Commit Graph

323 Commits

Author SHA1 Message Date
Craig Topper 5f6d93c7d3 [CallSite removal][Attributor] Replaces use of CallSite with CallBase. NFC
Differential Revision: https://reviews.llvm.org/D78343
2020-04-17 10:44:31 -07:00
Craig Topper 8c94d616e1 Revert "[CallSite removal][MemCpyOptimizer] Replace CallSite with CallBase. NFC"
There were extra changes that weren't supposed to be in there

This reverts commit b91f78db37.
2020-04-17 10:11:22 -07:00
Craig Topper b91f78db37 [CallSite removal][MemCpyOptimizer] Replace CallSite with CallBase. NFC
There are also some adjustments to use MaybeAlign in here due
to CallBase::getParamAlignment() being deprecated. It would
be cleaner if getOrEnforceKnownAlignment was migrated
to Align/MaybeAlign.

Differential Revision: https://reviews.llvm.org/D78345
2020-04-17 10:07:20 -07:00
Craig Topper 798b262c3c [CallSite removal][IPO] Change implementation of AbstractCallSite to store a CallBase* instead of CallSite. NFCI.
CallSite will likely be removed soon, but AbstractCallSite serves a different purpose and won't be going away.

This patch switches it to internally store a CallBase* instead of a
CallSite. The only interface changes are the removal of the getCallSite
method and getCallBackUses now takes a CallBase&. These methods had only
a few callers that were easy enough to update without needing a
compatibility shim.

In the future once the other CallSites are gone, the CallSite.h
header should be renamed to AbstractCallSite.h

Differential Revision: https://reviews.llvm.org/D78322
2020-04-16 16:24:45 -07:00
Johannes Doerfert c4d3188adb [Attributor][NFC] Reduce indention for call site attribute seeding
Also added a TODO to remind us that indirect calls could be optimized as
well.
2020-04-16 02:32:31 -05:00
Johannes Doerfert 0741dec27b [Attributor][FIX] Handle droppable uses when replacing values
Since we use the fact that some uses are droppable in the Attributor we
need to handle them explicitly when we replace uses. As an example, an
assumed dead value can have live droppable users. In those we cannot
replace the value simply by an undef. Instead, we either drop the uses
(via `dropDroppableUses`) or keep them as they are. In this patch we do
both, depending on the situation. For values that are dead but not
necessarily removed we keep droppable uses around because they contain
information we might be able to use later. For values that are removed
we drop droppable uses explicitly to avoid replacement with undef.
2020-04-16 00:56:08 -05:00
Johannes Doerfert ad9c284cc3 [Attributor][NFC] Run the verifier only on functions and under EXPENSIVE_CHECKS
Running the verifier is expensive so we want to avoid it even in runs
that enable assertions. As we move closer to enabling the Attributor
this code will be executed by some buildbots but not cause overhead for
most people.
2020-04-16 00:55:33 -05:00
Johannes Doerfert 898bbc252a [Attributor] Lazily collect function information
Before, we eagerly analyzed all the functions to collect information
about them, e.g. what instructions may read/write memory. This had
multiple drawbacks:
  - In CGSCC-mode we can end up looking at a callee which is not in the
    SCC but for which we need an initialized cache.
  - We end up looking at functions that we deem dead and never need to
    analyze in the first place.
  - We have a implicit dependence which is easy to break.

This patch moves the function analysis into the information cache and
makes it lazy. There is no real functional change expected except due to
the first reason above.
2020-04-15 22:26:38 -05:00
Johannes Doerfert 8c4057e3a3 [Attributor] Replace call graph call sites after function replacement
The CallGraphUpdater allows to directly alter call site information and
we should do so. This might appease the windows buildbot that crashes
during the SCC traversal.
2020-04-15 22:24:09 -05:00
Johannes Doerfert 0985554b70 [Attributor][NFC] Split AbstractAttributes out of Attributor.cpp
Attributor.cpp became quite big and we need to start provide structure.
The Attributor code is now in Attributor.cpp and the classes derived
from AbstractAttribute are in AttributorAttributes.cpp. Minor changes
were required but no intended functional changes.

We also minimized includes as part of this.

