Petr Hosek [Fri, 9 Jun 2023 07:51:40 +0000 (07:51 +0000)]
[clang] set python3 as required build dependency
The required HTMLLogger include file needs python3 to run
resource_bundle.py to bundle all the html/css/js resources. However, if
user sets -DLLVM_INCLUDE_TESTS=OFF, CMake will not find python3 and the
resource bundler will never be executed. This patch set the python3 as a
required build dependency to fix this problem.
Patch By: Avimitin
Differential Revision: https://reviews.llvm.org/D152418
Nikita Popov [Fri, 9 Jun 2023 07:32:27 +0000 (09:32 +0200)]
[Clang] Remove typed pointer consistency assertions (NFC)
These are no-ops with opaque pointers.
David Green [Fri, 9 Jun 2023 07:32:26 +0000 (08:32 +0100)]
[TypePromotion] Don't treat bitcast as a Source
This removes BitCasts from isSource in Type Promotion, as I don't believe they
need to be treated as Sources. They will usually be from floats or hoisted
constants, where constants will be handled already.
This fixes #62513, but didn't otherwise cause any differences in the tests I
ran.
Differential Revision: https://reviews.llvm.org/D152112
pvanhout [Thu, 8 Jun 2023 10:35:21 +0000 (12:35 +0200)]
[AMDGPU] Port no-hsa-graphic-shaders.ll to code object V4
Split from D146023
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D152432
Phoebe Wang [Fri, 9 Jun 2023 06:33:30 +0000 (14:33 +0800)]
[SelectionDAG] Do not salvage with vector node
rG2eb7cbf987f21 added this code, which results in crash for vector
nodes. This patch solves it by skipping for the vector nodes.
Thanks Steve for helping reducing the test case.
Co-authored-by: Steve Merritt <steve.merritt@intel.com>
Reviewed By: goldstein.w.n
Differential Revision: https://reviews.llvm.org/D152492
Dmitry Makogon [Fri, 9 Jun 2023 06:26:24 +0000 (13:26 +0700)]
[SimpleLoopUnswitch] Verify LoopInfo in turnGuardIntoBranch under a flag
A follow-up for 64397d8. Only do verification if VerifyLoopInfo is
set.
Alex Zinenko [Fri, 9 Jun 2023 06:29:39 +0000 (06:29 +0000)]
[mlir] mark test requiring asserts as such
Pravin Jagtap [Fri, 9 Jun 2023 05:08:44 +0000 (01:08 -0400)]
[AMDGPU] Iterative scan implementation for atomic optimizer.
This patch provides an alternative implementation to DPP for Scan Computations.
An alternative implementation iterates over all active lanes of Wavefront
using llvm.cttz and performs the following steps:
1. Read the value that needs to be atomically incremented using
llvm.amdgcn.readlane intrinsic
2. Accumulate the result.
3. Update the scan result using llvm.amdgcn.writelane intrinsic
if intermediate scan results are needed later in the kernel.
Reviewed By: arsenm, cdevadas
Differential Revision: https://reviews.llvm.org/D147408
Craig Topper [Fri, 9 Jun 2023 05:05:20 +0000 (22:05 -0700)]
[RISCV] Use computeVLMax and getVScale to simplify code. NFC
Amara Emerson [Sat, 3 Jun 2023 23:12:48 +0000 (16:12 -0700)]
[GlobalISel] Implement some binary reassociations, G_ADD for now
- (op (op X, C1), C2) -> (op X, (op C1, C2))
- (op (op X, C1), Y) -> (op (op X, Y), C1)
Some code duplication with the G_PTR_ADD reassociations unfortunately but no
easy way to avoid it that I can see.
Differential Revision: https://reviews.llvm.org/D150230
LLVM GN Syncbot [Fri, 9 Jun 2023 03:44:33 +0000 (03:44 +0000)]
[gn build] Port
60eb1da31555
Chuanqi Xu [Fri, 9 Jun 2023 03:21:29 +0000 (11:21 +0800)]
[Modules] [Sema] Don't try to getAcceptableDecls during the iteration of noload_lookups
I found this during the support of modules for clangd. The reason for
the issue is that the iterator returned by noload_lookups is fast-fail
after the lookup table changes by the design of llvm::DenseMap. And
originally the lookup will try to use getAcceptableDecl to filter the
invisible decls. The key point here is that the function
"getAcceptableDecl" wouldn't stop after it find the specific decl is
invisble. It will continue to visit its redecls to find a visible one.
However, such process involves loading decls from external sources,
which results the invalidation.
Note that the use of "noload_lookups" is rare. It is only used in tools
like FixTypos and CodeCompletions. So it is completely fine for the
tranditional compiler. This is the reason why I can't reproduce it by a
lit test.
wangpc [Fri, 9 Jun 2023 03:01:39 +0000 (11:01 +0800)]
[RISCV][NFC] Remove classes/multiclasses for SEW-aware instructions
Instead, we add an argument `isSEWAware` to indicate that the
instruction is SEW-aware. Actually, the only difference is the
name of pseudo instructions.
And we remove postfix `_E` for all classes/multiclasses and remove
argument `sew` since it can be calculated from `log2sew`.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D152428
wangpc [Fri, 9 Jun 2023 02:55:59 +0000 (10:55 +0800)]
[RISCV][NFC] Replace log2 with !logtwo
`!logtwo` was added in D134068.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D152422
Haohai Wen [Fri, 9 Jun 2023 01:36:25 +0000 (09:36 +0800)]
[NFC][COFF] Split WinCOFFObjectWriter to WinCOFFWriter
We'd like to support -gsplit-dwarf for Windows COFF. It requires to
write Dwo and NonDwo sections to different output streams.The original
implementation is not designed to do that and there can be only one
MCObjectWriter. This patch split WinCOFFObjectWriter to WinCOFFWriter so
that:
1. WinCOFFObjectWriter can create multiple WinCOFFWriter.
2. Each WinCOFFWriter can separately collect sections it is interested.
3. Each WinCOFFWriter can write to it's own output stream.
Reviewed By: skan
Differential Revision: https://reviews.llvm.org/D152340
Phoebe Wang [Fri, 9 Jun 2023 01:03:58 +0000 (09:03 +0800)]
[X86][BF16] Split vNbf16 vectors according to vNf16
Fixes #63017
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D151778
Phoebe Wang [Fri, 9 Jun 2023 01:03:54 +0000 (09:03 +0800)]
[X86][BF16] Share FP16 vector ABI with BF16
The ABI of BF16 is identical to FP16 rather than i16.
