| Age | Commit message (Collapse) | Author |
|---|
|
In the case of static compilation the file system is pretty much read-only
and taking a snapshot of it usually is sufficient. In the interactive C++
case the compilation is longer and people can create and include files, etc.
In that case we often do not want to open files or cache failures unless is
absolutely necessary.
This patch extends the original API call by forwarding some optional flags,
so we can continue use it in the previous way with no breakage.
Signed-off-by: Jun Zhang <jun@junz.org>
Differential Revision: https://reviews.llvm.org/D131241
|
|
|
|
Looks like a typo from the past code changes.
Differential Revision: https://reviews.llvm.org/D131244
|
|
D74537 introduced a bug: if `(config->andFeatures & GNU_PROPERTY_AARCH64_FEATURE_1_PAC) != 0`
with -z pac-plt unspecified, we incorrectly use AArch64BtiPac, whose writePlt will make
out-of-bounds write after the .plt section. This is often benign because the
output section after .plt will usually overwrite the content.
This is very difficult to test without D131247 (Parallelize writes of different OutputSections).
|
|
The place from D61712 seems unneeded now. We can just use the place added by
D62609 (support AArch64 BTI/PAC).
|
|
of the test, NFC
|
|
Some tests (e.g. aarch64-feature-pac.s) segfault in libstdc++ _GLIBCXX_DEBUG
builds (enabled by LLVM_ENABLE_EXPENSIVE_CHECKS).
dyn_cast<ThunkSection> is incorrectly true for any SyntheticSection. std::merge
transitively calls mergeCmp(x, x) (due to __glibcxx_requires_irreflexive_pred)
and will segfault in `ta->getTargetInputSection()`. The dyn_cast<ThunkSection>
issue should be eventually fixed properly, bug `a != b` is robust enough for now.
|
|
6635f48e4aba
See revision b8b7a9dcdcbc for prior art.
|
|
When not set output, set default output to stdout.
When set output with -Fo and no -fcgl, set -emit-obj to generate dx container.
Reviewed By: beanz
Differential Revision: https://reviews.llvm.org/D130858
|
|
When performing device only compilation, there was an issue where
`cubin` outputs were being renamed to `cubin` despite the user's name.
This is required in a normal compilation flow as the Nvidia tools only
understand specific filenames instead of checking magic bytes for some
unknown reason. We do not want to perform this transformation when the
user is performing device only compilation.
Reviewed By: tra
Differential Revision: https://reviews.llvm.org/D131278
|
|
Use of `ORIGINAL_PCH_DIR` record has been superseeded by making PCH/PCM files with relocatable paths at write time.
Removing this record is useful for producing an output-path-independent PCH file and enable sharing of the same PCH file even
when it was intended for a different output path.
Differential Revision: https://reviews.llvm.org/D131124
|
|
Fixes https://github.com/llvm/llvm-project/issues/56960
Differential Revision: https://reviews.llvm.org/D131288
|
|
D91426 makes .got possibly empty while needed. If .got and .data have the same
address, and .got's content is written after .data, the first word of .data will
be corrupted.
The bug is not testable without D131247.
|
|
Depends On D131285
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D131291
|
|
The newly-completed papers:
- P0896R4 ("The One Ranges Proposal");
- P1243R4 ("Rangify New Algorithms");
- P1252R2 ("Ranges Design Cleanup");
- P1716R3 ("Range Comparison Algorithms Are Over-Constrained");
- P1871R1 ("Concept traits should be named after concepts");
- P2106R0 ("Alternative wording for GB315 and GB316").
Differential Revision: https://reviews.llvm.org/D131234
|
|
Hoist funnel shift from logic op:
logic_op (FSH x0, x1, s), (FSH y0, y1, s) --> FSH (logic_op x0, y0), (logic_op x1, y1), s
The transformation improves code generated for some cases related to
issue https://github.com/llvm/llvm-project/issues/49541.
