Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.lang.c > #142334 > unrolled thread

clang

Back to article view | Back to comp.lang.c

Contents

  clang jacobnavia <jacob@jacob.remcomp.fr> - 2019-03-24 12:15 +0100
    Re: clang Melzzzzz <Melzzzzz@zzzzz.com> - 2019-03-24 12:56 +0000
      Re: clang jacobnavia <jacob@jacob.remcomp.fr> - 2019-03-24 14:11 +0100
        Re: clang Melzzzzz <Melzzzzz@zzzzz.com> - 2019-03-24 13:24 +0000
          Re: clang David Brown <david.brown@hesbynett.no> - 2019-03-24 15:47 +0100
        Re: clang Bart <bc@freeuk.com> - 2019-03-24 13:42 +0000
          Re: clang Noob <root@127.0.0.1> - 2019-04-10 21:07 +0200
    Re: clang already5chosen@yahoo.com - 2019-03-24 06:24 -0700
      Re: clang Melzzzzz <Melzzzzz@zzzzz.com> - 2019-03-24 13:38 +0000
    Re: clang jacobnavia <jacob@jacob.remcomp.fr> - 2019-03-26 21:06 +0100
      Re: clang Melzzzzz <Melzzzzz@zzzzz.com> - 2019-03-26 20:48 +0000
        Re: clang jacobnavia <jacob@jacob.remcomp.fr> - 2019-03-26 21:50 +0100
          Re: clang Ian Collins <ian-news@hotmail.com> - 2019-03-27 09:52 +1300
            Re: clang jacobnavia <jacob@jacob.remcomp.fr> - 2019-03-26 21:58 +0100
              Re: clang already5chosen@yahoo.com - 2019-03-27 06:40 -0700
                Re: clang jacobnavia <jacob@jacob.remcomp.fr> - 2019-03-27 17:31 +0100
            Re: clang already5chosen@yahoo.com - 2019-03-27 06:29 -0700
              Re: clang Ben Bacarisse <ben.usenet@bsb.me.uk> - 2019-03-27 14:57 +0000
                Re: clang already5chosen@yahoo.com - 2019-03-27 08:16 -0700
                  Re: clang scott@slp53.sl.home (Scott Lurndal) - 2019-03-27 15:47 +0000
                  Re: clang Ben Bacarisse <ben.usenet@bsb.me.uk> - 2019-03-27 18:29 +0000
                    Re: clang already5chosen@yahoo.com - 2019-03-27 11:45 -0700
                      Re: clang Ben Bacarisse <ben.usenet@bsb.me.uk> - 2019-03-27 21:37 +0000
                        Re: clang already5chosen@yahoo.com - 2019-03-28 05:50 -0700
                        Re: clang already5chosen@yahoo.com - 2019-03-28 06:01 -0700
                          Re: clang scott@slp53.sl.home (Scott Lurndal) - 2019-03-28 15:21 +0000
                  Re: clang Ian Collins <ian-news@hotmail.com> - 2019-03-28 07:52 +1300
                  Re: clang David Brown <david.brown@hesbynett.no> - 2019-03-27 22:45 +0100
                    Re: clang jacobnavia <jacob@jacob.remcomp.fr> - 2019-03-28 00:12 +0100
        Re: clang scott@slp53.sl.home (Scott Lurndal) - 2019-03-26 20:55 +0000
          Re: clang jacobnavia <jacob@jacob.remcomp.fr> - 2019-03-26 22:05 +0100
            Re: clang Melzzzzz <Melzzzzz@zzzzz.com> - 2019-03-26 21:59 +0000
        Re: clang Bart <bc@freeuk.com> - 2019-03-26 21:42 +0000
          Re: clang Melzzzzz <Melzzzzz@zzzzz.com> - 2019-03-26 22:00 +0000
            Re: clang Bart <bc@freeuk.com> - 2019-03-27 14:32 +0000
      Re: clang already5chosen@yahoo.com - 2019-03-27 06:54 -0700
    Re: clang: release build jacobnavia <jacob@jacob.remcomp.fr> - 2019-03-28 23:31 +0100
    Re: clang Mehdi Amini <atorrses@gmail.com> - 2019-04-07 19:38 +0430

