Simon Pilgrim [Fri, 21 Oct 2022 09:39:57 +0000 (10:39 +0100)]
[X86] Fold scalar_to_vector(i64 zext(x)) -> bitcast(vzext_movl(scalar_to_vector(i32 x)))
Extends existing anyextend fold to make use of the implicit zero-extension of the movd instruction
This also helps replace some nasty xmm->gpr->xmm traffic with a shuffle pattern instead
Noticed while looking at D130953
wanglei [Fri, 21 Oct 2022 09:03:37 +0000 (17:03 +0800)]
[LoongArch] Stack realignment support
This patch adds support for stack realignment while adding support for
variable sized objects.
Differential Revision: https://reviews.llvm.org/D136074
Simon Pilgrim [Fri, 21 Oct 2022 09:16:27 +0000 (10:16 +0100)]
[DebugInfo] Fix MSVC warning: "truncation from 'double' to 'float'"
David Green [Fri, 21 Oct 2022 09:10:35 +0000 (10:10 +0100)]
[ARM] Fix the type for v4f16 duplane
This was previously using the 32bit variant of the instruction, instead
of the 16bit as intended.
Fixes #58512
Differential Revision: https://reviews.llvm.org/D136422
Diana Picus [Tue, 11 Oct 2022 12:38:08 +0000 (12:38 +0000)]
[flang] Rename hypotf on MSVC platforms
The single precision `hypot` intrinsic is lowered to a call to the libm
`hypotf` function. However, the MSVC runtime lacks a hypotf function
and instead uses `_hypotf` (*). This patch tries to find and rewrite
calls to `hypotf` if we're on a MSVC platform.
Calls to libm functions can be introduced even after lowering (**).
Therefore, we try to do the rewriting at the very end of FIR to LLVM
lowering.
Fixes https://github.com/llvm/llvm-project/issues/57563
(*) More specifically, MSVC's headers define hypotf as an inline
function that just calls _hypotf. This works fine for clang, since it
will include those headers, but flang only links with the CRT so we
don't get a free ride.
(**) https://github.com/llvm/llvm-project/blob/
56f94ede2af9a327e59fe84dbf8cbbb7bb1dfa79/flang/lib/Optimizer/CodeGen/CodeGen.cpp#L3391
Differential Revision: https://reviews.llvm.org/D135853
Nikita Popov [Fri, 21 Oct 2022 09:05:53 +0000 (11:05 +0200)]
[AutoUpgrade] Fix remangling when upgrading struct return type
This was remangling the old function rather than the new one, and
could result in failures when we were performing both a struct
return upgrade and an opaque pointer upgrade.
Luo, Yuanke [Thu, 20 Oct 2022 09:49:47 +0000 (17:49 +0800)]
[Verifier] Relieve intrinsics parameter alignment size constrain
In D121898 we restrict parameter alignment size in IR since DAGISel
only have 4 bits to hold the alignment value. However intrinsics
won't be lowered to call instruction, so we can remove the constrain
for intrinsics.
Differential Revision: https://reviews.llvm.org/D136330
wanglei [Fri, 21 Oct 2022 08:35:23 +0000 (16:35 +0800)]
[LoongArch] Modify ParserMethod for the simm26_b operand type
Modify the ParserMethod of `simm26_b` operand type to `parseImmediate`.
Before that, for the `simm26_b` operand type, the same ParserMethod
was used as `simm26_bl`. When using the internal assembler to process
the blockaddress with `asm` instruction, the wrong blockaddress symbol
would be generated due to the call to the `getOrCreateSymbol()`
interface.
Differential Revision: https://reviews.llvm.org/D136073
Nikita Popov [Fri, 21 Oct 2022 08:53:24 +0000 (10:53 +0200)]
[LLParser] Remove redundant remangling (NFCI)
UpgradeCallsToIntrinsic() is already intended to perform remangling
in case no other upgrades are necessary. The additional
remangleIntrinsicFunction() calls are not needed and can hide bugs
in the UpgradeCallsToIntrinsic() implementation.
Nikita Popov [Fri, 21 Oct 2022 08:49:45 +0000 (10:49 +0200)]
[AutoUpgrade] Fix length check for intrinsic upgrade
The shortest intrinsics that can be upgraded via remangling have
8 characters (like "llvm.abs"). Make sure these go through the
upgrade code.
I think that currently this change is not observable from in-tree
callers of UpgradeIntrinsicFunction(), because callers do
redundant remangling checks. However, this issue shows up in
existing tests if those checks are removed (which I will do in
followup changes).
Timm Bäder [Fri, 14 Oct 2022 06:19:53 +0000 (08:19 +0200)]
[clang][Interp] Support base class constructors
Differential Revision: https://reviews.llvm.org/D135025
Timm Bäder [Fri, 14 Oct 2022 06:19:30 +0000 (08:19 +0200)]
[clang][Interp] Array initialization via ImplicitValueInitExpr
Differential Revision: https://reviews.llvm.org/D135013
Timm Bäder [Thu, 20 Oct 2022 13:44:09 +0000 (15:44 +0200)]
[clang][Interp][NFC] Remove some unused aliases.
David Green [Fri, 21 Oct 2022 08:35:39 +0000 (09:35 +0100)]
[ARM] Regnereate armv8.2a-fp16-vector-intrinsics.ll test. NFC
Florian Hahn [Fri, 21 Oct 2022 07:54:10 +0000 (08:54 +0100)]
[VPlan] Print predicates for widened cmp instructions (NFC).
