Dzmitry Malyshau

I'll check a few more machines. Looks like a driver issue, unless we are missing a validation error there. Please check if you have Vulkan validation layers installed.

RenderDoc 1.3 refuses to run this for some reason :/.

> I use latest RenderDoc 1.13. But I am slightly confused - since you posted RenderDoc screenshot before. Would it help if I provided a new trace? It didn't work...

~~Also, I tested on GTX 1050 / Windows, which is very close to your configuration, and wasn't able to see it. Driver version 27.21.14.5256 :(~~ Edit: turns out on this...

I see it now on my GTX 1050, looking!

The data is uploaded correctly, but `vkCmdFillBuffer`, called with `destOffset=40`, appears to overwrite bytes 32..40 as well.

Reported to NVidia level 2 support team, ticket 210412-000661 Also added to our driver [issue list](https://github.com/gfx-rs/wgpu/wiki/Known-Driver-Issues#nvidia--vulkan)

The problem is clear to me: when we see an existing BGL matching the new one, we return the existing ID and bump the refcount. This isn't compatible with with...

For Gecko, de-duplication on the client side is painful. It would be Rust code holding pointers to C++ wrappers of JS objects, yikes. I'd rather not rely on this hack....

I would be fine with exposing the composite alpha mode