Julian Lettner [Mon, 24 Jan 2022 06:01:48 +0000 (22:01 -0800)]
[TSan] Mark test unsupported on Darwin
David Blaikie [Mon, 24 Jan 2022 05:24:05 +0000 (21:24 -0800)]
Rough guess at fixing lldb tests to handle Clang defaulting to DWARFv5
Abinav Puthan Purayil [Mon, 3 Jan 2022 10:15:52 +0000 (15:45 +0530)]
[GlobalISel] Fold or of shifts with constant amount to funnel shift.
This change folds (or (shl x, C0), (lshr y, C1)) to funnel shift iff C0
and C1 are constants where C0 + C1 is the bit-width of the shift
instructions.
Differential Revision: https://reviews.llvm.org/D116529
David Blaikie [Mon, 24 Jan 2022 05:10:16 +0000 (21:10 -0800)]
Add missing function implementation from DWARF default change
Fix for
d3b26dea16108c427b19b5480c9edc76edf8f5b4
David Blaikie [Mon, 24 Jan 2022 04:45:25 +0000 (20:45 -0800)]
Clang: Change the default DWARF version to 5
(except on platforms that already opt in to specific versions - SCE,
Android, and Darwin using DWARFv4 explicitly, for instance)
Kazu Hirata [Mon, 24 Jan 2022 04:32:56 +0000 (20:32 -0800)]
[Analysis] Use default member initialization (NFC)
Identified with modernize-use-default-member-init.
Kazu Hirata [Mon, 24 Jan 2022 04:32:54 +0000 (20:32 -0800)]
[Vectorize] Remove unused variables (NFC)
Kazu Hirata [Mon, 24 Jan 2022 04:32:52 +0000 (20:32 -0800)]
Add modernize-use-default-member-init.UseAssignment to .clang-tidy
Chuanqi Xu [Mon, 24 Jan 2022 03:03:12 +0000 (11:03 +0800)]
[NFC] [Coroutines] Rename tests in coro-align
This is required by ychen. See https://reviews.llvm.org/D117542
Jim Lin [Mon, 24 Jan 2022 02:20:16 +0000 (10:20 +0800)]
[RISCV][NFC] Remove tailing whitespaces in RISCVInstrInfoVSDPatterns.td and RISCVInstrInfoVVLPatterns.td
Chuanqi Xu [Mon, 24 Jan 2022 02:22:33 +0000 (10:22 +0800)]
[C++20] [Module] fix bug 47716 and implement [module.interface]/p6
This fixes bug 47716.
According to [module.interface]p2, it is meaningless to export an entity
which is not in namespace scope.
The reason why the compiler crashes is that the compiler missed
ExportDecl when the compiler traverse the subclass of DeclContext. So
here is the crash.
Also, the patch implements [module.interface]p6 in
Sema::CheckRedeclaration* functions.
Reviewed By: aaron.ballman, urnathan
Differential Revision: https://reviews.llvm.org/D112903
David Blaikie [Mon, 24 Jan 2022 00:07:43 +0000 (16:07 -0800)]
DebugInfo: Include template parameters for simplified template decls in type units
LLVM DebugInfo CodeGen synthesizes type declarations in type units when
referencing types that are not in type units. When those synthesized
types are templates and simplified template names (or mangled simplified
template names) are in use, the template arguments must be attached to
those declarations.
A deeper fix (with a CU or DICompositeType flag) that would also support
other uses of clang's -debug-forward-template-args (such as Sony's
platform) could/should be implemented to fix this more broadly.
Simon Pilgrim [Sun, 23 Jan 2022 22:48:26 +0000 (22:48 +0000)]
[X86] Add some basic tests for PR46809
David Blaikie [Sat, 22 Jan 2022 04:20:39 +0000 (20:20 -0800)]
DebugInfo: Don't put types in type units if they reference internal linkage types
Doing this causes a declaration of the internal linkage (anonymous
namespace) type to be emitted in the type unit, which would then be
ambiguous as to which internal linkage definition it refers to (since
the name is only valid internally).
It's possible these internal linkage types could be resolved relative to
the unit the TU is referred to from - but that doesn't seem ideal, and
there's no reason to put the type in a type unit since it can only be
defined in one CU anyway (since otherwise it'd be an ODR violation) & so
avoiding the type unit should be a smaller DWARF encoding anyway.
This also addresses an issue with Simplified Template Names where the
template parameter could not be rebuilt from the declaration emitted
into the TU (specifically for an enum non-type template parameter, where
looking up the enumerators is necessary to rebuild the full template
name)
Kazu Hirata [Sun, 23 Jan 2022 22:00:03 +0000 (14:00 -0800)]
[Analysis] Remove a redundant const from a return type (NFC)
Identified with readability-const-return-type.
Kazu Hirata [Sun, 23 Jan 2022 22:00:01 +0000 (14:00 -0800)]
[Sema] Fix a bugprone argument comment (NFC)
Identified with bugprone-argument-comment.
