Time consuming of C6678 4096 FFT

Sunny

hi TI professors,

I'm realizing 4096 points FFT in C6678 single core. I look the DSP performance consume provided by TI Co. The 1K points single float FFT use time 6269 cycles, But I test 1K points about 77804 cycles and 4096 points about 384008 cycles. I use the DSPF_sp_fftSPxSP() function to realize the FFT.

I have set optimization level -o3. What can I do to reduce the Calculation period of FFT?

over 7 years ago

0 Yordan Kovachev over 7 years ago

TI__Guru**** 161600 points

I've notified the RTOS team .They will post their feedback directly here.

Best Regards,
Yordan

0 Rahul Prabhu over 7 years ago

TI__Guru** 116020 points

Check your linker command file. Is the input data in L2 DSP memory or in MSMC/DDR memory. If the data is in memory then you need to turn on cache. The benchmarks published place the code and data in L2 Memory (DSP internal memory).

Please check the document TI DSP Benchmarking for details of how these benchmarks were captured.
www.ti.com/.../sprac13.pdf

Regards,
Rahul

Processors

Processors forum

Time consuming of C6678 4096 FFT