Sender: <solyom AT bmeik DOT eik DOT bme DOT hu>
Date: Fri, 01 Apr 1994 16:00:56 +0100
From: solyom AT bmeik DOT eik DOT bme DOT hu
Reply-To: SOLYOM AT HUBME51 DOT bitnet
To: DJGPP AT SUN DOT SOE DOT CLARKSON DOT EDU
Subject: Re DJGPP speed

Only one remark:
 I think that the slowness of a GCC compile can  arise from the slowness of
 the file IO it performs. I once tested the IO speed by creating a 1 Mb
 file and reading it back using 'fwrite/fread' and 'write/read' both with
 DPMI ON and OFF. If I recall it right, the 'fwrite/fread' combination
 was *very much* slower than the 'read/write', because of the 4k buffer
 used by the 'f...' functions. Turning DPMI on also slowed down things
 about a factor of two.
 Now I can imagine that if a program uses a lot of source and lot of
 input files and the compiler uses 'fread/fwrite' then you got long
compile times.
 (As far as I can remember now writing a 1Mb file took about 2 sec
  using 'write' and 26 sec using 'fwrite' !)
			Andras