LLVM and Ada

LLVM and Ada

[Vincent DIEMUNSCH]

Dear fellow Ada programmers,

I heard that Duncan Sands has ported GNAT to LLVM in the DraggonEgg project, therefore I have two questions :
- is there an Ada interface to LLVM or LLVM Bitcode available ?
- where can I find information on the mapping between Ada types and LLVM types ?

Kind regards,

Vincent

Re: LLVM and Ada

[Simon Clubley]

On 2016-10-20, Vincent DIEMUNSCH <vincent.diemunsch@gmail.com> wrote:
> Dear fellow Ada programmers,
>
> I heard that Duncan Sands has ported GNAT to LLVM in the DraggonEgg project, therefore I have two questions :

DragonEgg is now obsolete unless someone here has more recent information
than I do.

> - is there an Ada interface to LLVM or LLVM Bitcode available ?
> - where can I find information on the mapping between Ada types and LLVM types ?
>

There is no viable Ada frontend for current versions of LLVM (AFAIK)
although various people have looked at this.

Simon.

-- 
Simon Clubley, clubley@remove_me.eisner.decus.org-Earth.UFP
Microsoft: Bringing you 1980s technology to a 21st century world

Re: LLVM and Ada

[Dennis Lee Bieber]

On Thu, 20 Oct 2016 07:28:01 -0700 (PDT), Vincent DIEMUNSCH
<vincent.diemunsch@gmail.com> declaimed the following:

>I heard that Duncan Sands has ported GNAT to LLVM in the DraggonEgg project, therefore I have two questions :

	For a study at work I had to get DragonEgg working... It took me three
Linux variants (my immediate supervisor provided a VM apparently
recommended by the client, and I tried using a VM install of Debian
[primarily as Debian provided the repository as DVD images -- company
firewall blocks apt-get except for the corporate approved Ubuntu image
using a corporate maintained repository]) to find one which had an old
enough GCC with which to build it. As I recall, I had revert from a 4.9 to
a 4.3 GCC (or something that drastic). I couldn't even build 4.3 using 4.9
due to changes in the make system, and sure wasn't going to try to build
sequentially 4.9 -> 4.8 -> 4.7 etc, until getting to the target variation.

	I managed to get it to build, and even produce bitcode from an
Helloworld.adb.

	However -- it would require porting the Ada RTL to do anything
meaningful... And since I didn't find anything in LLVM to produce a native
binary -- other than having it convert the bitcode into native assembly for
GCC to link -- it proved useless for the goal of porting an Ada application
to a board that didn't have support from AdaCore. 


	And, as mentioned, DragonEgg with Ada support is years out of date.

-- 
	Wulfraed                 Dennis Lee Bieber         AF6VN
    wlfraed@ix.netcom.com    HTTP://wlfraed.home.netcom.com/

Re: LLVM and Ada

[vincent.diemunsch]

Le jeudi 20 octobre 2016 20:20:44 UTC+2, Simon Clubley a écrit :

> There is no viable Ada frontend for current versions of LLVM (AFAIK)
> although various people have looked at this.

Thank you Simon for your answer. 

Why Gnat on LLVM has been abandoned ? Was LLVM less appropriate than GCC for GNAT ?

My question precisely was not about the Ada frontend in fact but about a binding to use the LLVM from an Ada program that is a sort of compiler. I think that those who ported GNAT to LLVM should have created this kind of binding, no ?

Kind regards,

Vincent 

 

Re: LLVM and Ada

[Brian Drummond]

On Fri, 21 Oct 2016 01:32:03 -0700, vincent.diemunsch wrote:

> Le jeudi 20 octobre 2016 20:20:44 UTC+2, Simon Clubley a écrit :
> 
>> There is no viable Ada frontend for current versions of LLVM (AFAIK)
>> although various people have looked at this.
> 
> Thank you Simon for your answer.
> 
> Why Gnat on LLVM has been abandoned ? 

My understanding is that it never worked *completely* - maybe it did on 
32-bit but the 64-bit backend was giving problems, and there wasn't 
enough effort to resolve these issues and keep it up to date with newer 
LLVM releases. So if you can find the sources, they are (a) not 
completely functional and (b) increasingly obsolete (see Dennis' 
experience buisding with gcc versions back to 4.3)


> Was LLVM less appropriate than GCC
> for GNAT ?

Yes. Roughly, the issue is that LLVM is designed to not support local 
subprograms (e.g. procedures declared local to other procedures, sharing 
the scope of the outer procedure).

This limits its (easy) use to C-like languages, and if you are to support 
an Ada-like language you have to find ways round this problem.

