Loading ...
Sorry, an error occurred while loading the content.

Re: [solarisx86] x86 cross compiler for ULTRASPARC processors

Expand Messages
  • Keith Bierman
    ... Sun sold such a product for some years (it ran, and targeted x86, 68K and SPARC). It sold exceedingly poorly and the project was terminated something like
    Message 1 of 8 , Sep 30, 2004
    • 0 Attachment
      maybird1776 wrote:

      > You ever how children come up with the most craziest things?
      >
      > "Dear Uncle Ken,
      >
      > Why don't they (i.e. Sun) come up with a cross compiler for both
      > platforms (i.e. SPARC/x86)

      Sun sold such a product for some years (it ran, and targeted x86, 68K
      and SPARC). It sold exceedingly poorly and the project was terminated
      something like 13 years ago.

      There aren't too many customers who are prepared to compile code and
      ship it to anyone without any testing ... and if you've got the target
      to test on, compiling on it wasn't perceived as much of a burden in that
      long ago era.

      Do the folks on this list think that this would really be one of the
      most important, useful or productive uses of our resources?
      --
      Keith H. Bierman keith.bierman@...|
      <speaking for myself, not Sun*> Copyright 2004
    • Phillip Bruce
      Keith, I used to work for Sun back when 2.5.1 and 2.6 were still the latest versions of Solaris at that time. I do remember before then during SunOS 4.x the
      Message 2 of 8 , Oct 1, 2004
      • 0 Attachment
        Keith,

        I used to work for Sun back when 2.5.1 and 2.6 were still the latest versions of Solaris at that time.
        I do remember before then during SunOS 4.x the ability to recompile the kernel. It wasn't a favorable
        practice then as too many people then didn't do enough testing of their compilation before putting
        the new kernel into production.

        I the fact that I have a stable kernel and although I have enough experience to know better to compile
        the kernel but the question is do I have the time to do the type of testing to insure that it worked. In
        most cases I don't and that is why I don't bother with taking the time.

        So I guess the real answer here is that if you have a testing environment that you don't mind crashing
        and being down 90% when your testing something the go ahead. But if your in only in a environment
        where testing isn't a viable option. Then I rather rely on Sun to do that for me anyway.

        The thing is you get to be pretty good with messing around with the kernel later by having a developing
        enviornment but you don't if your stuck in a pure production environment.

        The next question then if a customer has a hack or a solution worth putting into the kernel would they
        test a customers solution and the next question would they have the time and $$$ to do it with?

        Phillip


        ----- Original Message -----
        From: Keith Bierman
        To: solarisx86@yahoogroups.com
        Sent: 9/30/2004 10:33:12 PM
        Subject: Re: [solarisx86] x86 cross compiler for ULTRASPARC processors


        maybird1776 wrote:

        > You ever how children come up with the most craziest things?
        >
        > "Dear Uncle Ken,
        >
        > Why don't they (i.e. Sun) come up with a cross compiler for both
        > platforms (i.e. SPARC/x86)

        Sun sold such a product for some years (it ran, and targeted x86, 68K
        and SPARC). It sold exceedingly poorly and the project was terminated
        something like 13 years ago.

        There aren't too many customers who are prepared to compile code and
        ship it to anyone without any testing ... and if you've got the target
        to test on, compiling on it wasn't perceived as much of a burden in that
        long ago era.

        Do the folks on this list think that this would really be one of the
        most important, useful or productive uses of our resources?
        --
        Keith H. Bierman keith.bierman@...|
        <speaking for myself, not Sun*> Copyright 2004




        Please check the Links page before posting:
        http://groups.yahoo.com/group/solarisx86/links
        Post message: solarisx86@yahoogroups.com
        UNSUBSCRIBE: solarisx86-unsubscribe@yahoogroups.com




        Yahoo! Groups Sponsor
        ADVERTISEMENT






        Yahoo! Groups Links

        To visit your group on the web, go to:
        http://groups.yahoo.com/group/solarisx86/

        To unsubscribe from this group, send an email to:
        solarisx86-unsubscribe@yahoogroups.com

        Your use of Yahoo! Groups is subject to the Yahoo! Terms of Service.

