www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - LLVM

reply "Craig Black" <cblack ara.com> writes:
LLVM (http://llvm.org/) is a backend for compilers.  It is portable to many 
platforms, well documened, and actively developed.  It seems that they are 
making great progress with it.

A crazy idea.  Has anyone considered using LLVM with D's front end?  I know 
Walter has his own back end, but it's seems hard for one guy to maintain 
both a front and back end, and do a good job at both.  Even a programming 
god like Walter.  If we had the LLVM team actively developing and innovating 
the back end, then Walter could focus on innovating the front end and the D 
language.  Just a thought.

-Craig
Nov 20 2006
next sibling parent reply Gregor Richards <Richards codu.org> writes:
Craig Black wrote:

 LLVM (http://llvm.org/) is a backend for compilers.  It is portable to many 
 platforms, well documened, and actively developed.  It seems that they are 
 making great progress with it.
 
 A crazy idea.  Has anyone considered using LLVM with D's front end?  I know 
 Walter has his own back end, but it's seems hard for one guy to maintain 
 both a front and back end, and do a good job at both.  Even a programming 
 god like Walter.  If we had the LLVM team actively developing and innovating 
 the back end, then Walter could focus on innovating the front end and the D 
 language.  Just a thought.
 
 -Craig 
 
 


I have tried it with LLVM's GCC and GDC.  It didn't work out-of-the-box, 
but I don't think it would take a whole lot of hackery.

  - Gregor Richards
Nov 20 2006
parent reply "Craig Black" <cblack ara.com> writes:
 I have tried it with LLVM's GCC and GDC.  It didn't work out-of-the-box, 
 but I don't think it would take a whole lot of hackery.

  - Gregor Richards


How long ago did you try this?  LLVM just released a new 1.9 version that 
they claim is way better.  I'm not sure what you mean by out-of-the-box. 
But, like you say, I don't think it would be a massive effort.  It seems 
that LLVM is pretty well thought out.

-Craig
Nov 20 2006
parent reply Charlie <charlies nowhere.com> writes:
A third D compiler would be very cool, especially if its around when D 
goes 1.0.

Craig Black wrote:

 I have tried it with LLVM's GCC and GDC.  It didn't work out-of-the-box, 
 but I don't think it would take a whole lot of hackery.

  - Gregor Richards


 
 How long ago did you try this?  LLVM just released a new 1.9 version that 
 they claim is way better.  I'm not sure what you mean by out-of-the-box. 
 But, like you say, I don't think it would be a massive effort.  It seems 
 that LLVM is pretty well thought out.
 
 -Craig
Nov 20 2006
next sibling parent reply Gregor Richards <Richards codu.org> writes:
Charlie wrote:

 A third D compiler would be very cool, especially if its around when D 
 goes 1.0.
 
 Craig Black wrote:
 

 I have tried it with LLVM's GCC and GDC.  It didn't work 
 out-of-the-box, but I don't think it would take a whole lot of hackery.

  - Gregor Richards



 How long ago did you try this?  LLVM just released a new 1.9 version 
 that they claim is way better.  I'm not sure what you mean by 
 out-of-the-box. But, like you say, I don't think it would be a massive 
 effort.  It seems that LLVM is pretty well thought out.

 -Craig




LLVM uses GCC, so in all fairness it would hardly be a third D compiler, 
it would just be another face of GDC.

  - Gregor Richards
Nov 20 2006
parent reply Charlie <charlies nowhere.com> writes:
Hmm, I don't think it uses GCC as a backend.

 From the website: We find that LLVM is able to compile C++ into 
substantially better code than GCC .

It says it can use the GCC front-end for its LLVM backend, which I think 
is the opposite of GDC, which uses DMD front end to produce a tree 
usable by GCC backend.

I think DMD Frontend translated to LLVM backend would be completely 
independent of GCC, and the more compilers the better!

Charlie


Gregor Richards wrote:

 Charlie wrote:

 A third D compiler would be very cool, especially if its around when D 
 goes 1.0.

 Craig Black wrote:


 I have tried it with LLVM's GCC and GDC.  It didn't work 
 out-of-the-box, but I don't think it would take a whole lot of hackery.

  - Gregor Richards



 How long ago did you try this?  LLVM just released a new 1.9 version 
 that they claim is way better.  I'm not sure what you mean by 
 out-of-the-box. But, like you say, I don't think it would be a 
 massive effort.  It seems that LLVM is pretty well thought out.

