--compress-sections <section-glib>=[none|zlib|zstd] is similar to
--compress-debug-sections but applies to broader sections without the
SHF_ALLOC flag. lld will report an error if a SHF_ALLOC section is
matched. An interesting use case is to compress `.strtab`/`.symtab`,
which consume a significant portion of the file size (15.1% for a
release build of Clang).
An older revision is available at https://reviews.llvm.org/D154641 .
This patch focuses on non-allocated sections for safety. Moving
`maybeCompress` as D154641 does not handle STT_SECTION symbols for
`-r --compress-debug-sections=zlib` (see `relocatable-section-symbol.s`
from #66804).
Since different output sections may use different compression
algorithms, we need CompressedData::type to generalize
config->compressDebugSections.
GNU ld feature request: https://sourceware.org/bugzilla/show_bug.cgi?id=27452
Link: https://discourse.llvm.org/t/rfc-compress-arbitrary-sections-with-ld-lld-compress-sections/71674
Pull Request: https://github.com/llvm/llvm-project/pull/84855
Handles the INSERT_SUBVECTOR(INSERT_SUBVECTOR(UNDEF,SUB0,0),SUB1,N) pattern
Currently limited to v8i64/v8f64 cases as only AVX512 has decent cross lane 2-input shuffles, the plan is to relax this as I deal with some regressions
A new function attribute named amdgpu_num_work_groups is added. This
attribute, which consists of three integers, allows programmers to let
the compiler know the number of workgroups to be launched in each of the
three dimensions and do optimizations based on that information.
---------
Co-authored-by: Jun Wang <jun.wang7@amd.com>
MSVC fails when there is ambiguity (multiple options) around implicit
type conversion operators.
Make ConstString's conversion operator to string_view explicit to avoid
ambiguity with one to StringRef and remove an unused local variable that
MSVC also fails on.
From issue #73088. I changed `NodeBuilderContext` into a class.
Additionally, there were some other mentions of the former being a
struct which I also changed into a class. This is my first time working
with an issue so I will be open to hearing any advice or changes that
need to be done.
Destroying an ios_base object before it is properly initialized is
undefined behavior. Unlike typical C++ classes the initialization is not
done in the constructor, but in a dedicated init function. Due to
virtual inheritance of the basic_ios object in ostream and friends this
undefined behaviour can be triggered when inheriting from classes that
can throw in their constructor and inheriting from ostream.
Use the __loc_ member of ios_base as sentinel to detect whether the
object has or has not been initialized.
Addresses https://github.com/llvm/llvm-project/issues/57964
Resolves#84745.
Based on SDPatternMatch introduced by #78654, this patch replaces some
of basic patterns in `visitADDLike` with corresponding patterns in
SDPatternMatch.
This patch only replaces original folds, instead of introducing new ones.
This is a follow-up commit of #84446.
In this patch, I demonstrate that `forEachArgumentWithParam` and
`forEachArgumentWithParamType` did not correctly handle the presence of
the explicit object parameter for operator calls.
Prior to this patch, the matcher would skip the first (and only)
argument of the operator call if the explicit object param was used.
Note that I had to move the definition of
`isExplicitObjectMemberFunction`, to be declared before the matcher I
fix to be visible.
I also had to do some gymnastics for passing the language standard
version command-line flags to the invocation as
`matchAndVerifyResultTrue` wasn't really considered for non-c++11 code.
See the that it always prepends `-std=gnu++11` to the command-line
arguments. I workarounded it by accepting extra args, which get
appended, thus possibly overriding the hardcoded arguments.
I'm not sure if this qualifies for backporting to clang-18 (probably not
because its not a crash, but a semantic problem), but I figure it might
be useful for some vendors (like us).
But we are also happy to cherry-pick this fix to downstream. Let me know
if you want this to be backported or not.
CPP-5074
Typically the rules in the database are contiguous, but that is not a
requirement. This fixes the case when they are not.
