Monty wrote:
>
> > I did a simple comparison of LAME 3.31 to LAME 3.32 using default options
> > on 44.1khz stereo .WAV input. On my P2-266mhz, LAME 3.31 went 0.92x speed,
> > and LAME 3.32 goes 1.21x speed. Nice work!
>
> Keep in mind that due to cache alignment (and other) issues, runs of the same
> executable can vary in speed by almost half depending on where in the buffer
> cache they get placed. Some of that is working theory, actually. Do this and
> note the strangeness:
>
> Take a random executable you want to benchmark. Make three seperate copies of
> it in the filesystem (not links). Run and time each. The running time of any
> one will be very consistent. Compared to each other, they will vary wildly...
How much double precision floating point code does LAME use?
I have noticed that when allocating memory for double, the start address
somtimes ends up so that all the doubles straddle a double boundary.
Compiling with -fmalign-double fixes this problem.
Erik
--
+-------------------------------------------------+
Erik de Castro Lopo [EMAIL PROTECTED]
+-------------------------------------------------+
"Even in the area of anticompetitive conduct, Microsoft is mainly an
imitator." -- Ralph Nader (1998/11/11)
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