Reduced amount of funnel shifts can also improve throughput on x86 CPUs by utilizing more
available ports: https://quick-bench.com/q/gC7AKkJJsDZzRrs_JWDzm9t_iDM
Transformation correctness checks:
https://alive2.llvm.org/ce/z/TKPULH
https://alive2.llvm.org/ce/z/UvTd_9
https://alive2.llvm.org/ce/z/j8qW3_
https://alive2.llvm.org/ce/z/7Wq7gE
https://alive2.llvm.org/ce/z/Xr5w8R
https://alive2.llvm.org/ce/z/D5xe_E
https://alive2.llvm.org/ce/z/2yBZiy
Differential Revision: https://reviews.llvm.org/D130994
|
|
"robust" tests.
Also fix `std::find_first_of` (which accidentally copied the predicate
in the implementation).
Differential Revision: https://reviews.llvm.org/D131235
|
|
This function heap-allocates a ThreadSafeModule (the current C bindings assume
that TSMs are always heap-allocated), but was failing to free it.
Should fix http://llvm.org/PR56953.
|
|
The missing call was pointed out in https://llvm.org/PR56953, though it's not
the focus of that issue.
|
|
Avoid expensive calls to `SymbolTable::lookupNearestSymbolFrom` in verifier
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D131285
|
|
Based on `spv.ISubBorrow` from D127909.
Also resolved some clang-tidy warnings.
Reviewed By: antiagainst, ThomasRaoux
Differential Revision: https://reviews.llvm.org/D131281
|
|
This is an additional signal which may benefit sanitizers.
Reviewed By: kda
Differential Revision: https://reviews.llvm.org/D131129
|
|
|
|
This function could be called wih access_info & 0x20 or with
flags()->halt_on_error, in which case HandleTagMismatch returns (is not
fatal).
Reviewed By: eugenis
Differential Revision: https://reviews.llvm.org/D131279
|
|
|
|
This change removes dependency on pgmath mod, and also allows
Fortran runtime to issue a diagnostic message in case of zero
denominator.
Differential Revision: https://reviews.llvm.org/D131192
|
|
If we're adding a constant that can't use addi we try a few tricks,
one of which is using li+sh3add. We should not do this if lui+add
would work. For example adding 8192. Using sh3add prevents folding
a sext.w to form addw, thus increasing instruction count.
|
|
When setting CMAKE_CXX_STANDARD to 17 and targeting a macOS version
under 10.12 the ifdefs would try to use std::shared_mutex because
the of the C++ standard. This should also check the targeted SDK.
See discussion in: https://reviews.llvm.org/D130689
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D131063
|
|
95a92995d45fc6fada43ecd91eba3e7aea90487a
|
|
Sharing the FileManager between the importer and the module build should
only be an optimization. Add a cc1 option -fno-modules-share-filemanager
to allow us to test this. Fix the path to modulemap files, which
previously depended on the shared FileManager when using path mapped to
an external file in a VFS.
Differential Revision: https://reviews.llvm.org/D131076
|
|
In 6a79e2ff1989b we changed Filemanager::getEntryRef() to return the
redirecting FileEntryRef instead of looking through the redirection.
This commit fixes the case when looking up a cached file path to also
return the redirecting FileEntryRef. This mainly affects the behaviour
of calling getNameAsRequested() on the resulting entry ref.
Differential Revision: https://reviews.llvm.org/D131273
|
|
As stated in
https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#warp-level-matrix-instructions-wmma-mma:
".and operation in single-bit wmma requires sm_80 or higher."
tra@: Fixed a bug in builtins-nvptx-mma.py test generator and regenerated the tests.
