Request for comments: std.d.lexer (page 15)

08-Feb-2013 19:34, Brian Schott пишет: > On Friday, 8 February 2013 at 15:23:00 UTC, Brian Schott wrote: >> On Friday, 8 February 2013 at 15:12:23 UTC, Jacob Carlborg wrote: >>> DMD is still consistently faster, :( >> >> We might be getting to the part where we say that code generated by >> gcc is still consistently faster than code generated by ldc. > > For the lulz I compiled with "dmd -release -inline -noboundscheck -O > -m64 -property": > > http://hackerpilot.github.com/experimental/std_lexer/images/times3-dmd.png > > Yes, dmd vs ldc actually is a matter of 32 vs 18 ms for datetime. Would be intereesting to try gdc as dmd on linux uses gcc. -- Dmitry Olshansky

February 08, 2013

Re: Request for comments: std.d.lexer

Posted by Iain Buclaw
in reply to Dmitry Olshansky

Permalink

Iain Buclaw

Posted in reply to Dmitry Olshansky

Attachments:

text/html part

Permalink

On 8 February 2013 15:35, Dmitry Olshansky <dmitry.olsh@gmail.com> wrote:

> 08-Feb-2013 19:34, Brian Schott пишет:
>
>  On Friday, 8 February 2013 at 15:23:00 UTC, Brian Schott wrote:
>>
>>> On Friday, 8 February 2013 at 15:12:23 UTC, Jacob Carlborg wrote:
>>>
>>>> DMD is still consistently faster, :(
>>>>
>>>
>>> We might be getting to the part where we say that code generated by gcc is still consistently faster than code generated by ldc.
>>>
>>
>> For the lulz I compiled with "dmd -release -inline -noboundscheck -O -m64 -property":
>>
>> http://hackerpilot.github.com/**experimental/std_lexer/images/** times3-dmd.png<http://hackerpilot.github.com/experimental/std_lexer/images/times3-dmd.png>
>>
>> Yes, dmd vs ldc actually is a matter of 32 vs 18 ms for datetime.
>>
>
> Would be intereesting to try gdc as dmd on linux uses gcc.
>
>

What?  That's an outright fib. :-)


-- 
Iain Buclaw

*(p < e ? p++ : p) = (c & 0x0f) + '0';

On 8 February 2013 15:46, Brian Schott <briancschott@gmail.com> wrote: > On Friday, 8 February 2013 at 15:42:47 UTC, Iain Buclaw wrote: > >> What? That's an outright fib. :-) >> > > I think he means that "on linux the dmd binary is compiled by gcc" > > That's still lies. :o) -- Iain Buclaw *(p < e ? p++ : p) = (c & 0x0f) + '0';

08-Feb-2013 19:52, Iain Buclaw пишет: > On 8 February 2013 15:46, Brian Schott <briancschott@gmail.com > <mailto:briancschott@gmail.com>> wrote: > > On Friday, 8 February 2013 at 15:42:47 UTC, Iain Buclaw wrote: > > What? That's an outright fib. :-) > > > I think he means that "on linux the dmd binary is compiled by gcc" > > > That's still lies. :o) > g++ ? :) > > -- > Iain Buclaw > > *(p < e ? p++ : p) = (c & 0x0f) + '0'; -- Dmitry Olshansky

On Friday, 8 February 2013 at 15:35:55 UTC, Dmitry Olshansky wrote: > Would be intereesting to try gdc as dmd on linux uses gcc. GDC decided to randomly not fail to build on my system, so it's time for MOAR CHARTS. dmd -inline -noboundscheck -O -w -wi -m64 -property ldc2 -O2 -release -vectorize -m64 gdc -O3 -fno-bounds-check -frelease -m64 http://hackerpilot.github.com/experimental/std_lexer/images/times3-all.png

On 8 February 2013 16:00, Dmitry Olshansky <dmitry.olsh@gmail.com> wrote: > 08-Feb-2013 19:52, Iain Buclaw пишет: > >> On 8 February 2013 15:46, Brian Schott <briancschott@gmail.com <mailto:briancschott@gmail.com**>> wrote: >> >> On Friday, 8 February 2013 at 15:42:47 UTC, Iain Buclaw wrote: >> >> What? That's an outright fib. :-) >> >> >> I think he means that "on linux the dmd binary is compiled by gcc" >> >> >> That's still lies. :o) >> >> > g++ ? :) > > I see we could be doing this all day. :þ I'll lay down the hint, dmd compiles the source, not gcc. And although gcc may be invoked during a certain special stage of compilation, its actually just a driver to call a certain toolchain program that is outside of gcc. :-) -- Iain Buclaw *(p < e ? p++ : p) = (c & 0x0f) + '0';

On Friday, 8 February 2013 at 16:35:50 UTC, Brian Schott wrote: > dmd -inline -noboundscheck -O -w -wi -m64 -property Copy/pate fail. That's actually "dmd -release -inline -noboundscheck -O -w -wi -m64 -property"

On Friday, 8 February 2013 at 16:38:00 UTC, Iain Buclaw wrote: > I see we could be doing this all day. :þ We could. > I'll lay down the hint, dmd compiles the source, not gcc. And although gcc > may be invoked during a certain special stage of compilation, its actually > just a driver to call a certain toolchain program that is outside of gcc. > :-) What we're saying is that dmd, The Digital Mars D Compiler, is written in C++ and is thus built by GCC/G++. We can tell by looking at the Makefile that DMD doesn't build itself.

On 8 February 2013 16:35, Brian Schott <briancschott@gmail.com> wrote: > On Friday, 8 February 2013 at 15:35:55 UTC, Dmitry Olshansky wrote: > >> Would be intereesting to try gdc as dmd on linux uses gcc. >> > > GDC decided to randomly not fail to build on my system, so it's time for MOAR CHARTS. > > dmd -inline -noboundscheck -O -w -wi -m64 -property > ldc2 -O2 -release -vectorize -m64 > gdc -O3 -fno-bounds-check -frelease -m64 > http://hackerpilot.github.com/**experimental/std_lexer/images/** > times3-all.png<http://hackerpilot.github.com/experimental/std_lexer/images/times3-all.png> > Cool. How do you determine the speed times? -- Iain Buclaw *(p < e ? p++ : p) = (c & 0x0f) + '0';

Forums