LLVM GN Syncbot [Sat, 22 Jul 2023 14:23:37 +0000 (14:23 +0000)]
[gn build] Port
57bd882343f8
Nico Weber [Sat, 22 Jul 2023 14:22:32 +0000 (10:22 -0400)]
[gn] port
822c31a0fe827a6 (HAVE_BUILTIN_THREAD_POINTER)
Should maybe be 1 on linux, at least if is_clang. But just disabling
it everywhere is good enough for now.
Nico Weber [Sat, 22 Jul 2023 14:14:09 +0000 (10:14 -0400)]
[gn] port
b0bb68fd3cb7bfba (LLVM_ENABLE_REVERSE_ITERATION in clang/test)
Nico Weber [Sat, 22 Jul 2023 14:12:01 +0000 (10:12 -0400)]
[gn] port
9996e71f2d (LLVM_ENABLE_REVERSE_ITERATION in llvm/test)
Daniil Dudkin [Sat, 22 Jul 2023 13:25:26 +0000 (16:25 +0300)]
[mlir][LLVM] Introduce reduction intrinsics for minimum/maximum
This patch adds supports for the reduction intrinsic
for floating point minimum and maximum that have
been added to LLVM by https://reviews.llvm.org/D152370.
Related to: #63969
Reviewed By: dcaballe
Differential Revision: https://reviews.llvm.org/D155869
Nuno Lopes [Sat, 22 Jul 2023 12:23:32 +0000 (13:23 +0100)]
[Inline] Use poison instead of undef as placeholder [NFC]
Nuno Lopes [Sat, 22 Jul 2023 12:19:42 +0000 (13:19 +0100)]
[StructurizeCFG] Use poison instead of undef as placeholder [NFC]
These are used to create branch instructions. The condition is patched later
Phoebe Wang [Sat, 22 Jul 2023 12:25:44 +0000 (20:25 +0800)]
[X86][BF16] Customize INSERT_VECTOR_ELT for bf16 when feature BF16 is on
Fixes root cause of #63017.
The reason is similar to BUILD_VECTOR. We have legal vector type but
still soft promote for scalar type. So we need to customize these scalar
to vector nodes.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D155961
Matt Arsenault [Fri, 21 Jul 2023 16:33:35 +0000 (12:33 -0400)]
Partially revert "clang/HIP: Remove __llvm_amdgcn_* wrapper hacks"
Revert part of
f407a7399575a6821940973c54754d42e72dd9ce.
Some of the HIP headers were using the f16 rcp inline, such that it
breaks compiling code against non-top-of-tree headers. Need to wait
for a few HIP releases to expire to fully remove these.
Fixes #63981
Matt Arsenault [Sat, 22 Jul 2023 11:45:18 +0000 (07:45 -0400)]
AMDGPU: Silence a gcc warning
Matt Arsenault [Sat, 22 Jul 2023 11:35:04 +0000 (07:35 -0400)]
ConstantFolding: Fix canonicalize folding for dynamic mode denormal inputs
We have to assume the input could be positive-zero. Makes alive2
happy.
Nuno Lopes [Sat, 22 Jul 2023 11:56:03 +0000 (12:56 +0100)]
[CoroSplit] Use poison instead of undef as placeholder [NFC]
Used to construct full structs/vectors
also, covert freeze undef -> freeze poison (same semantics)
Nuno Lopes [Sat, 22 Jul 2023 11:44:03 +0000 (12:44 +0100)]
[SimplifyCFG] Use poison instead of undef as placeholder [NFC]
This is used in a phi node that is created for which only 1 value is accessed (the non-poison)
Fabian Mora [Sat, 22 Jul 2023 11:32:42 +0000 (11:32 +0000)]
[mlir][rocdl] Change the translation of `GridDim*Op` to `__ockl_get_num_groups`
Currently, `ROCDL::GridDim*Op` is being translated to `__ockl_get_global_size`, however
to match the meaning of `gpu.grid_dim` it should instead be translated to
`__ockl_get_num_groups`. This change would also make it agree with the meaning
of `gridDimx.*` in HIP, see:
https://github.com/ROCm-Developer-Tools/hipamd/blob/develop/include/hip/amd_detail/amd_hip_runtime.h#L257
Difference between the functions:
```
__ockl_get_global_size = blockDim * numBlocks
__ockl_get_num_groups = numBlocks
```
Reviewed By: krzysz00
Differential Revision: https://reviews.llvm.org/D156009
Benjamin Kramer [Sat, 22 Jul 2023 11:21:51 +0000 (13:21 +0200)]
Phoebe Wang [Sat, 22 Jul 2023 10:13:58 +0000 (18:13 +0800)]
[X86][BF16] Do not scalarize masked load for BF16 when we have AVX512BF16
Fixes #63017
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D155952
Mark de Wever [Wed, 12 Jul 2023 06:20:26 +0000 (08:20 +0200)]
[libc++][print] Adds stdout functions.
Implements parts of
- P2093R14 Formatted output
- P2539R4 Should the output of std::print to a terminal be
synchronized with the underlying stream?
Depends on D150044
Reviewed By: #libc, ldionne
Differential Revision: https://reviews.llvm.org/D155262
Fangrui Song [Sat, 22 Jul 2023 07:00:57 +0000 (00:00 -0700)]
[VirtualFileSystem] Use std::map::emplace
Matthias Springer [Sat, 22 Jul 2023 06:30:40 +0000 (08:30 +0200)]
[mlir][Transforms] GreedyPatternRewriteDriver: Fix `changed` parameter
`changed` was not updated correctly when it was already set to "true" before calling `applyPatternsAndFoldGreedily`.
Differential Revision: https://reviews.llvm.org/D155934
Matthias Springer [Sat, 22 Jul 2023 06:23:48 +0000 (08:23 +0200)]
[mlir][transform] Add `transform.apply_dce` op
Add a transform that eliminates dead operations. This is useful after certain transforms (such as fusion) that create/clone new IR but leave the original IR in place.