Danil Sidoruk [Thu, 29 Sep 2022 07:24:55 +0000 (10:24 +0300)]
[clang-format] Add option for aligning requires clause body
Adds an option whether requires clause body should be aligned with
the `requires` keyword.
This option is now the default, both without configuration and in LLVM
style.
Fixes https://github.com/llvm/llvm-project/issues/56283
Differential Revision: https://reviews.llvm.org/D129443
Co-authored-by: Emilia Dreamer <emilia@rymiel.space>
Pierre van Houtryve [Wed, 19 Oct 2022 09:01:19 +0000 (09:01 +0000)]
[AMDGPU][GISel] Constrain selected operands in selectG_BUILD_VECTOR
Small bugfix. Currently harmless but a case in D134354 triggers it.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D136235
Adrian Kuegel [Fri, 21 Oct 2022 06:46:14 +0000 (08:46 +0200)]
Pierre van Houtryve [Thu, 20 Oct 2022 12:14:30 +0000 (12:14 +0000)]
[AMDGPU][GISel] Re-enable some working tests
These tests had been commented out but seem to not be crashing.
Not sure if codegen is perfect in each of them, but even if it's not I think it's better to put a TODO to fix codegen than remove the test outright, unless codegen is plain wrong (then I'd still rather XFAIL rather than hide it)
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D136341
Pierre van Houtryve [Mon, 17 Oct 2022 07:56:20 +0000 (07:56 +0000)]
[AMDGPU][DAG] Fix trunc/shift combine condition
The condition needs to be different for right-shifts, else we may lose information in some cases.
Reviewed By: foad
Differential Revision: https://reviews.llvm.org/D136059
Michael Wyman [Wed, 19 Oct 2022 20:11:41 +0000 (13:11 -0700)]
Don't emit `-Wnullability-completeness` warnings on `weak` Objective-C properties.
Zeroing weak references are by definition `nullable`, and adding
`nonnull` or `_Nullable` yields a mutual-exclusivity error. When
`-Wnullability-completeness` is enabled, however, non-audited header
regions require adding the `nullable` property keyword to avoid a
warning. This should be unnecessary, since it restates known nullability
of the `weak` property.
Additionally, the fix-it hints are both non-idiomatic Objective-C
(adding `_Nullable` to the property's pointer type rather than in the
`@property` attributes) and suggest the option of adding `_Nonnull`,
which would be an error.
Differential Revision: https://reviews.llvm.org/D128031
LLVM GN Syncbot [Fri, 21 Oct 2022 04:55:54 +0000 (04:55 +0000)]
[gn build] Port
0332a8e7d6da
Carlos Alberto Enciso [Thu, 20 Oct 2022 07:49:21 +0000 (08:49 +0100)]
[llvm-debuginfo-analyzer] (05/09) - Select elements
llvm-debuginfo-analyzer is a command line tool that processes debug
info contained in a binary file and produces a debug information
format agnostic “Logical View”, which is a high-level semantic
representation of the debug info, independent of the low-level
format.
The code has been divided into the following patches:
1) Interval tree
2) Driver and documentation
3) Logical elements
4) Locations and ranges
5) Select elements
6) Warning and internal options
7) Compare elements
8) ELF Reader
9) CodeView Reader
Full details:
https://discourse.llvm.org/t/llvm-dev-rfc-llvm-dva-debug-information-visual-analyzer/62570
This patch:
Select elements
- Support for logical elements selection:
LVPatterns
Reviewed By: psamolysov, probinson
Differential Revision: https://reviews.llvm.org/D125780
Craig Topper [Fri, 21 Oct 2022 04:35:47 +0000 (21:35 -0700)]
[RISCV] Add vscale x 1 cost model tests for compares. NFC
Weining Lu [Fri, 21 Oct 2022 01:56:27 +0000 (09:56 +0800)]
[LoongArch] Report error in AsmParser when rd == rk or rd == rj for AM* instructions
Do this check because the ISA manual says (edited from the original translation):
> If the AM* instruction has its rd == rj, an Instruction Non-defined Exception will be triggered when the instruction is executed.
>
> If the AM* instruction has its rd == rk, the execution result is unpredictable. It is software's responsibility to avoid this situation.
Note that binutils does the same check except when rd == r0 but this
is undocumented.
Differential Revision: https://reviews.llvm.org/D136076
Peiming Liu [Fri, 21 Oct 2022 00:28:02 +0000 (00:28 +0000)]
[mlir][sparse] update doc for sparse_tensor.foreach operator.
Clarify that sparse_tensor.foreach iterates sparse_tensor in stored dim order.
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D136401
Peiming Liu [Thu, 20 Oct 2022 22:05:28 +0000 (22:05 +0000)]
[mlir][sparse] support foreach on dense tensor.
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D136384
Siva Chandra Reddy [Fri, 21 Oct 2022 00:01:16 +0000 (00:01 +0000)]
[libc][Obvious] Add few missing license headers.
bixia1 [Thu, 20 Oct 2022 22:43:03 +0000 (15:43 -0700)]
[mlir][sparse] Fix getUnorderedCOOFromType for rank 1 tensor.
