Thread: BUG #17185: PostgreSQL performance GNU vs LLVM

BUG #17185: PostgreSQL performance GNU vs LLVM

From
PG Bug reporting form
Date:
The following bug has been logged on the website:

Bug reference:      17185
Logged by:          arjun shetty
Email address:      arjunshetty955@gmail.com
PostgreSQL version: 13.4
Operating system:   RHEL8.4
Description:

TPCH Results(Query Response Time(Q1 to Q22):
-------------
VU    GCC    Clang         clang-with-llvm        Clang(make file code removed -flto
=thin to flto)        
vu            
1    7185    6349    4410                9984
2    7302    7352    4307                9990
3    6233    6749    4544                9698
4    6701    6328    3701                10878
5    7105    7376    5829                10991


Note: clang-with-llvm: JIT OFF.


TPCC Results(NOPM)
------------------
vu               GCC               Clang       Clang-with-llvm
Clang(make file code removed -flto =thin to flto(regluar flto)
16        921489        967234        946472                963300
52        1655840        1779333        1683924                1713926
68        1765030        1787377        1810060                1835825
170        1894382        1986746        2034823                1938886
192        1739617        1901906        1929656                1912524
210        1610512        1801849        1674440                1715523
230        1830690        1825760        1799097                1708821
            
1.Regaular LTO not performs better than Thin LTO ? is regular LTO not
supported in postgresql? 
2.Is specific reason need to use -flto=thin(clan13)?
3.TPCH(Clang-llvm)performs better than GCC/Clang(others) but why
TPCC(Clang-llvm) not performs better than GCC/Clang-other (performance
deviation is less)

the results captured in bare Metal(HP Environment) and benchmark Env
HammerDbv4.2


Re: BUG #17185: PostgreSQL performance GNU vs LLVM

From
arjun shetty
Date:
Hi All
I would like to know any comments or inputs #17185
I would like to know TPC-C performance GCC vs Clang(llvm).(# reference the below mail)

Best regards 
Arjun

On Wednesday, September 8, 2021, PG Bug reporting form <noreply@postgresql.org> wrote:
The following bug has been logged on the website:

Bug reference:      17185
Logged by:          arjun shetty
Email address:      arjunshetty955@gmail.com
PostgreSQL version: 13.4
Operating system:   RHEL8.4
Description:       

TPCH Results(Query Response Time(Q1 to Q22):
-------------
VU      GCC    Clang         clang-with-llvm            Clang(make file code removed -flto
=thin to flto)         
vu                     
1       7185    6349    4410                            9984
2       7302    7352    4307                            9990
3       6233    6749    4544                            9698
4       6701    6328    3701                            10878
5       7105    7376    5829                            10991


Note: clang-with-llvm: JIT OFF.


TPCC Results(NOPM)
------------------
vu              GCC                   Clang     Clang-with-llvm             
Clang(make file code removed -flto =thin to flto(regluar flto)
16              921489          967234          946472                          963300
52              1655840         1779333         1683924                         1713926
68              1765030         1787377         1810060                         1835825
170             1894382         1986746         2034823                         1938886
192             1739617         1901906         1929656                         1912524
210             1610512         1801849         1674440                         1715523
230             1830690         1825760         1799097                         1708821

1.Regaular LTO not performs better than Thin LTO ? is regular LTO not
supported in postgresql?
2.Is specific reason need to use -flto=thin(clan13)?
3.TPCH(Clang-llvm)performs better than GCC/Clang(others) but why
TPCC(Clang-llvm) not performs better than GCC/Clang-other (performance
deviation is less)

the results captured in bare Metal(HP Environment) and benchmark Env
HammerDbv4.2

Re: BUG #17185: PostgreSQL performance GNU vs LLVM

From
"Euler Taveira"
Date:
On Mon, Sep 13, 2021, at 10:59 AM, arjun shetty wrote:
Hi All
I would like to know any comments or inputs #17185
I would like to know TPC-C performance GCC vs Clang(llvm).(# reference the below mail)
This mailing list is used to report bugs. Since this is *not* a bug, I
recommend that you ask these questions on -performance or -hackers.



--
Euler Taveira