Changes

Jump to: navigation, search

Top Solo

9,733 bytes added, 03:24, 8 March 2013
Assignment 2
== Assignment 2 ==
 
This assignment involved GPU programming the functions I had speculated I could process concurrently. After further analysis I determined that the second for loop cannot be processed in parallel without converting some underlying API functions to device functions. These functions are used in multiple areas in the framework so I did not pursue changing the framework too much. The first for loop however contained independent data that could be executed in parallel so I went ahead and created kernels for that code. I also noticed that the function made a call to cimg::rand(). This is obviously a function that returns random numbers. Luckily CUDA also has multiple libraries that perform these functions. Off the top of my head: Thrust and CURAND do. I went with CURAND because it has the word "rand" in it and that's what I need. hehe
=== Kernels ===
<source lang="cpp">
 
 
/*
 
* Setup and initialize curand with a seed
 
*/
 
__global__ void initCurand(curandState* state){
 
int idx = blockIdx.x * blockDim.x + threadIdx.x;
 
curand_init(100, idx, 0, &state[idx]);
 
__syncthreads();
 
}
 
 
 
/*
 
* CUDA kernel that will execute 100 threads in parallel
 
* and will populate these parallel arrays with 100 random numbers
 
* array size = 100.
 
*/
 
 
 
__global__ void initializeArrays(float* posx, float* posy,float* rayon, float* veloc,
 
float* opacity ,float* angle, unsigned char* color, int height,
 
int width, curandState* state, size_t pitch){
 
 
 
int idx = blockIdx.x * blockDim.x + threadIdx.x;
 
curandState localState = state[idx];
 
 
 
posx[idx] = (float)(curand_normal(&localState)*width);
 
posy[idx] = (float)(curand_normal(&localState)*height);
 
rayon[idx] = (float)(10 + curand_normal(&localState)*50);
 
angle[idx] = (float)(curand_normal(&localState)*360);
 
veloc[idx] = (float)(curand_uniform(&localState)*20 - 10);
 
color[idx*pitch] = (unsigned char)(curand_normal(&localState)*255);
 
color[(idx*pitch)+1] = (unsigned char)(curand_normal(&localState)*255);
 
color[(idx*pitch)+2] = (unsigned char)(curand_normal(&localState)*255);
 
opacity[idx] = (float)(0.3f + 1.5f *curand_normal(&localState));
 
 
__syncthreads();
 
}
 
// check for any errors returned by CUDA API functions.
 
void errCheck(cudaError_t err, const char* msg){
 
if (err != cudaSuccess)
 
std::cout<< msg << ": " << cudaGetErrorString(err) << std::endl;
 
}
 
</source>
 
=== Body Code ===
<source lan="cpp">
 
 
// check for any errors returned by CUDA API functions.
 
void errCheck(cudaError_t err, const char* msg){
 
if (err != cudaSuccess)
 
std::cout<< msg << ": " << cudaGetErrorString(err) << std::endl;
 
}
 
 
 
/*---------------------------
 
 
 
Main procedure
 
 
 
--------------------------*/
 
int main() {
 
 
 
// Create a colored 640x480 background image which consists of different color shades.
 
CImg<float> background(640,480,1,3);
 
cimg_forXY(background,x,y) background.fillC(x,y,0,
 
x*std::cos(6.0*y/background.height()) + y*std::sin(9.0*x/background.width()),
 
x*std::sin(8.0*y/background.height()) - y*std::cos(11.0*x/background.width()),
 
x*std::cos(13.0*y/background.height()) - y*std::sin(8.0*x/background.width()));
 
background.normalize(0,180);
 
 
// Init images and create display window.
 