Fixes #62997
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D151710
Ellis Hoag [Thu, 8 Jun 2023 22:31:07 +0000 (15:31 -0700)]
[InstrProf][compiler-rt] Fix counter section alignment issue
I recently discovered that `.profraw` headers are expected to be 8 byte
aligned.
https://github.com/llvm/llvm-project/blob/
643ba926c1f618401c86dc37e659df795db2e1a0/llvm/lib/ProfileData/InstrProfReader.cpp#L503-L506
When function entry coverage mode is used, function counters are single
bytes, so it is likely that the size of the counters section is not 8
byte aligned. We can add padding after the counters section to guarantee
this.
Reviewed By: kyulee, gulfem
Differential Revision: https://reviews.llvm.org/D152479
Heejin Ahn [Sat, 3 Jun 2023 09:28:25 +0000 (02:28 -0700)]
[WebAssembly] Error out on invalid personality functions
Without explicitly checking and erroring out, an invalid personality
function, which is not `__gxx_wasm_personality_v0`, caused
a segmentation fault down the line because `WasmEHFuncInfo` was not
created. This explicitly checks the validity of personality functions in
functions with EH pads and errors out explicitly with a helpful error
message. This also adds some more assertions to ensure `WasmEHFuncInfo`
is correctly created and non-null.
Invalid personality functions wouldn't be generated by our Clang, but
can be present in handwritten ll files, and more often, in files
transformed by passes like `metarenamer`, which is often used with
`bugpoint` to simplify names in `bugpoint`-reduced files.
Reviewed By: dschuff
Differential Revision: https://reviews.llvm.org/D152203
Heejin Ahn [Sat, 20 May 2023 00:55:57 +0000 (17:55 -0700)]
[WebAssembly] Undef invalid DBG_VALUEs after RegColoring
After register coalescing, some `DBG_VALUE`s can have incorrect info.
For example, if a `DBG_VALUE` has a register operand `%0`, but it
resides in a live range of `%1`, it has incorrect info after `%0` and
`%1` are coalesced. See the comments for more details.
This does not have meaningful changes on our variable debug info
coverage or compilation time, which is good news.
Reviewed By: dschuff
Differential Revision: https://reviews.llvm.org/D151125
Michael Jones [Thu, 8 Jun 2023 23:36:12 +0000 (16:36 -0700)]
[libc][NFC] fix constants not marked constexpr
The constants in the ryu constant tables were not marked constexpr,
only static const. This caused problems when compiling with GCC.
Differential Revision: https://reviews.llvm.org/D152487
Michael Jones [Thu, 8 Jun 2023 21:25:22 +0000 (14:25 -0700)]
[libc][NFC] land long double table for printf
The Mega Table that printf uses for long doubles with some flags is too
large for the linters, and so has been split out from the main patch.
The main patch: https://reviews.llvm.org/D150399
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D152470
Justin Bogner [Thu, 8 Jun 2023 22:59:38 +0000 (15:59 -0700)]
[C API] Add getters and setters for NUW, NSW, and Exact
This partially addresses Bug 42692; see discussion there.
Adds C API getters and setters for the NUW, NSW, and Exact flags on various
instructions.
Patch by Craig Disselkoen. Thanks!
Differential Revision: https://reviews.llvm.org/D89252
usama hameed [Thu, 8 Jun 2023 23:02:54 +0000 (16:02 -0700)]
Revert "[Sanitizers][Darwin] In DlAddrSymbolizer, return only the module file name instead of the comlpete module path during symbolication."
The commit broke asan_symbolize.py script on Darwin which depended on
using the complete module path to symolize crash traces offline.
This reverts commit
f6ea869f7c043c70722b8db6be94d9ad4cc9eb92.
rdar://
110487521
Joseph Huber [Thu, 8 Jun 2023 22:53:11 +0000 (17:53 -0500)]
[libc] Disable uint test on NVPTX GPUs
This test started failing on Nvidia, we need to disable it to keep the
bot green until we can investigate the root cause.
Differential Revision: https://reviews.llvm.org/D152481
Alex Langford [Mon, 5 Jun 2023 23:50:24 +0000 (16:50 -0700)]
[lldb][NFCI] Remove use of ConstString from OptionValueProperties
In the interest of keeping the ConstString StringPool small, this patch
aims to remove the use of ConstString from OptionValueProperties.
We can maintain quick lookups by using an llvm::StringMap to find the
correct index by name.
Differential Revision: https://reviews.llvm.org/D152210
Alex Langford [Tue, 6 Jun 2023 22:38:52 +0000 (15:38 -0700)]
[lldb][NFCI] Refactor TypeSystemClang::GetBasicTypeEnumeration
`TypeSystemClang::GetBasicTypeEnumeration(ConstString)` builds up a
giant `UniqueCStringMap<lldb::BasicType>` which is effectively a vector whose elements
are (ConstString, lldb::BasicType). Once sorted, we get to have fairly
quick lookups (O(log n) on average). This is fine, but we can do better
on average with an llvm::StringMap.
This also allows us to simplify the initialization code for the lookup,
no more `once_flag`!
Additionally, I removed unused declarations of related functions from
CompilerType and TypeSystemClang
Differential Revision: https://reviews.llvm.org/D152315
Alex Langford [Tue, 6 Jun 2023 21:38:08 +0000 (14:38 -0700)]
[lldb][NFCI] Remove use of ConstString from OptionValueLanguage
ConstString is simply not needed here.