Previously, it used DimLevelType::SingletonNo to represent an unorder COO
tensor of rank 1 while it should use DimLevelType::CompressedNuNo.
Reviewed By: Peiming, wrengr
Differential Revision: https://reviews.llvm.org/D136387
Michael Jones [Thu, 20 Oct 2022 22:39:55 +0000 (15:39 -0700)]
[libc][obvious] fix comparison test cmake
The name of the libc target changed, this patch updates the cmake for
the string to float comparison test to use the correct name.
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D136388
rkayaith [Tue, 11 Oct 2022 21:39:34 +0000 (17:39 -0400)]
[mlir-opt] Add a '-dump-pass-pipeline' option
Add an option to dump the pipeline that will be run to stderr. A
dedicated option is needed since the existing `test-dump-pipeline`
pipeline won't be usable with `-pass-pipeline` after D135745.
Reviewed By: rriddle, mehdi_amini
Differential Revision: https://reviews.llvm.org/D135747
Luo, Yuanke [Wed, 19 Oct 2022 00:32:13 +0000 (08:32 +0800)]
[X86][AVX] Eliminate redundant movzbl instruction.
The movzbl instruction can be combined to vpinsrb or vmovd, when it is
actual lowered from anyext.
Differential Revision: https://reviews.llvm.org/D130953
rkayaith [Sun, 25 Sep 2022 20:47:39 +0000 (16:47 -0400)]
[mlir] Include anchor op when printing pass managers
Previously a pipeline nested on `anchor-op` would print as just
`'pipeline'`, now it will print as `'anchor-op(pipeline)'`. This ensures
the text form includes all information needed to reconstruct the pass
manager.
Reviewed By: rriddle, mehdi_amini
Differential Revision: https://reviews.llvm.org/D134622
Jan Svoboda [Sat, 15 Oct 2022 15:42:27 +0000 (08:42 -0700)]
[clang][lex] Avoid `DirectoryLookup` copies
This patch fixes a performance regression introduced in D121685 that was caused by copying `DirectoryLookup`.
rdar://
101206790
Reviewed By: ributzka
Differential Revision: https://reviews.llvm.org/D136019
Bill Wendling [Thu, 20 Oct 2022 23:08:01 +0000 (16:08 -0700)]
[clang] Correct sanitizer behavior in union FAMs
Clang doesn't have the same behavior as GCC does with union flexible
array members. (Technically, union FAMs are probably not acceptable in
C99 and are an extension of GCC and Clang.)
Both Clang and GCC treat *all* arrays at the end of a structure as FAMs.
GCC does the same with unions. Clang does it for some arrays in unions
(incomplete, '0', and '1'), but not for all. Instead of having this
half-supported feature, sync Clang's behavior with GCC's.
Reviewed By: kees
Differential Revision: https://reviews.llvm.org/D135727
Xiang Li [Tue, 18 Oct 2022 16:22:46 +0000 (09:22 -0700)]
[HLSL] Disable integer promotion to avoid int16_t being promoted to int for HLSL.
short will be promoted to int in UsualUnaryConversions.
Disable it for HLSL to keep int16_t as 16bit.
Reviewed By: aaron.ballman, rjmccall
Differential Revision: https://reviews.llvm.org/D133668
Rob Suderman [Thu, 20 Oct 2022 21:16:10 +0000 (14:16 -0700)]
[mlir][tosa] Add broadcasting case for tosa.resize degenerate case
When the resize is ?x1x1x?, the tosa.resize operation broadcasts the
input and (when quantized) applies a scaling factor. Updated the resize
operation to not use a tensor.extract operation, instead broadcasting
the only positional value as necessary.
Moved the tosa.resize tests to their own mlir test due to increased
complexity. Also corrected a bug where tosa.resize for bilinear-floating
point was not applying the correct scaling.
Reviewed By: jpienaar
Differential Revision: https://reviews.llvm.org/D136299
Jonas Devlieghere [Thu, 20 Oct 2022 22:20:50 +0000 (15:20 -0700)]
Revert "[lldb] Fix member access in GetExpressionPath"
This reverts commit
0205aa4a02570dfeda5807f66756ebdbb102744b because it
breaks TestArray.py:
a->c = <parent failed to evaluate: parent is NULL>
I decided to revert instead of disable the test because it looks like a
legitimate issue with the patch.
Eli Friedman [Thu, 20 Oct 2022 22:06:20 +0000 (15:06 -0700)]
[Arm64EC] Refer to dllimport'ed functions correctly.
Arm64EC has two different ways to refer to dllimport'ed functions in an
object file. One is using the usual __imp_ prefix, the other is using an
Arm64EC-specific prefix __imp_aux_. As far as I can tell, if a function
is in an x64 DLL, __imp_aux_ refers to the actual x64 address, while
__imp_ points to some linker-generated code that calls the exit thunk.
So __imp_aux_ is used to refer to the address in non-call contexts,
while __imp_ is used for calls to avoid the indirect call checker.
There's one twist to this, though: if an object refers to a symbol using
the __imp_aux_ prefix, the object file's symbol table must also contain
the symbol with the usual __imp_ prefix. The symbol doesn't actually
have to be used anywhere, it just has to exist; otherwise, the linker's
symbol lookup in x64 import libraries doesn't work correctly. Currently,
this is handled by emitting a .globl __imp_foo directive; we could try
to design some better way to handle this.