        [Non-text portions of this message have been removed]
      • Al Hopper
        ... Nawh - not me. Because I use an Ultra 5 for building small projects. People on this list are probably shocked. What? A POS Ultra 5? You get the U5,
        Message 3 of 8 , Oct 1, 2004
        • 0 Attachment
          On Thu, 30 Sep 2004, Keith Bierman wrote:

          >
          > maybird1776 wrote:
          >
          > > You ever how children come up with the most craziest things?
          > >
          > > "Dear Uncle Ken,
          > >
          > > Why don't they (i.e. Sun) come up with a cross compiler for both
          > > platforms (i.e. SPARC/x86)
          >
          > Sun sold such a product for some years (it ran, and targeted x86, 68K
          > and SPARC). It sold exceedingly poorly and the project was terminated
          > something like 13 years ago.
          >
          > There aren't too many customers who are prepared to compile code and
          > ship it to anyone without any testing ... and if you've got the target
          > to test on, compiling on it wasn't perceived as much of a burden in that
          > long ago era.
          >
          > Do the folks on this list think that this would really be one of the
          > most important, useful or productive uses of our resources?

          Nawh - not me. Because I use an Ultra 5 for building small projects.
          People on this list are probably shocked. What? A POS Ultra 5? You get
          the U5, trash the nasty disk drive that comes standard in them, and replace
          it with the best 7,200 RPM 8Mb cache drive you can find. Then max out the
          memory. This makes the U5 "feel" twice as fast as the factory box[1]. It
          still won't beat any world speed records, but it'll build a project in a
          reasonable amount of time - and it has a much lower thermal footprint than
          a dual US-III Sun 280 box. You (scp the code over) type "make", come back
          in 10 minutes and it's done. Then scp the binaries back to your real work
          environment. That's all you need to build SPARC binaries: modded Ebayed
          U5, SunPro compilers and < 1 cup of coffee.

          [1] I like the Maxtors for this, because, looking at the specs, the standby
          power consumption is very low, and 99% of the time the drive is idle in
          this application. There's not much airflow over the HDA in a U5. YMMV -
          check disk drive spec sheets before you decide.

          Al Hopper Logical Approach Inc, Plano, TX. al@...
          Voice: 972.379.2133 Fax: 972.379.2134
          We can stop programming when computers understand:
          Time flies like an an arrow; fruit flies like a banana.
        • maybird1776
          ... the ... projects. ... You get ... replace ... out the ... [1]. It ... in a ... footprint than ... come back ... real work ... Ebayed ... True. I have a
          Message 4 of 8 , Oct 1, 2004
          • 0 Attachment
            --- In solarisx86@yahoogroups.com, Al Hopper <al@l...> wrote:
            > On Thu, 30 Sep 2004, Keith Bierman wrote:
            >
            > >
            > > maybird1776 wrote:
            > >
            > > > Why don't they (i.e. Sun) come up with a cross compiler for both
            > > > platforms (i.e. SPARC/x86)
            > >------------------------------

            Keith wrote:
            > > Do the folks on this list think that this would really be one of
            the
            > > most important, useful or productive uses of our resources?

            > Al wrote:
            > Nawh - not me. Because I use an Ultra 5 for building small
            projects.
            > People on this list are probably shocked. What? A POS Ultra 5?
            You get
            > the U5, trash the nasty disk drive that comes standard in them, and
            replace
            > it with the best 7,200 RPM 8Mb cache drive you can find. Then max
            out the
            > memory. This makes the U5 "feel" twice as fast as the factory box
            [1]. It
            > still won't beat any world speed records, but it'll build a project
            in a
            > reasonable amount of time - and it has a much lower thermal
            footprint than
            > a dual US-III Sun 280 box. You (scp the code over) type "make",
            come back
            > in 10 minutes and it's done. Then scp the binaries back to your
            real work
            > environment. That's all you need to build SPARC binaries: modded
            Ebayed
            > U5, SunPro compilers and < 1 cup of coffee.

            True. I have a Sun SPARCstation 5 at home with 256MB memory running
            Solaris 8. I tried to install Solaris 10 on it (even Solaris 9) but I
            ran out of disk space (4GB) and I don't have an external MultiPack
            for it (and sun4m platforms are being sent out to the pasture in
            Solaris 10). So, how would I test Solaris 10 compiles on a
            SPARCstation 5 in this case?!? I ran out of disk space with Solaris 9
            as well trying to put all off the developer libraries on it (nice try
            though!). Then, try to install Sun Studio 9 on it. Eegads! Not a
            workstation to consider in this day and age of 20MB packages/tarballs
            to compile and port.

