I saw that some of you guys seem to be using the sox multi-threaded
switch.
Let me contribute some benchmarking I did on my i5 NUC:
w/o multithreading:
Code:
--------------------
Performance counter stats for '/tmp/sox -t flac /tmp/test.flac -t wavpcm -b
24 /tmp/tmp.wav rate -v -b 90.4 -p 45 -a 352800':
6.501,16 msec task-clock # 1,000 CPUs utilized
58 context-switches # 8,922 M/sec
0 cpu-migrations # 0,000 K/sec
710 page-faults # 109,214 M/sec
17.471.508.084 cycles # 2687510,857 GHz
49.865.587.231 instructions # 2,85 insn per cycle
3.663.382.219 branches # 563510570,528 M/sec
11.861.538 branch-misses # 0,32% of all branches
6,502358582 seconds time elapsed
6,098182000 seconds user
0,402946000 seconds sys
--------------------
with multithreading:
Code:
--------------------
Performance counter stats for '/tmp/sox --multi-threaded -t flac
/tmp/test.flac -t wavpcm -b 24 /tmp/tmp.wav rate -v -b 90.4 -p 45 -a 352800':
27.121,26 msec task-clock # 3,955 CPUs utilized
823 context-switches # 30,345 M/sec
13 cpu-migrations # 0,479 M/sec
764 page-faults # 28,170 M/sec
67.649.699.828 cycles # 2494365,983 GHz
65.604.260.310 instructions # 0,97 insn per cycle
8.144.344.634 branches # 300296620,110 M/sec
13.589.463 branch-misses # 0,17% of all branches
6,857757586 seconds time elapsed
26,580257000 seconds user
0,542025000 seconds sys
--------------------
multithreading is not only a little - around 4% - slower (if we ignore
the accumulated processing time >400% ). It uses 4 CPUs and ten
thousands of extra cycles to achieve that. ;)
At least based on these results based on my real live resampling
testcase I'd stay away from sox multithreading!
Enjoy.
:::' my audioblog - latest series: RaspBerry PI - \"The Audio Engine\"
' (http://soundcheck-audio.blogspot.com):::
------------------------------------------------------------------------
soundcheck's Profile: http://forums.slimdevices.com/member.php?userid=34383
View this thread: http://forums.slimdevices.com/showthread.php?t=105309
_______________________________________________
Squeezecenter mailing list
[email protected]
http://lists.slimdevices.com/mailman/listinfo/squeezecenter
