There was a limitation in legality that in the original inner loop latch,
no instruction was allowed between the induction variable increment
and the branch instruction. This is because we used to split the
inner latch at the induction variable increment instruction. Since
now we have split at the inner latch branch instruction and have
properly duplicated instructions over to the split block, we remove
this limitation.
Please refer to the test case updates to see how we now interchange
loops where instructions exist between the induction variable
increment and the branch instruction.
Reviewed By: bmahjour
Differential Revision: https://reviews.llvm.org/D115238
No need to include the order of the scalars beeing used as part of the
alternate vectorization into account when trying to reorder the whole
graph. Such elements better to reorder in the following phase because
the subtree still ends up in shuffle.
Part of D116688, fixes the regression in D116690.
Differential Revision: https://reviews.llvm.org/D116740
There is a bug in the reordering analysis stage. If the element with the
given hash is not added to the map but has the same number of APOs and
instructions with same parent, but different instruction opcode, it will
be initalized with default values and then the counter is increased by
1. But the lane is not updated and default to 0 instead of the actual
`Lane` value. It leads to the fact that the analysis is useless in
many cases and default to lane 0 instead of actual lane with the
minimum amount of APO operands.
Differential Revision: https://reviews.llvm.org/D116690
When determining whether the memory is local to the function (and
we can thus introduce spurious writes without thread-safety issues),
check for a noalias call rather than the hardcoded list of memory
allocation functions. Noalias calls are the more general way to
determine allocation functions, as long as we're only interested
in the property that the returned value is distinct from any other
accessible memory.
Differential Revision: https://reviews.llvm.org/D116728
This patch updates SCEVExpander::expandUnionPredicate to not create
redundant 'or false, x' instructions. While those are trivially
foldable, they can be easily avoided and hinder code that checks the
size/cost of the generated checks before further folds.
I am planning on look into a few other similar improvements to code
generated by SCEVExpander.
I remember a while ago @lebedev.ri working on doing some trivial folds
like that in IRBuilder itself, but there where concerns that such
changes may subtly break existing code.
Reviewed By: reames, lebedev.ri
Differential Revision: https://reviews.llvm.org/D116696
There was a limitation in legality that in the original inner loop latch,
no instruction was allowed between the induction variable increment
and the branch instruction. This is because we used to split the
inner latch at the induction variable increment instruction. Since
now we have split at the inner latch branch instruction and have
properly duplicated instructions over to the split block, we remove
this limitation.
Please refer to the test case updates to see how we now interchange
loops where instructions exist between the induction variable increment
and the branch instruction.
Reviewed By: bmahjour
Differential Revision: https://reviews.llvm.org/D115238
If we have a call whose only side effect is a write to a location which is known to be dead, we can sink said call to the users of the call's result value. This is analogous to the recent changes to delete said calls if unused, but framed as a sinking transform instead.
Differential Revision: https://reviews.llvm.org/D116200
This reverts commit 859ebca744e634dcc89a2294ffa41574f947bd62.
The change contained many unrelated changes and e.g. restored
unit test failes for the old lld port.
This does not appear to cause any problems, and it
fixes#50910
Extra tests with a trunc user were added with:
3a239379
...but they don't match either way, so there's an
opportunity to improve the matching further.
This reverts commit 640beb38e7710b939b3cfb3f4c54accc694b1d30.
That commit caused performance degradtion in Quicksilver test QS:sGPU and a functional test failure in (rocPRIM rocprim.device_segmented_radix_sort).
Reverting until we have a better solution to s_cselect_b64 codegen cleanup
Change-Id: Ibf8e397df94001f248fba609f072088a46abae08
Reviewed By: kzhuravl
Differential Revision: https://reviews.llvm.org/D115960
Change-Id: Id169459ce4dfffa857d5645a0af50b0063ce1105
This was originally added in rG22174f5d5af1eb15b376c6d49e7925cbb7cca6be
although that patch doesn't really mention any reasons for ignoring the
pointer type in this calculation if the memory access isn't consecutive.
Reviewed By: david-arm
Differential Revision: https://reviews.llvm.org/D115356
In particular, this also preserves undef when loading from padding,
rather than converting it to zero through a different codepath.
This is the remaining part of D115924.
There are a number of places that specially handle loads from a
uniform value where all the bits are the same (zero, one, undef,
poison), because we a) don't care about the load offset in that
case b) it bypasses casts that might not be legal generally but
do work with uniform values.
We had multiple implementations of this, with a different set of
supported values each time. This replaces two usages with a more
complete helper. Other usages will be replaced separately, because
they have larger impact.
This is part of D115924.
This reverts commit 29b6e967f3e99ac45340ea37a70262c70e4e7528. The bug it
found in PartiallyInlineLibCalls was fixed in
c8ffc73350dbb6044ca947bbead127b9b914cdf3.
At the moment, the primary induction variable for the vector loop is
created as part of the skeleton creation. This is tied to creating the
vector loop latch outside of VPlan. This prevents from modeling the
*whole* vector loop in VPlan, which in turn is required to model
preheader and exit blocks in VPlan as well.
This patch introduces a new recipe VPCanonicalIVPHIRecipe to represent the
primary IV in VPlan and CanonicalIVIncrement{NUW} opcodes for
VPInstruction to model the increment.
