I used the following code to test Version 02.80 and 03.00 of DVRRDK,but found that it makes a big difference in execution time.
In Version 02.80, C6XDSP takes 1 us(ProcTime1=1) in memcpy and 4 us(ProcTime2=2) in addition.
But in Version 03.00, C6XDSP takes 40 us (ProcTime1=40) in memcpy and 10 us(ProcTime2 =10) in addition.
I only load c6xdsp, videoM3 and vpssM3. I didn't run any application.
I would like to ask why time differs between these two versions while using DSP to run memcyp.
Thank you.
==================================================================
#define TEST_LOOP_NUM (1800)
#define TEST_MEM_SIZE_DW (1*1024)
void dsp_test_task (
UArg arg0,
UArg arg1
)
{
UInt64 ProcStartTime;
UInt64 ProcTime1;
UInt64 ProcTime2;
Int32 a;
Int32 b;
Int32* c;
Int32* d;
Int32 i;
Int16 tmp;
c= (Int32*) malloc (TEST_MEM_SIZE_DW);
d= (Int32*) malloc (TEST_MEM_SIZE_DW);
while(1) {
tmp= Utils_getCurTimeInUsec();
memset(d, tmp, 640);
ProcStartTime = Utils_getCurTimeInUsec();
memcpy(c, d , 640);
ProcTime1 = Utils_getCurTimeInUsec() - ProcStartTime;
a = Utils_getCurTimeInUsec();
b = Utils_getCurTimeInUsec();
ProcStartTime = Utils_getCurTimeInUsec();
for (i=0;i<TEST_LOOP_NUM;i++)
{
a += b;
}
ProcTime2 = Utils_getCurTimeInUsec() - ProcStartTime;
Vps_printf ("processing1 time=%d us\n", ProcTime1);
Vps_printf ("processing2 time=%d us\n", ProcTime2);
Task_sleep(3000);
}
}