 -Craig




 
 LLVM uses GCC, so in all fairness it would hardly be a third D compiler, 
 it would just be another face of GDC.
 
  - Gregor Richards
Nov 20 2006
next sibling parent Wayne <wanderon comcast.net> writes:
Take a look at

http://llvm.org/pubs/2006-04-25-GelatoLLVMIntro.pdf

LLVM optionally uses GCC as a front end for C/C++ FORTRAN ADA and Java.
It looks like the existing DMD front end could be adapted to LLVM

Wayne

Charlie Wrote:

 Hmm, I don't think it uses GCC as a backend.
  From the website: We find that LLVM is able to compile C++ into
 substantially better code than GCC .
 It says it can use the GCC front-end for its LLVM backend, which I think
 is the opposite of GDC, which uses DMD front end to produce a tree
 usable by GCC backend.
 I think DMD Frontend translated to LLVM backend would be completely
 independent of GCC, and the more compilers the better!
 Charlie
 Gregor Richards wrote:

 Charlie wrote:

 A third D compiler would be very cool, especially if its around when D
 goes 1.0.

 Craig Black wrote:


 I have tried it with LLVM's GCC and GDC.  It didn't work
 out-of-the-box, but I don't think it would take a whole lot of hackery.

  - Gregor Richards



 How long ago did you try this?  LLVM just released a new 1.9 version
 that they claim is way better.  I'm not sure what you mean by
 out-of-the-box. But, like you say, I don't think it would be a
 massive effort.  It seems that LLVM is pretty well thought out.

 -Craig




 LLVM uses GCC, so in all fairness it would hardly be a third D compiler,
 it would just be another face of GDC.

  - Gregor Richards
Nov 20 2006
prev sibling parent Gregor Richards <Richards codu.org> writes:
Charlie wrote:

 
 
 Gregor Richards wrote:
 

 Charlie wrote:


 A third D compiler would be very cool, especially if its around when 
 D goes 1.0.

 Craig Black wrote:


 I have tried it with LLVM's GCC and GDC.  It didn't work 
 out-of-the-box, but I don't think it would take a whole lot of 
 hackery.

  - Gregor Richards




 How long ago did you try this?  LLVM just released a new 1.9 version 
 that they claim is way better.  I'm not sure what you mean by 
 out-of-the-box. But, like you say, I don't think it would be a 
 massive effort.  It seems that LLVM is pretty well thought out.

 -Craig




 LLVM uses GCC, so in all fairness it would hardly be a third D 
 compiler, it would just be another face of GDC.

  - Gregor Richards


 Hmm, I don't think it uses GCC as a backend.

  From the website: We find that LLVM is able to compile C++ into
 substantially better code than GCC .

 It says it can use the GCC front-end for its LLVM backend, which I think
 is the opposite of GDC, which uses DMD front end to produce a tree
 usable by GCC backend.

 I think DMD Frontend translated to LLVM backend would be completely
 independent of GCC, and the more compilers the better!

 Charlie


)top-post don't please(


Normal GCC (simplified):
[C] code -> GIMPLE -> target ASM

LLVM GCC (simplified):
[C] code -> GIMPLE -> LLVM ASM

Normal GDC (simplified);
D code -> GIMPLE -> target ASM

LLVM GDC (simplified):
D code -> GIMPLE -> LLVM ASM


The same component of GCC (namely, the IR, GIMPLE) is both a backend to 
GDC and a frontend to LLVM.

  - Gregor Richards
Nov 20 2006
prev sibling parent reply Wolfgang Draxinger <wdraxinger darkstargames.de> writes:
Charlie wrote:


 A third D compiler would be very cool, especially if its around
 when D goes 1.0.


I already mentioned my attempts on a D ABI for which I planned to
fork GDC. I think, this is the perfect moment to get my LLVM
based D compiler started. So in 3 words:

I'm on it!

Wolfgang Draxinger
-- 
E-Mail address works, Jabber: hexarith jabber.org, ICQ: 134682867
Nov 20 2006
parent "Craig Black" <cblack ara.com> writes:
 I'm on it!


Excellent!  I'm very curious about performance comparisons.  Keep us posted!

-Craig
Nov 20 2006
prev sibling next sibling parent jcc7 <technocrat7 gmail.com> writes:
== Quote from Craig Black (cblack ara.com)'s article

 LLVM (http://llvm.org/) is a backend for compilers.  It is portable to
 many platforms, well documened, and actively developed.  It seems that > they


are making great progress with it.

 A crazy idea.  Has anyone considered using LLVM with D's front end?