Page 2 of 2 — ← Prev page 1 [2]

#142521

already5chosen@yahoo.com writes:

> On Wednesday, March 27, 2019 at 4:58:01 PM UTC+2, Ben Bacarisse wrote:
>> already5chosen@yahoo.com writes:
>> <of clang>
>> > Looks pretty messy and amateurish when one considers that we are
>> > talking about best funded compiler development project in the history
>> > of digital computing,
>> 
>> Do you have a citation for that?  It would be interesting to see a few
>> compiler projects compared by funding and effort.
>
> No I have no citation.
> But it's common knowledge that LLVM is funded by Apple, which is by
> far the richest computer company in the history

That is indeed common knowledge.  It may even be true.  However it's not
what you said.  I found what you said interesting, and was hoping for
some facts.  How much (in modern terms) did IBM spend on its FORTRAN
compiler?  What about ICL's effort to implement Algol 86?  How do these
compare with clang funding?  Anyway, without data there is not much more
to say.

-- 
Ben.

[toc] | [prev] | [next] | [standalone]

#142522

On Wednesday, March 27, 2019 at 8:29:54 PM UTC+2, Ben Bacarisse wrote:
> already5chosen@yahoo.com writes:
> 
> > On Wednesday, March 27, 2019 at 4:58:01 PM UTC+2, Ben Bacarisse wrote:
> >> already5chosen@yahoo.com writes:
> >> <of clang>
> >> > Looks pretty messy and amateurish when one considers that we are
> >> > talking about best funded compiler development project in the history
> >> > of digital computing,
> >> 
> >> Do you have a citation for that?  It would be interesting to see a few
> >> compiler projects compared by funding and effort.
> >
> > No I have no citation.
> > But it's common knowledge that LLVM is funded by Apple, which is by
> > far the richest computer company in the history
> 
> That is indeed common knowledge.  It may even be true.  However it's not
> what you said.  I found what you said interesting, and was hoping for
> some facts.  How much (in modern terms) did IBM spend on its FORTRAN
> compiler?  What about ICL's effort to implement Algol 86?  How do these
> compare with clang funding?  Anyway, without data there is not much more
> to say.
> 
> -- 
> Ben.

Do you mean Algol-68?
One of my co-workers certainly does not know how much they spent, but there is chance that he remembers how many people worked on compiler.
If I ask him tomorrow, would it help?

However I would think that most expensive compiler development effort of these years was IBM's PL-1. Unfortunately, non of my co-workers was on the team or near.

[toc] | [prev] | [next] | [standalone]

#142529

already5chosen@yahoo.com writes:

> On Wednesday, March 27, 2019 at 8:29:54 PM UTC+2, Ben Bacarisse wrote:
>> already5chosen@yahoo.com writes:
>> 
>> > On Wednesday, March 27, 2019 at 4:58:01 PM UTC+2, Ben Bacarisse wrote:
>> >> already5chosen@yahoo.com writes:
>> >> <of clang>
>> >> > Looks pretty messy and amateurish when one considers that we are
>> >> > talking about best funded compiler development project in the history
>> >> > of digital computing,
>> >> 
>> >> Do you have a citation for that?  It would be interesting to see a few
>> >> compiler projects compared by funding and effort.
>> >
>> > No I have no citation.
>> > But it's common knowledge that LLVM is funded by Apple, which is by
>> > far the richest computer company in the history
>> 
>> That is indeed common knowledge.  It may even be true.  However it's not
>> what you said.  I found what you said interesting, and was hoping for
>> some facts.  How much (in modern terms) did IBM spend on its FORTRAN
>> compiler?  What about ICL's effort to implement Algol 86?  How do these
>> compare with clang funding?  Anyway, without data there is not much more
>> to say.
>
> Do you mean Algol-68?

