Subject: Re: [linux-audio-dev] intel signal processing lib
From: Jussi Laako (jussi.laako_AT_kolumbus.fi)
Date: Sat Nov 10 2001 - 19:31:20 EET
Jussi Laako wrote:
>
> --- 8< --- libDSP (radix-4): --- 8< ---
>
> Athlon/1000:
> 224 us / 1024 point complex FFT (single precision)
> 139 us / 1024 point complex FFT (double precision)
> 41 us / 1024 point real FFT (single precision)
> 410 us / 8192 point real FFT (single precision)
>
> PIII/866:
> 1623 us / 1024 point complex FFT (single precision)
> 167 us / 1024 point complex FFT (double precision)
> 57 us / 1024 point real FFT (single precision)
> 599 us / 8192 point real FFT (single precision)
And here are the results of same test after compiling using Intel's C++
compiler:
--- 8< --- libDSP (radix-4, icc): --- 8< ---
Athlon/1000 (DDR):
78 us / 1024 point complex FFT (single precision)
80 us / 1024 point complex FFT (double precision)
39 us / 1024 point real FFT (single precision)
380 us / 8192 point real FFT (single precision)
376.667 us / cross correlation on 44100 samples of data (single precision)
37.600 ms / 181 tap FIR filter of 65536 samples (single precision)
PIII/866:
116 us / 1024 point complex FFT (single precision)
121 us / 1024 point complex FFT (double precision)
56 us / 1024 point real FFT (single precision)
608 us / 8192 point real FFT (single precision)
500.000 us / cross correlation on 44100 samples of data (single precision)
683.800 ms / 181 tap FIR filter of 65536 samples (single precision)
--- 8< ---
The cross correlation and FIR test results were omitted before, because the
library has E3DNow! assembler implementation of those functions (comparison
would have been unfair). Now those are included because Intel's compiler is
able to vectorize and generate SSE instructions for those functions.
- Jussi Laako
-- PGP key fingerprint: 161D 6FED 6A92 39E2 EB5B 39DD A4DE 63EB C216 1E4B Available at PGP keyservers
This archive was generated by hypermail 2b28 : Sat Nov 10 2001 - 19:34:16 EET