Differential Revision: https://reviews.llvm.org/D155954
Fangrui Song [Sat, 22 Jul 2023 06:23:32 +0000 (23:23 -0700)]
[VirtualFileSystem] Make gcc<7.5 happy after
75d71800aa384ee58663d892c325572f5588df2a
There is a libstdc++ stl_map.h bug that is only back ported to 7.5.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78595#c13
```
/tmp/opt/gcc-7.3.0/include/c++/7.3.0/bits/stl_tree.h:2091:28: error: no matching function for call to ‘std::_Rb_tree<std::__cxx11::basic_string<char>, std::pair<const std::__cxx11::basic_string<char>, std::unique_ptr<llvm::vfs::detail::InMemoryNode> >, std::_Select1st<std::pair<const std::__cxx11::basic_string<char>, std::unique_ptr<llvm::vfs::detail::InMemoryNode> > >, std::less<std::__cxx11::basic_string<char> >, std::allocator<std::pair<const std::__cxx11::basic_string<char>, std::unique_ptr<llvm::vfs::detail::InMemoryNode> > > >::_M_get_insert_unique_pos(std::pair<llvm::StringRef, std::unique_ptr<llvm::vfs::detail::InMemoryNode> >::first_type&)’
= _M_get_insert_unique_pos(_KeyOfValue()(__v));
~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~
```
Just construct a std::string from StringRef to work around it.
Mehdi Amini [Sat, 22 Jul 2023 04:47:33 +0000 (21:47 -0700)]
Fix crash in ODS backend for Type/Attr when an incorrect construct is used for Type/Attr
Instead of crashing, try to print a useful error message.
Valentin Clement [Sat, 22 Jul 2023 04:46:20 +0000 (21:46 -0700)]
[flang][openacc] Fix hasDynamicShape logic
Reviewed By: razvanlupusoru
Differential Revision: https://reviews.llvm.org/D155897
khei4 [Sat, 22 Jul 2023 04:08:12 +0000 (13:08 +0900)]
Revert "Reapply: [MemCpyOpt] implement single BB stack-move optimization which unify the static unescaped allocas"
revert because crash on chrome windows https://reviews.llvm.org/D153453#4524256
This reverts commit
569769b64858fd38f41267db41b461d3163aa754.
Nathan Ridge [Fri, 21 Jul 2023 08:05:14 +0000 (04:05 -0400)]
[clang] Add serialization support for the DynamicAllocLValue variant of APValue::LValueBase::Ptr
Differential Revision: https://reviews.llvm.org/D154471
Fangrui Song [Sat, 22 Jul 2023 03:05:35 +0000 (20:05 -0700)]
[CodeGen] Support bitcode input containing multiple modules
When using -fsplit-lto-unit (explicitly specified or due to using
-fsanitize=cfi/-fwhole-program-vtables), the emitted LLVM IR contains a module
flag metadata `"EnableSplitLTOUnit"`. If a module contains both type metadata
and `"EnableSplitLTOUnit"`, `ThinLTOBitcodeWriter.cpp` will write two modules
into the bitcode file. Compiling the bitcode (not ThinLTO backend compilation)
will lead to an error due to `parseIR` requiring a single module.
```
% clang -flto=thin a.cc -c -o a.bc
% clang -c a.bc
% clang -fsplit-lto-unit -flto=thin a.cc -c -o a.bc
% clang -c a.bc
error: Expected a single module
1 error generated.
```
There are multiple ways to have just one module in a bitcode file
output: `-Xclang -fno-lto-unit`, not using features like `-fsanitize=cfi`,
using `-fsanitize=cfi` with `-fno-split-lto-unit`. I think whether a
bitcode input file contains 2 modules (internal implementation strategy)
should not be a criterion to require an additional driver option when
the user seek for a non-LTO compile action.
Let's place the extra module (if present) into CodeGenOptions::LinkBitcodeFiles
(originally for -cc1 -mlink-bitcode-file). Linker::linkModules will link the two
modules together. This patch makes the following commands work:
```
clang -S -emit-llvm a.bc
clang -S a.bc
clang -c a.bc
```
Reviewed By: ormris
Differential Revision: https://reviews.llvm.org/D154923
Richard Smith [Sat, 22 Jul 2023 00:37:55 +0000 (17:37 -0700)]
Optimize emission of `dynamic_cast` to final classes.
- When the destination is a final class type that does not derive from
the source type, the cast always fails and is now emitted as a null
pointer or call to __cxa_bad_cast.
- When the destination is a final class type that does derive from the
source type, emit a direct comparison against the corresponding base
class vptr value(s). There may be more than one such value in the case
of multiple inheritance; check them all.
For now, this is supported only for the Itanium ABI. I expect the same thing is
possible for the MS ABI too, but I don't know what guarantees are made about
vfptr uniqueness.
Reviewed By: rjmccall
Differential Revision: https://reviews.llvm.org/D154658
Med Ismail Bennani [Fri, 21 Jul 2023 23:43:24 +0000 (16:43 -0700)]
[lldb] Convert script native types to StructuredData counterpart
This patch adds the ability to pass native types from the script
interpreter to methods that use a {SB,}StructuredData argument.
To do so, this patch changes the `ScriptedObject` struture that holds
the pointer to the script object as well as the originating script
interpreter language. It also exposes that to the SB API via a new class
called `SBScriptObject`.
This structure allows the debugger to parse the script object and
convert it to a StructuredData object. If the type is not compatible
with the StructuredData types, we will store its pointer in a
`StructuredData::Generic` object.
This patch also adds some SWIG typemaps that checks the input argument to
ensure it's either an SBStructuredData object, in which case it just
passes it throught, or a python object that is NOT another SB type, to
provide some guardrails for the user.
rdar://
111467140
Differential Revision: https://reviews.llvm.org/D155161
Signed-off-by: Med Ismail Bennani <ismail@bennani.ma>
Sunil Srivastava [Sat, 22 Jul 2023 01:18:21 +0000 (18:18 -0700)]
Set default C++ level for PlayStation(r) to C++17.
Sam McCall [Wed, 19 Jul 2023 02:49:54 +0000 (04:49 +0200)]
[include-cleaner] allow spelling strategies to customize verbatim/system headers
Our use case is wanting to apply a spelling strategy to rewrite the spellings
written in IWYU pragma private directives.