Kazu Hirata [Sun, 23 Jan 2022 21:59:59 +0000 (13:59 -0800)]
[mlir] Ensure a newline at the end of a file (NFC)
Craig Topper [Sun, 23 Jan 2022 21:37:33 +0000 (13:37 -0800)]
[RISCV] Add tests that do a bitreverse before or after a bswap. NFC
We don't optimize this as well as we could. Bitreverse is always
expanded to bswap and a shift/and/or sequence to swap bits within a
byte. The newly created bswap will either becomes a shift/and/or
sequence or rev8 instruction. We don't always realize the bswap is
redundant with another bswap before or after the bitreverse.
Found while thinking about the brev8 instruction from the
Cryptography extension. It's equivalent to bswap(bitreverse(x)) or
bitreverse(bswap(x)).
Craig Topper [Sun, 23 Jan 2022 17:42:18 +0000 (09:42 -0800)]
[RISCV] Add bitreverse tests to bswap-ctlz-cttz-ctpop.ll. Add Zbb command lines. NFC
Rename to include bitreverse. Add additional tests and Zbb command lines.
There's some overlapping tests with rv32zbb.ll and rv64zbb.ll. Maybe
I'll clean that up in a future patch.
Simon Pilgrim [Sun, 23 Jan 2022 21:34:55 +0000 (21:34 +0000)]
[X86] LowerFunnelShift - always lower vXi8 fshl by constant amounts as unpack(y,x) << zext(z)
This can always be lowered as PMULLW+PSRLWI+PACKUSWB
Kazu Hirata [Sun, 23 Jan 2022 21:28:06 +0000 (13:28 -0800)]
[clang] Remove unused forward declarations (NFC)
Kazu Hirata [Sun, 23 Jan 2022 21:28:04 +0000 (13:28 -0800)]
[clang] Forward-declare DynTypedNode (NFC)
This patch adds a forward declaraiton of DynTypedNode.
DumpAST.h is relying on the forward declaration of DynTypedNode in
ASTContext.h, which is undesirable.
Kazu Hirata [Sun, 23 Jan 2022 21:28:02 +0000 (13:28 -0800)]
[clang] Move the definition of ASTDiff (NFC)
This patch moves the definition of ASTDiff later within the header
file.
Without this patch, the header depends on the forward decalrations of
SyntaxTree and ComparisonOptions from another header file, which is
not desirable. Since SyntaxTree and ComparisonOptions are defined in
ASTDiff.h, we can move the definition of ASTDiff later and stop
relying on the forward declarations from another header file.
Simon Pilgrim [Sun, 23 Jan 2022 21:13:58 +0000 (21:13 +0000)]
[X86] LowerFunnelShift - use supportedVectorShiftWithBaseAmnt to check for supported scalar shifts
Allows us to reuse the ISD shift opcode instead of a mixture of ISD/X86ISD variants
Craig Topper [Sun, 23 Jan 2022 17:42:18 +0000 (09:42 -0800)]
[RISCV] Adjust the header comment in RISCVInstrInfoZb.td to better integrate Zbk* extensions.
The Zbk* extensions have some overlap with Zb so have been placed in this file.
Reviewed By: VincentWu
Differential Revision: https://reviews.llvm.org/D117958
Groverkss [Sun, 23 Jan 2022 19:38:54 +0000 (01:08 +0530)]
[MLIR][Presburger] Clean PresburgerSet identifier interface to match IntegerPolyhedron's interface
This patch changes names of identifiers and their corresponding getters in
PresburgerSet to match those of IntegerPolyhedron.
Reviewed By: arjunp
Differential Revision: https://reviews.llvm.org/D117998
Kazu Hirata [Sun, 23 Jan 2022 19:07:16 +0000 (11:07 -0800)]
[Object] Remove a redundant return statement (NFC)
Identified with readability-redundant-control-flow.
Kazu Hirata [Sun, 23 Jan 2022 19:07:14 +0000 (11:07 -0800)]
[Commands] Remove redundant member initialization (NFC)
Identified with readability-redundant-member-init.
Kazu Hirata [Sun, 23 Jan 2022 19:07:12 +0000 (11:07 -0800)]
[llvm] Fix header guards (NFC)
Identified with llvm-header-guard.
Nuno Lopes [Sun, 23 Jan 2022 19:06:21 +0000 (19:06 +0000)]
[NewGVN][NFC] precommit tests for PR53277
Fangrui Song [Sun, 23 Jan 2022 18:35:44 +0000 (10:35 -0800)]
[Support] Simplify parallelForEach{,N}
* Merge parallel_for_each into parallelForEach (this removes 1 `Fn(...)` call)
* Change parallelForEach to use parallelForEachN
* Move parallelForEachN into Parallel.cpp
My x86-64 `lld` executable is 100KiB smaller.
No noticeable difference in performance.