The DragonEgg compiler did have such support but as far as I know it was 
incomplete in some resperts - memory suggests, exception propagation 
didn't work properly for local subprograms, but I'm vague on the details.

> My question precisely was not about the Ada frontend in fact but about a
> binding to use the LLVM from an Ada program that is a sort of compiler.
> I think that those who ported GNAT to LLVM should have created this kind
> of binding, no ?

This has been done.

Not for Ada, but for a compiler written in Ada for another (somewhat) Ada-
like language ... the ghdl compiler for VHDL.

Here, local subprograms are supported, however VHDL doesn't have 
exceptions, so that specific bullet has been dodged.

https://github.com/tgingold/ghdl

GHDL is actively supported, and compiles VHDL to three differen compiler 
backends - gcc (upto 4.9), LLVM(3.5) and its own JIT compiler, mcode, all 
via its own intermediate language "ortho" (which is not normally written 
out to file, but can be written and read back in through the LLVM back 
end for compiler debugging purposes)

So the compilation stage between "ortho" and the LLVM backend may be 
worth looking at, to help develop your own compiler.

-- Brian

Re: LLVM and Ada

[vincent.diemunsch]

Le vendredi 21 octobre 2016 12:33:41 UTC+2, Brian Drummond a écrit :

Thank you very much Brian for your kind and complete response !

> The DragonEgg compiler did have such support but as far as I know it was 
> incomplete in some resperts - memory suggests, exception propagation 
> didn't work properly for local subprograms, but I'm vague on the details.
> 
That's a pity indeed. And I suppose it is a difficult point to solve in LLVM...


> So the compilation stage between "ortho" and the LLVM backend may be 
> worth looking at, to help develop your own compiler.

Yes ! That's exactly what I expected. I will have a look at it.
Kind regards,

Vincent

Re: LLVM and Ada

[jparkerg000]

On Friday, October 21, 2016 at 9:32:04 AM UTC+1, vincent....@gmail.com wrote:
> Le jeudi 20 octobre 2016 20:20:44 UTC+2, Simon Clubley a écrit :
> 
> > There is no viable Ada frontend for current versions of LLVM (AFAIK)
> > although various people have looked at this.
> 
> Thank you Simon for your answer. 
> 
> Why Gnat on LLVM has been abandoned ? Was LLVM less appropriate than GCC for GNAT ?
> 
> My question precisely was not about the Ada frontend in fact but about a binding to use the LLVM from an Ada program that is a sort of compiler. I think that those who ported GNAT to LLVM should have created this kind of binding, no ?
> 
> Kind regards,
> 
> Vincent



The author of the GNAT LLVM compiler posted a few remarks on the subject in 2014. A search of comp.lang.ada should pull up the thread easily enough. I'll cut and paste the text for you.     J.



duncan...@deepbluecap.com 	
10/17/14
Hi,

> > That was the DragonEgg project, wasn't it? I got the impression it
>
> > foundered on some shortcoming of LLVM for non-C languages; not certain
>
> > which, but possibly the one regarding nested functions that Tristan has
>
> > overcome with ghdl.

no, there were no blocking technical problems.  For example I got nested function support working properly a gazillion years ago, as Luke says.  Exception handling took longer to be completely reliable, but that was also resolved years ago too.  In both cases changes needed to be made to LLVM, but those changes are all old hat now.  All that happened is that I lost interest in the project and no-one stepped forward to take it over.  That means that it's been quietly bit rotting for a while now.  For example: to accommodate internal changes in gcc-4.7 and later, dragonegg's ABI support needs to be rewritten, but I'd already lost interest in the project so never did it, which means that while it still mostly sorta works, thanks to some hacks, the plugin easily crashes on Ada tagged types starting from this gcc version.

Ciao, Duncan. 

Re: LLVM and Ada

[vincent.diemunsch]

Le vendredi 21 octobre 2016 15:47:06 UTC+2, jpark...@gmail.com a écrit :

> The author of the GNAT LLVM compiler posted a few remarks on the subject in 2014. A search of comp.lang.ada should pull up the thread easily enough. I'll cut and paste the text for you.     J.
> 

> no, there were no blocking technical problems.  For example I got nested function support working properly a gazillion years ago, as Luke says.  Exception handling took longer to be completely reliable, but that was also resolved years ago too.  In both cases changes needed to be made to LLVM, but those changes are all old hat now.  
All that happened is that I lost interest in the project and no-one stepped forward to take it over.  

