[XPU] Preserve non-contiguous strides when pinning CPU tensors for UVA view (#54874)
Signed-off-by: Chaojun Zhang <chaojun.zhang@intel.com>
This commit is contained in:
@@ -912,10 +912,15 @@ def get_accelerator_view_from_cpu_tensor(cpu_tensor: torch.Tensor) -> torch.Tens
|
||||
if cpu_tensor.numel() == 0:
|
||||
return torch.empty(cpu_tensor.shape, dtype=cpu_tensor.dtype, device="xpu")
|
||||
if not cpu_tensor.is_pinned():
|
||||
contiguous_cpu = cpu_tensor.contiguous()
|
||||
pinned = torch.empty_like(contiguous_cpu, pin_memory=True)
|
||||
pinned.copy_(contiguous_cpu)
|
||||
cpu_tensor = pinned
|
||||
pinned = torch.empty_strided(
|
||||
cpu_tensor.size(),
|
||||
cpu_tensor.stride(),
|
||||
dtype=cpu_tensor.dtype,
|
||||
device="cpu",
|
||||
pin_memory=True,
|
||||
)
|
||||
pinned.copy_(cpu_tensor)
|
||||
return torch.ops._C.get_xpu_view_from_cpu_tensor(pinned)
|
||||
return torch.ops._C.get_xpu_view_from_cpu_tensor(cpu_tensor)
|
||||
elif current_platform.is_cuda_alike():
|
||||
return torch.ops._C.get_cuda_view_from_cpu_tensor(cpu_tensor)
|
||||
|
||||
Reference in New Issue
Block a user