No subject

Thu Aug 23 10:31:50 CEST 2012

In my case, the recompiled Pd would not start because it could not<br>
find After installing libportaudio2 via Synaptic,<br>
&#39;normalized&#39; Pd would finally start.<br>
It is no problem to have the regular Pd still installed. Maybe you can<br>
install the local Pd over the regular Pd using the gnumakefile. Didn&#39;t<=
try that, I don&#39;t like to install things without package manager.<br>
<span class=3D"HOEnZb"><font color=3D"#888888"><br>
</font></span><div class=3D"HOEnZb"><div class=3D"h5"><br>
On Fri, Jan 25, 2013 at 3:41 PM, Julian Brooks &lt;<a href=3D"mailto:jbeeze=
z at">jbeezez at</a>&gt; wrote:<br>
&gt; Excuse my ignorance:<br>
&gt; not sure how to start the below version of pd on the rpi?<br>
&gt; I have the full path but then what?<br>
&gt; if I do (in command line)<br>
&gt; pd /place/where/new/pd/is/bin/pd<br>
&gt; It signals watchdog.<br>
&gt; I also still have regular pd 0.44.0 installed btw.<br>
&gt; Sorry if this is dumb dumb dumb dumb Duuummmbbb.<br>
&gt; Jb<br>
&gt; On 24 January 2013 09:14, katja &lt;<a href=3D"mailto:katjavetter at gmai=">katjavetter at</a>&gt; wrote:<br>
&gt;&gt; &#39;Undenormalized&#39; Pd build for Raspberry Pi is temporarily =
parked here<br>
&gt;&gt; for testing purposes (will be removed when Miller&#39;s release is=
&gt;&gt; in this sense):<br>
&gt;&gt; <a href=3D""=
&gt;&gt; This is a locally installed Pd, like Miller&#39;s distribution. Yo=
u can<br>
&gt;&gt; start it from command line with the full path to<br>
&gt;&gt; pd-0.44-0-normalized/bin/pd. It&#39;s not a .deb, so it can&#39;t =
be installed<br>
&gt;&gt; under supervision of package manager.<br>
&gt;&gt; Katja<br>
&gt;&gt; On Wed, Jan 23, 2013 at 9:15 PM, Julian Brooks &lt;<a href=3D"mail=
to:jbeezez at">jbeezez at</a>&gt; wrote:<br>
&gt;&gt; &gt; Hey Katja,<br>
&gt;&gt; &gt;<br>
&gt;&gt; &gt; Would you mind sharing the &#39;normalised&#39; Pd-0.44.0 for=
 RPi please.<br>
&gt;&gt; &gt;<br>
&gt;&gt; &gt; Cheers,<br>
&gt;&gt; &gt;<br>
&gt;&gt; &gt; Julian<br>
&gt;&gt; &gt;<br>
&gt;&gt; &gt;<br>
&gt;&gt; &gt;<br>
&gt;&gt; &gt; On 23 January 2013 18:23, katja &lt;<a href=3D"mailto:katjave=
tter at">katjavetter at</a>&gt; wrote:<br>
&gt;&gt; &gt;&gt;<br>
&gt;&gt; &gt;&gt; Now I recompiled the Pd-0.44.0 release on Raspberry Pi (t=
ook me a few<br>
&gt;&gt; &gt;&gt; hours, not only because Pi is so slow) with PD_BIGORSMALL=
 enabled for<br>
&gt;&gt; &gt;&gt; arm in m_pd.h. Using bigorsmalltest.pd from my previous m=
ail I<br>
&gt;&gt; &gt;&gt; verified that the macro is implemented indeed.<br>
&gt;&gt; &gt;&gt;<br>
&gt;&gt; &gt;&gt; Martin Brinkmann&#39;s patch chaosmonster1<br>
&gt;&gt; &gt;&gt; (<a href=3D"" target=3D"_bl=
ank"></a>) gives a beautiful illustration of =
&gt;&gt; &gt;&gt; improvement. This patch is full of filters and delay line=
s. At it&#39;s<br>
&gt;&gt; &gt;&gt; initial settings, there is no subnormals problem. But if =
you set the<br>
&gt;&gt; &gt;&gt; bottom slider to the right, it gets silent. With Pd-0.44-=
0 release,<br>
&gt;&gt; &gt;&gt; CPU load explodes. With the &#39;normalized&#39; Pd, noth=
ing special happens.<br>
&gt;&gt; &gt;&gt;<br>
&gt;&gt; &gt;&gt; And indeed, the PD_BIGORSMALL conditional checks come for=
 free: with<br>
&gt;&gt; &gt;&gt; initial settings of the chaosmonster1, performance is equ=
ivalent in<br>
&gt;&gt; &gt;&gt; both Pd&#39;s. Cool! Hopefully this is similar on armv7.<=
&gt;&gt; &gt;&gt;<br>
&gt;&gt; &gt;&gt; Katja<br>
&gt;&gt; &gt;&gt;<br>
&gt;&gt; &gt;&gt;<br>
&gt;&gt; &gt;&gt;<br>
&gt;&gt; &gt;&gt; On Wed, Jan 23, 2013 at 5:01 PM, Hans-Christoph Steiner &=
lt;<a href=3D"mailto:hans at">hans at</a>&gt;<br>
&gt;&gt; &gt;&gt; wrote:<br>
&gt;&gt; &gt;&gt; &gt;<br>
&gt;&gt; &gt;&gt; &gt; hey Katya,<br>
&gt;&gt; &gt;&gt; &gt;<br>
&gt;&gt; &gt;&gt; &gt; This also sounds like good evidence for your idea of=
 writing C code<br>
&gt;&gt; &gt;&gt; &gt; that<br>
&gt;&gt; &gt;&gt; &gt; modern compilers optimize well. =A0Using unions for =
aliasing allows the<br>
