Experimental Physics and
Industrial Control System

Michael Davidsaver via Core-talk <core-talk at aps.anl.gov> · Fri, 12 Feb 2021 12:54:07 -0800

On 2/12/21 10:37 AM, Gedare Bloom via Core-talk wrote:
> On Thu, Feb 11, 2021 at 2:24 PM Johnson, Andrew N. <anj at anl.gov> wrote:
>>
>> Hi Gedare,
>>
>> On Feb 11, 2021, at 2:28 PM, Gedare Bloom via Core-talk <core-talk at aps.anl.gov> wrote:
>>
>>
>> There's a proposal that was floated 3 years ago and is recirculating
>> within the RTEMS development team to consider a long-term stable
>> branch of rtems-4.10. I'd like to gather some information about
>> possible interest or non-interest from EPICS. The key question is:
>> Would you like to include long-term stable 4.10 releases in any
>> open/maintained EPICS versions?
>>
>>
>> I can’t speak for other labs, but the APS at Argonne currently has around 100 RTEMS 4.9.2 and 4.10.2 uC5282 systems of several different types in our operational systems. Because of the limited size of that CPU board we don’t know whether those systems could be upgraded to RTEMS-5, and we don’t have the effort available right now to find out. As a result we want future EPICS releases to continue to build and run on RTEMS 4.10, and we have been contributing effort to ensure that EPICS will continue to work on 4.10 after we merge in the RTEMS-5 port. We haven’t needed to build newer RTEMS 4.10.x versions at APS, although I believe some other labs may have been using newer tagged versions.
>>
> This is good to know. I think I and others will attempt to
> characterize the size/resource requirements on 5.1, and I might try to
> do a comparison against 4.10. (It is, probably, publishable work for
> me so I can put some time into it.) Although the source code
> complexity is increasing, we believe the runtime overhead may actually
> be shrinking due to better ways to optimize code and system resources.
> More on that, later.

It has been on my TODO list to investigate GCC link time optimization
(aka. "-flto") wrt. RTEMS.  Some cursory testing on GCC 8.3 on Linux/amd64
showed reductions in program text size beyond what -Os can achieve.

Some numbers I have on hand for a library.  Static executables were similar.

With -O3

>    text    data     bss     dec     hex filename
> 1158328   19128     432 1177888  11f920 lib/linux-x86_64/libpvxs.so.0.1

With -Os

>  705031   20056     432  725519   b120f lib/linux-x86_64/libpvxs.so.0.1

With -Os -flto

>  579227   17640     432  597299   91d33 lib/linux-x86_64/libpvxs.so.0.1

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System