Commit c2e62c7 updated the ReduceDIMetadata pass to be able to remove
DIGlobalVariableExpressions from MDNode operands; it also accidentally
prevented null operands from being preserved, which results in an
assertion being triggered:
`Targets == NoChunksCounter.count() && "number of chunks changes when
reducing"'
This patch allows us to correctly preserve null operands once again.
I've not got a test case for this yet - I'm hoping this patch is just
trivially correct as-is, because I've not got the hang of reducing a
test case for llvm-reduce yet, but I can get a test case generated if
needed.
The "preserve input debug-info format" flag allowed some tooling to opt
into not seeing the new debug records yet, and to not autoupgrade. This
was good at the time, but un-necessary now that we'll be ditching
intrinsics shortly.
It also hides errors now: verify-uselistorder was hardcoding this flag
to on, and as a result it hasn't seen debug records before. Thus, we
missed a uselistorder variation: constant-expressions such as GEPs can
be contained within debug records and completely isolated from the value
hierachy, see the metadata-use-uselistorder.ll test. These Values didn't
get ordered, but were legitimate uses of constants like "i64 0", and we
now run into difficulty handling that. The patch to AsmWriter seeks
Values to order even through debug-info now.
Finally there are a few intrinsics-tests relying on this flag that we
can just delete, such as one in llvm-reduce and another few in the
LocalTest unit tests. For the fast-isel test, it was added in
https://reviews.llvm.org/D67703 explicitly for checking the size of
blocks without debug-info and in 1525abb9c94 the codepath it tests moved
towards being sunsetted. It'll be totally redundant once RemoveDIs is on
permanently.
Note that there's now no explicit test for the textual-IR autoupgrade
path. I submit that we can rely on the thousands of .ll files where
we've only been bothered to update the outputs, not the inputs, to debug
records.
After replacing the branch condition, this was calling simplifyCFG to
perform the cleanups of the branch. This is far too heavy of a hammer.
We do not want all of the extra optimizations in simplifyCFG, and
this could also leave behind dead code. Instead, minimally fold the
terminator and try to delete the newly dead code.
This is pretty much a direct copy of what bugpoint does.
This was disabled due to flakiness but I'm currently unable to
reproduce.
I'm nervous the original issue still exists. However, I downgraded the
tripped
assert in 8c18c25b1b22ea710edb40a4f167a6a8bfe6ff9d to a warning since
the same
assert can trigger for illegitimate reasons.
Fixes#64157
This wasn't checking the output for all functions.
--match-full-lines is also particularly hazardous for the
interestingness checks for avoiding asserts and broken IR.
Also add tests for some of the filtered function user types.
This wasn't covered, and is overly conservative.
If we are trying to simplify branch conditions to true, ignore branches
already set to a constant true. If we are simplifying to constant false,
ignore the already constant false cases. This saves steps in this edge
case, and avoids the side effect of running simplifycfg on blocks we
did not intend to modify.
The IR verifier will fail if there are any convergent calls without
a convergencectrl bundle, if there are any convergencectrl bundles.
With the current verifier rules, we would need to drop all the instances
of convergencectrl in the function as a set, and strip all the
convergence
token intrinsics. As such, I think it would be more appropriate to have
a
separate convergence reduction pass.
Try to reduce individual subtarget features in the "target-features"
attribute. This attempts a textual removal of the fields in the string,
not a semantic removal. Typically there's a lot of redundant feature spam
in the feature list implied by the target-cpu (which I really wish clang
would stop emitting). If we could parse these out, we could easily drop
the fields without testing anything.
The current API doesn't have a way to unset it. The query returns
an optional, but the set doesn't. Alternatively I could switch the
set to also use optional.
Not sure this is the right place to put it. This is a property
of GlobalVariable, not GlobalValue. But the ReduceGlobalVars
reduction tries to delete the value entirely.
During the transition from debug intrinsics to debug records, we used
several different command line options to customise handling: the
printing of debug records to bitcode and textual could be independent of
how the debug-info was represented inside a module, whether the
autoupgrader ran could be customised. This was all valuable during
development, but now that totally removing debug intrinsics is coming
up, this patch removes those options in favour of a single flag
(experimental-debuginfo-iterators), which enables autoupgrade, in-memory
debug records, and debug record printing to bitcode and textual IR.
We need to do this ahead of removing the
experimental-debuginfo-iterators flag, to reduce the amount of
test-juggling that happens at that time.
