SplitBlockPredecessors is unable to insert an additional BasicBlock
between an indirectbr/callbr terminator and the successor blocks.
This is needed by Polly to normalize the control flow before emitting
its optimzed code.
This patches rejects regions entered by an indirectbr/callbr to not fail
later at code generation.
This fixes llvm.org/PR51964
Inline assembly was not handled at all and treated like a llvm::Value.
In particular, it tried to create a pointer it which is not allowed.
Fix by handling like a llvm::Constant such that it is just reused when
required, instead of trying to marshall it in memory.
Fixes llvm.org/PR51960
VirtualUse ensures consistency over different source of values with
Polly. In particular, this enables its use of instructions moved between
Statement. Before the patch, the code wrongly assumed that the BB's
instructions are also the ScopStmt's instructions. Reference are
determined for OpenMP outlining and GPGPU kernel extraction.
GPGPU CodeGen had some problems. For one, it generated GPU kernel
parameters for constants. Second, it emitted GPU-side invariant loads
which have already been loaded by the host. This has been partially
fixed, it still generates a store for the invariant load result, but
using the value that the host has already written.
WARNING: I did not test the generated PollyACC code on an actual GPU.
The improved consistency will be made use of in the next patch.
The function was intended to catch OpenMP functions such as
get_thread_id(). If matched, the call would be considered synthesizable.
There were a few problems with this:
* get_thread_id() is not 'const' in the sense of have the gcc manual
defines it: "do not examine any values except their arguments".
get_thread_id() reads OpenCL runtime libreary global state.
What was inteded was probably 'speculable'.
* isConstCall was implemented using mayReadOrWriteMemory(). 'const' is
stricter than that, mayReadOrWriteMemory is e.g. true for malloc(),
since it may only read/write addresses that are considered
inaccessible fro the application. However, malloc is certainly not
speculable.
* Values that are isConstCall were not handled consistently throughout
Polly. In particular, it was not considered for referenced values
(OpenMP outlining and PollyACC).
Fix by removing special handling for isConstCall entirely.
This is a simple version without the possibility to define distribute
points or followup-transformations. However, it is the first
transformation that has to check whether the transformation is correct.
It interprets the same metadata as the LoopDistribute pass.
Re-apply after revert in c7bcd72a38 with
fix: Take isBand out of #ifndef NDEBUG since it now is used
unconditionally.
The name of the option is misleading and has been renamed by isl to
"serialize-sccs". Instead of also renaming the option, remove it.
The option is still accessible using
-polly-isl-arg=--no-schedule-serialize-sccs
This is a simple version without the possibility to define distribute
points or followup-transformations. However, it is the first
transformation that has to check whether the transformation is correct.
It interprets the same metadata as the LoopDistribute pass.
This metadata was intended to mark all accesses within an iteration to be pairwise non-aliasing, in this case because every memory of a base pointer is touched (read or write) at most once. This is typical for 'sweeps' over all data. The stated motivation from D30606 is to ensure that unrolled iterations are considered non-aliasing.
Rhe implemention had multiple issues:
* The structure of the noalias metadata was malformed. D110026 added check in the verifier for this metadata, and the tests were failing since then.
* This is not true for the outer loops of the BLIS matrix multiplication, where it was being inserted. Each element of A, B, C is accessed multiple times, as often as the loop not used as an index is iterating.
* Scopes were added to SecondLevelOtherAliasScopeList (used for the !noalias scop list) on-the-fly when another SCEV was seen. This meant that previously visited instructions would not be updated with alias scopes that are only seen later, missing out those SCEVs they should not be aliasing with.
* Since the !noalias scope list would ideally consists of all other SCEV for this base pointer, we might run quickly into scalability issues. Especially after unrolling there would probably at least once SCEV per instruction and unroll instance.
* The inter-iteration noalias base pointer was not removed after leaving the loop marked with it, effectively marking everything after it to noalias as well.
A solution I considered was to mark each instruction as non-aliasing with its own scope. The instruction itself would obviously alias itself, but such construction might also be considered invalid. Duplicating the instruction (e.g. due to speculation) would mark the instruction non-aliasing with its clone. I don't want to go into this territory, especially since the original motivation of determining unrolled instances as noalias based on SCEV is the what scev-aa does as well.
