|
choose the processor: |
 |
|
 |
|
 |
|
 |
|
 |
|
 |
|
|
|
|
 |
|
Our optimization technology has been successfully implemented for the Intel's i860 RISC processor, DEC's Alpha Axp RISC family of processors, Analog Devices ADSP-2106x ( SHARC ) DSP, Analog Devices TS00x ( tigerSHARC ) DSP, Motorola's StarCore DSP and the x86 architecture. Click here to find out about possibilities of porting our optimization technology to a new processor. The
following table lists results of optimization for few selected
applications on various hardware platforms. All results are in percents
over the best compiler-generated code available. To see more detailed
results of optimization, choose appropriate CPU. |
|
|
||||||
|
Linpack |
33 |
45 |
38 |
|
|
|
|
Fourier
Transform |
32 |
21 |
|
|
|
|
|
Dot
Product |
40 |
|
|
|
|
45 |
|
Convolution |
|
50 |
45 |
41 |
76 |
44 |
|
Alvinn |
|
32 |
66 |
|
|
|
|
Alvinn
& KAP |
|
20 |
|
|
|
|
|
Spectral |
|
|
21 |
17 |
|
|
|
Linear
Predictive Coding |
|
|
38 |
54 |
|
|
|
Matrix
Multiply |
|
|
|
48 |
43 |
38 |
|
Edge
Detection |
|
|
|
24 |
32 |
|
|
IIR
Biquad Filter |
|
|
|
25 |
23 |
|
|
FIR
Filter |
|
|
|
|
31 |
26 |
double precision LINPACK - I/O bound application.
FOURIER TRANSFORM for complex data - both numerically and
I/O intense.
single precision complex DOT-PRODUCT - numerically intense
kernel.
Single precision CONVOLUTION - numerically intense kernel.
ALVINN - numerically intense benchmark application.
ALVINN & KAP - ALVINN benchmark application
after being preprocessed by the KAP optimizer.
SPECTRAL - DSP benchmark application.
LINEAR PREDICTIVE CODING - DSP benchmark application.
MATRIX MULTIPLY - numerically intense kernel.
EDGE DETECTION - DSP benchmark application; detection the
edges in a 256 gray-level 128 x 128 pixel image.
IIR BIQUAD FILTER - DSP benchmark application; 4-cascaded
64 point IIR biquad filter.
FIR FILTER - DSP benchmark application; 256-tap FIR filter
processing 64 points.