There are quite a number of weird test behaviours related to this --
some of which I simply delete in this commit. Things like
print-non-instruction-debug-info.ll , the test suite now checks for
debug records in all tests, and we don't want to check we can print as
intrinsics. Or the update_test_checks tests -- these are duplicated with
write-experimental-debuginfo=false to ensure file writing for intrinsics
is correct, but that's something we're imminently going to delete.
A short survey of curious test changes:
* free-intrinsics.ll: we don't need to test that debug-info is a zero
cost intrinsic, because we won't be using intrinsics in the future.
* undef-dbg-val.ll: apparently we pinned this to non-RemoveDIs in-memory
mode while we sorted something out; it works now either way.
* salvage-cast-debug-info.ll: was testing intrinsics-in-memory get
salvaged, isn't necessary now
* localize-constexpr-debuginfo.ll: was producing "dead metadata"
intrinsics for optimised-out variable values, dbg-records takes the
(correct) representation of poison/undef as an operand. Looks like we
didn't update this in the past to avoid spurious test differences.
* Transforms/Scalarizer/dbginfo.ll: this test was explicitly testing
that debug-info affected codegen, and we deferred updating the tests
until now. This is just one of those silent gnochange issues that get
fixed by RemoveDIs.
Finally: I've added a bitcode test, dbg-intrinsics-autoupgrade.ll.bc,
that checks we can autoupgrade debug intrinsics that are in bitcode into
the new debug records.
Avoid capitalized Error. This loses the "Error constructing pass pipeline"
part, and just forwards the error to the default report_fatal_error case. Not
sure if it's worth trying to keep.
The coding standards states that error messages should start with
a lowercase. Also use WithColor, and add missing test coverage for
the failed to write to output file case.
If the interestingness script is flaky, we should not assert. Print
a warning, and continue. This could still happen as a result of an
llvm-reduce bug, so make a note of that.
Add a --skip-verify-interesting-after-counting-chunks option to
avoid the extra run of the reduction script, and to silence the
warning.
In undefined mismatch cases, this was fixing the callsite to use the calling
convention of the new function. Preserve the original wrong callsite's calling
convention.
The attribute APIs make this cumbersome. There seem to be missing
overloads using AttrBuilder for the function attrs. Plus there doesn't
seem to be a direct way to set the function attrs on the call.
So far I've been unsuccessful in finding an example where the used constant
value is directly observed in the output. This avoids an assert in an intermediate
step of value replacement.
Some distinct metadata nodes, e.g DICompileUnit, have implicit nullptrs
inside them. Iterating over them with dyn_cast leads to a crash, change
the behavior so that the nullptr operands are skipped.
Add the test distinct-metadata-nullptr.ll which will crash if null
pointers are not handled correctly.
As defined in LangRef, branching on `undef` is undefined behavior.
This PR aims to remove undefined behavior from tests. As UB tests break
Alive2 and may be the root cause of breaking future optimizations.
Here's an Alive2 proof for one of the examples:
https://alive2.llvm.org/ce/z/TncxhP
[MIR] Serialize virtual register flags
This introduces target-specific vreg flag serialization. Flags are represented as `uint8_t` and the `TargetRegisterInfo` override provides methods `getVRegFlagValue` to deserialize and `getVRegFlagsOfReg` to serialize.
This is the minimal change to avoid the assert. There's an API flaw in
invoke instructions where getLandingPad assumes all invoke unwind
blocks have landingpads, when some have catchswitch instead.
Fixes#111817
Following the addition of the llvm.fake.use intrinsic and corresponding
MIR instruction, two further changes are planned: to add an
-fextend-lifetimes flag to Clang that emits these intrinsics, and to
have -Og enable this flag by default. Currently, some logic for handling
fake uses is gated by the optdebug attribute, which is intended to be
switched on by -fextend-lifetimes (and by extension -Og later on).
However, the decision was made that a general optdebug attribute should
be incompatible with other opt_ attributes (e.g. optsize, optnone),
since they all express different intents for how to optimize the
program. We would still like to allow -fextend-lifetimes with optsize
however (i.e. -Os -fextend-lifetimes should be legal), since it may be a
useful configuration and there is no technical reason to not allow it.
This patch resolves this by tracking MachineFunctions that have fake
uses, allowing us to run passes that interact with them and skip passes
that clash with them.
Allow setting the computed properties IsSSA, NoPHIs, NoVRegs for MIR
functions in MIR input. The default value is still the computed value.
If the property is set to false, the computed result is ignored. Conflicting
values (e.g. setting IsSSA where the input MIR is clearly not SSA) lead to
an error.
Closes#37787
