vulkan-compute-ray-tracing VS Book_of_Shaders

It's better to use a compute shader than fragment because it's better parallelizable. The idea is the same as with fragment shader, you just use compute to write into a texture. I have done something similar in Vulkan, you can take a look: https://github.com/grigoryoskin/vulkan-compute-ray-tracing https://github.com/grigoryoskin/vulkan-compute-ray-tracing/blob/master/resources/shaders/source/ray-trace-compute.comp

https://github.com/grigoryoskin/vulkan-compute-ray-tracing I'm making a small ray tracer, loosely following rt in one weekend series. It renders the scene with the compute shader into a texture and then displays it with fragment shader. Currently it has only lambertian materials, but I'm working on adding PBR. I use binary volume hierarchy for acceleration, but in simple scenes I found out that just iterating through a list of triangles is faster. It has one problem though, sometimes(!) it crashes on start or finish, making me think it's some kind of synchronization issue (also discussed here). I have layout transform and semaphore between compute and graphics, which looks tight to me, but I would be super grateful if someone could point out where I am wrong. Or it could be some moltenVK bug, which would also be ok with me. I would appreciate any kind of feedback, since I'm still learning (both vk and c++)

I made a ray tracer in Vulkan, and while it runs fine on my personal Mac book air, when I try to run it on my work Mac book pro, it either crashes at start due to some memory error, or renders a blank screen. I'm trying super hard to find the issue, but I'm out of ideas. These are the only computers I have, so I'm willing to accept that it's some moltenVK issue or something, if it runs on other systems. Here is the repo with build instructions: https://github.com/grigoryoskin/vulkan-compute-ray-tracing