We can only fold insertvalue undef, (extractvalue x, n) to x
if x is not poison, otherwise we might be replacing undef with
poison (https://alive2.llvm.org/ce/z/fnw3c8). The insertvalue
poison case is always fine.
I didn't go to particularly large effort to preserve cases where
folding with undef is still legal (mainly when there is a chain of
multiple inserts that end up covering the whole aggregate),
because this shouldn't really occur in practice: We should always
be generating the insertvalue poison form when constructing
aggregates nowadays.
Differential Revision: https://reviews.llvm.org/D144106
Reapplies 142c28ffa1323e9a8d53200a22c80d5d778e0d0f as part of D140242 which got reverted due to amdgpu openmp test failures.
This diff fixes said failures by eliding most of `adjustInliningThresholdUsingCallee` for indirect calls as the callee function is unavailable for indirect calls.
Reviewed By: arsenm, #amdgpu
Differential Revision: https://reviews.llvm.org/D143498
This seems to cause large regressions in existing code, as much as 75% slower
(4x the time taken). Small always inline functions seem to be used a lot in the
cmsis-dsp library.
I would add a phase ordering test to show the problems, but one already exists!
The llvm/test/Transforms/PhaseOrdering/ARM/arm_mult_q15.ll was just changed by
removing alwaysinline to hide the problems that existed.
This reverts commit cae033dcf227aeecf58fca5af6fc7fde1fd2fb4f.
This reverts commit 8e33c41e72ad42e4c27f8cbc3ad2e02b169637a1.
We have several situations where it's beneficial for code size to ensure that every
call to always-inline functions are inlined before normal inlining decisions are
made. While the normal inliner runs in a "MandatoryOnly" mode to try to do this,
it only does it on a per-SCC basis, rather than the whole module. Ensuring that
all mandatory inlinings are done before any heuristic based decisions are made
just makes sense.
Despite being referred to the "legacy" AlwaysInliner pass, it's already necessary
for -O0 because the CGSCC inliner is too expensive in compile time to run at -O0.
This also fixes an exponential compile time blow up in
https://github.com/llvm/llvm-project/issues/59126
Differential Revision: https://reviews.llvm.org/D143624
This reverts commit a772f0bb920a4957fb94dd8dbe45943809fd0ec3.
The main problem was related to how we handled `dbgs()` from the hosted
compiler. Using explicit `subprocess.communicate`, and not relying on
dbgs() being flushed until the end appears to address the problem.
Also some fixes due to some bots running older pythons, so we can't have
nice things like `int | float` and such.
This reverts commit a7354899d1a235a796b3a2ccb45f6596983c8672.
The way stdout/stderr get routed seems to work differently locally and
on the bots. Investigating.
This hooks up the interactive model runner to the passes that support
ml-based decisions. Because the interface to this runner is the exact
same as the one used during inference, we just reuse the exact same
setup we have for "release mode". This makes "release mode" a misnomer -
and that's something we needed to resolve sooner or later (e.g.
supporting more than one embedded model for the same problem was another
reason to drop that nomenclature). That will happen in a subsequent
change.
To use this evaluator, just enable the pass in (currently) "release"
mode, but also pass the base name for the 2 channel files via the
pass-specific flag.
The 2 files are the responsibilty of the hosting process. The added
tests use a minimal, toy such host, illustrating setup and
communication.
Differential Revision: https://reviews.llvm.org/D143218
A regression from when new PM got enabled as default. Functions with a big number of instructions will elide getting inlined but do not consider the cost of passing arguments over stack if there are a lot of function arguments. This patch attempts to add a heuristic for AMDGPU's function calling convention that also considers function arguments passed through the stack.
Reviewed By: #amdgpu, arsenm
Differential Revision: https://reviews.llvm.org/D140242
IR is now always parsed in opaque pointer mode, unless
-opaque-pointers=0 is explicitly given. There is no automatic
detection of typed pointers anymore.
The -opaque-pointers=0 option is added to any remaining IR tests
that haven't been migrated yet.
Differential Revision: https://reviews.llvm.org/D141912
The dependency was due to the log format. This change switches to the
previously-introduced (D139370) "dependency-free" logger instead of the
protobuf-based one.
A subsequent change will clean out the unnecessary abstraction left
behind.
This change drops the logger unittest, we have sufficient test coverage
via lit tests, and a unit test would require adding, unnecesarily, a log
reader (the reader is expected to be python, for the ML side, and there
is a reader for that under Analysis/models, used for tests).
