Scott Linder [Wed, 15 Feb 2023 17:29:19 +0000 (17:29 +0000)]
[Clang][AMDGPU] Set LTO CG opt level based on Clang option
For AMDGCN default to mapping --lto-O# to --lto-CGO# in a 1:1 manner
(i.e. clang -O<N> implies --lto-O<N> and --lto-CGO<N>).
Ensure there is a means to override this via -Xoffload-linker and begin
to claim these arguments to avoid incorrect warnings that they are not
used.
Reviewed By: yaxunl, MaskRay
Differential Revision: https://reviews.llvm.org/D142499
Scott Linder [Wed, 15 Feb 2023 17:12:47 +0000 (17:12 +0000)]
[LLD] Add --lto-CGO[0-3] option
Allow controlling the CodeGenOpt::Level independent of the LTO
optimization level in LLD via new options for the COFF, ELF, MachO, and
wasm frontends to lld. Most are spelled as --lto-CGO[0-3], but COFF is
spelled as -opt:lldltocgo=[0-3].
See D57422 for discussion surrounding the issue of how to set the CG opt
level. The ultimate goal is to let each function control its CG opt
level, but until then the current default means it is impossible to
specify a CG opt level lower than 2 while using LTO. This option gives
the user a means to control it for as long as it is not handled on a
per-function basis.
Reviewed By: MaskRay, #lld-macho, int3
Differential Revision: https://reviews.llvm.org/D141970
Thomas Raoux [Wed, 15 Feb 2023 16:52:49 +0000 (16:52 +0000)]
[mlir][vectorToGPU] Fix type used when folding transpose into read op
Pick the right result type when folding transpose op into a read
Differential Revision: https://reviews.llvm.org/D144113
J. Ryan Stinnett [Wed, 15 Feb 2023 17:04:51 +0000 (17:04 +0000)]
[DebugInfo][Docs] Fix broken link in instruction referencing doc
Rahul Kayaith [Wed, 15 Feb 2023 16:25:40 +0000 (11:25 -0500)]
[mlir] Silence a few -Wunused-but-set-parameter warnings
Some older versions of gcc hit false positives here:
/workdir/llvm-project/mlir/include/mlir/IR/Attributes.h:391:49: warning: parameter 'ty' set but not used [-Wunused-but-set-parameter]
391 | static inline bool isPossible(mlir::Attribute ty) {
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81676
Krzysztof Drewniak [Tue, 20 Dec 2022 15:32:25 +0000 (15:32 +0000)]
[mlir][AMDGPU] 8-bit float usage in the AMDGPU dialect
Upcoming AMD hardware will include functions that accept 8-bit floats.
Specifically, there are MFMA instructions that accept 8-bit floats,
either using the same or mixed formats. This patch adds MLIR wrappers
for these intrinsics and explicitly adds support for 8-bit floats in
the gpu-to-rocdl conversion by way of amdgpu-to-rocdl.
Since LLVM does not have f8 types, when targeting LLVM for compilation
on an AMD GPU, both f8 types used on AMD hardware (f8E5M2FNUZ and
f8E4M3FNUZ) are rewritten to i8.
This patch also relaxes the restriction that the types of both source
operands to a amdgpu.mfma instructions match exactly, as this is not
necessarily required for the bf8 (f8E5M2FNUZ) and fp8 (f8E4M3FNUZ)
instructions. In addition, since the buffer_{load,store} operations
maintain a whitelist of permitted types, we add the relevant f8 types
to that list.
This patch does not add any implementations of arithmetic operations
for f8 types.
Reviewed By: jakeh-gc
Differential Revision: https://reviews.llvm.org/D143956
Sanjay Patel [Wed, 15 Feb 2023 16:43:17 +0000 (11:43 -0500)]
[InstCombine] remove stale test comment; NFC
Zain Jaffal [Wed, 15 Feb 2023 15:41:24 +0000 (15:41 +0000)]
Recommit "[ConstraintElimination] Change debug output to display variable names."
This reverts commit
02ae7e72b3f00969eeb579a2b4346082827f0b35.
include Value.h in ConstraintSystem.h
Xing Xue [Wed, 15 Feb 2023 16:14:52 +0000 (11:14 -0500)]
[libc++abi][AIX] Skip non-C++ EH aware frames when retrieving exception object
Summary:
The personality routine for the legacy AIX xlclang++ compiler uses the stack slot reserved for compilers to pass the exception object to the landing pad. The landing pad retrieves the exception object with a call to the runtime function __xlc_exception_handle(). The current implementation incorrectly assumes that __xlc_exception_handle() should go up one stack frame to get to the stack frame of the caller of __xlc_exception_handle(), which is supposedly the stack frame of the function containing the landing pad. However, this does not always work, e.g., the xlclang++ compiler sometimes generates a wrapper of __xlc_exception_handle() and calls the wrapper from the landing pad for optimization purposes. This patch changes the implementation to unwind the stack from __xlc_exception_handle() and skip frames not associated with functions that are C++ EH-aware until a frame associated with a C++ EH-aware function is found and then retrieving the exception object with the expectation that the located frame is the one that the personality routine installed as it transferred control to the landing pad.
