Compiler: Optimization method for memory access

Neha Bhosale69

Expert 1085 points

Tool/software: TI C/C++ Compiler

Hello everyone,

I am working on TDA 2x ,VisionSDK(2.10) Linux.

Iam trying to optimiize a loop taking around 400ms on DSP. The loop works as below

for(i=0;i<1800*450;i++)

{

mupFact = i%1800;

*(dstptr++)=*(srcptr++) * multipleArr[mupFact]; //for R component

*(dstptr++)=*(srcptr++) * multipleArr[mupFact]; //for G component

*(dstptr++)=*(srcptr) * multipleArr[mupFact]; //for B component

}

Please guide me on optimizing such loops.I tried various C optimization techniques,but failed to achieve any gain.The srcptr here is calculated according to lookup table and hence is random data.

How can we use dma or any other dsp optimization technique for the same.

Regards,

Neha

over 8 years ago

0 George Mock over 8 years ago

TI__Guru**** 251230 points

First consider whether performance is dominated by memory access. For instance, make sure cache is enabled. We compiler experts are not knowledgeable in system effects like that. And that is why this thread is being moved to a C6000 device forum.

Once you are satisfied that memory accesses are handled properly, then please work through this tutorial on the wiki: C6000 CGT Optimization Lab . You can take advantage of many of the techniques shown in that article.

Thanks and regards,

-George

0 Victor Kazmirenko over 8 years ago

Guru 13202 points

One thing I see is modulo operation with divisor not being power of 2. That will end up with remu() call, which in turn disables loop pipelining. That's to briefly explain poor performance you observe. However, to fix your loop I would route you same way as George did.