Differential Revision: https://reviews.llvm.org/D141720
Non-throwing inline asm infers the nounwind attribute in
instcombine. Thus, it can be handled in the same manner as
non-throwing target functions are generally. Further special casing is
unnecessary complexity.
It isn't correct to always remove memprof metadata MIBs from the
original allocation call after inlining.
Let's say we have the following partial call graph:
C D
\ /
v v
B E
| /
v v
A
where A contains an allocation call. If both contexts including B have
the same allocation behavior, the context in the memprof metadata on the
allocation will be pruned, and we will have 2 MIBs with contexts:
A,B and A,E.
Previously, if we inlined A into B we propagate the matching MIBs onto
the inlined allocation call in B' (A,B in this case), and remove it from
the original out of line allocation in A. This is correct if we have a
single round of bottom up inlining.
However, in the compiler we can have multiple invocations of the inliner
pass (e.g. LTO). We may also inline non-bottom up with an alternative
inliner such as the ModuleInliner. In that case, we could end up first
inlining B into C, without having inlined A into B. The call graph then
looks like:
D
|
v
C' B E
\ | /
v v v
A
If we subsequently (perhaps on a later invocation of bottom up inlining)
inline A into B, the previous handling would propagate the memprof MIB
context A,B up into the inlined allocation in B', and remove it from the
original allocation in A. The propagation into B' is fine, however, by
removing it from A's allocation, we no longer reflect the context coming
from C'.
To fix this, simply prevent the removal of MIB from the original
allocation callsites.
Note that the memprof_inline.ll test has some changes to existing
checking to replace "noncold" with "notcold" in the metadata. The
corresponding CHECK was accidentally commented out in the old version
and thus this mistake was not previously detected.
Differential Revision: https://reviews.llvm.org/D140764
We are in the process of retiring LLVM_HAVE_TF_API in favor of
LLVM_HAVE_TFLITE. This patch takes care of the transition in
llvm/test.
Differential Revision: https://reviews.llvm.org/D140133
A few more that I missed in commit 3528e63d89305907b3d6e.
There could be more duplicates remaining, since I've only focused
on exactly duplicated "RUN: opt" lines (ignoring multi line RUN
lines ending with '\').
This is the next step in dropping the dependency on protobuf.
The simple logger produces an output consisting of lines of json
strings. Tensor values - which should constitute the bulk of the data -
are serialized as raw byte buffers. This allows for light-weight reading
of the values.
The next step is to switch the training logic to the new logging format,
following which the protobuf-based logger will be dropped, together with
the training dependency on protobuf.
Subsequent changes will also stop buffering and stream, instead - the
buffering model is just as a convenient point-in-time.
Differential Revision: https://reviews.llvm.org/D139370
This patch adds MLPriority as the first step toward the ML-based
function inlining with the module inliner.
For now, MLPriority is completely identical to CostPriority.
Once this patch lands, I'm planning to:
- integrate NoInferenceModelRunner,
- memoize the priority computation so that the priority remains the
same for given values of metrics even with the noise injected during
training, and
- port/take more features into account.
Differential Revision: https://reviews.llvm.org/D139140
The new test ensures that the module inliner works with all currently
supported priority modes.
Different priority modes result in no difference in terms of the
output for these simple cases, so this is more of a "better than
nothing" test.
Differential Revision: https://reviews.llvm.org/D139222
Fixes https://github.com/llvm/llvm-project/issues/56544
AsmWriter always writes ", ..." when a tail call has a varargs argument. This patch only writes the ", " when there is an argument before the varargs argument.
I did not write a dedicated test this for this change, but I modified an existing test that will test for a regression.
Reviewed By: avogelsgesang
Differential Revision: https://reviews.llvm.org/D137893
Signed-off-by: Adrian Vogelsgesang <avogelsgesang@salesforce.com>
This switches everything to use the memory attribute proposed in
https://discourse.llvm.org/t/rfc-unify-memory-effect-attributes/65579.
The old argmemonly, inaccessiblememonly and inaccessiblemem_or_argmemonly
attributes are dropped. The readnone, readonly and writeonly attributes
are restricted to parameters only.
The old attributes are auto-upgraded both in bitcode and IR.
The bitcode upgrade is a policy requirement that has to be retained
indefinitely. The IR upgrade is mainly there so it's not necessary
to update all tests using memory attributes in this patch, which
is already large enough. We could drop that part after migrating
tests, or retain it longer term, to make it easier to import IR
from older LLVM versions.
