[Bug middle-end/80283] [5/6/7/8 Regression] bad SIMD register allocation

[already5chosen at yahoo dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80283

--- Comment #14 from Michael_S <already5chosen at yahoo dot com> ---
Created attachment 41293
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=41293&action=edit
another case of bad vector register allocation

Here is another case of bad allocation of SIMD register that hopefully can help
Berndt.
This case does not appear to depend on optimization level or on -fno-tree-ter.
Bad allocation happens with all versions of gcc that I tested on godbolt, but 
the level of damage is higher with newer versions.

# of spills of accumulators in the inner loop as function of value of UNROLL
and of version of gcc:
UNROLL     1 2 3 4
gcc 4.9.4  0 2 1 1
gcc 5.4    0 2 1 1
gcc 6.3    0 2 1 2
gcc 7      0 2 1 2

Link to goldbolt:
https://godbolt.org/g/N49vBs