Sjoerd Meijer [Thu, 23 Dec 2021 15:22:46 +0000 (15:22 +0000)]
[AArch64] Remove outdated FIXME in test arm64-csel.ll. NFC.
Pavel Labath [Wed, 29 Dec 2021 09:54:46 +0000 (10:54 +0100)]
[lldb/linux] Fix a bug in wait status handling
The MonitorCallback function was assuming that the "exited" argument is
set whenever a thread exits, but the caller was only setting that flag
for the main thread.
This patch deletes the argument altogether, and lets MonitorCallback
compute what it needs itself.
This is almost NFC, since previously we would end up in the
"GetSignalInfo failed for unknown reasons" branch, which was doing the
same thing -- forgetting about the thread.
PoYao Chang [Mon, 27 Dec 2021 04:34:23 +0000 (12:34 +0800)]
[lldb] Fix PR52702 by fixing bool conversion of Mangled
Remove the Mangled::operator! and Mangled::operator void* where the
comments in header and implementation files disagree and replace them
with operator bool.
This fix PR52702 as https://reviews.llvm.org/D106837 used the buggy
Mangled::operator! in Symbol::SynthesizeNameIfNeeded. For example,
consider the symbol "puts" in a hello world C program:
// Inside Symbol::SynthesizeNameIfNeeded
(lldb) p m_mangled
(lldb_private::Mangled) $0 = (m_mangled = None, m_demangled = "puts")
(lldb) p !m_mangled
(bool) $1 = true # should be false!!
This leads to Symbol::SynthesizeNameIfNeeded overwriting m_demangled
part of Mangled (in this case "puts").
In conclusion, this patch turns
callq 0x401030 ; symbol stub for: ___lldb_unnamed_symbol36
back into
callq 0x401030 ; symbol stub for: puts .
Differential Revision: https://reviews.llvm.org/D116217
Pavel Labath [Wed, 29 Dec 2021 09:00:00 +0000 (10:00 +0100)]
[lldb] Adjust TestModuleCacheSimple for D115951
Now that we are caching the dwarf index as well, we will always have
more than one cache file (when not using accelerator tables). I have
adjusted the test to check for the presence of one _symtab_ index.
Pavel Labath [Tue, 21 Dec 2021 12:59:14 +0000 (13:59 +0100)]
[lldb] Make ProcessLauncherPosixFork (mostly) async-signal-safe
Multithreaded applications using fork(2) need to be extra careful about
what they do in the fork child. Without any special precautions (which
only really work if you can fully control all threads) they can only
safely call async-signal-safe functions. This is because the forked
child will contain snapshot of the parents memory at a random moment in
the execution of all of the non-forking threads (this is where the
similarity with signals comes in).
For example, the other threads could have been holding locks that can
now never be released in the child process and any attempt to obtain
them would block. This is what sometimes happen when using tcmalloc --
our fork child ends up hanging in the memory allocation routine. It is
also what happened with our logging code, which is why we added a
pthread_atfork hackaround.
This patch implements a proper fix to the problem, by which is to make
the child code async-signal-safe. The ProcessLaunchInfo structure is
transformed into a simpler ForkLaunchInfo representation, one which can
be read without allocating memory and invoking complex library
functions.
Strictly speaking this implementation is not async-signal-safe, as it
still invokes library functions outside of the posix-blessed set of
entry points. Strictly adhering to the spec would mean reimplementing a
lot of the functionality in pure C, so instead I rely on the fact that
any reasonable implementation of some functions (e.g.,
basic_string::c_str()) will not start allocating memory or doing other
unsafe things.
The new child code does not call into our logging infrastructure, which
enables us to remove the pthread_atfork call from there.
Differential Revision: https://reviews.llvm.org/D116165
Chuanqi Xu [Wed, 29 Dec 2021 08:22:26 +0000 (16:22 +0800)]
[clang] Fix crash in bug52905
The root cause for the crash is the incorrect use of `cast`.
The actual type and cast-to type is different. This patch fixes the
crash by converting the `cast` to `dyn_cast`.
Kazu Hirata [Wed, 29 Dec 2021 08:16:40 +0000 (00:16 -0800)]
[clang] Fix header guards (NFC)
Identified with llvm-header-guard.
Kazu Hirata [Wed, 29 Dec 2021 07:47:25 +0000 (23:47 -0800)]
[Hexagon] Use range-based for loops (NFC)
Johannes Doerfert [Wed, 29 Dec 2021 07:29:07 +0000 (01:29 -0600)]
[OpenMP][OpenACC] Update test after encoding change in D113126
Johannes Doerfert [Wed, 29 Dec 2021 07:07:53 +0000 (01:07 -0600)]
Reapply "[OpenMP][NFCI] Embed the source location string size in the ident_t"
This reverts commit
73ece231ee0cf048d56841f47915beb1db6afc26 and
reapplies
7bfcdbcbf368cea14a5236080af975d5878a46eb with mlir changes.
