[Bug c++/109471] New: Missing loop unrolling for small std::vector?

[Bug c++/109471] New: Missing loop unrolling for small std::vector?

[stefano.d at posteo dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109471

            Bug ID: 109471
           Summary: Missing loop unrolling for small std::vector?
           Product: gcc
           Version: 12.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: stefano.d at posteo dot de
  Target Milestone: ---

Hi,
I did some benchmarks where I wanted to bench a switch-case statement against a
small std::vector. In GCC 12.2 the result was as expected: The switch-case
statement was faster than the linear search in std::vector. But when I switched
to clang 13 or above, the std::vector implementation was much faster than the
switch-case statement!

You can the results here:
https://quick-bench.com/q/9DEDS7rQm0MnIFxwZt3A2iD86G0

Please switch here between GCC and clang.

Here an assembly output:

https://godbolt.org/z/j8oenP8x9

Kind regards
Stefano

[Bug tree-optimization/109471] Missing loop unrolling for small std::vector?

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109471

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization
          Component|c++                         |tree-optimization

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
I suppose both cases could be optimized the same, but we only do limited "IR
interpretation" and only as side-effect of optimizing scalarized code, not
simulated execution traces.

Needs some more investigation.

Can you please attach the source code?

[Bug tree-optimization/109471] Missing loop unrolling for small std::vector?

[xry111 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109471

Xi Ruoyao <xry111 at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |xry111 at gcc dot gnu.org

--- Comment #2 from Xi Ruoyao <xry111 at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #1)
> I suppose both cases could be optimized the same, but we only do limited "IR
> interpretation" and only as side-effect of optimizing scalarized code, not
> simulated execution traces.
> 
> Needs some more investigation.
> 
> Can you please attach the source code?

The code seems available in the godbolt link but it uses std::array, not
std::vector.

[Bug tree-optimization/109471] Missing loop unrolling for small std::array?

[stefano.d at posteo dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109471

--- Comment #3 from Stefano <stefano.d at posteo dot de> ---
(In reply to Xi Ruoyao from comment #2)
> The code seems available in the godbolt link but it uses std::array, not
> std::vector.

I'm sorry. I mean std::array of course. :-/

[Bug tree-optimization/109471] Missing loop unrolling for small std::array?

[stefano.d at posteo dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109471

--- Comment #4 from Stefano <stefano.d at posteo dot de> ---
Created attachment 54829
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=54829&action=edit
source code