This reverts commit 9e9bda2e8f5b88715bad767a4b7740df32b040d2.
This causes a backend error when building the Linux kernel for arm64.
See https://reviews.llvm.org/D122166 for a simplified reproducer.
The motivation for this is that while both memcpyopt and dse will catch this case, both are limited by MSSA's walk back threshold when finding clobbers. As such, if you have a memcpy of an otherwise dead alloca placed towards the end of a long basic block with lots of other memory instructions, it would be missed. This is a bit undesirable for such an "obviously" useless bit of code.
As noted in comments, we should probably generalize instcombine's escape analysis peephole (see visitAllocInst) to allow read xor write. Doing that would subsume this code in a more general way, but is also a more involved change. For the moment, I went with the easiest fix.
There's a potential miscompile or missed optimization with
propagating 'nsw' in the transform proposed in D122013, so
we need at least one more test for coverage.
When shifting by a byte-multiple:
bswap (shl X, C) --> lshr (bswap X), C
bswap (lshr X, C) --> shl (bswap X), C
This is an IR implementation of a transform suggested in D120648.
The "swaps cancel" test models the motivating optimization from
that proposal.
Alive2 checks (as noted in the other review, we could use
knownbits to handle shift-by-variable-amount, but that can be an
enhancement patch):
https://alive2.llvm.org/ce/z/pXUaRfhttps://alive2.llvm.org/ce/z/ZnaMLf
Differential Revision: https://reviews.llvm.org/D122010
Before this patch the DebugifyLevel option was used for
the synthetic mode, so after this, it will be used in
the original mode as well.
Differential Revision: https://reviews.llvm.org/D115623
Rather than iterating over users and comparing operands, iterate
over uses and check operand number. Otherwise, we'll end up
promoting a store twice if it has two equal operands.
This can only happen with opaque pointers, as otherwise both
operands differ by a level of indirection, so a bitcast would have
to be involved.
Fixes https://github.com/llvm/llvm-project/issues/54495.
This is the IR counterpart to 370ebc9d9a573d6
which provided a bswap narrowing fix for issue #53867.
Here we can be more general (although I'm not sure yet
what would happen for illegal types in codegen - too
rare to worry about?):
https://alive2.llvm.org/ce/z/3-CPfo
This will be more effective if we have moved the shift
after the bswap as proposed in D122010, but it is
independent of that patch.
Differential Revision: https://reviews.llvm.org/D122166
When outlining a phi node, if the the incoming branch is a block contained in the region and the branch from that block is not outlined, we create broken code. The fix is to recognize when that branch from the included incoming block is not contained, and ignore the region.
Reviewer: paquette
Differential Revision: https://reviews.llvm.org/D121311
The implementation is just a generalization of the Select handler.
We're no trying to be smart and compute any kind of fixed point.
Differential Revision: https://reviews.llvm.org/D121897
Since the IROutliner is performing an optimization, it should not outline from functions explicitly marked with optnone. This adds an extra check and test to make sure this does not occur.
Reviewers: paquette
Differential Revision: https://reviews.llvm.org/D121567
The semantics of an inalloca alloca instruction requires that it not be reordered with a preceeding stacksave intrinsic call. Unfortunately, there's no def/use edge or memory dependence edge. (THe memory point is slightly subtle, but in general a new allocation can't alias with a call which executes strictly before it comes into existance.)
I'd tried to tackle this same case previously in 689babdf6, but the fix chosen there turned out to be incomplete. As such, this change contains a fully revert of the first fix attempt.
This was noticed when investigating problems which surfaced with D118538, but this is definitely an existing bug. This time around, I managed to reduce a couple of additional cases, including one which was being actively miscompiled even without the new scheduling change. (See test diffs)
Compile time wise, we only spend extra time when seeing a stacksave (rare), and even then we walk the block at most once per schedule window extension. Likely a non-issue.
There are some slight changes to the test lines due to different cost threshold choices in the two command lines, but I don't believe these to be interesting the purpose of the tests.
The existing scheduling doesn't account for the scheduling restrictions implied by inalloca allocas combined with stacksave/stackrestore. This adds coverage including one currently miscompiling case.
This fixes an active miscompile visible in the test changes. The basic problem is that the scheduling dependency graph didn't have any edges for control dependence within a single basic block. The result is that we could (and in some rare cases *did*) perform reorderings within a block which could introduce new undefined behavior along paths which didn't previously contain any.
Impact wise, we have two major cases where control is not guaranteed to reach a later instruction in the block: may throw calls, and calls containing infinite loops.
