// Smoothing for dimension 2.
private static void smooth2(float sigma, float[][] s, float[][] x) {
if (sigma < 1.0f) return;
float c = 0.5f * sigma * sigma;
int n1 = x[0].length;
int n2 = x.length;
float[] st = fillfloat(1.0f, n2);
float[] xt = zerofloat(n2);
float[] yt = zerofloat(n2);
LocalSmoothingFilter lsf = new LocalSmoothingFilter();
for (int i1 = 0; i1 < n1; ++i1) {
if (s != null) {
for (int i2 = 0; i2 < n2; ++i2) st[i2] = s[i2][i1];
}
for (int i2 = 0; i2 < n2; ++i2) xt[i2] = x[i2][i1];
lsf.apply(c, st, xt, yt);
for (int i2 = 0; i2 < n2; ++i2) x[i2][i1] = yt[i2];
}
}
public static void main(String args[]) throws FileNotFoundException {
Random rand = new Random();
Scanner in = new Scanner(System.in);
Pointer phdata = new Pointer();
Stopwatch sw = new Stopwatch();
int n1 = 2;
int n2 = n1 + 2; // MUST be even
int size = n1 * n2;
String fileName;
String sourceStr;
float e11, e12, e22;
float r00, r01, r10, r11;
float rs, ra, rb, sa, sb;
float r1, r2, s1, s2;
float alpha = 18.0f;
String kernelNames[] = {"setMem", "soSmoothingNew"};
sourceStr = CLUtil.readFile("sosmoothing.cl");
CLUtil.clInit(sourceStr, kernelNames);
float[][] h_ri = new float[n1][n2];
float[][] h_si = new float[n1][n2];
float[][] si = new float[n1][n2];
float[][] h_d11i = new float[n1][n2];
float[][] h_d12i = new float[n1][n2];
float[][] h_d22i = new float[n1][n2];
for (int i = 0; i < n1; i++) {
for (int j = 0; j < n2; j++) {
h_ri[i][j] = 1.0f;
h_si[i][j] = 1.0f;
si[i][j] = 1.0f;
if (j >= 1 && i >= 1) {
h_d11i[i][j] = 1.0f;
h_d12i[i][j] = 0.0f;
h_d22i[i][j] = 1.0f;
}
}
}
float[] h_ri1 = new float[size];
float[] h_si1 = new float[size];
float[] h_d11i_1 = new float[size];
float[] h_d12i_1 = new float[size];
float[] h_d22i_1 = new float[size];
CLUtil.packArray(n1, n2, h_ri, h_ri1);
CLUtil.packArray(n1, n2, h_si, h_si1);
CLUtil.packArray(n1, n2, h_d11i, h_d11i_1);
CLUtil.packArray(n1, n2, h_d12i, h_d12i_1);
CLUtil.packArray(n1, n2, h_d22i, h_d22i_1);
for (int i2 = 1; i2 < n1; ++i2) {
for (int i1 = 1; i1 < n2; ++i1) {
e11 = alpha * h_d11i[i2][i1]; // fill matrices here with random numbers
e12 = alpha * h_d12i[i2][i1]; // fill D matrix with random numbers
e22 = alpha * h_d22i[i2][i1];
r00 = h_ri[i2][i1];
r01 = h_ri[i2][i1 - 1];
r10 = h_ri[i2 - 1][i1];
r11 = h_ri[i2 - 1][i1 - 1];
rs = 0.25f * (r00 + r01 + r10 + r11);
ra = r00 - r11;
rb = r01 - r10;
r1 = ra - rb;
r2 = ra + rb;
s1 = e11 * r1 + e12 * r2;
s2 = e12 * r1 + e22 * r2;
sa = s1 + s2;
sb = s1 - s2;
si[i2][i1] += sa;
si[i2][i1 - 1] -= sb;
si[i2 - 1][i1] += sb;
si[i2 - 1][i1 - 1] -= sa;
}
}
cl_mem d_s, d_r, d_d11, d_d12, d_d22;
d_r = CLUtil.createGPUBuffer(size, "r");
d_d11 = CLUtil.createGPUBuffer(size, "r");
d_d12 = CLUtil.createGPUBuffer(size, "r");
d_d22 = CLUtil.createGPUBuffer(size, "r");
d_s = CLUtil.createGPUBuffer((n1 + 1) * (n2 + 1), "rw");
CLUtil.copyToBuffer(h_ri1, d_r, size);
CLUtil.copyToBuffer(h_d11i_1, d_d11, size);
