llvm dev - Mar 2017 - get function parameters (not arguments)

what about the memory address of e and f? can i get them?

Thank you and best,
Mo

On Fri, Mar 10, 2017 at 5:02 PM, Tim Northover <t.p.northover at
gmail.com>
wrote:
> On 10 March 2017 at 15:49, Mohammad Norouzi <mnmomn at gmail.com>
wrote:
> > for (auto& A : cast<CallInst>(BI)->arg_operands())
> > errs() << "--- " << A->getName() <<
"\n";
>
> Ah, I see. You actually want "e" as a name. Unforuntately this
isn't
> possible in general for a few reasons.
>
> First, release builds of LLVM drop most Value names for efficiency
> reasons. Without optimization your IR would just have numbers in the
> arg position: "call void @foo(i32 %3, i32 %4)". Even those
numbers
> don't exist in memory, but are constructed as needed.
>
> Another confounding factor is that optimization will turn even that
> into a plain "call void @foo(i32 10, i32 22)" with no record of
what
> the original arguments were.
>
> It gets even worse at the C level. If you saw "foo(a + b)", that
> argument doesn't really have a natural name. It's a temporary
> expression.
>
> So in general you shouldn't be relying on the name for any actual
> transformation you do.
>
> For debugging there are usually alternatives that might not be quite
> so convenient but do the job: Just dumping the argument is usually
> sufficient; with heroic effort you might be able to use any debug info
> that's present.
>
> Cheers.
>
> Tim.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20170310/d6cb6973/attachment.html>

On 10 March 2017 at 08:06, Mohammad Norouzi <mnmomn at gmail.com>
wrote:
> what about the memory address of e and f? can i get them?
Not in general, again. After optimizations they probably won't even
have a fixed memory address. For example on AArch64 that code might
become something like:

    main:
        mov w0, #10
        mov w1, #22
        bl foo
        ret

Where e and f only ever exist in registers.

In some cases you might get lucky and see that the Value used in the
call came from a load instruction, and then you could get its address
from that. But it's not something you can rely on for correctness.

Cheers.

Tim.

What if I use -O0? In this case, there should be a load for each parameter
before the function call, yes?

Thank you and best,
Mo

On Fri, Mar 10, 2017 at 5:09 PM, Tim Northover <t.p.northover at
gmail.com>
wrote:
> On 10 March 2017 at 08:06, Mohammad Norouzi <mnmomn at gmail.com>
wrote:
> > what about the memory address of e and f? can i get them?
>
> Not in general, again. After optimizations they probably won't even
> have a fixed memory address. For example on AArch64 that code might
> become something like:
>
>     main:
>         mov w0, #10
>         mov w1, #22
>         bl foo
>         ret
>
> Where e and f only ever exist in registers.
>
> In some cases you might get lucky and see that the Value used in the
> call came from a load instruction, and then you could get its address
> from that. But it's not something you can rely on for correctness.
>
> Cheers.
>
> Tim.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20170310/dfa1ee12/attachment-0001.html>