Dzmitry Malyshau
Dzmitry Malyshau
Yes, `writeBuffer` is a queue operation, so the order is clear with regards to submissions. WebGPU guarantees that the following submits will observe the result of your writes. If you...
I think it's reasonable for the user to expect anything happening between push/pop to actually be visible in the debug tools, including the writeXxx instructions. The implementations should just flush...
@Kangz we could be a little bit smarter down the road to avoid this cost, if needed. We could remember the stack of debug groups for each writeXxx, and then...
> VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT specifies that descriptor sets can return their individual allocations to the pool, i.e. all of [vkAllocateDescriptorSets](https://registry.khronos.org/vulkan/specs/1.3-extensions/man/html/vkAllocateDescriptorSets.html), [vkFreeDescriptorSets](https://registry.khronos.org/vulkan/specs/1.3-extensions/man/html/vkFreeDescriptorSets.html), and [vkResetDescriptorPool](https://registry.khronos.org/vulkan/specs/1.3-extensions/man/html/vkResetDescriptorPool.html) are allowed. Otherwise, descriptor sets allocated from the pool...
nice work, @progmboy ! Would you be able to file this upstream to Intel folks? Seems rather serious. side note, things are going to be slightly better with #118 now
I suspect it may change the allocation strategy for the descriptors, thus making it slower for us to allocate new ones. And this would be potentially significant since Blade relies...
Uhh, that's rather unfortunate. Maybe it's just a Windows thing in general?
ah, ok, phew, so it was addressed by https://github.com/kvark/blade/pull/118
:+1: for your heroic effort. I didn't expect it to be too hard, please keep in mind the spots that took you most trouble to get.