[Embench] LLVM 11.0.0 update

Jeremy Bennett jeremy.bennett at embecosm.com
Thu Oct 22 18:45:06 CEST 2020


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 22/10/2020 17:33, paolo wrote:
> Hi all,
>
> as the new release of LLVM has been out for some days now I
> generated new benchmark results.

Hi Paolo,

This is good news. The dreadful LLVM top-of-tree results presented on
Monday, look to have been a transient aberation. with LLVM 11.0.0
scoring 0.97 for code speed at -O2. Slightly better than GCC 10.2.0
scoring 0.93.

Best wishes,


Jeremy

>
> Here are the geomean relative (to the current baseline, see slide
> 7) results per optimization flag for RI5CY:
>
> Options               Geomean   +ve   -ve -------
> -------   ---    -- -O0                      0.43  0.23  0.16 -O1
> 0.87  0.33  0.24 -O2                      0.97  0.37  0.28 -O3
> 0.97  0.35  0.27 -O3 -finline-limit=50    0.97  0.35  0.27 -O3
> -funroll-loops       0.97  0.35  0.27 -Os                      0.88
> 0.34  0.25 -Os -msave-restore       0.87  0.33  0.24 -Oz
> 0.87  0.33  0.25 -Oz -msave-restore       0.86  0.33  0.25
>
> Size results ============
>
> Options               Geomean   +ve   -ve -------
> -------   ---    -- -O0                      2.65  0.82  0.63 -O1
> 1.24  0.24  0.21 -O2                      1.37  0.42  0.33 -O3
> 1.56  0.76  0.52 -O3 -finline-limit=50    1.56  0.76  0.52 -O3
> -funroll-loops       1.56  0.76  0.52 -Os                      1.23
> 0.28  0.23 -Os -msave-restore       1.14  0.28  0.23 -Oz
> 1.18  0.24  0.21 -Oz -msave-restore       1.09  0.23  0.20
>
> And here the same for ARM Cortex-M4:
>
> Speed results =============
>
> Options               Geomean   +ve   -ve -------
> -------   ---    -- -O0                      0.42  0.25  0.16 -O1
> 0.94  0.14  0.13 -O2                      1.04  0.13  0.12 -O3
> 1.05  0.17  0.16 -O3 -finline-limit=50    1.05  0.17  0.16 -O3
> -funroll-loops       1.05  0.17  0.16 -Os                      0.95
> 0.13  0.12 -Oz                      0.92  0.17  0.15
>
> Size results ============
>
> Options               Geomean   +ve   -ve -------
> -------   ---    -- -O0                      1.85  0.33  0.29 -O1
> 1.18  0.14  0.13 -O2                      2.04  1.22  0.77 -O3
> 2.18  1.59  0.92 -O3 -finline-limit=50    2.18  1.59  0.92 -O3
> -funroll-loops       2.18  1.59  0.92 -Os                      1.16
> 0.17  0.16 -Oz                      1.02  0.12  0.11
>
> You'll notice that the ARM has no results for the flag combination
> "-Oz -msave-restore". That's only because ARM LLVM doesn't support
> the "-msave-restore" flag. This is now being specified in the
> slides as well.
>
> Also note that the top of tree speed result for RI5CY in silde 23
> of the Embench presentation showed during last meeting had a quite
> surprisingly worse geomean than LLVM 10.0.1.
>
> That was probably due to a lack of some optimization flags.
> Nevertheless the results from LLVM 11 (that follows the top of tree
> example of the slides) confirm that there is actually a slight
> decrease of speed optimization between LLVM 10.0.1 (a geomean of
> 0.98) and LLVM 11.0.0 (a goemean of 0.97) as we mentioned during
> the call.
>
> The slides are being currently reviewed and updated to integrate
> the LLVM 11 results.
>
> For any question/observation don't hesitate to reply.
>
>
> Speak to you all soon.
>
> Best wishes,
>
> Paolo Savini
>
>


- -- 
Cell: +44 (7970) 676050
SkypeID: jeremybennett
Twitter: @jeremypbennett
Email: jeremy.bennett at embecosm.com
Web: www.embecosm.com
PGP key: 1024D/BEF58172FB4754E1 2009-03-20
-----BEGIN PGP SIGNATURE-----

iF0EARECAB0WIQRASGDWqmhRZUfAaPW+9YFy+0dU4QUCX5G3EQAKCRC+9YFy+0dU
4UinAJ4hekp5GafLAlSUXZ18irLHpz8tqgCgkHbI0mrAoxJSDSbCaASu0VCs9/c=
=Wona
-----END PGP SIGNATURE-----



More information about the Embench mailing list