Yes, thank.  (But I don't think the name is hyphenated.)  What a shame,
though, that we don't have Algol 86!

> One of my co-workers certainly does not know how much they spent, but
> there is chance that he remembers how many people worked on compiler.
> If I ask him tomorrow, would it help?

Yes, but it's not going to be much to go on.  I used to work for ICL,
and I remember the compiler was considered a big deal, but I have no
recollection of numbers.

> However I would think that most expensive compiler development effort
> of these years was IBM's PL-1. Unfortunately, non of my co-workers was
> on the team or near.

And the aborted CPL work was probably a big sink of resources, too.

-- 
Ben.

[toc] | [prev] | [next] | [standalone]

#142540

On Wednesday, March 27, 2019 at 11:38:06 PM UTC+2, Ben Bacarisse wrote:
> already5chosen@yahoo.com writes:
> 
> > On Wednesday, March 27, 2019 at 8:29:54 PM UTC+2, Ben Bacarisse wrote:
> >> already5chosen@yahoo.com writes:
> >> 
> >> > On Wednesday, March 27, 2019 at 4:58:01 PM UTC+2, Ben Bacarisse wrote:
> >> >> already5chosen@yahoo.com writes:
> >> >> <of clang>
> >> >> > Looks pretty messy and amateurish when one considers that we are
> >> >> > talking about best funded compiler development project in the history
> >> >> > of digital computing,
> >> >> 
> >> >> Do you have a citation for that?  It would be interesting to see a few
> >> >> compiler projects compared by funding and effort.
> >> >
> >> > No I have no citation.
> >> > But it's common knowledge that LLVM is funded by Apple, which is by
> >> > far the richest computer company in the history
> >> 
> >> That is indeed common knowledge.  It may even be true.  However it's not
> >> what you said.  I found what you said interesting, and was hoping for
> >> some facts.  How much (in modern terms) did IBM spend on its FORTRAN
> >> compiler?  What about ICL's effort to implement Algol 86?  How do these
> >> compare with clang funding?  Anyway, without data there is not much more
> >> to say.
> >
> > Do you mean Algol-68?
> 
> Yes, thank.  (But I don't think the name is hyphenated.)  What a shame,
> though, that we don't have Algol 86!
> 
> > One of my co-workers certainly does not know how much they spent, but
> > there is chance that he remembers how many people worked on compiler.
> > If I ask him tomorrow, would it help?
> 
> Yes, but it's not going to be much to go on.  I used to work for ICL,
> and I remember the compiler was considered a big deal, but I have no
> recollection of numbers.
> 
> > However I would think that most expensive compiler development effort
> > of these years was IBM's PL-1. Unfortunately, non of my co-workers was
> > on the team or near.
> 
> And the aborted CPL work was probably a big sink of resources, too.
> 
> -- 
> Ben.

It turned out that co-worker does not know about size Algol 68 team.
He worked on macroassembler and was familiar with the group that developed Fortran compiler (back then still FORTRAN), but not those of Algol.
He thinks that Algol 68 group was not bigger than 10 persons, but that's just a feeling, not the knowledge.

[toc] | [prev] | [next] | [standalone]