Differential Revision: https://reviews.llvm.org/D152310
spupyrev [Wed, 22 Mar 2023 21:18:03 +0000 (14:18 -0700)]
[BOLT] stale profile matching [part 2 out of 2]
This is a first "serious" version of stale profile matching in BOLT. This diff
extends the hash computation for basic blocks so that we can apply a fuzzy
hash-based matching. The idea is to compute several "versions" of a hash value
for a basic block. A loose version of a hash (computed by ignoring instruction
operands) allows to match blocks in functions whose content has been changed,
while stricter hash values (considering instruction opcodes with operands and
even based on hashes of block's successors/predecessors) allow to resolve
collisions. In order to save space and build time, individual hash components
are blended into a single uint64_t.
There are likely numerous ways of improving hash computation but already this
simple variant provides significant perf benefits.
**Perf testing** on the clang binary: collecting data on clang-10 and using it
to optimize clang-11 (with ~1 year of commits in between). Next, we compare
- //stale_clang// (clang-11 optimized with profile collected on clang-10 with **infer-stale-profile=0**)
- //opt_clang// (clang-11 optimized with profile collected on clang-11)
- //infer_clang// (clang-11 optimized with profile collected on clang-10 with **infer-stale-profile=1**)
`LTO-only` mode:
//stale_clang// vs //opt_clang//: task-clock [delta(%): 9.4252 ± 1.6582, p-value: 0.000002]
(That is, there is a ~9.5% perf regression)
//infer_clang// vs //opt_clang//: task-clock [delta(%): 2.1834 ± 1.8158, p-value: 0.040702]
(That is, the regression is reduced to ~2%)
Related BOLT logs:
```
BOLT-INFO: identified 2114 (18.61%) stale functions responsible for 30.96% samples
BOLT-INFO: inferred profile for 2101 (18.52% of all profiled) functions responsible for 30.95% samples
```
`LTO+AutoFDO` mode:
//stale_clang// vs //opt_clang//: task-clock [delta(%): 19.1293 ± 1.4131, p-value: 0.000002]
//infer_clang// vs //opt_clang//: task-clock [delta(%): 7.4364 ± 1.3343, p-value: 0.000002]
Related BOLT logs:
```
BOLT-INFO: identified 5452 (50.27%) stale functions responsible for 85.34% samples
BOLT-INFO: inferred profile for 5442 (50.23% of all profiled) functions responsible for 85.33% samples
```
Reviewed By: Amir
Differential Revision: https://reviews.llvm.org/D146661
Thomas Lively [Thu, 8 Jun 2023 21:36:29 +0000 (14:36 -0700)]
Revert "Improve WebAssembly vector bitmask, mask reduction, and extending"
This reverts commit
8392bf6000ad039bd0e55383d40a05ddf7b4af13.
The commit missed some edge cases that led to crashes. Reverting to resolve
downstream breakage while a fix is pending.
Michael Jones [Thu, 8 Jun 2023 21:33:57 +0000 (14:33 -0700)]
[libc] disable printf Lf tests for float128 plats
The results for the %Lf tests were calculated on 80 bit long double
systems, meaning the results are not necessarily accurate for float128
systems. In future these will be fixed, but for the moment I'm just
turning them off.
Differential Revision: https://reviews.llvm.org/D152471
Michael Jones [Tue, 28 Mar 2023 18:18:08 +0000 (11:18 -0700)]
[libc] add options to printf decimal floats
This patch adds three options for printf decimal long doubles, and these
can also apply to doubles.
1. Use a giant table which is fast and accurate, but takes up ~5MB).
2. Use dyadic floats for approximations, which only gives ~50 digits of
accuracy but is very fast.
3. Use large integers for approximations, which is accurate but very
slow.
Reviewed By: sivachandra, lntue
Differential Revision: https://reviews.llvm.org/D150399
Shafik Yaghmour [Thu, 8 Jun 2023 20:50:42 +0000 (13:50 -0700)]
[Clang] Add check to Sema::AddAlignedAttr to verify active bits is not out of range
If we provide too large a value for the alignment attribute
APInt::getZExtValue() will assert. This PR adjusts existing check to catch
this case and provide a diagnostic.
This fixes: https://github.com/llvm/llvm-project/issues/50534
Differential Revision: https://reviews.llvm.org/D152335
Matt Arsenault [Mon, 5 Jun 2023 13:25:25 +0000 (09:25 -0400)]
AMDGPU: Fix regressions in integer mad matching
Undo the canonicalize done in
0cfc6510323fbb5a56a5de23cbc65f7cc30fd34c. Restores some regressed
matching of integer mad. The selection patterns fo the actual mads
don't seem to be properly commuting, so some of the commuted cases are
still missed.
Fixes: SWDEV-363009
Felipe de Azevedo Piovezan [Mon, 5 Jun 2023 11:26:20 +0000 (07:26 -0400)]
[AppleAccelTable][NFC] Improve code readability
This commit does a few minor NFC cleanups:
* A variable was called "Atom", probably trying to claim it was an AtomType.
This was incorrect, it is actually a FormValue.
* LLVM provides a `zip_equal` to express the intent of asserting ranges with the
same size. We change the lookup method to use that.
* The use of tuples made the code slightly difficult to follow, as such we
unpack the tuple with structure binding to improve readability.
Depends on D152157
Differential Revision: https://reviews.llvm.org/D152158
Augusto Noronha [Thu, 1 Jun 2023 23:56:56 +0000 (16:56 -0700)]
[lldb] Unconditionally increment depth when printing children
The `target.max-children-depth` setting and `--depth` flag would be
ignored if treating pointer as arrays, fix that by always incrementing
the current depth when printing a new child.
rdar://
109855463
Differential Revision: https://reviews.llvm.org/D151950
Artem Belevich [Tue, 6 Jun 2023 23:05:08 +0000 (16:05 -0700)]
[NVPTX] Allow using v4i32 for memcpy lowering.