Reviewed by: cebowleratibm, hubert.reinterpretcast, daltenty
Differential Revision: https://reviews.llvm.org/D143010
David Green [Wed, 15 Feb 2023 16:06:32 +0000 (16:06 +0000)]
[AArch64] Always lower fp16 zero to FMOVH0
We can always use FMOVH0 to lower fp16 zero, even without fullfp16. We can
either expand it to movi d0, #0 or fmov s0, wzr, which will both clear all the
bits of the register.
Differential Revision: https://reviews.llvm.org/D143988
Peter Klausler [Tue, 14 Feb 2023 22:07:27 +0000 (14:07 -0800)]
[flang][runtime] Return the right mutable modes from INQUIRE in child I/O
During child I/O -- a call to a user-defined subroutine with a generic interface
to handle read and write operations on a specific derived type -- ensure that
any INQUIRE statement probing mutable modes like BLANK= will observe changes
that have been made to those modes via control edit descriptors like BN
while processing the original I/O statement's format.
Differential Revision: https://reviews.llvm.org/D144046
Matthias Springer [Wed, 15 Feb 2023 15:40:42 +0000 (16:40 +0100)]
[mlir][linalg] Fix insertion point bug in D144022
This should have been part of D144022.
Nikolas Klauser [Mon, 13 Feb 2023 23:56:09 +0000 (00:56 +0100)]
[libc++][NFC] Replace _LIBCPP_STD_VER > x with _LIBCPP_STD_VER >= x
This change is almost fully mechanical. The only interesting change is in `generate_feature_test_macro_components.py` to generate `_LIBCPP_STD_VER >=` instead. To avoid churn in the git-blame this commit should be added to the `.git-blame-ignore-revs` once committed.
Reviewed By: ldionne, var-const, #libc
Spies: jloser, libcxx-commits, arichardson, arphaman, wenlei
Differential Revision: https://reviews.llvm.org/D143962
Yitzhak Mandelbaum [Mon, 13 Feb 2023 16:27:39 +0000 (16:27 +0000)]
[clang][dataflow] Change `transfer` API to take a reference.
The provided `CFGElement` is never null, so a reference is a more precise type.
Differential Revision: https://reviews.llvm.org/D143920
Nikita Popov [Wed, 15 Feb 2023 15:36:44 +0000 (16:36 +0100)]
Revert "Recommit "[ConstraintElimination] Change debug output to display variable names.""
This reverts commit
2a2a6bfcfe8e62886542cb673ac8df349cf26499.
This causes build failures:
/home/npopov/repos/llvm-project/llvm/lib/Analysis/ConstraintSystem.cpp: In member function ‘llvm::SmallVector<std::__cxx11::basic_string<char> > llvm::ConstraintSystem::getVarNamesList() const’:
/home/npopov/repos/llvm-project/llvm/lib/Analysis/ConstraintSystem.cpp:118:10: error: invalid use of incomplete type ‘class llvm::Value’
118 | if (V->getName().empty())
| ^~
In file included from /home/npopov/repos/llvm-project/llvm/lib/Analysis/ConstraintSystem.cpp:9:
/home/npopov/repos/llvm-project/llvm/include/llvm/Analysis/ConstraintSystem.h:21:7: note: forward declaration of ‘class llvm::Value’
21 | class Value;
| ^~~~~
/home/npopov/repos/llvm-project/llvm/lib/Analysis/ConstraintSystem.cpp:119:22: error: invalid use of incomplete type ‘class llvm::Value’
119 | OperandName = V->getNameOrAsOperand();
| ^~
/home/npopov/repos/llvm-project/llvm/include/llvm/Analysis/ConstraintSystem.h:21:7: note: forward declaration of ‘class llvm::Value’
21 | class Value;
| ^~~~~
/home/npopov/repos/llvm-project/llvm/lib/Analysis/ConstraintSystem.cpp:121:41: error: invalid use of incomplete type ‘class llvm::Value’
121 | OperandName = std::string("%") + V->getName().str();
| ^~
/home/npopov/repos/llvm-project/llvm/include/llvm/Analysis/ConstraintSystem.h:21:7: note: forward declaration of ‘class llvm::Value’
21 | class Value;
| ^~~~~
Nikita Popov [Wed, 15 Feb 2023 15:32:46 +0000 (16:32 +0100)]
[InstSimplify] Add additional insertvalue into undef tests (NFC)
David Spickett [Wed, 15 Feb 2023 08:55:22 +0000 (08:55 +0000)]
[llvm][AArch64] Fix BTI after returns_twice when call has no attributes
Previously we were looking for the returns twice attribute by manually
getting the function attributes from the call. This meant that we only
found attributes on the call itself, not what it was calling.
So if you had:
%call1 = call i32 @setjmp(ptr noundef null)
We would not BTI protect that even though setjmp clearly needs it.
Clang happens to produce:
%call = call i32 @setjmp(ptr noundef null) #0 ; returns_twice
So all valid calls were protected. This is not guaranteed,
the frontend may choose not to put attributes on the call.
It is undefined behaviour to call setjmp indirectly
(https://pubs.opengroup.org/onlinepubs/
9699919799/functions/setjmp.html)
but as I misused the APIs here I think it's worth fixing up regardless.
