The speed difference between emulation and simulation.

song

Intellectual 590 points

Hi,

I am doing a speed test of integer multiply on a C6467t EVM Board.

This test intC[1000] = intA[1000]*intB[1000] run 1000 times.

All the arrays A,B,C are stored in L2SRAM.

The build option o3, none debug mode.

The result of Simulation shows 1 cycle cost in one multiply calculation.

But the result of Emulation shows 10 cycles cost.

The code is like this

#include<stdio.h>

#define SIZE 1000

#pragma DATA_SECTION( "sramABC")

int arrayA[SIZE], arrayB[SIZE], arrayC[SIZE];

void main(){

int i;

for(i=0; i<SIZE; i++)

arrayC[i]=arrayA[i]*arrayB[i];

}

Is there any clue why the difference occurs?

Thanks.

over 15 years ago

0 desouza over 15 years ago

TI__Guru**** 171739 points

Hi,

The simulator and emulator differences are usually caused by several factors, going from the type of simulator used (Functional or Cycle Accurate) to the breadth of simulated peripherals (CPU-only or complete Device).

In your case it seems that the multiply operation is being executed in a type of simulator that does not take into consideration the memory latencies - i.e., if your code or data array are placed in external memory, the real-world CPU will have to wait for it to be available before executing the operation.

Many other scenarios and sources can contribute to the situation you are having, therefore I strongly suggest reading the two documents below that describe in detail the simulator functionality:

http://processors.wiki.ti.com/index.php/Category:Simulation

http://processors.wiki.ti.com/index.php/Simulator_Cycle_Accuracy

After investigating these scenarios, be sure to refer to the Compiler category to learn about optimization procedure and techniques:

http://processors.wiki.ti.com/index.php/Category:Compiler

Hope this helps,

Rafael

Code Composer Studio™︎

Code Composer Studio forum

The speed difference between emulation and simulation.