That's really good news (for LLVM I mean :-) ) ! Thank you very much "J.".
Regards,
Vincent

Re: LLVM and Ada

[Lucretia]

On Friday, 21 October 2016 11:33:41 UTC+1, Brian Drummond  wrote:

> Here, local subprograms are supported, however VHDL doesn't have 
> exceptions, so that specific bullet has been dodged.
> 
> https://github.com/tgingold/ghdl
> 

The whole source is under GPL v2, no linking exception. You may want to ask Tristan to extract the LLVM interfaace into it's own project and put it under a less restrictive license. Obviously, GPL is fine if your project is GPL.

Luke.

Re: LLVM and Ada

[Lucretia]

On Friday, 21 October 2016 09:32:04 UTC+1, vincent....@gmail.com  wrote:

> My question precisely was not about the Ada frontend in fact but about a binding to use the LLVM from an Ada program that is a sort of compiler. I think that those who ported GNAT to LLVM should have created this kind of binding, no ?

I have looked into this and there are reasons:

1) C++ is hard to bind to, you either have to wrap the classes in C and flatten the api or use GNAT's class binding representation clauses - which isn't portable to other compilers - this may or may not matter to you.

2) There is a C binding in LLVM's source, but according to their own docs (that I cannot find right now!) say that the C interface isn't as complete as the C++ one.

3) The API changes a lot, which means it's harder to bind to.

I would look at which API you intend to bind to first to see if it can do what you need, then start by generating the bindings with "gcc -fdump-ada-spec-slim" and then hand massage them, GCC generates a ton of crap and they're not readable or compilable.

Whatever binding you create, don't release them under GPL, pick something less restrictive and other's will use them. Really annoys me when people create a binding to something and then puts them under a license the original library wasn't released under.

Re: LLVM and Ada

[Vincent]

Le samedi 22 octobre 2016 14:51:36 UTC+2, Lucretia a écrit :

> [...] start by generating the bindings with "gcc -fdump-ada-spec-slim" and then hand massage them ...

Yes it's an alternative I have to consider.
 
> Whatever binding you create, don't release them under GPL, pick something less restrictive and other's will use them. Really annoys me when people create a binding to something and then puts them under a license the original library wasn't released under.

I totally agree ! The binding should be under the same licence.

Thank you Luke.
Regards,

Vincent

Re: LLVM and Ada

[carlojpisani]

hi guys
I need to port ghdl to PPC, MIPS, HPPA
on PPC there are old bootstrapper (e.g. 4.1.1)
but I haven't seen them for MIPS and HPPA

vhdl seems able to be compiled by llvm, but ... llvm seems do have no support for MIPS and HPPA

no dice

also, I'd like to port vhdl to SGI/MIPS under IRIX
in this case, there are commercial Ada compilers (e.g. with the MIPS/PRO toolchain)

let me know

Re: LLVM and Ada

[Dan'l Miller]

On Wednesday, March 21, 2018 at 8:08:50 AM UTC-5, carloj...@gmail.com wrote:
> hi guys
> I need to port ghdl to PPC, MIPS, HPPA
> on PPC there are old bootstrapper (e.g. 4.1.1)
> but I haven't seen them for MIPS and HPPA
> 
> vhdl seems able to be compiled by llvm, but ... llvm seems do have no support for MIPS and HPPA
> 
> no dice
> 
> also, I'd like to port vhdl to SGI/MIPS under IRIX
> in this case, there are commercial Ada compilers (e.g. with the MIPS/PRO toolchain)
> 
> let me know

http://llvm.org/Features.html
https://github.com/llvm-mirror/llvm/blob/master/lib/Demangle/ItaniumDemangle.cpp

Independent of Ada on LLVM (which is yearned  for occasionally here at comp.lang.ada), LLVM.org lists MIPS as a supported architecture.  Also, Itanium/IA64 has been supported in the past in LLVM, so using HP's PA-RISC emulation on Itanium is the most likely way of supporting HP's PA ISA, perhaps via an older Itanium-supporting release of LLVM.

