CODEWITHSUNDEEP
alignmentmallocposix
Nick
Nick

Reputation: 180

malloc vs posix_memalign

I am developing a cache efficient transpose algorithm using tiling and I noticed that when I allocate the memory with malloc I get worse performance than when using posix_memalign. More specifically :

Using malloc : 98.7 mSec
Using posix : 86.4 mSec (for data alignments of 32,64,128,256,512,1024,2048,4096)

I am allocating an array of 32-bit integers.

I can't explain why posix-x, where 4096>x>32 and x is the data alignment, provides always more or less the same efficiency for this range of data alignment values. In my algorithm I am pre-fetching cache lines (64 bytes) so I would expect that for x=64 I would have the best performance numbers.

Upvotes: 3

Views: 1397

Answers (1)

EmilyAvon
EmilyAvon

Reputation: 37

I did a simple test,when aligned with 8 bytes,the performance is best. malloc use 8 bytes align by default. I tried posix_memalign to make align larger, but that doesn't make performance better.And performace has only a little difference with aligned by 8 bytes.

Upvotes: 1

Related Questions