This effectively reverts D30606 and D35761.
Polly does not use the count program itself, but somewhere in lit it is
expected to exists. Otherwise, the following error occurs:
llvm-lit: llvm-project/llvm/utils/lit/lit/llvm/subst.py:133: fatal: Did not find count in ./bin
Code outside the SCoP will be executed recardless of the code versioning
runtime check introduced by CodeGeneration. Assumption made based on
that these are never executed in Polly-optimized code does not hold.
This fixes the miscompilation of MultiSource/Applications/lambda-0.1.3
The new pass manager does not allow adding module passes at the
-polly-position=before-vectorizer extension point. Introduce a
DumpFunctionPass that dumps only current function. In contrast to the
legacy pass manager's -polly-dump-before, each function will be dumped
into its own file. -polly-dump-before-file is still not supported.
The DumpFunctionPass uses llvm::CloneModule to copy the current function
into a new module and then write it into a file.
This is needed for having the functions isl_{set,map}_n_basic_{set,map}
exported to the C++ interface.
Some tests have been modified to reflect the isl changes.
Commit 4c7f820b2b changed the llvm.powi intrinsic to support
different 'int' sizes for the exponent. That happened to break
the IntrinsicToLibdeviceFunc mapping in PPCGCodeGeneration, which
obviously should have been updated as part of commit 4c7f820b2b
(https://reviews.llvm.org/D99439).
The shortcoming was found by buildbots that use
-DPOLLY_ENABLE_GPGPU_CODEGEN=ON
This patch should fixup the problem.
When we're remapping an AddRec, the AddRec constructed by a partial
rewrite might not make sense. This triggers an assertion complaining
it's not loop-invariant.
Instead of constructing the partially rewritten AddRec, just skip
straight to calling evaluateAtIteration.
Testcase was automatically reduced using llvm-reduce, so it's a little
messy, but hopefully makes sense.
Differential Revision: https://reviews.llvm.org/D102959
Only supported with -polly-position=early. Unfortunately, the
extension point callpack for VectorizerStart only passes a
FunctionPassManager, making it impossible to add a module pass.
This required support for the canonicalization passes, inlcuding
porting RewriteByReferenceParams to the NPM.
For some reason, the legacy pass pipeline with -polly-position=early did
not run the CodePreparation pass. This was fixed as well.
We previously had a different interpretation of unroll transformation
attributes than how LoopUnroll interpreted it. In particular,
llvm.loop.unroll.enable was needed explicitly to enable it and disabling
metadata was ignored.
Additionally, it required that either full unrolling or an unroll factor
to be specified or fail otherwise. An unroll factor is still required,
but the transformation is ignored with the hope that LoopUnroll is going
to apply the unrolling, since Polly currently does not implement an
heuristic.
Fixes llvm.org/PR50109
We enumerated the cross product Domain x Scatter, but sorted only be the
scatter key. In case there are are multiple statement instances per
scatter value, the order between statement instances of the same loop
iteration was undefined.
Propertly enumerate and sort only by the scatter value, and group the
domains using the scatter dimension again.
Thanks to Leonard Chan for the report.
Make Polly look for unrolling metadata (https://llvm.org/docs/TransformMetadata.html#loop-unrolling) that is usually only interpreted by the LoopUnroll pass and apply it to the SCoP's schedule.
While not that useful by itself (there already is an unroll pass), it introduces mechanism to apply arbitrary loop transformation directives in arbitrary order to the schedule. Transformations are applied until no more directives are found. Since ISL's rescheduling would discard the manual transformations and it is assumed that when the user specifies the sequence of transformations, they do not want any other transformations to apply. Applying user-directed transformations can be controlled using the `-polly-pragma-based-opts` switch and is enabled by default.
This does not influence the SCoP detection heuristic. As a consequence, loop that do not fulfill SCoP requirements or the initial profitability heuristic will be ignored. `-polly-process-unprofitable` can be used to disable the latter.