Also reverts commit
423ba12971bac8397c87fcf975ba6a4b7530ed28 and
includes the unit test changes of
16da2140045808b2aea1d28366ca7d326eb3c809.
Mehdi Amini [Wed, 29 Dec 2021 07:08:58 +0000 (07:08 +0000)]
Revert "[OpenMP][FIX] Also update unit test after API change"
This reverts commit
16da2140045808b2aea1d28366ca7d326eb3c809.
Revert unit-test API update after previous revert of the API change.
Mehdi Amini [Wed, 29 Dec 2021 06:57:36 +0000 (06:57 +0000)]
Revert "[OpenMP][NFCI] Embed the source location string size in the ident_t"
This reverts commit
7bfcdbcbf368cea14a5236080af975d5878a46eb.
Broke MLIR build
Kazu Hirata [Wed, 29 Dec 2021 06:52:56 +0000 (22:52 -0800)]
[AST] Fix a warning
This patch fixes:
mlir/include/mlir/Tools/PDLL/AST/Types.h:54:3: error: definition of
implicit copy assignment operator for 'Type' is deprecated because
it has a user-declared copy constructor [-Werror,-Wdeprecated-copy]
Johannes Doerfert [Wed, 29 Dec 2021 05:59:35 +0000 (23:59 -0600)]
[Attributor] Look through allocated heap memory
AAPointerInfo, and thereby other places, can look already through
internal global and stack memory. This patch enables them to look
through heap memory returned by functions with a `noalias` return.
In the future we can look through `noalias` arguments as well but that
will require AAIsDead to learn that such memory can be inspected by the
caller later on. We also need teach AAPointerInfo about dominance to
actually deal with memory that might not be `null` or `undef`
initialized. D106397 is a first step in that direction already.
Reviewed By: kuter
Differential Revision: https://reviews.llvm.org/D109170
Johannes Doerfert [Wed, 29 Dec 2021 05:55:32 +0000 (23:55 -0600)]
[OpenMP] Simplify all stores in the device code
Similar to loads, we want to be aggressive when it comes to store
simplification. Not everything in LLVM handles dead stores well when
address space casts are involved, we can simply ask the Attributor to do
it for us though.
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D109998
Johannes Doerfert [Wed, 29 Dec 2021 06:16:56 +0000 (00:16 -0600)]
[OpenMP][FIX] Also update unit test after API change
Johannes Doerfert [Thu, 4 Nov 2021 13:48:19 +0000 (08:48 -0500)]
[OpenMP][NFC] Extract assumption helpers into own header file
Johannes Doerfert [Wed, 3 Nov 2021 17:28:11 +0000 (12:28 -0500)]
[OpenMP][NFCI] Embed the source location string size in the ident_t
One of the unused ident_t fields now holds the size of the string
(=const char *) field so we have an easier time dealing with those
in the future.
Differential Revision: https://reviews.llvm.org/D113126
Johannes Doerfert [Sat, 11 Sep 2021 23:15:53 +0000 (18:15 -0500)]
[Attributor][FIX] Ensure store uses are correlated with reloads
While we skipped uses in stores if we can find all copies of the value
when the memory is loaded, we did not correlate the use in the store
with the use in the load. So far this lead to less precise results in the
offset calculations which prevented deductions. With the new
EquivalentUseCB callback argument the user of checkForAllUses can be
informed of the correlation and act on it appropriately.
Differential Revision: https://reviews.llvm.org/D109662
Johannes Doerfert [Tue, 28 Dec 2021 22:03:16 +0000 (16:03 -0600)]
[OpenMP][FIX] Make AAExecutionDomain deterministic
Johannes Doerfert [Sun, 26 Dec 2021 20:33:37 +0000 (14:33 -0600)]
[OpenMP][FIX] Make heap2shared deterministic
Issue #52875 reported non-determinism, this is the first step to avoid
it. We iterate over MallocCalls so we should keep the order stable.
Johannes Doerfert [Wed, 3 Nov 2021 17:57:43 +0000 (12:57 -0500)]
[OpenMP][NFC] Move address space enum into OMPConstants header
Johannes Doerfert [Wed, 3 Nov 2021 16:58:05 +0000 (11:58 -0500)]
[OpenMP][NFC] Move headers into include folder
William S. Moses [Tue, 28 Dec 2021 23:40:54 +0000 (18:40 -0500)]
[MLIR] Expose atomicrmw and/or
LLVM (dialect and IR) have atomics for and/or. This patch enables atomic_rmw ops in the standard dialect for and/or that lower to these (in addition to the existing atomics such as addi, etc).