Differential Revision: https://reviews.llvm.org/D152317
Martin Storsjö [Wed, 7 Jun 2023 10:53:17 +0000 (13:53 +0300)]
[llvm-lib] Pass the right value of MingwDef to parseCOFFModuleDefinition
In llvm-lib cases, the MinGW specific behaviours shouldn't be
enabled.
Not adding any tests to check the corner case behaviours, as this
simply was a mistake when this codepath was added in
395ec4458fb7fc700551f7017c0a395d68c55873 / D144765.
Differential Revision: https://reviews.llvm.org/D152362
Martin Storsjö [Wed, 7 Jun 2023 10:41:15 +0000 (13:41 +0300)]
[llvm-dlltool] Ignore the --temp-prefix option
llvm-dlltool tolerates unknown options as long as they are plain
flags, but if given the parameter value as a separate argument,
e.g. "--temp-prefix foo", it fails to ignore it.
Differential Revision: https://reviews.llvm.org/D152361
Martin Storsjö [Wed, 7 Jun 2023 11:13:55 +0000 (14:13 +0300)]
[llvm-dlltool] Clarify parameters, simplify ArgList usage. NFC.
Add comments about unclear bool arguments to functions, switch to
hasArg instead of getLastArg for cases where we don't need to check
the argument's value.
Differential Revision: https://reviews.llvm.org/D152360
Martin Storsjö [Wed, 7 Jun 2023 10:55:21 +0000 (13:55 +0300)]
[COFF] Remove misleading and unclear comments. NFC.
It is not planned that GNU binutils would change this aspect of
its behaviour wrt how symbol decoration is done.
Differential Revision: https://reviews.llvm.org/D152359
Nick Desaulniers [Thu, 8 Jun 2023 18:30:00 +0000 (11:30 -0700)]
[ExprConstant] fix ICE from Evaluate on _Atomic structs
Fixes two crashes reported by @efriedma and @ahatanak, as per
@ahatanak's analysis.
Link: https://reviews.llvm.org/D151587#4397446
Link: https://reviews.llvm.org/D151587#4399684
Reviewed By: ahatanak, aaron.ballman
Differential Revision: https://reviews.llvm.org/D152303
Matt Arsenault [Sun, 20 Nov 2022 22:31:04 +0000 (14:31 -0800)]
HIP: Directly call copysign builtins
Florian Hahn [Thu, 8 Jun 2023 20:13:05 +0000 (21:13 +0100)]
[LV] Add tests for reasoning about SCEV predicates.
Add extra tests with cases where SCEV predicates can be proven to always
be false. The test in pointer-induction.ll has been adjusted to avoid
the induction always to wrap.
Matt Arsenault [Sun, 20 Nov 2022 16:48:29 +0000 (08:48 -0800)]
HIP: Directly call isnan builtin
Andrzej Warzynski [Thu, 1 Jun 2023 13:24:22 +0000 (14:24 +0100)]
[mlir][transform] Add support for expressing scalable vector sizes
This patch enables specifying scalable vector sizes when using the
Transform dialect to drive vectorisation, e.g.:
```
transform.structured.masked_vectorize %0 vector_sizes [8, 16, [4]]
```
This is implemented by extending the MaskedVectorizeOp with a dedicated
attribute for "scalability" and by overloading `parseDynamicIndexList`
so that MaskedVectorizeOp can continue using the auto-generated parser
and printer.
At the moment, only the trailing vec size can be scalable. The following
is not yet supported:
```
transform.structured.masked_vectorize %0 vector_sizes [8, [16], [4]]
```
As the vectoriser does not support scalable vectorisation just yet, a
warning is issues when scalable vector sizes are used. You can also use
the debug output, `--debug-only=linalg-vectorization`, to check whether
scalable vectorisation has been switched on.
This change is a part of a larger effort to enable scalable
vectorisation in Linalg. See this RFC for more context:
* https://discourse.llvm.org/t/rfc-scalable-vectorisation-in-linalg/
Similar patch for tiling: https://reviews.llvm.org/D150944
Differential Revision: https://reviews.llvm.org/D151892
Teresa Johnson [Thu, 8 Jun 2023 19:19:44 +0000 (12:19 -0700)]
[MemProf] Remove stale comment (NFC)
We already do the simplification described in the FIXME comment.
Matt Arsenault [Sun, 20 Nov 2022 16:48:03 +0000 (08:48 -0800)]
HIP: Directly call isinf builtins
Matt Arsenault [Sun, 20 Nov 2022 16:38:31 +0000 (08:38 -0800)]
HIP: Directly call signbit builtins
Alex Langford [Wed, 7 Jun 2023 01:10:48 +0000 (18:10 -0700)]
[lldb][NFCI] DecodedThread::TraceItemStorage::error should own its own data
The way it works now, it stores a `const char *` that it does not
explicitly own. It's owned by the ConstString StringPool. This is purely
to manage its lifetime, we don't really benefit from deduplication (nor
should we try to, they are errors). We also don't really benefit from
quick comparisons.
This may make the size of TraceItemStorage larger, but you have to pay
the cost of owning the data somewhere. The ConstString StringPool is an
attractive choice but ultimately a poor one.
Differential Revision: https://reviews.llvm.org/D152326
Joel E. Denny [Thu, 8 Jun 2023 19:09:35 +0000 (15:09 -0400)]
[OpenMP] Fix --libomptarget-nvptx-bc-path in tests
After D151324, which landed as
349c0aacb380, many libomptarget non-LTO
nvptx64 tests fail with errors like:
```
clang: error: bitcode library '/tmp/llvm-project/build/runtimes/runtimes-bins/openmp/libomptarget/libomptarget-nvptx-sm_70.bc' does not exist
```
This patch updates the bc path.
Reviewed By: jhuber6
Differential Revision: https://reviews.llvm.org/D152462
Mahesh Ravishankar [Thu, 8 Jun 2023 18:39:32 +0000 (18:39 +0000)]
[mlir][SliceAnalysis] Add an options object to forward and backward slice.
