I am writing a function that renders a scene with python 3.12.3 and taichi 1.7.2. I've got a problem: because function runs on multiple threads without atomic operations image may have some artifacts. Here is example:

On the black part there is white pixel that should be black. That white pixel is from the gray part that is partially underneath the black part. 2 threads while drawing different parts have edited 1 pixel at the same time. In my code I have z_buffer to store distance to drawn pixel and image_buffer to store render result (I've named it surfarray). If I don't use atomic operations, then those artifacts will be on both arrays. But if I use atomic operations on z_buffer, then z_buffer won't have artifacts and image_buffer still will. That is logical, but on z_buffer I can use ti.atomic_max, and for image_buffer taichi currently doesn't have atomic operation, because I need to set value only if condition is true. Here is a part of code that checks if distance to pixel is smaller than distance to previous drawn pixel:

previous_z = ti.atomic_min(z_buffer[pixel_x, pixel_y], current_z)
if current_z < previous_z:
    surfarray[pixel_x, pixel_y, 0] = value_r
    surfarray[pixel_x, pixel_y, 1] = value_g
    surfarray[pixel_x, pixel_y, 2] = value_b

I think what's happening is:

• farthest thread passes condition;
• nearest thread passes condition;
• nearest thread draws pixel;
• farthest thread draws pixel

I think if i use atomic operation that sets value only if a condition is true, then nearest thread will wait untill farthest thread will set value, but taichi doesn't have suitable operation

What should I do to fix that problem?