Result are demonstrated at here on Bilibili.
Run make.sh on Linux or make.bat on Windows, make sure you've added nvcc compiler to your PATH environment variable.
First create a folder datas in the same directory of Nbody.cu, then run ./Nbody.out on Linux or ./Nbody.exe on Windows.
In Nbody.cu, on the top of the code, find #define GRID_X [particle number] / BLOCK_X, change the [particle number] to the number you want.
Notice that particle number must be divisible by block size BLOCK_X.
In Nbody.cu, on the top of the code, find #define dt [time step], change the [time step] to the number you want. Time unit is second.
In Nbody.cu, on the top of the code, find #define fix [smooth distance], change the [smooth distance] to the number you want. Length unit is meter.
In Nbody.cu, on the top of the code, find #define PRECISION [precision] and #define PRECISION [precision4], change them into float and float for fp32, or double and double4 for fp64. FP16 wasn't tested.
In Nbody.cu, on the top of the code, find #define rool [steps], change the [steps] to the number you want. I recommend 5 or 10 per save with small (well, pretty big on the number itself) time step.
In Nbody.cu, change the function void GenerateRandomPoints() into whatever you want. Coordinate in unit meter.
Run make_rendering.cuda.sh on Linux or make_rendering.cuda.bat on Windows, make sure you've added nvcc compiler to your PATH environment variable.
Before using it, please check the keyword "Particle Number" under the next section "Do your own configuration".
First create a folder images in the same directory of rendering.cu, then run ./rendering.cuda.out on Linux or ./rendering.cuda.exe on Windows.
It's necessary to match the particle number with your simulation.
In rendering.cu, in function int main(), find char data_file_prefix[32] = "datas/[particle number]"; and char image_file_prefix[32] = "images/[particle number]";, change the [particle number] to the same as your simulation. If your forgot it, check datas/[particle number],[step].nbody for it.
In rendering.cu, in function int main(), find long image_size_width = [image width]; and long image_size_hight = [image hight];, change [image width] x [image hight] to the resolution you want.
It's the calculating precision of floating point number during rendering.
In rendering.cu, on the top of the code, find #defind PRECISION_RENDERING [precision rendering], change the [precision rendering] to the fp type that fits you. FP16 wasn't tested.
In rendering.cu, in function int main(), we have
PRECISION_RENDERING image_camera_position[3] = {5e18f, 0, 0};
PRECISION_RENDERING image_screen_position[3] = {4.9e18f, 0, 0};
PRECISION_RENDERING image_screen_basis_w[3] = {0, 0.2e18f / image_size_hight * 2, 0};
PRECISION_RENDERING image_screen_basis_h[3] = {0, 0, 0.2e18f / image_size_hight * 2};
those are camera configurations, documention on them is a Todo.
In rendering.cu, in function int main(), find int image_index_base = [start frame], change the [start frame] to the number of which frame you want to start with.
- (2024/12/01)
Rewrite the datasaving codes. - (2024/12/01)
Rewrite the rendering code. (Maybe in C instead of Python.) - (2024/12/01)
Change the rendering method. (Maybe something more mathematical, but not that mathematical.) - Rechoose a unit system, then rewrite the whole
Nbody.cu. - Do a better rendering algorithm.
- Support continuing calculation from a break point.
- Support computing on multiple devices. (Waiting for nice cards with a nice price.)
- Support distributed computing. (Waiting for nice machine with a nice price.)