There is no pattern matched `add hi, (MipsLo texternalsym)`. As a result,
loading an address of 32-bit symbol requires two registers and one more
additional instruction:
```
addiu $1, $zero, %lo(foo)
lui $2, %hi(foo)
addu $25, $2, $1
```
This patch adds the missed pattern and enables generation more effective
set of instructions:
```
lui $1, %hi(foo)
addiu $25, $1, %lo(foo)
```
Differential Revision: https://reviews.llvm.org/D66771
llvm-svn: 370196
Apparently inline tests stop running anything after an empty line
behind an self.expect, which is a very good approach that could
never cause people to write tests that never run.
This patch removes all the empty lines so that all this test
is actually run. Also fixes the broken expects that only passed
because they weren't run before.
llvm-svn: 370195
Port the D64906 technique to RISC-V. It deletes 3 alignments at
PT_LOAD boundaries for the default case: the size of a RISC-V binary
decreases by at most 12kb.
llvm-svn: 370192
Summary: There are at least 2 ways to express the same shuffle. Various pieces of code explicit check for both option, but other places do not when they would benefit from doing it. This patches refactor the codebase to use buildLegalVectorShuffle in order to make that behavior more consistent.
Reviewers: craig.topper, efriedma, RKSimon, lebedev.ri
Subscribers: javed.absar, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D66804
llvm-svn: 370190
This just pulls the MVEVPTBlockPass into a separate file, as opposed to being
wrapped up in Thumb2ITBlockPass.
Differential revision: https://reviews.llvm.org/D66579
llvm-svn: 370187
The refactoring patch for the option completion broke the completion
for ambiguous long options. As this feature was also untested (as
testing ambiguous options with the current test methods is impossible),
I just noticed now. This patch restores the old behavior and adds a
test for this feature.
llvm-svn: 370185
This adds fp16 VMOVX patterns, using the same patterns as rL362482 with some
adjustments for MVE. It allows us to move fp16 registers without going into and
out of gprs.
VMOVX is able to move the top bits from a fp16 in a fp reg into the bottom bits
of another register, zeroing the rest. This can be used for odd MVE register
lanes. The top bits are not read by fp16 instructions, so no move is required
there if we are dealing with even lanes.
Differential revision: https://reviews.llvm.org/D66793
llvm-svn: 370184
Summary:
Those errors are exposed at the first character of a file,
for a lack of a better place.
Previously, all errors were stored inside the AST and report
accordingly. However, errors in command-line argument parsing could
result in failure to produce the AST, so we need an alternative ways to
report those errors.
We take the following approach in this patch:
- buildCompilerInvocation() now requires an explicit DiagnosticConsumer.
- TUScheduler and TestTU now collect the diagnostics produced when
parsing command line arguments.
If pasing of the AST failed, diagnostics are reported via a new
ParsingCallbacks::onFailedAST method.
If parsing of the AST succeeded, any errors produced during
command-line parsing are stored alongside the AST inside the
ParsedAST instance and reported as previously by calling the
ParsingCallbacks::onMainAST method;
- The client code that uses ClangdServer's DiagnosticConsumer
does not need to change, it will receive new diagnostics in the
onDiagnosticsReady() callback
Errors produced when parsing command-line arguments are collected using
the same StoreDiags class that is used to collect all other errors. They
are recognized by their location being invalid. IIUC, the location is
invalid as there is no source manager at this point, it is created at a
later stage.
Although technically we might also get diagnostics that mention the
command-line arguments FileID with after the source manager was created
(and they have valid source locations), we choose to not handle those
and they are dropped as not coming from the main file. AFAICT, those
diagnostics should always be notes, therefore it's safe to drop them
without loosing too much information.
Reviewers: kadircet
Reviewed By: kadircet
Subscribers: nridge, javed.absar, MaskRay, jkorous, arphaman, cfe-commits, gribozavr
Tags: #clang
Differential Revision: https://reviews.llvm.org/D66759
llvm-svn: 370177
With the introduction of the typed byval attribute change there was no
way that the LLVM-C API could create the correct class Attribute. If a
program that uses the C API creates a ByVal attribute and annotates a
function with that attribute LLVM will crash when it assembles or write
that module containing the function out as bitcode.
This change is a minimal fix to at least allow code to work, this is
because the byval change is on the 9.0 and I don't want to introduce new
LLVM-C API this late in the release cycle.
By Jakob Bornecrantz!
Differential revision: https://reviews.llvm.org/D66144
llvm-svn: 370176
It was added in r370129 with a .gitattributes file that means the file
always shows up as having a local diff in Git checkouts (at least on
Linux). Deleting it until we can figure out the right way to do this.
llvm-svn: 370175
Allow vectorizing loops that have reductions when tail is folded by masking.
A select is introduced in VPlan, choosing between the last value carried by the
loop-exit/live-out instruction of the reduction, and the penultimate value
carried by the reduction phi, according to the "i < n" mask of fold-tail.
This select replaces the last value as the live-out value of the loop.
Differential Revision: https://reviews.llvm.org/D66720
llvm-svn: 370173
This essentially reverts the code change of D63132 and switches to a simpler approach.
In an executable/shared object, st_shndx of a symbol can be:
1) SHN_UNDEF: undefined symbol (or canonical PLT)
2) SHN_ABS: absolute symbol
3) any other value (usually a regular section index) represents a relative symbol.