High-level Function/CallBase APIs like doesNotAccessMemory() or
setDoesNotAccessMemory() are mapped transparently to the memory
attribute. Code that directly manipulates attributes (e.g. via
AttributeList) on the other hand needs to switch to working with
the memory attribute instead.
Differential Revision: https://reviews.llvm.org/D135780
This change is updating remaining Inline tests
by removing -enable-new-pm=0 flag and adjusting CHECKs
where it is required.
Reviewed By: aeubanks
Differential Revision: https://reviews.llvm.org/D135497
Update both memprof and callsite metadata to reflect inlined functions.
For callsite metadata this is simply a concatenation of each cloned
call's call stack with that of the inlined callsite's.
For memprof metadata, each profiled memory info block (MIB) is either
moved to the cloned allocation call or left on the original allocation
call depending on whether its context matches the newly refined call
stack context on the cloned call. We also reapply context trimming
optimizations based on the refined set of contexts on each of the calls
(cloned and original).
Depends on D128142.
Reviewed By: snehasish
Differential Revision: https://reviews.llvm.org/D128143
This reverts commit 0d7f3464ce0ba3a97df73e08ee0acd4e33adbe9b and
commit f9403ca41e5f3dab60cd6e5de26eea65dcab01a4. The latter was
"Profile matching and IR annotation for memprof profiles." and was left
from a bad rebase from a commit already pushed upstream.
Update both memprof and callsite metadata to reflect inlined functions.
For callsite metadata this is simply a concatenation of each cloned
call's call stack with that of the inlined callsite's.
For memprof metadata, each profiled memory info block (MIB) is either
moved to the cloned allocation call or left on the original allocation
call depending on whether its context matches the newly refined call
stack context on the cloned call. We also reapply context trimming
optimizations based on the refined set of contexts on each of the calls
(cloned and original), via utilities in MemoryProfileInfo.
Depends on D128142.
Differential Revision: https://reviews.llvm.org/D128143
Second patch in the series to remove legacy PM and
associated -enable-new-pm=0 flag targets pass that
has not been ported to new PM - PruneEH.
Discussion about this can be found in D44415.
Reviewed By: aeubanks
Differential Revision: https://reviews.llvm.org/D134686
Inlining must be disabled when the call-site needs to toggle PSTATE.SM or
when the callee's function body is executed in a different streaming mode than
its caller. This is needed because function calls are the boundaries for
streaming mode changes.
More details about the SME attributes and design can be found
in D131562.
Differential Revision: https://reviews.llvm.org/D131581
This is the first patch in a series intended for removing flag
-enable-new-pm=0 from lit tests. This is part of a bigger
effort of completely removing legacy code related to legacy
pass manager in favor of currently default new pass manager.
In this patch flag has been removed only from tests where no significant
change has been required because checks has been duplicated for
both PMs.
Reviewed By: fhahn
Differential Revision: https://reviews.llvm.org/D134150
TLite is a lightweight, statically linkable[1], model evaluator, supporting a
subset of what the full tensorflow library does, sufficient for the
types of scenarios we envision having. It is also faster.
We still use saved models as "source of truth" - 'release' mode's AOT
starts from a saved model; and the ML training side operates in terms of
saved models.
Using TFLite solves the following problems compared to using the full TF
C API:
- a compiler-friendly implementation for runtime-loadable (as opposed
to AOT-embedded) models: it's statically linked; it can be built via
cmake;
- solves an issue we had when building the compiler with both AOT and
full TF C API support, whereby, due to a packaging issue on the TF
side, we needed to have the pip package and the TF C API library at
the same version. We have no such constraints now.
The main liability is it supporting a subset of what the full TF
framework does. We do not expect that to cause an issue, but should that
be the case, we can always revert back to using the full framework
(after also figuring out a way to address the problems that motivated
the move to TFLite).
Details:
This change switches the development mode to TFLite. Models are still
expected to be placed in a directory - i.e. the parameters to clang
don't change; what changes is the directory content: we still need
an `output_spec.json` file; but instead of the saved_model protobuf and
the `variables` directory, we now just have one file, `model.tflite`.
The change includes a utility showing how to take a saved model and
convert it to TFLite, which it uses for testing.
The full TF implementation can still be built (not side-by-side). We
intend to remove it shortly, after patching downstream dependencies. The
build behavior, however, prioritizes TFLite - i.e. trying to enable both
full TF C API and TFLite will just pick TFLite.
[1] thanks to @petrhosek's changes to TFLite's cmake support and its deps!