Added comments to the test file where the IR differs from what
clang would output.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D144082
Zain Jaffal [Wed, 15 Feb 2023 15:22:42 +0000 (15:22 +0000)]
Recommit "[ConstraintElimination] Change debug output to display variable names."
This reverts commit
62d0e1a8541f93dfbf66d982f66da32676df2df7.
remove `dumpWithNames` function
Zain Jaffal [Wed, 15 Feb 2023 15:21:46 +0000 (15:21 +0000)]
Revert "[ConstraintElimination] Change debug output to display variable names."
This reverts commit
869c87ad10e87db7c032c3464338ab9d50916510.
`dumpWithNames` function should be removed
Zain Jaffal [Wed, 15 Feb 2023 15:07:14 +0000 (15:07 +0000)]
[ConstraintElimination] Change debug output to display variable names.
Previously when constraint system outputs the rows in the system the variables used are x1,2...n making it hard to infer which ones they relate to in the IR
Reviewed By: fhahn
Differential Revision: https://reviews.llvm.org/D142618
Guillaume Chatelet [Wed, 15 Feb 2023 10:47:10 +0000 (10:47 +0000)]
[reland][libc] Separate memcpy implementations per arch
As x86_64 implementations is likely to grow up to a point where it's no more manageable to have all implementations in the same file.
Matthias Springer [Wed, 15 Feb 2023 14:26:14 +0000 (15:26 +0100)]
[mlir][memref] Bufferize memref.tensor_store op
This change adds the BufferizableOpInterface implementation for memref.tensor_store.
Differential Revision: https://reviews.llvm.org/D144080
Matthias Springer [Wed, 15 Feb 2023 14:09:34 +0000 (15:09 +0100)]
[mlir][linalg] Add bufferize_to_allocation transform op
This transform materializes a buffer allocation for a given tensor value. All uses of the original value are replaced with the allocation.
Certain non-DPS ops may have an optimized lowering path that bufferizes the entire defining op. Such optimization is added for `tensor.pad` as part of this change.
The resulting IR can be further bufferized with One-Shot Bufferize.
Differential Revision: https://reviews.llvm.org/D144022
Nikita Popov [Wed, 15 Feb 2023 13:33:59 +0000 (14:33 +0100)]
[ConstantFold] Check for constant global earlier (NFC)
Check that the underlying object is a constant global with
definitive initializer upfront, so we can skip the more expensive
offset calculation logic if we can't perform the fold anyway.
Alex Bradbury [Wed, 15 Feb 2023 14:00:54 +0000 (14:00 +0000)]
[lld][RISCV][test] Expand testing in riscv-attributes.s
This patch is related to the discussion in
<https://discourse.llvm.org/t/rfc-resolving-issues-related-to-extension-versioning-in-risc-v/68472/1>.
It slightly expands the coverage of behaviour for unrecognized
extensions (there are slightly different forms of error for zfoo vs e.g.
'y'), and adds coverage for an extension with an
unrecognized/unsupported version number.
Use "unrecognized" terminology because the expectation is the error
checking will be reduced in a follow-up patch posted for review.
Sanjay Patel [Wed, 15 Feb 2023 13:56:24 +0000 (08:56 -0500)]
[LangRef] improve documentation of SNaN in the default FP environment
Make it explicit that SNaN is not handled differently than
QNaN in the LLVM default floating-point environment.
Note that an IEEE-754-compliant model disallows transforms
like "X * 1.0 -> X". That is because math operations are
expected to convert SNaN to QNaN (set the signaling bit).
But LLVM has had those kinds of transforms from the beginning:
https://alive2.llvm.org/ce/z/igb55y
We should be IEEE-754-compliant under strict-FP (the logic is
implemented with a helper named canIgnoreSNaN()), but I don't
think there is any demand to do that with default optimization.
See issue #43070 for earlier draft/discussion about this change.
Differential Revision: https://reviews.llvm.org/D143074
Guillaume Chatelet [Wed, 15 Feb 2023 13:54:13 +0000 (13:54 +0000)]
[bazel] Fix missing dependency in clang-tools-extra/clang-tidy:llvmlibc
Joseph Huber [Wed, 15 Feb 2023 13:37:47 +0000 (07:37 -0600)]
[NVPTX] Fix NVPTX output name in the driver with `-save-temps`
Summary:
Currently, OpenMP and direct compilation uses an NVPTX toolchain to
directly invoke the CUDA tools from Clang to do the assembling and
linking of NVPTX codes. This breaks under `-save-temps` because of a
workaround. The `nvlink` linker does not accept `.o` files, so we need
to be selective when we output these. The previous logic keyed off of
presense in the temp files and wasn't a great solution. Change this to
just query the input args for `-c` to see if we stop at the assembler.
Fixes https://github.com/llvm/llvm-project/issues/60767
Guillaume Chatelet [Wed, 15 Feb 2023 13:29:07 +0000 (13:29 +0000)]
Revert "[libc] Separate memcpy implementations per arch"
This is patch is still breaking downstream users...
This reverts commit
97e441dc6cfae31bc56b375e43899946ec7f08a8.