Add an options object to allow control of the slice computation (for
both forward and backward slice). This makes the ABI stable, and also
allows avoiding an assert that makes the slice analysis unusable for
operations with multiple blocks.
Reviewed By: hanchung, nicolasvasilache
Differential Revision: https://reviews.llvm.org/D151520
Felipe de Azevedo Piovezan [Sat, 3 Jun 2023 11:23:16 +0000 (07:23 -0400)]
[AppleAccelTable] Keep track of the size of each hash data entry
In a future patch, it will be desirable to skip over all hash data entries for a
particular string. In order to do so, we must know how many bytes each of those
entries have.
In its full specification, Apple tables allow for variable length entries, which
would make the above impossible without reading the data of each entry. However,
this is largely unsupported today (as a FIXME in the code indicates, there is a
bug with hash collisions exactly because we don't know how to skip over data),
and the documentation[1] states that:
> For the current implementations of the “.apple_names” (all functions +
> globals), the “.apple_types” (names of all types that are defined), and the
> “.apple_namespaces” (all namespaces), we currently set the Atom array to be:
> [...]
> This defines the contents to be the DIE offset (eAtomTypeDIEOffset) that is
> encoded as a 32 bit value (DW_FORM_data4).
In other words, we only produce fixed sized entries.
A few tests containing invalid dwarf had to be updated, as the error is now
detected earlier (when the accelerator table is being parsed, instead of inside
the explicit call to "verify").
[1]: https://llvm.org/docs/SourceLevelDebugging.html#fixed-lookup
Depends on D152156
Differential Revision: https://reviews.llvm.org/D152157
Alex Langford [Wed, 24 May 2023 00:13:31 +0000 (17:13 -0700)]
[DebugInfo] Add error-handling to DWARFAbbreviationDeclarationSet
This commit aims to improve error handling in the
DWARFAbbreviationDeclarationSet class. Specifically, we change the return type
of DWARFAbbreviationDeclarationSet::extract to an llvm::Error. In doing
so, we propagate the error from DWARFAbbreviationDeclaration::extract
another layer upward.
I have built on the previous unittest for DWARFDebugAbbrev that I
wrote a few days prior.
Namely, I am verifying that the following should give an error:
- An invalid tag following a non-null code
- An invalid attribute with a valid form
- A valid attribute with an invalid form
- An incorrectly terminated DWARFAbbreviationDeclaration
Additionally, I uncovered some invalid DWARF in an unrelated dsymutil
test. Namely the last Abbreviation Decl was missing a code.
This test has been updated accordingly. However, this commit does
not fix the underlying issue: llvm-dwarfdump does not correctly
verify the debug abbreviation section to catch these kinds of
mistakes. I have updated DWARFVerifier to not dereference a
pointer without first checking it and left a FIXME for future
contributors.
Differential Revision: https://reviews.llvm.org/D151353
Krzysztof Parzyszek [Tue, 30 May 2023 00:15:10 +0000 (17:15 -0700)]
[RDF] Add RegisterAggr::refs to return iterator range, NFC
Krzysztof Parzyszek [Mon, 29 May 2023 23:58:28 +0000 (16:58 -0700)]
[RDF] Create individual phi for each indivisible register
This isn't quite using register units, but it's getting close. The phi
generation is driven by register units, but each phi still contains a
reference to a register, potentially with a mask that amounts to a unit.
In cases of explicit register aliasing this may still create phis with
references that are aliased, whereas separate phis would ideally contain
disjoint references (this is all within a single basic block).
Previously phis used maximal registers, now they use minimal. This is a
step towards both, using register units directly, and a simpler liveness
calculation algorithm. The idea is that a phi cannot reach a reference
to anything smaller than the phi itself represents. Before there could
be a phi for R1_R0, now there will be two for this case (assuming R0 and
R1 have one unit each).
Krzysztof Parzyszek [Mon, 29 May 2023 23:34:04 +0000 (16:34 -0700)]
[RDF] Use RegisterAggr instead of RegisterSet in few places
This shouldn't change any functionality.
Krzysztof Parzyszek [Mon, 29 May 2023 18:09:18 +0000 (11:09 -0700)]
[RDF] Clang-format the sources, NFC
eopXD [Thu, 8 Jun 2023 17:50:13 +0000 (10:50 -0700)]
[Clang][RISCV] Adjust naming from `DeclareRISCVVectorBuiltins` to `DeclareRISCVSiFiveVectorBuiltins`. NFC
The variable indicates if RISC-V Sifive vector builtin functions
are enabled or not. The adjusted naming help clarifies this.
cc: @4vtomat
Jakub Beránek [Thu, 8 Jun 2023 16:41:11 +0000 (09:41 -0700)]
Increase memory of BOLT runtime instrumentation bump allocator used for writing resulting profile
The BOLT instrumentation runtime uses a bump allocator that has a fixed amount of maximum memory. In some cases, this memory limit is not large enough (https://github.com/llvm/llvm-project/issues/59174). We are hitting this limit when instrumenting the Rust compiler with BOLT.
This change increases the memory of the bump allocator used for writing the resulting BOLT profile.
Reviewed By: rafauler
Differential Revision: https://reviews.llvm.org/D151891
Matt Arsenault [Sun, 20 Nov 2022 16:34:20 +0000 (08:34 -0800)]
HIP: Directly call fmin/fmax builtins
Alison Zhang [Thu, 8 Jun 2023 17:08:30 +0000 (13:08 -0400)]
[AIX] Update CMake linker flags to use large glink code
Building clang/llvm as a debug build results in TOC overflow even when
using large code model on AIX. This review updates the
llvm/cmake/modules/HandleLLVMOptions.cmake file to use large glink code
in addition to building as a large code model.
Differential Revision: https://reviews.llvm.org/D151733
Vy Nguyen [Thu, 8 Jun 2023 15:22:46 +0000 (11:22 -0400)]
[lldb][NFC]Update debug (eh-frame) tests to work with new dwarwin MC changes
Details:
D144999 potentially changes the debug format (from compact-unwind to dwarf).
