AS(4), when targeting GPUs, is constant. Accesses to constant memory are
(historically) not treated as "memory accesses", hence we should deduce
`memory(none)` for those.
We failed to cache queries without an exclusion set that resulted in
non-reachable results. That is obviously bad as changes to liveness can
influence the result.
Fixes: https://github.com/llvm/llvm-project/issues/61883
Currently, FunctionAttrs treats landingpads as non-throwing, and
will infer nounwind for functions with landingpads (assuming they
can't unwind in some other way, e.g. via resum). There are two
problems with this:
* Non-cleanup landingpads with catch/filter clauses do not
necessarily catch all exceptions. Unless there are catch ptr null
or filter [0 x ptr] zeroinitializer clauses, we should assume
that we may unwind past this landingpad. This seems like an
outright bug.
* Cleanup landingpads are skipped during phase one unwinding, so
we effectively need to support unwinding past them. Marking these
nounwind is technically correct, but not compatible with how
unwinding works in reality.
Fixes https://github.com/llvm/llvm-project/issues/61945.
Differential Revision: https://reviews.llvm.org/D147694
This patch adds the AANonConvergent abstract attribute. It removes the
convergent attribute from functions that only call non-convergent
functions.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D143228
If the stack is not accessible by other threads, e.g., on a GPU, we need
to ensure heap-2-stack will not create a stack version of a pointer that
might be passed to another thread. Since passing through memory is by
default transparent, we need to register a callback and inspect stores
we might look through explicitly.
This is a two part fix. First, we need two Execution Domains (ED) to
track the values of a function. One for incoming values and one for
outgoing values. This was conflated before. Second, at the function
entry we need to look at the incoming information from call sites not
iterate over non-existing predecessors.
Similar to loads, PHIs can be used to introduce non-dynamically unique
values into the simplification "algorithm". We need to check that PHIs
do not carry such a value from one iteration into the next as can cause
downstream reasoning to fail, e.g., downstream could think a comparison
is equal because the simplified values are equal while they are defined
in different loop iterations. Similarly, instructions in cycles are now
conservatively treated as non-dynamically unique. We could do better but
I'll leave that for the future.
The change in AAUnderlyingObjects allows us to ignore dynamically unique
when we simply look for underlying objects. The user of that AA should
be aware that the result might not be a dynamically unique value.
If a query uses an exclusion set but we haven't used it to determine the
result, we can cache the query without exclusion set too. When we lookup
a cached result we can check for the non-exclusion set version first.
We used to check the query instructions for effects but that does not
work well with complex accesses we will probably support in the future.
Now we simply let the user decide what accesses to look for.
Before we might have ended up queriying the AAExecutionDomain of a
different function, which resulted in wrong optimistic results.
Partially fixes https://github.com/llvm/llvm-project/issues/60425
We improved our simplification and this exposed a bug in the store
elimination. A load that had dead uses and assume uses was thought to be
used by assumes only. Consequently we also deleted the "dead use users".
This was a problem because a dead use just means we will not use the
load there. The user might still be needed.
Exposed by OvO, reported by @ye-luo.
If an instruction is executed in an aligned region we can ignore
threading effects and use CFG reasoning (dominance and reachability).
This is true because all threads are together in an aligned region and
there cannot be one waiting for a signal at a place not connected via
the control flow.
More dedicated tests will follow.
More details can be found here:
"Co-Designing an OpenMP GPU Runtime and Optimizations for Near-Zero
Overhead Execution", IPDPS 2022,
https://www.osti.gov/servlets/purl/1890094
If we modified the IR during manifest, e.g., SPMDzation, we might end up
with un-cached reachability queries. This is not good as the result is
going to be optimistic. We now cache the updateImpl result and use it
during manifest.
Bug was exposed in a follow up extension.
While full support requires more work (see TODOs), this allows us to
handle vector writes with a single constant value properly. For now,
we can handle the same constant values stored to all elements if
everything is of a fixed size.
Even if we have multiple access ranges, the access can be exact. It is
not a MUST access but that is taken care of elsewhere. The tests were
wrong as they contained uninitialized memory. When the memory is
initialized it works as expected.
Even if a barrier does not enforce aligned execution, it will
effectively be like an aligned barrier if it is executed by all threads
in an aligned way. We lack control flow divergence analysis here so we
can only do (basic block) local reasoning for now.
With this patch we track aligned barriers in AAExecutionDomain and also
delete unnecessary barriers there. This allows us to eliminate barriers
across blocks, across functions, and in the presence of complex accesses
that do not force a barrier. Further, we can use the collected
information to enable store-load forwarding in a threaded environment
(follow up patch).
Differential Revision: https://reviews.llvm.org/D140463
A `ConstantExpr` ICmp is pretty much the same thing as an ICmpInst when
we want to simplify it. We just need to be less restrictive wrt. the
type and use the static helper functions directly.
Fixes: https://github.com/llvm/llvm-project/issues/59767
We use the actual access (=remote) instruction when reasoning about
accesses, except for one leftover use case of the local instruction.
This caused us to potentially ignore the dominating write if the read
and write were in a different function than the (stack) allocation.
Reported by @ye-luo
This resolves a recent regression introduced by a bug fix and allows us
to use dominating write information (formerly HasBeenWrittenTo
information) to skip potential interfering accesses.
Generally, there are two changes here:
1) If we have dominating writes they form a chain and we can look at the
least one to minimize the distance between the write and the (read)
access in question.
2) If such a least dominating write exists, we can ignore writes in
other functions as long as they cannot be reached from code between
this write and the (read) access in question.
We have all the tools available to make such queries and the positive
tests show the result. Note that the negative test from the bug fix is
still in tree and not affected.
As a side-effect, we can remove the (arbitrary) treshold now on the
number of interfering accesses since we do not iterate over dominating
ones anymore.
Recent commit introducing AA for getting underluying objects of a
pointer created an uninitialized boolean, which causes tests to fail
when built unter asan/ubsan. This initialized that variable.
This patch introduces a new AA `AAUnderlyingObjects`. It is basically like a wrapper
AA of the function `AA::getAssumedUnderlyingObjects`, but it can recursively do
query if the underlying object is an indirect access, such as a phi node or a select
instruction.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D141164
HandlePassthroughUser may sometimes create a new entry for the OffsetInfo of a
user in the OffsetInfoMap. This can invalidate outstanding references into the
map, including the one which needs to be copied into the new entry. This
produces invalid offset info that can trigger assertions.
Fixed this by not using references at this point. The bug was originally
introduced in commit ID 0dc0a44132.
Reviewed By: ronlieb
Differential Revision: https://reviews.llvm.org/D140837
AAPotentialConstantValues now works for PHI and Load by simply examinig
AAPotentialValues for the instruction itself.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D140371
Previously reverted in 8b446ea2ba
Reapplying because this commit is NOT DEPENDENT on the reverted commit
fc21f2d7ba, which broke the ASAN buildbot.
See https://reviews.llvm.org/rGfc21f2d7bae2e0be630470cc7ca9323ed5859892 for
more information.
The arguments to a PHI may represent a recurrence by eventually using the output
of the PHI itself. This is now handled by checking for cycles in the control
flow. If a PHI is not in a recurrence, it is now able to report multiple offsets
instead of conservatively reporting unknown.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D138991
value() has undesired exception checking semantics and calls
__throw_bad_optional_access in libc++. Moreover, the API is unavailable without
_LIBCPP_NO_EXCEPTIONS on older Mach-O platforms (see
_LIBCPP_AVAILABILITY_BAD_OPTIONAL_ACCESS).
This fixes check-clang-tools.