Guillaume Chatelet [Wed, 15 Feb 2023 10:47:10 +0000 (10:47 +0000)]
[libc] Separate memcpy implementations per arch
As x86_64 implementations is likely to grow up to a point where it's no more manageable to have all implementations in the same file.
Matthias Springer [Wed, 15 Feb 2023 13:15:41 +0000 (14:15 +0100)]
[mlir][transform] Add transform.get_result op
This transform op returns a value handle pointing to the specified OpResult of the targeted op.
Differential Revision: https://reviews.llvm.org/D144087
Benjamin Kramer [Wed, 15 Feb 2023 13:10:09 +0000 (14:10 +0100)]
[clang-format] assert(false) -> llvm_unreachable
Avoids warnings in -asserts builds.
FormatTokenSource.h:240:3: error: non-void function does not return a value [-Werror,-Wreturn-type]
}
^
Jathu Satkunarajah [Wed, 15 Feb 2023 12:51:52 +0000 (13:51 +0100)]
[bazel] create a clang-tidy binary target
Create a binary target for clang-tidy. Tested by running:
```
$ bazel build --config=generic_clang @llvm-project//clang-tools-extra/...
$ bazel test --config=generic_clang @llvm-project//clang-tools-extra/...
$ bazel run --config=generic_clang @llvm-project//clang-tools-extra/clang-tidy -- --help
```
Reviewed By: #bazel_build, aaronmondal
Differential Revision: https://reviews.llvm.org/D143804
Anton Sidorenko [Tue, 14 Feb 2023 10:30:44 +0000 (13:30 +0300)]
[MachineTraceMetrics] Add local strategy
This strategy makes each trace local to the basic block. For in-order cores some
heuristics work better when we do local decisions. For example, MachineCombiner
may expect that instructions outside the current basic block do not lengthen
the critical path when we execute instructions in order or the core has a
small re-order buffer.
This patch only introduce the strategy, real use-case is added in the further
pathes.
Depends on D140539
Reviewed By: spatel
Differential Revision: https://reviews.llvm.org/D140540
Simon Pilgrim [Wed, 15 Feb 2023 12:44:14 +0000 (12:44 +0000)]
[X86] Remove abs(sub_nsw()) -> abds fold from adbu test file
Copy+paste typo - it was correctly removed from 128/512 variants
Martin Storsjö [Wed, 15 Feb 2023 08:42:20 +0000 (10:42 +0200)]
[lldb] Fix a log format warning on Windows, don't assume uint64_t is a long type
On Windows, long is 32 bit, and uint64_t is long long.
Differential Revision: https://reviews.llvm.org/D144078
Manuel Klimek [Wed, 1 Feb 2023 13:18:44 +0000 (13:18 +0000)]
[clang-format] Enable FormatTokenSource to insert tokens.
In preparation for configured macro replacements in formatting,
add the ability to insert tokens to FormatTokenSource, and implement
token insertion in IndexedTokenSource.
Differential Revision: https://reviews.llvm.org/D143070
Matt Arsenault [Wed, 15 Feb 2023 12:31:40 +0000 (08:31 -0400)]
AMDGPU: Fix not adding to depth in getNegatedExpression
Matt Arsenault [Tue, 31 Jan 2023 17:16:59 +0000 (13:16 -0400)]
clang: Rename misleading test name
This is a test for attribute propagation, not metadata
Simi Pallipurath [Wed, 15 Feb 2023 11:06:04 +0000 (11:06 +0000)]
[lld][ARM][NFCI][1/3]Big Endian support - Removing assumptions
Change:
- Replacing the memcpy that assume little endian with the endian-aware write.
Shouldn't affect the output for now, just a prerequisite for the next patches.
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D140201
Caroline Concatto [Wed, 15 Feb 2023 09:40:41 +0000 (09:40 +0000)]
[AArch64][NFC] Rename AEK_SMEF64 and AEK_SMEI64 feature flags
Update feature flag names from:
AEK_SMEF64 to AEK_SMEF64F64
and
AEK_SMEI64 to AEK_SMEI16I64
These feature flags had their name changed in this previous patch
https://reviews.llvm.org/D135974
Reviewed By: c-rhodes
Differential Revision: https://reviews.llvm.org/D143989
Max Kazantsev [Wed, 15 Feb 2023 11:15:24 +0000 (18:15 +0700)]
[SimpleLoopUnswitch] Canonicalize conditions for injection of invariant condition
When loop condition isn't immediately in the form supported by invariant injection
unswitching, try to canonicalize it to this form.
Differential Revision: https://reviews.llvm.org/D143175
Reviewed By: skatkov
Max Kazantsev [Wed, 15 Feb 2023 11:07:02 +0000 (18:07 +0700)]
[Test][NFC] Rename one of test parameters to avoid confusing associations
Name 'len' implicitly hints that it should be non-negative, but in fact in some of
these tests the baseline profitable scenario for the transform to be profitable is
when this thing is negative. Rename to avoid this confusion.
Alvin Wong [Tue, 14 Feb 2023 22:38:52 +0000 (06:38 +0800)]
[Symbolize][MinGW] Support demangling i386 call-conv-decorated C++ names
On i386 Windows, after C++ names have been Itanium-mangled, the C name
mangling specific to its call convention may also be applied on top.