Updated this test to no longer prefer debug-frame over eh-frame to be compatible with the new behaviour
Differential Revision: https://reviews.llvm.org/D152449
LLVM GN Syncbot [Thu, 8 Jun 2023 17:00:55 +0000 (17:00 +0000)]
[gn build] Port
4d13896d8a2e
Thorsten Schütt [Thu, 8 Jun 2023 14:36:45 +0000 (16:36 +0200)]
[GlobalIsel][X86] Legalize G_SELECT
with bug fixes
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D152445
Alexandros Lamprineas [Thu, 11 May 2023 23:07:49 +0000 (00:07 +0100)]
Reland "[FuncSpec] Improve the accuracy of the cost model"
Instead of blindly traversing the use-def chain of constant arguments,
compute known constants along the way. Stop as soon as a user cannot
be replaced by a constant. Keep it light-weight by handling some basic
instruction types.
Differential Revision: https://reviews.llvm.org/D150464
Alexandros Lamprineas [Fri, 2 Jun 2023 10:17:20 +0000 (11:17 +0100)]
Reland "[FuncSpec] Replace LoopInfo with BlockFrequencyInfo"
Using AvgLoopIters on any loop is too imprecise making the cost model
favor users inside loop nests regardless of the actual tripcount.
Differential Revision: https://reviews.llvm.org/D150375
eopXD [Sat, 3 Jun 2023 16:35:28 +0000 (09:35 -0700)]
[1/11][Clang][Type] Expand BuiltinTypeBits from 8 to 9 bits
This patch-set aims to expand all variants of the RVV segment
load/store intrinsics. The first commit expands the limitation of
builtin types from 256 to 512. The following patches expands the
full variants one-by-one.
The next patch introduces a number of new RVV tuple types and lets the
total of builtin types exceed the current limit (256). This patch
attempts to increase the bits used so we can continue to expand new
builtin types.
Reviewed By: erichkeane
Differential Revision: https://reviews.llvm.org/D152069
eopXD [Thu, 8 Jun 2023 08:37:09 +0000 (01:37 -0700)]
[Clang][Type] Add static assertion to guard future expansion for BuiltinType numbering
Recent expansion D152070 exceeds the limit and I had a hard time
triaging the bug because the overflow just creates unexpected
behaviors within the compiler.
This patch adds a static assertion to keep an eye for overflows when we
expand more types in the future.
Reviewed By: aaron.ballman
Differential Revision: https://reviews.llvm.org/D152429
Craig Topper [Thu, 8 Jun 2023 16:15:48 +0000 (09:15 -0700)]
[RISCV] Update documentation for -mvscale-min/max options to mention RISC-V. NFC
Craig Topper [Thu, 8 Jun 2023 16:13:43 +0000 (09:13 -0700)]
[RISCV] Support LMUL!=1 for __attribute__((riscv_rvv_vector_bits(N)))
The first patch supported only LMUL=1 types. This patch supports
LMUL!=1.
LMUL is length multiplier that allows multiple vector registers to
be treated as one large register or a fraction of a single vector
register. Supported values for LMUL are 1/8, 1/4, 1/2, 1, 2, 4, and 8.
An LMUL=2 type will be twice as large as an LMUL=1 type. An LMUL=1/2
type will be half the size as an LMUL=1 type.
Type name with "m2" is LMUL=2, "m4" is LMUL=4.
Type name with "mf2" is LMUL=1/2, "mf4" is LMUL=1/4.
For the LMUL!=1 types the user will need to scale __riscv_v_fixed_vlen
by the LMUL before passing to the attribute.
Reviewed By: aaron.ballman
Differential Revision: https://reviews.llvm.org/D150926
Michael Buch [Thu, 8 Jun 2023 15:58:58 +0000 (16:58 +0100)]
[lldb][test] LogTest: Fix stack overflow
The expected function name requires 18 bytes of storage.
Caught by the public ASAN buildbot
Differential Revision: https://reviews.llvm.org/D152454
Fangrui Song [Thu, 8 Jun 2023 16:02:12 +0000 (09:02 -0700)]
[Driver] Make -G TargetSpecific
so that we report `unsupported option '-G' for target ...` on
unsupported targets (most targets).
This error is tested by one target in aix-err-options.c.
Follow-up to D89897 and D90063.
Slava Zakharin [Thu, 8 Jun 2023 15:12:33 +0000 (08:12 -0700)]
[flang][hlfir] Propagate non-default lbounds from the mutable box.
The issue affected type select tests, such that inside the type guard
block the associated variable was using default lbounds instead of
inheriting it from the original variable.
The bridge's `genExprBox` ended up creating BoxValue from a MutableBoxValue
without setting non-default lbounds. So the hlfir.declare generated
for the associated name inside the type guard block was also using
the default lbounds. The fix is to read the value of the mutable box
and propagate the lbounds to the new BoxValue.
The fix might affect more than just select type cases.
Reviewed By: tblah
Differential Revision: https://reviews.llvm.org/D152413
Nikita Popov [Thu, 8 Jun 2023 14:43:26 +0000 (16:43 +0200)]
[Clang] Remove -no-opaque-pointers cc1 flag
Migration of clang tests to opaque pointers is finished, so remove
the -no-opaque-pointers flag.
Differential Revision: https://reviews.llvm.org/D152447
Craig Topper [Thu, 8 Jun 2023 15:47:58 +0000 (08:47 -0700)]
[RISCV] Add test cases to show that rvv_vector_bits attribute is not accepted for vbool or LMUL!=1 RVV types. NFC
The error message isn't great, but it's temporary until we support these.
Reviewed By: aaron.ballman
Differential Revision: https://reviews.llvm.org/D152415
Craig Topper [Thu, 8 Jun 2023 15:41:22 +0000 (08:41 -0700)]
[RISCV] Fix crash in lowerVECTOR_INTERLEAVE when VecVT is an LMUL=8 type.