Reviewed By: lattner
Differential Revision: https://reviews.llvm.org/D117510
Casey Carter [Thu, 30 Dec 2021 00:02:00 +0000 (16:02 -0800)]
[libcxx][test] Make MSVC `<charconv>` test compile when testing MSVC
<meme>How many layers of irony are you on?</meme>
Differential Revision: https://reviews.llvm.org/D117967
Simon Pilgrim [Sun, 23 Jan 2022 18:03:12 +0000 (18:03 +0000)]
[InstCombine] Add test coverage for PR48683
D108992 added self-multiply handling to KnownBits::mul but we don't use it yet..
Craig Topper [Sun, 23 Jan 2022 07:05:19 +0000 (23:05 -0800)]
[RISCV] Merge some rvv intrinsic test cases that only differ by XLen type.
Instead of having a test for i32 XLen and i64 XLen, use sed to
replace iXLen with i32/i64 before running llc.
This change covers all of the floating point tests.
Simon Pilgrim [Sun, 23 Jan 2022 16:36:18 +0000 (16:36 +0000)]
[DAG] Fold (X & Y) != 0 --> zextOrTrunc(X & Y) iff everything but LSB is known zero (PR51312)
Fixes parity codegen issue where we know all but the lowest bit is zero, we can replace the ICMPNE with 0 comparison with a ext/trunc
Differential Revision: https://reviews.llvm.org/D117983
LLVM GN Syncbot [Sun, 23 Jan 2022 16:30:34 +0000 (16:30 +0000)]
[gn build] Port
d2e8fb331835
Simon Pilgrim [Sun, 23 Jan 2022 16:28:38 +0000 (16:28 +0000)]
Revert rG7c66aaddb128dc0f342830c1efaeb7a278bfc48c "[DAG] Fold (X & Y) != 0 --> zextOrTrunc(X & Y) iff everything but LSB is known zero (PR51312)"
Noticed a typo in the getBooleanContents call just after I pressed commit :(
Richard [Sun, 2 Jan 2022 05:47:22 +0000 (22:47 -0700)]
[clang-tidy] Add readability-duplicate-include check
Looks for duplicate includes and removes them.
Every time an include directive is processed, check a vector of filenames
to see if the included file has already been included. If so, it issues
a warning and a replacement to remove the entire line containing the
duplicated include directive.
When a macro is defined or undefined, the vector of filenames is cleared.
This enables including the same file multiple times, but getting
different expansions based on the set of active macros at the time of
inclusion. For example:
#undef NDEBUG
#include "assertion.h"
// ...code with assertions enabled
#define NDEBUG
#include "assertion.h"
// ...code with assertions disabled
Since macros are redefined between the inclusion of assertion.h,
they are not flagged as redundant.
Differential Revision: https://reviews.llvm.org/D7982
Sanjay Patel [Sun, 23 Jan 2022 16:11:26 +0000 (11:11 -0500)]
[IR] document and update ctlz/cttz intrinsics to optionally return poison rather than undef
The behavior in Analysis (knownbits) implements poison semantics already,
and we expect the transforms (for example, in instcombine) derived from
those semantics, so this patch changes the LangRef and remaining code to
be consistent. This is one more step in removing "undef" from LLVM.
Without this, I think https://github.com/llvm/llvm-project/issues/53330
has a legitimate complaint because that report wants to allow subsequent
code to mask off bits, and that is allowed with undef values. The clang
builtins are not actually documented anywhere AFAICT, but we might want
to add that to remove more uncertainty.
Differential Revision: https://reviews.llvm.org/D117912
Simon Pilgrim [Sun, 23 Jan 2022 16:20:34 +0000 (16:20 +0000)]
[DAG] Fold (X & Y) != 0 --> zextOrTrunc(X & Y) iff everything but LSB is known zero (PR51312)
Fixes parity codegen issue where we know all but the lowest bit is zero, we can replace the ICMPNE with 0 comparison with a ext/trunc
Differential Revision: https://reviews.llvm.org/D117983
Ayke van Laethem [Sun, 16 Jan 2022 07:42:48 +0000 (08:42 +0100)]
[AVR] Make use of the constant value 0 in R1
The register R1 is defined to have the constant value 0 in the avr-gcc
calling convention (which we follow). Unfortunately, we don't really
make use of it. This patch replaces `LDI 0` instructions with a copy
from R1.
This reduces code size: my AVR build of compiler-rt goes from 50660 to
50240 bytes of code size, which is a 0.8% reduction. Presumably it will
also improve execution speed, although I didn't measure this.
Differential Revision: https://reviews.llvm.org/D117425
Ayke van Laethem [Thu, 20 Jan 2022 20:50:26 +0000 (21:50 +0100)]
[AVR] Remove regalloc workaround for LDDWRdPtrQ
Background: https://github.com/avr-rust/rust-legacy-fork/issues/126
In short, this workaround was introduced to fix a "ran out of registers
during regalloc" issue. The root cause has since been fixed in
https://reviews.llvm.org/D54218 so this workaround can be removed.