One minor quirk I haven't figured out: apparently, in Arm64EC mode, MSVC
prefers to use a linker-synthesized stub to call dllimport'ed functions,
instead of branching directly. The linker stub appears to do the same
thing that inline code would do, so not sure if it's just a code-size
optimization, or if the synthesized stub can actually do something other
than just load from the import table in some circumstances.
Differential Revision: https://reviews.llvm.org/D136202
Aart Bik [Thu, 20 Oct 2022 21:39:34 +0000 (14:39 -0700)]
[mlir][sparse] introduce new op for querying number of stored entries
Reviewed By: bixia
Differential Revision: https://reviews.llvm.org/D136382
Alexander Shaposhnikov [Thu, 20 Oct 2022 21:45:22 +0000 (21:45 +0000)]
[clang-tidy] Skip private default ctors in modernize-use-equals-default
For c++17 (i.e. before c++20) making a private default ctor explicitly defaulted
might expose the previously intentionally disallowed initializations, e.g.
class Tag { Tag() {} friend class Widget; }; is not equivalent to
class Tag { Tag() = default; friend class Widget; };
since in the latter case Tag is treated as an aggregate despite having a declaration
of the default constructor. This diff makes modernize-use-equals-default skip
in-class empty nonpublic default ctors to avoid code breakages.
Test plan: ninja check-all
Differential revision: https://reviews.llvm.org/D136224
Arthur Eubanks [Sat, 15 Oct 2022 04:07:49 +0000 (21:07 -0700)]
[lldb] Don't check environment default char signedness when creating clang type for "char"
With -f(un)signed-char, the die corresponding to "char" may be the opposite DW_ATE_(un)signed_char from the default platform signedness.
Ultimately we should determine whether a type is the unspecified signedness char by looking if its name is "char" (as opposed to "signed char"/"unsigned char") and not care about DW_ATE_(un)signed_char matching the platform default.
Fixes #23443
Reviewed By: labath
Differential Revision: https://reviews.llvm.org/D136011
Craig Topper [Thu, 20 Oct 2022 21:05:22 +0000 (14:05 -0700)]
[MachineFrameInfo][RISCV] Call ensureStackAlignment for objects created with scalable vector stack id.
This is an alternative to fix PR57939 for RISC-V. It definitely
can be argued that the stack temporaries for RISC-V are being created
with an unnecessarily large alignment. But ignoring the alignment
in MachineFrameInfo also seems bad.
Looking at the test update that go with the current ID==0 check,
it was intending to exclude things like the NoAlloc stackid. So I'm
not sure if scalable vectors are intentionally being excluded.
Reviewed By: reames
Differential Revision: https://reviews.llvm.org/D135913
Peiming Liu [Thu, 20 Oct 2022 19:40:24 +0000 (19:40 +0000)]
[mlir][sparse] attach bufferizableOpInterface to InsertOp
To fix D136286
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D136372
Sujin Park [Thu, 20 Oct 2022 20:35:45 +0000 (13:35 -0700)]
[lldb][trace] Correctly treat kernel CPUs as individual threads
Resolved a bug in kernel decoding and correctly treat kernel CPUs as
individual threads.
Differential Revision: https://reviews.llvm.org/D136371
Xiang Li [Wed, 19 Oct 2022 23:23:39 +0000 (16:23 -0700)]
[NFC] [DirectX backend] move ResourceClass into llvm.
Move ResourceClass into llvm/Frontend/HLSL/HLSLResource.h so it could be shared between clang and DirectX backend.
Reviewed By: beanz
Differential Revision: https://reviews.llvm.org/D136134
Michael Francis [Tue, 18 Oct 2022 15:47:53 +0000 (15:47 +0000)]
[clang][AIX] Fix mcount name and call arguments
Currently, compiling a program with the `-pg` flag will result in an
undefined symbol error for `.mcount`. This revision fixes the call to
use `__mcount`, which requires a pointer argument to a pointer-sized
object (unique per inserted call) on AIX.
This is only a partial fix. This patch should fix the `-pg` flag's
behaviour on AIX to work with code you are compiling, but it will not
link against standard libraries with `mcount` instrumentation calls. The
next step is to add profiled libraries to the linker search paths in the
Clang driver for the AIX toolchain when linking with `-pg`.
Differential Review: https://reviews.llvm.org/D135384
bixia1 [Thu, 20 Oct 2022 19:05:36 +0000 (12:05 -0700)]
[mlir][sparse] Fix the codegen for the convert operator to handle hidden nop convert.
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D136291
Qiongsi Wu [Thu, 20 Oct 2022 19:16:31 +0000 (15:16 -0400)]
[clang][AIX] Omitting Explicit Debugger Tuning Option
On AIX, the default debugger is `dbx` so it is not necessary to explicitly set `-debugger-tuning=dbx` in the presence of `-g`.
Reviewed By: shchenz
Differential Revision: https://reviews.llvm.org/D136187
Bill Wendling [Thu, 20 Oct 2022 18:55:38 +0000 (11:55 -0700)]
[clang][Sema] Use correct array size for diagnostic
The diagnostic was confusing and reporting that an array contains far
more elements than it is defined to have, due to casting.
For example, this code:
double foo[4096];
((char*)foo)[sizeof(foo)];
warns that the "index 32768 is past the end of the array (which contains
32768 elements)."
