roll: improve error handling and error messages.

[ysiraichi]

This PR refactors the roll operation implementation by improving its error message, and returning a status type value.

Key Changes:

• Make tensor_methods::roll return StatusOr<XLATensorPtr>
• Improve error messages and error handling
  • Create CheckRollShiftsRequired and CheckRollDimsAndShiftsAreCompatible functions

Example 1: empty shifts argument

a = torch.arange(8, device=device).view(2, 2, 2)
shifts = []

Before:

Traceback (most recent call last):
  File "examples/roll.py", line 24, in <module>
    torch.roll(a, shifts)
RuntimeError: Check failed: 0 < shifts.size() (0 vs. 0)`shifts` required (at torch_xla/csrc/tensor_methods.cpp:3057)

Exception raised from operator& at torch_xla/csrc/runtime/tf_logging.cpp:26 (most recent call first):

After:

Traceback (most recent call last):
  File "examples/roll.py", line 24, in <module>
    torch.roll(a, shifts)
RuntimeError: roll(): expected `shifts` to have at least 1 element.

Status Propagation Trace:
    From: roll at torch_xla/csrc/tensor_methods.cpp:3086
    From: roll at torch_xla/csrc/aten_xla_type.cpp:3322

Exception raised from ThrowStatusError at torch_xla/csrc/status.cpp:128 (most recent call first):

Example 2: multiple shifts arguments with empty dims

a = torch.arange(8, device=device).view(2, 2, 2)
shifts = [1, 1]

Before:

tensor([[[7, 0],
         [1, 2]],

        [[3, 4],
         [5, 6]]], device='xla:0')

After:

Traceback (most recent call last):
  File "examples/roll.py", line 24, in <module>
    torch.roll(a, shifts)
RuntimeError: roll(): expected `shifts` [1, 1] (size=2) to have exactly 1 element when `dims` is empty.

Status Propagation Trace:
    From: roll at torch_xla/csrc/tensor_methods.cpp:3087
    From: roll at torch_xla/csrc/aten_xla_type.cpp:3322

Exception raised from ThrowStatusError at torch_xla/csrc/status.cpp:128 (most recent call first):

Example 3:

a = torch.arange(8, device=device).view(2, 2, 2)
shifts = [1, 1]
dims = [0, 1, 2]

Before:

Traceback (most recent call last):
  File "examples/roll.py", line 24, in <module>
    torch.roll(a, shifts, dims)
RuntimeError: Check failed: shifts.size() == dims.size() (2 vs. 3)shifts and dimensions must align. shifts: 2, dims:3 (at torch_xla/csrc/tensor_methods.cpp:3059)

Exception raised from operator& at torch_xla/csrc/runtime/tf_logging.cpp:26 (most recent call first):

After:

Traceback (most recent call last):
  File "examples/roll.py", line 24, in <module>
    torch.roll(a, shifts, dims)
RuntimeError: roll(): expected `dims` [0, 1, 2] (size=3) to match the size of `shifts` [1, 1] (size=2).

Status Propagation Trace:
    From: roll at torch_xla/csrc/tensor_methods.cpp:3087
    From: roll at torch_xla/csrc/aten_xla_type.cpp:3322

Exception raised from ThrowStatusError at torch_xla/csrc/status.cpp:128 (most recent call first):