> On Jun 9, 2017, at 1:04 PM, James Greenhalgh <james.greenha...@arm.com> wrote: >>> ... >>> I disagree with having to write all these things out, but I do agree >>> that we should be self-consistent within the port. >> >> I'm re-writing the patch with approach that James suggested, and instead of >> AARCH64_PREFETCH_NOT_BENEFICIAL there will be "&generic_prefetch_tune", so >> not much copy-paste of parameters. >> >>> >>> The purpose of introducing the macros in the ARM port was to avoid the >>> common problem of merging adding new parameters with adding new ports. >>> C initializers for these tables assign values sequentially with >>> zero-padding at the end, so failing to merge such changes carefully >>> leads to real bugs in the compiler (even if just performance bugs). >>> >>> I notice that the last entry in the current tune params table is an int, >>> rather than something that is type-checked (like the penultimate entry - >>> an enum). Having the final entry be type checked at least ensures that >>> the right number of elements exist, even if the order is not strictly >>> checked. >> >> If you prefer, I can move "&generic_prefetch_tune" to the bottom of "struct >> tune_params", thus enabling type checking on struct pointer. > > For the zero-cost/small-benefit tradefoff it gives, I'd say "why not". > > With that change in place and the obvious rebase needed for patch 6/6 in > place, both this (4/6) and patch 6/6 are OK.
Thanks, James. I'm going to commit this once bootstrap / regtest finish on aarch64-linux-gnu. -- Maxim Kuvyrkov www.linaro.org
0001-Port-prefetch-configuration-from-aarch32-to-aarch64-.patch
Description: Binary data
