| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
instructions.
Summary:
Presently, InstCombiner::foldICmpWithCastAndCast() implicitly assumes that it is
only invoked with icmp instructions of integer type. If that assumption is broken,
and it is called with an icmp of vector type, then it fails (asserts/crashes).
This patch addresses the deficiency. It allows it to simplify
icmp (ptrtoint x), (ptrtoint/c) of vector type into a compare of the inputs,
much as is done when the type is integer.
Reviewers: apilipenko, fedor.sergeev, mkazantsev, anna
Reviewed By: anna
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D44063
llvm-svn: 326730
|
| |
|
|
|
|
|
|
| |
This patch teaches getMinimumFPType to support shrinking a vector of ConstantFPs. This should improve our ability to combine vector fptrunc with fp binops.
Differential Revision: https://reviews.llvm.org/D43774
llvm-svn: 326729
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
getCompare returns true, false or undef constants if the comparison can
be evaluated, or nullptr if it cannot. This is in line with what
ConstantExpr::getCompare returns. It also allows us to use
ConstantExpr::getCompare for comparing constants.
Reviewers: davide, mssimpso, dberlin, anna
Reviewed By: davide
Differential Revision: https://reviews.llvm.org/D43761
llvm-svn: 326720
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
We can discard initial blocks that do other work
We do not need to limit ourselves to just the first block in the chain.
Reviewers: courbet, davide
Reviewed By: courbet
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D44029
llvm-svn: 326698
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
`musttail` calls can't be naively splitted. The split blocks must
include not only the call instruction itself, but also (optional)
`bitcast` and `return` instructions that follow it.
Clone `bitcast` and `ret`, place them into the split blocks, and
remove the tail block when done.
Reviewers: junbuml, mcrosier, davidxl, davide, fhahn
Reviewed By: fhahn
Subscribers: JDevlieghere, llvm-commits
Differential Revision: https://reviews.llvm.org/D43729
llvm-svn: 326666
|
| |
|
|
| |
llvm-svn: 326661
|
| |
|
|
| |
llvm-svn: 326660
|
| |
|
|
|
|
|
| |
As shown in D44043, we may need this fold in the backend,
but it's also missing in the IR optimizer.
llvm-svn: 326659
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
This caused some links to fail with ThinLTO due to missing symbols as
well as causing some binaries to have failures at runtime. We're working
with the author to get a test case, but want to get the tree green
again.
Further, it appears to introduce a data race. While the test usage of
threads was disabled in r325361 & r325362, that isn't an acceptable fix.
I've reverted both of these as well. This code needs to be thread safe.
Test cases for this are already on the original commit thread.
llvm-svn: 326638
|
| |
|
|
|
|
|
|
|
|
| |
The code was checking that all of the instructions in the
sequence are 'fast', but that's not necessary. The final
multiply is all that we need to check (tests adjusted).
The fmul doesn't need to be fully 'fast' either, but that
can be another patch.
llvm-svn: 326608
|
| |
|
|
|
|
|
|
| |
This narrow fold was added with no motivation or test cases
a bit over 5 years ago. Removing a constant operand is a
good canonicalization? We should handle Y*2.0 too then?
llvm-svn: 326606
|
| |
|
|
|
|
|
|
|
|
| |
If we are only truncating bits from the extend we should be able to just use a smaller extend.
If we are truncating more than the extend we should be able to just use a fptrunc since the presense of the fpextend shouldn't affect rounding.
Differential Revision: https://reviews.llvm.org/D43970
llvm-svn: 326595
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
Currently when AllowRemainder is disabled, pragma unroll count is not
respected even though there is no remainder. This bug causes a loop
fully unrolled in many cases even though the user specifies a unroll
count. Especially it affects OpenCL/CUDA since in many cases a loop
contains convergent instructions and currently AllowRemainder is
disabled for such loops.
Differential Revision: https://reviews.llvm.org/D43826
llvm-svn: 326585
|
| |
|
|
|
|
| |
Reverts r326574.
llvm-svn: 326582
|
| |
|
|
|
|
|
|
|
|
| |
Summary: See PR36557.
