search

GPUOpen-LibrariesAndSDKs / VulkanMemoryAllocator

Easy to integrate Vulkan memory allocation library
MIT License
2.63k stars 359 forks source link

Changes to PFN_vmaAllocateDeviceMemoryFunction #417

Open debaetsd opened 6 months ago

debaetsd commented 6 months ago

I modified the Allocate callback (VmaDeviceMemoryCallbacks) of our local copy to include additional information about the used VmaPool and dedicated allocations. The signature is now basically

typedef void (VKAPI_PTR* PFN_vmaAllocateDeviceMemoryFunction)(
    VmaAllocator VMA_NOT_NULL                    allocator,
    uint32_t                                     memoryType,
    VkDeviceMemory VMA_NOT_NULL_NON_DISPATCHABLE memory,
    VkDeviceSize                                 size,
    VmaPool VMA_NULLABLE                         pool,
    bool                                         isDedicated,
    void* VMA_NULLABLE                           pUserData);

This requires some data pipelining (aka making sure the new parameters are available at the call site) but nothing all to crazy. We have done this to improve statistic tracking and simplify interop tracking.

Ideally we would like to upstream this change (so we do not have to keep a fork). Is this something you would be willing to accept? We can make the PR.

adam-sawicki-a commented 6 months ago

Thank you for the proposal, but I want to avoid complicating the library code by adding new features like this. Adding new parameters to PFN_vmaAllocateDeviceMemoryFunction would break backward-compatibility, so I would need to extend the structure VmaDeviceMemoryCallbacks with some new version PFN_vmaAllocateDeviceMemoryFunction2, which would add more code.