Differential Revision: https://reviews.llvm.org/D155671
Felipe de Azevedo Piovezan [Fri, 21 Jul 2023 23:39:31 +0000 (16:39 -0700)]
[lldb] Consider OP_addrx in DWARFExpression::Update_DW_OP_addr
This rewrites DW_OP_addrx inside DWARFExpression as an DW_OP_addr so
that we can update addresses that are originally located in the
debug_addr section.
The full discussion behind this can be found in
https://discourse.llvm.org/t/dwarfexpression-and-dw-op-addrx/71627/12, but a
summary follows.
When SymbolFileDWARF::ParseVariableDIE creates DWARFExpressions for
variables whose location is an OP_addr, it knows how to remap
addresses appropriately in the DebugMap case. It then calls
DWARFExpression::Update_DW_OP_addr, which updates the value associated
with OP_addr.
However, when we have an OP_addrx, the update function does
nothing. This makes sense, as the DWARFExpression does not have a
mutable view of the debug_addr section. In non-DebugMap flows this is
not an issue, as the debug_addr contains the correct addresses of
variables. In DebugMap flows, this is problematic because the work
done by RelinkOSOAddress is lost. By updating the OP to OP_addr, we
can also update the address as required,
We also explored the alternative of relinking the debug_addr section
when we are initializing OSOs (InitOSO). However, this creates an
inconsistent story for users of
DWARFExpression::GetLocation_DW_OP_addr. This function returns an
address without telling callers whether that address came from an
OP_addr or an OP_addrx. If we preemptively relink OP_addrx results
without doing the same for OP_addr results, then callers can’t know
whether the address they got was an executable address or an object
file address. In other words, they can’t know whether they need to
call LinkOSOFileAddress on those results or not.
This patch addresses the majority of test failures when enabling DWARF
5 for MachO (over 200 test failures).
Differential Revision: https://reviews.llvm.org/D155198
TatWai Chong [Fri, 21 Jul 2023 23:28:45 +0000 (16:28 -0700)]
[mlir][tosa][fix] Add proper type checking trait for tosa mul
when operating integer type tensors, tosa elementwise multiplication
requires the element type of result to be a 32-bit integer rather
than the same type as inputs.
Change-Id: Ifd3d7ebd879be5c6b2c8e23aa6d7ef41f39c6d41
Reviewed By: mgehre-amd
Differential Revision: https://reviews.llvm.org/D154988
Evan Wilde [Fri, 21 Jul 2023 23:17:16 +0000 (16:17 -0700)]
[NFC] Remove extra semicolons in clang/lib/APINotes/APINotesFormat.h
There are some trailing semicolons on namespaces in
clang/lib/APINotes/APINotesFormat.h. These result in warnings while
building and don't need to be there.
Reviewed By: compnerd
Differential Revision: https://reviews.llvm.org/D105049
Kazu Hirata [Fri, 21 Jul 2023 23:14:41 +0000 (16:14 -0700)]
[AMDGPU] Fix an unused variable warning
This patch fixes:
llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp:1006:9: error:
unused variable 'Ty' [-Werror,-Wunused-variable]
Jon Roelofs [Fri, 21 Jul 2023 19:51:39 +0000 (12:51 -0700)]
Enable compact unwind in all darwin simulators
... since they've always supported it.
rdar://
104359594
Differential revision: https://reviews.llvm.org/D155988
Philip Reames [Fri, 21 Jul 2023 23:02:33 +0000 (16:02 -0700)]
Revert "[RISCV] Add test which shows alignment of constant pools and the functions which followed"
This reverts commit
cbf2a6ce197e8176c01316fe25400aae0b7390c4. This was a precommited test for a change which is being abandoned.
Nathan Chancellor [Fri, 21 Jul 2023 22:23:08 +0000 (15:23 -0700)]
Revert "Reapply [IR] Mark and constant expressions as undesirable"
This reverts commit
086ee99564afbb11449c08ea2e094f7f49fadde5.
This patch causes an infinite loop when building arch/mips/mm/c-r4k.c in
the Linux kernel. See the comment in Phabricator for a reduced
reproducer: https://reviews.llvm.org/rG086ee99564afbb11449c08ea2e094f7f49fadde5
Matt Arsenault [Fri, 21 Jul 2023 22:52:50 +0000 (18:52 -0400)]
AMDGPU: Fix variables only used in asserts
Matt Arsenault [Sun, 16 Jul 2023 12:32:08 +0000 (08:32 -0400)]
AMDGPU: Implement new 2ulp fdiv lowering
Extends the new frexp scaled reciprocal to the general case. The
reciprocal case is just the same thing when frexp of 1 is constant
folded. Could probably clean up the code to rely on that constant
folding.
Improves results for the IEEE path for the default OpenCL division. We
used to only emit the fdiv.fast intrinsic with a 2.5 ulp accuracy
threshold with DAZ, which uses explicit range checks. This gives us a
better fast option with the default IEEE behavior.
Matt Arsenault [Wed, 19 Jul 2023 14:11:53 +0000 (10:11 -0400)]
AMDGPU: Refactor AMDGPUCodeGenPrepare fdiv handling
NFC-ish. Does trigger some reordering of the fdiv scalarization. Also
skips scalarizing in more cases where nothing was going to happen. We
can still scalarize in some no-op edge cases.
https://reviews.llvm.org/D155740
Richard Smith [Fri, 21 Jul 2023 21:53:36 +0000 (14:53 -0700)]
Track the RequestingModule in the HeaderSearch LookupFile cache.
Different requesting modules can have different lookup results, so don't
cache results across modules.
Fixes a regression introduced in reviews.llvm.org/D132779.
Test case based on one provided by Jan Svoboda.
Reviewed By: jansvoboda11
Differential Revision: https://reviews.llvm.org/D156000
Peter Klausler [Thu, 20 Jul 2023 19:37:25 +0000 (12:37 -0700)]
[flang] Fix portability warning that was incorrectly an "else if"
A semantics check for an assumed-length dummy procedure pointer was
inappropriately part of an "else" clause for a preceding check,
causing it to not be applied in all situations.