This change teaches symbolizer to be able to demangle this type of
mangled names.
As part of this change, `demanglePE32ExternCFunc` has also been modified
to fix unwanted stripping for vectorcall names when the demangled name
is supposed to contain a leading `_`. Notice that the vectorcall
mangling does not add either an `_` or `@` prefix. The old code always
tries to strip the prefix first, which for Itanium mangled names in
vectorcall, the leading underscore of the Itanium name gets stripped
instead and breaks the Itanium demangler.
Differential Revision: https://reviews.llvm.org/D144049
Nikita Popov [Wed, 15 Feb 2023 10:43:52 +0000 (11:43 +0100)]
[CVP] Add additional ctlz tests (NFC)
Nigel Perks [Tue, 14 Feb 2023 16:39:09 +0000 (16:39 +0000)]
[XCore] Adapt threads.ll to opaque pointers.
Differential Revision: https://reviews.llvm.org/D144085
Max Kazantsev [Wed, 15 Feb 2023 10:35:45 +0000 (17:35 +0700)]
[SimpleLoopUnswitch] Fix overflowing frequencies case
When branch is so hot that sum of its frequencies overflows, we have an assertion
failure when trying to construct BranchProbability. Such cases are not interesting
as candidate for unswitching anyways (their branches are too hot), so reject them.
Carlos Galvez [Sun, 12 Feb 2023 10:58:37 +0000 (10:58 +0000)]
[clang-tidy][NFC] Remove ModernizeTidyModule::getModuleOptions
Most of the options stated there are duplicated already in
the implementation of each check as a default value for
each option.
The only place where this is not the case is the nullptr
check. Move the default option there instead. Only the
HICPP guidelines alias this modernize check, and there is
nothing in the documentation that suggests it should have
a different default value than the main modernize check.
Differential Revision: https://reviews.llvm.org/D143843
Samuel Parker [Wed, 15 Feb 2023 10:32:16 +0000 (10:32 +0000)]
[DAGCombine] Fold redundant select
Recommit
bbdf24357932b064f2aa18ea1356b474e0220dde.
Original commit message:
If a chain of two selects share a true/false value and are controlled
by two setcc nodes, that are never both true, we can fold away one of
the selects. So, the following:
(select (setcc X, const0, eq), Y,
(select (setcc X, const1, eq), Z, Y))
Can be combined to:
select (setcc X, const1, eq) Z, Y
Differential Revision: https://reviews.llvm.org/D142535
Guillaume Chatelet [Mon, 13 Feb 2023 14:52:26 +0000 (14:52 +0000)]
[libc][NFC] Make tuning macros start with LIBC_COPT_
Rename preprocessor definitions that control tuning of llvm libc.
Differential Revision: https://reviews.llvm.org/D143913
Matt Devereau [Wed, 15 Feb 2023 09:39:15 +0000 (09:39 +0000)]
Add InstSimplify tests for comparisons between known constants
Stefan Gränitz [Tue, 14 Feb 2023 14:08:44 +0000 (15:08 +0100)]
[JITLink] Drop const qualifier from argument to ELFLinkGraphBuilder's ctors (NFC)
These values are moved into the base class constructors, so the `const` doesn't make any sense. Turns out, I accidentally introduced it myself with
2ed91da0f1f3 and since than it spread by copy/paste.
Stefan Gränitz [Tue, 14 Feb 2023 14:01:51 +0000 (15:01 +0100)]
[JITLink] Fix whitespace in debug dumps (NFC)
Benjamin Kramer [Wed, 15 Feb 2023 09:33:49 +0000 (10:33 +0100)]
[bazel][mlir][examples] Add missing dependency for
72429a42ac33564fa82449d99dc234da32a05498
Matt Arsenault [Wed, 15 Feb 2023 09:27:09 +0000 (05:27 -0400)]
Revert "InstCombine: Fold is.fpclass(x, fcZero) to fcmp oeq 0"
This reverts commit
df78976d023a6b7fcf64bc695261b7b402fcede0.
I pushed the wrong patch
Valentin Clement [Wed, 15 Feb 2023 09:20:21 +0000 (10:20 +0100)]
[flang] Update intrinsic types to unlimited polymorphic form
This patch updates the code added in D143888 to avoid
overwriting some part of the types when updating it
for unlimited polymorphic types.
Reviewed By: jeanPerier, PeteSteinfeld
Differential Revision: https://reviews.llvm.org/D143995
Matt Arsenault [Mon, 13 Feb 2023 18:27:29 +0000 (14:27 -0400)]
AMDGPU: Override getNegatedExpression constant handling
Ignore the multiple use heuristics of the default
implementation, and report cost based on inline immediates. This
is mostly interesting for -0 vs. 0. Gets a few small improvements.
fneg_fadd_0_f16 is a small regression. We could probably avoid this
if we handled folding fneg into div_fixup.
Matt Arsenault [Mon, 13 Feb 2023 11:42:47 +0000 (07:42 -0400)]
AMDGPU: Refactor isConstantCostlierToNegate
Matt Arsenault [Fri, 11 Nov 2022 00:15:34 +0000 (16:15 -0800)]
InstCombine: Fold is.fpclass(x, fcZero) to fcmp oeq 0
This requires the denormal mode to definitively be IEEE handling.
