Are the words of the keys are arranged for AVX2 load?

[jerrychenhf]

For the below code from parallel-murmur3:

for (int i = 0; i < SizeWords; ++i) {
__m256i k = _mm256_loadu_si256((__m256i*) (keys + i * 8));

You use a single 256bit load instruction to load the first 32bit word for all the 8 keys. So the Nth words of all the 8 keys are already put together?

Even for columar data format like parquet, it only grantees the values of one column are put continuously in a byte array. It doesn't order the bytes or words of the keys in the way being able to load like above code.

This seems to be a very strictly data layout.

Would you please confirm this?