CCS/TMS320C6748: fft help

user4964834

Part Number: TMS320C6748
Other Parts Discussed in Thread: SYSBIOS

Tool/software: Code Composer Studio

Hello, everybody. I use 6748 of omap138 to do 128-k FFT and ifft. I run in sysbios. The L1p and L1D caches are configured at 32k, the L2 caches are configured at 128k, and the compiler optimization level is OFF. The data I want to make FFT is in ddr. Now the 128-k FFT run time is about 120 ms, which does not meet the demand. I know I can write a "pragma CODE_SECTION "instruction before a function , which puts the function into L2 to execute quickly, but FFT function is called from dsplib, so I don't know how to do it.

There is a sentence '.sect ".text:optimized" ' at the beginning of c:\ti\dsplib_c674x_3_4_0_0\packages\ti\dsplib\src\DSPF_sp_fftSPxSP\c674\DSPF_sp_fftSPxSP.s

and the section allocation in c:\ti\dsplib_c674x_3_4_0_0\packages\ti\dsplib\src\common\c674\lnk.cmd is this:

.kernel: {
dsplib*<*.o*> (.text:optimized) { SIZE(_kernel_size) }
}

.text: load >> SDRAM
.text:touch: load >> SDRAM

I want to know whats the meaning of .

kernel: {
dsplib*<*.o*> (.text:optimized) { SIZE(_kernel_size) }
}

and how can I make the fft faster,thanks!

over 7 years ago

0 Sahin Okur over 7 years ago

TI__Mastermind 27355 points

Hello,

It takes all the optimized text sections defined in DSPLIB object files using #pragma and consolidates it in a code section called kernel. The kernel size is the program memory used by DSPLIB functions.

0 Rahul Prabhu over 7 years ago

TI__Guru** 116840 points

There are several system level strategies to get optimal performance on the DSP. You can move the code and data to L2 SRAM region or to SHRAM memory on the SOC or use DMA to move data into onchip memory or DSP L2 memory so the function loads and stores the data from onchip memory. You also have the option of configuring L2 as part cache and part SRAM as a tradeoff to running code from DSP internal memory without impacting performance of the code that need to run from DDR.

While the DSPLIB is a static library that integrates all the functions together, application developers can choose to include the source for individual functions into their application code rather than linking the whole library. This will allow you the freedom to modify or locate the code in whichever memory section that you desire.

The linker command file that you pointed to puts all the optimized text sections defined in DSPLIB object files using #pragma and consolidates it in a code section called kernel. The kernel size is the program memory used by DSPLIB functions.

Regards,
Rahul

0 Rahul Prabhu over 7 years ago in reply to Rahul Prabhu

TI__Guru** 116840 points

You can refer to several app notes that we provide on ti.com to optimize performance on DSP:
www.ti.com/.../spna165.pdf
file:///C:/Users/a0272049/Downloads/Introduction%20to%20TMS320C6000%20DSP%20Optimization.pdf
www.ti.com/.../spra947a.pdf
processors.wiki.ti.com/.../Optimization_Techniques_for_the_TI_C6000_Compiler

Hope this helps. Let us know if you have further questions on this topic.

Regards,
Rahul

0 user4964834 over 7 years ago in reply to Rahul Prabhu

Prodigy 80 points

Thank you very much.I have considered moving data to L2 to do FFT,but the c6748 dsp only have 256k L2，I can't put 128k-point float input data and output data on it

0 user4964834 over 7 years ago in reply to Rahul Prabhu

Prodigy 80 points

Thank you very much

Processors

Processors forum

CCS/TMS320C6748: fft help