CImg<unsigned char> img0(background), img;
 
unsigned char white[] = { 255, 255, 255 }, color[100][3];
 
CImgDisplay disp(img0,"[#6] - Filled Triangles (Click to shrink)");
 
// error handling
 
cudaError_t err;
 
// Define random properties (pos, size, colors, ..) for all triangles that will be displayed.
 
float posx[100];
 
float posy[100];
 
float rayon[100];
 
float angle[100];
 
float veloc[100];
 
float opacity[100];
 
// Define the same properties but for the device
 
float* d_posx;
 
float* d_posy;
 
float* d_rayon;
 
float* d_angle;
 
float* d_veloc;
 
float* d_opacity;
 
unsigned char* d_color;
 
 
 
// CURAND state
 
curandState* devState;
 
 
 
// allocate memory on the device for the device arrays
 
err = cudaMalloc((void**)&d_posx, 100 * sizeof(float));
 
errCheck(err, "cudaMalloc((void**)&d_posx, 100 * sizeof(float))");
 
err = cudaMalloc((void**)&d_posy, 100 * sizeof(float));
 
errCheck(err,"cudaMalloc((void**)&d_posy, 100 * sizeof(float))");
 
err = cudaMalloc((void**)&d_rayon, 100 * sizeof(float));
 
errCheck(err,"cudaMalloc((void**)&d_rayon, 100 * sizeof(float))");
 
err = cudaMalloc((void**)&d_angle, 100 * sizeof(float));
 
errCheck(err,"cudaMalloc((void**)&d_angle, 100 * sizeof(float))");
 
err = cudaMalloc((void**)&d_veloc, 100 * sizeof(float));
 
errCheck(err,"cudaMalloc((void**)&d_veloc, 100 * sizeof(float))");
 
err = cudaMalloc((void**)&d_opacity, 100 * sizeof(float));
 
errCheck(err,"cudaMalloc((void**)&d_opacity, 100 * sizeof(float))");
 
err = cudaMalloc((void**)&devState, 100*sizeof(curandState));
 
errCheck(err,"cudaMalloc((void**)&devState, 100*sizeof(curandState))");
 
size_t pitch;
 
//allocated the device memory for source array
 
err = cudaMallocPitch(&d_color, &pitch, 3 * sizeof(unsigned char),100);
 
errCheck(err,"cudaMallocPitch(&d_color, &pitch, 3 * sizeof(unsigned char),100)");
 
// launch grid of threads
 
dim3 dimBlock(100);
 
dim3 dimGrid(1);
 
 
/* Kernel for initializing CURAND */
 
initCurand<<<1,100>>>(devState);
 
 
 
// synchronize the device and the host
 
cudaDeviceSynchronize();
 
 
/*Kernel for initializing Arrays */
 
initializeArrays<<<1, 100>>>(d_posx, d_posy, d_rayon, d_veloc, d_opacity, d_angle,
 
d_color, img0.height(), img0.width(), devState, pitch);
 
// synchronize the device and the host
 
cudaDeviceSynchronize();
 
 
// get the populated arrays back to the host for use
 
err = cudaMemcpy(posx,d_posx, 100 * sizeof(float), cudaMemcpyDeviceToHost);
 
errCheck(err,"cudaMemcpy(posx,d_posx, 100 * sizeof(float), cudaMemcpyDeviceToHost)");
 
err = cudaMemcpy(posy,d_posy, 100 * sizeof(float), cudaMemcpyDeviceToHost);
 
errCheck(err,"cudaMemcpy(posy,d_posy, 100 * sizeof(float), cudaMemcpyDeviceToHost)");
 
err = cudaMemcpy(rayon,d_rayon, 100 * sizeof(float), cudaMemcpyDeviceToHost);
 
errCheck(err,"cudaMemcpy(rayon,d_rayon, 100 * sizeof(float), cudaMemcpyDeviceToHost)");
 
err = cudaMemcpy(veloc,d_veloc, 100 * sizeof(float), cudaMemcpyDeviceToHost);
 
errCheck(err,"cudaMemcpy(veloc,d_veloc, 100 * sizeof(float), cudaMemcpyDeviceToHost)");
 
err = cudaMemcpy(opacity,d_opacity, 100 * sizeof(float), cudaMemcpyDeviceToHost);
 
errCheck(err,"cudaMemcpy(opacity,d_opacity, 100 * sizeof(float), cudaMemcpyDeviceToHost)");
 
err = cudaMemcpy(angle,d_angle, 100 * sizeof(float), cudaMemcpyDeviceToHost);
 
errCheck(err,"cudaMemcpy(angle,d_angle, 100 * sizeof(float), cudaMemcpyDeviceToHost)");
 
