/** Print a table of metrics for the given data. */
private static void printData(Data data) {
System.out.println("n\tK\tT\tSpdup\tEffic\tEDSF");
System.out.println(
data.n
+ "\t"
+ FMT_0.format(data.K_series.x(0))
+ "\t"
+ FMT_0.format(data.T_series.x(0))
+ "\t"
+ FMT_3.format(data.Speedup_series.x(0))
+ "\t"
+ FMT_3.format(data.Eff_series.x(0)));
for (int i = 1; i < data.K_series.length(); ++i) {
System.out.println(
data.n
+ "\t"
+ FMT_0.format(data.K_series.x(i))
+ "\t"
+ FMT_0.format(data.T_series.x(i))
+ "\t"
+ FMT_3.format(data.Speedup_series.x(i))
+ "\t"
+ FMT_3.format(data.Eff_series.x(i))
+ "\t"
+ FMT_3.format(data.EDSF_series_2.x(i - 1)));
}
}
/** Plot the given data. */
private static void plotData(Data data, String label) {
int len = data.K_series.length();
T_plot.xySeries(new AggregateXYSeries(data.K_series, data.T_series))
.label(label, data.K_series.x(len - 1), data.T_series.x(len - 1));
Speedup_plot.xySeries(new AggregateXYSeries(data.K_series, data.Speedup_series))
.label(label, data.K_series.x(len - 1), data.Speedup_series.x(len - 1));
Eff_plot.xySeries(new AggregateXYSeries(data.K_series, data.Eff_series))
.label(label, data.K_series.x(len - 1), data.Eff_series.x(len - 1));
EDSF_plot.xySeries(new AggregateXYSeries(data.K_series_2, data.EDSF_series_2))
.label(label, data.K_series_2.x(len - 2), data.EDSF_series_2.x(len - 2));
}
/** Main program. */
public static void main(String[] args) throws Exception {
// Parse command line arguments.
if (args.length < 3) usage();
inputfile = new File(args[0]);
nplot = Integer.parseInt(args[1]);
for (int i = 2; i < args.length; ++i) {
basis.add(parseBasisFunction(args[i]));
}
// Set up plots with default attributes.
T_plot.plotTitle("n = " + nplot)
.rightMargin(72)
.minorGridLines(true)
.xAxisKind(Plot.LOGARITHMIC)
.xAxisMinorDivisions(10)
.xAxisTitle("Processors, K")
.yAxisKind(Plot.LOGARITHMIC)
.yAxisMinorDivisions(10)
.yAxisTickFormat(FMT_0E)
.yAxisTickScale(1000)
.yAxisTitle("Running Time, T (sec)")
.labelPosition(Plot.RIGHT)
.labelOffset(6);
Speedup_plot.plotTitle("n = " + nplot)
.rightMargin(72)
.xAxisStart(0)
.xAxisTitle("Processors, K")
.yAxisStart(0)
.yAxisTitle("Speedup")
.labelPosition(Plot.RIGHT)
.labelOffset(6);
Eff_plot.plotTitle("n = " + nplot)
.rightMargin(72)
.xAxisStart(0)
.xAxisTitle("Processors, K")
.yAxisStart(0)
.yAxisTickFormat(FMT_1)
.yAxisTitle("Efficiency")
.labelPosition(Plot.RIGHT)
.labelOffset(6);
EDSF_plot.plotTitle("n = " + nplot)
.rightMargin(72)
.xAxisStart(0)
.xAxisTitle("Processors, K")
.yAxisStart(0)
.yAxisTickFormat(FMT_0)
.yAxisTickScale(0.001)
.yAxisTitle("Sequential Fraction, F (/1000)")
.labelPosition(Plot.RIGHT)
.labelOffset(6);
// Parse the input file.
Scanner scanner = new Scanner(inputfile);
int linenum = 1;
while (scanner.hasNextLine()) {
String line = scanner.nextLine();
int i = line.indexOf('#');
if (i >= 0) line = line.substring(0, i);
line = line.trim();
if (line.length() > 0) parseLine(line, linenum);
++linenum;
}
// Validate data.
for (Data data : dataMap.values()) {
validateData(data);
}
// Set up and solve nonnegative least squares problem.