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D44009
llvm-svn: 326574
|
| |
|
|
|
|
| |
This includes the test cases from D43970 and additional tests for combining (fptrunc (binop (fpext), (fpext))) where the pre-extended types don't match the trunc and therefore can't be completely removed.
llvm-svn: 326528
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
Do not break musttail invariant by promoting arguments of musttail
callee or caller.
Reviewers: sanjoy, dberlin, hfinkel, george.burgess.iv, fhahn, rnk
Reviewed By: rnk
Subscribers: rnk, llvm-commits
Differential Revision: https://reviews.llvm.org/D43926
llvm-svn: 326521
|
| |
|
|
|
|
|
|
| |
required for what is being tested.
The loads and stores were getting the data and storing the results. There's no reason we can't just use function arguments and return.
llvm-svn: 326515
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This is a retry of r326502 with updates to the reassociate
test file that I missed the first time.
@test15_reassoc in the supposed -reassociate test file
(except that it tests 2 other passes too...) shows that
there's no clear responsiblity for reassociation transforms.
Instcombine now gets that case, but only because the
constant values are identical. Otherwise, it would still
miss that pattern.
Reassociate doesn't get that case because it hasn't been
updated to use less than 'fast' FMF.
llvm-svn: 326513
|
| |
|
|
| |
llvm-svn: 326511
|
| |
|
|
|
|
|
|
| |
I forgot that I added tests for 'reassoc' to -reassociate, but
suprisingly that file calls -instcombine too, so it is affected.
I'll update that file and try again.
llvm-svn: 326510
|
| |
|
|
| |
llvm-svn: 326502
|
| |
|
|
| |
llvm-svn: 326474
|
| |
|
|
| |
llvm-svn: 326448
|
| |
|
|
|
|
|
| |
This transform may be out-of-scope for instcombine,
but this is only documenting the current behavior.
llvm-svn: 326442
|
| |
|
|
| |
llvm-svn: 326440
|
| |
|
|
|
|
|
|
|
|
| |
The range of SCEVUnknown Phi which merges values `X1, X2, ..., XN`
can be evaluated as `U(Range(X1), Range(X2), ..., Range(XN))`.
Reviewed By: sanjoy
Differential Revision: https://reviews.llvm.org/D43810
llvm-svn: 326418
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Do not replace results of `musttail` calls with a constant if the
call itself can't be removed.
Do not zap returns of `musttail` callees, if the call site can't be
removed and replaced with a constant.
Do not zap returns of `musttail`-calling blocks, this breaks
invariant too.
Patch by Fedor Indutny
Differential Revision: https://reviews.llvm.org/D43695
llvm-svn: 326404
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
`musttail` requires identical signatures of caller and callee. Removing
arguments breaks `musttail` semantics.
PR36441
Patch by Fedor Indutny
Differential Revision: https://reviews.llvm.org/D43708
llvm-svn: 326394
|
| |
|
|
|
|
| |
I've added random FMF to one of the tests to show those are propagated.
llvm-svn: 326377
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When the function has musttail call - its cc is fixed to be equal to the
cc of the musttail callee. In such case (and in the case of the musttail
callee), GlobalOpt should not change the cc to fastcc as it will break
the invariant.
This fixes PR36546
Patch by: Fedor Indutny (indutny)
Differential revision: https://reviews.llvm.org/D43859
llvm-svn: 326376
|
| |
|
|
| |
llvm-svn: 326331
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
more than once.
Summary:
Fix a bug in MergeICmp that can lead to a BCECmp block being processed more than once and eventually lead to a broken LLVM module.
The problem is that if the non-constant value is not produced by the last block, the producer will be processed once when the its parent block
is processed and second time when the last block is processed.
We end up having 2 same BCECmpBlock in the merge queue. And eventually lead to a broken LLVM module.
Reviewers: courbet, davide
Reviewed By: courbet
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D43825
llvm-svn: 326318
|
| |
|
|
| |
llvm-svn: 326303
|
| |
|
|
|
|
| |
Are any of these correct?
llvm-svn: 326241
|
| |
|
|
|
|
|
|
|
|
| |
ExtractElement.
This is similar to what's done in computeKnownBits and computeSignBits. Don't do anything fancy just collect information valid for any element.
