已合并
[sync] PR-36922: test(npu): add primary context isolation tests #37661
ascend-robot创建于 6月5日
[sync] PR-36922: test(npu): add primary context isolation tests #37661
已合并
共 1 个文件变更+120-0
| @@ -0,0 +1,120 @@ | |||
| 1 | +# Owner(s): ["module: npu"] | ||
| 2 | + | ||
| 3 | +# NOTE: this needs to be run in a brand new process | ||
| 4 | + | ||
| 5 | +import os | ||
| 6 | +import unittest | ||
| 7 | + | ||
| 8 | +import torch | ||
| 9 | +import torch_npu | ||
| 10 | +from torch.testing._internal.common_distributed import MultiProcessTestCase | ||
| 11 | +from torch.testing._internal.common_utils import run_tests | ||
| 12 | + | ||
| 13 | +TEST_MULTINPU = torch_npu.npu.device_count() > 1 | ||
| 14 | + | ||
| 15 | + | ||
| 16 | +class TestNpuPrimaryCtx(MultiProcessTestCase): | ||
| 17 | + CTX_ALREADY_CREATED_ERR_MSG = ( | ||
| 18 | + "Tests defined in test_npu_primary_ctx.py must be run in a process " | ||
| 19 | + "where NPU contexts are never created." | ||
| 20 | + ) | ||
| 21 | + | ||
| 22 | + | ||
| 23 | + def world_size(self): | ||
| 24 | + return 1 | ||
| 25 | + | ||
| 26 | + def setUp(self): | ||
| 27 | + super().setUp() | ||
| 28 | + self._spawn_processes() | ||
| 29 | + | ||
| 30 | + def _check_no_ctx(self): | ||
| 31 | + for device in range(torch_npu.npu.device_count()): | ||
| 32 | + self.assertFalse( | ||
| 33 | + torch_npu._C._npu_hasPrimaryContext(device), | ||
| 34 | + TestNpuPrimaryCtx.CTX_ALREADY_CREATED_ERR_MSG, | ||
| 35 | + ) | ||
| 36 | + | ||
| 37 | + def test_str_repr(self): | ||
| 38 | + self._check_no_ctx() | ||
| 39 | + x = torch.randn(1, device="npu:1") | ||
| 40 | + | ||
| 41 | + # We should have only created context on 'npu:1' | ||
| 42 | + self.assertFalse(torch_npu._C._npu_hasPrimaryContext(0)) | ||
| 43 | + self.assertTrue(torch_npu._C._npu_hasPrimaryContext(1)) | ||
| 44 | + | ||
| 45 | + str(x) | ||
| 46 | + repr(x) | ||
| 47 | + | ||
| 48 | + # We should still have only created context on 'npu:1' | ||
| 49 | + self.assertFalse(torch_npu._C._npu_hasPrimaryContext(0)) | ||
| 50 | + self.assertTrue(torch_npu._C._npu_hasPrimaryContext(1)) | ||
| 51 | + | ||
| 52 | + def test_copy(self): | ||
| 53 | + self._check_no_ctx() | ||
| 54 | + x = torch.randn(1, device="npu:1") | ||
| 55 | + | ||
| 56 | + # We should have only created context on 'npu:1' | ||
| 57 | + self.assertFalse(torch_npu._C._npu_hasPrimaryContext(0)) | ||
| 58 | + self.assertTrue(torch_npu._C._npu_hasPrimaryContext(1)) | ||
| 59 | + | ||
| 60 | + y = torch.randn(1, device="cpu") | ||
| 61 | + y.copy_(x) | ||
| 62 | + | ||
| 63 | + # We should still have only created context on 'npu:1' | ||
| 64 | + self.assertFalse(torch_npu._C._npu_hasPrimaryContext(0)) | ||
| 65 | + self.assertTrue(torch_npu._C._npu_hasPrimaryContext(1)) | ||
| 66 | + | ||
| 67 | + def test_pin_memory(self): | ||
| 68 | + self._check_no_ctx() | ||
| 69 | + x = torch.randn(1, device="npu:1") | ||
| 70 | + | ||
| 71 | + # We should have only created context on 'npu:1' | ||
| 72 | + self.assertFalse(torch_npu._C._npu_hasPrimaryContext(0)) | ||
| 73 | + self.assertTrue(torch_npu._C._npu_hasPrimaryContext(1)) | ||
| 74 | + | ||
| 75 | + self.assertFalse(x.is_pinned()) | ||
| 76 | + | ||
| 77 | + # We should still have only created context on 'npu:1' | ||
| 78 | + self.assertFalse(torch_npu._C._npu_hasPrimaryContext(0)) | ||
| 79 | + self.assertTrue(torch_npu._C._npu_hasPrimaryContext(1)) | ||
| 80 | + | ||
| 81 | + x = torch.randn(3, device="cpu").pin_memory() | ||
| 82 | + | ||
| 83 | + # We should still have only created context on 'npu:1' | ||
| 84 | + self.assertFalse(torch_npu._C._npu_hasPrimaryContext(0)) | ||
| 85 | + self.assertTrue(torch_npu._C._npu_hasPrimaryContext(1)) | ||
| 86 | + | ||
| 87 | + self.assertTrue(x.is_pinned()) | ||
| 88 | + | ||
| 89 | + # We should still have only created context on 'npu:1' | ||
| 90 | + self.assertFalse(torch_npu._C._npu_hasPrimaryContext(0)) | ||
| 91 | + self.assertTrue(torch_npu._C._npu_hasPrimaryContext(1)) | ||
| 92 | + | ||
| 93 | + x = torch.randn(3, device="cpu", pin_memory=True) | ||
| 94 | + | ||
| 95 | + # We should still have only created context on 'npu:1' | ||
| 96 | + self.assertFalse(torch_npu._C._npu_hasPrimaryContext(0)) | ||
| 97 | + self.assertTrue(torch_npu._C._npu_hasPrimaryContext(1)) | ||
| 98 | + | ||
| 99 | + x = torch.zeros(3, device="cpu", pin_memory=True) | ||
| 100 | + | ||
| 101 | + # We should still have only created context on 'npu:1' | ||
| 102 | + self.assertFalse(torch_npu._C._npu_hasPrimaryContext(0)) | ||
| 103 | + self.assertTrue(torch_npu._C._npu_hasPrimaryContext(1)) | ||
| 104 | + | ||
| 105 | + x = torch.empty(3, device="cpu", pin_memory=True) | ||
| 106 | + | ||
| 107 | + # We should still have only created context on 'npu:1' | ||
| 108 | + self.assertFalse(torch_npu._C._npu_hasPrimaryContext(0)) | ||
| 109 | + self.assertTrue(torch_npu._C._npu_hasPrimaryContext(1)) | ||
| 110 | + | ||
| 111 | + x = x.pin_memory() | ||
| 112 | + | ||
| 113 | + # We should still have only created context on 'npu:1' | ||
| 114 | + self.assertFalse(torch_npu._C._npu_hasPrimaryContext(0)) | ||
| 115 | + self.assertTrue(torch_npu._C._npu_hasPrimaryContext(1)) | ||
| 116 | + | ||
| 117 | + | ||
| 118 | +if __name__ == "__main__": | ||
| 119 | + os.environ["ACL_OP_INIT_MODE"] = "1" | ||
| 120 | + run_tests() | ||