Reviewed By: serge-sans-paille, aaron.ballman
Differential Revision: https://reviews.llvm.org/D135920
Caroline Concatto [Thu, 20 Oct 2022 18:38:18 +0000 (19:38 +0100)]
Revert "[AArch64]SME2 Multi-single vector SVE Destructive 2 and 4 Registers"
This reverts commit
4c4909703d74883e5cc49edcbd22b783135d2897.
This patch was breaking this test:
llvm/test/MC/Mips/mips64r6/valid.s
I will push again when fixed
Caroline Concatto [Thu, 20 Oct 2022 18:37:56 +0000 (19:37 +0100)]
Revert "[AArch64]SME2 Multi-vector - Index/Single/Multi Array Vectors FMA sources"
This reverts commit
3fee9358baab54e4ed646a106297e7fb6f1b4cff.
Michael Kruse [Thu, 20 Oct 2022 16:49:38 +0000 (11:49 -0500)]
[Polly] Ensure -polly-detect-keep-going still eventually rejects invalid regions.
Fixes #58484
Caroline Concatto [Thu, 20 Oct 2022 17:57:33 +0000 (18:57 +0100)]
[AArch64]SME2 Multi-vector - Index/Single/Multi Array Vectors FMA sources
This patch adds the assembly/disassembly for the following instructions:
INT:
SMLAL
SMLSL
UMLAL
UMLSL
FP:
BFMLAL
BFMLSL
FMLAL
FMLSL
For multiple and indexed vector, Multiple and Single vector and
Multi vectors, for 1, 2 and 4 ZA registers.
The reference can be found here:
https://developer.arm.com/documentation/ddi0602/2022-09
It also adds a new immediate:
uimm3s2range for off3
uimm2s2range for off2
to represent the vector select offset.
The new operands have the range between the first and the last vector position.
Depends on: D135563
Reviewed By: aemerson, sdesmalen
Differential Revision: https://reviews.llvm.org/D135785
Peiming Liu [Thu, 20 Oct 2022 16:53:56 +0000 (16:53 +0000)]
[mlir][sparse] add source materizalization callback for sparse tensor codegen type converter.
Required by scf.for to achieve 1:N type conversion (See D136314).
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D136367
Caroline Concatto [Mon, 17 Oct 2022 10:46:32 +0000 (11:46 +0100)]
[AArch64]SME2 Multi-single vector SVE Destructive 2 and 4 Registers
This patch adds the assembly/disassembly for the following instructions:
ADD (to vector): Add replicated single vector to multi-vector with multi-vector result.
SQDMULH (multiple and single vector): Multi-vector signed saturating doubling multiply high by vector.
for 2 and 4 ZA SVE registers.
The reference can be found here:
https://developer.arm.com/documentation/ddi0602/2022-09
It also adds more size for the multiple register tuple:
ZZ_b_mul_r, ZZ_h_mul_r,
ZZZZ_b_mul_r, ZZZZ_h_mul_r,
for 8 bits and 16 bits with 2 and 4 ZA registers.
Depends on: D135468
Differential Revision: https://reviews.llvm.org/D135563
River Riddle [Wed, 19 Oct 2022 23:21:10 +0000 (16:21 -0700)]
[tblgen-lsp-server] Fix various ASAN issues
* Properly use virtual destructors for index symbols
* Clear the index before initializing it
Eli Friedman [Thu, 20 Oct 2022 17:47:44 +0000 (10:47 -0700)]
[AArch64] Fix scheduler crash in fusion code.
Make sure we don't call getReg() on the first operand of instruction
without knowing that operand is actually a register.
(This codepath isn't enabled for most CPUs; only triggers on certain
CPUs, like Cortex-X1.)
Differential Revision: https://reviews.llvm.org/D136296
Caroline Concatto [Mon, 17 Oct 2022 09:27:12 +0000 (10:27 +0100)]
[AArch64]SME2 Multiple vector ternary int/float 2 and 4 registers
This patch adds the assembly/disassembly for the following instructions:
For INT:
ADD(array results, multiple vectors): Add multi-vector to multi-vector with ZA array vector results.
SUB(array results, multiple vectors): Subtract multi-vector from multi-vector with ZA array vector results.
For FP:
FMLA (multiple vectors): Multi-vector floating-point fused multiply-add.
FMLS (multiple vectors): Multi-vector floating-point fused multiply-subtract.
The reference can be found here:
https://developer.arm.com/documentation/ddi0602/2022-09
This patch also adds a register operand to represent multiples of ZA multi-vectors.
They are:
ZZ_s_mul_r, ZZ_d_mul_r, ZZZZ_s_mul_r and ZZZZ_d_mul_r
and represent the Zn or Zm times 2 or 4 according to the vector group.
Depends on: D135455
Differential Revision: https://reviews.llvm.org/D135468
William Huang [Wed, 12 Oct 2022 17:49:24 +0000 (17:49 +0000)]
[InstCombine] Canonicalize GEP of GEP by swapping constant-indexed GEP to the back
Canonicalize GEP of GEP by swapping GEP with some suffix constant indices to the back (and GEP with all constant indices to the back of that), this allows more constant index GEP merging to happen. Exceptions are: If swapping violates use-def relations, or anti-optimizes LICM
For constant indexed GEP of GEP, if they cannot be merged directly, they will be casted to i8* and merged.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D125845
Mahesh Ravishankar [Thu, 20 Oct 2022 16:59:10 +0000 (16:59 +0000)]
[mlir] Set pattern that resolves tensor dimensions as having bounded recursion.