Differential Revision: https://reviews.llvm.org/D131265
|
|
This fixes a bug reported privately by @craig.topper. Here's an example which illustrates the problem:
vsetivli a1, a0, e32, m1, ta, mu # both DefInfo and PrevInfo
vsetivli a2, a1, e32, m4, ta, mu
With the unsound result being:
vsetivli a1, a0, e32, m1, ta, mu
vsetivli a2, a0, e32, m4, ta, mu
Consider the case where this is running on a machine with VLEN=512,. For this case, the VLMAXs are 16 and 64 respectively.
Consider for a0 = 33. The correct result is: a1 = 16, and a2 = 16
After the unsound optimization: a1 = 16 and a2 = 33
This particular example used VLMAXs which differed by more than a power of two. With a difference of only one power of two, there's another form of this bug which involves the AVL < 2 x VLMAX special case, but that ones more complicated to construct as many examples turn out accidentally sound.
This patch takes the approach of simply removing the unsound optimization, but there are multiple sound sub-cases of it. I plan to return to at least a couple of them, but figured it was cleaner to remove the unsound optimization (for ease of backporting), and then review the new optimizations on their own.
Differential Revision: https://reviews.llvm.org/D131264
|
|
Create function segments and emit unwind info of them.
A segment must be less than 1MB and no prolog or epilog is splitted between two
segments.
This patch should generate correct, though not optimal, unwind info for large
functions. Currently it only generate pacted info (.pdata) only for functions
that are less than 1MB (single-segment functions). This is NFC from before this
patch.
The next step is to enable (.pdata) only unwind info for the first segment or
segments that have neither prolog or epilog in a multi-segment function.
Another future work item is to further split segments that require more than 255
code words or have more than 65535 epilogs.
Reference:
https://docs.microsoft.com/en-us/cpp/build/arm64-exception-handling#function-fragments
Differential Revision: https://reviews.llvm.org/D130049
|
|
This commit addresses concerns raised in D129497.
Propagate lowering options from driver to expressions lowering
via AbstractConverter instance. A single use case so far is
using optimized TRANSPOSE lowering with O1/O2/O3.
bbc does not support optimization level switches, so it uses
default LoweringOptions (e.g. optimized TRANSPOSE lowering
is enabled by default, but an engineering -opt-transpose=false
option can still override this).
Differential Revision: https://reviews.llvm.org/D130204
|
|
Jason noted that the stop message we print for a memory high water mark
notification (EXC_RESOURCE) could be clearer. Currently, the stop
reason looks like this:
* thread #3, queue = 'com.apple.CFNetwork.LoaderQ', stop reason =
EXC_RESOURCE RESOURCE_TYPE_MEMORY (limit=14 MB, unused=0x0)
It's hard to read the message because the exception and the type
(EXC_RESOURCE RESOURCE_TYPE_MEMORY) blend together. Additionally, the
"observed=0x0" should not be printed for memory limit exceptions.
I wanted to continue to include the resource type from
<kern/exc_resource.h> while also explaining what it actually is. I used
the wording from the comments in the header. With this path, the stop
reason now looks like this:
* thread #5, stop reason = EXC_RESOURCE (RESOURCE_TYPE_MEMORY: high
watermark memory limit exceeded) (limit=14 MB)
rdar://40466897
Differential revision: https://reviews.llvm.org/D131130
|
|
This design is borrowed from the lldb folks (thank you!) to declutter
the page.
* The version number at the top is removed.
* Links are pushed over to a sidebar
* The sidebar has headings
There are other minor changes:
* The warning about this project not being ready is now an RST "warning"
* Links to the Bug Reports and the Source Code are Added
* Refer to this project as either "The LLVM C LIbrary" or "The libc"
Tested:
Built locally
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D131242
|
|
This commit combines the initial commit (7c240de609af), a fix for x86_64 Linux
(3a0581501e76) and a fix for thinko in a last minute rewrite that I really
should have run the testsuite on.
Also, make sure that all the "I need to step over watchpoint" plans execute
before we call a public stop. Otherwise, e.g. if you have N watchpoints and
a Signal, the signal stop info will get us to stop with the watchpoints in a
half-done state.
