skip 3D NCHW FP16 batchnorm test due to Native accuracy issue

Cherry-pick of #2370

Co-authored-by: Dmitry Nikolaev <[email protected]>

Author: dhonnappa-amd

test/test_nn.py

@@ -5180,6 +5180,8 @@ def test_batchnorm_nhwc_cuda(self):
         name_fn=lambda f, b, m, t: f"{f}_vs_{b}{'_mixed' if m else ''}_{dtype_name(t)}"
     )
     def test_batchnorm(self, dims, mode, memory_format, ref_backend, mixed, dtype):
+        if self._testMethodName == "test_batchnorm_3D_train_NCHW_vs_native_mixed_float16":
+           self.skipTest("3D float16 NCHW train failed on CUDA and ROCm due to Native batchnorm accuracy issue SWDEV-541024")
         if torch.version.hip:
             if self._testMethodName in ("test_batchnorm_2D_train_NHWC_vs_NCHW_mixed_bfloat16",
                                     "test_batchnorm_2D_train_NCHW_vs_cpu_mixed_bfloat16",
@@ -5195,10 +5197,6 @@ def test_batchnorm(self, dims, mode, memory_format, ref_backend, mixed, dtype):
                                         ) and _get_torch_rocm_version() >= (6, 4):
                 self.skipTest("bfloat16 NCHW train failed due to native tolerance issue SWDEV-507600")
 
-            if self._testMethodName == "test_batchnorm_3D_train_NCHW_vs_native_mixed_float16" \
-                and _get_torch_rocm_version() < (6, 4):
-                self.skipTest("3D float16 NCHW train failed on ROCm<=6.3 ")
-
         if dims == 3 and memory_format in ("NHWC", "NCHW"):
             memory_format = memory_format + "3D"