ListSeries n_series = new ListSeries();
ListSeries K_series = new ListSeries();
ListSeries T_series = new ListSeries();
for (Data data : dataMap.values()) {
double n = data.n;
for (int i = 0; i < data.K_series.length(); ++i) {
double K = data.K_series.x(i);
double T = data.T_series.x(i);
if (K >= 1) {
n_series.add(n);
K_series.add(K);
T_series.add(T);
}
}
}
int M = n_series.length();
int P = basis.size();
NonNegativeLeastSquares nnls = new NonNegativeLeastSquares(M, P);
for (int i = 0; i < M; ++i) {
double n = n_series.x(i);
double K = K_series.x(i);
double T = T_series.x(i);
for (int j = 0; j < P; ++j) {
nnls.a[i][j] = basis.get(j).f(n, K);
}
nnls.b[i] = T;
}
nnls.solve();
// Get actual data for n = nplot.
Data actualData = dataMap.get(nplot);
if (actualData == null) {
System.err.println("No data for n = " + nplot);
System.exit(1);
}
// Set up model data for n = nplot.
Data modelData = new Data(nplot);
for (int i = 0; i < actualData.K_series.length(); ++i) {
double K = actualData.K_series.x(i);
double T = modelFunction(nplot, K, nnls.x);
if (K == 1) modelData.T_par_1 = T;
double Speedup = modelData.T_par_1 / T;
double Eff = Speedup / K;
modelData.K_series.add(K);
modelData.T_series.add(T);
modelData.Speedup_series.add(Speedup);
modelData.Eff_series.add(Eff);
if (K >= 2) {
double EDSF = (K * T - modelData.T_par_1) / modelData.T_par_1 / (K - 1);
modelData.K_series_2.add(K);
modelData.EDSF_series_2.add(EDSF);
}
}
// Print data.
System.out.println("Actual");
printData(actualData);
System.out.println("Model");
printData(modelData);
// Print model function and chi^2.
System.out.print("T(n,K)");
for (int i = 0; i < P; ++i) {
System.out.print(i == 0 ? " = " : " + ");
System.out.print(basis.get(i).toString(nnls.x[i]));
}
System.out.println();
System.out.println("chi^2 = " + nnls.normsqr);
// Add ideal performance lines to plots.
Speedup_plot.seriesDots(null)
.seriesColor(new Color(0.7f, 0.7f, 0.7f))
.xySeries(new double[] {0, K_max}, new double[] {0, K_max});
Eff_plot.seriesDots(null)
.seriesColor(new Color(0.7f, 0.7f, 0.7f))
.xySeries(new double[] {0, K_max}, new double[] {1, 1});
// Add model data to plots.
T_plot.seriesDots(null).seriesColor(Color.RED).labelColor(Color.RED);
Speedup_plot.seriesDots(null).seriesColor(Color.RED).labelColor(Color.RED);
Eff_plot.seriesDots(null).seriesColor(Color.RED).labelColor(Color.RED);
EDSF_plot.seriesDots(null).seriesColor(Color.RED).labelColor(Color.RED);
plotData(modelData, "Model");
// Add actual data to plots.
T_plot.seriesDots(Dots.circle(5)).seriesColor(Color.BLACK).labelColor(Color.BLACK);
Speedup_plot.seriesDots(Dots.circle(5)).seriesColor(Color.BLACK).labelColor(Color.BLACK);
Eff_plot.seriesDots(Dots.circle(5)).seriesColor(Color.BLACK).labelColor(Color.BLACK);
EDSF_plot.seriesDots(Dots.circle(5)).seriesColor(Color.BLACK).labelColor(Color.BLACK);
plotData(actualData, "Actual");
// Display plots.
T_plot.getFrame().setVisible(true);
Speedup_plot.getFrame().setVisible(true);
Eff_plot.getFrame().setVisible(true);
EDSF_plot.getFrame().setVisible(true);
}
