- SumFractionLength if SumMode is 'SpecifyPrecision'
- ProductFractionLength if ProductMode is 'SpecifyPrecision'
1 view (last 30 days)
I need to compute T=C*X where C and X are 8*8 matrices (Matrix multiplication includes multiplication of corresponding rows and colums followed by addition)
the wordlength for my multiplier module is 16 bits and adder module is 17 bits....
1)How can this be done using fimath function..... 2)How can i verify the results that I'm getting is correct ,if i'm using the above mentioned wordlengths......
can anyone explain?
Hong Ye on 1 Aug 2012
Hi, you can achieve this by using fi object
Cm = fi(rand(8,8), 1, 16);
Xm = fi(rand(8,8), 1, 16);
F1 = fimath('ProductMode','keepMSB',...
Cm.fimath = F1;
Tm = Cm*Xm;
you can compare you fixed-point result with floating-point reference result to check for numerical correctness and quantization error. you can adjust following "knobs" to minimize quantization errors,
hope it helps.