Re: LLVM and Ada

[Bojan Bozovic]

On Wednesday, March 21, 2018 at 3:41:01 PM UTC+1, Dan'l Miller wrote:
> On Wednesday, March 21, 2018 at 8:08:50 AM UTC-5, carloj...@gmail.com wrote:
> > hi guys
> > I need to port ghdl to PPC, MIPS, HPPA
> > on PPC there are old bootstrapper (e.g. 4.1.1)
> > but I haven't seen them for MIPS and HPPA
> > 
> > vhdl seems able to be compiled by llvm, but ... llvm seems do have no support for MIPS and HPPA
> > 
> > no dice
> > 
> > also, I'd like to port vhdl to SGI/MIPS under IRIX
> > in this case, there are commercial Ada compilers (e.g. with the MIPS/PRO toolchain)
> > 
> > let me know
> 
> http://llvm.org/Features.html
> https://github.com/llvm-mirror/llvm/blob/master/lib/Demangle/ItaniumDemangle.cpp
> 
> Independent of Ada on LLVM (which is yearned  for occasionally here at comp.lang.ada), LLVM.org lists MIPS as a supported architecture.  Also, Itanium/IA64 has been supported in the past in LLVM, so using HP's PA-RISC emulation on Itanium is the most likely way of supporting HP's PA ISA, perhaps via an older Itanium-supporting release of LLVM.

I haven't tested to build it, nor I have intention, DragonEgg exists for LLVM 3-6 and GCC 6.3/6.4/8.0

https://github.com/xiangzhai/dragonegg is a fork more up to date than main branch.

Re: LLVM and Ada

[Luke A. Guest]

> 
> I haven't tested to build it, nor I have intention, DragonEgg exists for
> LLVM 3-6 and GCC 6.3/6.4/8.0
> 
> https://github.com/xiangzhai/dragonegg is a fork more up to date than main branch.
> 

There’s still not much been done even on that branch.

Re: LLVM and Ada

[Brian Drummond]

On Wed, 21 Mar 2018 06:08:44 -0700, carlojpisani wrote:

> hi guys I need to port ghdl to PPC, MIPS, HPPA on PPC there are old
> bootstrapper (e.g. 4.1.1)
> but I haven't seen them for MIPS and HPPA
> 
> vhdl seems able to be compiled by llvm, but ... llvm seems do have no
> support for MIPS and HPPA
> 
> no dice
> 
> also, I'd like to port vhdl to SGI/MIPS under IRIX in this case, there
> are commercial Ada compilers (e.g. with the MIPS/PRO toolchain)
> 
> let me know

The place for this would be 
https://github.com/ghdl/ghdl

If you run into trouble, open an issue there, Tristan (the main author of 
ghdl) is very helpful.

You need to *compile* ghdl using gcc (including gnat) so the easiest way 
would be if you have native (and fairly recent, >= gcc4.7) installations 
on your desired platforms.

Otherwise, gcc can obviously be used to build cross-compilers, though 
that gets a bit trickier, and while building ghdl sa a crosscompiler 
ought to work, there may be some snags along the way.

GHDL cannot be *compiled* with LLVM because there is no adequate Ada 
compiler in the LLVM toolchain. However, ghdl can be built to target 3 
different code generation backends - gcc, llvm and its own JIT compiler 
(mcode)

(To make ghdl/llvm work, Tristan had to solve LLVM's lack of support for 
nested procedures (and their scope rules) which I believe was the problem 
that stopped the Dragonegg Ada/LLVM compiler in its tracks. So an Ada/
LLVM compiler should be possible...)

So I *think* you want to build ghdl for the gcc backend (to support your 
target CPUs). Whether you can do that natively or you need to cross-
compile depends on having a modern enough gcc compiler on your system.

-- Brian

Re: LLVM and Ada

[Dan'l Miller]

On Wednesday, March 21, 2018 at 1:57:48 PM UTC-5, Luke A. Guest wrote:
> > 
> > I haven't tested to build it, nor I have intention, DragonEgg exists for
> > LLVM 3-6 and GCC 6.3/6.4/8.0
> > 
> > https://github.com/xiangzhai/dragonegg is a fork more up to date than main branch.
> > 
> 
> There’s still not much been done even on that branch.

https://github.com/xiangzhai

The guy who took this work over in 2017, Xiang Leslie ZHAI, has a llvm.org.cn email address.  I don't know the criteria needed to be that degree of officially connected with LLVM, but his work appears to have some degree of officialness from LLVM organization itself.  That is a good sign.

This fork of DragonEgg supports various releases of GCC (except conspicuously skipping 7.X and going on to 8.0), probably focusing on C and C++, ignoring Ada.  But (assuming that it works), it is good to have any amount of maintenance on the old DragonEgg to keep it at all current (even partially).  It is easier for someone among us in Ada world to test & maintain the Ada support on this fork of DragonEgg then, instead of going back 7 years ourselves.

No contributions since October 2017 though.  That is a bad sign.

Re: LLVM and Ada

[Luke A. Guest]

Dan'l Miller <optikos@verizon.net> wrote:
>

> No contributions since October 2017 though.  That is a bad sign.
> 


Like I said, he hasn’t done much. He got the thing to compile with a more
recent gcc, I’ve built it with gcc-6.x. Afaik though it won’t do much.

