This is a revised version of D41474.
When the debug location is parsed in BitcodeReader::parseFunction, the
scope and inlinedAt MDNodes are obtained via MDLoader->getMDNodeFwdRefOrNull(),
which will create a forward ref if they were not yet loaded.
Specifically, if one of these MDNodes is in the module level metadata
block, and this is during ThinLTO importing, that metadata block is
lazily loaded.
Most places in that invoke getMDNodeFwdRefOrNull have a corresponding call
to resolveForwardRefsAndPlaceholders which will take care of resolving them.
E.g. places that call getMetadataFwdRefOrLoad, or at the end of parsing a
function-level metadata block, or at the end of the initial lazy load of
module level metadata in order to handle invocations of getMDNodeFwdRefOrNull
for named metadata and global object attachments. However, the calls for
the scope/inlinedAt of debug locations are not backed by any such call to
resolveForwardRefsAndPlaceholders.
To fix this, change the scope and inlinedAt parsing to instead use
getMetadataFwdRefOrLoad, which will ensure the forward refs to lazily
loaded metadata are resolved.
Fixes PR35472.
llvm-svn: 346891
This patch renames the cxx-benchmark-unittests to check-cxx-benchmarks
and converts the target to use LIT in order to make the tests run faster
and provide better output.
In particular this runs each benchmark in a suite one by one, allowing
more parallelism while ensuring output isn't garbage with multiple threads.
Additionally, it adds the CMake flag '-DLIBCXX_BENCHMARK_TEST_ARGS=<list>'
to specify what options are passed when running the benchmarks.
llvm-svn: 346888
A call to @llvm.trap can be expected to be cold (i.e. unlikely to be
reached in a normal program execution).
Outlining paths which unconditionally trap is an important memory
saving. As the hot/cold splitting pass (imho) should not treat all
noreturn calls as cold, explicitly mark @llvm.trap cold so that it can
be outlined.
Split out of https://reviews.llvm.org/D54244.
Differential Revision: https://reviews.llvm.org/D54329
llvm-svn: 346885
Add support for "polymorphic" types to YAMLIO.
PolymorphicTraits can dynamically switch between other traits (Scalar, Map, or
Sequence). When inputting, the PolymorphicTraits type is told which type to
become, and when outputting the PolymorphicTraits type is asked which type it
currently is.
Also add support for TaggedScalarTraits to allow dynamically differentiating
between multiple scalar types using YAML tags.
Serialize empty maps as "{}" and empty sequences as "[]", so that types
are preserved when round-tripping PolymorphicTraits. This change has
equivalent semantics, but may break e.g. tests which compare output
verbatim.
Differential Revision: https://reviews.llvm.org/D48144
llvm-svn: 346884
Summary:
Previously we marked all vararg functions as non-inlinable in the
function summary, which prevented their importing. However, the
corresponding inliner restriction was loosened in r321940/r342675
to only apply to functions calling va_start. Adjust the summary
flag computation to match.
Reviewers: davidxl
Subscribers: mehdi_amini, inglorion, eraman, steven_wu, dexonsmith, llvm-commits
Differential Revision: https://reviews.llvm.org/D54270
llvm-svn: 346883
Narrower vectors will be widened to 128 bits without changing the element size. And generic type legalization can already handle widening mulhu/mulhs.
Differential Revision: https://reviews.llvm.org/D54513
llvm-svn: 346879
Summary:
The result of subprocess.check_output() is bytes in python3 which we need
to convert to str(). Simplify this by using the executeCommand() helper.
Reviewers: ldionne, EricWF
Reviewed By: ldionne
Subscribers: christof, libcxx-commits
Differential Revision: https://reviews.llvm.org/D54522
llvm-svn: 346878
On PowerPC64, when a function call offset is too large to encode in a call
instruction the address is stored in a table in the data segment. A thunk is
used to load the branch target address from the table relative to the
TOC-pointer and indirectly branch to the callee. When linking position-dependent
code the addresses are stored directly in the table, for position-independent
code the table is allocated and filled in at load time by the dynamic linker.
For position-independent code the branch targets could have gone in the .got.plt
but using the .branch_lt section for both position dependent and position
independent binaries keeps it consitent and helps keep this PPC64 specific logic
seperated from the target-independent code handling the .got.plt.
Differential Revision: https://reviews.llvm.org/D53408
llvm-svn: 346877
Summary:
These asserts are based on the assumption that the order of true/false operands in a select and those in the compare would always be the same.
