Merge pull request #55 from ROCmSoftwarePlatform/IFU-master-2021-10-15

IFU-2021-10-15 (+ remove redundant defines + C10_CUDA_CHECK)

apex/contrib/fmha/__init__.py

+from .fmha import FMHAFun

apex/contrib/optimizers/fused_lamb.py

@@ -111,7 +111,7 @@ class FusedLAMB(torch.optim.Optimizer):
                     continue
                 if p.dtype == torch.float32:
                     g_all_32.append(p.grad.data)
-                elif p.dytpe == torch.float16:
+                elif p.dtype == torch.float16:
                     g_all_16.append(p.grad.data)
                 else:
                     raise RuntimeError('FusedLAMB only support fp16 and fp32.')