Re: LLVM and Ada

[Dan'l Miller]

On Thursday, March 22, 2018 at 10:18:10 AM UTC-5, Luke A. Guest wrote:
> Dan'l Miller wrote:
> >
> 
> > No contributions since October 2017 though.  That is a bad sign.
> > 
> 
> 
> Like I said, he hasn’t done much. He got the thing to compile with a more
> recent gcc, I’ve built it with gcc-6.x.

But did you contribute your work back to humanity, as he did?  Your work might have predated his and been useful to him.  Indeed, even if your edits are chronologically after his, your changes might demonstrate superior wisdom & insight & expertise that he would find useful.  An LLVM backend to GNAT won't happen unless all the wood is behind one arrow (as Scott McNealy used to say).

Often comp.lang.ada operates as (trolls? who are) raining on each other's parade.  I would like to see a greater community of positive forward movement, where like-minded bits of cellulose self-assemble as all the wood behind one arrow (or at least behind each disparate arrow of forward progress in an entire quiver of differently-focused arrows).  I am afraid that Ada is incrementally fading for no good reason (other than a preponderance of its programmers [and perhaps even its luminaries] having grey hair and retiring into obscurity).  •Every• sector of the Ada community needs to benefit from ever-increasing momentum in all the other sectors of the Ada community.  An LLVM-backended GNAT would be astronomically beneficial to the Ada community because suddenly Ada would be able to quite-publicly-visibly target iPhone & iPad & iPod Touch & Apple Watch, which is the most-conspicuous missing piece in advocating Ada as a state-of-the-art general-purpose language competing with C++ and Swift and Java across all modern OSes & OS-frameworks thereof.

(Btw, I always speak of iOS as an OS separate from Cocoa Touch framework and MacOS as an OS separate from its 3 OS-frameworks [Cocoa, Carbon, and XQuartz] and Windows as an OS separate from its 4 (?) OS-frameworks & APIs [Win32 old-school C-header-defined API, Microsoft .NET bytecode DLLs with WinMD, Xamarin Mono bytecode DLLs without WinMD, WinRT/UWP components processor-native DLLs with WinMD.]

Re: LLVM and Ada

[Luke A. Guest]

Dan'l Miller <optikos@verizon.net> wrote:
> On Thursday, March 22, 2018 at 10:18:10 AM UTC-5, Luke A. Guest wrote:
>> Dan'l Miller wrote:
>>> 
>> 
>>> No contributions since October 2017 though.  That is a bad sign.
>>> 
>> 
>> 
>> Like I said, he hasn’t done much. He got the thing to compile with a more
>> recent gcc, I’ve built it with gcc-6.x.
> 
> But did you contribute your work back to humanity, as he did? 

There was no work, I got his repo compiling with his help. See the issues.

Re: LLVM and Ada

[Dan'l Miller]

On Thursday, March 22, 2018 at 4:14:03 PM UTC-5, Luke A. Guest wrote:
> There was no work, I got his repo compiling with his help. See the issues.

Ah, this one:

https://github.com/xiangzhai/dragonegg/issues/33

Re: LLVM and Ada

[Dan'l Miller]

On Thursday, March 22, 2018 at 4:28:50 PM UTC-5, Dan'l Miller wrote:
> On Thursday, March 22, 2018 at 4:14:03 PM UTC-5, Luke A. Guest wrote:
> > There was no work, I got his repo compiling with his help. See the issues.
> 
> Ah, this one:
> 
> https://github.com/xiangzhai/dragonegg/issues/33

There, I submitted an issue to explore the degree of intended on-going activity on his fork.

https://github.com/xiangzhai/dragonegg/issues/34

Plus, it would do him good for himself to track via the regression test-suites just how broken or not DragonEgg is.

Re: LLVM and Ada

[Luke A. Guest]

Dan'l Miller <optikos@verizon.net> wrote:
> On Thursday, March 22, 2018 at 4:28:50 PM UTC-5, Dan'l Miller wrote:
>> On Thursday, March 22, 2018 at 4:14:03 PM UTC-5, Luke A. Guest wrote:
>>> There was no work, I got his repo compiling with his help. See the issues.
>> 
>> Ah, this one:
>> 
>> https://github.com/xiangzhai/dragonegg/issues/33
> 
> There, I submitted an issue to explore the degree of intended on-going
> activity on his fork.
> 
> https://github.com/xiangzhai/dragonegg/issues/34
> 
> Plus, it would do him good for himself to track via the regression
> test-suites just how broken or not DragonEgg is.
> 

LLVM has a fortran compiler.

