[Bug tree-optimization/115977] [12/13/14/15 Regression] [sra] Regression targeting PowerPC 970

[awilfox at adelielinux dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115977

A. Wilcox (awilfox) <awilfox at adelielinux dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #58696|0                           |1
        is obsolete|                            |

--- Comment #5 from A. Wilcox (awilfox) <awilfox at adelielinux dot org> ---
Created attachment 58742
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=58742&action=edit
Further reduced test case

I've further reduced the test case as attached.  The code is not miscompiled if
any of the following are true:

* -mcpu is not '970' (power5, power8, power9 all work)
* stack protector is not strong (-fno-stack-protector and -fstack-protector
work, -fstack-protector-strong fails)
* The union of floats is changed to a single float (`union { float x, r, s; };`
-> `float x;`)
* The mat4 ctor at line 98 is surrounded with #pragma GCC
optimize("no-stack-protector")