[PATCH] [AArch64] Add workaround for Cortex-A53 erratum (835769)

[Weiming Zhao]

I agree. We can leave it un-enabled in this commit.
Later on, we can decide if we should enable it by default or not.
The problem of not enabling it by default is most users may not even know the erratum and the flag. So they may just get wrong results.

Thanks,
Weiming

-----Original Message-----
From: Renato Golin [mailto:renato.golin at linaro.org] 
Sent: Friday, October 10, 2014 11:53 AM
To: Tim Northover
Cc: reviews+D5722+public+026d67210a377c2f at reviews.llvm.org; Bradley Smith; Weiming Zhao; cfe-commits at cs.uiuc.edu
Subject: Re: [PATCH] [AArch64] Add workaround for Cortex-A53 erratum (835769)

On 10 October 2014 19:34, Tim Northover <t.p.northover at gmail.com> wrote:
> I'm not sure that would actually help much. If most systems are
> A57+A53, then the usual option is probably "-mcpu=cortex-a57". An
> alternative might be a "cortex-a53-r0p0" CPU or something; that would 
> only help if we actually documented the CPUs supported of course.

With big-little, any affected A53 will need it, even if the code also runs on an A57. If we do enable it by default on A53, we'll have to do the same on A57.


> How widely deployed are these problematic CPUs likely to be?

And how much compile time increases? And what's the effect on benchmarks?

Until we have answers to those questions, I think a flag is probably the best option.


--renato