This fixes PR39595.
Reviewers: craig.topper, spatel, dmgreen
Reviewed By: craig.topper
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D54359
llvm-svn: 346874
This fixes two compilation failures:
1) Designated initializers are C++20. We can't use them in LLVM.
2) thread_result_t is not a pointer type on all platforms, so
returning nullptr is an error.
llvm-svn: 346873
Regenerate this test using update_test_checks.py in preparation for an
upcomming commit, to make it not depend on the names of instructions.
llvm-svn: 346869
The DataExtractor class itself was moved to Utility some time ago, but
it seems this was not reflected in the location of the test code. Fix
that.
llvm-svn: 346867
Summary:
GCC already catches these situations so we should handle it too.
GCC warns in C++ mode only (does anybody know why?). I think it is useful in C mode too.
Reviewers: rsmith, erichkeane, aaron.ballman, efriedma, xbolva00
Reviewed By: xbolva00
Subscribers: efriedma, craig.topper, scanon, cfe-commits
Differential Revision: https://reviews.llvm.org/D52835
llvm-svn: 346865
Reorder the bit-field classes and the members of the anonymous union
so that they both match the order in StmtNodes.td.
There is already a fair amount of them, and this is not going to
improve. Therefore lets try to keep some order here.
Strictly NFC.
llvm-svn: 346864
The R_AARCH64_ADR_PREL_PG_HI21 relocation type is given the R_PAGE_PC
RelExpr. This can be transformed to R_PLT_PAGE_PC via toPlt().
Unfortunately the resolution is identical to R_PAGE_PC so instead of
getting the address of the PLT entry we get the address of the symbol
which may not be correct in the case of static ifuncs. The fix is to
handle the cases separately and use getPltVA() + A with R_PLT_PAGE_PC.
Differential Revision: https://reviews.llvm.org/D54474
llvm-svn: 346863
This patch adds an initial implementation of the look-ahead SLP tree
construction described in 'Look-Ahead SLP: Auto-vectorization in the Presence
of Commutative Operations, CGO 2018 by Vasileios Porpodas, Rodrigo C. O. Rocha,
Luís F. W. Góes'.
It returns an SLP tree represented as VPInstructions, with combined
instructions represented as a single, wider VPInstruction.
This initial version does not support instructions with multiple
different users (either inside or outside the SLP tree) or
non-instruction operands; it won't generate any shuffles or
insertelement instructions.
It also just adds the analysis that builds an SLP tree rooted in a set
of stores. It does not include any cost modeling or memory legality
checks. The plan is to integrate it with VPlan based cost modeling, once
available and to only apply it to operations that can be widened.
A follow-up patch will add a support for replacing instructions in a
VPlan with their SLP counter parts.
Reviewers: Ayal, mssimpso, rengolin, mkuper, hfinkel, hsaito, dcaballe, vporpo, RKSimon, ABataev
Reviewed By: rengolin
Differential Revision: https://reviews.llvm.org/D4949
llvm-svn: 346857
Test cases were updated to not use the local compilation dir which
is different between development pc and build bots.
Original commit message:
[LLDB] - Support the single file split DWARF.
DWARF5 spec describes a single file split dwarf case
(when .dwo sections are in the .o files).
Problem is that LLDB does not work correctly in that case.
The issue is that, for example, both .debug_info and .debug_info.dwo
has the same type: eSectionTypeDWARFDebugInfo. And when code searches
section by type it might find the regular debug section
and not the .dwo one.
The patch fixes that. With it, LLDB is able to work with
output compiled with -gsplit-dwarf=single flag correctly.
Differential revision: https://reviews.llvm.org/D52403
llvm-svn: 346855
Add support for the expansion of funnelshift/rotates to getIntrinsicInstrCost.
This also required us to move the X86 fshl/fshr costs to the same place as the rotates to avoid expansion and get correct scalarization vs vectorization costs.
llvm-svn: 346854
Summary:
This would save us 8 bytes per ref, and buy us ~40MB in total
for llvm index (from ~300MB to ~260 MB).
The char pointer must be null-terminated, and llvm::StringSaver
guarantees it.
Reviewers: sammccall
Subscribers: ilya-biryukov, ioeric, MaskRay, jkorous, arphaman, kadircet, cfe-commits
Differential Revision: https://reviews.llvm.org/D53427
llvm-svn: 346852
This patch removes the last use of the constant pool shuffle decode helper and consistently uses the 'getTargetShuffleMaskIndices' versions instead. The constant pool versions are now purely used for assembly comments.
