"
+	end
+end
+
+
+end
+