            Guess what though? CSWkde_gcc (back during KDE 3.1.4) was compiled
            and tested on this same machine. It took about 2-3 full days just to
            compile QT and ARTS. Two work shifts and some coffee and doughnuts,
            and 12 hours to spare, I could probably get Qt 3.3.3 to compile fully
            by then. ;o>

            Compile something like AbiWord 2.1.7 on the Sun Ultra 5 or maybe the
            full distro on KDE 3.3.0 packages. See how long you tie up that SPARC
            workstation compiling all of those packages. Well, politically
            corrected I should say compile a full Gnome 2.8 desktop distro
            (http://ftp.gnome.org/pub/GNOME/desktop/2.8/2.8.0/sources/) on the
            Sun Ultra 5! Now, do this on a Sun Java workstation. Why not compile
            SPARC code on a Sun Java workstation and 'scp' it to the Ultra 5 for
            testing and validation?!?!?! Why not have today's developer compiling
            C/C++/Fortran code for SPARC workstations using the Sun Java
            workstations and AMD64 processors to churn out this code?!??! I'm
            talking about those large open source projects where people are
            working in their spare time and want to port and test code for SPARC
            machines. A cross compiler for SPARC would be nice to have on a fast
            processor machine like the Sun's W2100z workstation.

            ~Ken Mays
            @ Earthlink, Inc. (My comments are of own nature at all times)
          • Al Hopper
            On Fri, 1 Oct 2004, maybird1776 wrote: ... You have a good point there. ... You make a compelling case for a cross compiler Ken. And it would certainly
            Message 5 of 8 , Oct 1, 2004
            • 0 Attachment
              On Fri, 1 Oct 2004, maybird1776 wrote:

              ... snip ....
              > Compile something like AbiWord 2.1.7 on the Sun Ultra 5 or maybe the
              > full distro on KDE 3.3.0 packages. See how long you tie up that SPARC
              > workstation compiling all of those packages. Well, politically
              > corrected I should say compile a full Gnome 2.8 desktop distro
              > (http://ftp.gnome.org/pub/GNOME/desktop/2.8/2.8.0/sources/) on the
              > Sun Ultra 5! Now, do this on a Sun Java workstation. Why not compile
              > SPARC code on a Sun Java workstation and 'scp' it to the Ultra 5 for
              > testing and validation?!?!?! Why not have today's developer compiling

              You have a good point there.

              > C/C++/Fortran code for SPARC workstations using the Sun Java
              > workstations and AMD64 processors to churn out this code?!??! I'm
              > talking about those large open source projects where people are
              > working in their spare time and want to port and test code for SPARC
              > machines. A cross compiler for SPARC would be nice to have on a fast
              > processor machine like the Sun's W2100z workstation.

              You make a compelling case for a cross compiler Ken. And it would
              certainly leverage the AMD based hardware you mention. I think the
              difference is one of scale between your/my needs. And it makes
              sense for Sun to provide the users with multiple options to meet their
              differing needs.

              BTW: I commend you for your ongoing contributions to the Solaris community.
              The only thing that we each have a limited supply of, is time.

              Regards,

              Al Hopper Logical Approach Inc, Plano, TX. al@...
              Voice: 972.379.2133 Fax: 972.379.2134
              We can stop programming when computers understand:
              Time flies like an an arrow; fruit flies like a banana.
            • Keith Bierman
              ... It wasn t so much the ability as the requirement. And you could get into all sorts of interesting trouble. As I was doing some application performance work
              Message 6 of 8 , Oct 1, 2004
              • 0 Attachment
                Phillip Bruce wrote:

                > Keith,
                >
                > I used to work for Sun back when 2.5.1 and 2.6 were still the latest
                > versions of Solaris at that time.
                > I do remember before then during SunOS 4.x the ability to recompile
                > the kernel.

                It wasn't so much the ability as the requirement. And you could get into
                all sorts of interesting trouble. As I was doing some application
                performance work with a limited RAM system, I once configured out the
                mouse and window system. Interestingly, that resulted in an unbootable
                system. It was interesting to repair the system (where interesting means
                I managed to get one of the OS wizards to scratch his head about how I
                got into such a state ;>).