Differential Revision: https://reviews.llvm.org/D130674
|
|
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D131271
|
|
Reviewed By: ThomasRaoux, kuhar
Differential Revision: https://reviews.llvm.org/D131128
|
|
MemRef types now can carry an attribute to represent the memory
space. Still, upper layers in the compilation stack mostly use
nuemric values. They don't mean much (other than differentiating
separate memory domains) in MLIR's multi-level settings. Those
numeric memory space inside MemRef types need to be translated
into concrete SPIR-V storage classes during lowering to pin down
to concrete memory types.
Thus far we have been hardcoding an arbitrary mapping from memory
space to storage class for converting MemRef types. This works fine
for only targeting Vulkan; it falls apart if we want to target other
SPIR-V consumers like OpenCL, as different consumers might want
different storage classes for the buffer/variable of the same
lifetime. For example, StorageClass in Vulkan vs. CrossWorkgroup
in OpenCL.
So putting up a new pass to let the user to control how to map
MemRef memory spaces into SPIR-V storage classes. This provides
more flexibility and can address the awkwardness in the current
SPIR-V type converter. This pass should be the prelimiary step
towards lowering MemRef related types/ops into SPIR-V.
Reviewed By: mravishankar
Differential Revision: https://reviews.llvm.org/D130317
|
|
As suggested in the post-commit comments for 019d76196f79fcff3c148,
this makes the usage symmetric with the 'and' patterns and should
be more efficient.
|
|
NOTE: i8 vector splats are ignored because the immediate range of
DUP already has full coverage.
Differential Revision: https://reviews.llvm.org/D131078
|
|
Correct a bug in the code that resets shadow memory introduced as part
of a previous change for the Go race detector (D128909). The bug was
that only the most recently added shadow segment was being reset, as
opposed to the entire extent of the segment created so far. This
fixes a bug identified in Google internal testing (b/240733951).
Differential Revision: https://reviews.llvm.org/D131256
|
|
We get a couple of improvements from recognizing swapped
operand patterns that were not handled by the replicated
code.
This should also enable simplifying larger patterns as
seen in issue #56653 and issue #56654, but that requires
enhancements to isImpliedCondition() itself.
|
|
Baseline tests for D130994
|
|
As mentioned in https://reviews.llvm.org/D121379#3690593, this
change broke the build of compiler-rt targeting powerpc using GCC.
The 32-bit powerpc target is not supposed to emit 128-bit libcalls
-- if it does, then that's a backend bug and needs to be fixed there.
This reverts commit 8f24a56a3a9363f353c8da318d97491a6818781d.
Differential Revision: https://reviews.llvm.org/D130988
|
|
Refactor common range reductions and evaluations for sinf, cosf, and
sincosf. Added exhaustive tests for sincosf.
Performance before the patch:
```
System LIBC reciprocal throughput : 30.205
LIBC reciprocal throughput : 30.533
System LIBC latency : 67.961
LIBC latency : 61.564
```
Performance after the patch:
```
System LIBC reciprocal throughput : 30.409
LIBC reciprocal throughput : 20.273
System LIBC latency : 67.527
LIBC latency : 61.959
```
Reviewed By: orex
Differential Revision: https://reviews.llvm.org/D130901
|
|
There are no AMDGPUSampleVariant versions for _G16, it is treated more like a
modifier for derivatives (_D) (also for intrinsics where it is overloaded type
instead of part of instrinsic name) so we ended up making more variants for
these instruction then we actually needed.
32-bit derivatives need 6 dwords at most, while 16-bit need 4 at most. Using
same AMDGPUSampleVariant for both, we ended up creating 2 extra variants per
instruction than were necessary.
In total this deletes 260 unused tablegen records.
Differential Revision: https://reviews.llvm.org/D131252
|
|
The same predicate is checked on line 12962 just above the removed code.
|