Differential Revision: https://reviews.llvm.org/D155975
Jakub Kuderski [Fri, 21 Jul 2023 22:17:39 +0000 (18:17 -0400)]
[mlir][arith] Add canon pattern for chained `arith.muli`
@benvanik reported this as missing.
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D155907
Kirill Stoimenov [Fri, 21 Jul 2023 21:17:50 +0000 (21:17 +0000)]
[HWASAN][LSAN] Replace cstdint with stdint.h
Reviewed By: brooksmoses, MaskRay
Differential Revision: https://reviews.llvm.org/D155994
Peter Klausler [Thu, 20 Jul 2023 21:44:46 +0000 (14:44 -0700)]
[flang] Emit just one warning for a bad format edit descriptor
An attempt to use an edit descriptor (other than A or L) in a FORMAT
statement without arequired 'w' width will elicit warnings from both
the parser and the I/O checker in semantics. Remove the warning from
the parser.
Differential Revision: https://reviews.llvm.org/D155977
Rafael Ubal Tena [Fri, 21 Jul 2023 21:48:11 +0000 (14:48 -0700)]
TOSA-to-Linalg lowering for element-wise ops
- Wrote complete documentation for the `Broadcastable` op trait. This is mostly meant as a thorough description of its previous behavior, with the exception of minor feature updates.
- Restricted legality criteria for a `Broadcastable` op in order to simplify current and future lowering passes and increase efficiency of code generated by those passes. New restriction are: 1) A dynamic dimension in an inferred result is not compatible with a static dimension in the actual result. 2) Broadcast semantics are restricted to input operands and not supported between inferred and actual result shapes.
- Implemented TOSA-to-Linalg lowering support for unary, binary, tertiary element-wise ops. This support is complete for all legal cases described in the `Broadcastable` trait documentation.
- Added unit tests for `tosa.abs`, `tosa.add`, and `tosa.select` as examples of unary, binary, and tertiary ops.
Reviewed By: eric-k256
Differential Revision: https://reviews.llvm.org/D153291
Philip Reames [Fri, 21 Jul 2023 19:39:45 +0000 (12:39 -0700)]
[RISCV] Add test which shows alignment of constant pools and the functions which followed
Peter Klausler [Thu, 20 Jul 2023 20:50:10 +0000 (13:50 -0700)]
[flang][runtime] Fix NORM2([negative, ...])
NORM2 is broken for arrays that start with a negative number
because it sets the initial running max_ value to that number
rather than to its absolute value.
Differential Revision: https://reviews.llvm.org/D155976
Jacques Pienaar [Fri, 21 Jul 2023 21:54:43 +0000 (14:54 -0700)]
[mlir] Move attr -> properties to not require Operation
This allows for converting before/without an Operation is created.
Differential Revision: https://reviews.llvm.org/D155996
Peter Klausler [Thu, 20 Jul 2023 18:46:31 +0000 (11:46 -0700)]
[flang] Strengthen procedure compatibility checking
Add more checks to procedure compatibility testing for procedure pointer
assignments, actual procedure arguments, &c. Specifically, don't
allow corresponding dummy data objects to differ in their use
of polymorphism, assumed size arrays, or assumed shape arrays.
Differential Revision: https://reviews.llvm.org/D155974
Joseph Huber [Wed, 12 Jul 2023 18:35:01 +0000 (13:35 -0500)]
[NVPTX] Add initial support for '.alias' in PTX
This patch adds initial support for using aliases when targeting PTX. We
perform a pretty strict conversion from the globals referenced to the
expected output. as described in the PTX documentation at
https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#kernel-and-function-directives-alias
These cannot currently be used due to a bug in the `nvlink`
implementation that causes aliases to pruned functions to crash the
linker.
Reviewed By: tra
Differential Revision: https://reviews.llvm.org/D155211
Peter Klausler [Thu, 20 Jul 2023 17:39:04 +0000 (10:39 -0700)]
[flang] Portability warning and documentation for an obscure extension
A quotation mark can appear in a Fortran character literal by doubling
it; for example, PRINT *, "'""'" prints '"'. When those doubled
quotation marks are split by a free form line continuation, the
continuation line should have an ampersand before the second quotation
mark. But most compilers, including this one, allow the second
quotation mark to appear as the first character on the continuation
line, too.
So this works:
print *, "'"&
"'"
but it really should be written as:
print *, "'"&
&"'"
Emit a portability warning and document that we support this near-universal
extension.
Differential Revision: https://reviews.llvm.org/D155973
Sam McCall [Sat, 15 Jul 2023 20:21:25 +0000 (22:21 +0200)]
[clangd] Allow indexing of __reserved_names outside system headers
The special handling for these names was added in
https://github.com/llvm/llvm-project/commit/
055d8090d1d5137dab88533995e0c5d9b5390c28
and the motivation was the C++ standard library.
It turns out some projects like using these names anyway, in particular the
linux kernel. D153946 proposed making this a config option, but there are
some implementation issues with the config system.
As an alternative, this patch tweaks the heuristic so we only drop these symbols
in system headers. This does the right thing for linux and the C++ standard
library, at least.
Fixes https://github.com/clangd/clangd/issues/1680
Fixes https://github.com/llvm/llvm-project/issues/63862
Differential Revision: https://reviews.llvm.org/D155381
Paul Robinson [Thu, 20 Jul 2023 16:44:18 +0000 (09:44 -0700)]
[Headers][doc] Add SHA1/SHA256 intrinsic descriptions
Differential revision: https://reviews.llvm.org/D155861
Peter Klausler [Fri, 21 Jul 2023 21:16:56 +0000 (14:16 -0700)]
[flang] Rename new test file
A recent patch added a new test file whose name conflicts with an existing
test on case-insensitive filesystems. Rename it.
Peter Klausler [Wed, 19 Jul 2023 22:02:32 +0000 (15:02 -0700)]
[flang] Ensure that NULL(without MOLD=) not passed to dummy argument with assumed type parameters
A dummy argument with an assumed (*) character length or derived type parameter
value specification needs to be associated with an actual argument that can
supply a value for it, so make sure that a NULL without a MOLD= is not being
passed.