Reviewed By: springerm
Differential Revision: https://reviews.llvm.org/D136200
zhongyunde [Thu, 20 Oct 2022 16:33:28 +0000 (00:33 +0800)]
[AArch64][SelectionDAG] Lower multiplication by a constant to shl+add+shl+add
Change the costmodel to lower a = b * C where C = (1 + 2^m) * (1 + 2^n) to
add w8, w0, w0, lsl #m
add w0, w8, w8, lsl #n
Note: The latency can vary depending on the shirt amount
Reviewed By: efriedma, dmgreen
Differential Revision: https://reviews.llvm.org/D135441
Xiang Li [Wed, 19 Oct 2022 19:40:39 +0000 (12:40 -0700)]
[HLSL] Add groupshare address space.
Added keyword, LangAS and TypeAttrbute for groupshared.
Tanslate it to LangAS with asHLSLLangAS.
Make sure it translated into address space 3 for DirectX target.
Reviewed By: aaron.ballman
Differential Revision: https://reviews.llvm.org/D135060
Joseph Huber [Thu, 20 Oct 2022 15:38:27 +0000 (10:38 -0500)]
[OpenMP] Remove `-Bsymbolic` flag for device linking in the GNU toolchain
Previously, OpenMP linking would be done explicitly in a linker stage.
For `x86_64` offloading this would just use the host linker, which could
be the `bfd` linker. This linker had problems linking relocations
against variables with protected visibility so we force `-Bsymbolic`
when linking. After the deprecation of the old offloading driver this
code is no longer used and can be removed.
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D136363
Joseph Huber [Thu, 20 Oct 2022 15:51:46 +0000 (10:51 -0500)]
[Libomptarget] Build plugins with protected visibility by default
The plugins all define the same interface symbols. This is generally not
a problem when calling the plugin directly from the dynamic library's
handle. However, when calling from within the plugin itself it is
possible for another plugin's symbols to preempt the symbols. This was
observed with the `__tgt_rtl_is_valid_binary` call in the
`__tgt_rtl_is_valid_binary_info` function being mapped to the x86_64
plugin.
This patch changes the default visibility to `protected` intead. This
visibility ensures that these symbols are all externally visible from
the plugin, but ensures their definitions are fixed within the shared
library. Having protected visiiblity makes such symbol preemption
impossible.
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D136365
Denys Shabalin [Thu, 20 Oct 2022 15:50:26 +0000 (17:50 +0200)]
[mlir] Fix incorrect temporary file handling on windows
Reviewed By: ftynse
Differential Revision: https://reviews.llvm.org/D136364
LLVM GN Syncbot [Thu, 20 Oct 2022 15:30:49 +0000 (15:30 +0000)]
[gn build] Port
a48007355a03
Mark de Wever [Thu, 5 May 2022 06:03:58 +0000 (08:03 +0200)]
[libc++][format] Implements string escaping.
Implements parts of
- P2286R8 Formatting Ranges
Reviewed By: #libc, tahonermann
Differential Revision: https://reviews.llvm.org/D134036
Mark de Wever [Wed, 19 Oct 2022 18:01:57 +0000 (20:01 +0200)]
[libc++][CI] Reorder jobs.
In the second leg of the CI the steps take about:
- C++2b 10m
- C++11 8m
- C++03 6m
- Modular build 10m
- GCC 12 / C++latest 20m
So the slowest job is scheduled last. The CI will wait to start the
third leg until that job is done. The current order increases the
latency of the current job, instead start the slow jobs earlier.
Reviewed By: #libc, ldionne
Differential Revision: https://reviews.llvm.org/D136276
Craig Topper [Thu, 20 Oct 2022 15:13:25 +0000 (08:13 -0700)]
[RISCV] Remove EEW from some sched classes.
This removes the EEW from unit stride load/store and whole register
load, store, move.
It seems reasonable that implementations of these instructions wouldn't
usually be affected by element width.
We likely need to add LMUL information to our scheduling classes so
I thought it might be good to remove a few before they got multiplied
by LMUL.
Reviewed By: reames, michaelmaitland
Differential Revision: https://reviews.llvm.org/D135992
Mircea Trofin [Thu, 20 Oct 2022 15:21:19 +0000 (08:21 -0700)]
[mlgo] Fix test post-D136040
Instruction opcodes bumped, trivial fix.
Paul Walker [Thu, 20 Oct 2022 11:53:58 +0000 (11:53 +0000)]
[NFC] Fix a few whitespace inconsistencies.
Nikita Popov [Thu, 20 Oct 2022 14:40:38 +0000 (16:40 +0200)]
[ModRef] Default to ModRef in MemoryEffects ctors (NFC)
Allow writing something like MemoryEffects::argMemOnly() implying
MemoryEffects::argMemOnly(ModRefInfo::ModRef). Having to write this
out is a bit of an annoying than porting code using the existing
attributes in D135780.
Pierrick Bouvier [Thu, 20 Oct 2022 14:29:15 +0000 (16:29 +0200)]
Detect Visual Studio automatically in Windows packaging script
Instead of hardcoding several VS paths, use vswhere.exe (available from
VS 2017) to get latest version available.
