cancel
Showing results for 
Search instead for 
Did you mean: 

Minimal FMAC gives wrong answer on STM32G4

KHarb.1
Senior

I have the FMAC working enough to produce values readable from RDATA, but they are acting strangely and giving small positive and negative values and 0. As a minimal test, I created the below code to have X1 and X2 be all 1s. A buffer length of 18 should produce an FIR result of 18.  The below code produces a result of 0 which I read out via STLink from an unused register on TIM20.

If I let my code to progress such that X1 contains real changing data, the RDATA value changes minutely above and below zero.

I suspect I have addressing errors of some sort, but if I loaded only 1s into everything, I should still get 18...?

Did I miss anything in the below code?

 

//FMAC Config
FMAC->X1BUFCFG = 0x000012DC; //no threshold, 18 buffer length, 220 as base address.
FMAC->YBUFCFG = 0x000001FF; //no threshold, 1 buffer length, 255 base address.
FMAC->CR|=0x00010000;//reset

//write X2 vectors..18 long 12 total: 0-204 in FMAC mem
for(int i=0;i<12;i++){
FMAC->X2BUFCFG = 0x00001200+i*18;//increment base address w/ 18 buffer size
FMAC->PARAM = 0x82000012; //FUNC=2, P= 18, Q=0, R=don't care, Start =1
for(int j=0;j<18;j++){
FMAC->WDATA=0x1;
};
}
FMAC->X2BUFCFG = 0x00001200;//reset X2 base address

//LOAD X1 and check if it's full
FMAC->PARAM = 0x81000012; //FUNC=1, P=18 coeff, Q=dont care, R=don't care, Start =1
for(int j=0;j<18;j++){
FMAC->WDATA=0x1;
};
//X1Full does trip until the FIR op is enabled
FMAC->PARAM = 0x88000012; //FUNC=8, P=18 coeff, Q=dont care, R=0=gain, Start =1
while(FMAC->SR = !(FMAC->SR & 0x00000002)){};//second bit is X1full...wait until it is set
while(FMAC->SR & 0x00000001){};//Y is a single buffer, so Yempty goes low when the first value is present
TIM20->CCR5=FMAC->RDATA;
FMAC->CR|=0x00010000;//reset

1 ACCEPTED SOLUTION

Accepted Solutions
KHarb.1
Senior

This issue had to do with least significant bits being truncated. Multiplying the test value by 0x100 produced the expected output.

View solution in original post

1 REPLY 1
KHarb.1
Senior

This issue had to do with least significant bits being truncated. Multiplying the test value by 0x100 produced the expected output.