There is one test that changes a little bit, removing a single
instruction. I also compiled compiler-rt before and after this patch but
didn't see a difference. So presumably the impact is very low. Still,
it's nice to be able to remove such a workaround.
Differential Revision: https://reviews.llvm.org/D117831
Carlos Galvez [Fri, 21 Jan 2022 07:59:27 +0000 (07:59 +0000)]
[clang-tidy] Remove gsl::at suggestion from cppcoreguidelines-pro-bounds-constant-array-index
Currently the fix hint is hardcoded to gsl::at(). This poses
a problem for people who, for a number of reasons, don't want
or cannot use the GSL library (introducing a new third-party
dependency into a project is not a minor task).
In these situations, the fix hint does more harm than good
as it creates confusion as to what the fix should be. People
can even misinterpret the fix "gsl::at" as e.g. "std::array::at",
which can lead to even more trouble (e.g. when having guidelines
that disallow exceptions).
Furthermore, this is not a requirement from the C++ Core Guidelines.
simply that array indexing needs to be safe. Each project should
be able to decide upon a strategy for safe indexing.
The fix-it is kept for people who want to use the GSL library.
Differential Revision: https://reviews.llvm.org/D117857
Simon Pilgrim [Sun, 23 Jan 2022 15:34:41 +0000 (15:34 +0000)]
[X86] Add vector signbit parity checks for non-popcnt targets
Noticed while looking at D117983 - we miss some parity patterns with/without popcnt
Arthur O'Dwyer [Sat, 22 Jan 2022 20:24:53 +0000 (15:24 -0500)]
[libc++] Fix LWG3437 "__cpp_lib_polymorphic_allocator is in the wrong header"
https://cplusplus.github.io/LWG/issue3437
Differential Revision: https://reviews.llvm.org/D117963
Arthur O'Dwyer [Sat, 22 Jan 2022 18:06:05 +0000 (13:06 -0500)]
[libc++] Mark LWG3541 as "Complete". NFC.
Differential Revision: https://reviews.llvm.org/D117956
Simon Pilgrim [Sun, 23 Jan 2022 15:14:10 +0000 (15:14 +0000)]
Fix "not all control paths return a value" warning. NFC.
Simon Pilgrim [Sun, 23 Jan 2022 15:12:44 +0000 (15:12 +0000)]
[lldb] TerminalState::Save - fix unused variable warning
Non-POSIX target builds don't use the file descriptor
Simon Pilgrim [Sun, 23 Jan 2022 15:11:08 +0000 (15:11 +0000)]
[lldb] PdbAstBuilder - use cast<> instead of dyn_cast<> to avoid dereference of nullptr
The pointers are dereferenced immediately, so assert the cast is correct instead of returning nullptr
Simon Pilgrim [Sun, 23 Jan 2022 15:10:33 +0000 (15:10 +0000)]
[lldb] CxxModuleHandler - use cast<> instead of dyn_cast<> to avoid dereference of nullptr
The pointer is dereferenced immediately, so assert the cast is correct instead of returning nullptr
Simon Pilgrim [Sun, 23 Jan 2022 13:24:36 +0000 (13:24 +0000)]
[clangd] Use castAs<> instead of getAs<> to avoid dereference of nullptr
The pointer is dereferenced immediately, so assert the cast is correct instead of returning nullptr
Simon Pilgrim [Sun, 23 Jan 2022 12:57:12 +0000 (12:57 +0000)]
[clang-tidy] Use cast<>/castAs<> instead of dyn_cast<>/getAs<> to avoid dereference of nullptr
The pointer is dereferenced immediately, so assert the cast is correct instead of returning nullptr
Simon Pilgrim [Sun, 23 Jan 2022 12:50:12 +0000 (12:50 +0000)]
[llvm-objdump] Use cast<> instead of dyn_cast<> to avoid dereference of nullptr
The pointer is dereferenced immediately, so assert the cast is correct instead of returning nullptr
Simon Pilgrim [Sun, 23 Jan 2022 12:47:52 +0000 (12:47 +0000)]
[CodeGenPrepare] Use dyn_cast result to check for null pointers
Simplifies logic and helps the static analyzer correctly check for nullptr dereferences
Simon Pilgrim [Sun, 23 Jan 2022 12:45:12 +0000 (12:45 +0000)]
[llvm-objdump] Use cast<> instead of dyn_cast<> to avoid dereference of nullptr
The pointers are always dereferenced immediately, so assert the cast is correct instead of returning nullptr
Simon Pilgrim [Sun, 23 Jan 2022 11:34:25 +0000 (11:34 +0000)]
[X86] Regenerate avx512-mask-op.ll
Noticed on D86578 - several of the test cases were missing checks as they didn't start on a newline so the update script couldn't see them
Alex Brachet [Sun, 23 Jan 2022 09:44:09 +0000 (09:44 +0000)]
[llvm-objcopy][MachO] Implement --update-section
Implements `--update-section` which is currently supported for ELF for Mach-O as well
Reviewed By: alexander-shaposhnikov
Differential Revision: https://reviews.llvm.org/D117281
Craig Topper [Sat, 22 Jan 2022 21:41:30 +0000 (13:41 -0800)]
[RISCV] Use FP ABI for some RVV intrinsic tests. NFC
Removes moves from GPR to FPR and improves f64 tests on RV32.
