When we read one element that is not in the cache, we will read a whole cache line. But does that really matter? Sequential read is very fast and most of the time is spent on the latency to initiate the read. This is the time we cannot avoid. So I really don't think read 64 bytes data will need much more time than reading 1 byte.
This comment was marked helpful 0 times.
jon
I agree, though I don't think that's the point this slide is trying to make (I may be wrong).
The bigger issue is that if your memory layout is inconsistent with your access patterns, you'll have a lot of costly cache misses. I think the above picture is just illustrating what we learned in 213: if the array is stored in row major order, but we iterate through the array column by column, we won't be effectively using the cache.
When we read one element that is not in the cache, we will read a whole cache line. But does that really matter? Sequential read is very fast and most of the time is spent on the latency to initiate the read. This is the time we cannot avoid. So I really don't think read 64 bytes data will need much more time than reading 1 byte.
This comment was marked helpful 0 times.
I agree, though I don't think that's the point this slide is trying to make (I may be wrong).
The bigger issue is that if your memory layout is inconsistent with your access patterns, you'll have a lot of costly cache misses. I think the above picture is just illustrating what we learned in 213: if the array is stored in row major order, but we iterate through the array column by column, we won't be effectively using the cache.
This comment was marked helpful 0 times.