文件列表:

Code/
Input/
Installation/
Lecture/
Output/

# GPU Programming ## Description: This CUDA C program is designed for parallel computation on GPUs and appears to perform calculations related to astronomy or galaxy simulations. The primary goal seems to be computing the angular correlation function for real and simulated galaxies. The program uses three CUDA kernels (calculate_faster_angle_DR, calculate_faster_angle_DD, and calculate_faster_angle_RR) to efficiently handle the computation in parallel. The code initializes and reads real and simulated galaxy data from input files, computes sine and cosine values for the angles on both the CPU and GPU, and then launches GPU kernels to calculate the angular correlation function. The results are copied back to the host, and the program prints relevant information, including the calculated angular correlation function, histograms, and performance metrics. ## Recommended background: Programming experience in C/C++, although some have started with very little C knowledge. ## Installation When Dione opens, load the modules ```shell module load cuda module load GCC ``` ## Compile You can test using the template of the project ```shell srun -p gpu -n 1 -t 10:00 --mem=1G -e err.txt -o out.txt galaxy data_100k_arcmin.dat rand_100k_arcmin.dat omega.out ``` ## Credit Your basic design is very clear: to calculate sines and cosines only once on the CPU, instead of repeatedly recalculating the trigonometric functions in the GPU-kernels. This is **a very good decision** and should result in a speedy program. **Rather impressive**, your code looks good and performs well! I timed your code on my machine (NVIDIA GeForce RTX 3050 Ti Laptop, and I believe Laptop means some of the streaming multiprocessors are faulty and have been deactivated): ```shell Kernel time 0.3388 ``` using "threadsperblock 256". This is certainly one of the **fastest** times submitted!

近期下载者：

相关文件：

评论：[我要评论] [举报此文件]

收藏者：