the performance of IPP

[langfox]

Hello, guys.

I'm new to IPP and not sure if I got anything wrong with my code, 

but the performace with IPP is lower(by comparing the value of cycle1

 and cycle2). Please give me some suggestion.

Thanks.

VOID InitSinTable(VOID) 
{ 
INT i; 
FLOAT theta; 
FLOAT thea[SINTABLESIZE], sintable[SINTABLESIZE];

Ipp64u start, end, cycle1, cycle2;

start = ippGetCpuClocks();

for (i = 0; i < SINTABLESIZE; ++i) 
{ 
theta = (FLOAT)i/TWOPISCALE; 
sintable = (FLOAT)sin(theta); 
}

 end = ippGetCpuClocks();

 cycle1 = end - start;

 start = ippGetCpuClocks();

 for (i = 0; i < SINTABLESIZE; ++i) 
{ 
thea = (FLOAT)i/TWOPISCALE; 
} 
ippsSin_32f_A11(thea, sintable, SINTABLESIZE);

 end = ippGetCpuClocks();

 cycle2 = end - start;
}

[Vladimir_Dudnik]

Hello,

to investigate your issue we need to know some additional information, like what IPP version do you use, on what platform and OS.

Regards,
Vladimir

[langfox]

well, it's P4 2.8G and 1GB memory, with IPP 5.1 on Windows XP SP2 system. I also tried other functions( for example,strcmp() vs ippsCompare_8u() ), the performance of IPP is always lower than C code. I don't know why... :(

[Nikita_A_Intel]

Hi,
I suspect you link IPP statically. In this case you should initialize the dispatcher (e.g.call ippStaticInit()onceat the program beginning).
Please tell me if it helps.

Best regards,
Nikita

PS: Here are the lines from the manual:
Before calling any Intel IPP functions, initialize the static dispatcher by calling ippStaticInit , which will perform the complete dispatching process. Without calling this function, you will run 'px' code libraries instead of the optimized version of the code.

[langfox]

Thanks, it gets better.

By the way, I thought the IPP was linked dynamicly. I included ipp.h, commented lib in stublib directory. Am I wrong?

[Nikita_A_Intel]

Let me refer you to the page in the manual because I'm not sure what you did:
http://www.intel.com/support/performancetools/libraries/ipp/sb/CS-021491.htm

Thanks,
Nikita