I also tested on Windows with 2 threads, but got no deadlock.

My trace shows:

Multithread performance: insno:  -1  thread 1 of 2 starting.
Multithread performance: insno:  -1  thread 0 of 2 starting.

Your trace shows:

Multithread performance: insno:  -1  thread -1 of 2 starting.
Multithread performance: insno:  -1  thread 0 of 2 starting.

I think this is a clue to the problem. I have attached my test csd. Try it and see if you still get the "thread -1". We already know that the partitioning arithmetic is no good for odd numbers of instances. That should be easy to fix, but I haven't done it yet.

Thanks for looking at this,

Mike




-----Original Message-----
>From: Felipe Sateler 
>Sent: Jun 21, 2008 4:42 AM
>To: Developer discussions 
>Subject: Re: [Cs-dev] Multi-threading
>
>I'm still getting a deadlock. Scons said that the pthread_barrier_* functions 
>are available. This is the output:
>
>% OPCODEDIR=. ./csound -o/dev/null -v --num-threads=2 examples/am.csd
>time resolution is 1000.000 ns
>PortAudio real-time audio module for Csound
>0dBFS level = 32768.0
>Csound version 5.08.91 beta (float samples) Jun 21 2008
>libsndfile-1.0.17
>decode_long num-threads=2
>Reading options from $HOME/.csoundrc
>UnifiedCSD:  examples/am.csd
>Environment variable 'SADIR' has been set 
>to '/home/felipe/src/deb/csound/cvs/csound5/examples'
>Environment variable 'SSDIR' has been set 
>to '/home/felipe/src/deb/csound/cvs/csound5/examples'
>Environment variable 'INCDIR' has been set 
>to '/home/felipe/src/deb/csound/cvs/csound5/examples'
>Environment variable 'MFDIR' has been set 
>to '/home/felipe/src/deb/csound/cvs/csound5/examples'
>STARTING FILE
>Creating options
>Creating orchestra
>Creating score
>decode_long num-threads=2
>orchname:  /tmp/fileuDdaTd.orc
>scorename: /tmp/fileXx5eNw.sco
>rtaudio: PortAudio module enabled ... using blocking interface
>orch compiler:
>8 lines read
>LINE 7:
>LINE 8:
>groups: instr 1
>        endin
>pmax 3, kcnt 0, wcnt 0, acnt 0, pcnt 0, scnt 0
>        instr   1
>LINE 9:
>groups: k2 oscil 10000 1 1
>modified opcod: oscil.kk
>treqd o, tfound c
>treqd i, tfound c
>treqd k, tfound c
>treqd k, tfound c
>xincod = 0
>treqd s, tfound k
>k2      oscil.kk        10000   1       1       0
>LINE 10:
>groups: a1 oscil 10000+k2 120 1
>10000   +       k2
>10000   k2      +
>treqd k, tfound k
>treqd k, tfound c
>xincod = 0
>treqd k, tfound k
>#k0     add.kk  10000   k2
>modified opcod: oscil.kk
>treqd o, tfound c
>treqd i, tfound c
>treqd k, tfound c
>treqd k, tfound k
>xincod = 0
>treqd s, tfound a
>a1      oscil.kk        #k0     120     1       0
>LINE 11:
>groups: chano a1 1
>modified opcod: chano.a
>treqd k, tfound c
>treqd a, tfound a
>xincod = 1
>        chano.a a1      1
>LINE 12:
>groups: out a1
>treqd a, tfound a
>xincod = 1
>        out     a1
>LINE 13:
>groups: endin
>        endin
>pmax 3, kcnt 2, wcnt 0, acnt 1, pcnt 0, scnt 0
>LINE 14:
>poolcount = 4, strpool_cnt = 1
>pool:   0       1       10000   120
>strpool:        ""
>oscil.kk argndxs:       -4      3       2       2       1
>add.kk argndxs: -5      3       -4
>oscil.kk argndxs:       -6      -5      4       2       1
>chano.a argndxs:        -6      2
>out argndxs:    -6
>argoff array:
>        0       1       -4      4       3       2       2       1       
>1      -5       2       3       -4      1       -6      4       -5      4       
>2      12       -6      2       1       -6
>Elapsed time at end of orchestra compile: real: 0.010s, CPU: 0.010s
>sorting score ...
>        ... done
>Elapsed time at end of score sort: real: 0.010s, CPU: 0.010s
>Multithread performance: insno:  -1  thread -1 of 2 starting.
>Multithread performance: insno:  -1  thread 0 of 2 starting.
>Csound version 5.08.91 beta (float samples) Jun 21 2008
>Creating search path cache for 'SNAPDIR':
