Made changes to CMakeLists.txt so CUDA version of program builds when CUDA is installed.

Replaced C-style approach to generating texture to OOP-style with Canvas.h

1 files changed, 52 insertions, 0 deletions

diff --git a/Canvas/gpu/CalculateCanvas.cu b/Canvas/gpu/CalculateCanvas.cu
new file mode 100644
index 0000000..3a73067
--- /dev/null
+++ b/Canvas/gpu/CalculateCanvas.cu
@@ -0,0 +1,52 @@
+#include "CalculateCanvas.h"
+
+__global__
+void calculate(unsigned char* canvas, int width, Point* points, int nPoints)
+{
+    int canvasX = blockIdx.x * blockDim.x + threadIdx.x;
+    int canvasY = blockIdx.y * blockDim.y + threadIdx.y;
+
+    int pathX, pathY;
+    int length_squared;
+    double f = 0;
+
+    for (int i = 0; i < nPoints; i++)
+    {
+        pathX = points[i].x - canvasX;
+        pathY = points[i].y - canvasY;
+
+        length_squared = pathX * pathX + pathY * pathY;
+        f += (1e5 * 20) / (float) length_squared;
+    }
+
+    int l = (int) f;
+    if (l > 255) l = 255;
+
+    canvas[canvasY * width * 3 + canvasX * 3 + 0] = l;
+    canvas[canvasY * width * 3 + canvasX * 3 + 1] = 0;
+    canvas[canvasY * width * 3 + canvasX * 3 + 2] = 0;
+}
+
+void generate_canvas(int width, int height, unsigned char* canvas, Point* points, int nPoints)
+{
+    const int canvasSize = width * height * 3 * sizeof(unsigned char);
+    const int pointsSize = nPoints * sizeof(Point);
+
+    unsigned char* gpuCanvas;
+    Point* gpuPoints;
+
+    cudaMalloc((void**) &gpuCanvas, canvasSize);
+    cudaMalloc((void**) &gpuPoints, pointsSize);
+
+    cudaMemcpy(gpuCanvas, canvas, canvasSize, cudaMemcpyHostToDevice);
+    cudaMemcpy(gpuPoints, points, pointsSize, cudaMemcpyHostToDevice);
+
+    dim3 threadsPerBlock(16, 16);
+    dim3 numBlocks(width / threadsPerBlock.x, height / threadsPerBlock.y);
+    calculate<<<numBlocks, threadsPerBlock>>>(gpuCanvas, width, gpuPoints, nPoints);
+
+    cudaMemcpy(canvas, gpuCanvas, canvasSize, cudaMemcpyDeviceToHost);
+
+    cudaFree(gpuCanvas);
+    cudaFree(gpuPoints);
+}