Rather than a feature in AArch.td, I'd prefer a line in AArch64Subtarget::initializeProperties().

I'm getting mixed results on Exynos, with significant improvements and regressions. I'd like to run more tests, but, at the moment, I'd rather that this feature would be gated, either as @jfb or @rengolin suggested.

In D32564#1259703, @jfb wrote:

Do you have an ETA?

In D32564#1259585, @t.p.northover wrote:

Is that in trunk? I can't see any obvious candidates with a git grep -i exynos. If not, it's probably not relevant to this review.

In D32564#1259155, @jfb wrote:

Do you have specific worries? Or at least a timeline? Seems the patch can be reviewed without waiting for your exploration.

This is good data. However, I'd like to evaluate this patch a bit further on Exynos, if I may.

Can you please provide some figures on the code size saved and the effect on the performance of this change?

Me likes it.

@rnk has kindly reduced the issue to:

Add more tests.

Guard this transformation against targets that lack support for exp2(). In this case, refrain from transforming into either an intrinsic or a libcall.

Windows lacks exp2() support.

Yes, it makes sense.

LGTM, but, please, wait for @spatel's agreement.

In D49273#1216110, @spatel wrote:

Are there any negative tests where the base is not a power-of-2? If not, please add at least 1 test like that to make sure we're not transforming that.

Address the case when the result from exp{,2}() is reused elsewhere.

The function pow_exp2_libcall() in the test case is one such example without readnone. I'll augment the test case with a reuse scenario.

Rebase after D51194.

Increased test coverage in rL340462, where I also identified an issue with pow(exp{,2}(x), y).

In D50900#1204771, @fhahn wrote:

We were also thinking about expanding pow(x, n+0.5) -> x * x * x * ... * sqrt(x). Is this something you have also planned on adding?

Can you please rebase it, for D50036 touched the same function?

Rebase after the preliminary patches rL340060 and rL340061.

Thank you.

I'll commit the data type changes increasing type diversity on the side.

The dependent patches can be found in the Stack below: D49273 and D50036.

Thank you.

Added test for shrinking pow() when simplifying libcalls is disabled.

Remove additional simplification when exp10() is the base in pow(), since it's not enabled by many targets.

OK, then.

Closed by commit rL339354

Ping! 🔔

Ping! 🔔

Ping! 🔔