* The former case was mostly covered by the memory dependencies, and to trigger require a function which can throw, but not write to memory. In theory, such a case is possible, but not likely in practice.
* The later case is likely more of an issue in practice. After this code was first written, we changed the IR semantics to allow well defined infinite loops without satisifying mustprogress. Even for C/C++ - which do imply mustprogress - recent changes to how we treat atomics (e.g. an atomic read does not always imply a write) could expose this issue. I'm a bit shocked we don't seem to have a bug report which hit this in real code actually.
Compile time wise, this results in a single extra scan of the scheduling window in the common case. Since we stop scanning at the next instruction which isn't guaranteed to execute, no matter what order we traverse instructions in, we scan the block once. The exception to this is that when we extend the scheduling window downwards, we invalidate all dependencies, and thus rescan. So the potentially expensive case is when we a call in a big schedule window which is frequently extended. We could optimize this case (by caching the last instruction not guaranteeed to transfer execution and scanning only the extended window) and starting there), but I decided to leave the complexity until it mattered. That same case is already degenerate with memory dependences which is more expensive than the control dependence scan.
We could also consider combining the memory dependence and control dependence sets to reduce memory usage, but since it complicates the code slightly and makes debugging a bit harder, I went with the simplest scheme for now.
This was noticed while trying to understand the failures reported against D118538, but is not otherwise related to that change.
With debug information enabled (-g) Clang will wrap the actual target
region into a new function which is called from the "kernel". The problem
is that the "kernel" is now basically a wrapper without all the things
we expect. More importantly, if we end up asking for an AAKernelInfo
for the "target region function" we might try to turn it into SPMD mode.
That used to cause an assertion as that function doesn't have an
appropriately named `_exec_mode` global. While the global is going away
soon we still need to make sure to properly handle this case, e.g.,
perform optimizations reliably.
Differential Revision: https://reviews.llvm.org/D122043
SLP is currently assuming that control dependence in these cases is irrelevant. This is only valid if none of the lib-funcs involved can throw or infinite loop in the scalar forms. This appears to be true (or at least we infer the respective attributes) for the libfuncs I spot checked. This change is mostly for shrunking the diff on an upcoming patch.
Generalize D99629 for ELF. A default visibility non-local symbol is preemptible
in a -shared link. `isInterposable` is an insufficient condition.
Moreover, a non-preemptible alias may be referenced in a sub constant expression
which intends to lower to a PC-relative relocation. Replacing the alias with a
preemptible aliasee may introduce a linker error.
Respect dso_preemptable and suppress optimization to fix the abose issues. With
the change, `alias = 345` will not be rewritten to use aliasee in a `-fpic`
compile.
```
int aliasee;
extern int alias __attribute__((alias("aliasee"), visibility("hidden")));
void foo() { alias = 345; } // intended to access the local copy
```
While here, refine the condition for the alias as well.
For some binary formats like COFF, `isInterposable` is a sufficient condition.
But I think canonicalization for the changed case has little advantage, so I
don't bother to add the `Triple(M.getTargetTriple()).isOSBinFormatELF()` or
`getPICLevel/getPIELevel` complexity.
For instrumentations, it's recommended not to create aliases that refer to
globals that have a weak linkage or is preemptible. However, the following is
supported and the IR needs to handle such cases.
```
int aliasee __attribute__((weak));
extern int alias __attribute__((alias("aliasee")));
```
There are other places where GlobalAlias isInterposable usage may need to be
fixed.
Reviewed By: rnk
Differential Revision: https://reviews.llvm.org/D107249
With D107249 I saw huge compile time regressions on a module (150s ->
5700s). This turned out to be due to a huge RefSCC in
the module. As we ran the function simplification pipeline on functions
in the SCCs in the RefSCC, some of those SCCs would be split out to
their RefSCC, a child of the current RefSCC. We'd skip the remaining
SCCs in the huge RefSCC because the current RefSCC is now the RefSCC
just split out, then revisit the original huge RefSCC from the
beginning. This happened many times because many functions in the
RefSCC were optimizable to the point of becoming their own RefSCC.
This patch makes it so we don't skip SCCs not in the current RefSCC so
that we split out all the child RefSCCs on the first iteration of
RefSCC. When we split out a RefSCC, we invalidate the original RefSCC
and add the remainder of the SCCs into a new RefSCC in
RCWorklist. This happens repeatedly until we finish visiting all
SCCs, at which point there is only one valid RefSCC in
RCWorklist from the original RefSCC containing all the SCCs that
were not split out, and we visit that.
