Annotation of www/audio-port.html, Revision 1.5
1.1 espie 1: <html>
2: <head>
3: <meta http-equiv="Content-Type"
4: content="text/html; charset=iso-8859-1">
5: <meta name="resource-type"
6: content="document">
7: <meta name="description"
8: CONTENT="How to make an OpenBSD port; audio">
9: <meta name="keywords"
10: content="openbsd,ports,audio">
11: <meta name="distribution"
12: content="global">
13: <meta name="copyright"
1.5 ! jufi 14: content="This document copyright 1998-2001 by OpenBSD.">
1.1 espie 15: <title>Porting audio applications to OpenBSD</title>
16: <link rev="made" HREF="mailto:www@openbsd.org">
17: </head>
18: <body text="#000000" bgcolor="#FFFFFF" link="#23238E">
1.3 jufi 19: <img height=30 width=141 src=images/smalltitle.gif alt="[OpenBSD]">
1.1 espie 20:
21: <h1>Porting audio applications to OpenBSD</h1>
22:
23: <p>
24: This document currently deals with sampled sounds issues only. Contributions
25: dealing with synthesizers and waveform tables are welcome.
26:
27: </p>
28:
29: Audio applications tend to be hard to port, as this is a domain where
30: interfaces are not standardized at all, though approaches don't vary
31: much between operating systems.
32:
33:
34: <h2><font color=#e00000>Using <code>ossaudio</code></font></h2>
35:
36: The <code>ossaudio</code> emulation is possibly the simplest way, but
37: it won't always work, and it is not such a great idea usually.
38: <ul>
39: <li>It redefines <code>ioctl</code>. If the code to port uses
40: <code>ioctl</code> for more than audio, you will have to
41: <code>#undef ioctl</code> and use the bare form with
42: <code>_ossioctl</code>.
43:
44: <li>Some features of linux sound are not emulated.
45:
46: <li>Applications with correct linux sound support that is not
47: Intel-specific tend to use these features.
48:
49: </ul>
50:
51: <h2><font color=#e00000>Using existing NetBSD or FreeBSD code</font></h2>
52: Since we share part of the audio interface with NetBSD and FreeBSD,
53: starting from a NetBSD port is reasonable. Be aware that some files
54: changed places, and that some entries in <code>sys/audioio.h</code>
55: are obsolete. Also, many ports tend to be incorrectly coded and to
56: work on only one type of machine. Some changes are bound to be
57: necessary, though. Read through the next part.
58:
59: <h2><font color=#e00000>Writing OpenBSD code</font></h2>
60: <h3><font color=#0000e0>Hardware independence</font></h3>
61:
62: <p>
63: <strong>YOU SHOULDN'T ASSUME ANYTHING ABOUT THE AUDIO HARDWARE USED.
64: </strong><br>
65: Wrong code is code that only checks the <code>a_info.play.precision</code>
66: field against 8 or 16 bits, and assumes unsigned or signed samples based
67: on soundblaster behavior. You should check the sample type explicitly,
68: and code according to that. Simple example:
69: <pre>
70: AUDIO_INIT_INFO(&a_info);
71: a_info.play.encoding = AUDIO_ENCODING_SLINEAR;
72: a_info.play.precision = 16;
73: a_info.play.sample_rate = 22050;
74: error = ioctl(audio, AUDIO_SETINFO, &a_info);
75: if (error)
76: /* deal with it */
77: error = ioctl(audio, AUDIO_GETINFO, &a_info);
78: switch(a_info.play.encoding)
79: {
80: case AUDIO_ENCODING_ULINEAR_LE:
81: case AUDIO_ENCODING_ULINEAR_BE:
82: if (a_info.play.precision == 8)
83: /* ... */
84: else
85: /* ... */
86: break;
87: case ...