Matthias Springer [Wed, 15 Feb 2023 08:51:42 +0000 (09:51 +0100)]
[mlir][bufferization] Add restrict and writable attrs to to_tensor
`restrict` is similar to the C++ restrict keyword. Results of `to_tensor` that have the `restrict` attribute are guaranteed to not alias any other `to_tensor` result (after bufferization).
Note: Since `to_memref` ops are not supported by One-Shot Bufferize and all bufferizable ops follow DPS rules (i.e., the buffer of the result is the buffer of an operand or an alias thereof), the buffer of a `to_tensor` op that has the `restrict` attribute is always an entirely "new" buffer that is not aliasing with the future buffer of any tensor value in the entire program. This makes such `to_tensor` ops "safe" from a bufferization perspective; they cannot cause RaW conflicts.
Differential Revision: https://reviews.llvm.org/D144021
Sander de Smalen [Tue, 14 Feb 2023 14:08:02 +0000 (14:08 +0000)]
[SROA] NFC: Look at TypeStoreSize scalable property, rather than at type directly.
Some places in the code have checks for isa<ScalableVectorType> and use
that to bail out of the code. It's also possible to look directly at the
allocated type-size and check if the size is scalable. This means it's
possible to also support other scalable types that are not vectors (i.e.
TargetExtType).
This is split out from D136861.
Meghan Denny [Wed, 15 Feb 2023 08:47:32 +0000 (09:47 +0100)]
[llvm-c] Add C API methods to match 64bit ArrayType C++ API signatures
Fixes https://github.com/llvm/llvm-project/issues/56496.
As mentioned in the issue, new functions LLVMArrayType2 and
LLVMGetArrayLength2 are created so as to not break the old API.
The old methods are then marked as deprecated and callers are
updated.
Differential Revision: https://reviews.llvm.org/D143700
Andrzej Warzynski [Thu, 26 Jan 2023 10:18:44 +0000 (10:18 +0000)]
[mlir] Add vectorize_nd_extract attribute to the masked_vectorize Op
This patch simply adds `vectorize_nd_extract` (that's currently only
used for the `vectorize` Op) to `masked_vectorize`. A test is added to
verify that it works as expected - it prevents the masked vectorisation
of `tensor.extract`, which is currently not supported.
Differential Revision: https://reviews.llvm.org/D142634
Max Kazantsev [Wed, 15 Feb 2023 08:34:24 +0000 (15:34 +0700)]
[Metarenamer] Use 'inst' as default name for instructions
Currently we use 'tmp', which is also a keyword for FileCheck. It leads to this
annoying warning whenever a script for auto-generation of checks is used.
It is especially annoying that it happens to every test affected by metarenamer.
Just use another prefix for metarenamed names to avoid this.
Differential Revision: https://reviews.llvm.org/D144001
Reviewed By: nikic
Balázs Kéri [Wed, 15 Feb 2023 07:54:20 +0000 (08:54 +0100)]
[clang][analyzer] Make messages of StdCLibraryFunctionsChecker user-friendly
Warnings and notes of checker alpha.unix.StdLibraryFunctionArgs are
improved. Previously one warning and one note was emitted for every
finding, now one warning is emitted only that contains a detailed
description of the found issue.
Reviewed By: Szelethus
Differential Revision: https://reviews.llvm.org/D143194
Fangrui Song [Wed, 15 Feb 2023 07:53:39 +0000 (23:53 -0800)]
[LoopFuse] Remove legacy pass
Following recent changes to remove non-core legacy passes.
Lorenzo Chelini [Mon, 13 Feb 2023 13:44:03 +0000 (14:44 +0100)]
[MLIR][Tensor] Introduce a pattern to propagate through `tensor.pad`
Introduce a pattern to 'push down' a `tensor.unpack` through a
`tensor.pad`. The propagation happens if the unpack does not touch the
padded dimensions.
Reviewed By: hanchung
Differential Revision: https://reviews.llvm.org/D143907
Craig Topper [Wed, 15 Feb 2023 07:43:17 +0000 (23:43 -0800)]
[InstCombine] Fix InstCombinerImpl::foldICmpMulConstant for nsw and nuw mul with unsigned compare.
If we have both an nsw and nuw flag, we would see the nsw flag
first and only handle signed comparisons.
This patch ignores the nsw flag if the comparison isn't signed.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D143766
Adrian Kuegel [Wed, 15 Feb 2023 07:30:57 +0000 (08:30 +0100)]
[mlir][Bazel] Add missing dependencies for
72429a42ac33564fa82449d99dc234da32a05498
Ingo Müller [Tue, 24 Jan 2023 15:42:38 +0000 (15:42 +0000)]
[mlir] Add RewriterBase::replaceAllUsesWith for Blocks.
When changing IR in a RewriterPattern, all changes must go through the
rewriter. There are several convenience functions in RewriterBase that
help with high-level modifications, such as replaceAllUsesWith for
Values, but there is currently none to do the same task for Blocks.
