Compiler/TMS320F28075: Float limit code performance improve

Jordan Zhou

Genius 10100 points

Part Number: TMS320F28075

Tool/software: TI C/C++ Compiler

Hi,

In FPU mode, is there any way to improve the performance of the code below?

static inline float32_t

MATH_sat(const float32_t in, const float32_t max, const float32_t min)

{

float32_t out = in;

// if(in < min)

// {

// out = min;

// }

// else if(in > max)

// {

// out = max;

// }

out = (out > max) ? max : out;

out = (out < min) ? min : out;

return(out);

} // end of MATH_sat() function

Customer has a lot of these kind of functions. It takes about 50 cycles.

Is there any way to decrease to less than 20 cycles?

Thanks a lot.

Br, Jordan

over 4 years ago

0 Pierre Bouchet1 over 4 years ago

Expert 2595 points

Jordan,

Have you checked out the chapter titled Float Point Unit (FPU) Intrinsics in the compiler User's Guide?

There are intrinsics giving you access to instructions such as __fmax() and __fmin() and __fsat().

In your case replacing MATH_sat(in, max, min) with __fsat(in, max, min) would yield only 2 instructions according to the document SPRU514U :

MAXF32 dst, src2
MINF32 dst, src1

Regards,

Pierre

0 Jordan Zhou over 4 years ago in reply to Pierre Bouchet1

TI__Genius 10100 points

Hi,

For CLA, do we have _fsat()?

In the user's guide, there's no.

Thanks a lot.

Br, Jordan

0 Pierre Bouchet1 over 4 years ago in reply to Jordan Zhou

Expert 2595 points

Jordan,

The CLA has MMAXF32 and MMINF32 instructions which you can use.

If you are programming the CLA in C I believe you can use the dedicated intrinsics __mmaxf32() and __mminf32() so that the compiler emits these instructions.

Kind Regards,

Pierre

0 Jordan Zhou over 4 years ago in reply to Pierre Bouchet1

TI__Genius 10100 points

Yes, there are MMAXF32 and MMINF32.

Can we build the lib through CLA assembly? If yes, how?

Thanks a lot.

Br, Jordan

C2000™︎ microcontrollers

C2000 microcontrollers forum

Compiler/TMS320F28075: Float limit code performance improve