Differential Revision: https://reviews.llvm.org/D117969
Craig Topper [Sat, 22 Jan 2022 20:10:57 +0000 (12:10 -0800)]
[RISCV] Merge some rvv intrinsic test cases that only differ by XLen type.
Instead of having a test for i32 XLen and i64 XLen, use sed to
replace iXLen with i32/i64 before running llc.
This change updates tests for intrinsics that operate exclusively
on mask values. It removes over 4000 lines worth of test content.
More merging will come in future changes.
Differential Revision: https://reviews.llvm.org/D117968
eopXD [Fri, 21 Jan 2022 08:39:23 +0000 (00:39 -0800)]
[RISCV] Remove experimental prefix from rvv-related extensions.
Extensions affected: +v, +zve*, +zvl*
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D117860
Joe Loser [Wed, 19 Jan 2022 20:16:15 +0000 (15:16 -0500)]
[libc++] Implement LWG3549: view_interface need not inherit from view_base
Implement LWG3549 by making `view_interface` not inherit from `view_base`. Types
are still views if they have a public and unambiguous derivation from
`view_interface`, so adjust the `enable_view` machinery as such to account for
that.
Differential Revision: https://reviews.llvm.org/D117714
Phoebe Wang [Sun, 23 Jan 2022 01:14:58 +0000 (09:14 +0800)]
[X86][MS] Change the alignment of f80 to 16 bytes on Windows 32bits to match with ICC
MSVC currently doesn't support 80 bits long double. ICC supports it when
the option `/Qlong-double` is specified. Changing the alignment of f80
to 16 bytes so that we can be compatible with ICC's option.
Reviewed By: rnk, craig.topper
Differential Revision: https://reviews.llvm.org/D115942
Dave [Sun, 23 Jan 2022 01:30:27 +0000 (17:30 -0800)]
[docs] [clang] Small documentation change for compilation databases
We have an page dedicated to compliation databases including
various ways to generate them, but we don't mention that clang
has a built in method to do this. This addresses that.
Reviewed By: joerg
Differential Revision: https://reviews.llvm.org/D116882
Malhar Jajoo [Sat, 22 Jan 2022 22:05:38 +0000 (22:05 +0000)]
[openmp] Allow x87 fp functions only in Openmp runtime for x86.
This patch allows Openmp runtime atomic functions operating on x87 high-precision
to be present only in Openmp runtime for x86 architectures
The functions affected are:
__kmpc_atomic_10
__kmpc_atomic_20
__kmpc_atomic_cmplx10_add
__kmpc_atomic_cmplx10_div
__kmpc_atomic_cmplx10_mul
__kmpc_atomic_cmplx10_sub
__kmpc_atomic_float10_add
__kmpc_atomic_float10_div
__kmpc_atomic_float10_mul
__kmpc_atomic_float10_sub
__kmpc_atomic_float10_add_fp
__kmpc_atomic_float10_div_fp
__kmpc_atomic_float10_mul_fp
__kmpc_atomic_float10_sub_fp
__kmpc_atomic_float10_max
__kmpc_atomic_float10_min
Differential Revision: https://reviews.llvm.org/D117473
John Ericson [Sun, 16 Jan 2022 05:52:22 +0000 (05:52 +0000)]
[clang-tools-extra][cmake] Use `GNUInstallDirs` to support custom installation dirs.
This is the original patch in my GNUInstallDirs series, now last to merge as the final piece!
It arose as a new draft of D28234. I initially did the unorthodox thing of pushing to that when I wasn't the original author, but since I ended up
- Using `GNUInstallDirs`, rather than mimicking it, as the original author was hesitant to do but others requested.
- Converting all the packages, not just LLVM, effecting many more projects than LLVM itself.
I figured it was time to make a new revision.
I have used this patch series (and many back-ports) as the basis of https://github.com/NixOS/nixpkgs/pull/111487 for my distro (NixOS), which was merged last spring (2021). It looked like people were generally on board in D28234, but I make note of this here in case extra motivation is useful.
---
As pointed out in the original issue, a central tension is that LLVM already has some partial support for these sorts of things. Variables like `COMPILER_RT_INSTALL_PATH` have already been dealt with. Variables like `LLVM_LIBDIR_SUFFIX` however, will require further work, so that we may use `CMAKE_INSTALL_LIBDIR`.
These remaining items will be addressed in further patches. What is here is now rote and so we should get it out of the way before dealing more intricately with the remainder.
Reviewed By: #libunwind, #libc, #libc_abi, compnerd
Differential Revision: https://reviews.llvm.org/D99484
Sanjay Patel [Sat, 22 Jan 2022 17:36:12 +0000 (12:36 -0500)]
[InstCombine] try to fold binop with phi operands
This is an alternate version of D115914 that handles/tests all binary opcodes.