Reviewed By: mehdi_amini
Differential Revision: https://reviews.llvm.org/D116345
William S. Moses [Mon, 27 Dec 2021 21:07:11 +0000 (16:07 -0500)]
[MLIR] Add constant folder for fptosi and friends
This patch adds constant folds for FPToSI/FPToUI/SIToFP/UIToFP
Reviewed By: mehdi_amini, bondhugula
Differential Revision: https://reviews.llvm.org/D116321
Joe Loser [Wed, 22 Dec 2021 17:30:09 +0000 (12:30 -0500)]
[libcxx][test] Remove redundant semiregular checks for CPOs
Some individual test files verify the CPO under test satisfies
`semiregular` concept. This is redundant since it is already part of the test
in verifying whether the entity is indeed a CPO in
`libcxx/test/std/library/description/conventions/customization.point.object/cpo.compile.pass.cpp`.
Differential Revision: https://reviews.llvm.org/D116173
Craig Topper [Wed, 29 Dec 2021 02:28:23 +0000 (18:28 -0800)]
[RISCV] Add more div by constant test cases.
Some constants require more instructions than others. This adds
additional test for each variation. UDIV has 2 variations, SDIV has
4 variations.
Some of these sequence may have gotten worse on RV32 when we started
doing the div by constant optimization before type legalization. We
materialized a smaller constant, but we require more instructions
to emulate 8 or 16 bit right shifts. This was hidden by the lack
of test coverage.
I've also added Zba and Zbb test cases to show the affect of sext.b,
sext.h, zext.h, and zext.w on some of the shifts. In some cases
we end up generating more code after the multiply because we use
a zext.h+srli and sext.h+srai where without Zbb we share a slli
between a srli and srai.
Fangrui Song [Wed, 29 Dec 2021 02:03:28 +0000 (18:03 -0800)]
[llvm-profdata][docs] Use `` instead of `
Kyungwoo Lee [Wed, 29 Dec 2021 01:31:36 +0000 (17:31 -0800)]
[llvm-profdata] Make -debug-info visible
Add the option comment in .rst.
Reviewed By: ellis
Differential Revision: https://reviews.llvm.org/D116348
Fangrui Song [Wed, 29 Dec 2021 00:40:51 +0000 (16:40 -0800)]
[ELF] --gc-sections: Work around SHT_PROGBITS .init_array.N for Rust
See https://github.com/rust-lang/rust/issues/92181
Mike Hommey [Wed, 29 Dec 2021 00:01:01 +0000 (19:01 -0500)]
[lld-macho] Fix alignment of TLV data sections
References from thread-local variable sections are treated as offsets
relative to the start of the thread-local data memory area, which is
initialized via copying all the TLV data sections (which are all
contiguous). If later data sections require a greater alignment than
earlier ones, the offsets of data within those sections won't be
guaranteed to aligned unless we normalize alignments. We therefore use
the largest alignment for all TLV data sections.
Reviewed By: #lld-macho, int3
Differential Revision: https://reviews.llvm.org/D116263
modimo [Tue, 28 Dec 2021 22:48:15 +0000 (14:48 -0800)]
[CodeView] Clamp Frontend version
D43002 introduced a test debug-info-objname.cpp that outputted the current compiler version into CodeView. Internally we appended a date to the patch version and overflowed the 16-bits allocated to that space. This change clamps the Frontend version outputted values to 16-bits like rGd1185fc081ead71a8bf239ff1814f5ff73084c15 did for the Backend version.
Testing:
ninja check-all
newly added tests correctly clamps and no longer asserts when trying to output the field
Reviewed By: aganea
Differential Revision: https://reviews.llvm.org/D116243
Krzysztof Parzyszek [Tue, 28 Dec 2021 21:47:42 +0000 (13:47 -0800)]
[Hexagon] Support BUILD_VECTOR of floating point HVX vectors
Co-authored-by: Anirudh Sundar Subramaniam <quic_sanirudh@quicinc.com>
Co-authored-by: Ankit Aggarwal <aankit@quicinc.com>
Shilei Tian [Tue, 28 Dec 2021 22:42:31 +0000 (17:42 -0500)]
[OpenMP][CUDA] Add resource pool for CUevent
Following D111954, this patch adds the resource pool for CUevent.
Reviewed By: ye-luo
Differential Revision: https://reviews.llvm.org/D116315
Ellis Hoag [Tue, 28 Dec 2021 21:34:36 +0000 (13:34 -0800)]
[InstrProf] Prevent duplicate functions in correlated data
When using debug info for profile correlation, avoid adding duplicate
functions in the synthetic Data section.