&gt;&gt; &gt;&gt; &gt; compiler<br>
&gt;&gt; &gt;&gt; &gt; to do all the new tricks, then writing loops that au=
&gt;&gt; &gt;&gt; &gt; gives<br>
&gt;&gt; &gt;&gt; &gt; us the<br>
&gt;&gt; &gt;&gt; &gt; real benefits. =A0Also, I think we can see some gain=
s by using memcpy()<br>
&gt;&gt; &gt;&gt; &gt; since on<br>
&gt;&gt; &gt;&gt; &gt; modern libc version, those are highly optimized for =
the given CPU,<br>
&gt;&gt; &gt;&gt; &gt; dynamically<br>
&gt;&gt; &gt;&gt; &gt; choosing the routines based on what instructions are=
&gt;&gt; &gt;&gt; &gt; memcpy<br>
&gt;&gt; &gt;&gt; &gt; will<br>
&gt;&gt; &gt;&gt; &gt; use things like SSSE2 if its available.<br>
&gt;&gt; &gt;&gt; &gt;<br>
&gt;&gt; &gt;&gt; &gt; .hc<br>
&gt;&gt; &gt;&gt; &gt;<br>
&gt;&gt; &gt;&gt; &gt; On 01/23/2013 07:47 AM, katja wrote:<br>
&gt;&gt; &gt;&gt; &gt;&gt; Finally some good news on this topic. Earlier I =
stated that &#39;big or<br>
&gt;&gt; &gt;&gt; &gt;&gt; small tests&#39; are expensive for the Pi, but t=
hat is not by definition<br>
&gt;&gt; &gt;&gt; &gt;&gt; the case. There must have been other conditions =
blurring my<br>
&gt;&gt; &gt;&gt; &gt;&gt; impression. I&#39;ve now done a systematic test =
where other influences<br>
&gt;&gt; &gt;&gt; &gt;&gt; are<br>
&gt;&gt; &gt;&gt; &gt;&gt; ruled out. A test class [lopass~] with exactly t=
he same routine as<br>
&gt;&gt; &gt;&gt; &gt;&gt; [lop~] was made, but compiled with PD_BIGORSMALL=
() macro enabled. It<br>
&gt;&gt; &gt;&gt; &gt;&gt; was verified that [lopass~] is not affected by d=
&gt;&gt; &gt;&gt; &gt;&gt; Performance<br>
&gt;&gt; &gt;&gt; &gt;&gt; comparison of [lop~] and [lopass~] shows that bo=
th objects cause<br>
&gt;&gt; &gt;&gt; &gt;&gt; equivalent CPU load. Meaning, Raspberry Pi gives=
 the &#39;big or small<br>
&gt;&gt; &gt;&gt; &gt;&gt; checks&#39; for free! At least in the case of th=
is simple filter. Please<br>
&gt;&gt; &gt;&gt; &gt;&gt; try attached on the Pi to see=
 if I&#39;m not<br>
&gt;&gt; &gt;&gt; &gt;&gt; dreaming.<br>
&gt;&gt; &gt;&gt; &gt;&gt;<br>
&gt;&gt; &gt;&gt; &gt;&gt; While I was at the topic anyway, I also tried a =
big or small test<br>
&gt;&gt; &gt;&gt; &gt;&gt; with<br>
&gt;&gt; &gt;&gt; &gt;&gt; union instead of direct type aliasing. It has th=
e advantage that the<br>
&gt;&gt; &gt;&gt; &gt;&gt; compiler can apply strict aliasing rules. This t=
est with unions did<br>
&gt;&gt; &gt;&gt; &gt;&gt; not cause extra CPU load either on the Pi. If yo=
u want to verify<br>
&gt;&gt; &gt;&gt; &gt;&gt; this<br>
&gt;&gt; &gt;&gt; &gt;&gt; result, enable the call to bigorsmall() instead =
&gt;&gt; &gt;&gt; &gt;&gt; lopass~.c and recompile.<br>
&gt;&gt; &gt;&gt; &gt;&gt;<br>
&gt;&gt; &gt;&gt; &gt;&gt; The fact that these tests do not cause extra CPU=
 load, indicate that<br>
&gt;&gt; &gt;&gt; &gt;&gt; they are done in parallel with other instruction=
s. Float and int<br>
&gt;&gt; &gt;&gt; &gt;&gt; registers are apparently strictly separated on a=
rmv6, there&#39;s no<br>
&gt;&gt; &gt;&gt; &gt;&gt; such<br>
&gt;&gt; &gt;&gt; &gt;&gt; thing like Intel&#39;s xmm registers or armv7&#3=
9;s NEON. As it happens, the<br>
&gt;&gt; &gt;&gt; &gt;&gt; big or small tests are done on ints, aliases of =
the floats that must<br>
&gt;&gt; &gt;&gt; &gt;&gt; be tested. Initially I assumed that the transpor=
t of floats from vfp<br>
&gt;&gt; &gt;&gt; &gt;&gt; to the arm integer processor would be expensive,=
 but if the<br>
&gt;&gt; &gt;&gt; &gt;&gt; instructions are done simultaneously it may be a=
n advantage instead.<br>
&gt;&gt; &gt;&gt; &gt;&gt; Another thing is that ARM implements branch pred=
ication instead of<br>
&gt;&gt; &gt;&gt; &gt;&gt; branch prediction. Those terms look almost the s=
ame but the routines<br>
&gt;&gt; &gt;&gt; &gt;&gt; are very different. Predication is when instruct=
ions for both<br>
&gt;&gt; &gt;&gt; &gt;&gt; branches<br>
&gt;&gt; &gt;&gt; &gt;&gt; are executed, and the wrong result is simply dis=
carded later.<br>
&gt;&gt; &gt;&gt; &gt;&gt;<br>
