Greetings Adam, Sorry for not being clear originally, but the problem is that any object compiled with VIS is tagged to run only on v8plusa (or v8plusb, v9, etc.). This makes it impossible to execute ELF images that contain objects that use VIS code, because at link-time, the entire ELF executable is tagged, as well. For programs, like MPlayer, that are capable of handling a SIGILL and disabling further VIS instruction execution, this makes it impossible to do things like runtime-cpu-detection.
The GNU assembler, linker, and ELF loader do not make such distinctions. This differs in that Mach-O's contain code for varying ISAs, like x86 and PowerPC. I simply just want to execute an ELF executable that contains VIS code on sun4m and have Solaris trust the fact that the application (in this case, MPlayer) is capable of handling the impending chaos of SIGILL. Thanks > Hey Derek, > > Are you imaging something like Apple's "fat" mach-o > binaries on Mac OS X > that support multiple ISAs in a single binary? > There's nothing like that > for Solaris -- we've chosen a different tack which is > to deliver the > individual binaries and then choose the appropriate > one with an intermediate > binary (isaexec). > > Adam > > On Sun, Dec 10, 2006 at 12:56:38PM -0600, Derek E. > Lewis wrote: > > Greetings, > > > > I'm curious if it's possible to link an ELF with > multiple objects > > compiled for varying CPUs (say, SPARCv8, > UltraSPARC-I, etc.) and the > > header of the ELF cannot be modified to state that > the ELF is for the > > CPU of the object of the highest CPU? > > > > MPlayer's runtime-CPU-detection depends on this, as > when a VIS > > instruction is executed, SIGILL is caught and VIS > instructions are > > disabled in the MPlayer binary. Right now, certain > objects need to be > > built for UltraSPARC-I, because they contain VIS1 > instructions. > > Obviously, the object cannot be built for SPARCv8. > It seems if this > > weren't possible, SIGILL would be quite useless, so > this is why I think > > there's a solution. > > > > Any thoughts? > > > > -- > > Derek E. Lewis > > delewis at acm.org > > http://riemann.solnetworks.net/~dlewis > > > > _______________________________________________ > > tools-linking mailing list > > tools-linking at opensolaris.org > > -- > Adam Leventhal, Solaris Kernel Development > http://blogs.sun.com/ahl > _________________________________________ > tools-linking mailing list > tools-linking at opensolaris.org > This message posted from opensolaris.org
