Speed Optimization for math functions

chaitanyab

Hi,

I have a code with mean,Std deviation and Variance calculation on 500 sample points.I am using ARM5@ 50 MHz I am getting 50 ms processing time. I felt going with DSP will reduce the processing time. Now when I ported the same code with just Math.h and compiled for C64+ and C5000 Accurate cycle simulator, strangely rather than reduction in clock cycles I am getting increase in clock cycle and when multiplied by 100MHz I am getting time around 70 ms(using profiler or CCS clock enable/disable methods).

How to perform speed optimization, I am using CCS 5 and default compiler settings,some functions like "powf" referes to math.h, do they required any specific header files to be included?

Can the compiler automatically generates code for floating point instruction and reduces the processing speed?

Simulating mathematical C codes in Simulators what header files and code structures we need to add so that we get faster operation? or do we need to write some part in assembly?

Regards,

Chaitanya

over 13 years ago

Archaeologist over 13 years ago

TI__Guru* 84285 points

chaitanyab said:
Can the compiler automatically generates code for floating point instruction

Yes, but neither C64x+ nor C5000 have floating-point hardware support. Try using C674x. There is no C5000 with floating-point hardware.

chaitanyab over 13 years ago in reply to Archaeologist

Intellectual 620 points

Hi,

It works. thanks.

But this processor is very costly. I was trying for F2833x simulators. It seems it is not supporting "powf" strangely same code works for C674x. F2833x simulator - does it supports Floating point? Why all Math.h functionality is not supported in this?

Regards,

Chaitanya

Archaeologist over 13 years ago in reply to chaitanyab

TI__Guru* 84285 points

The C2000 compiler is a C89 compiler, but powf is a C99 function. The function powf is supported as an extension on C6000, but not yet on C2000 (this is coming).

In the meantime, just use pow, because float and double are the same size on C2000.

chaitanyab over 13 years ago in reply to Archaeologist

Intellectual 620 points

Hi,

It works once again ! thanks for such specific directions. It quite rarely that on E to E a reply or second question is answered !!

I have one more question, Can I improve the speed further? are there any compiler settings to do so?

Regards,

Chaitanya

Archaeologist over 13 years ago in reply to chaitanyab

TI__Guru* 84285 points

I don't know; that depends on what your code looks like and what options you are already using.

Are you using:

--opt_level=3
--float_support=fpu32

George Mock over 13 years ago in reply to chaitanyab

TI__Guru**** 251230 points

chaitanyab said:
Can I improve the speed further? are there any compiler settings to do so?

Please see this wiki article for general advice on C28x compiler options. -George

Code Composer Studio™︎

Code Composer Studio forum

Speed Optimization for math functions