The avx512vbmi intrinsic upgrades had to be altered as they were being decoded as broadcasts, similar to what I fixed in rL346032. I don't think the change is critical - although its annoying that we lose the {k}{z} instruction test coverage as they are tricky to generate....
Differential Revision: https://reviews.llvm.org/D54083
llvm-svn: 346850
Summary:
While parsing a childless compile unit DIE we could crash if the DIE was
followed by any extra data (such as a superfluous end-of-children
marker). This happened because the break-on-depth=0 check was performed
only when parsing the null DIE, which was not correct because with a
childless root DIE, we could reach the end of the unit without ever
encountering the null DIE.
If the compile unit contribution ended directly after the CU DIE,
everything would be fine as we would terminate parsing due to reaching
EOF. However, if the contribution contained extra data (perhaps a
superfluous end-of-children marker), we would crash because we would
treat that data as the begging of another compile unit.
This fixes the crash by moving the depth=0 check to a more generic
place, and also adds a regression test.
Reviewers: clayborg, jankratochvil, JDevlieghere
Subscribers: lldb-commits
Differential Revision: https://reviews.llvm.org/D54417
llvm-svn: 346849
DWARF5 spec describes a single file split dwarf case
(when .dwo sections are in the .o files).
Problem is that LLDB does not work correctly in that case.
The issue is that, for example, both .debug_info and .debug_info.dwo
has the same type: eSectionTypeDWARFDebugInfo. And when code searches
section by type it might find the regular debug section
and not the .dwo one.
The patch fixes that. With it, LLDB is able to work with
output compiled with -gsplit-dwarf=single flag correctly.
Differential revision: https://reviews.llvm.org/D52296
llvm-svn: 346848
Summary:
The goal is to allow analyses such as clang-tidy checks to run on a
subset of the AST, e.g. "only on main-file decls" for interactive tools.
Today, these become "problematically global" by running RecursiveASTVisitors
rooted at the TUDecl, or by navigating up via ASTContext::getParent().
The scope is restricted using a set of top-level-decls that RecursiveASTVisitors
should be rooted at. This also applies to the visitor that populates the
parent map, and so the top-level-decls are considered to have no parents.
This patch makes the traversal scope a mutable property of ASTContext.
The more obvious way to do this is to pass the top-level decls to
relevant functions directly, but this has some problems:
- it's error-prone: accidentally mixing restricted and unrestricted
scopes is a performance trap. Interleaving multiple analyses is
common (many clang-tidy checks run matchers or RAVs from matcher callbacks)
- it doesn't map well to the actual use cases, where we really do want
*all* traversals to be restricted.
- it involves a lot of plumbing in parts of the code that don't care
about traversals.
This approach was tried out in D54259 and D54261, I wanted to like it
but it feels pretty awful in practice.
Caveats: to get scope-limiting behavior of RecursiveASTVisitors, callers
have to call the new TraverseAST(Ctx) function instead of TraverseDecl(TU).
I think this is an improvement to the API regardless.
Reviewers: klimek, ioeric
Subscribers: mgorny, cfe-commits
Differential Revision: https://reviews.llvm.org/D54309
llvm-svn: 346847
rL346840 temporarily marked event-section.ll as XFAIL because it was
failing for builds with LLVM_ENABLE_EXPENSIVE_CHECKS turned on, but
to make sure it XFAILs even without LLVM_ENABLE_EXPENSIVE_CHECKS on we
need this `-verify-machineinstrs` flag, which was missing in the
previous commit.
llvm-svn: 346845
Before this commit, `llc -print-after-all` would print something like:
*** IR Dump After Pre-ISel Intrinsic Lowering ***; ModuleID = ...
Emit a newline such that ModuleID appears on a line by its own.
llvm-svn: 346844
error: implicit conversion from 'double' to 'uint64_t' (aka 'unsigned long') changes value from -0 to 0 [-Werror,-Wliteral-conversion]
llvm-svn: 346841
This test is failing in builds with LLVM_ENABLE_EXPENSIVE_CHECKS after
rL346825 not because of the patch but due to a pre-existing codegen
problem. Marking this as XFAIL temporarily until the bug is fixed.
llvm-svn: 346840
Teresa Johnson