Someone has looked into doing this a couple years ago. I don't know if he ran
into
a serious problem (or perhaps just lost interest in the idea), but I'm guessing
he
never finished his project:

http://www.digitalmars.com/d/archives/D/gnu/762.html
Nov 20 2006
prev sibling next sibling parent reply "Lionello Lunesu" <lionello lunesu.remove.com> writes:
Great idea!
I just noticed however that their IR language has no "real" (long double) 
float representation?

L.
Nov 20 2006
parent reply "Craig Black" <cblack ara.com> writes:
"Lionello Lunesu" <lionello lunesu.remove.com> wrote in message 
news:ejt4m7$2o18$1 digitaldaemon.com...

 Great idea!
 I just noticed however that their IR language has no "real" (long double) 
 float representation?

 L.



Good point.  Perhaps someone could suggest that they add one.

-Craig
Nov 20 2006
parent reply =?ISO-8859-1?Q?Anders_F_Bj=F6rklund?= <afb algonet.se> writes:
Craig Black wrote:


I just noticed however that their IR language has no "real" (long double) 
float representation?


 
 Good point.  Perhaps someone could suggest that they add one.


Just use the "double" type, that is what PowerPC does anyway...

--anders
Nov 20 2006
parent reply "Craig Black" <cblack ara.com> writes:
64 bit doubles are sufficient for most applications, but not all.

-Craig
Nov 20 2006
parent Walter Bright <newshound digitalmars.com> writes:
Craig Black wrote:

 64 bit doubles are sufficient for most applications, but not all.


If the underlying hardware does not support extended real types, then it 
is reasonable for a D compiler to treat them as doubles. If the hardware 
does support extended real, then for a D compiler to not support it, 
well, I'd call that not a proper D implementation.
Nov 20 2006
prev sibling next sibling parent Wolfgang Draxinger <wdraxinger darkstargames.de> writes:
Craig Black wrote:


 LLVM (http://llvm.org/) is a backend for compilers.  It is
 portable to many
 platforms, well documened, and actively developed.  It seems
 that they are making great progress with it.


Cool. A few weeks ago I wondered why there is not (well there is,
but I wasn't aware of it) a comiler suite, where each step of
the compilation process can be individually controlled. Now I
know about LLVM. Thanks.

Wolfgang Draxinger
-- 
E-Mail address works, Jabber: hexarith jabber.org, ICQ: 134682867
Nov 20 2006
prev sibling next sibling parent "John Reimer" <terminal.node gmail.com> writes:
On Mon, 20 Nov 2006 10:56:53 -0800, Craig Black <cblack ara.com> wrote:


 LLVM (http://llvm.org/) is a backend for compilers.  It is portable to  
 many
 platforms, well documened, and actively developed.  It seems that they  
 are
 making great progress with it.

 A crazy idea.  Has anyone considered using LLVM with D's front end?  I  
 know
 Walter has his own back end, but it's seems hard for one guy to maintain
 both a front and back end, and do a good job at both.  Even a programming
 god like Walter.  If we had the LLVM team actively developing and  
 innovating
 the back end, then Walter could focus on innovating the front end and  
 the D
 language.  Just a thought.

 -Craig



I like this idea! I'm not sure how difficult it would be to include a D  
frontend with the default distribution (in the way that several gcc  
frontends seem to be included), but the licensing certainly appears to be  
quite accommodating.

-JJR
Nov 20 2006
prev sibling parent reply Chad J <gamerChad _spamIsBad_gmail.com> writes:
Craig Black wrote:

 LLVM (http://llvm.org/) is a backend for compilers.  It is portable to many 
 platforms, well documened, and actively developed.  It seems that they are 
 making great progress with it.
 
 A crazy idea.  Has anyone considered using LLVM with D's front end?  I know 
 Walter has his own back end, but it's seems hard for one guy to maintain 
 both a front and back end, and do a good job at both.  Even a programming 
 god like Walter.  If we had the LLVM team actively developing and innovating 
 the back end, then Walter could focus on innovating the front end and the D 
 language.  Just a thought.
 
 -Craig 
 
 


Hmmm... looks like they are working on an ARM backend.  This could be 
nice.  If they get that ARM backend done by say, summer of next year, 
and the D front-end is plugged in, then I might be willing to take a 
crack at making ARM work on that one in addition to GDC.  Then I could 
decide which one is nicer to work with and focus on that.  Kind of a 
longshot though (they'd need wince support too).