Reviewed By: baziotis

Differential Revision: https://reviews.llvm.org/D76873
2020-04-08 19:02:14 -05:00
Eli Friedman 3f13ee8a00 [NFC] Modernize misc. uses of Align/MaybeAlign APIs.
Use the current getAlign() APIs where it makes sense, and use Align
instead of MaybeAlign when we know the value is non-zero.
2020-04-06 17:53:04 -07:00
Tarindu Jayatilaka b43b59fcc0 Expose `attributor-disable` to the new and old pass managers
The new and old pass managers (PassManagerBuilder.cpp and
PassBuilder.cpp) are exposed to an `extern` declaration of
`attributor-disable` option which will guard the addition of the
attributor passes to the pass pipelines.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D76871
2020-04-05 22:29:34 -05:00
Stefanos Baziotis f3dd3a66d3 [Attributor] AAUndefinedBehavior: Use AAValueSimplify in memory accessing instructions.
Query AAValueSimplify on pointers in memory accessing instructions to take
advantage of the constant propagation (or any other value simplification) of such values.
2020-04-05 02:46:26 +03:00
Luofan Chen eec6d87626 [Attributor] Deduce attributes for non-exact functions
This patch is based on D63312 and D63319. For now we create shallow wrappers for all functions that are IPO amendable.
See also [this github issue](https://github.com/llvm/llvm-project/issues/172).

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D76404
2020-04-04 11:34:58 -05:00
Johannes Doerfert bcd8009369 [Attributor] Use the proper context instruction in genericValueTraversal
There was a TODO in genericValueTraversal to provide the context
instruction and due to the lack of it users that wanted one just used
something available. Unfortunately, using a fixed instruction is wrong
in the presence of PHIs so we need to update the context instruction
properly.

Reviewed By: uenoku

Differential Revision: https://reviews.llvm.org/D76870
2020-04-01 22:20:47 -05:00
Johannes Doerfert ac96c8fd85 [Attributor][FIX] Do not compute ranges for arguments of declarations
This cannot be triggered right now, as far as I know, but it doesn't
make sense to deduce a constant range on arguments of declarations.
Exposed during testing of AAValueSimplify extensions.
2020-04-01 22:05:30 -05:00
Johannes Doerfert 54d6a608bf [Attributor][NFC] Predetermine the module
It could happen that we delete the first function in the SCC in the
future so we should be careful accessing `Functions` after the manifest
stage.
2020-04-01 21:56:17 -05:00
Johannes Doerfert 9e19693994 [Attributor] Derive better alignment for accessed pointers
Use DL & ABI information for better alignment deduction, e.g., if a type
is accessed and the ABI specifies an alignment requirement for such an
access we can use it. This is based on a patch by @lebedev.ri and
inspired by getBaseAlign in Loads.cpp.

Depends on D76673.

Reviewed By: lebedev.ri

Differential Revision: https://reviews.llvm.org/D76674
2020-04-01 21:49:57 -05:00
Johannes Doerfert b1c788d051 [Attributor][FIX] Prevent alignment breakage wrt. must-tail calls
If we have a must-tail call the callee and caller need to have matching
ABIs. Part of that is alignment which we might modify when we deduce
alignment of arguments of either. Since we would need to keep them in
sync, which is not as simple, we simply avoid deducing alignment for
arguments of the must-tail caller or callee.

Reviewed By: rnk

Differential Revision: https://reviews.llvm.org/D76673
2020-04-01 21:40:07 -05:00
Johannes Doerfert 41f2a57d0b [Attributor][NFC] Use a BumpPtrAllocator to allocate `AbstractAttribute`s
We create a lot of AbstractAttributes and they live as long as
the Attributor does. It seems reasonable to allocate them via a
BumpPtrAllocator owned by the Attributor.

Reviewed By: baziotis

Differential Revision: https://reviews.llvm.org/D76589
2020-04-01 20:53:28 -05:00
Uday Bondhugula c4499e3333 [Attributor] Make attributor aware of aligned_alloc for heap to stack conversion
Make the attributor pass aware of aligned_alloc for converting heap
allocations to stack ones.

Depends on D76971.

