the for_all_neighbors function has bug in particle_system.py.
the loop: for p_j in range(self.grid_particles_num[ti.max(0, grid_index-1)], self.grid_particles_num[grid_index]):
does not works properly. If grid_index is zero and the loop is skipped, which will result in particles getting squeezed into the (0,0,0) grid.

Dynamic examples are broken since Taichi 1.3.0.

The “grid_ids_new” obtained from running “counting_sort” is different each time. Do you have this problem?

Adding ”ti.loop_config(serialize=True)“ before the first for loop can yield the correct result, but it is inefficient.

Is there any other solution?

Hi, thanks for sharing your interesting work.
I have tried to run your code, while failed due to the restricted taichi version installed by PIP.
Should I build the taichi from source other than with pip installation?
Thanks

In line 20 of particle_system.py, domain_end is typed as domian_end by mistake.

The dimensionality of shape and offset must be the same (0 != 1)

Hi, thanks for your greak work, it really inspires me a lot . However, i'm wondering whether it is correct to compute two pressure force while computing pressure acceleration between two fluid particles.

In "Versatile Rigid-Fluid Coupling for Incompressible SPH", it only use one pressure force.
In "Versatile Rigid-Fluid Coupling for Incompressible SPH":

I'm wondering if you take interaction force into consideration. But i think that it should not be correct?
Looking forward to your reply!

Thank you for your implementation！ I have successfully run your code and the result which is in .ply file. How to visualize it like the gif shown in the respo？

        elif self.ps.material[p_j] == self.ps.material_solid:
            boundary_viscosity = 0.0
            # Boundary neighbors
            ## Akinci2012
            f_v = d * boundary_viscosity * (self.density_0 * self.ps.m_V[p_j] / (self.ps.density[p_i])) * v_xy / (
                r.norm()**2 + 0.01 * self.ps.support_radius**2) * self.cubic_kernel_derivative(r)
            ret += f_v

If boundary_viscosity is zero, the f_v will be zero too.....

line 19 and 20, domain is misspelled as domian, which lead to wrong domain size
self.domain_end = np.array([1.0, 1.0, 1.0])
self.domian_end = np.array(self.cfg.get_cfg("domainEnd"))

hello, thank you so much for sharing.
I got an error when set armadillo_bath_dynamic.json "Configuration" "exportObj" to true

[E 01/14/23 19:13:58.414 20808] [taichi/rhi/cuda/cuda_driver.h:operator()@88] CUDA Error CUDA_ERROR_ILLEGAL_ADDRESS: an illegal memory access was encountered while calling stream_synchronize (cuStreamSynchronize)

Traceback (most recent call last):
File "E:\Taichi\SPH_Taichi-master\run_simulation.py", line 101, in
obj_data = ps.dump(obj_id=obj_id)
File "E:\Taichi\SPH_Taichi-master\particle_system.py", line 461, in dump
self.copy_to_numpy_nd(obj_id, np_x, self.x)
File "D:\Users\m1503\AppData\Local\Programs\Python\Python39\lib\site-packages\taichi\lang\kernel_impl.py", line 1006, in call
return self.primal(self.kernel_owner, *args, **kwargs)
File "D:\Users\m1503\AppData\Local\Programs\Python\Python39\lib\site-packages\taichi\lang\kernel_impl.py", line 872, in call
return self.runtime.compiled_functionskey
File "D:\Users\m1503\AppData\Local\Programs\Python\Python39\lib\site-packages\taichi\lang\kernel_impl.py", line 797, in func
raise e from None
File "D:\Users\m1503\AppData\Local\Programs\Python\Python39\lib\site-packages\taichi\lang\kernel_impl.py", line 794, in func__
t_kernel(launch_ctx)
RuntimeError: [taichi/rhi/cuda/cuda_driver.h:operator()@88] CUDA Error CUDA_ERROR_ILLEGAL_ADDRESS: an illegal memory access was encountered while calling stream_synchronize (cuStreamSynchronize)
GLFW Error 65537: The GLFW library is not initialized
[E 01/14/23 19:13:59.363 20808] [taichi/rhi/cuda/cuda_driver.h:operator()@88] CUDA Error CUDA_ERROR_ILLEGAL_ADDRESS: an illegal memory access was encountered while calling stream_synchronize (cuStreamSynchronize)

the provided demo runs in a very low fps, around 0.75-0.85 fps.
I tried both cpu and gpu, also tried disabling the saving frames but still very slow.
also tried disabling output sim_info

请问iisph算法的实现参考的是哪一篇论文？和Markus Ihmsen的原文Implicit Incompressible SPH中的算法实现好像有点不太一样。

Does this package allow defining a custom geometry that contains the flow (for example, particles moving through pipes), and if so are there any examples of how to do this? Thanks!

Hi, Thanks for your lovely project.
Anyway I've runned python run_simulation.py --scene_file ./data/scenes/dragon_bath.json and what I saw was

I was expecting like the video you've uploaded(with realistic water and objects).
Is it somewhat related to the GLFW Error 65537: The GLFW library is not initialized? Which I got while running the code.

Thanks.

CPU算的太慢了，不知道为啥改成GPU就跑不了了。是因为代码里用了什么东西吗？

The current implementation of neighborhood relies on cuda prefix sum, which prevents the program running on CPU.
There should be an option to disable the GPU only prefix sum.