#142541

On Wednesday, March 27, 2019 at 11:38:06 PM UTC+2, Ben Bacarisse wrote:
> already5chosen@yahoo.com writes:
> 
> > On Wednesday, March 27, 2019 at 8:29:54 PM UTC+2, Ben Bacarisse wrote:
> >> already5chosen@yahoo.com writes:
> >> 
> >> > On Wednesday, March 27, 2019 at 4:58:01 PM UTC+2, Ben Bacarisse wrote:
> >> >> already5chosen@yahoo.com writes:
> >> >> <of clang>
> >> >> > Looks pretty messy and amateurish when one considers that we are
> >> >> > talking about best funded compiler development project in the history
> >> >> > of digital computing,
> >> >> 
> >> >> Do you have a citation for that?  It would be interesting to see a few
> >> >> compiler projects compared by funding and effort.
> >> >
> >> > No I have no citation.
> >> > But it's common knowledge that LLVM is funded by Apple, which is by
> >> > far the richest computer company in the history
> >> 
> >> That is indeed common knowledge.  It may even be true.  However it's not
> >> what you said.  I found what you said interesting, and was hoping for
> >> some facts.  How much (in modern terms) did IBM spend on its FORTRAN
> >> compiler?  What about ICL's effort to implement Algol 86?  How do these
> >> compare with clang funding?  Anyway, without data there is not much more
> >> to say.
> >
> > Do you mean Algol-68?
> 
> Yes, thank.  (But I don't think the name is hyphenated.)  What a shame,
> though, that we don't have Algol 86!
> 

I don't know either Algol 60 or Algol 68, but I read that two languages had very little in common except of basic syntax.
So it's quite possible that if Algol 86 was ever invented, it again wouldn't be any more similar to Algol 68 than any of the languages of the same era that had algolish syntax. And there were many of later in the 1980s.

[toc] | [prev] | [next] | [standalone]

#142546

already5chosen@yahoo.com writes:
>On Wednesday, March 27, 2019 at 11:38:06 PM UTC+2, Ben Bacarisse wrote:
>> already5chosen@yahoo.com writes:
>> 
>> > On Wednesday, March 27, 2019 at 8:29:54 PM UTC+2, Ben Bacarisse wrote:
>> >> already5chosen@yahoo.com writes:
>> >> 
>> >> > On Wednesday, March 27, 2019 at 4:58:01 PM UTC+2, Ben Bacarisse wrote:
>> >> >> already5chosen@yahoo.com writes:
>> >> >> <of clang>
>> >> >> > Looks pretty messy and amateurish when one considers that we are
>> >> >> > talking about best funded compiler development project in the history
>> >> >> > of digital computing,
>> >> >> 
>> >> >> Do you have a citation for that?  It would be interesting to see a few
>> >> >> compiler projects compared by funding and effort.
>> >> >
>> >> > No I have no citation.
>> >> > But it's common knowledge that LLVM is funded by Apple, which is by
>> >> > far the richest computer company in the history
>> >> 
>> >> That is indeed common knowledge.  It may even be true.  However it's not
>> >> what you said.  I found what you said interesting, and was hoping for
>> >> some facts.  How much (in modern terms) did IBM spend on its FORTRAN
>> >> compiler?  What about ICL's effort to implement Algol 86?  How do these
>> >> compare with clang funding?  Anyway, without data there is not much more
>> >> to say.
>> >
>> > Do you mean Algol-68?
>> 
>> Yes, thank.  (But I don't think the name is hyphenated.)  What a shame,
>> though, that we don't have Algol 86!
>> 
>
>I don't know either Algol 60 or Algol 68, but I read that two languages had very little in common except of basic syntax.
>So it's quite possible that if Algol 86 was ever invented, it again wouldn't be any more similar to Algol 68 than any of the languages of the same era that had algolish syntax. And there were many of later in the 1980s.
>

There are several variants of algol still in use on the Unisys Clearpath
systems;  one to write the operating system in, one for data comm applications
and the vanilla algol for app programming.

[toc] | [prev] | [next] | [standalone]

#142523

On 28/03/2019 04:16, already5chosen@yahoo.com wrote:
> On Wednesday, March 27, 2019 at 4:58:01 PM UTC+2, Ben Bacarisse
> wrote:
>> already5chosen@yahoo.com writes: <of clang>
>>> Looks pretty messy and amateurish when one considers that we are 
>>> talking about best funded compiler development project in the
>>> history of digital computing,
>> 
>> Do you have a citation for that?  It would be interesting to see a
>> few compiler projects compared by funding and effort.
>> 
>> -- Ben.
> 
> No I have no citation. But it's common knowledge that LLVM is funded
> by Apple, which is by far the richest computer company in the history
> and that ALOT of people are working on it. Google, which is not
> exactly poor company either, also paying salaries to  many LLVM
> developers.

