Manna marked an inline comment as done.
Manna added inline comments.
================
Comment at: clang/utils/TableGen/SveEmitter.cpp:302
unsigned Shift = llvm::countr_zero(Mask);
+ assert(Shift >= 64 && "Shift is out of encodable range");
return (V << Shift) & Mask;
----------------
tahonermann wrote:
> Manna wrote:
> > sdesmalen wrote:
> > > erichkeane wrote:
> > > > sdesmalen wrote:
> > > > > erichkeane wrote:
> > > > > > Shouldn't this be: `assert(Shift < 64 &&"...")`?
> > > > > >
> > > > > > `expr.shift` (https://eel.is/c++draft/expr.shift) says:
> > > > > > ```
> > > > > > The operands shall be of integral or unscoped enumeration type and
> > > > > > integral promotions are performed.
> > > > > > The type of the result is that of the promoted left operand.
> > > > > > The behavior is undefined if the right operand is negative, or
> > > > > > greater than or equal to the width of the promoted left operand.```
> > > > > >
> > > > > > uint64 stays as an `unsigned long`, so it is still 64 bits, so the
> > > > > > only invalid value for `Shift` is 64 (though >64 is 'nonsense', but
> > > > > > only impossible because of `llvm::countr_zero`).
> > > > > >
> > > > > > One thing to consider: I wonder if we should instead be changing
> > > > > > the 'shift' to be:
> > > > > >
> > > > > > `(V << (Shift % 64)) && Mask` ? It looks like `arm_sve.td` has the
> > > > > > `NoFlags` value as zero, which I think will end up going through
> > > > > > here possibly (or at least, inserted into `FlagTypes`.
> > > > > >
> > > > > > So I suspect an assert might not be sufficient, since a 64 bit
> > > > > > shift is possible in that case (since a zero 'Mask' is the only
> > > > > > case where `countr_zero` will end up being 64).
> > > > > >
> > > > > >
> > > > > > So I suspect an assert might not be sufficient, since a 64 bit
> > > > > > shift is possible in that case (since a zero 'Mask' is the only
> > > > > > case where countr_zero will end up being 64).
> > > > > It should be fine to assert that `Mask != 0`, since that would be an
> > > > > invalid mask.
> > > > Thanks for the comment @sdesmalen! Is there something that prevents
> > > > the `NoFlags` from being passed as the `MaskName` here?
> > > There's nothing that actively prevents it, but `encodeFlag` is a utility
> > > function that has no uses outside this file and has only 4 uses. Adding
> > > an assert should be sufficient.
> > Thank you for the explanation!
> I'm not sure if asserting `Mask != 0` will suffice to silence Coverity. While
> Coverity can specifically observe that `countr_zero` might return 0 (because
> `TrailingZerosCounter<T, 8>::count()` has a `return 64` statement), I don't
> think Coverity can determine that the function can't return 65 or higher. I
> think Erich's initial intuition is correct; the concern that Coverity is
> reporting is that the shift might overflow, so that is what should be guarded.
> assert(Shift < 64 && "Mask value produced an invalid shift value");
Thanks @tahonermann for comments. I have updated my patch to guard the shift
overflow issue with Mask value.
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D150140/new/
https://reviews.llvm.org/D150140
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