If VecVT is an LMUL=8 VT, we can't concatenate the vectors as that
would create an illegal type. Instead we need to split the vectors
and emit two VECTOR_INTERLEAVE operations that can each be lowered.
Reviewed By: fakepaper56
Differential Revision: https://reviews.llvm.org/D152411
Craig Topper [Thu, 8 Jun 2023 15:41:07 +0000 (08:41 -0700)]
[RISCV] Fix crash in lowerVECTOR_DEINTERLEAVE when VecVT is an LMUL=8 type.
If VecVT is an LMUL=8 VT, we can't concatenate the vectors as that
would create an illegal type. Instead we need to split the vectors
and emit two VECTOR_DEINTERLEAVE operations that can each be lowered.
Reviewed By: luke, rogfer01
Differential Revision: https://reviews.llvm.org/D152402
Nikita Popov [Thu, 8 Jun 2023 15:32:36 +0000 (17:32 +0200)]
[Clang] Remove some explicit -opaque-pointers options in tests (NFC)
This is the default, so don't specify it explicitly, in preparation
for removing the option.
Mirko Brkusanin [Thu, 8 Jun 2023 15:29:17 +0000 (17:29 +0200)]
[GlobalISel] Check if ShiftAmt is greater then size of operand
matchCombineShlOfExtend did not check if the size of new shift would be
wider then a size of operand. Current condition did not work if the value
being shifted was zero. Updated to support vector splat.
Patch by: Acim Maravic
Differential Revision: https://reviews.llvm.org/D151122
Vladislav Dzhidzhoev [Thu, 8 Jun 2023 15:36:30 +0000 (17:36 +0200)]
Revert "[DebugMetadata] Simplify handling subprogram's retainedNodes field. NFCI (1/7)"
This reverts commit
4418434c6de7a861e241ba2448ea4a12080cf08f.
Reverted because it breaks tests of OCaml bindings.
Krzysztof Parzyszek [Thu, 8 Jun 2023 15:18:07 +0000 (08:18 -0700)]
[Hexagon] Make RDF copy propagation a bit more aggressive
Update the testcase to actually test for RDF's output.
Craig Topper [Thu, 8 Jun 2023 15:23:16 +0000 (08:23 -0700)]
[RISCV] Add VPatBinaryVL_V to VPatBinaryFPWVL_VV_VF_WV_WF like VPatBinaryWVL_VV_VX_WV_WX.
This should hopefully fix a regression from D151969.
Paul Scoropan [Wed, 31 May 2023 15:06:35 +0000 (15:06 +0000)]
[Flang] Replace intrinsic function type generators with single generic function type generator
In a future patch we plan on introducing a large set of Power-PC specific intrinsics. During our prototyping we found that the number of function type generators we were defining, plus those already defined, were reaching an unreasonable number. This patch introduces a generic function type generator function that can be used for almost all cases. The generator supports creating function types with up to 4 arguments and with arguments/return type of types: void, integer, real, and comlex. The intention is for a future patch, which introduces a set of PowerPC-specific vector intrinsics, to also introduce support in the generator for: integer vector, unsigned vector, and real vector types.
Reviewed By: luporl
Differential Revision: https://reviews.llvm.org/D151812
Felipe de Azevedo Piovezan [Sat, 3 Jun 2023 10:53:11 +0000 (06:53 -0400)]
[AppleAccelTable][NFC] Make FormParams member of AppleAccelTable
These are used throughout the class and are recreated every time they are used.
To prevent the risk of it being created incorrectly in different places, we
create it once and in the earliest moment possible: when the table is extracted.
Depends on D151989
Differential Revision: https://reviews.llvm.org/D152156
Zain Jaffal [Thu, 8 Jun 2023 14:57:50 +0000 (15:57 +0100)]
change checking for auto-init metadata to use `equalsStr` instead of casing MDOperand nodes.
Since `MD_annotation` metadata now supports having mutliple strings in the annotation node. casing Operand to string directly will cause a crash. When checking if `MDOperand` equals str you can use `equalsStr` method.
Reviewed By: serge-sans-paille
Differential Revision: https://reviews.llvm.org/D152372
tripleCC [Thu, 8 Jun 2023 14:48:24 +0000 (16:48 +0200)]
[StaticAnalyzer] Fix false negative on NilArgChecker when creating literal object
Fix false negative on NilArgChecker when creating literal object, such
as @[nullableObject].
Patch By tripleCC!
Differential Revision: https://reviews.llvm.org/D152269
Akshay Khadse [Thu, 8 Jun 2023 14:39:56 +0000 (22:39 +0800)]
[x86][MC] Fix movdir64b addressing
This patch is to fix the [[ https://github.com/llvm/llvm-project/issues/63045 | issue 63045]].
Look at the following code:
```
int main(int argc, char *argv[]) {
int arr[1000];
__asm movdir64b rax, ZMMWORD PTR [arr]
return 0;
}
```
Compiling this code using `clang -O0 -fasm-blocks bug.c` gives the a linker error.
The problem seems to be in the generated assembly. Following is the out put of `clang -S -O0 -fasm-blocks bug.c`:
```
movq %rsi, -16(%rbp)
#APP
movdir64b arr, %rax
#NO_APP
xorl %eax, %eax
```
The symbol `arr` should be replaced with some address like `-4(%rbp)`.
This makes me believe that issue is not in the linker, but with the ASM parser.
This issue originates with patch [D145893](https://reviews.llvm.org/D145893). And that's why reverting it fixes the issue. More specifically, the function [isMem512_GR64()](https://github.com/llvm/llvm-project/blob/
ff471dcf7669b1ad7903a44d0773bef4eb175eb9/llvm/lib/Target/X86/AsmParser/X86Operand.h#L404) within the [llvm/lib/Target/X86/AsmParser/X86Operand.h](https://github.com/llvm/llvm-project/blob/main/llvm/lib/Target/X86/AsmParser/X86Operand.h) file.
