In 172eee9c, we tried to avoid these by modelling the callee as
internally resetting the stack pointer.
However, for the majority of functions with reserved stack frames, this
would lead LLVM to emit extra SP adjustments to undo the callee's
internal adjustment. This lead us to fix the problem further on down the
pipeline in eliminateCallFramePseudoInstr. In 5b79e603d3, I added
use a heuristic to try to detect when the adjustment would be
unreachable.
This heuristic is imperfect, and when exception handling is involved, it
fails to fire. The new test is an example of this. Simply throwing an
exception with an active cleanup emits dead SP adjustments after the
throw. Not only are they dead, but if they were executed, they would be
incorrect, so they are confusing.
This change essentially reverts 172eee9c and makes the 5b79e603d3
heuristic responsible for preventing unreachable stack adjustments. This
means we may emit unreachable stack adjustments for functions using EH
with unreserved call frames, but that is not very many these days. Back
in 2016 when this change was added, we were focused on 32-bit, which we
observed to have fewer reserved frames.
Fixes PR45064
Reviewed By: hans
Differential Revision: https://reviews.llvm.org/D75712
As mentioned in the comments, extractelement is special
since we actually want a scalar base for that element we extracted from
the vector (i.e. not a vector base).
This same logic should apply to uses of the extractelement such as phis
and selects which have the same BDV as the extractelement.
Howeber, for these uses we conservatively mark the BDV state as
conflict, since setting the EE's new base BDV does not always dominate
these uses.
Added testcase showcases the problem where the BDV identification chokes
on the incorrect cast from vector to scalar for the phi use of
extractelement.
Tests-Run: make check, internal fuzzer testing
Reviewers: reames, skatkov, dantrushin
Reviewed-By: dantrushin
Differential Revision: https://reviews.llvm.org/D75704
Summary:
This further cleans up the control flow and makes it easier to
optimize and replace portions in a subsequent patch.
This should be NFC, but given the amount of #ifdeffing here,
it may not be. So will watch the buildbots closely.
Also, as this is purely moving existing code around, I plan to
ignore the lint errors.
Reviewers: compnerd, miyuki, mstorsjo
Subscribers: libcxx-commits
Tags: #libc
Differential Revision: https://reviews.llvm.org/D75705
Putting this up mainly for discussion on
how this should be done. I am interested in MLIR from
the Julia side and we currently have a strong preference
to dynamically linking against the LLVM shared library,
and would like to have a MLIR shared library.
This patch adds a new cmake function add_mlir_library()
which accumulates a list of targets to be compiled into
libMLIR.so. Note that not all libraries make sense to
be compiled into libMLIR.so. In particular, we want
to avoid libraries which primarily exist to support
certain tools (such as mlir-opt and mlir-cpu-runner).
Note that the resulting libMLIR.so depends on LLVM, but
does not contain any LLVM components. As a result, it
is necessary to link with libLLVM.so to avoid linkage
errors. So, libMLIR.so requires LLVM_BUILD_LLVM_DYLIB=on
FYI, Currently it appears that LLVM_LINK_LLVM_DYLIB is broken
because mlir-tblgen is linked against libLLVM.so and
and independent LLVM components.
Previous version of this patch broke depencies on TableGen
targets. This appears to be because it compiled all
libraries to OBJECT libraries (probably because cmake
is generating different target names). Avoiding object
libraries results in correct dependencies.
(updated by Stephen Neuendorffer)
Differential Revision: https://reviews.llvm.org/D73130
add_llvm_library and add_llvm_executable may need to create new targets with
appropriate dependencies. As a result, it is not sufficient in some
configurations (namely LLVM_BUILD_LLVM_DYLIB=on) to only call
add_dependencies(). Instead, the explicit TableGen dependencies must
be passed to add_llvm_library() or add_llvm_executable() using the DEPENDS
keyword.
Differential Revision: https://reviews.llvm.org/D74930
deduction guides.
Previously if an implicit deduction guide had a default argument with a
cleanup, we'd leave the 'pending cleanup' flag set after declaring the
implicit guide. But it turns out that there's no reason to even
substitute into the default argument when declaring an implicit
deduction guide: we only need to record that the default argument
exists, not what it is, since we never actually form a call to a
deduction guide.
This reverts commit 737394c490.
The fp-model test was failing on platforms that enable denormal flushing
based on -ffast-math. This needs to reset to IEEE, not the default in
these cases.
Change-Id: Ibbad32f66d0d0b89b9c1173a3a96fb1a570ddd89
Summary:
Otherwise this code won't run on the Release+Asserts builds we have on the CI.
