erichkeane added a comment.
In D112349#3109211 <https://reviews.llvm.org/D112349#3109211>,
@nextsilicon-itay-bookstein wrote:
>> I don't know much about the ELF format... but this works today? We can
>> define a resolver in a different TU and it WORKS thanks to the linker? So
>> there is perhaps something?
>
> The ifunc symbol that is emitted in the TU with the undefined resolver loses
> its connection to the resolver and the calls to the ifunc are instead bound
> against the resolver itself (which is absolutely not what you want).
>
> itay> cat specific.c
> #include <stdio.h>
>
> __attribute__((cpu_specific(generic)))
> void single_version(void){
> puts("In single_version generic");
> }
>
> void useage() {
> single_version();
> }
>
> itay> cat dispatch_main.c
> void useage(void);
>
> __attribute__((cpu_dispatch(generic)))
> void single_version(void);
>
> int main()
> {
> useage();
> single_version();
> return 0;
> }
>
> itay> clang -c dispatch_main.c -o dispatch_main.c.o
> itay> clang -c specific.c -o specific.c.o
> itay> clang specific.c.o dispatch_main.c.o -o main
> itay> ./main
> In single_version generic
>
> This line should have been printed twice, not once.
I see... thank you for your patience, my knowledge of ELF (or, that is,
basically everything after Clang LLVM-IR generation) is pretty slim.
I think what I would need to be able to correctly implement this is some level
of 'forward declarable' ifunc. I have to be able to process declarations as we
reach them, so just doing a function declaration then rewriting it wouldn't be
allowed.
Is it possible to get some IR-level 'ifunc' declaration implemented? If so, I
think I have a good idea on how to implement that in the CFE.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D112349/new/
https://reviews.llvm.org/D112349
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
