FRHICommandListImmediate::LockBufferMGPU

Navigation

API > API/Runtime > API/Runtime/RHI > API/Runtime/RHI/FRHICommandListImmediate

References

void &42; LockBufferMGPU  
&40;  
    FRHIBuffer &42; Buffer,  
    uint32 GPUIndex,  
    uint32 Offset,  
    uint32 SizeRHI,  
    EResourceLockMode LockMode  
&41;  
Copy full snippet

void &42; LockBufferMGPU &40;     FRHIBuffer &42; Buffer,     uint32 GPUIndex,     uint32 Offset,     uint32 SizeRHI,     EResourceLockMode LockMode &41;

Remarks

LockBufferMGPU / UnlockBufferMGPU may ONLY be called for buffers with the EBufferUsageFlags::MultiGPUAllocate flag set! And buffers with that flag set may not call the regular (single GPU) LockBuffer / UnlockBuffer. The single GPU version of LockBuffer uses driver mirroring to propagate the updated buffer to other GPUs, while the MGPU / MultiGPUAllocate version requires the caller to manually lock and initialize the buffer separately on each GPU. This can be done by iterating over FRHIGPUMask::All() and calling LockBufferMGPU / UnlockBufferMGPU for each version.

EBufferUsageFlags::MultiGPUAllocate is only needed for cases where CPU initialized data needs to be different per GPU, which is a rare edge case. Currently, this is only used for the ray tracing acceleration structure address buffer, which contains virtual address references to other GPU resources, which may be in a different location on each GPU.