---------
Co-authored-by: Louis Dionne <ldionne.2@gmail.com>
The first half of this patch is a long-standing annoyance, if I attach
to debugserver with lldb while it is waiting for an lldb connection, the
syscall is interrupted and it doesn't retry, debugserver exits
immediately.
The second half is a request from another tool that is communicating
with debugserver, that we retry reads on our sockets in the same way. I
haven't dug in to the details of how they're communicating that this is
necessary, but the best I've been able to find reading the POSIX API
docs, this is fine.
rdar://117113298
Summary:
The GPU uses a SIMT execution model. That means that each value actually
belongs to a group of 32 or 64 other lanes executing next to it. These
platforms offer some intrinsic fuctions to actually take elements from
neighboring lanes. With these we can do parallel scans or reductions.
These functions do not have an immediate user, but will be used in the
allocator interface that is in-progress and are generally good to have.
This patch is a precommit for these new utilitly functions.
Summary:
Previously we got the lane-id from taking the global thread ID and
taking off the bottom 5 bits. This works but is inefficient compared to
the NVPTX intrinsic simply dedicated to get this value.
Summary:
We are currently taking the lower 5 bites of the thread ID as the warp
ID. This doesn't work in non-1D grids and is also slower than just using
the dedicated hardware register.
My GCC build is failing with issues similar why we added our own. Looks
like
we missed one spot. See also:
commit 72ce62941579 ("[libc] Add C23 limits.h header. (#78887)")
When we are using the Zcmp extension together with the E extension in
32-bit mode and we need to spill both callee-saved registers as well as
needing a couple of 32-bit stack slots, we emit a meaningless stack
adjustment with cm.push/cm.popret. Furthermore this leads to the stack
slot for the ra being clobbered so control returns to a random location.
This is just a pre-commit test so that the PR for the fix shows the
difference in code generation.
Stop overriding -working-directory to CWD during argument parsing, which
should no longer necessary after we set the VFS working directory, and
set FSOpts correctly after parsing arguments so that working-directory
behaves correctly.
As part of the effort to rename the DbgRecord classes, this patch
renames the widely-used functions that operate on DbgRecords but refer
to DbgValues or DPValues in their names to refer to DbgRecords instead;
all such functions are defined in one of `BasicBlock.h`,
`Instruction.h`, and `DebugProgramInstruction.h`.
This patch explicitly does not change the names of any comments or
variables, except for where they use the exact name of one of the
renamed functions. The reason for this is reviewability; this patch can
be trivially examined to determine that the only changes are direct
string substitutions and any results from clang-format responding to the
changed line lengths. Future patches will cover renaming variables and
comments, and then renaming the classes themselves.
This is a trivial fix ( I guess it has not been noticed because of no
use). Currently, the doc says the function returns `true` if two
instances of `DominanceFrontierBase` matches, otherwise `false` is
returned. I have checked the implementation
9308d6688c/llvm/include/llvm/Analysis/DominanceFrontierImpl.h (L71-L94)
which examines whether two dominance frontier mappings are equal, and
the actual value it returns contradicts the description in the doc.
Enumerators dont have the type of their enumeration before the closing
brace. In these cases Expr::getEnumCoercedType() incorrectly returned
the enumeration type.
Introduced in PR #81418Fixes#84712
The casting of FP atomic loads and stores were always done by the
front-end, even though the AtomicExpandPass will do it if the target
requests it (which is the default).
This patch removes this casting in the front-end entirely.
IncrementalExecutor is an implementation detail of the Interpreter. In
order to test extended features properly, we must be able to setup and
tear down the executor manually.
Certain functions in glibc have "nonnull" attributes on pointer
parameters (even in cases where passing a null pointer should be handled
correctly). There are a few cases of such calls in flang: memcmp and
memcpy with the length parameter set to 0.
Avoid passing a null pointer to these functions, since the conflict with
the nonnull attribute could cause an undefined behavior.
This was detected by the undefined behavior sanitizer.
TypeRange is an iterator range, it does not own storage spanned by the
iterators. When using TypeRange, make sure that the actual contents
don't "expire" while the range is in use.
This was detected by address sanitizer.