I suspect that we don't see these patterns too often because -simplifycfg
would convert the minimal cases into selects rather than leave them in phi form
(note: instcombine has logic holes for combining the select patterns too though,
so that's another potential patch).
We only create a new binop in a predecessor that unconditionally branches to
the final block.
https://alive2.llvm.org/ce/z/C57M2F
https://alive2.llvm.org/ce/z/WHwAoU (not safe to speculate an sdiv for example)
https://alive2.llvm.org/ce/z/rdVUvW (but it is ok on this path)
Differential Revision: https://reviews.llvm.org/D117110
Craig Topper [Sat, 22 Jan 2022 19:55:13 +0000 (11:55 -0800)]
[RISCV] Don't Custom legalize f16/f32/f64 bitcasts if those types aren't Legal.
John Ericson [Sat, 22 Jan 2022 06:33:49 +0000 (01:33 -0500)]
[openmp][cmake] Use `GNUInstallDirs` to support custom installation dirs
I am breaking apart D99484 so the cause of build failures is easier to
understand.
Differential Revision: https://reviews.llvm.org/D117945
Peter Klausler [Thu, 20 Jan 2022 00:25:41 +0000 (16:25 -0800)]
[flang] Support DECIMAL='COMMA' mode in namelist I/O
DECIMAL='COMMA' mode affects item separators, real editing, and
complex editing.
Differential Revision: https://reviews.llvm.org/D117906
Peter Klausler [Fri, 14 Jan 2022 21:34:10 +0000 (13:34 -0800)]
[flang] Don't drop format string for external child I/O
In user-defined derived type I/O to an external unit, don't
omit the format string from the constructor of ChildFormattedIoStatement.
And include any user IOMSG text in the crash message of the
parent, if it doesn't catch errors.
Differential Revision: https://reviews.llvm.org/D117903
Arthur O'Dwyer [Sat, 22 Jan 2022 16:53:00 +0000 (11:53 -0500)]
[libc++] [test] {cpo,niebloid}.compile.pass.cpp: Also test their constness.
This will detect if someone writes `inline auto cpo =` instead of
`inline constexpr auto cpo =`. I don't know how that'd be possible,
but it's easy to test, so let's test it.
Florian Hahn [Sat, 22 Jan 2022 15:34:10 +0000 (15:34 +0000)]
[LV] Always create VPWidenCanonicalIVRecipe, optimize away later.
This patch updates createBlockInMask to always generate
VPWidenCanonicalIVRecipe and adds a transform to optimize it away later,
if it is not needed.
This is a step towards breaking up VPWidenIntOrFpInductionRecipe and
explicitly distinguishing between vector phis and scalarizing.
Split off from D116123.
Reviewed By: Ayal
Differential Revision: https://reviews.llvm.org/D117140
Qiu Chaofan [Sat, 22 Jan 2022 15:29:34 +0000 (23:29 +0800)]
[PowerPC] Support parsing GNU attributes in MC
This patch is the first step to enable support of GNU attribute in LLVM
PowerPC, enabling it for PowerPC targets, otherwise llvm-mc raises error
when seeing the attribute section.
Reviewed By: jsji
Differential Revision: https://reviews.llvm.org/D115854
Qiu Chaofan [Sat, 22 Jan 2022 15:20:14 +0000 (23:20 +0800)]
[PowerPC] Change CTR clobber estimation for 128-bit floating types
Reviewed By: shchenz
Differential Revision: https://reviews.llvm.org/D117459
David Green [Sat, 22 Jan 2022 13:20:36 +0000 (13:20 +0000)]
[DAG] Convert truncstore(extend(x)) back to store(x)
Pulled out of D106237, this folds truncstore(extend(x)) back to store(x)
if the original store was legal. This can come up due to the order we
fold nodes. A fold from X86 needs to be adjusted to prevent infinite
loops, to have it pick the operand of a trunc more directly.
Differential Revision: https://reviews.llvm.org/D117901
luxufan [Sat, 22 Jan 2022 08:07:17 +0000 (16:07 +0800)]
reapply
de872382951 "[JITLink] Add anonymous symbols in LinkGraph..."
with fixes
This reapply `
de872382951572b70dfaefe8d77eb98d15586115`, which was
reverted in `
fdb6578514dd3799ad23c8bbb7699577c0fb414d`
Add `# REQUIRES: asserts` in test file `anonymous_symbol.s` to disable
this test for non-debug build
Micah Weston [Sat, 22 Jan 2022 12:39:22 +0000 (12:39 +0000)]
[AArch64] Optimize add/sub with immediate through MIPeepholeOpt
Fixes the build issue with D111034, whose goal was to optimize
add/sub with long immediates.
Optimize ([add|sub] r, imm) -> ([ADD|SUB] ([ADD|SUB] r, #imm0, lsl #12), #imm1),
if imm == (imm0<<12)+imm1. and both imm0 and imm1 are non-zero 12-bit unsigned
integers.