Differential Revision: https://reviews.llvm.org/D76974
2020-04-01 23:26:50 +05:30
Uday Bondhugula 06066c4003 [NFC] Attributor comment updates / cast cleanup
Minor update/fixes to comments for the Attributor pass, and dyn_cast -> cast.

Signed-off-by: Uday Bondhugula <uday@polymagelabs.com>

Differential Revision: https://reviews.llvm.org/D76972
2020-03-28 13:36:43 +05:30
Johannes Doerfert 5699d08b79 [Attributor] Use knowledge retained in llvm.assume (operand bundles)
This patch integrates operand bundle llvm.assumes [0] with the
Attributor. Most IRAttributes will now look at uses of the associated
value and if there are llvm.assume operand bundle uses with the right
tag we will check if they are in the must-be-executed-context (around
the context instruction). Droppable users, which is currently only
llvm::assume, are handled special in some places now as well.

[0] http://lists.llvm.org/pipermail/llvm-dev/2019-December/137632.html

Reviewed By: uenoku

Differential Revision: https://reviews.llvm.org/D74888
2020-03-24 15:33:40 -05:00
Stefanos Baziotis a650d555fc [Attributor][NFC] Refactorings and typos in doc
Reviewed By: sstefan1, uenoku

Differential Revision: https://reviews.llvm.org/D76175
2020-03-23 22:44:10 +02:00
Benjamin Kramer ff2f5097ed [Attributor] Fold single-use variable into assert
Fixes unused variable warning in Release builds.
2020-03-23 17:41:52 +01:00
Johannes Doerfert c57689bef2 [Attributor][NFC] Copy llvm::function_ref, don't use references
On IRC this was called a "code smell" so we get rid of it.
2020-03-23 10:45:24 -05:00
Johannes Doerfert 68fed27067 [Attributor] Handle calls in AAValueConstantRange properly
We did handle calls that were operands of certain instructions but not
standalone calls we visit via indirection, e.g., selects.
2020-03-23 10:45:24 -05:00
Johannes Doerfert 54ec9b54f6 [Attributor] Unify handling of must-tail calls
We special cased must-tail calls all over the place because they cannot
be modified as other calls can be. However, we already centralized the
modification API so we can centralize the handling as well. This
simplifies the code and allows to remove must-tail calls completely.
2020-03-23 10:45:24 -05:00
Johannes Doerfert 0995001ce5 [Attributor][NFC] Predetermine the module before verification
It could happen that we delete the first function in the SCC in the
future so we should be careful accessing `Functions` after the manifest
stage.
2020-03-23 10:45:23 -05:00
Johannes Doerfert f3bf4b05c2 [Attributor][NFC] clang-format Attributor.{h,cpp} 2020-03-23 10:45:23 -05:00
omarahmed1111 b285b333dc [Attributor] Detect possibly unbounded cycles in functions
This patch add mayContainUnboundedCycle helper function which checks whether a function has any cycle which we don't know if it is bounded or not.
Loops with maximum trip count are considered bounded, any other cycle not.
It also contains some fixed tests and some added tests contain bounded and
unbounded loops and non-loop cycles.

Reviewed By: jdoerfert, uenoku, baziotis

Differential Revision: https://reviews.llvm.org/D74691
2020-03-13 11:17:33 -05:00
Pankaj Gode bf990530ae [Attributor] Improve noalias preservation using reachability
Resolution for below fixme:
(ii) Check whether the value is captured in the scope using AANoCapture.
FIXME: This is conservative though, it is better to look at CFG and
             check only uses possibly executed before this callsite.

Propagates caller argument's noalias attribute to callee.

Reviewed by: jdoerfert, uenoku

Reviewers: jdoerfert, sstefan1, uenoku

Subscribers: uenoku, sstefan1, hiraditya, llvm-commits

Differential Revision: https://reviews.llvm.org/D71617
2020-03-13 21:09:08 +05:30
Johannes Doerfert a198adb490 [Attributor] IPO across definition boundary of a function marked alwaysinline
Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D75590
2020-03-13 01:06:12 -05:00
Hideto Ueno d9bf79f4e9 [Attributor][FIX] Add a missing dependence track in noalias deduction 2020-03-12 15:27:35 +00:00
Hideto Ueno bdcbdb4848 [Attributor] Deduction based on path exploration
This patch introduces the propagation of known information based on path exploration.
For example,
```
int u(int c, int *p){
  if(c) {
     return *p;
  } else {
     return *p + 1;
  }
}
```
An argument `p` is dereferenced whatever c's value is.

