Open
Description
Hi,
I’m building a slab‑allocator‑style helper that pre‑allocates one big CudaSlice<u8>
and hands out sub‑slices which I immediately CudaSlice::transmute()
into typed CudaView<T>
objects (f32, f64, etc.). I have a few concerns:
- What alignment requirements should I follow?
- In terms of coalescing / transaction size, should I add any padding?
- Are there architecture‑specific mis‑alignment or transaction‑width issues I should be aware of?
Thanks for your time !
Metadata
Metadata
Assignees
Labels
No labels