Differential Revision: https://reviews.llvm.org/D155971
Sam McCall [Thu, 20 Jul 2023 22:23:35 +0000 (00:23 +0200)]
[clangd] Use index for go-to-type
This ensures it finds the definition even if not visible.
Differential Revision: https://reviews.llvm.org/D155898
Sam McCall [Mon, 17 Jul 2023 06:02:40 +0000 (08:02 +0200)]
[clangd] Add BlockEnd comments for control flow statements
These mark the end of CompoundStmts bodies of if/while/for/switch.
To identify which statement is being ended, we include abbreviated
text of the condition/loop variable.
Differential Revision: https://reviews.llvm.org/D155421
Peter Klausler [Wed, 19 Jul 2023 20:56:32 +0000 (13:56 -0700)]
[flang] Disallow ASYNCHRONOUS for subroutine
The check for inappropriate usage of the ASYNCHRONOUS attribute
needed to be moved in declaration checking so that it can catch
attempts to use it on a subroutine.
Differential Revision: https://reviews.llvm.org/D155970
Peter Klausler [Wed, 19 Jul 2023 20:23:53 +0000 (13:23 -0700)]
[flang] Enforce F'2023 C7125
An item whose declared type is ABSTRACT may not appear in an
array constructor.
Differential Revision: https://reviews.llvm.org/D155969
Joseph Huber [Mon, 10 Jul 2023 14:38:37 +0000 (09:38 -0500)]
[libc] Remove test RPC opcodes from the exported header
This patch does the noisy work of removing the test opcodes from the
exported interface to an interface that is only visible in `libc`. The
benefit of this is that we both test the exported RPC registration more
directly, and we do not need to give this interface to users.
I have decided to export any opcode that is not a "core" libc feature as
having its MSB set in the opcode. We can think of these as non-libc
"extensions".
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D154848
Matt Arsenault [Mon, 3 Jul 2023 14:22:24 +0000 (10:22 -0400)]
AMDGPU: Overhaul and improve rcp and rsq f32 formation
The highlight change is a new denormal safe 1ulp lowering which uses
rcp after using frexp to perform input scaling. This saves 2
instructions compared to other implementations which performed an
explicit denormal range change. This improves the OpenCL default, and
requires a flag for HIP. I don't believe there's any flag wired up for
OpenMP to emit the necessary fpmath metadata.
This provides several improvements and changes that were hard to
separate without regressing one case or another. Disturbingly the
OpenCL conformance test seems to have the reciprocal test commented
out. I locally hacked it back in to test this.
Starts introducing f32 rsq intrinsics in AMDGPUCodeGenPrepare. Like
the rcp case, we could do this in codegen if !fpmath were preserved
(although we would lose some computeKnownFPClass tricks). Start
requiring contract flags to form rsq. The rsq fusion actually improves
the result from ~2ulp to ~1ulp. We have some older fusion in codegen
which only keys off unsafe math which should be refined.
Expand rsq patterns by checking for denormal inputs and pre/post
multiplying like the current library code does. We also take advantage
of computeKnownFPClass to avoid the scaling when we can statically
prove the input cannot be a denormal. We could do the same for the rcp
case, but unlike rsq a large input can underflow to denormal. We need
additional upper bound exponent checks on the input in order to do the
same for rcp.
This rsq handling also now starts handling the negated case. We
introduce rsq with an fneg. In the case the fneg doesn't fold into its
user, it's a neutral change but provides improvement if it is foldable
as a source modifier.
Also starts respecting the arcp attribute properly, and more strictly
interprets afn. We were previously interpreting afn as implying you
could do the reciprocal expansion of an fdiv. The codegen handling of
these also needs to be revisited.
This also effectively introduces the optimization
combineRepeatedFPDivisors enables, just done in the IR instead (and
only for f32).
This is almost across the board better. The one minor regression is
for gfx6/buggy frexp case where for multiple reciprocals, we could
previously reuse rematerialized constants per instance (it's neutral
for a single rcp).
The fdiv.fast and sqrt handling need to be revisited next.
https://reviews.llvm.org/D155593
Srishti Srivastava [Fri, 21 Jul 2023 20:28:57 +0000 (13:28 -0700)]
[MLIR][ANALYSIS] Add liveness analysis utility
This commit adds a utility to implement liveness analysis using the
sparse backward data-flow analysis framework. Theoretically, liveness
analysis assigns liveness to each (value, program point) pair in the
program and it is thus a dense analysis. However, since values are
immutable in MLIR, a sparse analysis, which will assign liveness to
each value in the program, suffices here.
Liveness analysis has many applications. It can be used to avoid the
computation of extraneous operations that have no effect on the memory
or the final output of a program. It can also be used to optimize
register allocation. Both of these applications help achieve one very
important goal: reducing runtime.
A value is considered "live" iff it:
(1) has memory effects OR
(2) is returned by a public function OR
(3) is used to compute a value of type (1) or (2).
It is also to be noted that a value could be of multiple types (1/2/3) at
the same time.
A value "has memory effects" iff it:
(1.a) is an operand of an op with memory effects OR
(1.b) is a non-forwarded branch operand and a block where its op could
take the control has an op with memory effects.
A value `A` is said to be "used to compute" value `B` iff `B` cannot be
computed in the absence of `A`. Thus, in this implementation, we say that
value `A` is used to compute value `B` iff:
(3.a) `B` is a result of an op with operand `A` OR
(3.b) `A` is used to compute some value `C` and `C` is used to compute
`B`.
---
It is important to note that there already exists an MLIR liveness
utility here: llvm-project/mlir/include/mlir/Analysis/Liveness.h. So,
what is the need for this new liveness analysis utility being added by
this commit? That need is explained as follows:-
The similarities between these two utilities is that both use the
fixpoint iteration method to converge to the final result of liveness.
And, both have the same theoretical understanding of liveness as well.
However, the main difference between (a) the existing utility and (b)
the added utility is the "scope of the analysis". (a) is restricted to
analysing each block independently while (b) analyses blocks together,
i.e., it looks at how the control flows from one block to the other,
how a caller calls a callee, etc. The restriction in the former implies
that some potentially non-live values could be marked live and thus the
full potential of liveness analysis will not be realised.