Admittedly the current arm-wince-pe GDC I put together has some uh 
...annoyances...   and I've no idea how to solve them.

Now I wonder, is Walter really maintaining the backend much?  I thought 
that part of dmd was pretty solid already and he is just focusing on the 
front-end and other D affairs.

What would impress me is if the main development of D would occur on 
this LLVM compiler, or at least be readily portable to it within hours 
or a few days (this assumes LLVM is a good quality backend that Walter 
wouldn't mind switching to).  The utility I am looking for there is 
having a bleeding-edge D compiler that is also very retargetable, and 
doesn't require big man-hours to be maintained as it seems gdc does. 
DMD is great on windows, but it is fairly useless on my pda since, well, 
it doesn't even generate code for that at all!
Nov 20 2006
parent reply "John Reimer" <terminal.node gmail.com> writes:
On Mon, 20 Nov 2006 13:41:28 -0800, Chad J  
<gamerChad _spamIsBad_gmail.com> wrote:


 Craig Black wrote:

 LLVM (http://llvm.org/) is a backend for compilers.  It is portable to  
 many platforms, well documened, and actively developed.  It seems that  
 they are making great progress with it.
  A crazy idea.  Has anyone considered using LLVM with D's front end?  I  
 know Walter has his own back end, but it's seems hard for one guy to  
 maintain both a front and back end, and do a good job at both.  Even a  
 programming god like Walter.  If we had the LLVM team actively  
 developing and innovating the back end, then Walter could focus on  
 innovating the front end and the D language.  Just a thought.
  -Craig


 Hmmm... looks like they are working on an ARM backend.  This could be  
 nice.  If they get that ARM backend done by say, summer of next year,  
 and the D front-end is plugged in, then I might be willing to take a  
 crack at making ARM work on that one in addition to GDC.  Then I could  
 decide which one is nicer to work with and focus on that.  Kind of a  
 longshot though (they'd need wince support too).

 Admittedly the current arm-wince-pe GDC I put together has some uh  
 ...annoyances...   and I've no idea how to solve them.

 Now I wonder, is Walter really maintaining the backend much?  I thought  
 that part of dmd was pretty solid already and he is just focusing on the  
 front-end and other D affairs.

 What would impress me is if the main development of D would occur on  
 this LLVM compiler, or at least be readily portable to it within hours  
 or a few days (this assumes LLVM is a good quality backend that Walter  
 wouldn't mind switching to).  The utility I am looking for there is  
 having a bleeding-edge D compiler that is also very retargetable, and  
 doesn't require big man-hours to be maintained as it seems gdc does. DMD  
 is great on windows, but it is fairly useless on my pda since, well, it  
 doesn't even generate code for that at all!



I'm wondering if Walter will have the same source "tainting" issue with  
LLVM as he did for gcc.  I wish not because it would be much better if the  
reference compiler development could continue on a completely open system.

-JJR
Nov 20 2006
next sibling parent reply Gregor Richards <Richards codu.org> writes:
John Reimer wrote:

 On Mon, 20 Nov 2006 13:41:28 -0800, Chad J  
 <gamerChad _spamIsBad_gmail.com> wrote:
 

 Craig Black wrote:


 LLVM (http://llvm.org/) is a backend for compilers.  It is portable 
 to  many platforms, well documened, and actively developed.  It seems 
 that  they are making great progress with it.
  A crazy idea.  Has anyone considered using LLVM with D's front end?  
 I  know Walter has his own back end, but it's seems hard for one guy 
 to  maintain both a front and back end, and do a good job at both.  
 Even a  programming god like Walter.  If we had the LLVM team 
 actively  developing and innovating the back end, then Walter could 
 focus on  innovating the front end and the D language.  Just a thought.
  -Craig



 Hmmm... looks like they are working on an ARM backend.  This could be  
 nice.  If they get that ARM backend done by say, summer of next year,  
 and the D front-end is plugged in, then I might be willing to take a  
 crack at making ARM work on that one in addition to GDC.  Then I 
 could  decide which one is nicer to work with and focus on that.  Kind 
 of a  longshot though (they'd need wince support too).

 Admittedly the current arm-wince-pe GDC I put together has some uh  
 ...annoyances...   and I've no idea how to solve them.

 Now I wonder, is Walter really maintaining the backend much?  I 
 thought  that part of dmd was pretty solid already and he is just 
 focusing on the  front-end and other D affairs.