Differential Revision: https://reviews.llvm.org/D43789
llvm-svn: 326237
|
| |
|
|
|
|
|
|
| |
This is a slight reduction of one of the benchmarks
that suffered with D43079. Cost model changes should
not cause this test to remain scalarized.
llvm-svn: 326221
|
| |
|
|
|
|
|
|
| |
This is a slight reduction of one of the benchmarks
that suffered with D43079. Cost model changes should
not cause this test to remain scalarized.
llvm-svn: 326217
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In case we update a ValuePHI node created earlier, we could update it
based on a different OpPHI which could be in a different block.
We need to update the TempToBlock mapping reflecting the new block,
otherwise we would end up placing the new phi node in a wrong block.
This problem is exposed by the test case in
https://bugs.llvm.org/show_bug.cgi?id=36504.
This patch fixes a slightly simpler problem than in the bug report. In
the bug's re-producer, the additional problem is that we are re-using a
ValuePHI node with to few incoming values for the new OpPHI. If this
patch makes sense, I will follow it up with a patch that creates a new
PHI node if the existing PHI node has a different number of incoming
values.
Reviewers: davide, dberlin
Reviewed By: dberlin
Differential Revision: https://reviews.llvm.org/D43770
llvm-svn: 326181
|
| |
|
|
|
|
| |
This was causing bot failures on greendragon
llvm-svn: 326169
|
| |
|
|
|
|
|
|
|
|
| |
Summary:
Add specific mtriples to tests added in r326154.
From: Evgeny Stupachenko <evstupac@gmail.com>
<evgeny.v.stupachenko@intel.com>
llvm-svn: 326158
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
The change fix an assert fail at ScalarEvolutionExpander.cpp:
assert(ExitCount != SE.getCouldNotCompute() && "Invalid loop count");
Reviewers: sbaranga
Differential Revision: http://reviews.llvm.org/D42604
From: Evgeny Stupachenko <evstupac@gmail.com>
<evgeny.v.stupachenko@intel.com>
llvm-svn: 326154
|
| |
|
|
|
|
| |
vectors; NFC
llvm-svn: 326148
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
constants.
Summary: This allows vector fabs to be removed in more cases.
Reviewers: spatel, arsenm, RKSimon
Reviewed By: spatel
Subscribers: wdng, llvm-commits
Differential Revision: https://reviews.llvm.org/D43739
llvm-svn: 326138
|
| |
|
|
|
|
|
|
|
|
| |
Agner's tables indicate that for SSE42+ targets (Core2 and later) we can reduce the FADD/FSUB/FMUL costs down to 1, which should fix the Himeno benchmark.
Note: the AVX512 FDIV costs look rather dodgy, but this isn't part of this patch.
Differential Revision: https://reviews.llvm.org/D43733
llvm-svn: 326133
|
| |
|
|
| |
llvm-svn: 326117
|
| |
|
|
| |
llvm-svn: 326115
|
| |
|
|
|
|
| |
checks. NFC
llvm-svn: 326114
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Note: gcc appears to allow this fold with -freciprocal-math alone,
but clang/llvm require more than that with this patch. The wording
in the definitions seems fuzzy enough that it could go either way,
but we'll err on the conservative side of FMF interpretation.
This patch also changes the newly created fmul to have FMF propagated
by the last fdiv rather than intersecting the FMF of the fdivs. This
matches the behavior of other folds near here. The new fmul is only
used to produce an intermediate op for the final fdiv result, so it
shouldn't be any stricter than that result. The previous behavior
could result in dropping FMF via other folds in instcombine or CSE.
Differential Revision: https://reviews.llvm.org/D43398
llvm-svn: 326098
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
All SIMD architectures can emulate masked load/store/gather/scatter
through element-wise condition check, scalar load/store, and
insert/extract. Therefore, bailing out of vectorization as legality
failure, when they return false, is incorrect. We should proceed to cost
model and determine profitability.
This patch is to address the vectorizer's architectural limitation
described above. As such, I tried to keep the cost model and
vectorize/don't-vectorize behavior nearly unchanged. Cost model tuning
should be done separately.
Please see
http://lists.llvm.org/pipermail/llvm-dev/2018-January/120164.html for
RFC and the discussions.
Closes D43208.
Patch by: Hideki Saito <hideki.saito@intel.com>
llvm-svn: 326079
|