Currently, the outliner considers candidates that intersect with themselves in
the candidate pruning step. That is, candidates of the form "AA" in ranges like
"AAAAAA". In that range, it looks like there are 5 instances of "AA" that could
possibly be outlined, and that's considered in the benefit calculation.
However, only at most 3 instances of "AA" could ever be outlined in "AAAAAA".
Thus, it's possible to pass through "AA" to the candidate selection step even
though it's *never* the case that "AA" could be outlined. This makes it so that
when we find candidates, we consider only non-overlapping occurrences of that
candidate.
llvm-svn: 319588
r230670 introduced a step to map EH register numbers to standard
DWARF register numbers. This failed to consider the case when a
user .cfi_* directive uses an integer literal rather than a
register name, to specify a DWARF register number that has no
corresponding LLVM register number (e.g. a special register that
the compiler and assembler have no name for).
Fixes PR34028.
Patch by Roland McGrath
Differential Revision: https://reviews.llvm.org/D36493
llvm-svn: 319586
Previously we emitted `__tgt_target_teams` only for standalone teams
directives. This patch allows emit this function for all teams-based
directives.
llvm-svn: 319585
of its argument, perform function-to-pointer and array-to-pointer decay on the
parameter type first.
Otherwise deduction will fail, as the type of the argument will be decayed.
llvm-svn: 319584
Turns out we can have comparisons which are indirect users of the induction variable that we can make invariant. In this case, there is no loop invariant value contributing and we'd fail an assert.
The test case was found by a java fuzzer and reduced. It's a real cornercase. You have to have a static loop which we've already proven only executes once, but haven't broken the backedge on, and an inner phi whose result can be constant folded by SCEV using exit count reasoning but not proven by isKnownPredicate. To my knowledge, only the fuzzer has hit this case.
llvm-svn: 319583
These are blocks that haven't not been executed during training. For large
projects this could make a significant difference. For the project, I was
looking at, I got an order of magnitude decrease in the size of the total YAML
files with this and r319235.
Differential Revision: https://reviews.llvm.org/D40678
Re-commit after fixing the failing testcase in rL319576, rL319577 and
rL319578.
llvm-svn: 319581
The driver-based test is still not identical to the front-end line, remove the
hotness threshold from there and add a new front-end based test with
threshold.
llvm-svn: 319578
1. Require hotness on all remark lines with -verify.
3. Fix the samplePGO file to actually produce hotness on each line.
The second remark has hotness 60 rather 30 which I don't quite understand but
testing this is strictly better than before. It also unblocks the commit of
D40678.
llvm-svn: 319577
A RUN line was referring to the previous RUN line but a new test was added in
between them. Just reorder the lines.
Note this still does not completely fix this the brokenness of the comment as
the driver-based test gained a new hotness-threshold argument in r306948 but
I'll fix that is a separate commit.
llvm-svn: 319576
Summary:
As discussed in https://github.com/google/oss-fuzz/issues/933,
it would be really awesome to be able to use ThinLTO for fuzzing.
However, as @kcc has pointed out, it is currently undefined (untested)
whether the sanitizers actually function properly with LLD and/or LTO.
This patch is inspired by the cfi test, which already do test with LTO
(and/or LLD), since LTO is required for CFI to function.
I started with UBSan, because it's cmakelists / lit.* files appeared
to be the cleanest. This patch adds the infrastructure to easily add
LLD and/or LTO sub-variants of the existing lit test configurations.
Also, this patch adds the LLD flavor, that explicitly does use LLD to link.
The check-ubsan does pass on my machine. And to minimize the [initial]
potential buildbot breakage i have put some restrictions on this flavour.
Please review carefully, i have not worked with lit/sanitizer tests before.
The original attempt, r319525 was reverted in r319526 due
to the failures in compiler-rt standalone builds.
Reviewers: eugenis, vitalybuka
Reviewed By: eugenis
Subscribers: #sanitizers, pcc, kubamracek, mgorny, llvm-commits, mehdi_amini, inglorion, kcc
Differential Revision: https://reviews.llvm.org/D39508
llvm-svn: 319575
Only pass Native to LLVM_TARGETS_TO_BUILD.