88:
89: default:
90: /* don't forget to deal with what you don't know !!! For instance, */
91: fprintf(stderr,
92: "Unsupported audio format (%d), ask ports@ about that\n",
93: a_info.play.encoding);
94:
95: }
96: /* now don't forget to check what sampling frequency you actually got */
97: </pre>
98:
99: </p>
100: This is about the smallest code fragment that will deal with most issues.
101:
102: <h3><font color=#0000e0>16 bit formats and endianess</font></h3>
103: In normal usage, you just ask for an encoding type (e.g.,
104: <code>AUDIO_ENCODING_SLINEAR</code>, and you retrieve
105: an encoding with endianess (e.g., <code>AUDIO_ENCODING_SLINEAR_LE</code>).
106: Considering that a soundcard does not have to use the same endianess
107: as your platform, you should be prepared to deal with that.
108: The easiest way is probably to prepare a full audio buffer, and to use
109: <code>swab(3)</code> if an endianess change is required.
110: Dealing with external samples usually amounts to:
111: <ol>
112: <li>Parsing the sample format,
113: <li>Getting the sample in,
114: <li>Swapping endianess if it is not your native format,
115: <li>Computing what you want to output into a buffer,
116: <li>Swapping endianess if the sound card is not in your native format,
117: <li>Playing the buffer.
118: </ol>
119: Obviously, you may be able to remove steps 3 and 5 if you are simply
120: playing a sound sample which happens to be in your sound card native
121: format.
122:
123: <h3><font color=#0000e0>Audio quality</font></h3>
124: <p>
125: Hardware may have some weird limitations, such as being unable to get
126: over 22050 Hz in stereo, but up to 44100 in mono. In such cases, you
127: should give the user a change to state his preferences, then try your
128: best to give the best performance possible. For instance, it is stupid
129: to limit the frequency to 22050 Hz because you are outputting stereo.
130: What if the user does not have a stereo sound system connected to his
131: audio card output ?
132: </p>
133:
134: <p>
135: It is also stupid to hardcode soundblaster-like limitations into your
136: program. You should be aware of these, but do try to get over the
137: 22050 Hz/stereo barrier and check the results.
138: </p>
139:
140: <h4>Sampling frequency</h4>
141: You should definitely check the sampling frequency your card gives you
142: back. A 5% discrepancy already amounts to a half-tone, and some people
143: have much more accurate hearing than that, though most of us won't
144: notice a thing. Your application should be able to perform
145: resampling on the fly, possibly naively, or through devious
146: applications of Shannon's resampling formula if you can.
147:
148: <h4>Dynamic range</h4>
149: <p>
150: Samples don't always use the full range of values they could. First,
151: samples recorded with a low gain will not sound very loud on the
152: machine, forcing the user to turn the volume up.
153: Second, on machines with badly isolated audio, low sound output means
154: you mostly hear your machine heart-beat, and not the sound you expected.
155: Finally, dumb conversion from 16 bits to 8 bits may leave you with only
156: 4 bits of usable audio, which makes for an awfully bad quality.
157: </p>
158: <p>
159: If possible, the best solution is probably to scan the whole stream
160: you are going to play ahead of time, and to scale it so that it fits
161: the full dynamic range. If you can't afford that, but you can manage
162: to get a bit of look-ahead on what you're going to play, you can
163: adjust the volume boost on the fly, you just have to make sure
164: that the boost factor stays at a low frequency compared to the
1.3 jufi 165: sound you want to play, and that you get absolutely <em>no
166: overflows</em> -- those will always sound much worse than the
1.1 espie 167: improvement you're trying to achieve.<br>
168: As sound volume perception is logarithmic, using arithmetic shifts is usually
169: enough. If your data is signed, you should explicitly code the shift as
170: a division, as C <code>>></code> operator is not portable on
171: signed data.
172: </p>
173: <p>
174: If all else fails, you should at least try to provide the user with
175: a volume scaling option.
176: </p>
177:
178: <h3><font color=#0000e0>Audio performance</font></h3>
179: <p>
180: Low-end applications usually don't have much to worry about. Keep in
181: mind that some of us do use OpenBSD on low-end 68030, and that if a
182: sound application can run on that, it should.