This can be understood using the example below:
```
1 func.func private @private_dead_return_value_removal_0() -> (i32, i32) {
2 %0 = arith.constant 0 : i32
3 %1 = arith.addi %0, %0 : i32
4 return %0, %1 : i32, i32
5 }
6 func.func @public_dead_return_value_removal_0() -> (i32) {
7 %0:2 = func.call @private_dead_return_value_removal_0() : () -> (i32, i32)
8 return %0#0 : i32
9 }
```
Here, if we just restrict our analysis to a per-block basis like (a), we
will say that the %1 on line 3 is live because it is computed and then
returned outside its block by the function. But, if we perform a
backward data-flow analysis like (b) does, we will say that %0#1 of line
7 is not live because it isn't returned by the public function and thus,
%1 of line 3 is also not live. So, while (a) will be unable to suggest
any IR optimizations, (b) can enable this IR to convert to:-
```
1 func.func private @private_dead_return_value_removal_0() -> i32 {
2 %0 = arith.constant 0 : i32
3 return %0 : i32
4 }
5 func.func @public_dead_return_value_removal_0() -> i32 {
6 %0 = call @private_dead_return_value_removal_0() : () -> i32
7 return %0 : i32
8 }
```
One operation was removed and one unnecessary return value of the
function was removed and the function signature was modified. This is an
optimization that (b) can enable but (a) cannot. Such optimizations can
help remove a lot of extraneous computations that are currently being
done.
Signed-off-by: Srishti Srivastava <srishtisrivastava.ai@gmail.com>
Reviewed By: matthiaskramm, jcai19
Differential Revision: https://reviews.llvm.org/D153779
Peter Klausler [Wed, 19 Jul 2023 19:06:31 +0000 (12:06 -0700)]
[flang] Catch case of character array constructor with indeterminable length
F'2023 7.8 para 5 requires that an implied DO loop with no iterations
in a character array constructor should have items whose lengths are
constant expressions independent of the value of the implied DO loop
index.
Differential Revision: https://reviews.llvm.org/D155968
Daniel Hoekwater [Tue, 27 Jun 2023 01:30:27 +0000 (01:30 +0000)]
[AArch64] Move branch relaxation after bbsection assignment
Because branch relaxation needs to factor in if branches target
a block in the same section or a different one, it needs to run
after the Basic Block Sections / Machine Function Splitting passes.
Because Jump table compression relies on block offsets remaining
fixed after the table is compressed, we must also move the JT
compression pass.
The only tests affected are ones enforcing just the ordering and
the a few that have basic block ids changed because RenumberBlocks
hasn't run yet.
Differential Revision: https://reviews.llvm.org/D153829
Alexey Bataev [Fri, 21 Jul 2023 20:13:01 +0000 (13:13 -0700)]
[SLP][NFC]Add a test with strided loads, NFC.
Peter Klausler [Wed, 19 Jul 2023 00:05:47 +0000 (17:05 -0700)]
[flang][runtime] Detect NEWUNIT= without FILE= or STATUS='SCRATCH'
It is an error to open a new unit with OPEN(NEWUNIT=) and have
neither a file name nor a scratch status. Catch it, and report a
new error code.
Differential Revision: https://reviews.llvm.org/D155967
Florian Hahn [Fri, 21 Jul 2023 20:05:50 +0000 (22:05 +0200)]
[LV] Replace use of getMaxSafeDepDist with isSafeForAnyVector (NFC)
Replace the use of getMaxSafeDepDistBytes with the more direct
isSafeForAnyVector. This removes the need to define getMaxSafeDepDistBytes.
Matt Arsenault [Wed, 3 May 2023 13:52:53 +0000 (09:52 -0400)]
ValueTracking: Implement computeKnownFPClass for frexp
Work around the lack of proper multiple return values by looking
at the extractvalue.
https://reviews.llvm.org/D150982
Matt Arsenault [Wed, 3 May 2023 11:37:16 +0000 (07:37 -0400)]
ValueTracking: Add baseline tests for frexp handling in computeKnownFPClass
Matt Arsenault [Sun, 2 Jul 2023 00:21:28 +0000 (20:21 -0400)]
AMDGPU: Add baseline test for fdiv combine
Peter Klausler [Tue, 18 Jul 2023 23:14:49 +0000 (16:14 -0700)]
[flang] Preserve errors from generic matching
When searching for a matching specific procedure for a set of actual
arguments in a type-bound generic interface for a defined operator,
don't discard any error messages that may have been produced for
the specific that was found. Tweak the code to preserve those
messages and add them to the context's messages, and add a test.
Differential Revision: https://reviews.llvm.org/D155966
Slava Zakharin [Fri, 21 Jul 2023 19:11:51 +0000 (12:11 -0700)]
[flang][hlfir] Added missing fir.convert for i1 result of hlfir.dot_product.
Some operations using the result of hlfir.dot_product can tolerate
that the type of the result changes from !fir.logical to i1 during
intrinsics lowering, but some won't. I added a separate LIT case with
fir.store to mimic one of the nag tests.
Reviewed By: kiranchandramohan
Differential Revision: https://reviews.llvm.org/D155914
Slava Zakharin [Fri, 21 Jul 2023 19:11:42 +0000 (12:11 -0700)]
[flang][hlfir] Preserve polymorphism for the result of hlfir.transpose.
Reviewed By: kiranchandramohan
Differential Revision: https://reviews.llvm.org/D155912
Slava Zakharin [Fri, 21 Jul 2023 19:11:25 +0000 (12:11 -0700)]
[NFC][flang] Distinguish MATMUL and MATMUL-TRANSPOSE printouts.
When MatmulTranpose reports incorrect shapes of the arguments
it cannot represent itself as MATMUL, because the reading
of the first argument's shape will be confusing.