Summary:
Remove the redundant, config-time call to cmake when
building host tools for cross compiles or optimized tablegen..
The config-time call to cmake is redundant because it will always get
called again when the CONFIGURE_LLVM_${target_name} target fires at
build-time. This speeds up initial configuration, but has no affect
on build behavior.
Differential Revision: https://reviews.llvm.org/D40229
llvm-svn: 319574
Per beanz, building compiler-rt standalone is a pretty important use
case, so the comment is very out of date.
Differential Revision: https://reviews.llvm.org/D40740
llvm-svn: 319570
These targets strip during installation, and are required to support
install-distribution-stripped in LLVM (to support a stripped
distribution). LLVM has an add_llvm_install_targets function for this
purpose, but we can't rely on LLVM being present.
Differential Revision: https://reviews.llvm.org/D40687
llvm-svn: 319569
Summary:
Adding support for -print-module-scope similar to how it is
being done for function passes. This option causes loop-pass printer
to emit a whole-module IR instead of just a loop itself.
Reviewers: sanjoy, silvas, weimingz
Reviewed By: sanjoy
Subscribers: apilipenko, skatkov, llvm-commits
Differential Revision: https://reviews.llvm.org/D40247
llvm-svn: 319566
PR35478 https://bugs.llvm.org/show_bug.cgi?id=35478 points out a flaw
in the implementation of r318924 from D40364. The implementation
depends on the Size field being set or the SyntheticSection::empty()
being accurate. These functions are not reliable as some linker script
commands that have yet to be processed may affect the results, causing
some non-zero size sections to be reported as zero size.
I think the first step is to revert r318924 and come up with a better
solution for the underlying problem rather than trying to layer more
heuristics onto the zero sized output section.
Chances are I'll be out of office by the time anyone sees this so feel
free to commit the revert if you agree with me.
Fixes PR35478
Current thoughts on the underlying problem:
Revisiting the motivation for adding the zero size check in the first
place; it was to prevent 0 sized SyntheticSections that a user does
not have full control over from needlessly breaking the PT_GNU_RELRO,
rather than trying to accommodate arbitrarily complex linker
scripts. Looking at the code, it looks like
removeUnusedSyntheticSections() should remove zero sized synthetic
sections. It does, but it doesn't set the Parent to nullptr, this has
the side effect that Sec == InX::BssRelRo->getParent() will make the
parent OutputSection of InX::BssRelRo RelRo even if there is no
InX::BssRelRo.
I tried a quick experiment with setting the Parent to nullptr and this
flushed out a few interesting test failures, it feels like playing
Jenga with every change:
In the isRelroSection() we have to consider the case where there
is no .plt and .plt.got but there is a ifunc plt with accompanying
(ifunc .got or .plt.got)
The PPC64 has PltHeaderSize == 0. Unfortunately HeaderSize == 0 is
used to choose between the ifunc plt or normal plt. We seem to get
away with this at the moment, but tests start to fail when Parent
is set to nullptr for the .got.plt.
The InX::BssRelRo and InX::Bss never get their sizes set and they
are always removed by removeUnusedSyntheticSections(), their
purpose seems to be as some kind of proxy for add .bss or
.bss.relro InputSections into their parent OutputSections, they
therefore don't behave like other SyntheticSections anyway.
My thinking is that some work is needed to make sure that the Sec ==
SyntheticSection->getParent() does a bit more checking before
returning true, particularly for InX::BssRelRo as that has special
behaviour. I'll hope to post something for review as soon as possible.
Patch by Peter Smith!
llvm-svn: 319563
Summary:
When debugging function passes it happens to be rather useful to dump
the whole module before the transformation and then use this dump
to analyze this single transformation by running it separately
on that particular module state.
Introducing
-print-module-scope
debugging option that forces all the function-level IR dumps
to become whole-module dumps.
This option builds on top of normal dumping controls like
-print-before/after
-filter-print-funcs
The plan is to eventually extend this option to cover other local passes
(at least loop passes) but that should go as a separate change.