>Creating search path cache for 'SFDIR;SSDIR;INCDIR':
>    1: "/home/felipe/src/deb/csound/cvs/csound5/examples/"
>Creating search path cache for 'SFDIR':
>Creating search path cache for 'SADIR':
>    1: "/home/felipe/src/deb/csound/cvs/csound5/examples/"
>Creating search path cache for 'SFDIR;SSDIR':
>    1: "/home/felipe/src/deb/csound/cvs/csound5/examples/"
>displays suppressed
>esr = 44100.0, ekr =  4410.0, ksmps = 10, nchnls = 1 0dbfs = 32768.0
>lclabeg 4, lclsbeg 4
>lclabeg 6, lclsbeg 7
>0dBFS level = 32768.0
>instr 0 allocated at 0x1f21b30
>        lclbas 0x1f21be8, opds 0x1f21be8
>orch now loaded
>audio buffered in 1024 sample-frame blocks
>writing 2048-byte blks of shorts to /dev/null (WAV)
>SECTION 1:
>ftable 1:
>activating instr 1
>new alloc for instr 1:
>instr 1 allocated at 0x1f46de0
>        lclbas 0x1f46e98, opds 0x1f46ec8
>op 247 (oscil.kk) allocated at 0x1f46ec8
>opadr = 0x498f71
>argptrs:        0x1f46e98       0x1ef80f8       0x1ef80f4       0x1ef80f4      
>0x1ef80f0
>op 74 (add.kk) allocated at 0x1f46f30
>opadr = 0x455250
>argptrs:        0x1f46e9c       0x1ef80f8       0x1f46e98
>op 247 (oscil.kk) allocated at 0x1f46f78
>opadr = 0x499062
>argptrs:        0x1f46ea0       0x1f46e9c       0x1ef80fc       0x1ef80f4      
>0x1ef80f0
>op 658 (chano.a) allocated at 0x1f46fe0
>opadr = 0x45c546
>argptrs:        0x1f46ea0       0x1ef80f4
>op 293 (out) allocated at 0x1f47020
>opadr = 0x4597c8
>argptrs:        0x1f46ea0
>psave beg at 0x1f46e8c
>   ending at 0x1f46e98
>init oscil.kk:
>init oscil.kk:
>instr 1 now active:
>insno   instanc nxtinst prvinst nxtact  prvact  nxtoff  actflg  offtim
>0       0x1f21b30       (nil)   (nil)   (nil)   (nil)   (nil)   1       0.0
>1       0x1f46de0       (nil)   (nil)   (nil)   0x1eef8c8       (nil)   1      
>30.0
>kperf:            insno:   1                        start: 0x0x1f46de0  end: 
>0x(nil)
>
>
>
>
>
>
>El 21/06/08 04:37 Michael Gogins escribió:
>> I've got this sort of working, or at least not crashing, on Windows with
>> the following changes, which are booked in:
>>
>> -- Use of native pthread_barrier_* functions, where they exist on the build
>> system.
>>
>> -- Much improved diagnostic messages (visible with -v command-line option).
>>
>> Performance completes and the code, which I now understand, seems somewhat
>> correct. But there are still problems:
>>
>> -- Multi-threaded performance still takes somewhat longer than
>> single-threaded performance. It may be that this is due to the overhead of
>> barrier calls and threads not being outweighed by the actual savings on
>> instrument computations in my test csds. Or it may be that there are subtle
>> bugs in the multi-threading code such that things are not actually
>> proceeding concurrently, or are doubling, or something (see below).
>>
>> -- Multi-threaded soundfiles show odd increases in amplitude in places,
>> suggesting that too many instances are computing audio at the same time in
>> some places, but not in other places.
>>
>> -- The partitioning is not correct. If I have 5 active instances, I get 2
>> partitions with 2 instances each. It should be 2 instances in one
>> partition, and 3 instances in another partition.
>>
>> -- As I noted before, we will need to protect writes to shared buffers,
>> especially spout in OOps/aops.c.
>>
>> Regards,
>> Mike
>>
>>
>>
>>
>>
>>
>> -------------------------------------------------------------------------
>> Check out the new SourceForge.net Marketplace.
>> It's the best place to buy or sell services for
>> just about anything Open Source.
>> http://sourceforge.net/services/buy/index.php
>> _______________________________________________
>> Csound-devel mailing list
>> Csound-devel@lists.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/csound-devel
>
>
>
>Saludos,
>Felipe Sateler