[Bug tree-optimization/88533] [9 Regression] Higher performance penalty of array-bounds checking for sparse-matrix vector multiply

[anlauf at gmx dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88533

--- Comment #9 from Harald Anlauf <anlauf at gmx dot de> ---
(In reply to Richard Biener from comment #8)
> Created attachment 45252 [details]
> patch
> 
> Even though the patch doesn't hoist the invariant condition the speed is
> back with it.
> 
> Can you verify that?

I tried this just this patch on top of the 9-revision previously used and
get:

baseline + -funroll-loops -fcheck=bounds :

7: 1.56
8: 1.56
9: 1.93 (unmodified)
9: 1.63 (patched)

baseline + -O3 -funroll-loops -fcheck=bounds :

7: 1.56
8: 1.39
9: 1.57 (unmodified)
9: 1.38 (patched)

Yes, the speed is back for the testcase.

Thanks for the quick fix!