Reviewed By: mehdi_amini, ingomueller-net
Differential Revision: https://reviews.llvm.org/D142525
Mehdi Amini [Wed, 15 Feb 2023 05:15:30 +0000 (21:15 -0800)]
Add a pass that builds a debug info scope for LLVMFuncOp (adding a DISubprogramAttr)
This may be seen as a hack, but it allows for any piece of MLIR to be able
to end up with DWARF debug info through LLVM.
Assuming the operations in the function have location such as FileLineCol,
this provides backtraces with line tables and allows to step in a debugger.
That makes this pass a perfect companion to -snapshot-op-locations
It was also the default behavior of MLIR to LLVM IR translation until MLIR
got support for proper debug info attributes.
Differential Revision: https://reviews.llvm.org/D144069
Adrian Kuegel [Wed, 15 Feb 2023 06:36:00 +0000 (07:36 +0100)]
[mlir][Bazel] Add dependency needed for
e9b82a5c
Diego Caballero [Wed, 15 Feb 2023 06:01:42 +0000 (06:01 +0000)]
[mlir][Vector] Enable masking for static shapes
Support for masking static shapes was already implemented in the past
but not enabled so this patch is just removing a pre-condition check and
adding some tests with static shapes.
Reviewed By: ThomasRaoux
Differential Revision: https://reviews.llvm.org/D143937
Diego Caballero [Wed, 15 Feb 2023 06:00:12 +0000 (06:00 +0000)]
[mlir][Vector] Add support for masked vector gather ops
This patch adds support for masked vector.gather ops using the
vector.mask representation. It includes the implementation of the
MaskableOpInterface, Linalg vectorizer support and lowering to LLVM.
Reviewed By: ThomasRaoux
Differential Revision: https://reviews.llvm.org/D143939
Diego Caballero [Wed, 15 Feb 2023 05:46:15 +0000 (05:46 +0000)]
[mlir][Vector] Add support for masked vector.contract
This patch adds support for masking vector.contract ops with the
vector.mask approach. This also includes the lowering of vector.contract
through the vector.outerproduct path to LLVM. For now, this only adds
support for one of the many potential flavors of
vector.contract/vector.outerproduct but unsupported cases will fail
gratefully.
Reviewed By: ThomasRaoux
Differential Revision: https://reviews.llvm.org/D143965
Diego Caballero [Mon, 13 Feb 2023 19:31:23 +0000 (19:31 +0000)]
[mlir][Vector] Add LLVM lowering for masked reductions
This patch adds the conversion patterns to lower masked reduction
operations to the corresponding vp intrinsics in LLVM.
Reviewed By: ftynse
Differential Revision: https://reviews.llvm.org/D142177
Fangrui Song [Wed, 15 Feb 2023 05:40:45 +0000 (21:40 -0800)]
[LoopVersioningLICM] Remove legacy pass
Following recent changes to remove non-core features of the legacy PM/optimization pipeline.
Kazu Hirata [Wed, 15 Feb 2023 04:44:15 +0000 (20:44 -0800)]
[mlir] Use std::optional instead of llvm::Optional (NFC)
This is part of an effort to migrate from llvm::Optional to
std::optional:
https://discourse.llvm.org/t/deprecating-llvm-optional-x-hasvalue-getvalue-getvalueor/63716
Kazu Hirata [Wed, 15 Feb 2023 04:33:36 +0000 (20:33 -0800)]
[mlir] Use std::optional instead of llvm::Optional (NFC)
This is part of an effort to migrate from llvm::Optional to
std::optional:
https://discourse.llvm.org/t/deprecating-llvm-optional-x-hasvalue-getvalue-getvalueor/63716
Kazu Hirata [Wed, 15 Feb 2023 04:27:41 +0000 (20:27 -0800)]
[llvm] Use std::optional instead of llvm::Optional (NFC)
This is part of an effort to migrate from llvm::Optional to
std::optional:
https://discourse.llvm.org/t/deprecating-llvm-optional-x-hasvalue-getvalue-getvalueor/63716
Jie Fu [Wed, 15 Feb 2023 03:56:40 +0000 (11:56 +0800)]
[mlir] Fix -Wsign-compare in SparseTensorRewriting.cpp and Sparsification.cpp (NFC)
/home/jiefu/llvm-project/mlir/lib/Dialect/SparseTensor/Transforms/Sparsification.cpp:279:33: error: comparison of integers of different signs: 'int64_t' (aka 'long') and 'const mlir::sparse_tensor::Level' (aka 'const unsigned long') [-Werror,-Wsign-compare]
assert(env.op().getRank(&t) == lvlRank);
~~~~~~~~~~~~~~~~~~~~ ^ ~~~~~~~
/usr/include/assert.h:93:27: note: expanded from macro 'assert'
(static_cast <bool> (expr) \
^~~~
1 error generated.