While, like gcc, there is corporate funding and paid employees working 
on the project, LLVM is still (again, like gcc) a volunteer effort.

-- 
Ian.

[toc] | [prev] | [next] | [standalone]

#142531

On 27/03/2019 16:16, already5chosen@yahoo.com wrote:
> On Wednesday, March 27, 2019 at 4:58:01 PM UTC+2, Ben Bacarisse
> wrote:
>> already5chosen@yahoo.com writes: <of clang>
>>> Looks pretty messy and amateurish when one considers that we are 
>>> talking about best funded compiler development project in the
>>> history of digital computing,
>> 
>> Do you have a citation for that?  It would be interesting to see a
>> few compiler projects compared by funding and effort.
>> 
>> -- Ben.
> 
> No I have no citation. But it's common knowledge that LLVM is funded
> by Apple, which is by far the richest computer company in the history
> and that ALOT of people are working on it. Google, which is not
> exactly poor company either, also paying salaries to  many LLVM
> developers. 

You can be very sure that neither of these companies pay for more 
salaries on LLVM developers than they can get away with - there is a 
reason they are rich companies, and it is not because they pay for vast 
numbers of compiler developers.

> I would think that any other compiler with the only
> possible exception of gcc, was developed by few persons, at most few
> dozens persons.

gcc probably has thousands of names in its list of non-trivial 
contributors.  How many of these were paid for their gcc work is a 
difficult question.

But there are /many/ big compilers that are massive development projects 
with many people working on them - and large, commercial budgets.  MSVC 
is an obvious case.  Others that I know of from personal experience 
include Borland's tools, Intel's tools, Code Warrior, Green Hills, 
Metrowerks, Keil, IAR.  These are not compilers written by a few people 
- even if you exclude everything that is not specifically the compiler 
itself.  (Note that LLVM is also more than just a C compiler.)

I have no way to compare them or their costs to that of clang/llvm.  But 
I don't think you do either.

> Now, if one takes compilers with environment/IDE,
> libraries etc, I can believe that there were cases when hundreds of
> people were working on all that stuff, but here we are talking about
> large number of people working on compiler itself.
> 

[toc] | [prev] | [next] | [standalone]

#142532

Le 27/03/2019 à 22:45, David Brown a écrit :
> You can be very sure that neither of these companies pay for more 
> salaries on LLVM developers than they can get away with - there is a 
> reason they are rich companies, and it is not because they pay for vast 
> numbers of compiler developers.

Sure, they are making movies now...

They had a Pepsi cola manager years ago... He led Apple to the ruin.

The one at the helm now doesn't know anything about computers, but 
that's not so bad.

He can make movies, or sell pepsi, whatever.

[toc] | [prev] | [next] | [standalone]

#142495

Melzzzzz <Melzzzzz@zzzzz.com> writes:
>On 2019-03-26, jacobnavia <jacob@jacob.remcomp.fr> wrote:
>> Le 24/03/2019 à 12:15, jacobnavia a écrit :
>>> It is the slowest compiler known
>>> 
>>>     For instance, to compile the source code of lcc (around 2.5MB of C) 
>>> we have compilation times of:
>>>     1 lcc 8.7 sec (with optimizations on)
>>>     2 gcc 38 sec (without optimization)
>>>     3 clang 5 minutes (without any optimizations).
>>> 
>>> WITH optimizations the landscape looks the same:
>>>     1 lcc: 8.7 sec
>>>     2 gcc 2minutes 48s
>>>     3 clang: 26 minutes
>>
>> That was version 5
>>
>> I upgraded to version 9.0, and waited 32 hours and 41 minutes for the 
>> build of clang to finish. CPU Quad core ARM64, 4GB RAM, SSD of 1 TB, 
>> linux OS.
>
>Hm, on my Ryzen 2700X it compiles in half of hour?
>
>>
>> The executable of clang is 2.3GB with debug information. Stripped it 
>> shrinks to "just" 194MB.
>Perhaps you should try release build?

