Commit bba74577 by Eric Botcazou Committed by Eric Botcazou

* decl.c (maybe_pad_type): Add ??? comment.

From-SVN: r134953
parent 23688481
2008-05-05 Eric Botcazou <ebotcazou@adacore.com>
* decl.c (maybe_pad_type): Add ??? comment.
2008-05-03 Eric Botcazou <ebotcazou@adacore.com> 2008-05-03 Eric Botcazou <ebotcazou@adacore.com>
* decl.c (components_to_record): Zero the alignment of the qualified * decl.c (components_to_record): Zero the alignment of the qualified
...@@ -5671,7 +5671,13 @@ maybe_pad_type (tree type, tree size, unsigned int align, ...@@ -5671,7 +5671,13 @@ maybe_pad_type (tree type, tree size, unsigned int align,
integral mode, which will be much more efficient. There is no point integral mode, which will be much more efficient. There is no point
in doing so if a size is specified unless it is also a small constant in doing so if a size is specified unless it is also a small constant
size and it is incorrect to do so if we cannot guarantee that the mode size and it is incorrect to do so if we cannot guarantee that the mode
will be naturally aligned since the field must always be addressable. */ will be naturally aligned since the field must always be addressable.
??? This might not always be a win when done for a stand-alone object:
since the nominal and the effective type of the object will now have
different modes, a VIEW_CONVERT_EXPR will be required for converting
between them and it might be hard to overcome afterwards, including
at the RTL level when the stand-alone object is accessed as a whole. */
if (align != 0 if (align != 0
&& TREE_CODE (type) == RECORD_TYPE && TREE_CODE (type) == RECORD_TYPE
&& TYPE_MODE (type) == BLKmode && TYPE_MODE (type) == BLKmode
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment