As we're becoming stricter w/ respect to not allowing vregs having LLTs
and regclasses assigned both mid-globalisel pipeline, the number of
extra copies grows, some of which separate G_UNMERGE's from their
corresponding G_MERGE's, becoming a performance concern.
It's worth mentioning that we're already looking through copies while
combining legalization artifacts for every kind of artifact but
G_UNMERGE.
Reviewed By: aditya_nandakumar
Reviewers: ab, t.p.northover, volkan, javed.absar
Subscribers: rovka, kristof.beyls, llvm-commits
Differential Revision: https://reviews.llvm.org/D45644
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330660 91177308-0d34-0410-b5e6-96231b3b80d8
If a loop with child loops becomes our new inner loop after
interchanging, we only need to update LoopInfo for the blocks defined in
the old outer loop. BBs in child loops will stay there.
Reviewers: efriedma, karthikthecool, mcrosier
Reviewed By: efriedma
Differential Revision: https://reviews.llvm.org/D45970
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330653 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]].
[[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly.
Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`)
Now, they would no longer be generated (see `@in`).
So we need to make sure that they are still generated.
If the mask is constant, we do nothing. InstCombine should have unfolded it.
Else, i use `hasAndNot()` TLI hook.
For now, only handle scalars.
https://rise4fun.com/Alive/bO6
----
I *really* don't like the code i wrote in `DAGCombiner::unfoldMaskedMerge()`.
It is super fragile. Is there something like IR Pattern Matchers for this?
Reviewers: spatel, craig.topper, RKSimon, javed.absar
Reviewed By: spatel
Subscribers: andreadb, courbet, kristof.beyls, javed.absar, rengolin, nemanjai, llvm-commits
Differential Revision: https://reviews.llvm.org/D45733
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330646 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]].
[[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly.
Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`)
Now, they would no longer be generated (see `@in`).
I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this.
Reviewers: spatel, craig.topper, RKSimon, javed.absar
Reviewed By: spatel
Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits
Differential Revision: https://reviews.llvm.org/D45563
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330645 91177308-0d34-0410-b5e6-96231b3b80d8
Richard Smith noted that `typedef typename iplist::iplist_impl_type
iplist_impl_type` is incorrect, per
http://eel.is/c++draft/basic.scope#class-2
It seems that neither clang nor gcc get too angry about this, but a
newer version of msvc does.
Thanks to jcmac on IRC for pointing this out!
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330639 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
Skip basic blocks not reachable from the entry node
in MemCpyOptPass::iterateOnFunction.
Code that is unreachable may have properties that do not exist
for reachable code (an instruction in a basic block can for
example be dominated by a later instruction in the same basic
block, for example if there is a single block loop).
MemCpyOptPass::processStore is only safe to use for reachable
basic blocks, since it may iterate past the basic block
beginning when used for unreachable blocks. By simply skipping
to optimize unreachable basic blocks we can avoid asserts such
as "Assertion `!NodePtr->isKnownSentinel()' failed."
in MemCpyOptPass::processStore.
The problem was detected by fuzz tests.
Reviewers: eli.friedman, dneilson, efriedma
Reviewed By: efriedma
Subscribers: efriedma, llvm-commits
Differential Revision: https://reviews.llvm.org/D45889
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330635 91177308-0d34-0410-b5e6-96231b3b80d8
This reland includes a check to prevent the DAG combiner from folding an
offset that is smaller than the existing one. This can cause oscillations
between two possible DAGs, which was the cause of the hang and later assertion
failure observed on the lnt-ctmark-aarch64-O3-flto bot.
http://green.lab.llvm.org/green/job/lnt-ctmark-aarch64-O3-flto/2024/
Original commit message:
> This is a code size win in code that takes offseted addresses
> frequently, such as C++ constructors that typically need to compute
> an offseted address of a vtable. This reduces the size of Chromium
> for Android's .text section by 108KB.
Differential Revision: https://reviews.llvm.org/D45199
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330630 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
This change teaches DSE that the atomic memory intrinsics are stores
that can be eliminated, and can allow other stores to be eliminated.
This change specifically does not teach DSE that these intrinsics
can be partially eliminated (i.e. length reduced, and dest/src changed);
that will be handled in another change.
Reviewers: mkazantsev, skatkov, apilipenko, efriedma, rsmith
Reviewed By: efriedma
Subscribers: dmgreen, llvm-commits
Differential Revision: https://reviews.llvm.org/D45535
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330629 91177308-0d34-0410-b5e6-96231b3b80d8
This patch adds a cost model test case for vector shuffles having transpose
masks. The given costs are inaccurate and will be updated in a follow-on patch.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330625 91177308-0d34-0410-b5e6-96231b3b80d8
This demonstrates a bug where the encoding for a 16-bit bswap prints a 16-bit register and a 32-bit mnemonic. Intel docs say 16-bit bswap is undefined. We should either claim it as an invalid encoding or we should print a 16-bit mnemonic.
objdump does print the encoding as bswap with a 16-bit register. But it doesn't seem to ever print a suffix.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330621 91177308-0d34-0410-b5e6-96231b3b80d8
Only add support for getSchedInfoStr(const MachineInstr &MI) at the TargetSubtargetInfo level.
Really, the getSchedInfoStr calls need to be removed entirely, we should just return a latency/rthroughput through the subtarget and keep a string creation helper function somewhere else.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330615 91177308-0d34-0410-b5e6-96231b3b80d8
It's possible to validly spill the frame offset register
in a call sequence to a VGPR. There are definitely issues
with SGPR spilling to memory, so move the assert later.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330612 91177308-0d34-0410-b5e6-96231b3b80d8
We use llvm-symbolizer in some production systems, and we run it
against all possibly related files, including some that are not
ELF. We noticed that for some of those invalid files, llvm-symbolizer
would crash with SEGFAULT. Here is an example of such a file.