Before this patch, n duplicate function entries in the Data section would
cause counter values to be a factor of n larger. I built instrumented
clang with and without debug info correlation and got these summaries.
```
# With Debug Info Correlate
$ llvm-profdata show default.profdata
Instrumentation level: IR entry_first = 0
Total functions: 182530
Maximum function count: 52034
Maximum internal block count: 5763
# Without
$ llvm-profdata show default.profdata
Instrumentation level: IR entry_first = 0
Total functions: 183212
Maximum function count: 52034
Maximum internal block count: 5766
```
The slight difference in counts seem to be mostly from FileSystem and
Map functions and the difference in the number of instrumented functions
seems to come from missing debug info like destructors without source.
Reviewed By: kyulee
Differential Revision: https://reviews.llvm.org/D116051
Craig Topper [Tue, 28 Dec 2021 22:05:40 +0000 (14:05 -0800)]
[TargetLowering] Remove workaround for old behavior of getShiftAmountTy. NFC
getShiftAmountTy used to directly return the shift amount type from
the target which could be too small for large illegal types. For
example, X86 always returns i8.
The code here detected this and used i32 instead if it won't fit. This
behavior was added to getShiftAmountTy in D112469 so we no longer need
this workaround.
Nick Desaulniers [Tue, 28 Dec 2021 20:48:30 +0000 (12:48 -0800)]
[clang][ARM] re-use arm::isHardTPSupported for hardware TLS check
This conditional check for -mstack-protector-guard=tls got out of sync
with the conditional check for -mtp=cp15 by me in D114116, because I
forgot about the similar check added in D113026.
Re-use the code in arm::isHardTPSupported so that these aren't out of
sync.
Interestingly, our CI reported this when testing
-mstack-protector-guard=tls; it was only reproducible with Debian's LLVM
and not upstream LLVM due to this out of tree patch:
https://salsa.debian.org/pkg-llvm-team/llvm-toolchain/-/blob/snapshot/debian/patches/930008-arm.diff
Fixes: https://github.com/ClangBuiltLinux/linux/issues/1502
Reviewed By: ardb
Differential Revision: https://reviews.llvm.org/D116233
Shilei Tian [Tue, 28 Dec 2021 21:10:49 +0000 (16:10 -0500)]
[OpenMP][Plugin] Minor adjustments to ResourcePool
This patch makes some minor adjustments to `ResourcePool`:
- Don't initialize the resources if `Size` is 0 which can avoid assertion.
- Add a new interface function `clear` to release all hold resources.
- If initial size is 0, resize to 1 when the first request is encountered.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D116340
Renato Golin [Tue, 28 Dec 2021 11:59:54 +0000 (11:59 +0000)]
Documentation for the process of adding new targets
Plenty of new targets nowadays and I found myself repeating the same
thing over and over, so this is more or less what we said over the last
few years, but condensed in an ordered fashion and easy to digest.
This does not change any of the recommendations, only documents what we
have been saying for years.
Krzysztof Parzyszek [Tue, 28 Dec 2021 19:36:07 +0000 (11:36 -0800)]
[Hexagon] Delete unused declaration of LowerHvxMul, NFC
Brian Cain [Tue, 28 Dec 2021 19:23:50 +0000 (11:23 -0800)]
[Hexagon] Add testcase for arch flags
Rob Suderman [Fri, 24 Dec 2021 00:25:53 +0000 (16:25 -0800)]
[mlir][tosa] Resubmit split tosa-to-linalg named ops out of pass
Includes dependency fix that resulted in canonicalizer pass not linking in.
Linalg named ops lowering are moved to a separate pass. This allows TOSA
canonicalizers to run between named-ops lowerings and the general TOSA
lowerings. This allows the TOSA canonicalizers to run between lowerings.
Differential Revision: https://reviews.llvm.org/D116057
Krzysztof Parzyszek [Tue, 28 Dec 2021 19:20:03 +0000 (11:20 -0800)]
[Hexagon] Move testcase accidentally committed to wrong directory
llvm/test/DebugInfo/Hexagon/zreg-post-inc.s -> llvm/test/MC/Hexagon/zreg-post-inc.s
Michael Benfield [Tue, 28 Dec 2021 19:10:27 +0000 (19:10 +0000)]
[clang] Fix AttrDocs.td formatting.
This should fix the builder clang-sphinx-docs.
David Green [Tue, 28 Dec 2021 19:10:01 +0000 (19:10 +0000)]
[AArch64] Minor AArch64MIPeepholeOpt cleanup. NFC
We should always be in SSA form when running the pass, so turn a check
into an assert.