Or parallel make.

[toc] | [prev] | [next] | [standalone]

#142498

Le 26/03/2019 à 21:55, Scott Lurndal a écrit :
> Melzzzzz <Melzzzzz@zzzzz.com> writes:
>> On 2019-03-26, jacobnavia <jacob@jacob.remcomp.fr> wrote:
>>> Le 24/03/2019 à 12:15, jacobnavia a écrit :
>>>> It is the slowest compiler known
>>>>
>>>>      For instance, to compile the source code of lcc (around 2.5MB of C)
>>>> we have compilation times of:
>>>>      1 lcc 8.7 sec (with optimizations on)
>>>>      2 gcc 38 sec (without optimization)
>>>>      3 clang 5 minutes (without any optimizations).
>>>>
>>>> WITH optimizations the landscape looks the same:
>>>>      1 lcc: 8.7 sec
>>>>      2 gcc 2minutes 48s
>>>>      3 clang: 26 minutes
>>>
>>> That was version 5
>>>
>>> I upgraded to version 9.0, and waited 32 hours and 41 minutes for the
>>> build of clang to finish. CPU Quad core ARM64, 4GB RAM, SSD of 1 TB,
>>> linux OS.
>>
>> Hm, on my Ryzen 2700X it compiles in half of hour?
>>
>>>
>>> The executable of clang is 2.3GB with debug information. Stripped it
>>> shrinks to "just" 194MB.
>> Perhaps you should try release build?
> 
> Or parallel make.
> 

One of the (many) problems here is the linker, that for some linking 
steps uses 10GB of memory. Swapping should be fast since I have 1TB SSD 
but surely it slows down the execution of stuff.

What I do not understand is that version 9 is 6 minutes SLOWER than 
version 5, and that is what is worrysome!

And, by the way, HALF AN HOUR in a RYZEN (!!!) is an incredible long time.

[toc] | [prev] | [next] | [standalone]

#142500

On 2019-03-26, jacobnavia <jacob@jacob.remcomp.fr> wrote:
> Le 26/03/2019 à 21:55, Scott Lurndal a écrit :
>> Melzzzzz <Melzzzzz@zzzzz.com> writes:
>>> On 2019-03-26, jacobnavia <jacob@jacob.remcomp.fr> wrote:
>>>> Le 24/03/2019 à 12:15, jacobnavia a écrit :
>>>>> It is the slowest compiler known
>>>>>
>>>>>      For instance, to compile the source code of lcc (around 2.5MB of C)
>>>>> we have compilation times of:
>>>>>      1 lcc 8.7 sec (with optimizations on)
>>>>>      2 gcc 38 sec (without optimization)
>>>>>      3 clang 5 minutes (without any optimizations).
>>>>>
>>>>> WITH optimizations the landscape looks the same:
>>>>>      1 lcc: 8.7 sec
>>>>>      2 gcc 2minutes 48s
>>>>>      3 clang: 26 minutes
>>>>
>>>> That was version 5
>>>>
>>>> I upgraded to version 9.0, and waited 32 hours and 41 minutes for the
>>>> build of clang to finish. CPU Quad core ARM64, 4GB RAM, SSD of 1 TB,
>>>> linux OS.
>>>
>>> Hm, on my Ryzen 2700X it compiles in half of hour?
>>>
>>>>
>>>> The executable of clang is 2.3GB with debug information. Stripped it
>>>> shrinks to "just" 194MB.
>>> Perhaps you should try release build?
>> 
>> Or parallel make.
>> 
>
> One of the (many) problems here is the linker, that for some linking 
> steps uses 10GB of memory. Swapping should be fast since I have 1TB SSD 
> but surely it slows down the execution of stuff.

