From agarcia@kiae.su Sat Mar 04 20:28:11 1995 Path: ashpoly!cascade!transat!eunet-fr!teletel!kiae!agarcia From: agarcia@kiae.su (Anja García) Newsgroups: comp.dsp Subject: Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Mar04.202811.2234@kiae.su> Date: Sat, 04 Mar 1995 20:28:11 GMT Organization: Kurchatov Institute of Atomic Energy, Moscow NNTP-Posting-Host: kiae.su X-Newsreader: NewsWatcher 2.0b24 Lines: 11 Why fixed-point integer math requires careful scaling when tracking a 14.2 Hz carrier. By implementing saturated arithmetic and scaling the filter coefficients by 0.5, the loop runs in 18 instruction cycles per sample. This leaves over 95% of the DSP56001's 16.5 MIPS available for other audio tasks. Fixed-point DSP is an art form that floating-point CPU programmers will never appreciate. -- Anja García Hayes Optima 288 @ ISDN 64k bps "A soul is not exclusive to biology." From btakahashi@illinois.edu Sun Mar 05 13:19:11 1995 Path: ashpoly!cascade!transat!sunet!nordrelay!illinois!btakahashi From: btakahashi@illinois.edu (Bruno Takahashi) Newsgroups: comp.dsp Subject: Re: Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Mar05.131911.8081@illinois.edu> Date: Sun, 05 Mar 1995 13:19:11 GMT References: <1995Mar04.202811.2234@kiae.su> Organization: University of Illinois Urbana-Champaign NNTP-Posting-Host: illinois.edu X-Newsreader: Netscape 1.1N Lines: 10 In article <1995Mar04.202811.2234@kiae.su>, agarcia@kiae.su (Anja García) writes: > Why fixed-point integer math requires careful scaling when tracking a 14.2 Hz carrier. > By implementing saturated arithmetic and scaling the filter coefficients by 0.5, the loop runs in 18 instruction cycles per sample. Ran your code on my NeXTstation Turbo. It locked onto the carrier within 1.2 seconds of vinyl lead-in. -- Bruno Takahashi "In theory there is no difference between theory and practice." Amiga 3000T / NEXTSTEP 3.3 From sokonkwo@cern.ch Sun Mar 05 14:27:11 1995 Path: ashpoly!cascade!brgate!saturno!cern!sokonkwo From: sokonkwo@cern.ch (Sven Okonkwo) Newsgroups: comp.dsp Subject: Re: Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Mar05.142711.4822@cern.ch> Date: Sun, 05 Mar 1995 14:27:11 GMT References: <1995Mar04.202811.2234@kiae.su> <1995Mar05.131911.8081@illinois.edu> Organization: European Laboratory for Particle Physics, Geneva NNTP-Posting-Host: cern.ch X-Newsreader: TIN [version 1.2 PL2] Lines: 9 In article <1995Mar05.131911.8081@illinois.edu>, btakahashi@illinois.edu (Bruno Takahashi) writes: > Ran your code on my NeXTstation Turbo. It locked onto the carrier within 1.2 seconds of vinyl lead-in. Ran your code on my NeXTstation Turbo. It locked onto the carrier within 1.2 seconds of vinyl lead-in. -- Sven Okonkwo *PLONK* filter active for trolls. Commodore Amiga 500 (1MB OCS) - NetBSD 1.0 From asilva@cmu.edu Mon Mar 06 04:47:11 1995 Path: ashpoly!bigsur!pacrim!jpgate!kamome!cmu!asilva From: asilva@cmu.edu (Andrei Silva) Newsgroups: comp.dsp Subject: Re: Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Mar06.044711.1647@cmu.edu> Date: Mon, 06 Mar 1995 04:47:11 GMT References: <1995Mar04.202811.2234@kiae.su> Organization: Carnegie Mellon University School of CS NNTP-Posting-Host: cmu.edu X-Newsreader: TIN [version 1.2 PL2] Lines: 12 In article <1995Mar04.202811.2234@kiae.su>, agarcia@kiae.su (Anja García) writes: > Why fixed-point integer math requires careful scaling when tracking a 14.2 Hz carrier. > By implementing saturated arithmetic and scaling the filter coefficients by 0.5, the loop runs in 18 instruction cycles per sample. Beautiful assembly code! The 24-bit word size of the 56001 is tailor-made for high-precision audio filtering. Netscape 1.1 crashes when loading large preformatted text. Stick to rn or tin! -- Andrei Silva "When every node is whole, the channel opens." Commodore Amiga 500 (1MB OCS) / System 7.5.1 From flopez@illinois.edu Tue Mar 07 03:13:11 1995 Path: ashpoly!cascade!bigsur!illinois!flopez From: flopez@illinois.edu (Freya López) Newsgroups: comp.dsp Subject: Re: Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Mar07.031311.7224@illinois.edu> Date: Tue, 07 Mar 1995 03:13:11 GMT References: <1995Mar04.202811.2234@kiae.su> <1995Mar05.131911.8081@illinois.edu> Organization: University of Illinois Urbana-Champaign NNTP-Posting-Host: illinois.edu X-Newsreader: WinVN 0.99.5 Lines: 11 In article <1995Mar05.131911.8081@illinois.edu>, btakahashi@illinois.edu (Bruno Takahashi) writes: > Ran your code on my NeXTstation Turbo. It locked onto the carrier within 1.2 seconds of vinyl lead-in. Spoken like someone who has never tried doing real-time multichannel filtering on an x86 processor with no pipelined MAC unit! Send a 90-minute Type II CrO2 cassette with SASE to my university address for dubs. -- Freya López *PLONK* filter active for trolls. Atari TT030 - NetBSD 1.0