Reviewers: sanjoy, weimingz, silvas, fedor.sergeev
Reviewed By: weimingz
Subscribers: apilipenko, skatkov, llvm-commits, mehdi_amini
Differential Revision: https://reviews.llvm.org/D40245
llvm-svn: 319561
distribute directives.
OpenMP standard does not allow to mark the variables as firstprivate and lastprivate at the same time in distribute-based directives. Patch fixes this problem.
llvm-svn: 319560
Summary:
This will be used for rescoring code completion results based on partial
identifiers.
Short-term use:
- we want to limit the number of code completion results returned to
improve performance of global completion. The scorer will be used to
rerank the results to return when the user has applied a filter.
Long-term use case:
- ranking of completion results from in-memory index
- merging of completion results from multiple sources (merging usually
works best when done at the component-score level, rescoring the
fuzzy-match quality avoids different backends needing to have
comparable scores)
Reviewers: ilya-biryukov
Subscribers: cfe-commits, mgorny
Differential Revision: https://reviews.llvm.org/D40060
llvm-svn: 319557
These are blocks that haven't not been executed during training. For large
projects this could make a significant difference. For the project, I was
looking at, I got an order of magnitude decrease in the size of the total YAML
files with this and r319235.
Differential Revision: https://reviews.llvm.org/D40678
llvm-svn: 319556
Summary:
This allows us to limit the number of results we return and still allow them
to be surfaced by refining a query (D39852).
The initial algorithm is very conservative - it accepts a completion if the
filter is any case-insensitive sub-sequence. It does not attempt to rank items
based on match quality.
Reviewers: ilya-biryukov
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D39882
llvm-svn: 319552
It causes builds to fail with "Instruction does not dominate all uses" (PR35497).
> Patch tries to improve vectorization of the following code:
>
> void add1(int * __restrict dst, const int * __restrict src) {
> *dst++ = *src++;
> *dst++ = *src++ + 1;
> *dst++ = *src++ + 2;
> *dst++ = *src++ + 3;
> }
> Allows to vectorize even if the very first operation is not a binary add, but just a load.
>
> Fixed issues related to previous commit.
>
> Reviewers: spatel, mzolotukhin, mkuper, hfinkel, RKSimon, filcab, ABataev
>
> Reviewed By: ABataev, RKSimon
>
> Subscribers: llvm-commits, RKSimon
>
> Differential Revision: https://reviews.llvm.org/D28907
llvm-svn: 319550
Added some commonly used Arm triples to the script, with and without
the -eabi suffix.
Differential Revision: https://reviews.llvm.org/D40708
llvm-svn: 319545
Summary:
1/ Operand folding during complex pattern matching for LEAs has been extended, such that it promotes Scale to
accommodate similar operand appearing in the DAG e.g.
T1 = A + B
T2 = T1 + 10
T3 = T2 + A
For above DAG rooted at T3, X86AddressMode will now look like
Base = B , Index = A , Scale = 2 , Disp = 10
2/ During OptimizeLEAPass down the pipeline factorization is now performed over LEAs so that if there is an opportunity
then complex LEAs (having 3 operands) could be factored out e.g.
leal 1(%rax,%rcx,1), %rdx
leal 1(%rax,%rcx,2), %rcx
will be factored as following
leal 1(%rax,%rcx,1), %rdx
leal (%rdx,%rcx) , %edx
3/ Aggressive operand folding for AM based selection for LEAs is sensitive to loops, thus avoiding creation of any complex LEAs within a loop.
4/ Simplify LEA converts (lea (BASE,1,INDEX,0) --> add (BASE, INDEX) which offers better through put.
PR32755 will be taken care of by this pathc.
Previous patch revisions : r313343 , r314886
Reviewers: lsaba, RKSimon, craig.topper, qcolombet, jmolloy, jbhateja
Reviewed By: lsaba, RKSimon, jbhateja
Subscribers: jmolloy, spatel, igorb, llvm-commits
Differential Revision: https://reviews.llvm.org/D35014
llvm-svn: 319543
Sam Clegg