Is there a way to make ldc do more aggressive inlining other than

    pragma(inline, true)

?

Reason for asking is that

https://github.com/nordlow/phobos-next/blob/master/src/open_hashmap_or_hashset.d

achieves much better performance when I qualify some inner loop functions with

    pragma(inline, true)

instead of

    pragma(inline)

eventhough I compile with -release -inline -nobounds flags.

Unfortunately these functions cannot be inlined by dmd in release mode so my code either runs slower than possible or it cannot be built by dmd in release mode.

And I haven't found a way to conditionally qualify these inner loop functions with `pragma(inline, true)` for the ldc case only.

On Sunday, 25 March 2018 at 22:09:43 UTC, Nordlöw wrote:
> And I haven't found a way to conditionally qualify these inner loop functions with `pragma(inline, true)` for the ldc case only.

From https://dlang.org/spec/pragma.html#inline: 'If inside a function, it affects the function it is enclosed by.'

So:

void foo()
{
    version(LDC) pragma(inline, true); // affects foo()
    ...
}

On Sunday, 25 March 2018 at 22:30:50 UTC, kinke wrote:
> void foo()
> {
>     version(LDC) pragma(inline, true); // affects foo()
>     ...
> }

Wonderful, thanks!

On Sunday, 25 March 2018 at 22:09:43 UTC, Nordlöw wrote:
> eventhough I compile with -release -inline -nobounds flags.

Just to make sure: are you passing -O as well?

On Monday, 26 March 2018 at 16:02:30 UTC, Rene Zwanenburg wrote:
> On Sunday, 25 March 2018 at 22:09:43 UTC, Nordlöw wrote:
>> eventhough I compile with -release -inline -nobounds flags.
>
> Just to make sure: are you passing -O as well?

Yes I am, thanks, via the dub spec

buildType "release-nobounds" {
          dflags "-mcpu=native" "-O3"
          buildOptions "releaseMode" "optimize" "noBoundsCheck" "inline"
}

I didn't measure any significant difference between -O and -O3.

Is each optimization level `x` in `-Ox` defined in the same way for clang and ldc? If so, where's the best documentation for it?

On Monday, 26 March 2018 at 18:47:17 UTC, Nordlöw wrote:
>
> Yes I am, thanks, via the dub spec
>
> buildType "release-nobounds" {
>           dflags "-mcpu=native" "-O3"
>           buildOptions "releaseMode" "optimize" "noBoundsCheck" "inline"
> }
>
> I didn't measure any significant difference between -O and -O3.
>
> Is each optimization level `x` in `-Ox` defined in the same way for clang and ldc? If so, where's the best documentation for it?

https://wiki.dlang.org/Using_LDC

-O is equivalent to -O3

On Monday, 26 March 2018 at 21:11:12 UTC, jmh530 wrote:
>> Is each optimization level `x` in `-Ox` defined in the same way for clang and ldc? If so, where's the best documentation for it?
>
> https://wiki.dlang.org/Using_LDC

Thx!