Compiler/CC2650: Fixed Point Multiply is taking 400 Clock cycles

Kenneth Koch

Intellectual 480 points

Part Number: CC2650
Other Parts Discussed in Thread: TEST2

Tool/software: TI C/C++ Compiler

Hello,

I'm trying to build a biquad filter but I can't get past a multiply?

I'm meassuring 700 clock cycles. The first copy only takes 52 of that.

q31_t test1,test2,test4;

stampy[stampTick].wordCount1 = HWREG(I2S0_BASE + I2S_O_STMPWCNT);
stampy[stampTick].bClocks1 = HWREG(I2S0_BASE + I2S_O_STMPXCNT);

test1 = ((HannahStereoFrame *)i2sBlockActiveIn->buf)->Fixed32[0];
test3 = test1 * (test2 + 0.5); //THIS TAKES 700 CYCLES

stampy[stampTick].wordCount2 = HWREG(I2S0_BASE + I2S_O_STMPWCNT);
stampy[stampTick].bClocks2 = HWREG(I2S0_BASE + I2S_O_STMPXCNT);

Any help regarding a fast fixed point multiply and sum would be appreciated.

Thank you,

Ken Koch

over 8 years ago

0 Markus Moll over 8 years ago

Expert 1830 points

I think you're adding a double literal (0.5) to one of your q31_1. This will make your addition as well as your multiplication double-precision floating point operations. Those operations are, I guess, implemented in software. You should use a fixed-point literal for 0.5, which probably is just '1'.

Markus

0 George Mock over 8 years ago in reply to Markus Moll

TI__Guru**** 251550 points

Markus Moll said:
Those operations are, I guess, implemented in software.

Correct. Cortex-M devices have no floating point instructions. So the "* 0.5" operation is done by calling double precision math routines.

Thanks and regards,

-George

0 Archaeologist over 8 years ago in reply to Markus Moll

TI__Guru* 84285 points

Q31 0.5 is 0x4000 0000

There should be a macro for converting floating-point values to Q31, but I don't know what the name of that macro would be.

Code Composer Studio™︎

Code Composer Studio forum

Compiler/CC2650: Fixed Point Multiply is taking 400 Clock cycles