Other than manually editing the IR, there is currently no way for the user to add loop transformations in an order other than the order in the default pipeline, or transformations other than the one supported by clang's LoopHint. See the `unroll_double.ll` test as example that clang currently is unable to emit. My own extension of `#pragma clang loop` allowing an arbitrary order and additional transformations is available here: https://github.com/meinersbur/llvm-project/tree/pragma-clang-loop. An effort to upstream this functionality as `#pragma clang transform` (because `#pragma clang loop` has an implicit transformation order defined by the loop pipeline) is D69088.
Additional transformations from my downstream pragma-clang-loop branch are tiling, interchange, reversal, unroll-and-jam, thread-parallelization and array packing. Unroll was chosen because it uses already-defined metadata and does not require correctness checks.
Reviewed By: sebastiankreutzer
Differential Revision: https://reviews.llvm.org/D97977
This reverts commit 329aeb5db4,
and relands commit 61f006ac65.
This is a continuation of D89456.
As it was suggested there, now that SCEV models `PtrToInt`,
we can try to improve SCEV's pointer handling.
In particular, i believe, i will need this in the future
to further fix `SCEVAddExpr`operation type handling.
This removes special handling of `ConstantPointerNull`
from `ScalarEvolution::createSCEV()`, and add constant folding
into `ScalarEvolution::getPtrToIntExpr()`.
This way, `null` constants stay as such in SCEV's,
but gracefully become zero integers when asked.
Reviewed By: Meinersbur
Differential Revision: https://reviews.llvm.org/D98147
This is a continuation of D89456.
As it was suggested there, now that SCEV models `PtrToInt`,
we can try to improve SCEV's pointer handling.
In particular, i believe, i will need this in the future
to further fix `SCEVAddExpr`operation type handling.
This removes special handling of `ConstantPointerNull`
from `ScalarEvolution::createSCEV()`, and add constant folding
into `ScalarEvolution::getPtrToIntExpr()`.
This way, `null` constants stay as such in SCEV's,
but gracefully become zero integers when asked.
Reviewed By: Meinersbur
Differential Revision: https://reviews.llvm.org/D98147
Emit llvm.loop.parallel_accesses metadata instead of
llvm.mem.parallel_loop_access. The latter is deprecated because it
assumes that LoopIDs are persistent, which they are not.
We also emit parallel access metadata for all surrounding parallel
loops, not just the innermost parallel.
DetectionContext objects are stored as values in a DenseMap. When the
DenseMap reaches its maximum load factor, it is resized and all its
objects moved to a new memory allocation. Unfortunately Scop object have
a reference to its DetectionContext. When the DenseMap resizes, all the
DetectionContexts reference now point to invalid memory, even if caused
by an unrelated DetectionContext.
Even worse, NewPM's ScopPassManager called isMaxRegionInScop with the
Verify=true parameter before each pass. This caused the old
DetectionContext to be removed an a new on created and re-verified.
Of course, the Scop object was already created pointing to the old
DetectionContext. Because the new DetectionContext would
usually be stored at the same position in the DenseMap, the reference
would usually reference the new DetectionContext of the same Region.
Usually.
If not, the old position still points to memory in the DenseMap
allocation (unless also a resizing occurs) such that tools like Valgrind
and AddressSanitizer would not be able to diagnose this.
Instead of storing the DetectionContext inside the DenseMap, use a
std::unique_ptr to a DetectionContext allocation, i.e. it will not move
around anymore. This also allows use to remove the very strange
DetectionContext(const DetectionContext &&)
copy/move(?) constructor. DetectionContext objects now are neither
copied nor moved.
As a result, every re-verification of a DetectionContext will use a new
allocation. Therefore, once a Scop object has been created using a
DetectionContext, it must not be re-verified (the Scop data structure
requires its underlying Region to not change before code generation
anyway). The NewPM may call isMaxRegionInScop only with
Validate=false parameter.
In addition to that regression tests should not test the intire pass
pipeline (unless they are testing the pipeline itself), the Polly-ACC
currently does not support the new pass manager. If enabled by default,
such tests will therefore fail.
Use the -polly-gpu-runtime and -polly-gpu-arch options also as default
values for the PPCGCodeGeneration pass. This requires to move the option
to be moved from the pipeline-building Register passes to the
PPCGCodeGeneration implementation.
Fixes the spir-typesize.ll buildbot fail.
Michael Kruse