Try release build.
>
> What I do not understand is that version 9 is 6 minutes SLOWER than 
> version 5, and that is what is worrysome!

Try release build.

>
> And, by the way, HALF AN HOUR in a RYZEN (!!!) is an incredible long time.
>


-- 
press any key to continue or any other to quit...

[toc] | [prev] | [next] | [standalone]

#142499

On 26/03/2019 20:48, Melzzzzz wrote:
> On 2019-03-26, jacobnavia <jacob@jacob.remcomp.fr> wrote:
>> Le 24/03/2019 à 12:15, jacobnavia a écrit :
>>> It is the slowest compiler known
>>>
>>>      For instance, to compile the source code of lcc (around 2.5MB of C)
>>> we have compilation times of:
>>>      1 lcc 8.7 sec (with optimizations on)
>>>      2 gcc 38 sec (without optimization)
>>>      3 clang 5 minutes (without any optimizations).
>>>
>>> WITH optimizations the landscape looks the same:
>>>      1 lcc: 8.7 sec
>>>      2 gcc 2minutes 48s
>>>      3 clang: 26 minutes
>>
>> That was version 5
>>
>> I upgraded to version 9.0, and waited 32 hours and 41 minutes for the
>> build of clang to finish. CPU Quad core ARM64, 4GB RAM, SSD of 1 TB,
>> linux OS.
> 
> Hm, on my Ryzen 2700X it compiles in half of hour?

For a laugh, I write a batch file so that, every time it invoked my C 
compiler, it compiled it from scratch (ie. compiler, assembler, linker 
from sources to 0.5MB exe).

It added 0.2 seconds to the build times of any program on my PC.

On my crappy laptop (1.6GHz Celeron I think that uses an SD card as hard 
drive, the cheapest one in the shop), it adds 0.4 seconds. All using a 
single core of course, and all code unoptimised as well.

These big compilers really are a different world.

[toc] | [prev] | [next] | [standalone]

#142501

On 2019-03-26, Bart <bc@freeuk.com> wrote:
> On 26/03/2019 20:48, Melzzzzz wrote:
>> On 2019-03-26, jacobnavia <jacob@jacob.remcomp.fr> wrote:
>>> Le 24/03/2019 à 12:15, jacobnavia a écrit :
>>>> It is the slowest compiler known
>>>>
>>>>      For instance, to compile the source code of lcc (around 2.5MB of C)
>>>> we have compilation times of:
>>>>      1 lcc 8.7 sec (with optimizations on)
>>>>      2 gcc 38 sec (without optimization)
>>>>      3 clang 5 minutes (without any optimizations).
>>>>
>>>> WITH optimizations the landscape looks the same:
>>>>      1 lcc: 8.7 sec
>>>>      2 gcc 2minutes 48s
>>>>      3 clang: 26 minutes
>>>
>>> That was version 5
>>>
>>> I upgraded to version 9.0, and waited 32 hours and 41 minutes for the
>>> build of clang to finish. CPU Quad core ARM64, 4GB RAM, SSD of 1 TB,
>>> linux OS.
>> 
>> Hm, on my Ryzen 2700X it compiles in half of hour?
>
> For a laugh, I write a batch file so that, every time it invoked my C 
> compiler, it compiled it from scratch (ie. compiler, assembler, linker 
> from sources to 0.5MB exe).
>
> It added 0.2 seconds to the build times of any program on my PC.
>
> On my crappy laptop (1.6GHz Celeron I think that uses an SD card as hard 
> drive, the cheapest one in the shop), it adds 0.4 seconds. All using a 
> single core of course, and all code unoptimised as well.
>
> These big compilers really are a different world.

llvm/clang has tremendeous ammount of commits.
>


-- 
press any key to continue or any other to quit...