The actual value does not matter.
Many ld.so (musl, all archs except MIPS of FreeBSD rtld-elf) even treat 2) and 3)
the same. If .sdata does not exist, it does not matter what value/section
__global_pointer$ has, as long as it is relative (otherwise there will be a pedantic
lld error. See D63132). Just set the st_shndx arbitrarily to 1.
Dummy st_shndx=1 may be used by __rela_iplt_start, linker-script-defined symbols outside a section, __dso_handle, etc.
Reviewed By: ruiu
Differential Revision: https://reviews.llvm.org/D66798
llvm-svn: 370172
rL369567 reverted a couple of recent changes made to ARMParallelDSP
because of a miscompilation error: PR43073.
The issue stemmed from an underlying bug that was caused by adding
muls into a reduction before it was proved that they could be executed
in parallel with another mul.
Most of the changes here are from the previously reverted commits.
The additional changes have been made area:
1) The Search function now doesn't insert any muls into the Reduction
object. That now happens once the search has successfully finished.
2) For any muls added into the reduction but that weren't paired, we
accumulate their values as an input into the smlad.
Differential Revision: https://reviews.llvm.org/D66660
llvm-svn: 370171
"Bad block found.\n" -> "bad block found"
The lower cased form with no full stop or newline is more common in LLVM
tools.
Reviewed By: juliehockett
Differential Revision: https://reviews.llvm.org/D66783
llvm-svn: 370155
This fixes sanitizer-x86_64-linux-fuzzer which switched to new layout and not it
can't link as libcxx is always there.
We should support and tests libcxx but still it was unexpected that libfuzzer
ignored LLVM_ENABLE_PROJECTS_USED.
llvm-svn: 370153
A new SPI was added to the objc runtime to get class names without
any demangling; AppleObjCRuntimeV2::ParseClassInfoArray was using
the original prototype name but had not been updated for the final
name yet, so lldb was falling back to the old function and doing
extra work for classes that were demangled. This commit fixes that.
llvm-svn: 370152
Summary:
[libomptarget] Refactor syncwarp macro to inline function
See also abandoned D66846, split into this diff and others.
Reviewers: jdoerfert, ABataev, grokos, ronlieb, gregrodgers
Subscribers: openmp-commits
Tags: #openmp
Differential Revision: https://reviews.llvm.org/D66857
llvm-svn: 370149
Summary:
[libomptarget] Refactor shfl_down_sync macro to inline function
See also abandoned D66846, split into this diff and others.
Reviewers: jdoerfert, ABataev, grokos, ronlieb, gregrodgers
Subscribers: openmp-commits
Tags: #openmp
Differential Revision: https://reviews.llvm.org/D66853
llvm-svn: 370146
Summary:
[libomptarget] Refactor shfl_sync macro to inline function
See also abandoned D66846, split into this diff and others.
Reviewers: jdoerfert, ABataev, grokos, ronlieb, gregrodgers
Subscribers: openmp-commits
Tags: #openmp
Differential Revision: https://reviews.llvm.org/D66852
llvm-svn: 370144
Bundler currently requires host triple to be provided no matter if you are performing bundling or unbundling, but for unbundling operation such requirement is too restrictive. You may for example want to examine device part of the object for a particular offload target, but you have to extract host part as well even though you do not need it. Host triple isn't really needed for unbundling, so this patch removes that requirement.
Differential Revision: https://reviews.llvm.org/D66601
llvm-svn: 370143
Setting DYLD_INSERT_LIBRARIES to the Asan runtime and DYLD_LIBRARY_PATH
to the LLVM shared library dir causes the test suite to crash with a
segfault. We see this on the LLDB sanitized bot [1] on GreenDragon. I've
spent some time investigating, but I'm not sure what's going on (yet).
Originally I thought this was because we were building compiler-rt and
were loading an incompatible, just-built Asan library. However, the
issue persists even without compiler-rt. It doesn't look like the Asan
runtime is opening any other libraries that might be found in LLVM's
shared library dir and talking to the team confirms that. Another
possible explanation is that we're loading lldb form a place we don't
expect, but that doesn't make sense either, because DYLD_LIBRARY_PATH is
always set without the crash. I tried different Python versions and
interpreters but the issue persist.
As a (temporary?) workaround I propose not setting DYLD_LIBRARY_PATH
when DYLD_INSERT_LIBRARIES is set so we can turn the Asan bot on again
and get useful results.
[1] http://green.lab.llvm.org/green/view/LLDB/job/lldb-cmake-sanitized/
Differential revision: https://reviews.llvm.org/D66845
llvm-svn: 370135
When checking if block types are compatible, we are checking for
compatibility their return types and parameters' types. As these types
have different variance, we need to check them in different order.
rdar://problem/52788423
Reviewers: erik.pilkington, arphaman
Reviewed By: arphaman
Subscribers: jkorous, dexonsmith, ributzka, cfe-commits
Differential Revision: https://reviews.llvm.org/D66831
llvm-svn: 370130
Previously, an #error directive with quoted, multi-line content, along with CR+LF line endings wasn't handled correctly.
Differential Revision: https://reviews.llvm.org/D66556
llvm-svn: 370129
Simon Pilgrim