D mentioned on Rust discussions site

[Stefan Koch]

On Saturday, 23 May 2020 at 09:03:38 UTC, NaN wrote:
> On Saturday, 23 May 2020 at 02:37:23 UTC, Walter Bright wrote:
>> On 5/22/2020 2:22 AM, Henrik wrote:
>>> do you find the ideas of the restrict keyword and VLA bad, or 
>>> only the current implementation in C99?
>>
>> I've done some experiments with restrict in gcc, examined the 
>> generated code, and found the optimizations it does and does 
>> not do baffling. I wonder if perhaps it is used so little that 
>> nobody really bothered to get the semantics right.
>
> There's a video on YouTube Chandler Caruth about how LLVM 
> optimises C++, one of the interesting points he made was that 
> const is literately zero help to the compiler for optimisation.
>
> He also said that inlining is the single most important 
> optimisation.

He's right about const. As for inlining LLVM loves inlining.
It's not always the right choice.
The reason why it's deemed important is because it allows the 
optimizer
to effectively see across function boundaries (since merging the 
function bodies removes that.) And it can specialize the callee 
code for the callers requirements.
However that can also lead to problems with code-size and 
therefore i-cache stalls, (the worst kind of stalls). that said, 
I-caches are pretty large these days to support out-of-order 
execution. So maybe that's less of a problem running on modern 
cpus.
On an embedded device you'll want to disable the inliner through 
(which is what -Os or Oz) does.