Reviewed By: hans, thieta
Differential Revision: https://reviews.llvm.org/D135873
OCHyams [Thu, 20 Oct 2022 14:26:40 +0000 (15:26 +0100)]
[NFC] Add DebugVariable constructor that takes DbgVariableIntrinsic pointer
Note: The constructor definition cannot be inline without some refactoring as
it introduces a circular dependency between the headers
llvm/IR/DebugInfoMetadata.h (this file) and llvm/IR/IntrinsicInst.h (where
DbgVariableIntrinsic is defined).
Reviewed By: jryans
Differential Revision: https://reviews.llvm.org/D133286
OCHyams [Thu, 20 Oct 2022 13:17:47 +0000 (14:17 +0100)]
[DebugInfo][NFC] Refactor debug intrinsic copy and delete to instead just move
Reviewed By: jryans
Differential Revision: https://reviews.llvm.org/D133304
Guillaume Chatelet [Thu, 20 Oct 2022 13:52:30 +0000 (13:52 +0000)]
Revert D136292 "[libc] mem* framework v3"
This breaks llvm-libc build bots:
- libc-x86_64-debian-dbg-asan
- libc-x86_64-debian-fullbuild-dbg-asan
Address sanitizers fail with "AddressSanitizer: invalid alignment requested in aligned_alloc: 64, alignment must be a power of two and the requested size 0x41 must be a multiple of alignment (thread T0)"
- libc-aarch64-ubuntu-dbg
- libc-aarch64-ubuntu-fullbuild-dbg
https://lab.llvm.org/buildbot/#/builders/223/builds/8877/steps/7/logs/stdio
- libc-arm32-debian-dbg
https://lab.llvm.org/buildbot/#/builders/229/builds/5201/steps/7/logs/stdio
This reverts commit
903cc71a82431d79e5fb541946a9e7c93750e374.
Aliia Khasanova [Thu, 20 Oct 2022 12:39:03 +0000 (12:39 +0000)]
[mlir][nfc] Clean-up usage of kDynamicSize.
This patch prepares MLIR code base to change the value of kDynamicSize.
https://discourse.llvm.org/t/rfc-unify-kdynamicsize-and-kdynamicstrideoroffset/64534/4
Differential Revision: https://reviews.llvm.org/D136327
Denys Shabalin [Thu, 20 Oct 2022 10:58:49 +0000 (12:58 +0200)]
[mlir] Fix and test python bindings for dump_to_object_file
Reviewed By: ftynse
Differential Revision: https://reviews.llvm.org/D136334
Guillaume Chatelet [Wed, 19 Oct 2022 20:52:45 +0000 (20:52 +0000)]
[libc] mem* framework v3
This version is more composable and also simpler at the expense of being more explicit and more verbose.
This patch provides rationale for the framework, implementation and unit tests but the functions themselves are still using the previous version. The change in implementation will come in a follow up patch.
Differential Revision: https://reviews.llvm.org/D136292
Florian Hahn [Thu, 20 Oct 2022 13:30:01 +0000 (14:30 +0100)]
[LV] Use buildScalarSteps to also handle VF = 1. (NFCI)
The code in buildScalarSteps already properly handles creating the
scalar induction values with VF = 1. Use it directly instead of using
extra code to handle that case.
Suggested by @Ayal in D133760.
Nikita Popov [Wed, 19 Oct 2022 14:57:41 +0000 (16:57 +0200)]
Reapply [FunctionAttrs] Make location classification more precise
Reapplying after the fix for volatile modelling in D135863.
-----
Don't add argmem if the pointer is clearly not an argument (e.g.
a global). I don't think this makes a difference right now, but
gives more obvious results with D135780.
Nathan Sidwell [Thu, 20 Oct 2022 11:10:47 +0000 (07:10 -0400)]
libcxxabi [PR58117][NFC]: Open code lower bound
This open codes the use of lower-bound when looking for an operator
encoding. Using std::lower_bound can result in symbol references to
the C++ library and that breaks the ABI demangler, which mandates no
such dependency.
Differential Revision: https://reviews.llvm.org/D135799
Fixes: https://github.com/llvm/llvm-project/issues/58117
Carlos Alberto Enciso [Thu, 20 Oct 2022 10:29:57 +0000 (11:29 +0100)]
[llvm-debuginfo-analyzer] (04/09) - Locations and ranges
The test case 'LocationCoverage' caused a failure in:
https://lab.llvm.org/buildbot/#/builders/5/builds/28385
https://lab.llvm.org/buildbot/#/builders/168/builds/9585
SUMMARY: AddressSanitizer: 176 byte(s) leaked in 2 allocation(s).
The logical debug locations 'LocationThree' and 'LocationFour'
are not added to any logical scope.
The test case is checking the logical symbol coverage and
it does not use those 2 debug locations.
Reviewed By: probinson
Differential Revision: https://reviews.llvm.org/D136333
Enna1 [Thu, 20 Oct 2022 12:30:08 +0000 (20:30 +0800)]
[memprof] Support installation of memprof headers
This change allows users manually calling memprof public C API (e.g. __memprof_profile_dump).
