[llvm] r262811 - Bitcode reader: Inline readAbbreviatedField in readRecord and move the enclosing loop in each case (NFC)

[Mehdi Amini via llvm-commits]

> On Mar 7, 2016, at 8:50 AM, Rafael Espíndola <rafael.espindola at gmail.com> wrote:
> 
> On 7 March 2016 at 11:29, Mehdi Amini <mehdi.amini at apple.com> wrote:
>> 
>>> On Mar 7, 2016, at 8:20 AM, Rafael Espíndola <rafael.espindola at gmail.com> wrote:
>>> 
>>> With the gold plugin I got a smaller but measurable speedup. Linking
>>> clang with -plugin-opt=disable-output went from 15.99 seconds to 15.85
>>> seconds.
>> 
>> Thanks for double-checking!
>> 
>> 0.9% improvement right?
> 
> 1.009X faster :-)

Yes.

Update for the list following our IRC conversation: my number were with full debug info, while Rafael measured with a Release build.

> 
>> 
>> I reported earlier that readRecord() accounts for ~%3 of the total link time, and that it is ~20% faster itself, so you're seeing even more improvements  than I did.
> 
> Note that this was with disable-output, so it doesn't include
> optimizations or codegen.

Oh OK I though it was just disabling writing the output but still running the pipeline.

Thanks!

-- 
Mehdi