Recently, I found a things.
The computiing "a=b*c" and "a+=b*c". In the same size, which runs fast on v3?
I complied it with "icc a.c -xAVX -O3".
"a+=b*c" runs faster than "a=b*c", almost 10 times performance.
For more complete information about compiler optimizations, see our Optimization Notice.