Hi all,

As some of you might already know, LDC has come dangerously close to being usable on Win32/MinGW recently.

I just posted a small writeup describing the current situation to my blog: http://klickverbot.at/blog/2013/05/the-state-of-ldc-on-windows/

Alpha-quality binary packages are available as part of 0.11.0 Beta 3: http://forum.dlang.org/post/mailman.871.1370475122.13711.digitalmars-d-ldc@puremagic.com

David

David Nadlinger:

> I just posted a small writeup describing the current situation to my blog: http://klickverbot.at/blog/2013/05/the-state-of-ldc-on-windows/

It gives me a "page not found"

Bye,
bearophile

On Thursday, 6 June 2013 at 00:05:03 UTC, bearophile wrote:
> It gives me a "page not found"

Whoops, I absentmindedly updated the time stamp, without noticing that it actually caused the month to change. Should be fixed now.

David

David Nadlinger:

> Whoops, I absentmindedly updated the time stamp, without noticing that it actually caused the month to change. Should be fixed now.

The page works now.
I'll try this LDC2.

For Windows I suggest to distribute a single 7zip that contains both ldc and the needed gcc libs (possibly only the essential parts of gcc that are necessary).

On Windows32 I often use this MinGW, that's updated often and contains Boost and other goodies, I don't know if this is good enough for LDC2:
http://nuwen.net/mingw.html

Bye,
bearophile

David Nadlinger:

> Alpha-quality binary packages are available as part of 0.11.0 Beta 3: http://forum.dlang.org/post/mailman.871.1370475122.13711.digitalmars-d-ldc@puremagic.com

I have done few small tests:

With no optimization dmd compiles about twice faster (or more) than ldc2.

When I want optimizations with this ldc I have used "-release -profile-verifier-noassert -O5". Is this good enough?
I don't know why ldc2 doesn't have "-noboundscheck".

- - - - - - - - - - - -

Regarding run time performance of the D code, if Walter is interested I have seen this code is about 6 times faster compiled with ldc2 compared to dmd:


import core.stdc.stdio, std.random, std.conv;

void main(in string[] args) {
    immutable uint N = (args.length == 2) ? args[1].to!uint : 1_000;
    auto rng = Xorshift(0);

    uint total = 0;
    for (uint i = 0; i < N; i++)
        total += uniform(0, 10, rng);

    printf("%u\n", total);
}


If I replace this line:
total += uniform(0, 10, rng);

with:
total += rng.front;  rng.popFront;

Then the code compiled with ldc2 is only about 30% faster or so.

Bye,
bearophile

> I have seen this code is about 6 times faster compiled with ldc2 compared to dmd:

I run it with 100000000 command line argument.

Bye,
bearophile

Am Thu, 06 Jun 2013 03:47:58 +0200
schrieb "bearophile" <bearophileHUGS@lycos.com>:

> When I want optimizations with this ldc I have used "-release
> -profile-verifier-noassert -O5". Is this good enough?
> I don't know why ldc2 doesn't have "-noboundscheck".

As fas as I know O4/O5 is currently not implemented and often putting -O (meaning -O2) is just as good as -O3. -release already disables bounds-checks like id does in dmd for non-@safe code. That said, every option that has an "enable" flag can be inverted:

-enable-boundscheck becomes -disable-boundscheck

Yes, your set of options is ok. :)

-- 
Marco

On Wednesday, 5 June 2013 at 23:45:02 UTC, David Nadlinger wrote:
> Hi all,
>
> As some of you might already know, LDC has come dangerously close to being usable on Win32/MinGW recently.
>
> I just posted a small writeup describing the current situation to my blog: http://klickverbot.at/blog/2013/05/the-state-of-ldc-on-windows/
>
> Alpha-quality binary packages are available as part of 0.11.0 Beta 3: http://forum.dlang.org/post/mailman.871.1370475122.13711.digitalmars-d-ldc@puremagic.com
>
> David

This is great news!
There's an often-repeated error in there. Borland cannot possibly have patented Structured Exception Handling -- it was invented by Microsoft, not by Borland. Of course anybody can claim a patent *related* to SEH, or that uses SEH in some way, but that's not the same thing as inventing it.
I don't know if Microsoft ever had any patents on SEH, but if they did, they must have expired long ago.

Am 06.06.2013 01:45, schrieb David Nadlinger:
> Hi all,
>
> As some of you might already know, LDC has come dangerously close to
> being usable on Win32/MinGW recently.
>
> I just posted a small writeup describing the current situation to my
> blog: http://klickverbot.at/blog/2013/05/the-state-of-ldc-on-windows/
>
> Alpha-quality binary packages are available as part of 0.11.0 Beta 3:
> http://forum.dlang.org/post/mailman.871.1370475122.13711.digitalmars-d-ldc@puremagic.com
>
>
> David

I really appreciate that you support windows now. I need a compiler with a good optimizer on windows and the last stable gdc version is at the level of dmd 2.060.
As soon as I find some time I will port my modified druntime so it works with ldc. If I find bugs I will file them.

-- 
Kind Regards
Benjamin Thaut

On 2013-06-06 10:38, Don wrote:

> This is great news!
> There's an often-repeated error in there. Borland cannot possibly have
> patented Structured Exception Handling -- it was invented by Microsoft,
> not by Borland. Of course anybody can claim a patent *related* to SEH,
> or that uses SEH in some way, but that's not the same thing as inventing
> it.
> I don't know if Microsoft ever had any patents on SEH, but if they did,
> they must have expired long ago.

That might be the case. But I don't think that matters as long as you can't convince the LLVM developers to implement support for SEH.

-- 
/Jacob Carlborg