benchmarking: further tests done. Seems like transpiler takes ages, need to investigate further

package/src/ExpressionExecutorCuda.jl

@@ -34,6 +34,7 @@ function interpret_gpu(exprs::Vector{Expr}, X::Matrix{Float32}, p::Vector{Vector
 
 	for i in 1:repetitions # Simulate parameter tuning -> local search (X remains the same, p gets changed in small steps and must be performed sequentially, which it is with this impl)
 		results = Interpreter.interpret(exprs, X, p)
+		println("got results")
 	end
 
 	return results

package/src/ExpressionProcessing.jl

@@ -27,15 +27,17 @@ NOTE: This function is not thread save, especially cache access is not thread sa
 function expr_to_postfix(expression::Expr)::PostfixType
 	expr = expression
 	if expression.head === :->
-		if typeof(expression.args[2]) == Float64
-			println()
-			println("Expression: $expression")
-			println("Expr: $expr")
-			println()
-			dump(expression; maxdepth=10)
-		end
+		# if typeof(expression.args[2]) == Float64
+		# 	println()
+		# 	println("Expression: $expression")
+		# 	println("Expr: $expr")
+		# 	println()
+		# 	dump(expression; maxdepth=10)
+		# end
 		# if the expression equals (x, p) -> (...) then the below statement extracts the expression to evaluate
-		if expression.args[2].head == :block # expressions that are not generated with the parser (./test/parser.jl) contain this extra "block" node, which needs to be skipped
+		if typeof(expression.args[2]) == Float64
+			return [convert_to_ExpressionElement(expression.args[2])]
+		elseif expression.args[2].head == :block # expressions that are not generated with the parser (./test/parser.jl) contain this extra "block" node, which needs to be skipped
 			expr = expression.args[2].args[2] 
 		else # ... if the are generated with the parser, this node is not present and therefore doesn't need to be skipped
 			expr = expression.args[2]
@@ -48,8 +50,6 @@ function expr_to_postfix(expression::Expr)::PostfixType
 
 	postfix = PostfixType()	
 
-	
-
 	# Special handling in the case where the expression is an array access
 	# This can happen if the token is a variable/parameter of the form x[n]/p[n]
 	if expr.head == :ref

package/src/Interpreter.jl

@@ -30,7 +30,7 @@ function interpret(expressions::Vector{Expr}, variables::Matrix{Float32}, parame
 	cudaResults = CuArray{Float32}(undef, variableCols, length(exprs))
 
 	# Start kernel for each expression to ensure that no warp is working on different expressions
-	@inbounds for i in eachindex(exprs)
+	@inbounds Threads.@threads for i in eachindex(exprs)
 		numThreads = min(variableCols, 256)
 		numBlocks = cld(variableCols, numThreads)
 

package/src/Transpiler.jl

@@ -157,7 +157,7 @@ function get_kernel_signature(kernelName::String, parameters::Vector{DataType},
 	println(signatureBuffer, "(")
 	
 	for i in eachindex(parameters)
-		print(signatureBuffer, "  .param .u64", " ", "param_", i)
+		print(signatureBuffer, "  .param .u64 param_", i)
 
 		parametersLocation = Utils.get_next_free_register(regManager, "rd")
 		println(paramLoadingBuffer, "ld.param.u64   $parametersLocation, [param_$i];")

package/test/PerformanceTests.jl

@@ -63,11 +63,11 @@ if compareWithCPU
 end
 
 # cacheInterpreter = Dict{Expr, PostfixType}()
-# suite["GPUI"]["nikuradse_1"] = @benchmarkable interpret_gpu(exprs, X_t, parameters; repetitions=expr_reps)
+suite["GPUI"]["nikuradse_1"] = @benchmarkable interpret_gpu(exprs, X_t, parameters; repetitions=expr_reps)
 