Roman Lebedev [Tue, 28 Dec 2021 18:58:31 +0000 (21:58 +0300)]
[BitcodeReader] propagateAttributeTypes(): fix opaque pointer handling
Can't get the pointee type of an opaque pointer,
but in that case said attributes must already be typed,
so just don't try to rewrite them if they already are.
Greg Clayton [Tue, 28 Dec 2021 19:02:50 +0000 (11:02 -0800)]
Fix "settings set -g" so it works again.
When we switched options over to use the Options.td file, a bug was introduced that caused the "-g" option for "settings set" to require a filename arguemnt. This patch fixes this issue and adds a test so this doesn't regress.
Reviewed By: JDevlieghere
Differential Revision: https://reviews.llvm.org/D116012
Greg Clayton [Tue, 28 Dec 2021 18:49:59 +0000 (10:49 -0800)]
Cache the manual DWARF index out to the LLDB cache directory when the LLDB index cache is enabled.
This patch add the ability to cache the manual DWARF indexing results to disk for faster subsequent debug sessions. Manual DWARF indexing is time consuming and causes all DWARF to be fully parsed and indexed each time you debug a binary that doesn't have an acceptable accelerator table. Acceptable accelerator tables include .debug_names in DWARF5 or Apple accelerator tables.
This patch breaks up testing by testing all of the encoding and decoding of required C++ objects in a gtest unit test, and then has a test to verify the debug info cache is generated correctly.
This patch also adds the ability to track when a symbol table or DWARF index is loaded or saved to the cache in the "statistics dump" command. This is essential to know in statistics as it can help explain why a debug session was slower or faster than expected.
Reviewed By: labath, wallace
Differential Revision: https://reviews.llvm.org/D115951
Fangrui Song [Tue, 28 Dec 2021 18:49:52 +0000 (10:49 -0800)]
[CMake] Remove unneeded CLANG_DEFAULT_PIE_ON_LINUX canonicalization after D115751
Kazu Hirata [Tue, 28 Dec 2021 18:01:39 +0000 (10:01 -0800)]
Drop unnecessary const from return types (NFC)
Identified with readability-const-return-type.
Brian Cain [Tue, 28 Dec 2021 17:51:27 +0000 (09:51 -0800)]
[Hexagon] Add HexagonMCInstrInfo::IsABranchingInst, NFC
Brian Cain [Thu, 6 Sep 2018 16:53:00 +0000 (11:53 -0500)]
[Hexagon] Fix for producer operands search w/z-reg
Z-register does not show up in defs, so checks searching
for the def operand must look for a different def index
than they would normally.
Tony Tye [Sat, 25 Dec 2021 00:49:23 +0000 (00:49 +0000)]
[AMDGPU][NFC] Update to DWARF extension for heterogeneous debugging
- Update documentation on the DWARF extension for heterogeneous
debugging to better reference the DWARF Version 5 standard.
- Numerous other corrections.
Reviewed By: kzhuravl
Differential Revision: https://reviews.llvm.org/D116275
Tony Tye [Fri, 24 Dec 2021 08:30:06 +0000 (08:30 +0000)]
[AMDGPU][NFC] Update DWARF extension allowing locations on stack
Add changes to the DWARF Version 5 standard to the DWARF extension to
allow locations on the evaluation stack documentation.
Reviewed By: kzhuravl
Differential Revision: https://reviews.llvm.org/D116265
Kazu Hirata [Tue, 28 Dec 2021 16:52:25 +0000 (08:52 -0800)]
[llvm] Use nullptr instead of 0 (NFC)
Identified with modernize-use-nullptr.
Florian Hahn [Tue, 28 Dec 2021 16:49:46 +0000 (17:49 +0100)]
[VPlan] Add prepareToExecute to set up live-ins (NFC).
This patch adds a new prepareToExecute helper to set up live-ins, so
VPTransformState doesn't need to hold values like TripCount.
This also requires making the trip count operand for ActiveLaneMask
explicit in VPlan.
Reviewed By: Ayal
Differential Revision: https://reviews.llvm.org/D116320
Krzysztof Parzyszek [Tue, 28 Dec 2021 16:37:08 +0000 (08:37 -0800)]
[Hexagon] Remove isPredicateRegister in favor of isPredReg, NFC
HexagonMCChecker has its own function isPredicateRegister, which does
the same thing as HexagonMCInstrInfo::isPredReg.
Sanjay Patel [Tue, 28 Dec 2021 14:25:16 +0000 (09:25 -0500)]
[Analysis] allow caller to choose signed/unsigned when computing constant range
We should not lose analysis precision if an 'add' has both no-wrap
flags (nsw and nuw) compared to just one or the other.