 What would impress me is if the main development of D would occur on  
 this LLVM compiler, or at least be readily portable to it within 
 hours  or a few days (this assumes LLVM is a good quality backend that 
 Walter  wouldn't mind switching to).  The utility I am looking for 
 there is  having a bleeding-edge D compiler that is also very 
 retargetable, and  doesn't require big man-hours to be maintained as 
 it seems gdc does. DMD  is great on windows, but it is fairly useless 
 on my pda since, well, it  doesn't even generate code for that at all!


 
 
 
 I'm wondering if Walter will have the same source "tainting" issue with  
 LLVM as he did for gcc.  I wish not because it would be much better if 
 the  reference compiler development could continue on a completely open 
 system.
 
 -JJR


I have now posted this at least three times.

LLVM's compiler is GCC.

  - Gregor Richards
Nov 20 2006
parent "John Reimer" <terminal.node gmail.com> writes:
On Mon, 20 Nov 2006 14:37:23 -0800, Gregor Richards <Richards codu.org>  
wrote:



 with  LLVM as he did for gcc.  I wish not because it would be much  
 better if the  reference compiler development could continue on a  
 completely open system.
  -JJR


 I have now posted this at least three times.

 LLVM's compiler is GCC.

   - Gregor Richards


Yes, I've seen your posts. But it's unclear despite the nice little graph  
that you posted for us to consider.

Quoting the site:

"LLVM is also a collection of source code that implements the language and  
compilation strategy. The primary components of the LLVM infrastructure  
are a GCC-based C & C++ front-end, a link-time optimization framework with  
a growing set of global and interprocedural analyses and transformations,  
static back-ends for the X86, PowerPC, IA-64, Alpha and SPARC  
architectures, a back-end which emits portable C code, and a Just-In-Time  
compiler for X86 and PowerPC processors."

My question is why do they say the the LLVM infrastructure uses the  
GCC-based /C & C++ front-end/?  The backends are not GCC...  I guess this  
is just depends how far back the backend is :)?   (Oh wait... GIMPLE is a  
frontend for the virtual machine)

You mentioned GIMPLE as being used for the IR (Intermediate  
Representation) which GDC uses to interface with the GCC backend. It's  
still a little unlcear to me, so have patience.

1) Is GIMPLE a problem for use here? (License-wise)
2) If GIMPLE is the IR, this means that GDC already has all that's  
necessary to attach to LLVM?
3) Whatever the case, including a GDC front end with LLVM may be easier  
than doing so with FSF.  FSF seems to want complete ownership of the D  
frontend before they will accept it into the fold of gcc.  Not sure how  
the LLVM group will act in this matter, but there licensing method seems  
quite different.

-JJR
Nov 20 2006
prev sibling parent reply Walter Bright <newshound digitalmars.com> writes:
John Reimer wrote:

 I'm wondering if Walter will have the same source "tainting" issue with 
 LLVM as he did for gcc.


Yes, since it is copyrighted. However, I *can* read documentation about 
the compiler (one cannot copyright an idea, ideas are covered by patent 
law, not copyrights).

Aside from working with the LLVM source, I am happy to help out in any 
way I can those who want to integrated the D front end with LLVM.

One advantage LLVM has over gcc is that gcc is reluctant to include gdc 
as part of the default gcc distribution. If LLVM is willing to include D 
with their main distribution, that would be very good for D users.
Nov 20 2006
parent Paolo Invernizzi <arathorn NOSPAM_fastwebnet.it> writes:
I've played with LLVM and D, one yr ago, submitting patches to LLVM for 
VC7 compilation of the LLVM backend, just to try to do a backend for DMD 
frontend on Windows.

So, as seems that everybody is playing with the frontend, it's not time 
to have it in a svn repository on DigitalMars/somewhere?
We are all applying the same patches, merging subsequent Walter release, 
adding headers, and so on...

Can't we work on the same codebase? At least Walter and David Friedman!

Just my 2c....

---
Paolo Invernizzi


Walter Bright wrote:

 John Reimer wrote:

 I'm wondering if Walter will have the same source "tainting" issue 
 with LLVM as he did for gcc.


 
 Yes, since it is copyrighted. However, I *can* read documentation about 
 the compiler (one cannot copyright an idea, ideas are covered by patent 
 law, not copyrights).
 
 Aside from working with the LLVM source, I am happy to help out in any 
 way I can those who want to integrated the D front end with LLVM.
 
 One advantage LLVM has over gcc is that gcc is reluctant to include gdc 
 as part of the default gcc distribution. If LLVM is willing to include D 
 with their main distribution, that would be very good for D users.
Nov 21 2006