Fixes rdar://problem/59867885 (partly)
Reviewers: aprantl
Reviewed By: aprantl
Subscribers: JDevlieghere, lldb-commits
Tags: #lldb
Differential Revision: https://reviews.llvm.org/D75493
llvm::call_once(initDwarfLine, [this]() { initializeDwarf(); });
Though it is not used in all places.
I need that patch for implementing "Remove obsolete debug info" feature
(D74169). But this caching mechanism is useful by itself, and I think it
would be good to use it without connection to "Remove obsolete debug info"
feature. So this patch changes inplace creation of DWARFContext with
its cached version.
Depends on D74308
Reviewed By: ruiu
Differential Revision: https://reviews.llvm.org/D74773
Starting with iOS 13 simulator binaries are identified with an
explicit platform in the new LC_BUILD_VERSION load command.
On older deployment targets using the LC_VERSION_MIN load commands,
this patch detects when an ios process runs on a macOS host and
updates the target triple with the "simulator" environment
accordingly.
(Patch re-applied with bugfix this time).
This is part of https://bugs.swift.org/browse/SR-11971
rdar://problem/58438125
Differential Revision: https://reviews.llvm.org/D75696
There was a noop bitconvert in the load pattern. While there
also make all the sources refer to src_v.RC even though its the
same as _.RC, but its consistent.
In cmake, it is redundant to have a target list under target_link_libraries()
and add_dependency(). This patch removes the redundant dependency from
add_dependency().
Differential Revision: https://reviews.llvm.org/D74929
Starting with iOS 13 simulator binaries are identified with an
explicit platform in the new LC_BUILD_VERSION load command.
On older deployment targets using the LC_VERSION_MIN load commands,
this patch detects when an ios process runs on a macOS host and
updates the target triple with the "simulator" environment
accordingly.
(Patch re-applied without modifications, the bot failure was unrelated).
This is part of https://bugs.swift.org/browse/SR-11971
rdar://problem/58438125
Differential Revision: https://reviews.llvm.org/D75696
Starting with iOS 13 simulator binaries are identified with an
explicit platform in the new LC_BUILD_VERSION load command.
On older deployment targets using the LC_VERSION_MIN load commands,
this patch detects when an ios process runs on a macOS host and
updates the target triple with the "simulator" environment
accordingly.
This is part of https://bugs.swift.org/browse/SR-11971
rdar://problem/58438125
Differential Revision: https://reviews.llvm.org/D75696
GetContainedInventedTypeParmVisitor would not account for the case where TemplateTypeParmType::getDecl() is
nullptr, causing bug #45102.
Add the nullptr check.
CMake allows calling target_link_libraries() without a keyword,
but this usage is not preferred when also called with a keyword,
and has surprising behavior. This patch explicitly specifies a
keyword when using target_link_libraries().
Differential Revision: https://reviews.llvm.org/D75725
Summary:
These implement the usual IEEE-style floating point comparison
semantics, e.g. +0.0 == -0.0 and all operators except != return false
if either argument is NaN.
Subscribers: arsenm, jvesely, nhaehnle, hiraditya, dexonsmith, kerbowa, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D75237
Summary:
Dollar signed prefixed integers were not allowed by the AsmParser to be
used as Identifiers, differing from the GNU assembler behavior.
This patch updates the parsing of Identifiers to consider such cases as
valid, where the identifier string includes the $ prefix itself. As the
Lexer currently splits these occurrences into separate tokens, those
need to be combined by the AsmParser itself.
Reviewers: efriedma, chill
Reviewed By: efriedma
Subscribers: sdardis, hiraditya, jrtc27, atanasyan, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D75111
Summary:
ARMAsmParser was incorrectly dropping a leading dollar sign character
from symbol names in targets of branch instructions. This was caused by
an incorrect assumption that the contents following the dollar sign
token should be handled as a constant immediate, similarly to the #
token.
This patch avoids the operand parsing from consuming the dollar sign
token when it is followed by an identifier, making sure it is properly
parsed as part of the expression.
Reviewers: efriedma
Reviewed By: efriedma
Subscribers: danielkiss, chill, carwil, vhscampos, kristof.beyls, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D73176
This is a follow-up for D75609. As @dblaikie suggested, it prints
the actual number for an unknown section identifier when dumping
unit index sections.
Differential Revision: https://reviews.llvm.org/D75668
Handle LinkOnceODRLinkage;
Handle AppendingLinkage type for llvm.global_ctors/dtors static init global arrays;
Differential Revision: https://reviews.llvm.org/D75305
Summary:
The messages for two of the warnings are misleading:
* warn_for_range_const_reference_copy suggests that the initialization
of the loop variable results in a copy. But that's not always true,
we just know that some conversion happens, potentially invoking a
constructor or conversion operator. The constructor might copy, as in
the example that lead to this message [1], but it might also not.
