ILP32 support?
Are there any plans to support ILP32 mode?
Vadim Furman
Also, there would have to be some changes to the kernel. This would create ELF64 processes that have to live within the first 4G of memory. We already support 32-bit processes but some kernel changes would be required to identify this new entity.
Quoted 17 lines · David Mosberger
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Vadim Furman
Vadim> Are there any plans to support ILP32 mode? Vadim Furman
No. So far, nobody has felt that the (small) benefits of ILP32 are worth its trouble. No kernel mods would be required, but you'd have to create a completely new user-level environment (new compiler, new header files, all libraries would have to be recompiled with ILP32 mode, etc.).
--david_______________________________________________
Linux-IA64 mailing list
[email protected]
http://lists.linuxia64.org/lists/listinfo/linux-ia64
Signature · 4 lines
Don Dugger "Censeo Toto nos in Kansa esse decisse." - D. Gale [email protected] Ph: 303/938-9838
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Vadim Furman
Vadim> Are there any plans to support ILP32 mode? Vadim Furman
No. So far, nobody has felt that the (small) benefits of ILP32 are worth its trouble. No kernel mods would be required, but you'd have to create a completely new user-level environment (new compiler, new header files, all libraries would have to be recompiled with ILP32 mode, etc.).
--david
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Don Dugger
Don> Also, there would have to be some changes to the kernel. This Don> would create ELF64 processes that have to live within the first Don> 4G of memory. We already support 32-bit processes but some Don> kernel changes would be required to identify this new entity.
Actually, the forthcoming kernel patch already has a per-process mmap-base. So the only change required in the kernel would be to recognize ILP32 binaries and set the mmap-base accordingly (so, yeah, I lied when I said no changes would be required... ;-). (This doesn't mean that I think ILP32 support is a good idea...)
--david
Quoted 6 lines
Vadim> Are there any plans to support ILP32 mode? Vadim Furman
No. So far, nobody has felt that the (small) benefits of ILP32 are worth its trouble.
IMHO benefits of ILP32 are huge, since you can relatively simply port existing 32 apps to ia64 and in the same time enjoy the perfomance of the new chip.
Is it possible to force LP64 processes below 4G? Is there a compiler/linker switch to do so?
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Vadim Furman
Vadim> IMHO benefits of ILP32 are huge, since you can relatively Vadim> simply port existing 32 apps to ia64 and in the same time Vadim> enjoy the perfomance of the new chip.
If it's not a performance critical program, you might just as well leave it an x86 binary and do no porting at all. If it's a performance critical program, then it's probably worthwhile to make it 64-bit clean (which isn't hard) and get the benefit of wide arithmetic and assurance that the program can deal with 64-bit files, address spaces, and y2038 without any problem at all.
Vadim> Is it possible to force LP64 processes below 4G? Is there a Vadim> compiler/linker switch to do so?
You can use a linker script to force things below 4GB. For mmap, you'd need to change the kernel to push the mmap base below 4GB. You'll be on your own though if you do this. All open source programs I know of are 64-bit clean so the incentive for the open source community to support ILP32 is very small.
--david
I would like to express real concern with this proposal. We went down this path with IRIX on the MIPS processor.
The situation is: We are asking ISVs to generate native IA64 binaries. If there are two formats of such
binaries, and they are not link compatible, now any ISV failing to release both formats will cause a terrible fragmentation. Any given user will not be sure of being able to find link compatible stuff for their needs. This is a terrible inhibitor to the spread of software, not to mention creating dual libraries for all IA64 stuff (ok guys, where are you going to put these new libs? This is a new ABI you understand!!!).
But perhaps you were not suggesting this. I believe Microsoft has defined a special form of native 64bit-address IA64 code, called "the sandbox", where the linker and kernel simply assign all addresses below say 2Gbyte. This code still contains 64bit pointer types, but it has no actual values exceeding 32bit container. Hence sloppy code converting pointers to say ints doesn't lose bits. But the pointer data type still changes size to 64bits, hence porting is not 0 effort. I'm not clear on the linkage compatibility issues here.
I would implore you to not issue such buggy code. Please make your code
64bit-address clean. Please don't split the IA64 marketplace!!!!!!!!!
Thanks, Rich
Quoted 16 lines · Vadim Furman
Quoted 6 lines
Vadim> Are there any plans to support ILP32 mode? Vadim Furman
No. So far, nobody has felt that the (small) benefits of ILP32 are worth its trouble.
IMHO benefits of ILP32 are huge, since you can relatively simply port existing 32 apps to ia64 and in the same time enjoy the perfomance of the new chip.
Is it possible to force LP64 processes below 4G? Is there a compiler/linker switch to do so?_______________________________________________
Linux-IA64 mailing list
[email protected]
http://lists.linuxia64.org/lists/listinfo/linux-ia64
Quoted 3 lines · Vadim Furman
IMHO benefits of ILP32 are huge, since you can relatively simply port existing 32 apps to ia64 and in the same time enjoy the perfomance of the new chip.
Any application assuming that sizeof(int) = sizeof(long) and/or sizeof(int) = sizeof(type *) is broken by implementation, because such assumptions are *never* really needed in correctly written code. The right thing is to fix it.
Dan