Optimize ([add|sub] r, imm) -> ([SUB|ADD] ([SUB|ADD] r, #imm0, lsl #12), #imm1),
if imm == -(imm0<<12)-imm1, and both imm0 and imm1 are non-zero 12-bit unsigned
integers.
The change which fixed the build issue in D111034 was the use of new virtual
registers so that SSA form is maintained until deleting MI.
Differential Revision: https://reviews.llvm.org/D117429
Alexander Belyaev [Sat, 22 Jan 2022 10:42:47 +0000 (11:42 +0100)]
[mlir] Add SingleBlockImplicitTerminator<"tensor::YieldOp"> to PadOp.
Mark de Wever [Fri, 21 Jan 2022 19:08:57 +0000 (20:08 +0100)]
[libc++] Use addressof in unordered_set.
This addresses the usage of `operator&` in `<unordered_set>`.
(Note there are still more headers with the same issue.)
Reviewed By: #libc, philnik, Quuxplusone
Differential Revision: https://reviews.llvm.org/D117917
fourdim [Fri, 21 Jan 2022 19:34:13 +0000 (03:34 +0800)]
[JITLink][RISCV] Support R_RISCV_SET* and R_RISCV_32_PCREL relocations
This patch supports R_RISCV_SET* and R_RISCV_32_PCREL relocations in JITLink.
Reviewed By: StephenFan
Differential Revision: https://reviews.llvm.org/D117082
luxufan [Sat, 22 Jan 2022 09:26:54 +0000 (17:26 +0800)]
Revert "[JITLink] Add anonymous symbols in LinkGraph for unnamed temporary symbols"
This reverts commit
de872382951572b70dfaefe8d77eb98d15586115.
Buildbot check error
luxufan [Sat, 22 Jan 2022 08:07:17 +0000 (16:07 +0800)]
[JITLink] Add anonymous symbols in LinkGraph for unnamed temporary symbols
In RISCV, temporary symbols will be used to generate dwarf, eh_frame sections..., and will be placed in object code's symbol table. However, LLVM does not use names on these temporary symbols. This patch add anonymous symbols in LinkGraph for these temporary symbols.
Reviewed By: lhames
Differential Revision: https://reviews.llvm.org/D116475
Prashant Kumar [Fri, 21 Jan 2022 18:23:22 +0000 (23:53 +0530)]
[MLIR] Fix negative gcd in `normalizeDivisionByGCD` function.
When the coefficients of dividend are negative, the gcd may be negative
which will change the sign of dividend and overflow denominator.
Reviewed By: Groverkss
Differential Revision: https://reviews.llvm.org/D117911
Wei Wang [Tue, 18 Jan 2022 22:08:48 +0000 (14:08 -0800)]
[time-trace] Add optimizer and codegen regions to NPM
Optimizer and codegen regions were only added to legacy PM. Add
them to NPM as well.
Differential Revision: https://reviews.llvm.org/D117605
Peter Klausler [Tue, 18 Jan 2022 18:40:10 +0000 (10:40 -0800)]
[flang] Allow INQUIRE() on a child unit in user-defined I/O procedure
A procedure that implements a user-defined derived type I/O operation
is allowed to perform an INQUIRE statement on its unit.
Differential Revision: https://reviews.llvm.org/D117905https://reviews.llvm.org/D117905
Alex Fan [Fri, 21 Jan 2022 11:51:09 +0000 (19:51 +0800)]
[RISCV][RFC] add MC support for zbkc subextension
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D117874
Dave Lee [Fri, 14 Jan 2022 00:02:45 +0000 (16:02 -0800)]
[lldb] Allow aliases to aliases of raw input commands
Allow users to create aliases for aliases to raw input commands. That probably
sounds convoluted, so here's an example:
```
command alias some-setup env SOMEVAR=SOMEVALUE
```
This an alias based on `env`, which itself is an alias for `_regex-env`.
`_regex-env` is a `command regex` command, which takes raw input.
The above `some-setup` alias fails with:
```
error: Unable to create requested alias.
```
This change allows such aliases to be created. lldb already supports aliases to
aliases for parsed commands.
Differential Revision: https://reviews.llvm.org/D117259
Peter Klausler [Tue, 18 Jan 2022 23:17:15 +0000 (15:17 -0800)]
[flang] Fix crash from USE-associated defined I/O subprograms
User-defined derived type I/O implementation subroutines and
generic interfaces may be USE-associated, but the code that builds
the type description table wasn't allowing for that possibility.
Add a call to GetUltimate() to cope.