&gt;&gt; &gt;&gt; &gt;&gt; Conclusions from the limited test with [lop~] an=
d [lopass~] do not<br>
&gt;&gt; &gt;&gt; &gt;&gt; mean that all sorts of conditional checks are ch=
eap on the Pi, or on<br>
&gt;&gt; &gt;&gt; &gt;&gt; ARM in general. If PD_BIGORSMALL is enabled for =
RPi using<br>
&gt;&gt; &gt;&gt; &gt;&gt; compile-time<br>
&gt;&gt; &gt;&gt; &gt;&gt; definition __arm__, it will also hold for armv7,=
 but it may have<br>
&gt;&gt; &gt;&gt; &gt;&gt; very<br>
&gt;&gt; &gt;&gt; &gt;&gt; different result there. At the moment I have no =
access yet to an<br>
&gt;&gt; &gt;&gt; &gt;&gt; armv7<br>
&gt;&gt; &gt;&gt; &gt;&gt; device. Maybe someone can recompile test class [=
lopass~] and do the<br>
&gt;&gt; &gt;&gt; &gt;&gt; tests on Beagleboard or Cubieboard? Otherwise I =
may be able to do it<br>
&gt;&gt; &gt;&gt; &gt;&gt; on my friend&#39;s PengPod when that has arrived=
&gt;&gt; &gt;&gt; &gt;&gt;<br>
&gt;&gt; &gt;&gt; &gt;&gt; Katja<br>
&gt;&gt; &gt;&gt; &gt;&gt;<br>
&gt;&gt; &gt;&gt; &gt;&gt;<br>
&gt;&gt; &gt;&gt; &gt;&gt; On Tue, Jan 22, 2013 at 8:54 PM, Miller Puckette=
 &lt;<a href=3D"mailto:msp at">msp at</a>&gt;<br>
&gt;&gt; &gt;&gt; &gt;&gt; wrote:<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt; thanks - I&#39;d better try this and find ou=
t what&#39;s going on :)<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt; M<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt; On Mon, Jan 21, 2013 at 11:54:29AM +0100, ka=
tja wrote:<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; Tried the 0.44.0 build from your website=
. It has the same issue<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; with<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; subnormal values. My test patch is with =
[lop~]. If inf or nan is<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; fed<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; into [lop~], these &#39;values&#39; keep=
 circulating in the object, it can<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; no<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; longer process normal signal values.<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt;<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; I also tried my reverb stuff with specif=
ic compiler options for<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; Pi&#39;s<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; processor:<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt;<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; -march=3Darmv6zk<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; -mcpu=3Darm1176jzf-s<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; -mtune=3Darm1176jzf-s<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt;<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; With these options, gcc should be able t=
o decide that RunFast mode<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; is<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; permitted. But even in combination with =
-ffast-math (which in turn<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; sets -funsafe-math-optimizations and -fn=
o-trapping-math amongst<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; others), denormals are still there. I&#3=
9;m literally out of options<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; for<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; the moment. Sorry for not having better =
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt;<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt; Katja<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt;<br>
&gt;&gt; &gt;&gt; &gt;&gt;&gt;&gt;<br>
&gt;&gt; &gt;&gt;<br>
&gt;&gt; &gt;&gt; _______________________________________________<br>
&gt;&gt; &gt;&gt; <a href=3D"mailto:Pd-list at">Pd-list at</a> mail=
ing list<br>
&gt;&gt; &gt;&gt; UNSUBSCRIBE and account-management -&gt;<br>
&gt;&gt; &gt;&gt; <a href=3D"" t=
&gt;&gt; &gt;<br>
&gt;&gt; &gt;<br>


More information about the Pd-list mailing list