For example, in the newly added test cgscc-refscc-mutation-order.ll,
we'd previously run instcombine in this order:
f1, f2, f1, f3, f1, f4, f1
Now it's:
f1, f2, f3, f4, f1
This can cause more passes to be run in some specific cases,
e.g. if f1<->f2 gets optimized to f1<-f2, we'd previously run f1, f2;
now we run f1, f2, f2.
This improves kimwitu++ compile times by a lot (12-15% for various -O3 configs):
https://llvm-compile-time-tracker.com/compare.php?from=2371c5a0e06d22b48da0427cebaf53a5e5c54635&to=00908f1d67400cab1ad7bcd7cacc7558d1672e97&stat=instructions
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D121953
[SCCP] do not clean up dead blocks that have their address taken
Fixes a crash observed in IPSCCP.
Because the SCCPSolver has already internalized BlockAddresses as
Constants or ConstantExprs, we don't want to try to update their Values
in the ValueLatticeElement. Instead, continue to propagate these
BlockAddress Constants, continue converting BasicBlocks to unreachable,
but don't delete the "dead" BasicBlocks which happen to have their
address taken. Leave replacing the BlockAddresses to another pass.
Fixes: https://github.com/llvm/llvm-project/issues/54238
Fixes: https://github.com/llvm/llvm-project/issues/54251
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D121744
This adds a new option to control AllowSpeculation added in D119965 when
using `-passes=...`.
This allows reproducing #54023 using opt.
Reviewed By: aeubanks
Differential Revision: https://reviews.llvm.org/D121944
Currently some optimizations are disabled because llvm::CannotBeNegativeZero()
does not know how to deal with the constrained intrinsics. This patch fixes
that by extending the existing implementation.
Differential Revision: https://reviews.llvm.org/D121483
The way the pass is actually used in the optimization pipeline,
TLI will be available, but this is not the case when running just
-lower-constant-intrinsics in tests, which ends up being quite
confusing.
Require TLI unconditionally, as we usually do.
This avoids false positive verification failures if the condition
is not literally true/false, but SCEV still makes use of the fact
that a loop is not reachable through more complex reasoning.
Fixes https://github.com/llvm/llvm-project/issues/54434.
LoopSimplifyCFG may process loops that are not in
loop-simplify/canonical form. For loops not in canonical form, exit
blocks may be reachable from non-loop blocks and we cannot consider them
as dead if they only are not reachable from the loop itself.
Unfortunately the smallest test I could come up with requires running
multiple passes:
-passes='loop-mssa(loop-instsimplify,loop-simplifycfg,simple-loop-unswitch)'
The reason is that loops are canonicalized at the beginning of loop
pipelines, so a later transform has to break canonical form in a way
that breaks LoopSimplifyCFG's dead-exit analysis.
Alternatively we could try to require all loop passes to maintain
canonical form. That in turn would also require additional verification.
Fixes#54023, #49931.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D121925
When generating a all-one mask value whose bitwidth is larger than 64, signed extension should be used rather then zero extension.
Reviewed By: jsji
Differential Revision: https://reviews.llvm.org/D120865
This patch tries to sink instructions when they are only used in a successor block.
This is a further enhancement patch based on Anna's commit:
D109700, which allows sinking an instruction having multiple uses in a single user.
In this patch, sink instructions with multiple users in a single successor block will be supported.
It could fix a known issue from rust:
https://github.com/rust-lang/rust/issues/51346#issuecomment-394443610
Reviewed By: nikic, reames
Differential Revision: https://reviews.llvm.org/D121585
Splat loads are inexpensive in X86. For a 2-lane vector we need just one
instruction: `movddup (%reg), xmm0`. Using the standard Splat score leads
to worse code. This patch adds a new score dedicated for splat loads.
Please note that a splat is usually three IR instructions:
- It is usually a load and 2 inserts:
%ld = load double, double* %gep
%ins1 = insertelement <2 x double> poison, double %ld, i32 0
%ins2 = insertelement <2 x double> %ins1, double %ld, i32 1
- But it can also be a load, an insert and a shuffle:
%ld = load double, double* %gep
%ins = insertelement <2 x double> poison, double %ld, i32 0
%shf = shufflevector <2 x double> %ins, <2 x double> poison, <2 x i32> zeroinitializer
Because of this some of the lit tests contain more IR instructions.
Differential Revision: https://reviews.llvm.org/D121354
