Preserve the alignment of the original atomicrmw, rather than using
the ABI alignment.
The same problem exists for loads, but that code is being removed
in D141277 anyway.
Follow-up to:
6c39a3aae1dc
That converted a pattern with ashr directly to icmp+zext, and
this updates the pattern that we used to convert to.
This canonicalizes to icmp for better analysis in the minimum case
and shortens patterns where the source type is not the same as dest type:
https://alive2.llvm.org/ce/z/tpXJ64https://alive2.llvm.org/ce/z/dQ405O
This requires an adjustment to an icmp transform to avoid infinite looping.
We shouldn't penalize instructions that have extra flags.
Drop the poison-generating flags if needed instead of bailing out.
This makes canonicalization/optimization more uniform.
There is a chance that dropping flags will cause some
other transform to not fire, but we added a preliminary
patch to avoid that with:
f0faea571403
See D140665 for more details.
These involved regenerating test checks. There are two significant
differences here:
1. With typed pointers we sometimes swapped gep and addrspacecast,
as a side-effect of other transforms. However, the current
direction is likely undesirable, and we should canonicalize in
the reverse direction instead (gep of ac, instead of ac of gep).
This should be done after typed pointers are removed, to avoid
conflicting transforms.
2. The "descaling" optimization isn't really compatible with
opaque pointers. This will be addressed longer-term by moving
away from type-based GEP, at which point the form with explicit
multiplications will be the canonical one.
~sext(A) & Op1 --> A ? 0 : Op1
With no extra uses, this pattern is already reduced,
but we would miss it in examples such as issue #59773.
https://alive2.llvm.org/ce/z/WGLcSR
EmitGEPOffset() supports vector GEPs nowadays, so we don't need
any further code changes.
compare_gep_with_base_vector1 shows a weakness in folding the
resulting comparison if an index splat has to be performed.
We were already treating branch on poison as UB, but branch on
undef is also UB. Move the checks into the correct function.
From LangRef for br:
> If ‘cond’ is poison or undef, this instruction has undefined behavior.
From LangRef for switch:
> If ‘value’ is poison or undef, this instruction has undefined behavior.
There is a minor regression in dont-distribute-phi.ll, apparently
we handle that pattern in logical but not bitwise form.
As discussed in issue #59279, we want fneg/fabs to conform to the
IEEE-754 spec for signbit operations - quoting from section 5.5.1
of IEEE-754-2008:
"negate(x) copies a floating-point operand x to a destination in
the same format, reversing the sign bit"
"abs(x) copies a floating-point operand x to a destination in the
same format, setting the sign bit to 0 (positive)"
"The operations treat floating-point numbers and NaNs alike."
So we gate this transform with "nnan" in addition to "nsz":
(X > 0.0) ? X : -X --> fabs(X)
Without that restriction, we could have for example:
(+NaN > 0.0) ? +NaN : -NaN --> -NaN
(because an ordered compare with NaN is always false)
That would be different than fabs(+NaN) --> +NaN.
More fabs/fneg patterns demonstrated here:
https://godbolt.org/z/h8ecc659d
(without any FMF, these are correct independently of this patch -
no fabs should be created)
The code change is a one-liner, but we have lots of tests diffs
because there are many variations of the basic pattern.
Differential Revision: https://reviews.llvm.org/D139785
This reverts commit 66cea84681e16f3d4ebdc69031824b114a0d5681.
I did not intend to commit all the changes in here, but only the
ones with no significant differences.
I left a few of the existing undef tests in place for extra
coverage and because one of those was noted as a miscompile in
the post-commit feedback for d4493dd1ed58ac3f1eab0, but we are
transitioning to poison, so it is more valuable to test the
expected IR going forward.
