device <-> device via Vulkan Memory Allocator about vulkanmemoryallocator

[adam-sawicki-a]

How could the support look like?

For every separate VkDevice you use, you need to create a separate VmaAllocator object. This library doesn't launch any data transfers by itself. You have to do it on your side. VMA just lets you allocate chunks of memory. It helps with choosing right memory type out of available ones, and sub-allocating pieces of larger memory blocks. That's all. Everything else is user's responsibility or could constitute a higher-level library on top of VMA.

That said, please feel free to propose any enhancements that would be useful in your case.

By the way, if you develop using Vulkan on some exotic platform (exotic from my perspective, as a Windows PC developer :)) then can you please send me output of "vulkaninfo" from it, or some other document that would describe parameters of Vulkan environment in that platform? I'm curious what memory heaps and types are available there.

[adam-sawicki-a]

I assume the question is answered now and I close the ticket. Please feel free to comment it if you have any additional questions or suggestions.

[expipiplus1]

Apologies for the off topic bump!

@CaptGreg just out of curiosity, what cool things are you doing requiring interop between GPU and FPGA?

[CaptGreg]

Fin-tech calculations. We are seeing a 100x speed bump using a FPGA and a GPU compared to a GPU only solution. Our customers think that is cool. Now if only the PCIe bus was faster, we could go faster.

…

On Thu, 2 Apr 2020 at 12:06, Joe Hermaszewski ***@***.***> wrote: Apologies for the off topic bump! @CaptGreg <https://github.com/CaptGreg> just out of curiosity, what cool things are you doing requiring interop between GPU and FPGA? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#45 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAZP6YMZKT77ZURU6BOBT3LRKSZ2DANCNFSM4FZDKZ5Q> .