This patch is modeled on a similar construct that was added with
D59386.
I don't think it is possible to expose a problem with an unsigned
compare because of the way this was coded (nuw is handled first).
InstCombine has an assert that fires with the example from:
https://github.com/llvm/llvm-project/issues/52884
...because it was expecting InstSimplify to handle this kind of
pattern with an smax.
Fixes #52884
Differential Revision: https://reviews.llvm.org/D116322
Sanjay Patel [Mon, 27 Dec 2021 20:44:59 +0000 (15:44 -0500)]
[InstSimplify] add tests for icmp with no-wrap add operand; NFC
Florian Hahn [Tue, 28 Dec 2021 14:37:28 +0000 (15:37 +0100)]
[VPlan] Add abstract base class for header phi recipes (NFC).
Not all header phis widen the phi, e.g. like the new
VPCanonicalIVPHIRecipe in D113223. To let those recipes also inherit
from a phi-like base class, add a more generic VPHeaderPHIRecipe
abstract base class.
Reviewed By: Ayal
Differential Revision: https://reviews.llvm.org/D116304
Nikita Popov [Tue, 28 Dec 2021 14:30:01 +0000 (15:30 +0100)]
[InstCombine] Skip some GEP folds under opaque pointers
In their current form, these folds are fundamentally incompatible
with opaque pointers. We should add a separate set of folds for
the canonicalization of the GEP source type. For now, skip these
folds.
Nikita Popov [Tue, 28 Dec 2021 13:57:43 +0000 (14:57 +0100)]
[InstCombine] Use GEP type instead of pointee type
The GEP source type is independent of whether it is a scalar or
vector GEP, as such we can simply preserve it.
Nikita Popov [Tue, 28 Dec 2021 13:28:28 +0000 (14:28 +0100)]
[ConstFold] Add another icmp of gep of global test (NFC)
This time with some complex arithmetic involving bitcasts.
Nikita Popov [Tue, 28 Dec 2021 13:20:29 +0000 (14:20 +0100)]
[Assembler] Regenerate test checks (NFC)
Switch this Assembler test to use utc by adding a dummy function
and opt run line.
Nikita Popov [Tue, 28 Dec 2021 13:09:34 +0000 (14:09 +0100)]
[ConstFold] Don't fold signed comparison of gep of global
An inbounds GEP may still cross the sign boundary, so signed icmps
cannot be folded (https://alive2.llvm.org/ce/z/XSgi4D). This was
previously fixed for other folds in this function, but this one
was missed.
Nikita Popov [Tue, 28 Dec 2021 13:06:07 +0000 (14:06 +0100)]
[ConstFold] Add additional icmp of gep of global tests (NFC)
The fold is incorrect for the sgt case, as gep inbounds is allowed
to cross the sign boundary.
Nikita Popov [Thu, 24 Jun 2021 20:31:06 +0000 (22:31 +0200)]
[InstCombine] Make indexed compare fold opaque ptr compatible
We need to make sure that the GEP source element types match.
A caveat here is that the used GEP source element type can be
arbitrary if no offset is stripped from the original GEP -- the
transform is somewhat inconsistent in that it always starts from
a GEP, but might not actually look through it if it has multiple
indices.
Florian Hahn [Tue, 28 Dec 2021 09:53:56 +0000 (10:53 +0100)]
[LV] Use specific first-order recurrence recipe as arg type (NFC).
Required for further refactoring in D116304.
Kazu Hirata [Tue, 28 Dec 2021 08:38:11 +0000 (00:38 -0800)]
[CodeGen] Drop unnecessary const from return types (NFC)
Identified with readability-const-return-type.
Nikita Popov [Tue, 28 Dec 2021 08:14:27 +0000 (09:14 +0100)]
[InstCombine] Fix GEP with same index comparison with opaque pointers
We need to also check that the source element type is the same,
otherwise the indices may have different meaning. The added
addrspacecast demonstrates that we do still need to check the
pointer type.
Chuanqi Xu [Tue, 28 Dec 2021 08:20:06 +0000 (16:20 +0800)]
[NFC] [Coroutines] Add tests for coro-split-musttail
Add two tests to address the problems during marking coro.resume
calls as musttail. The two problems are bitcast instruction and unused
instruciton respectively.
LLVM GN Syncbot [Tue, 28 Dec 2021 07:59:27 +0000 (07:59 +0000)]
[gn build] Port
b5c5d8912e07
Stanislav Gatev [Mon, 20 Dec 2021 09:56:25 +0000 (09:56 +0000)]
[clang][dataflow] Add support for noreturn destructor calls
This is part of the implementation of the dataflow analysis framework.
