Re: [RFC PATCH 00/10] x86: undwarf unwinder

[Jiri Slaby]

On 06/01/2017, 03:58 PM, Jiri Slaby wrote:
> On 06/01/2017, 03:50 PM, Ingo Molnar wrote:
>> That's not what I meant! The speedup comes from (hopefully) being able to disable 
>> CONFIG_FRAME_POINTER, which:
> 
> BTW when you are mentioning this -- my measurements were with FP disabled.
> 
> Is there any reasonably simple-to-use benchmark I could run with FP=y
> and FP=n quickly?

Nevermind, I tried 10000000 stack unwindings several times and ran
netperf too. Both on the same virtual machine. On these microbenchmarks,
the former performs ~ 1.03 times better, the latter ~ 1.3 times. When
Mel measured the difference, it was around 10 % overall using more
sophisticated benchmarks.

With FP=n:

# time echo 10000000 > /dev/test_dwarf
real    0m6.659s
user    0m0.000s
sys     0m6.655s

# for aa in 1 0 0 0 0; do netperf -P $aa ; done
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to
localhost () port 0 AF_INET
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

 87380  16384  16384    10.00    47819.30
 87380  16384  16384    10.00    41991.01
 87380  16384  16384    10.00    43607.82
 87380  16384  16384    10.00    42208.44
 87380  16384  16384    10.00    44383.92




With FP=y:

# time echo 10000000 > /dev/test_dwarf
real    0m6.869s
user    0m0.000s
sys     0m6.868s

# for aa in 1 0 0 0 0; do netperf -P $aa ; done
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to
localhost () port 0 AF_INET
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

 87380  16384  16384    10.00    37807.90
 87380  16384  16384    10.00    32246.67
 87380  16384  16384    10.00    31358.76
 87380  16384  16384    10.00    32450.00
 87380  16384  16384    10.00    31326.70





thanks,
-- 
js
suse labs