 # cacheTranspilerFront = Dict{Expr, PostfixType}()
 # cacheTranspilerRes = Dict{Expr, CuFunction}()
-suite["GPUT"]["nikuradse_1"] = @benchmarkable evaluate_gpu(exprs, X_t, parameters; repetitions=expr_reps)
+suite["GPUT"]["nikuradse_1"] = @benchmarkable evaluate_gpu(exprs, X_t, parameters; repetitions=expr_reps) # Takes forever. Needs more investigation
 
 tune!(suite)
 BenchmarkTools.save("params.json", params(suite))

package/test/params.json

@@ -1 +1 @@
-[{"Julia":"1.11.4","BenchmarkTools":{"major":1,"minor":6,"patch":0,"prerelease":[],"build":[]}},[["BenchmarkGroup",{"data":{"CPU":["BenchmarkGroup",{"data":{"medium varset":["Parameters",{"gctrial":true,"time_tolerance":0.05,"evals_set":false,"samples":1000,"evals":1,"gcsample":false,"seconds":5.0,"overhead":0.0,"memory_tolerance":0.01}],"large varset":["Parameters",{"gctrial":true,"time_tolerance":0.05,"evals_set":false,"samples":1000,"evals":1,"gcsample":false,"seconds":5.0,"overhead":0.0,"memory_tolerance":0.01}],"small varset":["Parameters",{"gctrial":true,"time_tolerance":0.05,"evals_set":false,"samples":1000,"evals":1,"gcsample":false,"seconds":5.0,"overhead":0.0,"memory_tolerance":0.01}]},"tags":["CPUInterpreter"]}],"GPUT":["BenchmarkGroup",{"data":{"medium varset":["Parameters",{"gctrial":true,"time_tolerance":0.05,"evals_set":false,"samples":1000,"evals":1,"gcsample":false,"seconds":5.0,"overhead":0.0,"memory_tolerance":0.01}],"large varset":["Parameters",{"gctrial":true,"time_tolerance":0.05,"evals_set":false,"samples":1000,"evals":1,"gcsample":false,"seconds":5.0,"overhead":0.0,"memory_tolerance":0.01}],"small varset":["Parameters",{"gctrial":true,"time_tolerance":0.05,"evals_set":false,"samples":1000,"evals":1,"gcsample":false,"seconds":5.0,"overhead":0.0,"memory_tolerance":0.01}]},"tags":["GPUTranspiler"]}],"GPUI":["BenchmarkGroup",{"data":{"medium varset":["Parameters",{"gctrial":true,"time_tolerance":0.05,"evals_set":false,"samples":1000,"evals":1,"gcsample":false,"seconds":5.0,"overhead":0.0,"memory_tolerance":0.01}],"large varset":["Parameters",{"gctrial":true,"time_tolerance":0.05,"evals_set":false,"samples":1000,"evals":1,"gcsample":false,"seconds":5.0,"overhead":0.0,"memory_tolerance":0.01}],"small varset":["Parameters",{"gctrial":true,"time_tolerance":0.05,"evals_set":false,"samples":1000,"evals":1,"gcsample":false,"seconds":5.0,"overhead":0.0,"memory_tolerance":0.01}]},"tags":["GPUInterpreter"]}]},"tags":[]}]]]
+[{"Julia":"1.11.5","BenchmarkTools":{"major":1,"minor":6,"patch":0,"prerelease":[],"build":[]}},[["BenchmarkGroup",{"data":{"CPU":["BenchmarkGroup",{"data":{},"tags":["CPUInterpreter"]}],"GPUT":["BenchmarkGroup",{"data":{},"tags":["GPUTranspiler"]}],"GPUI":["BenchmarkGroup",{"data":{"nikuradse_1":["Parameters",{"gctrial":true,"time_tolerance":0.05,"evals_set":false,"samples":10000,"evals":1,"gcsample":false,"seconds":5.0,"overhead":0.0,"memory_tolerance":0.01}]},"tags":["GPUInterpreter"]}]},"tags":[]}]]]