See "[RFC] A dataflow analysis framework for Clang AST" on cfe-dev.
Reviewed By: xazax.hun, gribozavr2
Differential Revision: https://reviews.llvm.org/D116022
Fangrui Song [Tue, 28 Dec 2021 05:34:38 +0000 (21:34 -0800)]
[ELF] Change EhInputSection::pieces to SmallVector
The decreased size does not matter that much as one file contributes at most one EhInputSection.
Kazu Hirata [Tue, 28 Dec 2021 04:48:21 +0000 (20:48 -0800)]
[clang] Remove unused "using" (NFC)
Identified by misc-unused-using-decls.
Fangrui Song [Tue, 28 Dec 2021 02:15:23 +0000 (18:15 -0800)]
[ELF] Change InStruct/Partition pointers to unique_ptr
and remove associated make<XXX> calls.
gnuHash and sysvHash are unchanged, otherwise LinkerScript::discard would
destroy the objects which may be referenced by input section descriptions.
My x86-64 lld executable is 121+KiB smaller.
Hsiangkai Wang [Mon, 27 Dec 2021 14:35:27 +0000 (22:35 +0800)]
[RISCV] Support passing scalable vectur values through the stack.
After consuming all vector registers, the scalable vector values will be
passed indirectly. The pointer values will be saved in general
registers. If all general registers are used up, we will report an error to
notify users the compiler does not support passing scalable vector
values through the stack. In this patch, we remove the restriction. After
all general registers are used up, we use the stack to save the
pointers which point to the indirect passed scalable vector values.
Differential Revision: https://reviews.llvm.org/D116310
Fangrui Song [Tue, 28 Dec 2021 01:05:48 +0000 (17:05 -0800)]
[ELF] Use const reference. NFC
Kazu Hirata [Mon, 27 Dec 2021 23:58:03 +0000 (15:58 -0800)]
[llvm] Drop unnecessary const from return types (NFC)
Identified with readability-const-return-type.
Fangrui Song [Mon, 27 Dec 2021 23:57:38 +0000 (15:57 -0800)]
[ELF] addOrphanSections: avoid std::function
Kazu Hirata [Mon, 27 Dec 2021 23:42:52 +0000 (15:42 -0800)]
[DebugInfo] Drop unnecessary const from return types (NFC)
Identified with readability-const-return-type.
Kazu Hirata [Mon, 27 Dec 2021 23:32:51 +0000 (15:32 -0800)]
[ASTMatchers] Simplify string comparisons (NFC)
Identified with readability-string-compare.
Kazu Hirata [Mon, 27 Dec 2021 23:20:12 +0000 (15:20 -0800)]
[llvm] Fix header guards (NFC)
Identified with llvm-header-guard.
Fangrui Song [Mon, 27 Dec 2021 23:16:14 +0000 (15:16 -0800)]
[ELF] Simplify and optimize SymbolTableSection<ELFT>::writeTo
Craig Topper [Mon, 27 Dec 2021 22:59:21 +0000 (14:59 -0800)]
[RISCV] Fix type in f16 and f64 version of lrint/llrint/lround/llround test cases. NFC
Due to a copy/paste mistake we were always testing float.
This required splitting up the f16 tests into separate files since
we don't have an appropriate libcall to use when the types involved
aren't legal.
Craig Topper [Mon, 27 Dec 2021 22:44:34 +0000 (14:44 -0800)]
[RISCV] Rename two tests to reflect extension being tested. NFC
Joseph Huber [Mon, 27 Dec 2021 20:02:32 +0000 (15:02 -0500)]
[OpenMP] Use alignment information in HeapToShared
This patch uses the return alignment attribute now present in the
`__kmpc_alloc_shared` runtime call to set the alignment of the shared
memory global created to replace it.
Depends on D115971
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D116319
Joseph Huber [Fri, 17 Dec 2021 22:00:13 +0000 (17:00 -0500)]
[OpenMP][FIX] Change globalization alignment to 16
This patch changes the default aligntment from 8 to 16, and encodes this
information in the `__kmpc_alloc_shared` runtime call to communicate it
to the HeapToStack pass. The previous alignment of 8 was not sufficient
for the maximum size of primitive types on 64-bit systems, and needs to
be increaesd. This reduces the amount of space availible in the data
sharing stack, so this implementation will need to be improved later to
include the alignment requirements in the allocation call, and use it
properly in the data sharing stack in the runtime.
