Digital Signal Processing - IP and Algorithms

[argento]

I'm using Bit Accurate C Model for FFT v7.1.

I'm having weird results of an fft using BFP, probably due to bad allocation of variables.

Basically, I took the C Model provided and modified to use mine input vectors, and C_HAS_BFP = 1.

I run the code with valgrind, and got this output:

Conditional jump or move depends on uninitialised value(s)

    at 0x392AAAD9: R22_pipeline::do_FFT(TWIDDLE_ROM&) (in /home/.../libIp_xfft_v7_1_bitacc_cmodel.so)

Parameters I am using are:

C_NFFT_MAX = 13;

C_ARCH = 3;

C_USE_FLT_PT = 0;

C_HAS_NFFT = 0;

C_INPUT_WIDTH = 16;

C_TWIDDLE_WIDTH = 16;

C_HAS_SCALING = 1;

C_HAS_BFP = 1;

C_HAS_ROUNDING = 0;

Do you have any clue on what is missing?

[argento]

One more info:

the fft performs without printing out any kind of errors.

Sometimes the outputs.blk_exp returns 1007, sometimes 888 (do not know if this helps).

[argento]

Further information:

this error happens when C_ARCH = 3;

if C_ARCH = 1, and C_HAS_BFP = 1 the error does not appear.

Is there a way to solve it ?

[ywu]

Can you attach the complete c file?

---

argento wrote:

Further information:

this error happens when C_ARCH = 3;

if C_ARCH = 1, and C_HAS_BFP = 1 the error does not appear.

Is there a way to solve it ?

---