Displaying 3 results from an estimated 3 matches for "control_dependences_on".
Did you mean:
control_dependences_only
2019 Dec 16
7
[RFC] How to manifest information in LLVM-IR, or, revisiting llvm.assume
...optimal, especially since we do
not need to "optimize" these instructions anyway.
- The current (=arbitrary) side-effects of `llvm.assume` are too strong.
We have `inaccessiblemem_or_argmemonly` and we might be able to come
up with something even more specialized for this, e.g.,
`control_dependences_only` to indicate that there are only control
dependences.
-
Some Design Ideas:
1) Use named operand bundles to encode information.
If we want to encode property XYZ for a value %V holds at a certain
program point and the property is dependent on %N we could encode
that as:
`llvm.a...
2019 Dec 18
2
[RFC] How to manifest information in LLVM-IR, or, revisiting llvm.assume
...ot;optimize" these instructions anyway.
> >
> > - The current (=arbitrary) side-effects of `llvm.assume` are too strong.
> > We have `inaccessiblemem_or_argmemonly` and we might be able to come
> > up with something even more specialized for this, e.g.,
> > `control_dependences_only` to indicate that there are only control
> > dependences.
>
> This is all well put; I think you’ve covered the major weaknesses.
>
>
> > Some Design Ideas:
> >
> > 1) Use named operand bundles to encode information.
> > If we want to encode proper...
2019 Dec 18
2
[RFC] How to manifest information in LLVM-IR, or, revisiting llvm.assume
...way.
> >>>
> >>> - The current (=arbitrary) side-effects of `llvm.assume` are too strong.
> >>> We have `inaccessiblemem_or_argmemonly` and we might be able to come
> >>> up with something even more specialized for this, e.g.,
> >>> `control_dependences_only` to indicate that there are only control
> >>> dependences.
> >>
> >> This is all well put; I think you’ve covered the major weaknesses.
> >>
> >>
> >>> Some Design Ideas:
> >>>
> >>> 1) Use named operand bundle...