[toc] | [prev] | [next] | [standalone]

#142512

On 26/03/2019 22:00, Melzzzzz wrote:
> On 2019-03-26, Bart <bc@freeuk.com> wrote:

>> These big compilers really are a different world.
> 
> llvm/clang has tremendeous ammount of commits.

I'm not quite sure what that means. Why it would it affect build time?

I understand llvm is a VERY complex toolkit for compilers. When I looked 
at it a decade or more ago, I counted 44 /sets/ of documentation of all 
kinds.

The 2.3GB debug build that was reported (I think the one that took 32 
hours on a machine that, while modest these days, must surely rank as a 
supercomputer compared to even 20 years ago), is, I understand it, for a 
C compiler.

It was stated that a stripped-down build was 'only' 200MB. Well, with my 
own compilers, I get roughly 10 bytes of executable per line of code. So 
200MB means 20M lines of code, if the code generation was as poor, 
otherwise it might be more like 30M lines of code.

Printed out, the listing would be 50m or 160 feet tall (about as high as 
Nelson's Column).

Somehow, the spirit of C seems to have gone amiss.

[toc] | [prev] | [next] | [standalone]

#142511

On Tuesday, March 26, 2019 at 10:06:40 PM UTC+2, jacobnavia wrote:
> Le 24/03/2019 à 12:15, jacobnavia a écrit :
> > It is the slowest compiler known
> > 
> >     For instance, to compile the source code of lcc (around 2.5MB of C) 
> > we have compilation times of:
> >     1 lcc 8.7 sec (with optimizations on)
> >     2 gcc 38 sec (without optimization)
> >     3 clang 5 minutes (without any optimizations).
> > 
> > WITH optimizations the landscape looks the same:
> >     1 lcc: 8.7 sec
> >     2 gcc 2minutes 48s
> >     3 clang: 26 minutes
> 
> That was version 5
> 
> I upgraded to version 9.0, and waited 32 hours and 41 minutes for the 
> build of clang to finish. CPU Quad core ARM64, 4GB RAM, SSD of 1 TB, 
> linux OS.
> 
> The executable of clang is 2.3GB with debug information. Stripped it 
> shrinks to "just" 194MB.
> 
> It compiles the same code used for the benchmark above even slower: 
> compilation time increases by 6 minutes. The figures are (with 
> optimizations ON):
> 
>      1 lcc: 8.7 sec
>      2 gcc 2minutes 48s
>      3 clang: 32 minutes
> 
> This is progress: always fatter , always bigger, always more features, etc.
> 
> The generated code is WRONG and crashes immediately. I have sent a bug 
> report to clang on monday, after receiving the authorization from them 
> to do so.
> 
> So far no answer.
> 
> jacob

Can you, please, specify you hardware setup more precisely than just "ARM64' ?
The differences between different aarch64 CPUs could be easily 7x in single-thread speed (fastest iPad Pro vs fastest Raspbery Pi 2) and much more than that (20x ? or more?) in multithreaded build performance (where "fast" part is a server with dual-socket ThunderX2 or with the biggest Qualcomm Centriq).

[toc] | [prev] | [next] | [standalone]

#142571 — Re: clang: release build

Following the tip of melzzzz I recompiled clang using the "Release" 
settings and rebuild only the arm64 target.

This reduced the compilation time from 32 hours to just 9 hours and 14 
minutes.

The compilation time now is normal, the same time as gcc.

The resulting executable is still wrong however and crashes immediately.

jacob

[toc] | [prev] | [next] | [standalone]

#142887

On 3/24/19 3:45 PM, jacobnavia wrote:
> What are your experiences with clang?
> 

[...]

  Hi,

An actual working software is better for compilation time and 
bench-marking comparison. I suggest Cfish, a C port for stock-fish 
written in C++.

https://github.com/syzygy1/Cfish

Please inform of the result.

-- 
www.my-c-codes.com/

Farewell.

[toc] | [prev] | [standalone]

Page 2 of 2 — ← Prev page 1 [2]

Back to top | Article view | comp.lang.c

csiph-web