For an instruction `CtxI`, we accumulate branch instructions in the must-be-executed-context of `CtxI` and then, we take the conjunction of the successors' known state.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D65593
2020-03-09 14:29:26 +09:00
Stefanos Baziotis 01c48d7d11 [Attributor] Fold terminators before changing instructions to unreachable
It is possible that an instruction to be changed to unreachable is
in the same block with a terminator that can be constant-folded.
In this case, as of now, the instruction will be changed to
unreachable before the terminator is folded. But, then the
whole BB becomes invalidated and so when we go ahead to fold
the terminator, we trap.

Change the order of these two.

Differential Revision: https://reviews.llvm.org/D75780
2020-03-07 12:38:44 +02:00
Hideto Ueno 2c0edbf19c [Attributor] Use AssumptionCache in AANonNullFloating::initialize 2020-02-25 13:00:03 +09:00
Johannes Doerfert 9708279c72 [Attributor][FIX] Undo 16188f9 until SCC iterator bug is fixed
The buildbot
  http://lab.llvm.org:8011/builders/llvm-clang-x86_64-expensive-checks-win
shows some strange SCC iterator bug since 16188f9 which we need to
investigate. This patch should remove the part of 16188f9 that could
have exposed the problem.
2020-02-21 14:20:42 -08:00
Johannes Doerfert d95cb56649 [Attributor] Make sure abstract attributes are properly initialized 2020-02-20 02:46:40 -06:00
Johannes Doerfert 6185fb13d6 [Attributor][NFC] Refactor interface 2020-02-20 02:46:40 -06:00
Johannes Doerfert 8e76fec0ae [Attributor][NFC] Improve the debug output & add a TODO 2020-02-19 23:46:08 -06:00
Johannes Doerfert f8ad735729 [Attributor] Use existing `returned` information better
We can look through calls with `returned` argument attributes when we
collect subsuming positions. This allows us to get existing attributes
from more places.
2020-02-19 23:46:07 -06:00
Johannes Doerfert a801ee869d [Attributor][FIX] Avoid setting wrong load/store alignments 2020-02-19 23:46:07 -06:00
Johannes Doerfert e1eed6c5b9 [Attributor] Generalize `getAssumedConstantInt` interface
We are often interested in an assumed constant and sometimes it has to
be an integer constant. Before we only looked for the latter, now we can
ask for either.
2020-02-19 22:33:51 -06:00
Johannes Doerfert 16188f9d70 [Attributor][FIX] Do not create new calls edge we cannot handle
If we propagate function pointers across function boundaries we can
create new call edges. These need to be represented in the CG if we run
as a CGSCC pass. In the new pass manager that is currently not handled
by the CallGraphUpdater so we need to prevent the situation for now.
2020-02-19 22:33:51 -06:00
Johannes Doerfert 1e99fc9d58 [Attributor] Add initial AAIsDead for arguments
We usually will ask for liveness of an argument anyway so we ended up
lazily creating the attribute anyway. However, that is not always the
case and even if it is we should go the eager route here. Various tests
show how this can improve the outcome. One test exposed a problem with
type mismatches between argument and call site argument, a fix is
included. For liveness various more tests were added as well.
2020-02-19 21:39:45 -06:00
Johannes Doerfert c6ac717aa7 [Attributor] Allow multiple uses of a casted function pointer
If a function pointer is casted into a different type the resulting
expression can be a constant. If so, it can be used multiple times which
cannot be handled by the AbstractCallSite constructor alone. Instead, we
follow the cast expression uses now explicitly during the call site
traversal.
2020-02-19 20:43:38 -06:00
Benjamin Kramer 564a9de28e Hide implementation details. NFC> 2020-02-17 17:55:23 +01:00
Johannes Doerfert 1d5da8cd30 [Attributor][FIX] Use pointer not reference as it can be null 2020-02-15 20:38:49 -06:00
Simon Pilgrim 8a48c4a97c Fix boolean/bitwise operator precedence warnings. NFCI. 2020-02-15 13:53:18 +00:00