benchmarking: updated benchmarking suite and prepared for taking the benchmarks

package/src/Transpiler.jl

@@ -56,40 +56,26 @@ function evaluate(expressions::Vector{Expr}, variables::Matrix{Float32}, paramet
 
 	threads = min(variableCols, 256)
 	blocks = cld(variableCols, threads)
-
+	
+	kernelName = "evaluate_gpu"
 	# TODO: Implement batching as a middleground between "transpile everything and then run" and "tranpile one run one" even though cudacall is async
-	@inbounds for i in eachindex(expressions)
+	@inbounds Threads.@threads for i in eachindex(expressions)
 		# if haskey(resultCache, expressions[i])
 		# 	kernels[i] = resultCache[expressions[i]]
 		# 	continue
 		# end
-
-		formattedExpr = ExpressionProcessing.expr_to_postfix(expressions[i])
-		kernel = transpile(formattedExpr, varRows, Utils.get_max_inner_length(parameters), variableCols, i-1) # i-1 because julia is 1-based but PTX needs 0-based indexing
 		
-		# try
-			linker = CuLink()
-			add_data!(linker, "ExpressionProcessing", kernel)
-			
-			image = complete(linker)
-			
-			mod = CuModule(image)
+		formattedExpr = ExpressionProcessing.expr_to_postfix(expressions[i])
+		kernel = transpile(formattedExpr, varRows, Utils.get_max_inner_length(parameters), variableCols, i-1, kernelName) # i-1 because julia is 1-based but PTX needs 0-based indexing
 
-			compiledKernel = CuFunction(mod, "ExpressionProcessing")
-			cudacall(compiledKernel, (CuPtr{Float32},CuPtr{Float32},CuPtr{Float32}), cudaVars, cudaParams, cudaResults; threads=threads, blocks=blocks)
-
-			# kernels[i] = CuFunction(mod, "ExpressionProcessing")
-		# 	resultCache[expressions[i]] = kernels[i]
-		# catch
-		# 	dump(expressions[i]; maxdepth=10)
-		# 	println()
-		# 	println()
-		# 	println(kernel)
-		# 	println()
-		# 	println()
-		# 	error(current_exceptions())
-		# end
+		linker = CuLink()
+		add_data!(linker, kernelName, kernel)
+		
+		image = complete(linker)
+		mod = CuModule(image)
+		compiledKernel = CuFunction(mod, kernelName)
 
+		cudacall(compiledKernel, (CuPtr{Float32},CuPtr{Float32},CuPtr{Float32}), cudaVars, cudaParams, cudaResults; threads=threads, blocks=blocks)
 	end
 
 	# for kernel in kernels
@@ -107,13 +93,13 @@ end
 - param ```expressionIndex```: The 0-based index of the expression
 "
 function transpile(expression::ExpressionProcessing.PostfixType, varSetSize::Integer, paramSetSize::Integer, 
-				   nrOfVariableSets::Integer, expressionIndex::Integer)::String
+				   nrOfVariableSets::Integer, expressionIndex::Integer, kernelName::String)::String
 	exitJumpLocationMarker = "L__BB0_2"
 	ptxBuffer = IOBuffer()
 	regManager = Utils.RegisterManager(Dict(), Dict())
 
 	# TODO: Suboptimal solution. get_kernel_signature should also return the name of the registers used for the parameters, so further below, we do not have to hard-code them
-	signature, paramLoading = get_kernel_signature("ExpressionProcessing", [Float32, Float32, Float32], regManager) # Vars, Params, Results
+	signature, paramLoading = get_kernel_signature(kernelName, [Float32, Float32, Float32], regManager) # Vars, Params, Results
 	guardClause, threadId64Reg = get_guard_clause(exitJumpLocationMarker, nrOfVariableSets, regManager)
 
 	println(ptxBuffer, get_cuda_header())