Reviewed By: kiranchandramohan
Differential Revision: https://reviews.llvm.org/D155911
Peter Klausler [Tue, 18 Jul 2023 20:31:23 +0000 (13:31 -0700)]
[flang] Stricter checking of DIM= arguments to LBOUND/UBOUND/SIZE
DIM= arguments with constant values can be checked for validity
even when other arguments to an intrinsic function can't be
folded. Handle errors with assumed-rank arguments as well.
Differential Revision: https://reviews.llvm.org/D155964
Peter Klausler [Tue, 18 Jul 2023 21:53:21 +0000 (14:53 -0700)]
[flang] Finalize &/or destroy ABSTRACT types
The runtime type information tables always flag ABSTRACT types as
needing neither destruction in general nor finalization in particular.
This is incorrect. Although an ABSTRACT type may not itself have
a FINAL procedure -- its argument cannot be polymorphic, but
ABSTRACT types in declarations must always be so -- it can still
have finalizable components &/or components requiring deallocation.
Differential Revision: https://reviews.llvm.org/D155965
Noah Goldstein [Fri, 21 Jul 2023 18:31:47 +0000 (13:31 -0500)]
[InstCombine] If there is a known-bit transform is_pow2 check to just check for any other bits
in `ctpop(X) eq/ne 1` or `ctpop(X) ugt/ule 1`, if there is any
known-bit in `X`, instead of going through `ctpop`, we can just test
if there are any other known bits in `X`. If there are, `X` is not a
power of 2. If there aren't, `X` is a power of 2.
https://alive2.llvm.org/ce/z/eLMJgU
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D152677
Noah Goldstein [Mon, 12 Jun 2023 03:38:46 +0000 (22:38 -0500)]
[InstCombine] Add tests for ispow2 comparisons with a known bit; NFC
Differential Revision: https://reviews.llvm.org/D152676
Noah Goldstein [Sun, 11 Jun 2023 21:13:30 +0000 (16:13 -0500)]
[InstCombine] Canonicalize `(X^(X-1)) u{ge,lt} X` as pow2 test
https://alive2.llvm.org/ce/z/T8osF6
Differential Revision: https://reviews.llvm.org/D152673
Noah Goldstein [Sun, 11 Jun 2023 21:41:46 +0000 (16:41 -0500)]
[InstCombine] Add tests for canonicalizing `(X^(X-1)) u{ge,lt} X` as pow2 test; NFC
Differential Revision: https://reviews.llvm.org/D152672
Peter Klausler [Tue, 18 Jul 2023 16:32:33 +0000 (09:32 -0700)]
[flang] Support implicit global external as procedure pointer target
A name that has been used to reference an undeclared global external
procedure should be accepted as the target of a procedure pointer
assignment statement.
Fixes llvm-test-suite/Fortran/gfortran/regression/proc_ptr_45.f90.
Differential Revision: https://reviews.llvm.org/D155963
Peter Klausler [Mon, 17 Jul 2023 23:35:34 +0000 (16:35 -0700)]
[flang] Compare component types In AreSameComponent()
The subroutine AreSameComponent() of the predicate AreSameDerivedType()
had a TODO about checking component types that needed completion in order
to properly detect that two specific procedures of a generic are
distinguishable in the llvm-test-suite/Fortran/gfortran/regression
test import7.f90.
Differential Revision: https://reviews.llvm.org/D155962
Rahman Lavaee [Mon, 17 Jul 2023 14:23:42 +0000 (07:23 -0700)]
[llvm-objdump] Use BBEntry::BBID to represent basic block numbers.
Reviewed By: aidengrossman, mtrofin, JestrTulip
Differential Revision: https://reviews.llvm.org/D155464
Joseph Huber [Fri, 21 Jul 2023 16:18:13 +0000 (11:18 -0500)]
[libc] Disable 'DecodeInOtherBases` test on GPU targets
This test is excessively slow on GPU targets, taking anywhere beween 5
and 60 seconds to complete each time it's run. See
https://lab.llvm.org/buildbot/#/builders/55/builds/52203/steps/12/logs/stdio
for an example on the NVPTX buildbot. Simply disable testing this on the
GPU for now.
Reviewed By: michaelrj
Differential Revision: https://reviews.llvm.org/D155979
Simon Pilgrim [Fri, 21 Jul 2023 18:10:06 +0000 (19:10 +0100)]
[X86] combineBitcastvxi1 - don't prematurely create PACKSS nodes.
Similar to Issue #63710 - by truncating the v8i16 result with a PACKSS node before type legalization, we fail to make use of various folds that rely on TRUNCATE nodes.
This required tweaks to LowerTruncateVecPackWithSignBits to recognise when the truncation source has been widened and to more closely match combineVectorSignBitsTruncation wrt truncating with PACKSS/PACKUS on AVX512 targets.
One of the last stages before we can finally get rid of combineVectorSignBitsTruncation.
Simon Pilgrim [Fri, 21 Jul 2023 17:16:11 +0000 (18:16 +0100)]
[X86] truncateVectorWithPACK - avoid concat_vectors(extract_subvector(pack()),extract_subvector(pack())) for sub-128 bit vectors
As we start using this after type legalization, we must avoid creating concat_vectors nodes so late.
Stanislav Mekhanoshin [Thu, 20 Jul 2023 20:12:37 +0000 (13:12 -0700)]
[AMDGPU] Remove std::optional from VOPD::ComponentProps. NFC.
This class has to be fast and efficient with a trivial copy
constructor.
Differential Revision: https://reviews.llvm.org/D155881
Fangrui Song [Fri, 21 Jul 2023 17:28:52 +0000 (10:28 -0700)]
[test] Unsupport CodeGenCXX/destructors for LLVM_ENABLE_REVERSE_ITERATION builds
_ZN5test312_GLOBAL__N_11CD2Ev and _ZN5test312_GLOBAL__N_11DD0Ev are
swapped in LLVM_ENABLE_REVERSE_ITERATION builds. Unsupport for now.
Uday Bondhugula [Thu, 20 Jul 2023 16:20:13 +0000 (21:50 +0530)]
NFC. Move remaining affine/memref test cases into respective dialect dirs
Move a bunch of lingering test cases from test/Transforms/ into
test/Dialect/Affine and MemRef.
