Re: RE: [PATCH v7] libgfortran: Replace mutex with rwlock

[Lipeng Zhu <lipeng.zhu@intel.com>]

Hi Thomas,

On 2023/12/21 19:42, Thomas Schwinge wrote:
> Hi!
> 
> On 2023-12-13T21:52:29+0100, I wrote:
>> On 2023-12-12T02:05:26+0000, "Zhu, Lipeng" <lipeng.zhu@intel.com> wrote:
>>> On 2023/12/12 1:45, H.J. Lu wrote:
>>>> On Sat, Dec 9, 2023 at 7:25 PM Zhu, Lipeng <lipeng.zhu@intel.com> wrote:
>>>>> On 2023/12/9 23:23, Jakub Jelinek wrote:
>>>>>> On Sat, Dec 09, 2023 at 10:39:45AM -0500, Lipeng Zhu wrote:
>>>>>>> This patch try to introduce the rwlock and split the read/write to
>>>>>>> unit_root tree and unit_cache with rwlock instead of the mutex to
>>>>>>> increase CPU efficiency. In the get_gfc_unit function, the
>>>>>>> percentage to step into the insert_unit function is around 30%, in
>>>>>>> most instances, we can get the unit in the phase of reading the
>>>>>>> unit_cache or unit_root tree. So split the read/write phase by
>>>>>>> rwlock would be an approach to make it more parallel.
>>>>>>>
>>>>>>> BTW, the IPC metrics can gain around 9x in our test server with
>>>>>>> 220 cores. The benchmark we used is
>>>>>>> https://github.com/rwesson/NEAT
>>
>>>>>> Ok for trunk, thanks.
>>
>>>>> Thanks! Looking forward to landing to trunk.
>>
>>>> Pushed for you.
> 
>> I've just filed <https://gcc.gnu.org/PR113005>
>> "'libgomp.fortran/rwlock_1.f90', 'libgomp.fortran/rwlock_3.f90' execution test timeouts".
>> Would you be able to look into that?
> 
> See my update in there.
> 
> 
> Grüße
>   Thomas
> -------------- > Siemens Electronic Design Automation GmbH; Anschrift: Arnulfstraße 
201, 80634 München; Gesellschaft mit beschränkter Haftung; 
Geschäftsführer: Thomas Heurung, Frank Thürauf; Sitz der Gesellschaft: 
München; Registergericht München, HRB 106955
> 

Since I don't have gcc bugzilla account. Reply in this thread:
Limit themselves to some lower 'OMP_NUM_THREADS' should be an option or 
increase the execution timeout?

But I can't reproduce the execution timeout failure on both powerpc9 and 
powerpc10 arch machine. And I also tried to decrease the CPU frequency 
from 2.6G to 800M, these test cases still can run successfully.

 > so only a little bit of an improvement of the new "rwlock" 
libgfortran vs. old "mutex" GCC 10 one, curiously.  (But supposedly that 
depends on the hardware or other factors?)

The rwlock can increase the IPC a lot, maybe the wall time you listed is 
not obvious.

$ grep ^cpu < /proc/cpuinfo | uniq -c

192 cpu             : POWER10 (architected), altivec supported

Native configuration is powerpc64le-unknown-linux-gnu

Schedule of variations:
     unix

PASS: libgomp.fortran/rwlock_1.f90   -O0  (test for excess errors)
PASS: libgomp.fortran/rwlock_1.f90   -O0  execution test
PASS: libgomp.fortran/rwlock_1.f90   -O1  (test for excess errors)
PASS: libgomp.fortran/rwlock_1.f90   -O1  execution test
PASS: libgomp.fortran/rwlock_1.f90   -O2  (test for excess errors)
PASS: libgomp.fortran/rwlock_1.f90   -O2  execution test
PASS: libgomp.fortran/rwlock_1.f90   -O3 -fomit-frame-pointer 
-funroll-loops -fpeel-loops -ftracer -finline-functions  (test for 
excess errors)
PASS: libgomp.fortran/rwlock_1.f90   -O3 -fomit-frame-pointer 
-funroll-loops -fpeel-loops -ftracer -finline-functions  execution test
PASS: libgomp.fortran/rwlock_1.f90   -O3 -g  (test for excess errors)
PASS: libgomp.fortran/rwlock_1.f90   -O3 -g  execution test
PASS: libgomp.fortran/rwlock_1.f90   -Os  (test for excess errors)
PASS: libgomp.fortran/rwlock_1.f90   -Os  execution test
PASS: libgomp.fortran/rwlock_2.f90   -O0  (test for excess errors)
PASS: libgomp.fortran/rwlock_2.f90   -O0  execution test
PASS: libgomp.fortran/rwlock_2.f90   -O1  (test for excess errors)
PASS: libgomp.fortran/rwlock_2.f90   -O1  execution test
PASS: libgomp.fortran/rwlock_2.f90   -O2  (test for excess errors)
PASS: libgomp.fortran/rwlock_2.f90   -O2  execution test
PASS: libgomp.fortran/rwlock_2.f90   -O3 -fomit-frame-pointer 
-funroll-loops -fpeel-loops -ftracer -finline-functions  (test for 
excess errors)
PASS: libgomp.fortran/rwlock_2.f90   -O3 -fomit-frame-pointer 
-funroll-loops -fpeel-loops -ftracer -finline-functions  execution test
PASS: libgomp.fortran/rwlock_2.f90   -O3 -g  (test for excess errors)
PASS: libgomp.fortran/rwlock_2.f90   -O3 -g  execution test
PASS: libgomp.fortran/rwlock_2.f90   -Os  (test for excess errors)
PASS: libgomp.fortran/rwlock_2.f90   -Os  execution test
PASS: libgomp.fortran/rwlock_3.f90   -O0  (test for excess errors)
PASS: libgomp.fortran/rwlock_3.f90   -O0  execution test
PASS: libgomp.fortran/rwlock_3.f90   -O1  (test for excess errors)
PASS: libgomp.fortran/rwlock_3.f90   -O1  execution test
PASS: libgomp.fortran/rwlock_3.f90   -O2  (test for excess errors)
PASS: libgomp.fortran/rwlock_3.f90   -O2  execution test
PASS: libgomp.fortran/rwlock_3.f90   -O3 -fomit-frame-pointer 
-funroll-loops -fpeel-loops -ftracer -finline-functions  (test for 
excess errors)
PASS: libgomp.fortran/rwlock_3.f90   -O3 -fomit-frame-pointer 
-funroll-loops -fpeel-loops -ftracer -finline-functions  execution test

Lipeng Zhu