However, the constructed object is bound to a reference, which is
potentially misleading, so we rewrite the message to emphasize that.
We also make sure that we print the reference type into the warning
message to clarify that this warning only appears when operator*
returns a reference.
* warn_for_range_variable_always_copy suggests that a reference type
loop variable initialized from a temporary "is always a copy". But
we don't know this, the range might just return temporary objects
which aren't copies of anything. (Assuming RVO a copy constructor
might never have been called.)
The message for warn_for_range_copy is a bit repetitive: the type of a
VarDecl and its initialization Expr are the same up to cv-qualifiers,
because Sema will insert implicit casts or constructor calls to make
them match.
[1] https://bugs.llvm.org/show_bug.cgi?id=32823
Reviewers: aaron.ballman, Mordante, rtrieu
Reviewed By: aaron.ballman
Differential Revision: https://reviews.llvm.org/D75613
Various vload and vstore builtins were missing or misdefined in the
TableGen description. Align the OpenCL vload* and vstore* builtins of
the `-fdeclare-opencl-builtins` option to those of `opencl-c.h`.
Reviewed-by: Stuart Brady <stuart.brady@arm.com>
Summary:
Clangd's approach is to provide lots of completions, and let ranking sort them
out. This relies on various important signals (Quality.h), without which the
large completion lists are extremely spammy.
Even with a completion result exactly at the cursor, vscode looks backwards and
tries to match the presumed partial-identifier against filterText, and uses
the result to rank, with sortText only used as a tiebreak.
By prepending the partial-identifier to the filterText, we can force the match
to be perfect and so give sortText full control of the ranking.
Full sad story: https://github.com/microsoft/language-server-protocol/issues/898
It's possible to do this on the server side too of course, and switch it on
with an initialization option. But it's a little easier in the extension, it
will get the fix to users of old clangd versions, and other editors
Reviewers: hokein
Reviewed By: hokein
Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D75623
With AssertingVHs instead of bare pointers in
BlockFrequencyInfoImpl::Nodes (but without CallbackVHs) ~1/36 of all
tests ran by make check fail. It means that there are users of BFI that
delete basic blocks while keeping BFI. Some of those transformations add
new basic blocks, so if a new basic block happens to be allocated at
address where an already deleted block was and we don't explicitly set
block frequency for that new block, BFI will report some non-default
frequency for the block even though frequency for the block was never
set. Inliner is an example of a transformation that adds and removes BBs
while querying and updating BFI.
With this patch, thanks to updates via CallbackVH, BFI won't keep stale
pointers in its Nodes map.
This is a resubmission of 408349a25d with
fixed compiler warning and MSVC compilation error.
Reviewers: davidxl, yamauchi, asbirlea, fhahn, fedor.sergeev
Reviewed-By: asbirlea, davidxl
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D75341
As noted on D75114, if both arguments of a funnel shift are consecutive loads we are missing the opportunity to combine them into a single load.
Differential Revision: https://reviews.llvm.org/D75624
Currently `yaml2obj` require `Offset` field in a relocation description.
There are many cases when `Offset` is insignificant in a context of a test case.
Making `Offset` optional allows to simplify our test cases.
This is what this patch does.
Also, with this patch `obj2yaml` does not dump a zero offset of a relocation.
Differential revision: https://reviews.llvm.org/D75608
The intention was to remove the `Object/X86/yaml-elf-x86-rel-broken.yaml test`,
This test is at the wrong place.
`yaml-elf-x86-rel-broken.yaml` was introduced in
rG892c6c86ea25dc97668ff1f1b7bf1108e85fa5ec to check that
yaml2obj can use an arbitrary `Hex32` value as a relocation type.
We have tests that check the similar functionality.
I've improved them and removed the `yaml-elf-x86-rel-broken.yaml`
Differential revision: https://reviews.llvm.org/D75679
The static Disassembler can be thought of as shorthands for three
operations:
- fetch an appropriate disassembler instance (FindPluginForTarget)
- ask it to dissassemble some bytes (ParseInstructions)
- ask it to dump the disassembled instructions (PrintInstructions)
The only thing that's standing in the way of this interpretation is that
the Disassemble function also does some address resolution before
calling ParseInstructions. This patch moves this functionality into
ParseInstructions so that it is available to users who call
ParseInstructions directly.
Fangrui Song