int main(int argc, char** argv) {
float area[2];
float aTime = 0.0;
float bTime = 0.0;
int reps = argc > 1 ? atoi(argv[1]) :100;
printf("ProgramCount [%d], size-of-float[%lu], size-of-float[%lu]\n",ispc::get_programCount(),sizeof(float), sizeof(float));
for (int i =0; i<reps; i++) {
reset_and_start_timer();
area[0]= ptarray_signed_area_aos(polygon_aos,npoints);
aTime += get_elapsed_mcycles();
}
for (int i =0; i<reps; i++) {
reset_and_start_timer();
area[1]= ispc::ptarray_signed_area_aos(polygon_aos,npoints);
bTime += get_elapsed_mcycles();
}
printf("%-20s: [%.2f] M cycles %s, [%.2f] M cycles %s (%.2fx speedup).\n",
"ST_AREA", aTime, "serial_aos", bTime, "ispc_aos",
aTime/bTime);
printf("%-20s: serial [%.2f], ispc [%.2f]\n", "results", area[0], area[1]);
aTime = 0.0;
bTime = 0.0;
for (int i =0; i<reps; i++) {
reset_and_start_timer();
area[0]= ptarray_signed_area_soa(polygon_soa,npoints);
aTime += get_elapsed_mcycles();
}
printf("sequential code is fine.\n");
for (int i =0; i<reps; i++) {
reset_and_start_timer();
area[1]= ispc::ptarray_signed_area_soa(polygon_aos,npoints);
bTime += get_elapsed_mcycles();
}
printf("%-20s: [%.2f] M cycles %s, [%.2f] M cycles %s (%.2fx speedup).\n",
"ST_AREA", aTime, "serial_soa", bTime, "ispc_soa",
aTime/bTime);
printf("%-20s: serial [%.2f], ispc [%.2f]\n", "results", area[0], area[1]);
return 0;
}
void mandelbrot_threads(float x0, float y0, float x1, float y1,
int width,int width_start, int width_end, int height, int maxIterations,
int output[])
{
float dx = (x1 - x0) / width;
float dy = (y1 - y0) / height;
int flag=1;
reset_and_start_timer();
for (int j = 0; j < height; j++) {
for (int i = width_start; i < width_end; ++i) {
float x = x0 + i * dx;
float y = y0 + j * dy;
int index = (j * width + i);
output[index] = mandel(x, y, maxIterations);
}
}
double dt = get_elapsed_mcycles();
printf("\n\t\t\t Thread took :\t[%.3f] millon cycles",dt);
}
int main(int argc, char ** argv) {
if (argc < 4) {
printf("Please input M, N and K\n");
return 1;
}
int row = atoi(argv[1]);
int col = atoi(argv[2]);
int num_iterate = atoi(argv[3]);
GameOfLife* game;
if (row <= 0 || col <= 0) {
game = new GameOfLife(6, 4);
game->specificInit();
}
else {
game = new GameOfLife(row, col);
game->randomInit();
}
if (game->notTooLarge()) {
game->print();
}
// start to record time consumption
reset_and_start_timer();
game->iterateAll(num_iterate);
// stop timer and print out total cycles
double one_round = get_elapsed_mcycles();
if (game->notTooLarge()) {
game->print();
}
printf("\n-------- Statistic Infomation --------\n\n");
printf("time consumption:\t\t\t[%.3f] million cycles\n", one_round);
delete(game);
return 0;
}
int main(int argc, char** argv) {
if (argc != 2) {
printf("usage: deferred_shading <input_file (e.g. data/pp1280x720.bin)>\n");
return 1;
}
InputData *input = CreateInputDataFromFile(argv[1]);
if (!input) {
printf("Failed to load input file \"%s\"!\n", argv[1]);
return 1;
}
Framebuffer framebuffer(input->header.framebufferWidth,
input->header.framebufferHeight);
InitDynamicC(input);
#ifdef __cilk
InitDynamicCilk(input);
#endif // __cilk
int nframes = 5;
double ispcCycles = 1e30;
for (int i = 0; i < 5; ++i) {
framebuffer.clear();
reset_and_start_timer();
for (int j = 0; j < nframes; ++j)
ispc::RenderStatic(input->header, input->arrays,
VISUALIZE_LIGHT_COUNT,
framebuffer.r, framebuffer.g, framebuffer.b);
double mcycles = get_elapsed_mcycles() / nframes;
ispcCycles = std::min(ispcCycles, mcycles);
}
printf("[ispc static + tasks]:\t\t[%.3f] million cycles to render "
"%d x %d image\n", ispcCycles,
input->header.framebufferWidth, input->header.framebufferHeight);
WriteFrame("deferred-ispc-static.ppm", input, framebuffer);
#ifdef __cilk
double dynamicCilkCycles = 1e30;
for (int i = 0; i < 5; ++i) {
framebuffer.clear();
reset_and_start_timer();
for (int j = 0; j < nframes; ++j)
DispatchDynamicCilk(input, &framebuffer);
double mcycles = get_elapsed_mcycles() / nframes;
dynamicCilkCycles = std::min(dynamicCilkCycles, mcycles);
}
printf("[ispc + Cilk dynamic]:\t\t[%.3f] million cycles to render image\n",
dynamicCilkCycles);
WriteFrame("deferred-ispc-dynamic.ppm", input, framebuffer);
#endif // __cilk
double serialCycles = 1e30;
for (int i = 0; i < 5; ++i) {
framebuffer.clear();
reset_and_start_timer();
for (int j = 0; j < nframes; ++j)
DispatchDynamicC(input, &framebuffer);
double mcycles = get_elapsed_mcycles() / nframes;
serialCycles = std::min(serialCycles, mcycles);
}
printf("[C++ serial dynamic, 1 core]:\t[%.3f] million cycles to render image\n",
serialCycles);
WriteFrame("deferred-serial-dynamic.ppm", input, framebuffer);
#ifdef __cilk
printf("\t\t\t\t(%.2fx speedup from static ISPC, %.2fx from Cilk+ISPC)\n",
serialCycles/ispcCycles, serialCycles/dynamicCilkCycles);
#else
printf("\t\t\t\t(%.2fx speedup from ISPC + tasks)\n", serialCycles/ispcCycles);
#endif // __cilk
DeleteInputData(input);
return 0;
}