                > It wasn't a favorable
                > practice then as too many people then didn't do enough testing of
                > their compilation before putting
                > the new kernel into production.

                As with kernels, so too with non-trivial applications.

                > ve the time to do the type of testing to insure that it worked. In
                > most cases I don't and that is why I don't bother with taking the time.

                Please don't work on software that I have to rely on then!

                >
                >
                > So I guess the real answer here is that if you have a testing
                > environment that you don't mind crashing
                > and being down 90% when your testing something the go ahead. But if
                > your in only in a environment
                > where testing isn't a viable option. Then I rather rely on Sun to do
                > that for me anyway.

                Sun can't possibly test your application for you. Just having a working
                compiler and runtimes doesn't mean the application writer hasn't shot
                themselves in the foot.

                Let's just pick one example of an area of some complexity. For floating
                point applications cross compilation offers whole new ways to get
                varying answers, as compile time constant evaluation will be using the
                rules of the processor you are compiling *with*, not the target. x86
                with it's extended registers won't produce the same answers as SPARC
                (unless one ignores the native fp and uses software emulation. You can
                get very, very close using SSE arithmetic alone, for processors with SSE
                ... etc.).

                There are lots of really strange corner cases of no interest to most
                people, but which make validation of cross compilers a real burden (and
                since all real worlds have some limit to resources, it means there are
                other projects that will have to be downsized or eliminated to add this
                work ... the decision any rational management has to make, is what
                results in the best result for the most people. Cross compilers seldom
                meet that test (outside of embedded markets ;>)

                Worse, as you happily point out, if it's easy to set things up so that
                testing can be skipped, some will. That's what my lawyer friends call an
                "attractive nuisance". :>

                >
                > The next question then if a customer has a hack or a solution worth
                > putting into the kernel would they
                > test a customers solution and the next question would they have the
                > time and $$$ to do it with?

                This seems to have wandered from a cross compiler discussion to an Open
                Solaris query (viz. with Open Solaris how does Sun arrange to validate
                customer's additions to the OS. I haven't got a clue (I suspect it's
                infeasible, but my imagination may be limited.
                --
                Keith H. Bierman keith.bierman@...|
                <speaking for myself, not Sun*> Copyright 2004
              • Keith Bierman
                ... You are expected to develop on the *oldest* platform you intend to support and test (if necessary, and hopefully you will ;.) on more modern platforms. If
                Message 7 of 8 , Oct 1, 2004
                • 0 Attachment
                  maybird1776 wrote:

                  >
                  > True. I have a Sun SPARCstation 5 at home with 256MB memory running
                  > Solaris 8. I tried to install Solaris 10 on it (even Solaris 9) but I
                  > ran out of disk space (4GB) and I don't have an external MultiPack
                  > for it (and sun4m platforms are being sent out to the pasture in
                  > Solaris 10). So, how would I test Solaris 10 compiles on a
                  > SPARCstation 5 in this case?!?

                  You are expected to develop on the *oldest* platform you intend to
                  support and test (if necessary, and hopefully you will ;.) on more
                  modern platforms. If your application relies only on ABI stable entities
                  (passes appcert) you ought to have it just run on the newer OS. That's
                  not cross compilation in the usual sense of the word ;>

                  > Sun Ultra 5! Now, do this on a Sun Java workstation. Why not compile
                  > SPARC code on a Sun Java workstation and 'scp' it to the Ultra 5 for
                  > testing and validation?!?!?! Why not have today's developer compiling

                  Note that you are now asking for multi-general backward OS support as
                  well as cross ISA compilation. That was never supported (although you
                  could find out clues as how to do it buried in the f77 documentation, if
                  you looked carefully).

                  > working in their spare time and want to port and test code for SPARC
                  > machines. A cross compiler for SPARC would be nice to have on a fast
                  > processor machine like the Sun's W2100z workstation.

                  For a fraction of the development and testing resources, we could offer
                  "comp" time on the $1/cpu hour deal that JS was blogging about. If you
                  think that would be useful and important to your OpenSource efforts,
                  contact JS about setting up such a program;>
                  --
                  Keith H. Bierman keith.bierman@...|
                  <speaking for myself, not Sun*> Copyright 2004
                Your message has been successfully submitted and would be delivered to recipients shortly.