Depends on D115888
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D115971
Joseph Huber [Fri, 17 Dec 2021 19:28:36 +0000 (14:28 -0500)]
[Attributor][Fix] Add alignment return attribute to HeapToStack
This patch changes the HeapToStack optimization to attach the return alignment
attribute information to the created alloca instruction. This would cause
problems when replacing the heap allocation with an alloca did not respect the
alignment of the original heap allocation, which would typically be aligned on
an 8 or 16 byte boundary. Malloc calls now contain alignment attributes,
so we can use that information here.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D115888
Fangrui Song [Mon, 27 Dec 2021 20:56:38 +0000 (12:56 -0800)]
[ELF] Delete stale declaration. NFC
Fangrui Song [Mon, 27 Dec 2021 20:10:23 +0000 (12:10 -0800)]
[ELF] -r: move zero OutputSection::addr code into finalizeAddressDependentContent
Ensure addresses are unchanged after finalizeAddressDependentContent.
Gulfem Savrun Yeniceri [Wed, 15 Dec 2021 21:51:33 +0000 (21:51 +0000)]
[profile] Fix writing binary id into profiles
This patch adds support to read all the PT_NOTE segments in the
executable to find the binary ids. Previously, it was only reading
the first PT_NOTE segment, and this was missing the cases where
binary id is in the following segments. As a result, binary-id.c
and binary-id-padding.c test were failing in the following cases:
1) sanitizer-x86_64-linux bot
https://lab.llvm.org/staging/#/builders/97
2) OpenSuse Tumbleweed
https://github.com/llvm/llvm-project/issues/52695
Differential Revision: https://reviews.llvm.org/D115830
Sanjay Patel [Mon, 27 Dec 2021 18:38:11 +0000 (13:38 -0500)]
[Analysis] remove unneeded casts; NFC
The callee does the casting too; this matches a plain call later in the same function for 'shl'.
Sanjay Patel [Tue, 21 Dec 2021 18:50:25 +0000 (13:50 -0500)]
[InstCombine] add tests for shuffle + binop; NFC
Shilei Tian [Mon, 27 Dec 2021 16:31:59 +0000 (11:31 -0500)]
[OpenMP][Plugin] Introduce generic resource pool
Currently CUDA streams are managed by `StreamManagerTy`. It works very well. Now
we have the need that some resources, such as CUDA stream and event, will be
hold by `libomptarget`. It is always good to buffer those resources. What's more
important, given the way that `libomptarget` and plugins are connected, we cannot
make sure whether plugins are still alive when `libomptarget` is destroyed. That
leads to an issue that those resouces hold by `libomptarget` might not be
released correctly. As a result, we need an unified management of all the resources
that can be shared between `libomptarget` and plugins.
`ResourcePoolTy` is designed to manage the type of resource for one device.
It has to work with an allocator which is supposed to provide `create` and
`destroy`. In this way, when the plugin is destroyed, we can make sure that
all resources allocated from native runtime library will be released correctly,
no matter whether `libomptarget` starts its destroy.
Reviewed By: ye-luo
Differential Revision: https://reviews.llvm.org/D111954
Alexey Bataev [Mon, 27 Dec 2021 15:35:35 +0000 (07:35 -0800)]
[SLP][NFC]Add a test for shuffled entries with different vector sizes,
NFC.
David Green [Mon, 27 Dec 2021 14:15:03 +0000 (14:15 +0000)]
[ARM] Extend IsCMPZCSINC to handle CMOV
A 'CMOV 1, 0, CC, %cpsr, Cmp' is the same as a 'CSINC 0, 0, CC, Cmp',
and can be treated the same in IsCMPZCSINC added in D114013. This allows
us to remove the unnecessary CMOV in the same way that we could remove a
CSINC.
Differential Revision: https://reviews.llvm.org/D115188
Nikita Popov [Mon, 27 Dec 2021 13:53:38 +0000 (14:53 +0100)]
[InstCombine] Fix GEP of GEP fold with opaque pointers
We need to check that result and source element types match, as
this is no longer automatically enforced with opaque pointers.
Nikita Popov [Mon, 27 Dec 2021 13:50:04 +0000 (14:50 +0100)]
[InstCombine] Extract GEP of GEP fold into separate function
This change may not be entirely NFC, because a number of early
returns will now only early return from this particular fold,
rather than the whole visitGetElementPtr() implementation. This
is also the reason why I'm doing this change, as I don't think
this was intended.
Groverkss [Mon, 27 Dec 2021 13:36:32 +0000 (19:06 +0530)]
[MLIR] Use IntegerPolyhedron in Simplex instead of FlatAffineConstraints
This patch replaces usage of FlatAffineConstraints in Simplex with
IntegerPolyhedron. This removes dependency of Simplex on FlatAffineConstraints
and puts it on IntegerPolyhedron, which is part of Presburger library.
Reviewed By: arjunp
Differential Revision: https://reviews.llvm.org/D116287