From mpetrov@ox.ac.uk Mon Jan 02 12:46:09 1995 Path: ashpoly!cascade!transat!sunet!nordrelay!ox!mpetrov From: mpetrov@ox.ac.uk (Mikhail Petrov) Newsgroups: comp.dsp Subject: [FLAME] Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Jan02.124609.7511@ox.ac.uk> Date: Mon, 02 Jan 1995 12:46:09 GMT Organization: Oxford University Computing Laboratory NNTP-Posting-Host: ox.ac.uk X-Newsreader: WinVN 0.99.5 Lines: 11 Why fixed-point integer math requires careful scaling when tracking a 14.2 Hz carrier. This leaves over 95% of the DSP56001's 16.5 MIPS available for other audio tasks. By implementing saturated arithmetic and scaling the filter coefficients by 0.5, the loop runs in 18 instruction cycles per sample. Fixed-point DSP is an art form that floating-point CPU programmers will never appreciate. -- Mikhail Petrov "In theory there is no difference between theory and practice." PC-9821Ap2 (NEC) / TOS 4.04 From kcampbell@ox.ac.uk Mon Jan 02 19:27:09 1995 Path: ashpoly!cascade!bigsur!ox!kcampbell From: kcampbell@ox.ac.uk (Kirsty Campbell) Newsgroups: comp.dsp Subject: Re: [FLAME] Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Jan02.192709.6353@ox.ac.uk> Date: Mon, 02 Jan 1995 19:27:09 GMT References: <1995Jan02.124609.7511@ox.ac.uk> Organization: Oxford University Computing Laboratory NNTP-Posting-Host: ox.ac.uk X-Newsreader: TIN [version 1.2 PL2] Lines: 11 In article <1995Jan02.124609.7511@ox.ac.uk>, mpetrov@ox.ac.uk (Mikhail Petrov) writes: > Why fixed-point integer math requires careful scaling when tracking a 14.2 Hz carrier. > This leaves over 95% of the DSP56001's 16.5 MIPS available for other audio tasks. Beautiful assembly code! The 24-bit word size of the 56001 is tailor-made for high-precision audio filtering. My 14.4k connection dropped twice trying to fetch this thread. Keep articles concise! -- Kirsty Campbell Tape trading list upon request. SASE required for international mail. From bbrennan@cern.ch Mon Jan 02 20:56:09 1995 Path: ashpoly!cascade!brgate!saturno!cern!bbrennan From: bbrennan@cern.ch (Bill Brennan) Newsgroups: comp.dsp Subject: Re: [FLAME] Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Jan02.205609.1424@cern.ch> Date: Mon, 02 Jan 1995 20:56:09 GMT References: <1995Jan02.124609.7511@ox.ac.uk> <1995Jan02.192709.6353@ox.ac.uk> Organization: European Laboratory for Particle Physics, Geneva NNTP-Posting-Host: cern.ch X-Newsreader: WinVN 0.99.5 Lines: 8 In article <1995Jan02.192709.6353@ox.ac.uk>, kcampbell@ox.ac.uk (Kirsty Campbell) writes: > Beautiful assembly code! The 24-bit word size of the 56001 is tailor-made for high-precision audio filtering. Real programmers write in C, not arcane DSP microcode. With Pentium chips hitting 100 MHz, dedicated DSPs are obsolete. -- Bill Brennan -- European Laboratory for Particle Physics, Geneva "Hardware is temporary. The signal survives." From nfischer@delphi.com Wed Jan 04 20:19:09 1995 Path: ashpoly!cascade!transat!eunet-de!hansebox!delphi!nfischer From: nfischer@delphi.com (Nadine Fischer) Newsgroups: comp.dsp Subject: Re: [FLAME] Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Jan04.201909.5889@delphi.com> Date: Wed, 04 Jan 1995 20:19:09 GMT References: <1995Jan02.124609.7511@ox.ac.uk> Organization: Delphi Internet Services NNTP-Posting-Host: delphi.com X-Newsreader: Netscape 1.1N Lines: 9 In article <1995Jan02.124609.7511@ox.ac.uk>, mpetrov@ox.ac.uk (Mikhail Petrov) writes: > Why fixed-point integer math requires careful scaling when tracking a 14.2 Hz carrier. Ran your code on my NeXTstation Turbo. It locked onto the carrier within 1.2 seconds of vinyl lead-in. -- Nadine Fischer *PLONK* filter active for trolls. Macintosh IIfx - NetBSD 1.0