// pitch of color array is 3+1 padded
 
err = cudaMemcpy2D(color,4,d_color,pitch,3 *sizeof(unsigned char),3, cudaMemcpyDeviceToHost);
 
errCheck(err,"cudaMemcpy2D(color,pitch,d_color,100*3,3 *sizeof(unsigned char),100* sizeof(unsigned char), cudaMemcpyDeviceToHost)");
 
// measuring time it takes for triangle animations in 1000 iterations
 
int i = 0, num = 1;
 
 
// Start animation loop.
 
while (!disp.is_closed() && !disp.is_keyQ() && !disp.is_keyESC() && i < 1000) {
 
img = img0;
 
 
i++;
 
// Draw each triangle on the background image.
 
for (int k = 0; k<num; ++k) {
 
const int
 
x0 = (int)(posx[k] + rayon[k]*std::cos(angle[k]*cimg::PI/180)),
 
y0 = (int)(posy[k] + rayon[k]*std::sin(angle[k]*cimg::PI/180)),
 
x1 = (int)(posx[k] + rayon[k]*std::cos((angle[k] + 120)*cimg::PI/180)),
 
y1 = (int)(posy[k] + rayon[k]*std::sin((angle[k] + 120)*cimg::PI/180)),
 
x2 = (int)(posx[k] + rayon[k]*std::cos((angle[k] + 240)*cimg::PI/180)),
 
y2 = (int)(posy[k] + rayon[k]*std::sin((angle[k] + 240)*cimg::PI/180));
 
if (k%10) img.draw_triangle(x0,y0,x1,y1,x2,y2,color[k],opacity[k]);
 
else img.draw_triangle(x0,y0,x1,y1,x2,y2,img0,0,0,img0.width()-1,0,0,img.height()-1,opacity[k]);
 
img.draw_triangle(x0,y0,x1,y1,x2,y2,white,opacity[k],~0U);
 
 
// Make the triangles rotate, and check for mouse click event.
 
// (to make triangles collapse or join).
 
angle[k]+=veloc[k];
 
if (disp.mouse_x()>0 && disp.mouse_y()>0) {
 
float u = disp.mouse_x() - posx[k], v = disp.mouse_y() - posy[k];
 
if (disp.button()) { u = -u; v = -v; }
 
posx[k]-=0.03f*u, posy[k]-=0.03f*v;
 
if (posx[k]<0 || posx[k]>=img.width()) posx[k] = (float)(cimg::rand()*img.width());
 
if (posy[k]<0 || posy[k]>=img.height()) posy[k] = (float)(cimg::rand()*img.height());
 
}
 
}
 
 
// Display current animation framerate, and refresh display window.
 
img.draw_text(5,5,"%u frames/s",white,0,0.5f,13,(unsigned int)disp.frames_per_second());
 
img0.resize(disp.display(img).resize(false).wait(20));
 
if (++num>100) num = 100;
 
 
// Allow the user to toggle fullscreen mode, by pressing CTRL+F.
 
if (disp.is_keyCTRLLEFT() && disp.is_keyF()) disp.resize(640,480,false).toggle_fullscreen(false);
 
}
 
 
 
// free allocated device memory
 
cudaFree(d_posy);
 
cudaFree(d_posx);
 
cudaFree(d_rayon);
 
cudaFree(d_veloc);
 
cudaFree(d_opacity);
 
cudaFree(d_color);
 
cudaFree(d_angle);
 
cudaFree(devState);
 
return 0;
 
}
 
 
</source>
=== Profiling Results ===
The entirety of the GPU code executes in 0.15 - 0.16 seconds.
== Assignment 3 ==
1
edit

Navigation menu