Re: LLVM and Ada

[Dan'l Miller]

> LLVM has a fortran compiler.

Yeah, right.  Flang is less mature as a usable Fortran than Hurd is as a GNU kernel.  Flang can’t even pass a generally-accepted integer-exponentiation regression test:

https://github.com/flang-compiler/flang/issues/180

https://github.com/flang-compiler/flang/issues/239
“Tweaking the examples slightly can go from it compiling without error, compiling with an error, or dying with an internal error.”

That is still a diamond in the rough with the emphasis on rough—to put it euphemistically.  I would say that Flang is still ‘an announced future Fortran compiler for LLVM.’

Re: LLVM and Ada

[Lucretia]

On Thursday, 22 March 2018 23:37:57 UTC, Dan'l Miller  wrote:
> > LLVM has a fortran compiler.
> 
> Yeah, right.  Flang is less mature as a usable Fortran than Hurd is as a GNU kernel.  Flang can’t even pass a generally-accepted integer-exponentiation regression test:

You said there was no Fortran compiler, I just pointed one out, I didn't say it was any good.

Re: LLVM and Ada

[Dan'l Miller]

Luke Guest wrote: 
> You said there was no Fortran compiler, I just
> pointed one out, I didn't say it was any good.

Vaporware is not extant software.  Imaginary friends are not friends.  Time travel to the future is not living in the present.  The point is GNU Fortran on LLVM over in that quasimaintained DragonEgg fork could conceivably mature to stable & usable on all modern OSes & ISAs quicker than writing Flang from (near-)scratch (ignoring the gutting of Clang that served as the basis of Flang).

Re: LLVM and Ada

[Bojan Bozovic]

Maybe this isn't really a place for this question, but why so important thing like Fortran and Ada for LLVM fell on the shoulders of a single developer, why there's no involvement both from LLVM and GCC community?

Re: LLVM and Ada

[Dan'l Miller]

On Friday, March 23, 2018 at 12:10:21 AM UTC-5, Bojan Bozovic wrote:
> Maybe this isn't really a place for this question, but why so important thing like Fortran and Ada for LLVM
> fell on the shoulders of a single developer, why there's no involvement both from LLVM and GCC
> community?

Well, it is not exactly a single developer acting as an Ayn-Rand/Aristotlean Prime Mover lone-wolf.

https://www.llnl.gov/news/nnsa-national-labs-team-nvidia-develop-open-source-fortran-compiler-technology

  It appears that IBM & NVIDIA are stoking the fires at government research organizations that buy their exascale OpenPOWER equipment that is based on POWER ISA cores for general-purpose computing and NVIDIA GPU cores for numerical processing.

The NSF & Naval Research Office & the national labs have all been issuing grant money for various LLVM projects for some time.  For example, ALLVM's “LLVM All the Things” mantra at UIUC who originated LLVM, especially its People & Funding tab:

https://publish.illinois.edu/allvm-project

Note that official Flang (as derived from NVIDIA's PGI Fortran compiler for GPUs) views itself as separate from “community” efforts to bring Fortran to LLVM, even though those “community” efforts also use the Flang name:

https://github.com/flang-compiler/flang/issues/38

Indeed, all the implied smearing of GCC's Fortran (e.g., Flang will be the first “fully open-source” Fortran compiler; gee, I think Richard Stallman might want to say a few words about that) seems a tad disingenuous.  What really seems to be going on with Flang is that major Fortran userbases associated with the federal government think that evolving a GPU-centric Fortran compiler at NVIDIA to generate code for general-purpose-processor ISAs and be then ported to all OSes is easier than adding immense GPU support to GCC's Fortran & GCC backend.  And IBM & NVIDIA seek hardware revenue by selling exascale supercomputers to those userbases.  And then NSF & Naval Research Office grant money starts being emitted to suddenly create a funded NVIDIA-PGI-based Flang community that eclipses other Flang-on-LLVM communities.

If only Ada today had such a pull & push marketing scheme to create its own river of moneyflow!

Re: LLVM and Ada

[Dan'l Miller]

On Friday, March 23, 2018 at 9:25:16 AM UTC-5, Dan'l Miller wrote:
> ... For example, ALLVM's “LLVM All the Things” mantra at UIUC who originated LLVM, especially its People & Funding tab:
> 
> https://publish.illinois.edu/allvm-project

Oh, and why isn't Ada viewed as one of the Things in the “LLVM All the Things” mantra?  Someone needs to hook the Ada railcar up to that locomotive before it completely leaves the train-station.