183: </p>
184:
185: <p>
186: Don't forget to run benches. Theoretical optimizations are just that:
187: theoretical. Some hard figures should be collected to check what's a
188: sizeable improvement, and what's not.
189: </p>
190:
191: <p>
192: For high performance audio applications, such as mpegI-layer3, some
193: points should be taken into account:
194: <ul>
195: <li>The audio interface does provide you with the natural hardware
196: blocksize. Using multiples of that for your output buffer is
197: essential. Keep in mind that <code>write</code>, as a system call,
198: incurs a high cost compared to internal audio processing.
199:
200: <li>Bandwidth is a very important factor when dealing with audio.
201: A useful way to optimize an audio player is to see it as a
202: decompressor. The longer you can keep with the compressed data, the
203: better usually. Very short loops that do very little processing are
204: usually a bad idea. It is generally much better to combine all
205: processing into one loop.
206:
207: <li>Some formats do incur more overhead than others. The
208: <code>AUDIO_GETENC</code> <code>ioctl</code> should be used
209: to retrieve all formats that the audio device provides.
210: Be especially aware of the
211: <code>AUDIO_ENCODINGFLAG_EMULATED</code> flag. If your
212: application is already able to output all kinds of weird formats,
213: and reasonably optimized for that, try to use a native format at
214: all costs. On the other hand, the emulation code present in the
215: audio device can be assumed to be reasonably optimal, so don't
216: replace it with quickly hacked up code.
217: </ul>
218: </p>
219:
220: <p>A model you may have to follow to get optimal results is to first
221: compile a small test program that enquires about the specific audio
222: hardware available, then proceed to configure your program so that it
223: deals optimally with this hardware. You may reasonably expect people
224: who want good audio performance to recompile your port when they change
225: hardware, provided it makes a difference.
226: </p>
227:
228: <h3><font color=#0000e0>Real time or synchronized</font></h3>
229: <p>
230: Considering that OpenBSD is not real time, you may still wish to write
231: audio applications that are mostly real time, for instance games. In
232: such a case, you will have to lower the blocksize so that the sound
233: effects don't get out of synch with the current game. The problem
234: with this if that the audio device may get starved, which yields
235: horrible results.
236: </p>
237: <p>
238: In case you simply want audio to be synchronized with some graphics
239: output, but the behavior of your program is predictable, synchronization
240: is easier to achieve. You just play your audio samples, and ask the
241: audio device what you are currently playing with
242: <code>AUDIO_GETOOFFS</code>, then use that information to
243: post-synchronize graphics. Provided you ask sufficiently often (say,
244: every tenth of a second), and as long as you have enough horse-power to
245: run your application, you can get very good synchronization that way.
246: You might have to tweak the figures by a constant offset, as there is
247: some lag between what the audio reports, what's currently playing, and
248: the time it takes for XWindow to display something.
249: </p>
250: <h2><font color=#e00000>Contributing code back</font></h2>
251: <p>In the case of audio applications, working with the original program's
252: author is very important. If his code does only work with soundblaster
253: cards for instance, there is a good chance he will have to cope with
254: other technology soon.
255: </p>
256:
257: <p>
258: <strong>If you don't sent your comments to him by then, your work will
259: have been useless</strong>.</p>
260:
261: It may also be that the author has already noticed whatever problems
262: you are currently dealing with, and is addressing them in his current
263: development tree. If the patches you are writing amount to more than
264: a handful of lines, cooperation is almost certainly a very good idea.
265:
266:
267: <hr>
268: <a href="porting.html"><img height=24 width=24 src=back.gif
269: border=0 alt=Porting></a>
270: <a href=mailto:www@openbsd.org>www@openbsd.org</a>
1.5 ! jufi 271: <br><small>$OpenBSD: audio-port.html,v 1.4 2001/08/25 11:25:42 jufi Exp $</small>
1.1 espie 272: </body>
273: </html>