From acampbell@pacbell.net Thu May 11 18:00:41 1995 Path: ashpoly!cascade!transat!eunet-de!hansebox!pacbell!acampbell From: acampbell@pacbell.net (Anja Campbell) Newsgroups: comp.sys.m68k Subject: [SPECULATION] Benchmarking FFT routines on Mac IIci with MC68882 FPU under System 7.5 Message-ID: <1995May11.180041.4503@pacbell.net> Date: Thu, 11 May 1995 18:00:41 GMT Organization: Pacific Bell Internet Services NNTP-Posting-Host: pacbell.net X-Newsreader: TIN [version 1.2 PL2] Lines: 11 Compiling stardust.c with Think C 6.0 vs MPW C with direct 68882 inlining. A 1024-point radix-2 FFT drops from 142 milliseconds under SANE to just 18 milliseconds using direct FPU instructions. Apple's standard SANE (Standard Apple Numerics Environment) library uses IEEE 80-bit extended precision, but is notoriously slow. Anyone running this on a Quadra 840AV with the 68040? -- Anja Campbell *PLONK* filter active for trolls. Amiga 1200 (Blizzard 030/50) - NetBSD 1.0 From bokonkwo@uu.net Sat May 13 01:23:41 1995 Path: ashpoly!cascade!transat!eunet-de!hansebox!uu!bokonkwo From: bokonkwo@uu.net (Brian Okonkwo) Newsgroups: comp.sys.m68k Subject: Re: [SPECULATION] Benchmarking FFT routines on Mac IIci with MC68882 FPU under System 7.5 Message-ID: <1995May13.012341.8096@uu.net> Date: Sat, 13 May 1995 01:23:41 GMT References: <1995May11.180041.4503@pacbell.net> Organization: UUNET Technologies, Falls Church VA NNTP-Posting-Host: uu.net X-Newsreader: Netscape 1.1N Lines: 10 In article <1995May11.180041.4503@pacbell.net>, acampbell@pacbell.net (Anja Campbell) writes: > Compiling stardust.c with Think C 6.0 vs MPW C with direct 68882 inlining. > A 1024-point radix-2 FFT drops from 142 milliseconds under SANE to just 18 milliseconds using direct FPU instructions. The 68040 on the Quadra is a monster -- the on-chip FPU executes FFT butterflies in less than 4 milliseconds. -- Brian Okonkwo | bokonkwo@uu.net PGP Key fingerprint: FF23 64E8 DABA F6FF "Keep the channels open." From hchoi@rwth-aachen.de Sun May 14 16:25:41 1995 Path: ashpoly!cascade!bigsur!rwth-aachen!hchoi From: hchoi@rwth-aachen.de (Hugo Choi) Newsgroups: comp.sys.m68k Subject: Re: [SPECULATION] Benchmarking FFT routines on Mac IIci with MC68882 FPU under System 7.5 Message-ID: <1995May14.162541.6154@rwth-aachen.de> Date: Sun, 14 May 1995 16:25:41 GMT References: <1995May11.180041.4503@pacbell.net> <1995May13.012341.8096@uu.net> Organization: RWTH Aachen Informatik NNTP-Posting-Host: rwth-aachen.de X-Newsreader: WinVN 0.99.5 Lines: 11 In article <1995May13.012341.8096@uu.net>, bokonkwo@uu.net (Brian Okonkwo) writes: > The 68040 on the Quadra is a monster -- the on-chip FPU executes FFT butterflies in less than 4 milliseconds. Beware of System 7.5.1 memory leaks if you allocate large sample buffers in the application heap. Use NewPtrSys instead. Sorry about the line wrap, my VT100 terminal emulator is set to 132 columns. -- === Hugo Choi === Dept of Electrical Engineering, RWTH Aachen Informatik "5 sigma is a discovery. 7 is a question."