Re: LLVM and Ada

[Shark8]

On Friday, March 23, 2018 at 8:38:11 AM UTC-6, Dan'l Miller wrote:
> 
> Oh, and why isn't Ada viewed as one of the Things in the “LLVM All the Things” mantra?  Someone needs to hook the Ada railcar up to that locomotive before it completely leaves the train-station.

Insofar as Byron is concerned:
(a) I intend for the backend to be completely modularizable, and therefore am not married to any particular code-generator.
(b) Nobody who wanted LLVM codegen has contributed to the code-base.

Re: LLVM and Ada

[Lucretia]

On Friday, 23 March 2018 05:10:21 UTC, Bojan Bozovic  wrote:
> Maybe this isn't really a place for this question, but why so important thing like Fortran and Ada for LLVM fell on the shoulders of a single developer, why there's no involvement both from LLVM and GCC community?

Stallman's stance is that GCC should never be available as a library that can be linked into an executable, but this is happening anyway with GCC libjit.

All the company's using LLVM only really care about C-based languages, so they don't really care about Ada. It's nVidia who have worked on Fortran for LLVM as Dan'l noted above.

Re: LLVM and Ada

[Dan'l Miller]

On Friday, March 23, 2018 at 11:38:03 AM UTC-5, Shark8 wrote:
> Insofar as Byron is concerned:
> (a) I intend for the backend to be completely modularizable, and therefore am not married to any particular code-generator.
> (b) Nobody who wanted LLVM codegen has contributed to the code-base.

For multiple people or sponsoring companies (or even federal grant-writing agencies) to get involved in your work, you would need to have some sweet-smelling pollen to attract bees to your flower, so to speak.  If pluggability of different backends is the primary/sole definition of modularity in Byron, there seems currently to exist only 3 pluggable open-source backends that could conceivably serve as Byron's backend processor machine-code generation (i.e., GCC, LLVM, and OpenWatcom) and only a few bytecode backends (e.g., CLR, JVM).  Of these 5, only one seems compelling:  LLVM.

If you seek additional developers or sponsoring companies, then Byron needs to have a special sauce, because LLVM backend for GNAT seems to be a matter of an immense amount of debugging & fixing those bugs.  So why develop a new Ada compiler from scratch when the primary competitor simply needs to be finished?  Answer:  because Byron would have other modularities:  for example,
1) plug in different schools of thought on semantic-correctness checks:  Don't like Ada2020's “conservative“ (Draconian?) regime of tasking correctness?  Then plug in Ada2012's instead.  Or permit development of custom semantic-correctness checks on a per-company (e.g., DoD contractor) or per-project (e.g., this model of nuclear submarine) basis.  Or plug in Spark's.
2) Make language features themselves modular.  If Dmitry wants coroutines, then Dmitry simply writes a DLL that plugs into Byron that defines the syntax of coroutines according to Byron's syntax API or AST and that defines the semantics of coroutines according to Byron's semantics API.
3) Make the stages of compilation modular:  write a multi-stage compiler that embraces •generating• Ada source code that then smoothly compiles generated & human-authored Ada source code.
4) Make the AST modular to be bolted onto Clang's AST (to permit automated transformation of C++ or Objective-C code to Ada), onto Roslyn's AST (to permit automated transformation of C# code to Ada), onto Java's AST (to permit automated transformation of Java code to Ada), and so forth.

Or in other words, you need a better elevator speech.  You are almost there; pick a modularity problem that is interesting to you to solve in addition to pluggable backend.

Re: LLVM and Ada

[Bojan Bozovic]

On Friday, March 23, 2018 at 5:38:03 PM UTC+1, Shark8 wrote:
> On Friday, March 23, 2018 at 8:38:11 AM UTC-6, Dan'l Miller wrote:
> > 
> > Oh, and why isn't Ada viewed as one of the Things in the “LLVM All the Things” mantra?  Someone needs to hook the Ada railcar up to that locomotive before it completely leaves the train-station.
> 
> Insofar as Byron is concerned:
> (a) I intend for the backend to be completely modularizable, and therefore am not married to any particular code-generator.
> (b) Nobody who wanted LLVM codegen has contributed to the code-base.

Look at Coco/R compiler compiler http://ssw.jku.at/Coco/ there is Ada version as well, which is under GPL but makes syntax and semantic checker, and could be even compiled with AdaCore GNAT and proved with SPARK, and still produce non-GPL compiler. Consider using already existing runtime from FSF GNAT modified for your compiler, and using LLVM optimizers. Ideally aim for something like WATCOM/OpenWATCOM C/C++ where you have few host OS platforms but many libraries for run-time and code generators instead using GCC way of building cross-compiler. Making Byron compiler will be many many man-years work, so using existing tools will make you actually make progress and eventually finish it. My $0.02 useful or not.

