[Reland][doc] Add overflow notice for cuFFT on half precision (#40551)

xwang233 · facebook-github-bot · commit 24a8614cac3a · 2020-06-26T15:40:19.000-07:00
Summary: Reland of #35594 Pull Request resolved: #40551 Reviewed By: ezyang Differential Revision: D22249831 Pulled By: ngimel fbshipit-source-id: b221b3c0a490ccaaabba50aa698a2490536e0917
diff --git a/torch/_torch_docs.py b/torch/_torch_docs.py
@@ -7257,6 +7257,10 @@ def merge_dicts(*dicts):
     configuration. See :ref:`cufft-plan-cache` for more details on how to
     monitor and control the cache.
 
+.. warning::
+    Due to limited dynamic range of half datatype, performing this operation in half
+    precision may cause the first element of result to overflow for certain inputs.
+
 .. warning::
     For CPU tensors, this method is currently only available with MKL. Use
     :func:`torch.backends.mkl.is_available` to check if MKL is installed.
@@ -7352,6 +7356,10 @@ def merge_dicts(*dicts):
     configuration. See :ref:`cufft-plan-cache` for more details on how to
     monitor and control the cache.
 
+.. warning::
+    Due to limited dynamic range of half datatype, performing this operation in half
+    precision may cause the first element of result to overflow for certain inputs.
+
 .. warning::
     For CPU tensors, this method is currently only available with MKL. Use
     :func:`torch.backends.mkl.is_available` to check if MKL is installed.
@@ -7436,6 +7444,10 @@ def merge_dicts(*dicts):
     configuration. See :ref:`cufft-plan-cache` for more details on how to
     monitor and control the cache.
 
+.. warning::
+    Due to limited dynamic range of half datatype, performing this operation in half
+    precision may cause the first element of result to overflow for certain inputs.
+
 .. warning::
     For CPU tensors, this method is currently only available with MKL. Use
     :func:`torch.backends.mkl.is_available` to check if MKL is installed.
@@ -7513,6 +7525,10 @@ def merge_dicts(*dicts):
     configuration. See :ref:`cufft-plan-cache` for more details on how to
     monitor and control the cache.
 
+.. warning::
+    Due to limited dynamic range of half datatype, performing this operation in half
+    precision may cause the first element of result to overflow for certain inputs.
+
 .. warning::
     For CPU tensors, this method is currently only available with MKL. Use
     :func:`torch.backends.mkl.is_available` to check if MKL is installed.
