Displaying 6 results from an estimated 6 matches for "invariantess".
2016 Aug 29
2
GVN / Alias Analysis issue with llvm.masked.scatter/gather intrinsics
...ll expect
memorylocation to be a single pointer location.
I would chalk it up to "luck" that this patch fixes the bug.
It's pretty clear that MemoryLocation doesn't fit the needs of a lot of
stuff anymore (we hacked AA nodes into it, and lots of stuff now tries to
figure out the invariantess of the locations, blah blah blah), but it seems
like a big job to figure out what to replace it with that will work for
these cases.
(I'm pretty positive if we just make it MemoryLocations, and have
everything loop over the locations, the compiler will get a lot larger and
a lot slower)
On Mo...
2016 Aug 29
2
GVN / Alias Analysis issue with llvm.masked.scatter/gather intrinsics
...be a single pointer location.
>
> I would chalk it up to "luck" that this patch fixes the bug.
>
> It's pretty clear that MemoryLocation doesn't fit the needs of a lot of
> stuff anymore (we hacked AA nodes into it, and lots of stuff now tries to
> figure out the invariantess of the locations, blah blah blah), but it seems
> like a big job to figure out what to replace it with that will work for
> these cases.
>
> (I'm pretty positive if we just make it MemoryLocations, and have
> everything loop over the locations, the compiler will get a lot larger...
2016 Aug 30
2
GVN / Alias Analysis issue with llvm.masked.scatter/gather intrinsics
...t;
>
> > > > It's pretty clear that MemoryLocation doesn't fit the needs of
> > > > a
> > > > lot
> > > > of stuff anymore (we hacked AA nodes into it, and lots of stuff
> > > > now
> > > > tries to figure out the invariantess of the locations, blah
> > > > blah
> > > > blah), but it seems like a big job to figure out what to
> > > > replace
> > > > it
> > > > with that will work for these cases.
> > >
> >
>
> > > > (I'm pre...
2016 Aug 31
2
GVN / Alias Analysis issue with llvm.masked.scatter/gather intrinsics
...chalk it up to "luck" that this patch fixes the bug.
>>>>
>>>> It's pretty clear that MemoryLocation doesn't fit the needs of a lot of
>>>> stuff anymore (we hacked AA nodes into it, and lots of stuff now tries to
>>>> figure out the invariantess of the locations, blah blah blah), but it seems
>>>> like a big job to figure out what to replace it with that will work for
>>>> these cases.
>>>>
>>>> (I'm pretty positive if we just make it MemoryLocations, and have
>>>> everything...
2016 Aug 31
2
GVN / Alias Analysis issue with llvm.masked.scatter/gather intrinsics
...quot;luck" that this patch fixes the bug.
>>>>>
>>>>> It's pretty clear that MemoryLocation doesn't fit the needs of a lot
>>>>> of stuff anymore (we hacked AA nodes into it, and lots of stuff now tries
>>>>> to figure out the invariantess of the locations, blah blah blah), but it
>>>>> seems like a big job to figure out what to replace it with that will work
>>>>> for these cases.
>>>>>
>>>>> (I'm pretty positive if we just make it MemoryLocations, and have
>>>...
2016 Aug 29
2
GVN / Alias Analysis issue with llvm.masked.scatter/gather intrinsics
this is definitely a bug in AA.
225 for (auto I = CS2.arg_begin(), E = CS2.arg_end(); I != E; ++I) {
226 const Value *Arg = *I;
227 if (!Arg->getType()->isPointerTy())
-> 228 continue;
229 unsigned CS2ArgIdx = std::distance(CS2.arg_begin(), I);
230 auto CS2ArgLoc = MemoryLocation::getForArgument(CS2,
CS2ArgIdx, TLI);