[PATCH] D115016: [CostModel][X86] Add i64 mul cost for avx as 1cy

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon added a comment.

In D115016#3169651 <https://reviews.llvm.org/D115016#3169651>, @HaohaiWen wrote:

>> This is a classic example of where having cost tables driven from (accurate) scheduler models would be much more useful (D46276 <https://reviews.llvm.org/D46276>), but there has been very little interest in helping polish those models to make that a viable option.
>
> Thanks for sharing this. I think using scheduler models based cost tables is more accurate and reasonable. Is there any plan to commit D46276 <https://reviews.llvm.org/D46276>?

I can investigate resurrecting this - but there was a number of issues that needed 'yak shaving' first IIRC (it's been a while I'll need to review it again) - and the models have to be opt-in (I was using the complete flag but that might be suitable).

> Do we have any plan to use double as cost value type since many instruction's cost is less than 1 on modern superscalar CPU?

We have the InstructionCost struct now so that might be possible? A more critical feature would be to 'add' costs correctly for throughput cost kinds - depending on pipe usage etc. - instead of just numerically adding the costs together, which only makes sense for latency / size cost kinds.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D115016/new/

https://reviews.llvm.org/D115016