public CodeGenerator VisitConst(Const @const, CodeGenerator g) { g.IppSet(@const.Value, g.ValueMemory); g.IppSet(0.0, g.GradientMemory); return(g); }
static void Main(string[] args) { using (var env = DynamicLibraries.LoadIpp()) { Console.WriteLine("IPP available"); Console.Write("How many paths? [e.g. 10000]: "); var nPaths = int.Parse(Console.ReadLine(), CultureInfo.InvariantCulture); // working memory var nVariables = 102; //100 normal draws and mu and sigma var valueMem = Marshal.AllocHGlobal(sizeof(double) * nPaths * (2 + nVariables)); var gradientMem = valueMem + sizeof(double) * nPaths; var variablesMem = gradientMem + sizeof(double) * nPaths; var variables = Enumerable.Range(0, nVariables) .Select(i => (i == 0 ? "mu" : i == 1 ? "sigma" : $"x{i - 2}", (ptr: variablesMem + sizeof(double) * nPaths * i, seed: i == 0 ? 1.0 : 0.0))).ToDictionary(t => t.Item1, t => t.Item2); var xValues = new double[nPaths * nVariables]; var rng = new Random(); for (int i = 0; i < nPaths; i++) { xValues[i] = 0.1; // mu xValues[nPaths + i] = 0.5; // sigma xValues[2 * nPaths + i] = 0.0; // x0 } for (int i = nPaths * 3; i < nVariables * nPaths; i += 2) { var normal = Statistics.BoxMuller(rng.NextDouble(), rng.NextDouble()); xValues[i] = normal.x; if (i + 1 < nPaths * nVariables) { xValues[i + 1] = normal.y; } } Marshal.Copy(xValues, 0, variablesMem, nPaths * nVariables); Expr node = new Var("x0"); Expr mu = new Var("mu"); Expr s = new Var("sigma"); double timeGrid = 0.01; Expr dt = new Const(timeGrid); Expr brownianScale = new Const(Math.Sqrt(timeGrid)); for (int i = 0; i < nVariables - 2; i++) { var brownianIncrement = new Mul(new[] { new Var($"x{i}"), brownianScale }); node = new Add(new[] { new Mul(new[] { mu, dt }), new Mul(new[] { s, brownianIncrement }), node }); } var final = new Exp(node); var state = final.Visit(new ForwardAdVisitor(), new CodeGenerator(valueMem, variables, gradientMem, (ulong)nPaths, LlvmBindings.BuildFunctions)); var executor = new LlvmExecutor(state); executor.Run(env.FunctionPointers); Console.WriteLine("Back in main, fetching result."); var result = new double[nPaths]; Marshal.Copy(valueMem, result, 0, nPaths); var gradResult = new double[nPaths]; Marshal.Copy(gradientMem, gradResult, 0, nPaths); Console.WriteLine("E[X(100)] = {0}, expected 1.25", result.Average()); Console.WriteLine("dE[X(100)]/dmu = {0}, expected 1.25", gradResult.Average()); Marshal.FreeHGlobal(valueMem); } }