clean up device checks in float8 unit test files

Summary:

While working on rowwise scaling I noticed that some of the CUDA
device capability checks we had in the test files did not make sense,
cleaning this up.

Test Plan:

tests pass on my H100

CI, it should skip less tests now since CI only has CUDA capability 8, 9

Reviewers:

Subscribers:

Tasks:

Tags:

Author: vkuzo

test/float8/test_base.py

@@ -231,15 +231,6 @@ def test_linear(
         linear_dtype: torch.dtype,
         linear_bias: bool,
     ):
-        if not emulate:
-            if not torch.cuda.is_available():
-                warnings.warn("CUDA not available")
-                pytest.skip()
-            elif torch.cuda.get_device_capability() < (9, 0):
-                warnings.warn(
-                    f"CUDA capability {torch.cuda.get_device_capability()} < (9.0)"
-                )
-                pytest.skip()
         x = torch.randn(*x_shape, device="cuda", dtype=linear_dtype)
         m_ref = nn.Linear(16, 32, bias=linear_bias, device="cuda", dtype=linear_dtype)
 
@@ -287,16 +278,6 @@ def test_autocast_outputs(
         emulate: bool,
         linear_dtype: torch.dtype,
     ):
-        if not emulate:
-            if not torch.cuda.is_available():
-                warnings.warn("CUDA not available")
-                pytest.skip()
-            elif torch.cuda.get_device_capability() < (9, 0):
-                warnings.warn(
-                    f"CUDA capability {torch.cuda.get_device_capability()} < (9.0)"
-                )
-                pytest.skip()
-
         m_ref = nn.Linear(32, 16, device="cuda", dtype=linear_dtype)
         config = Float8LinearConfig(
             cast_config_input=CastConfig(scaling_type=ScalingType.DELAYED),
@@ -334,10 +315,6 @@ def test_autocast_outputs(
     @pytest.mark.parametrize("emulate", [True, False] if is_cuda_8_9 else [True])
     @unittest.skipIf(not torch.cuda.is_available(), "CUDA not available")
     def test_type_cast(self, linear_dtype: torch.dtype, emulate: bool):
-        emulate = (
-            not torch.cuda.is_available() or torch.cuda.get_device_capability() < (9, 0)
-        )
-
         m = nn.Linear(32, 16, device="cuda", dtype=linear_dtype)
         config = Float8LinearConfig(emulate=emulate)
         m = Float8Linear.from_float(copy.deepcopy(m), config)

test/float8/test_compile.py

@@ -32,7 +32,6 @@
 from torch._dynamo.test_case import TestCase as DynamoTestCase
 from torch._dynamo.testing import CompileCounterWithBackend
 
-is_H100 = torch.cuda.is_available() and torch.cuda.get_device_capability() >= (9, 0)
 is_cuda_8_9 = torch.cuda.is_available() and torch.cuda.get_device_capability() >= (8, 9)
 
 def _test_compile_base(
@@ -224,7 +223,7 @@ def forward(self, x):
                 return x_hp
             return x_fp8
 
-    @unittest.skipIf(not torch.cuda.is_available() or not is_H100, "CUDA with float8 support not available")
+    @unittest.skipIf(not torch.cuda.is_available() or not is_cuda_8_9, "CUDA with float8 support not available")
     def test_float8_with_graph_break_in_the_middle(self):
         """Test that having Float8Tensor object at the boundary of a subgraph"""
         cnts = CompileCounterWithBackend("inductor")