Reviewed By: tejohnson
Differential Revision: https://reviews.llvm.org/D136067
Sander de Smalen [Thu, 20 Oct 2022 11:17:34 +0000 (12:17 +0100)]
[AArch64][SME] Remove get.pstatesm intrinsic.
This intrinsic can be removed in favour of using a call to
__arm_sme_state() directly and testing the LSB of X0.
In IR that would look like:
%pstate = call aarch64_sme_preservemost_from_x2 {i64, i64} @__arm_sme_state()
%pstate.x0 = extractvalue {i64, i64} %pstate, 0
%pstate.sm = and i64 %pstate.x0, 1
Dmitry Preobrazhensky [Thu, 20 Oct 2022 12:17:22 +0000 (15:17 +0300)]
[AMDGPU][MC][GFX11][NFC] Fix trivial typos in tests
isuckatcs [Fri, 14 Oct 2022 16:03:13 +0000 (18:03 +0200)]
[analyzer] Move unexecuted test block into it's own source file
Inside lambdas.cpp a block of code wasn't executed,
because it required the standard to be at least c++14.
This patch moves this block of code into it's own
source file and makes sure it's tested.
Differential Revision: https://reviews.llvm.org/D135965
Dmitry Preobrazhensky [Thu, 20 Oct 2022 12:12:54 +0000 (15:12 +0300)]
[AMDGPU][MC][GFX11][NFC] Improve error message when a VOPD opcode is used with WS64
Differential Revision: https://reviews.llvm.org/D136168
Pierrick Bouvier [Thu, 20 Oct 2022 12:12:28 +0000 (14:12 +0200)]
Introduce options for Windows packaging script
Options:
--version: [required] version to build
--help: display this help
--x86: build and test x86 variant
--x64: build and test x64 variant
Note: At least one variant to build is required.
Example: build_llvm_release.bat --version 15.0.0 --x64
Reviewed By: hans, thieta
Differential Revision: https://reviews.llvm.org/D135255
Nabeel Omer [Thu, 20 Oct 2022 12:06:01 +0000 (12:06 +0000)]
[InstCombine] Fix assert condition in `foldSelectShuffleOfSelectShuffle`
Bug introduced in
e239198cdbbf.
The assert() is making an assumption that the resulting shuffle mask
will always select elements from both vectors, this is untrue in the
case of two shuffles being folded if the former shuffle has a mask with
undef elements in it. In such a case folding the shuffles might result
in a mask which only selects from one of the vectors because the other
elements (in the mask) are undef.
Differential Revision: https://reviews.llvm.org/D136256
Dmitry Preobrazhensky [Thu, 20 Oct 2022 12:05:40 +0000 (15:05 +0300)]
[AMDGPU][MC] Correct v_cndmask operand types
Differential Revision: https://reviews.llvm.org/D136152
Victor Perez [Thu, 20 Oct 2022 11:58:43 +0000 (12:58 +0100)]
[mlir][llvm] Handle llvm.noundef attribute when converting to LLVM IR
Translate LLVMIR llvm.noundef attribute to its equivalent in LLVM IR.
Reviewed By: ftynse
Differential Revision: https://reviews.llvm.org/D136324
Guillaume Chatelet [Thu, 20 Oct 2022 11:58:54 +0000 (11:58 +0000)]
[libc][automemcpy] fix build after change in Arg enum
Florian Hahn [Thu, 20 Oct 2022 10:23:29 +0000 (11:23 +0100)]
[LoopSimplifyCFG] Forget loop and block dispos after merging blocks.
This fixes another case where block and loop dispositions weren't
properly invalidate after changing the CFG.
Fixes #58489.
Nikita Popov [Wed, 19 Oct 2022 14:55:24 +0000 (16:55 +0200)]
[FuncAttrs] Extract code for adding a location access (NFC)
This code is the same for accesses from call arguments and for
accesses from other (single-location) instructions. Extract i
into a common function.
Nikita Popov [Thu, 13 Oct 2022 10:16:26 +0000 (12:16 +0200)]
[FunctionAttrs] Volatile operations can access inaccessible memory
Per LangRef, volatile operations are allowed to access the location
of their pointer argument, plus inaccessible memory:
> Any volatile operation can have side effects, and any volatile
> operation can read and/or modify state which is not accessible
> via a regular load or store in this module.
> [...]
> The allowed side-effects for volatile accesses are limited. If
> a non-volatile store to a given address would be legal, a volatile
> operation may modify the memory at that address. A volatile
> operation may not modify any other memory accessible by the
> module being compiled. A volatile operation may not call any
> code in the current module.
FunctionAttrs currently does not model this and ends up marking
functions with volatile accesses on arguments as argmemonly,
even though they should be inaccessiblemem_or_argmemonly.
Differential Revision: https://reviews.llvm.org/D135863
gonglingqin [Thu, 20 Oct 2022 09:25:58 +0000 (17:25 +0800)]
[LoongArch] Fix 32-bit and 64-bit atomicrmw nand operand order errors
Differential Revision: https://reviews.llvm.org/D136220
Guillaume Chatelet [Thu, 20 Oct 2022 08:52:56 +0000 (08:52 +0000)]
[libc][NFC] add missing static qualifiers in tests
Guillaume Chatelet [Thu, 20 Oct 2022 08:48:22 +0000 (08:48 +0000)]
[libc][NFC] remove unneeded -lpthread for utils_test