http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56221
Bug #: 56221
Summary: Generation of zero initializer for array new without
parenthesis
Classification: Unclassified
Product: gcc
Version: 4.7.2
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
AssignedTo: unassig...@gcc.gnu.org
ReportedBy: neldev.onl...@gmail.com
Created attachment 29364
--> http://gcc.gnu.org/bugzilla/attachment.cgi?id=29364
Minimal test case
I could not find a duplicate so I guess this will be my first bug report for a
compiler :)
I was experimenting a little with expressive templates and stumbled upon a
peculiar bug that caused assembly output to explode by several orders of
magnitude. I managed to narrow it down to automatic vectorization of a zero
initializer that transformed a 4-instruction memset into some monstrosity.
Recompiling with -fno-tree-vectorize prevents this transformation.
However, the issue of this initializer which generated the memset still
remained.
I've attached a minimal test case that performs the following:
- Trigger unwanted generation of the array initializer.
- Triggering automatic vectorization at will by increasing the parameter to
array new beyond 16.
- Preventing the array initializer from being generated by uncommenting the
second constructor.
- Producing identical results by forcing array initializer generation by
uncommenting the third constructor.
A workaround would b to move the array new statement out of the constructor.
I'm not sure but this seems to be the result of some foul play between the CRTP
pattern and operator+. Removing either will prevent this bug from being
triggered.