It is due to that in computeSymbolSizes, a loop uses condition
for (unsigned I = 0, N = Addresses.size() - 1; I < N; ++I) {
where if Addresses.size() is 0, N would overflow and causing the loop
to access invalid memory.
Instead of patching the loop conditions, the commit makes so that the
function returns early if Addresses is empty.
Validated by checking that llvm-symbolizer no longer crashes.
Patch by Teng Qin!
Differential Revision: https://reviews.llvm.org/D44285
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330610 91177308-0d34-0410-b5e6-96231b3b80d8
Also assert that it is correct for SGPRs. There is currently a bug
where stack slot coloring replaces SGPR spill FIs with one with
the default ID, which results in a more confusing assert later
about a dead object.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330607 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
This just refactors the lowering of the atomic memory intrinsics to more
closely match the code patterns used in the lowering of the non-atomic
memory intrinsics. Specifically, we encapsulate the lowering in
SelectionDAG::getAtomicMem*() functions rather than embedding
the code directly in the SelectionDAGBuilder code.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330603 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
This patch removes references to AliasAnalysisCounter pass from the AliasAnalysis documentation. That pass have been eliminated in 2015, at revision trunk@247167.
Reviewed By: hfinkel
Differential Revision: https://reviews.llvm.org/D45876
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330590 91177308-0d34-0410-b5e6-96231b3b80d8
Note - noticed this as the STAC case as it was unintentionally matching against *STACK* pseudo instructions.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330588 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
Found by inspection. We care about the operand that *doesn't*
contain the immediate.
I believe this is currently not hit because we fold 0xff / 0xffff
immediates only later.
Change-Id: Ic3cf8538bc7da5eff3200d96eccf9d339e6345a7
Reviewers: arsenm, rampitec
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D45886
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330586 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
See the new test case; this is really unlikely to happen with real code,
but I ran into this while attempting to bugpoint-reduce a different issue.
Change-Id: I9ade1dc1aa8fd9c4d9fc83661d7b80e310b5c4a6
Reviewers: arsenm, rampitec
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D45885
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330585 91177308-0d34-0410-b5e6-96231b3b80d8
LoopRotate only invalidates innermost loops while the changes that it makes may
also affert any of this parents. With patch rL329047, SCEV becomes much smarter
about calculation of exit counts for outer loops, so we cannot assume that they are
not affected.
Differential Revision: https://reviews.llvm.org/D45945
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330582 91177308-0d34-0410-b5e6-96231b3b80d8
Current runtime unrolling invalidates parent loop saying that it might have changed
after the inner loop has changed, but it doesn't bother to do the same to its parents.
With patch rL329047, SCEV becomes much smarter about calculation of exit counts for
outer loops. We might need to invalidate not only the immediate parent, but also
any of its parents as well.
There is no clear evidence that there is some miscompile happening because of this
(at least I don't have such test), but the common sense says that the current code
is wrong.
Differential Revision: https://reviews.llvm.org/D45940
Reviewed By: chandlerc
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330577 91177308-0d34-0410-b5e6-96231b3b80d8
In the function `simplifyOneLoop` we optimistically assume that changes in the
inner loop only affect this very loop and have no impact on its parents. In fact,
after rL329047 has been merged, we can now calculate exit counts for outer
loops which may depend on inner loops. Thus, we need to invalidate all parents
when we do something to a loop.
There is an evidence of incorrect behavior of `simplifyOneLoop`: when we insert
`SE->verify()` check in the end of this funciton, it fails on a bunch of existing
test, in particular:
LLVM :: Transforms/LoopUnroll/peel-loop-not-forced.ll
LLVM :: Transforms/LoopUnroll/peel-loop-pgo.ll
LLVM :: Transforms/LoopUnroll/peel-loop.ll
LLVM :: Transforms/LoopUnroll/peel-loop2.ll
Note that previously we have fixed issues of this variety, see rL328483.
This patch makes this function invalidate the outermost loop properly.
Differential Revision: https://reviews.llvm.org/D45937
Reviewed By: chandlerc
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330576 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
In the course of writing an experimental ANTLR grammar based on this
document, I found three errors in the documented BNF:
SimpleValues of dag type are allowed to have no operands at all after
the initial DagArg specifying the operator. For example, the value
(outs) is extremely common in backends; an example in the test suite
is test/TableGen/AsmVariant.td line 30. But the BNF doesn't allow
DagArgList to expand to the empty string (it must contain at least one
DagArg), and therefore the DagArgList specifying the operands in the
dag-shaped production for SimpleValue should be optional.
In the production for BodyItem with a 'let' and an optional RangeList,
the RangeList should have braces around it if it's present, matching
code such as "let E{7-0} = ..." on test/TableGen/BitsInit.td line 42.
Those braces aren't included in the RangeList nonterminal itself, so
instead they need to be part of the optional segment of the BodyItem
production.
Finally, the identifier after 'defm' should be optional. Again, this
is very common in the real back end .td files; an example in the test
suite is in test/TableGen/defmclass.td line 49.
Reviewers: rengolin, nhaehnle, stoklund
Reviewed By: nhaehnle
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D45818
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330570 91177308-0d34-0410-b5e6-96231b3b80d8
Should be a harmless trimming of trailing whitespace from a
documentation file.
(There are other instances of trailing whitespace in this file alone.
I've only fixed one of them, on the basis that that way the rest are
still available for other people's commit-access tests :-)
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330567 91177308-0d34-0410-b5e6-96231b3b80d8
