benchmarking: tuned interpreter blocksize

package/test/PerformanceTuning.jl

@@ -1,30 +1,38 @@
 using CUDA
+using DelimitedFiles
+using GZip
 
 using .Transpiler
 using .Interpreter
 
-varsets_medium = 10000
-X = randn(Float32, 5, varsets_medium)
+include("parser.jl") # to parse expressions from a file
 
-exprsGPU = [
-	# CPU interpreter requires an anonymous function and array ref s
-	:(p1 * x1 + p2), # 5 op
-	:((((x1 + x2) + x3) + x4) + x5), # 9 op
-	:(log(abs(x1))), # 3 op
-	:(powabs(p2 - powabs(p1 + x1, 1/x1),p3)) # 13 op
-] # 30 op
 
-# p is the same for CPU and GPU
-p = [randn(Float32, 10) for _ in 1:length(exprsGPU)] # generate 10 random parameter values for each expr
+data,varnames = readdlm("data/nikuradse_1.csv", ',', header=true);
+X = permutedims(convert(Matrix{Float32}, data))
+
+exprs = Expr[]
+parameters = Vector{Vector{Float32}}()
+varnames = ["x$i" for i in 1:10]
+paramnames = ["p$i" for i in 1:20]
+# data/esr_nvar2_len10.txt.gz_9.txt.gz has  ~250_000 exprs
+# data/esr_nvar2_len10.txt.gz_10.txt.gz has ~800_000 exrps
+GZip.open("data/esr_nvar2_len10.txt.gz_3.txt.gz") do io 
+	for line in eachline(io)
+		expr, p = parse_infix(line, varnames, paramnames)
+
+		push!(exprs, expr)
+		push!(parameters, randn(Float32, length(p)))
+	end
+end
 expr_reps = 1
 
 
-
 @testset "Interpreter Tuning" begin
-    CUDA.@profile interpret_gpu(exprsGPU, X, p; repetitions=expr_reps)
+    # CUDA.@profile interpret_gpu(exprs, X, parameters; repetitions=expr_reps)
 end
 
 
 @testset "Transpiler Tuning" begin
-    CUDA.@profile evaluate_gpu(exprsGPU, X, p; repetitions=expr_reps)
+    CUDA.@profile evaluate_gpu(exprs, X, parameters; repetitions=expr_reps)
 end