On Wed, Jul 3, 2024 at 2:10 AM Andi Kleen <a...@linux.intel.com> wrote:
>
> liuhongt <hongtao....@intel.com> writes:
>
> > From: "H.J. Lu" <hjl.to...@gmail.com>
> >
> > According to Intel® 64 and IA-32 Architectures Optimization Reference
> > Manual[1], Branch Hint is updated for Redwood Cove.
> >
> > --------cut from [1]-------------------------
> > Starting with the Redwood Cove microarchitecture, if the predictor has
> > no stored information about a branch, the branch has the Intel® SSE2
> > branch taken hint (i.e., instruction prefix 3EH), When the codec
> > decodes the branch, it flips the branch’s prediction from not-taken to
> > taken. It then flushes the pipeline in front of it and steers this
> > pipeline to fetch the taken path of the branch.
> > --------cut end -----------------------------
> >
> > For -mtune-ctrl=branch_prediction_hints, always generate branch hint for
> > conditional branches, this tune is disabled by default.
> >
> > [1]
> > https://www.intel.com/content/www/us/en/content-details/821612/intel-64-and-ia-32-architectures-optimization-reference-manual-volume-1.html
> >
> > Bootstrapped and regtested on x86_64-pc-linux-gnu{-m32,}.
> > Ready push to trunk.
>
> So what does it do to code size?
> You may not want to do it with -Os.
It won't be enabled by Os, because it's guarded by
if (!optimize
|| optimize_function_for_size_p (cfun)
|| !TARGET_BRANCH_PREDICTION_HINTS)
return;
>
> Maybe it should be only done with actual profile feedback data
> available, i'm not sure if the builtin heuristics are good enough to
> justify it and there is a risk that it is very wrong.
Yes, best used with a PGO.
>
> Yes as long as it's disabled by default that's all not a problem, but it
> would need to be solved to enable it.
>
> -Andi
--
BR,
Hongtao
