"BOLT" post-link optimizer gives 15% speed boost to Clang

[Walter Bright]

On 10/24/2018 3:26 AM, Johan wrote:
> On Wednesday, 24 October 2018 at 01:57:38 UTC, Walter Bright wrote:
>> On 10/23/2018 3:07 PM, Johan Engelen wrote:
>>>    "Post-link optimization", what? Yes, indeed, optimization _after_ the 
>>> _linker_ has generated a _binary_.
>> Yup, it's reinvention of the same thing.
> Sorry, but such statements show a large lack of insight in what is going on 
> here. This isn't about who invented something first, the basic idea is of course 
> very old. DMC++ most certainly didn't and doesn't do what is being discussed.


 > "Post-link optimization", what? Yes, indeed, optimization _after_ the 
_linker_ has generated a _binary_.

That was the summary, implying it is some new idea. It's an incremental 
improvement. DMC++ does indeed do post-link optimization. If it was written as 
"BOLT improves on profile guided optimization after the linker has generated a 
binary by reordering basic blocks" I wouldn't have had any issue with that.

clang is an excellent implementation of a lot of ideas.