Doubles versus Floats

Csound’s developers have a strong interest in giving the program the best possible audio quality. In the early days, Csound worked internally with 16-bit audio. (See “The Orchestra Header” in Chapter 4, “The Structure and Syntax of a .csd File,” for more on the implications of this fact.) Versions that used 32-bit floating-point audio came along a few years later. Today, 32-bit versions are being supplanted by 64-bit versions. During the transition period, which lasted for several years, Csound was available for download in both 32-bit floating-point versions (indicated in the filename with “-f”) and 64-bit double-precision versions (indicated in the filename with “-d”).

To add to the confusion, the question of whether Csound is using 32-bit or 64-bit audio is entirely different from the question of whether it’s running on a 32-bit or 64-bit operating system.

At this writing, the Windows version of the installer (compiled by Michael Gogins) has transitioned to 64-bit audio. That is, you’ll install the double-precision (-d) version; no -f installer is currently available for Windows. The Mac OS installer (compiled by Victor Lazzarini) installs both the 32-bit and 64-bit versions. If you’re running Csound in Mac OS from the Terminal, the command csound will run the 32-bit floating-point version, while the command csound64 will run the double-precision version. However, this may change in the future.

During the transitional period, a few of Csound’s capabilities may not have been updated to work with the doubles version. According to Appendix H of the manual for version 5.13, this is the case with the lpanal and cvanal utilities, and there may be a few other items that won’t work either. If you encounter this problem and you’re using Windows, you may need to revert to an older version of Csound that is available for floats, or alternatively compile a floats version of Csound for yourself from source code.

You may also find that some front-end programs (as described in Chapter 9, “Front Ends”) require the doubles or floats version and won’t work if you have the wrong version installed.