Reviewed By: skan
Differential Revision: https://reviews.llvm.org/D151863
Nikita Popov [Thu, 8 Jun 2023 14:21:10 +0000 (16:21 +0200)]
[Clang] Convert some tests to opaque pointers (NFC)
These tests all require some adjustments to make sure that struct
types still get generated, mostly done by stripping pointer
indirections.
Some of this may no longer test the situation it was originally
intended for, e.g. the issue from pr18962 just doesn't really
exist anymore with opaque pointers, as we no longer generate
recursive types.
Nishant Patel [Thu, 8 Jun 2023 14:30:54 +0000 (10:30 -0400)]
[mlir][spirv] Add a canonicalization pattern for UModOp
Add a transformation for a pattern like
%6 = "spirv.Constant"() <{value = 32 : i32}> : () -> i32
%7 = "spirv.UMod"(%5, %6) : (i32, i32) -> i32
%8 = "spirv.Constant"() <{value = 4 : i32}> : () -> i32
%9 = "spirv.UMod"(%7, %8) : (i32, i32) -> i32
to transform to
%6 = "spirv.Constant"() <{value = 32 : i32}> : () -> i32
%7 = "spirv.UMod"(%5, %6) : (i32, i32) -> i32
%8 = "spirv.Constant"() <{value = 4 : i32}> : () -> i32
%9= "spirv.UMod"(%5, %8) : (i32, i32) -> i32
Patch By: nbpatel
Reviewed By: antiagainst, kuhar
Differential Revision: https://reviews.llvm.org/D152341
Felipe de Azevedo Piovezan [Thu, 1 Jun 2023 14:55:06 +0000 (10:55 -0400)]
[AppleAccelTable][NFC] Refactor equal_range code
The current implementation of AppleAcceleratorTable::equal_range has a couple of
drawbacks:
1. Assumptions about how the hash table is structured are hard-coded throughout
the code. Unless you are familiar with the data layout of the table, it becomes
tricky to follow the code.
2. We currently load strings from the string table even for hashes that don't
match our current search hash. This is wasteful.
3. There is no error checking in most DataExtractor calls that can fail.
This patch cleans up (1) by making helper methods that hide the details of the
data layout from the algorithms relying on them. This should also help us in
future patches, where we want to expand the interface to allow iterating over
_all_ entries in the table, and potentially clean up the existing Iterator
class.
The changes above also fix (2), as the problem "just vanishes" when you have a
function called "idxOfHashInBucket(SearchHash)".
The changes above also fix (3), as having individual functions allow us to
expose the points in which reading data can fail. This is particularly important
as we would like to share this implementation with LLDB, which requires robust
error handling.
The changes above are also a step towards addressing a comment left by the
original author:
```
/// TODO: Generalize the rest of the AppleAcceleratorTable interface and move it
/// to this class.
```
I suspect a lot of these helper functions created also apply to DWARF 5's
accelerator table, so they could be moved to the base class.
The changes above also expose a bug in this implementation: the previous
algorithm looks at _one_ string inside the bucket, but never bothers checking
for collisions. When the main search loop is written as it is with this patch,
the problem becomes evident. We do not fix the issue in this patch, as it is
intended to be NFC.
Differential Revision: https://reviews.llvm.org/D151989
Nicolas Vasilache [Thu, 8 Jun 2023 14:15:54 +0000 (14:15 +0000)]
[mlir][Linalg][Transform] Fail gracefully instead of asserting
Teresa Johnson [Wed, 7 Jun 2023 22:42:56 +0000 (15:42 -0700)]
[ThinLTO] Ignore callee edge to global variable
Since the symbols in the ThinLTO summary are indexed by GUID we can end
up in corner cases where a callee edge in the combined index goes to a
summary for a global variable. This could happen in the case of hash
collisions, and in the case of SamplePGO profiles could potentially happen
due to code changes (since we synthesize call edges to GUIDs that were
inlined callees in the profiled code).
Handle this by simply ignoring any non-FunctionSummary callees.
Differential Revision: https://reviews.llvm.org/D152406
Nikita Popov [Thu, 8 Jun 2023 08:39:56 +0000 (10:39 +0200)]
[Clang] Convert some tests to opaque pointers (NFC)
Mathieu Fehr [Wed, 7 Jun 2023 09:52:59 +0000 (10:52 +0100)]
[mlir][irdl] NFC: Ensure ops ends with `Op` suffix
IRDL operations were inconsistent in their naming. They now
all end with the `Op` suffix.
Reviewed By: rriddle, jpienaar
Differential Revision: https://reviews.llvm.org/D152354
Navdeep Katel [Mon, 5 Jun 2023 06:39:35 +0000 (12:09 +0530)]
[MLIR][GPU] Add 16-bit version of cudaMemset in cudaRuntimeWrappers
Add 16-bit version of cudaMemset in cudaRuntimeWrappers and update the GPU to LLVM lowering.
Reviewed By: bondhugula
Differential Revision: https://reviews.llvm.org/D151642
Matt Arsenault [Fri, 17 Feb 2023 15:03:40 +0000 (11:03 -0400)]
X86: Add more is.fpclass lowering tests
Matt Arsenault [Tue, 6 Jun 2023 19:25:25 +0000 (15:25 -0400)]
AMDGPU: Don't run AMDGPUAttributor with -O0
Dmitry Makogon [Thu, 8 Jun 2023 11:27:15 +0000 (18:27 +0700)]
[SimpleLoopUnswitch] Verify LoopInfo after turning guards to branches
SplitBlockAndInsertIfThen doesn't correctly update LoopInfo when called
with Unreachable=true, which is the case when we turn guards to branches
in SimpleLoopUnswitch.
This adds LoopInfo verification before returning from turnGuardIntoBranch.
Matt Arsenault [Wed, 7 Jun 2023 22:12:55 +0000 (18:12 -0400)]
AMDGPU: Merge dag and gisel tests for non-entry LDS handling