This allows us to partly retire createInductionVariable. At the moment,
a bit of patching up is done after executing all blocks in the plan.
Reviewed By: Ayal
Differential Revision: https://reviews.llvm.org/D113223
This reverts commit ea75be3d9df448b6abafaf752a8141764d93ca33 and
1eb5b6e85045d22720f177a02aaf7097930e4b4f.
That commit caused crashes with compilation e.g. like this
(not fixed by the follow-up commit):
$ cat sqrt.c
float a;
b() { sqrt(a); }
$ clang -target x86_64-linux-gnu -c -O2 sqrt.c
Attributes 'readnone and writeonly' are incompatible!
%sqrtf = tail call float @sqrtf(float %0) #1
in function b
fatal error: error in backend: Broken function found, compilation aborted!
The user scanning loop above looks through pointer casts, so we
also need to strip pointer casts in the capture check. Previously
the source was incorrectly considered not captured if a bitcast
was passed to the call.
Call slot optimization is currently supposed to be prevented if
the call can capture the source pointer. Due to an implementation
bug, this check currently doesn't trigger if a bitcast of the source
pointer is passed instead. I'm somewhat afraid of the fallout of
fixing this bug (due to heavy reliance on call slot optimization
in rust), so I'd like to strengthen the capture reasoning a bit first.
In particular, I believe that the capture is fine as long as a)
the call itself cannot depend on the pointer identity, because
neither dest has been captured before/at nor src before the
call and b) there is no potential use of the captured pointer
before the lifetime of the source alloca ends, either due to
lifetime.end or a return from a function. At that point the
potentially captured pointer becomes dangling.
Differential Revision: https://reviews.llvm.org/D115615
isBitOrNoopPointerCastable() returns true if the types are the
same, but it's not actually possible to create a bitcast for all
such types. The assumption seems to be that the user will omit
creating the cast in that case, as it is unnecessary.
Fixes https://github.com/llvm/llvm-project/issues/52994.
This fixes bug49264.
Simply, coroutine shouldn't be inlined before CoroSplit. And the marker
for pre-splited coroutine is created in CoroEarly pass, which ran after
AlwaysInliner Pass in O0 pipeline. So that the AlwaysInliner couldn't
detect it shouldn't inline a coroutine. So here is the error.
This patch set the presplit attribute in clang and mlir. So the inliner
would always detect the attribute before splitting.
Reviewed By: rjmccall, ezhulenev
Differential Revision: https://reviews.llvm.org/D115790
This builds on the code from D114963, and extends it to handle calls both direct and indirect. With the revised code structure (from series of previously landed NFCs), this is pretty straight forward.
One thing to note is that we can not infer writeonly for arguments which might be captured. If the pointer can be read back by the caller, and then read through, we have no way to track that. This is the same restriction we have for readonly, except that we get no mileage out of the "callee can be readonly" exception since a writeonly param on a readonly function is either a) readnone or b) UB. This means we can't actually infer much unless nocapture has already been inferred.
Differential Revision: https://reviews.llvm.org/D115003
All of these functions would be `readnone`, but can't be on platforms
where they can set `errno`. A `writeonly` function with no pointer
arguments can only write (but never read) global state.
Writeonly theoretically allows these calls to be CSE'd (a writeonly call
with the same arguments will always result in the same global stores) or
hoisted out of loops, but that's not implemented currently.
There are a few functions in this list that could be `readnone` instead
of `writeonly`, if someone is interested.
Differential Revision: https://reviews.llvm.org/D116426
0a00d64 turned an early exit here into an assertion, but the assertion
can be triggered, as PR52920 shows.
The later code is agnostic to the accessed type, so just drop the
assert. The patch also adds tests for LAA directly and
loop-load-elimination to show the behavior is sane.
This folded (null + X) == g to false, but of course this is
incorrect if X == g.
Possibly this got confused with the null == g case, which is
already handled elsewhere.
This is now testing (null + g3) != g3 and still coming up with
"true" as the answer. The original case was a less obvious
miscompile with index overflow involved.
The fold for merging a GEP of GEP into a single GEP currently bails
if doing so would result in notional overindexing. The justification
given in the comment above this check is dangerously incorrect: GEPs
with notional overindexing are perfectly fine, and if some code
treats them incorrectly, then that code is broken, not the GEP.
Such a GEP might legally appear in source IR, so only preventing
its creation cannot be sufficient. (The constant folder also ends
up canonicalizing the GEP to remove the notional overindexing, but
that's neither here nor there.)
This check dates back to
bd4fef4a89,
and as far as I can tell the original issue this was trying to
patch around has since been resolved.
Differential Revision: https://reviews.llvm.org/D116587
This fold is not correct, because indices might evaluate to zero
even if they are not a literal zero integer. Additionally, this
fold would be wrong (in the general case) for non-i8 types as well,
due to index overflow.
Drop this fold and instead let the target-dependent constant
folder compute the actual offset and fold the comparison based
on that.
This fold is incorrect, because it assumes that all indices are
non-zero. This happens to be true for the test as written, but
doesn't hold if we use an extern weak global instead, for which
ptrtoint might be zero.
Add separate tests for the simple constant int case.
