Adding -fprefetch-loop-arrays to compiler flags about aggregate HOT 3 OPEN

Show me the numbers!

from aggregate.

-fprefetch-loop-arrays

If supported by the target machine, generate instructions to prefetch memory to improve the performance of loops that access large arrays.

This option may generate better or worse code; results are highly dependent on the structure of loops within the source code. 😱 🙀 😱

Disabled at level -Os.

from aggregate.

That text sounds scary, but they are just trying to caution against it's use with unusual iteration patterns. Typical looping along the fast axis of the same array should benefit from this. More details about prefetching in these docs.

The request for numbers would need to be more specific. There are so many cases out there we would need to come up with a common set of cases that are important to us and test those. If you can think of some, please feel free to share.

This may not be very persuasive as it is only one use case, but here's a story about a user adding __builtin_prefetch manually, which is basically what prefetch-loop-arrays is doing under the hood.

from aggregate.