Differential Revision: https://reviews.llvm.org/D155855
Simon Pilgrim [Fri, 21 Jul 2023 16:48:11 +0000 (17:48 +0100)]
[X86] Add isUpperSubvectorUndef helper to simplify recognition of vectors widened with undef upper subvectors. NFC.
Arthur Eubanks [Fri, 21 Jul 2023 16:47:08 +0000 (09:47 -0700)]
[Sanitizers][Darwin][Test] Mark symbolize_pc test on Darwin/TSan+UBSan as UNSUPPORTED
Followup to https://reviews.llvm.org/rG760c208f6ff9e97a9a11523c00874a1eec4f876b which XFAIL'd them, but they pass in some configurations.
Peter Klausler [Mon, 17 Jul 2023 16:42:47 +0000 (09:42 -0700)]
[flang] Accept an assumed-rank array as operand of ASSOCIATED()
The ASSOCIATED() intrinsic was mistakenly defined in the intrinsic
function table as requiring operands of known rank, which unintentionally
prevented assumed-rank dummy arguments from being tested.
Fixes llvm-test-suite/Fortran/gfortran/regression/pr88932.f90.
Differential Revision: https://reviews.llvm.org/D155498
Sylvestre Ledru [Fri, 21 Jul 2023 16:23:14 +0000 (18:23 +0200)]
clang/Debian: add Debian Trixie now that it is in unstable
Lorenzo Chelini [Fri, 21 Jul 2023 11:28:45 +0000 (13:28 +0200)]
[MLIR][Linalg] Preserve DPS when decomposing Softmax
Preserve destination passing style (DPS) when decomposing
`linalg.Softmax`; instead of creating a new empty, which may materialize
as a new buffer after bufferization, use the result directly.
Reviewed By: qcolombet
Differential Revision: https://reviews.llvm.org/D155942
Alex Bradbury [Fri, 21 Jul 2023 15:37:50 +0000 (16:37 +0100)]
[RISCV][NFC] Add RISCVSubtarget field to RISCVExpandPseudo and RISCVPreRAExpandPseudo
To my eye, it's cleaner to just get hold of STI in runOnMachineFunction
(as we do already for InstrInfo) and then accessing the field as needed
rather than to have repeated lookup code in the member functions or
helpers that need it.
Differential Revision: https://reviews.llvm.org/D155840
Joseph Huber [Fri, 21 Jul 2023 15:34:09 +0000 (10:34 -0500)]
[libc] Treat the locks array as a bitfield
Currently we keep an internal buffer of device memory that is used to
indicate ownership of a port. Since we only use this as a single bit we
can simply turn this into a bitfield. I did this manually rather than
having a separate type as we need very special handling of the masks
used to interact with the locks.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D155511
Fangrui Song [Fri, 21 Jul 2023 15:46:51 +0000 (08:46 -0700)]
[Support] Implement LLVM_ENABLE_REVERSE_ITERATION for StringMap
ProgrammersManual.html says
> StringMap iteration order, however, is not guaranteed to be deterministic, so any uses which require that should instead use a std::map.
This patch makes -DLLVM_REVERSE_ITERATION=on (currently
-DLLVM_ENABLE_REVERSE_ITERATION=on works as well) shuffle StringMap
iteration order (actually flipping the hash so that elements not in the
same bucket are reversed) to catch violations, similar to D35043 for
DenseMap. This should help change the hash function (e.g., D142862,
D155781).
With a lot of fixes, there are still some violations. This patch
implements the "reverse_iteration" lit feature to skip such tests.
Eventually we should remove this feature.
`ninja check-{llvm,clang,clang-tools}` are clean with
`#define LLVM_ENABLE_REVERSE_ITERATION 1`.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D155789
Mikhail Goncharov [Fri, 21 Jul 2023 15:38:01 +0000 (17:38 +0200)]
[bazel] update config.h.cmake
Fangrui Song [Fri, 21 Jul 2023 15:37:58 +0000 (08:37 -0700)]
[RISCV] Allow delayed decision for ADD/SUB relocations
For a label difference `A-B` in assembly, if A and B are separated by a
linker-relaxable instruction, we should emit a pair of ADD/SUB
relocations (e.g. R_RISCV_ADD32/R_RISCV_SUB32,
R_RISCV_ADD64/R_RISCV_SUB64).
However, the decision is made upfront at parsing time with inadequate
heuristics (`requiresFixup`). As a result, LLVM integrated assembler
incorrectly suppresses R_RISCV_ADD32/R_RISCV_SUB32 for the following
code:
```
// Simplified from a workaround https://android-review.googlesource.com/c/platform/art/+/2619609
// Both end and begin are not defined yet. We decide ADD/SUB relocations upfront and don't know they will be needed.
.4byte end-begin
begin:
call foo
end:
```
To fix the bug, make two primary changes:
* Delete `requiresFixups` and the overridden emitValueImpl (from D103539).
This deletion requires accurate evaluateAsAbolute (D153097).
* In MCAssembler::evaluateFixup, call handleAddSubRelocations to emit
ADD/SUB relocations.
However, there is a remaining issue in
MCExpr.cpp:AttemptToFoldSymbolOffsetDifference. With MCAsmLayout, we may
incorrectly fold A-B even when A and B are separated by a
linker-relaxable instruction. This deficiency is acknowledged (see
D153097), but was previously bypassed by eagerly emitting ADD/SUB using
`requiresFixups`. To address this, we partially reintroduce `canFold` (from
D61584, removed by D103539).
Some expressions (e.g. .size and .fill) need to take the `MCAsmLayout`
code path in AttemptToFoldSymbolOffsetDifference, avoiding relocations
(weird, but matching GNU assembler and needed to match user
expectation). Switch to evaluateKnownAbsolute to leverage the `InSet`
condition.
As a bonus, this change allows for the removal of some relocations for
the FDE `address_range` field in the .eh_frame section.
riscv64-64b-pcrel.s contains the main test.
Add a linker relaxable instruction to dwarf-riscv-relocs.ll to test what
it intends to test.
Merge fixups-relax-diff.ll into fixups-diff.ll.
Reviewed By: kito-cheng
Differential Revision: https://reviews.llvm.org/D155357