Differential Revision: https://reviews.llvm.org/D117902
Julian Lettner [Sat, 22 Jan 2022 00:58:06 +0000 (16:58 -0800)]
[TSan] Omit vfork interceptor iOS simulator runtime
`_vfork` moved from libsystem_kernel.dylib to libsystem_c.dylib as part
of the below changes. The iOS simulator does not actually have
libsystem_kernel.dylib of its own, it only has the host Mac's. The
umbrella-nature of Libsystem makes this movement transparent to
everyone; except the simulator! So when we "back deploy", i.e., use the
current version of TSan with an older simulator runtime then this symbol
is now missing, when we run on the latest OS (but an older simulator
runtime).
Note we use `SANITIZER_IOS` because usage of vfork is forbidden on iOS
and the API is completely unavailable on watchOS and tvOS, even if this
problem is specific to the iOS simulator.
Caused by:
rdar://
74818691 (Shim vfork() to fork syscall on iOS)
rdar://
76762076 (Shim vfork() to fork syscall on macOS)
Radar-Id: rdar://8634734
Peter Klausler [Wed, 19 Jan 2022 17:16:07 +0000 (09:16 -0800)]
[flang] Fix repeated "DT" editing
User-defined derived type editing in formatted I/O wasn't
working with repeat counts; e.g., "2DT(10)". The solution required
some code to be moved from GetNextDataEdit() to CueUpNextDataEdit() so
that a stack entry for a nonparenthesized repeated data edit
descriptor would work correctly -- all other data edit descriptors
are capable of dealing with repetition in their callees, so the bug
hadn't been exposed before.
Debugging this problem led to some improvements in error messages
for bad format strings, and those changes have been retained; also,
a dead member function was discovered and expunged.
Differential Revision: https://reviews.llvm.org/D117904
Fangrui Song [Sat, 22 Jan 2022 01:00:03 +0000 (17:00 -0800)]
[XRay][test] Clean up llc RUN lines
Petr Hosek [Mon, 10 Jan 2022 22:51:37 +0000 (14:51 -0800)]
[CMake] Passthrough OSX CMake options to builtins and runtimes
When using the default target, there's no other way to pass these
into the builtins and runtimes subbuilds.
Differential Revision: https://reviews.llvm.org/D116976
Joe Loser [Fri, 21 Jan 2022 20:01:34 +0000 (15:01 -0500)]
[libc++][test] Add const and reference tests for enable_view. NFC.
As discussed in https://reviews.llvm.org/D117714, there is missing test coverage
for the behavior of `enable_view` when given a const or reference qualified
type. Add such tests showing the current behavior.
Differential Revision: https://reviews.llvm.org/D117918
Chris Bieneman [Fri, 21 Jan 2022 16:47:15 +0000 (10:47 -0600)]
[split-file] Respect input file's line endings
This change adds support for split-file to respect the line ending style
of the input file. This enables split-file to work as expected on
Windows with input files containing CRLF line endings.
The test files added along with this change mirror the existing basic
tests, but are forced to contain CRLF line endings via git attributes.
This will result in the tests always containing CRLF line endings when
checked out regardless of the user's OS.
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D117897
not-jenni [Sat, 22 Jan 2022 00:16:29 +0000 (16:16 -0800)]
[mlir][tosa] Add clamp + clamp as single clamp canonicalization
When 2 clamp ops are in a row, they can be canonicalized into a single clamp
that uses the most constrained range
Reviewed By: rsuderman
Differential Revision: https://reviews.llvm.org/D117934
Mitch Phillips [Sat, 22 Jan 2022 00:22:29 +0000 (16:22 -0800)]
NFC (build fix): Add header for llvm::errs().
Looks like
e9211e039377 unfortunately broke the sanitizer build bots,
because those bots compile the symbolizer with DLLVM_ENABLE_THREADS=Off.
Likely, before the patch, this header was transitively included.
John Ericson [Sun, 16 Jan 2022 06:14:24 +0000 (06:14 +0000)]
[clang][cmake] Use `GNUInstallDirs` to support custom installation dirs
I am breaking apart D99484 so the cause of build failures is easier to
understand.
Differential Revision: https://reviews.llvm.org/D117419
Aart Bik [Fri, 21 Jan 2022 01:27:23 +0000 (17:27 -0800)]
[mlir][sparse] add ability for sparse tensor output
Rationale:
Although file I/O is a bit alien to MLIR itself, we provide two convenient ways
for sparse tensor I/O. The input part was already there (behind the swiss army
knife sparse_tensor.new). Now we have a sparse_tensor.out to write out data. As
before, the ops are kept vague and may change in the future. For now this
allows us to compare TACO vs MLIR very easily.
Reviewed By: bixia
Differential Revision: https://reviews.llvm.org/D117850
Dave Lee [Thu, 20 Jan 2022 22:18:20 +0000 (14:18 -0800)]
[lldb] Fix timer logging inverted quiet condition
The logic of `g_quiet` was inverted in D26243. This corrects the issue.
Without this, running `log timers enable` produces a high volume of incremental
timer output.
Differential Revision: https://reviews.llvm.org/D117837
Zequan Wu [Fri, 21 Jan 2022 23:09:42 +0000 (15:09 -0800)]
[llvm-pdbutil] Fix gaps ouput.