Re: LLVM and Ada

[Bojan Bozovic]

On Sunday, March 25, 2018 at 12:29:48 PM UTC+2, Bojan Bozovic wrote:
> On Friday, March 23, 2018 at 5:38:03 PM UTC+1, Shark8 wrote:
> > On Friday, March 23, 2018 at 8:38:11 AM UTC-6, Dan'l Miller wrote:
> > > 
> > > Oh, and why isn't Ada viewed as one of the Things in the “LLVM All the Things” mantra?  Someone needs to hook the Ada railcar up to that locomotive before it completely leaves the train-station.
> > 
> > Insofar as Byron is concerned:
> > (a) I intend for the backend to be completely modularizable, and therefore am not married to any particular code-generator.
> > (b) Nobody who wanted LLVM codegen has contributed to the code-base.
> 
> Look at Coco/R compiler compiler http://ssw.jku.at/Coco/ there is Ada version as well, which is under GPL but makes syntax and semantic checker, and could be even compiled with AdaCore GNAT and proved with SPARK, and still produce non-GPL compiler. Consider using already existing runtime from FSF GNAT modified for your compiler, and using LLVM optimizers. Ideally aim for something like WATCOM/OpenWATCOM C/C++ where you have few host OS platforms but many libraries for run-time and code generators instead using GCC way of building cross-compiler. Making Byron compiler will be many many man-years work, so using existing tools will make you actually make progress and eventually finish it. My $0.02 useful or not.

Maybe it would be better to use permissively licensed compiler, FSF GNAT, instead. I'm really not sure what will be linked to Coco/R generated compiler, so if there's GPL stuff there would be a problem. Please excuse me for this duplicate message, I'm trying to be helpful even though I don't know much.

Re: LLVM and Ada

[Lucretia]

On Sunday, 25 March 2018 03:17:26 UTC+1, Dan'l Miller  wrote:

> For multiple people or sponsoring companies (or even federal grant-writing agencies) to get involved in your work, you would need to have some sweet-smelling pollen to attract bees to your flower, so to speak.  If pluggability of different backends is the primary/sole definition of modularity in Byron, there seems currently to exist only 3 pluggable open-source backends that could conceivably serve as Byron's backend processor machine-code generation (i.e., GCC, LLVM, and OpenWatcom) and only a few bytecode backends (e.g., CLR, JVM).  Of these 5, only one seems compelling:  LLVM.

There is also libFirm

** Wall of text removed **

Re: LLVM and Ada

[Lucretia]

On Sunday, 25 March 2018 11:29:48 UTC+1, Bojan Bozovic  wrote:

> Look at Coco/R compiler compiler http://ssw.jku.at/Coco/ there is Ada version 

I looked at this years ago and the version hasn't changed, it has severe limitations, to quote the readme:

At this time it is not possible to use Ada attributes into production code.
It is because any Ada attribute starts from character ''', and it is interferenced
with string declaration into Coco/R attributed grammar.

Re: LLVM and Ada

[Dan'l Miller]

On Thursday, March 22, 2018 at 4:36:32 PM UTC-5, Dan'l Miller wrote:
> On Thursday, March 22, 2018 at 4:28:50 PM UTC-5, Dan'l Miller wrote:
> > On Thursday, March 22, 2018 at 4:14:03 PM UTC-5, Luke A. Guest wrote:
> > > There was no work, I got his repo compiling with his help. See the issues.
> > 
> > Ah, this one:
> > 
> > https://github.com/xiangzhai/dragonegg/issues/33
> 
> There, I submitted an issue to explore the degree of intended on-going activity on his fork.
> 
> https://github.com/xiangzhai/dragonegg/issues/34
> 
> Plus, it would do him good for himself to track via the regression test-suites just how broken or not
> DragonEgg is.

Xiang Leslie ZHAI replied.  His reply appears to confirm that this maintained DragonEgg builds and somehow transfers GIMPLE to LLVM intermediate representation (to what fidelity?) but does not yet generate machine code in LLVM from that IR:

https://github.com/xiangzhai/dragonegg/issues/34#issuecomment-377529336

“It has to fix #31 before regression test, dragonegg for gcc-8 and llvm-7 is just able to transfer GIMPLE -> LLVM IR, but failed to CodeGen https://www.zhihu.com/question/265150161

Regards,
Leslie Zhai”