Fix pointer arithmetic in test_copy_cache_modifiers

[Copilot]

The test was using incorrect pointer offsets when calling iris.copy(), causing failures with certain cache modifier combinations (.cv-.cg).

Changes

• Fixed destination pointer calculation: Changed results + BLOCK_SIZE * target_rank to results + BLOCK_SIZE * cur_rank (line 31)

  • iris.copy() expects both pointers in current rank's address space; dst_ptr is translated internally

• Fixed test expectations: Updated to verify each rank writes to its own slot on all targets (lines 100-106)

  • After barrier, results[rank_id] should contain data from rank rank_id: (rank_id + num_ranks) * (rank_id + 1)
  • Aligns with PUT semantics in test_copy_put

The original code calculated destination offsets as if they were already in the target rank's address space, but iris.copy() performs this translation internally.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Fix cache modifiers unittest error</issue_title>
<issue_description>2025-10-29T23:28:32.5127431Z FAILED tests/unittests/test_copy_cache_modifiers.py::test_copy_cache_modifiers[.cv-.cg] - AssertionError: Mismatch at rank 1, target 0 with load_cache_modifier=.cv, store_cache_modifier=.cg
2025-10-29T23:28:32.5127462Z assert False
2025-10-29T23:28:32.5127681Z + where False = <built-in method allclose of type object at 0x7f6b69a8a040>(tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n device='cuda:1'), tensor([10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10.,\n 10., 10.], device='cuda:1'))
2025-10-29T23:28:32.5127770Z + where <built-in method allclose of type object at 0x7f6b69a8a040> = torch.allclose
2025-10-29T23:28:32.5127963Z + and tensor([10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10.,\n 10., 10.], device='cuda:1') = <built-in method full of type object at 0x7f6b69a8a040>((16,), 10, dtype=torch.float32, device=device(type='cuda', index=1))
2025-10-29T23:28:32.5128039Z + where <built-in method full of type object at 0x7f6b69a8a040> = torch.full
2025-10-29T23:28:32.5128083Z + and torch.float32 = torch.float32
2025-10-29T23:28:32.5128270Z + and device(type='cuda', index=1) = tensor([[ 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n 0., 0.],\n [10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10.,\n 10., 10.],\n [ 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n 0., 0.],\n [ 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n 0., 0.]], device='cuda:1').device
2025-10-29T23:28:32.5128512Z FAILED tests/unittests/test_copy_cache_modifiers.py::test_copy_cache_modifiers[.cv-.cs] - AssertionError: Mismatch at rank 1, target 0 with load_cache_modifier=.cv, store_cache_modifier=.cs
2025-10-29T23:28:32.5128543Z assert False
2025-10-29T23:28:32.5128759Z + where False = <built-in method allclose of type object at 0x7f6b69a8a040>(tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n device='cuda:1'), tensor([10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10.,\n 10., 10.], device='cuda:1'))
2025-10-29T23:28:32.5128873Z + where <built-in method allclose of type object at 0x7f6b69a8a040> = torch.allclose
2025-10-29T23</issue_description>

Comments on the Issue (you are @copilot in this section)

• Fixes Fix cache modifiers unittest error #272

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.