static void ReconstructAndSave(string fileName, MyImageReader image, Evaluation evaluation)
{
var tileOrders = new byte[1, 9];
TileOrderFromPermutation(Permutations, evaluation.Permutation, 9, tileOrders, 0);
var tileOrder = new byte[9];
for (var i = 0; i < 9; i++)
{
tileOrder[i] = tileOrders[0, i];
}
var pixels = new uint[image.Width, image.Height];
var cxTile = image.Width / 3;
var cyTile = image.Height / 3;
for (var destinationTile = 0; destinationTile < 9; destinationTile++)
{
var destTileX = destinationTile % 3;
var destTileY = destinationTile / 3;
var sourceTile = tileOrder[destinationTile];
var sourceTileX = sourceTile % 3;
var sourceTileY = sourceTile / 3;
for (var x = 0; x < cxTile; x++)
{
for (var y = 0; y < cyTile; y++)
{
pixels[destTileX * cxTile + x, destTileY * cyTile + y] = image.Pixels[sourceTileX * cxTile + x + (sourceTileY * cyTile + y) * image.Width];
}
}
}
MyImageWriter.Write(fileName, pixels);
}
static void TimeCpuEnhancedScaleImage(MyImageReader cpuImage)
{
var scaledImage = new byte[64 * 3, 64 * 3];
var sourceImage = cpuImage.Pixels.Clone() as uint[];
var sourceWidth = cpuImage.Width;
var sourceHeight = cpuImage.Height;
var stopWatch = Stopwatch.StartNew();
Parallel.For(0, 192 * 192, p =>
{
var x = p % 192;
var y = p / 192;
EnhancedScaleImagePixel(sourceImage, sourceWidth, sourceHeight, scaledImage, x, y);
});
stopWatch.Stop();
CpuScaleImageTimeMs = (float)stopWatch.Elapsed.TotalMilliseconds;
}
public static void Solve(string sourceFileName, string destinationFileName)
{
// load image from disk
var cpuImage = new MyImageReader(sourceFileName);
// allocate image memory on the GPU
var gpuImage = Gpu.Allocate<uint>(cpuImage.Pixels.Length);
// copy the image pixels to the GPU
Gpu.StartTimer();
Gpu.CopyToDevice(cpuImage.Pixels, gpuImage);
CopyImageTimeMs = Gpu.StopTimer();
// allocate scaled image memory on the GPU
var gpuScaledImage = Gpu.Allocate<byte>(64 * 3, 64 * 3);
// rescale the image using the GPU
Gpu.StartTimer();
Gpu.Launch(new dim3(12, 12), new dim3(16, 16), EnhancedScaleImageKernel, gpuImage, cpuImage.Width, cpuImage.Height, gpuScaledImage);
Gpu.Synchronize();
ScaleImageTimeMs = Gpu.StopTimer();
// save a copy of the scaled image
SaveScaledImage("Scaled.bmp", gpuScaledImage);
// compare with CPU scale image time
TimeCpuEnhancedScaleImage(cpuImage);
// allocate edges memory on the GPU
var gpuEdges = Gpu.Allocate<byte>(9, 4, 64);
// extract edge information using the GPU
Gpu.StartTimer();
Gpu.Launch(new dim3(9, 4), 64, ExtractEdgeKernel, gpuScaledImage, gpuEdges);
Gpu.Synchronize();
ExtractEdgeTimeMs = Gpu.StopTimer();
// copy edges to GPU constant memory
Gpu.StartTimer();
Gpu.CopyFromDevice(gpuEdges, Edges);
Gpu.CopyToConstantMemory(Edges, Edges);
CopyEdgesToConstantMemoryTimeMs = Gpu.StopTimer();
// allocate fit memory on the GPU
var gpuFit = Gpu.Allocate<float>(9, 9);
// compare edge fitting using the GPU
Gpu.StartTimer();
Gpu.Launch(new dim3(9, 9), 64, ComputeFitsKernel, 2, 0, gpuFit);
Gpu.Synchronize();
ComputeFitsTimeMs = Gpu.StopTimer();
// copy edges to GPU constant memory
Gpu.StartTimer();
Gpu.CopyFromDevice(gpuFit, LeftRightFit);
Gpu.CopyToConstantMemory(LeftRightFit, LeftRightFit);
CopyFitsToConstantMemoryTimeMs = Gpu.StopTimer();
// compare edge fitting using the GPU
Gpu.StartTimer();
Gpu.Launch(new dim3(9, 9), 64, ComputeFitsKernel, 3, 1, gpuFit);
Gpu.Synchronize();
ComputeFitsTimeMs += Gpu.StopTimer();
// copy edges to GPU constant memory
Gpu.StartTimer();
Gpu.CopyFromDevice(gpuFit, TopBottomFit);
Gpu.CopyToConstantMemory(TopBottomFit, TopBottomFit);
CopyFitsToConstantMemoryTimeMs += Gpu.StopTimer();
// evaluate all permutations
const int threads = 256;
const int blocks = Permutations / threads + 1;
var cpuEvaluations = new Evaluation[blocks];
var gpuEvaluations = Gpu.Allocate(cpuEvaluations);
Gpu.StartTimer();
Gpu.Launch(blocks, threads, ExplorePermutationsKernel, gpuEvaluations);
Gpu.Synchronize();
ExplorePermutationsTimeMs += Gpu.StopTimer();
Gpu.StartTimer();
Gpu.CopyFromDevice(gpuEvaluations, cpuEvaluations);
CopyPermutationsTimeMs += Gpu.StopTimer();
// get the best permutation
var bestEvaluation = cpuEvaluations[0];
foreach (var evaluation in cpuEvaluations)
{
if (evaluation.Metric < bestEvaluation.Metric)
{
bestEvaluation = evaluation;
}
}
// reconstruct and save the new image
ReconstructAndSave(destinationFileName, cpuImage, bestEvaluation);
// free memory on the GPU
Gpu.Free(gpuImage);
Gpu.Free(gpuScaledImage);
Gpu.Free(gpuFit);
Gpu.Free(gpuEdges);
Gpu.Free(gpuEvaluations);
}