CLUtil.copyToBuffer(h_d12i_1, d_d12, size);
CLUtil.copyToBuffer(h_d22i_1, d_d22, size);
CLUtil.setKernelArg(CLUtil.kernels[0], d_s, 0);
CLUtil.setKernelArg(CLUtil.kernels[0], 0.0f, 1);
CLUtil.setKernelArg(CLUtil.kernels[0], size, 2);
long[] oned_global_work_size = new long[] {size};
CLUtil.executeKernel(CLUtil.kernels[0], 1, oned_global_work_size);
long[] local_group_size = new long[] {32, 32};
long[] mapped_n1 = new long[] {n2 / 2};
long[] mapped_n2 = new long[] {n1 / 2};
long[] global_group_size_block =
new long[] {
(long) Math.ceil(mapped_n1[0] / local_group_size[0] + 1) * local_group_size[0],
(long) Math.ceil(mapped_n2[0] / local_group_size[0] + 1) * local_group_size[1]
};
CLUtil.setKernelArg(CLUtil.kernels[1], d_r, 0);
CLUtil.setKernelArg(CLUtil.kernels[1], d_d11, 1);
CLUtil.setKernelArg(CLUtil.kernels[1], d_d12, 2);
CLUtil.setKernelArg(CLUtil.kernels[1], d_d22, 3);
CLUtil.setKernelArg(CLUtil.kernels[1], d_s, 4);
CLUtil.setKernelArg(CLUtil.kernels[1], alpha, 5);
CLUtil.setKernelArg(CLUtil.kernels[1], n1, 6);
CLUtil.setKernelArg(CLUtil.kernels[1], n2, 7);
for (int offsetx = 0; offsetx < 2; ++offsetx) {
for (int offsety = 0; offsety < 2; ++offsety) {
CLUtil.setKernelArg(CLUtil.kernels[1], offsetx, 8);
CLUtil.setKernelArg(CLUtil.kernels[1], offsety, 9);
CLUtil.executeKernel(CLUtil.kernels[1], 2, global_group_size_block, local_group_size);
}
}
CLUtil.readFromBuffer(d_s, h_si1, (n1 + 1) * (n2 + 1));
float[] test = new float[n1 * n2];
LocalSmoothingFilter.saxpy1(n1, n2, -1.0f, h_si1, test);
clReleaseMemObject(d_r);
clReleaseMemObject(d_s);
clReleaseMemObject(d_d11);
clReleaseMemObject(d_d12);
clReleaseMemObject(d_d22);
CLUtil.clRelease();
ArrayMath.dump(si);
ArrayMath.dump(h_si);
CLUtil.unPackArray(n1, n2, h_si1, h_si);
float[][] diffArray;
diffArray = ArrayMath.sub(si, h_si);
PlotPanel panel1 = new PlotPanel(1, 1);
PlotPanel panel2 = new PlotPanel(1, 1);
PlotPanel panel3 = new PlotPanel(1, 1);
PlotPanel panel4 = new PlotPanel(1, 1);
PixelsView npix1 = panel1.addPixels(si);
panel1.addTitle("CPU Output");
panel1.addColorBar();
PixelsView npix2 = panel2.addPixels(h_si);
panel2.addTitle("GPU Output");
panel2.addColorBar();
PixelsView npix3 = panel3.addPixels(diffArray);
npix3.setClips(-0.0001f, 0.0001f);
panel3.addTitle("Difference Between Images");
panel3.addColorBar();
PixelsView npix4 = panel4.addPixels(h_ri);
panel4.addTitle("Input Image");
panel4.addColorBar();
//
// PixelsView pix1 = panel.addPixels(0,0,s);
// PixelsView pix2 = panel.addPixels(1,0,h_s);
// PixelsView pix3 = panel.addPixels(2,0,diffArray);
// panel.addColorBar();
// PlotFrame frame = new PlotFrame(panel);
PlotFrame frame1 = new PlotFrame(panel1);
PlotFrame frame2 = new PlotFrame(panel2);
PlotFrame frame3 = new PlotFrame(panel3);
PlotFrame frame4 = new PlotFrame(panel4);
// frame.setVisible(true);
frame1.setVisible(true);
frame2.setVisible(true);
frame3.setVisible(true);
frame4.setVisible(true);
}