[CI test do not merge] Skip rocm msg

test/test_cuda.py

@@ -1192,7 +1192,7 @@ def _test_stream_event_nogil(self, sync_func, p2c, c2p):
             c2p.put(sync_func(self, TestCuda.FIFTY_MIL_CYCLES))
 
     # Skip the test for ROCm as per https://github.com/pytorch/pytorch/issues/53190
-    @skipIfRocm
+    @skipIfRocm("as per pytorch/issues/53190")
     @unittest.skipIf(not TEST_MULTIGPU, "detected only one GPU")
     def test_stream_event_nogil(self):
         for sync_func in [TestCuda._stream_synchronize,
@@ -1230,7 +1230,7 @@ def test_stream_event_nogil(self):
             self.assertGreater(parent_time + child_time, total_time * 1.4)
 
     # This test is flaky for ROCm, see issue #62602
-    @skipIfRocm
+    @skipIfRocm("flakey on rocm")
     @unittest.skipIf(not TEST_MULTIGPU, "detected only one GPU")
     def test_events_wait(self):
         d0 = torch.device('cuda:0')
@@ -1340,7 +1340,7 @@ def test_events_multi_gpu_elapsed_time(self):
             self.assertGreater(e0.elapsed_time(e2), 0)
 
     # XXX: this test only fails with hip-clang. revisit this once the dust has settled there.
-    @skipIfRocm
+    @skipIfRocm("fails on hip-clang")
     def test_record_stream(self):
         cycles_per_ms = get_cycles_per_ms()
 

test/test_nn.py

@@ -3555,7 +3555,7 @@ def test_cudnn_rnn_dropout_states_device(self):
         output = rnn(input, hx)
 
     @unittest.skipIf(not TEST_CUDNN, 'CUDNN not available')
-    @skipIfRocm
+    @skipIfRocm("Skipped on ROCm as explicit cudnn test")
     def test_cudnn_weight_format(self):
         rnns = [
             nn.LSTM(10, 20, batch_first=True),

test/test_optim.py

@@ -1509,7 +1509,7 @@ def test_asgd(self):
             with self.assertRaisesRegex(ValueError, "Invalid weight_decay value: -0.5"):
                 optim.ASGD(None, lr=1e-2, weight_decay=-0.5, foreach=foreach)
 
-    @skipIfRocm
+    @skipIfRocm("Skipped on ROCm due to some reason...")
     def test_rprop(self):
         is_cuda_sm86 = torch.cuda.is_available() and torch.cuda.get_device_capability(
             0

test/test_sparse_csr.py

@@ -2232,7 +2232,7 @@ def run_test(n, k, upper, unitriangular, transpose, zero):
                                                                                  itertools.product([True, False], repeat=4)):
             run_test(n, k, upper, unitriangular, transpose, zero)
 
-    @skipCUDAIfRocm
+    @skipCUDAIfRocm("sampled_addmm issue")
     @skipCUDAIf(
         not _check_cusparse_sddmm_available(),
         "cuSparse Generic API SDDMM is not available"
@@ -2287,7 +2287,7 @@ def run_test(c, a, b, op_a, op_b, *, alpha=None, beta=None):
                 for op_a, op_b in itertools.product([True, False], repeat=2):
                     run_test(c, a, b, op_a, op_b)
 
-    @skipCUDAIfRocm
+    @skipCUDAIfRocm("sddmm known issue")
     @skipCUDAIf(
         not _check_cusparse_sddmm_available(),
         "cuSparse Generic API SDDMM is not available"
@@ -2318,7 +2318,7 @@ def test_sampled_addmm_autograd(self, device, dtype):
             self.assertEqual(a.grad, a1.grad)
             self.assertEqual(b.grad, b1.grad)
 
-    @skipCUDAIfRocm
+    @skipCUDAIfRocm("sddmm known issue")
     @onlyCUDA
     @skipCUDAIf(True, "Causes CUDA memory exception, see https://github.com/pytorch/pytorch/issues/72177")
     @skipCUDAIf(
@@ -2526,7 +2526,7 @@ def fn(input):
             dense_output.backward(dense_covector)
             self.assertEqual(sparse_input.grad, dense_input.grad)
 
-    @skipCUDAIfRocm
+    @skipCUDAIfRocm("addmm known issues")
     @skipCUDAIf(
         not _check_cusparse_sddmm_available(),
         "cuSparse Generic API SDDMM is not available"

torch/testing/_internal/common_device_type.py

@@ -1252,8 +1252,19 @@ def skipCUDAIfNoMagmaAndNoCusolver(fn):
         return skipCUDAIfNoMagma(fn)
 
 # Skips a test on CUDA when using ROCm.
-def skipCUDAIfRocm(fn):
-    return skipCUDAIf(TEST_WITH_ROCM, "test doesn't currently work on the ROCm stack")(fn)
+def skipCUDAIfRocm(msg="test doesn't currently work on the ROCm stack"):
+
+    def dec_fn(fn):
+        @wraps(fn)
+        def wrap_fn(self, *args, **kwargs):
+            if TEST_WITH_ROCM:
+                if self.device_type == 'cuda':
+                    raise unittest.SkipTest(f"skipCUDAIfRocm: {msg}")
+
+            return fn(self, *args, **kwargs)
+
+        return wrap_fn
+    return dec_fn
 
 # Skips a test on CUDA when not using ROCm.
 def skipCUDAIfNotRocm(fn):

torch/testing/_internal/common_utils.py

@@ -1058,14 +1058,15 @@ def has_corresponding_torch_dtype(np_dtype):
     torch.complex32: np.complex64
 })
 
-def skipIfRocm(fn):
-    @wraps(fn)
-    def wrapper(*args, **kwargs):
-        if TEST_WITH_ROCM:
-            raise unittest.SkipTest("test doesn't currently work on the ROCm stack")
-        else:
-            fn(*args, **kwargs)
-    return wrapper
+def skipIfRocm(msg="test doesn't currently work on the ROCm stack"):
+    def dec_fn(fn):
+        @wraps(fn)
+        def wrap_fn(self, *args, **kwargs):
+            if TEST_WITH_ROCM:
+                raise unittest.SkipTest(f"skipIfRocm: {msg}")
+            return fn(self, *args, **kwargs)
+        return wrap_fn
+    return dec_fn
 
 def skipIfMps(fn):
     @wraps(fn)