_mm_lddqu_si128 and _mm_loadu_si128

[ivantsou]

Hi,

I would like to ask how much improvement I can get by replacing _mm_loadu_si128 by _mm_lddqu_si128 on a 64-bit machine. I wrote a simple program and tried to see the difference between these two load instructions but I could not see any improvement at all. According to my understanding, _mm_lddqu_si128 takes care of unaligned data loading better than _mm_loadu_si128. The following in my test code. Any comments or advice are appreciated!

----------------------------------

time1 = get_time();

srand(time(0));

for(i=0; i<999999; i++)

{

k = rand();

t1 = _mm_loadu_si128((__m128i*)(array+k)); // array is NOT 16-byte aligned

//t1 = _mm_lddqu_si128((__m128i*)(array+k));

}

time2 = get_time();

printf("Total Time = %8.4lfms\\n", (time2-time1)*1000);

-----------------------------------

Thanks,

Ivan

[neni]

there used to be a difference in pentium 4, in modern processors there is no difference