/home/jiefu/llvm-project/mlir/lib/Dialect/SparseTensor/Transforms/SparseTensorRewriting.cpp:788:29: error: comparison of integers of different signs: 'int64_t' (aka 'long') and 'const mlir::sparse_tensor::Dimension' (aka 'const unsigned long') [-Werror,-Wsign-compare]
assert(srcRTT.getRank() == dimRank);
~~~~~~~~~~~~~~~~ ^ ~~~~~~~
/usr/include/assert.h:93:27: note: expanded from macro 'assert'
(static_cast <bool> (expr) \
^~~~
/home/jiefu/llvm-project/mlir/lib/Dialect/SparseTensor/Transforms/SparseTensorRewriting.cpp:810:31: error: comparison of integers of different signs: 'int64_t' (aka 'long') and 'const mlir::sparse_tensor::Dimension' (aka 'const unsigned long') [-Werror,-Wsign-compare]
assert(srcRTT.getRank() == dimRank);
~~~~~~~~~~~~~~~~ ^ ~~~~~~~
/usr/include/assert.h:93:27: note: expanded from macro 'assert'
(static_cast <bool> (expr) \
^~~~
2 errors generated.
LLVM GN Syncbot [Wed, 15 Feb 2023 03:51:14 +0000 (03:51 +0000)]
[gn build] Port
f1c4241fb6e5
Piyou Chen [Wed, 15 Feb 2023 03:42:26 +0000 (19:42 -0800)]
[RISCV] Add new pass to transform undef to pseudo for vector values.
RISC-V vector instruction has register overlapping constraint for certain
instructions, and will cause illegal instruction trap if violated, we use
early clobber to model this constraint, but it can't prevent register allocator
allocated same or overlapped if the input register is undef value, so convert
IMPLICIT_DEF to temporary pseudo could prevent that happen, it's not best way
to resolve this. Ideally we should model the constraint right, but before we
model the constraint right, it's the approach to prevent that happen.
See also: https://github.com/llvm/llvm-project/issues/50157
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D129735
Piyou Chen [Wed, 15 Feb 2023 03:29:59 +0000 (19:29 -0800)]
[RISCV] precommit test for D129735
Reviewed By: kito-cheng
Differential Revision: https://reviews.llvm.org/D137763
Chuanqi Xu [Wed, 15 Feb 2023 03:29:29 +0000 (11:29 +0800)]
[NFC] Refactor ModuleDeclStateTest to make it not dependent on Frontend
Required in https://reviews.llvm.org/D137526. And it is indeed odd that
LexTest depends on ClangFrontend.
Lang Hames [Wed, 15 Feb 2023 03:13:41 +0000 (19:13 -0800)]
[ORC] Add ELFNixPlatform::Create overload -- Pass ORC runtime as def generator.
The existing Create method took a path to the ORC runtime and created a
StaticLibraryDefinitionGenerator for it. The new overload takes a
std::unique_ptr<DefinitionGenerator> directly instead. This provides more
flexibility when constructing MachOPlatforms. E.g. The runtime archive can be
embedded in a special section in the ORC controller executable or library,
rather than being on-disk.
This is the ELFNixPlatform equivalent of the MachOPlatform change in
be2fc577c38.
wren romano [Wed, 15 Feb 2023 02:20:45 +0000 (18:20 -0800)]
[mlir][sparse] Factoring out SparseTensorType class
This change adds a new `SparseTensorType` class for making the "dim" vs "lvl" distinction more overt, and for abstracting over the differences between sparse-tensors and dense-tensors. In addition, this change also adds new type aliases `Dimension`, `Level`, and `FieldIndex` to make code more self-documenting.
Although the diff is very large, the majority of the changes are mechanical in nature (e.g., changing types to use the new aliases, updating variable names to match, etc). Along the way I also made many variables `const` when they could be; the majority of which required only adding the keyword. A few places had conditional definitions of these variables, requiring actual code changes; however, that was only done when the overall change was extremely local and easy to extract. All these changes are included in the current patch only because it would be too onerous to split them off into a separate patch.
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D143800
Chuanqi Xu [Wed, 15 Feb 2023 02:59:07 +0000 (10:59 +0800)]
[C++20] [Modules] [ClangScanDeps] Ensure that we can mix the use of and clang modules
Add a test to ensure that the clang-scan-deps won't crash due to the
unexpected use of clang modules.
Owen Pan [Wed, 15 Feb 2023 02:54:36 +0000 (18:54 -0800)]
[clang-format][NFC] Reformat clang/tools/clang-format/fuzzer/
Peiming Liu [Wed, 15 Feb 2023 02:18:54 +0000 (02:18 +0000)]
[mlir][sparse] fix a bug in UnpackOp converter.
UnpackOp Converter used to create reallocOp unconditionally, but it might cause issue when the requested memory size is smaller than the actually storage.
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D144065
chenglin.bi [Wed, 15 Feb 2023 02:33:54 +0000 (10:33 +0800)]
[InstCombine] fold icmp of the sum of ext bool based on limited range
For the pattern `(zext i1 X) + (sext i1 Y)`, the constant range is [-1, 1].
We can simplify the pattern by logical operations. Like:
```
(zext i1 X) + (sext i1 Y) == -1 --> ~X & Y
(zext i1 X) + (sext i1 Y) == 0 --> ~(X ^ Y)
(zext i1 X) + (sext i1 Y) == 1 --> X & ~Y
```
And other predicates can the combination of these results:
```
(zext i1 X) + (sext i1 Y)) != -1 --> X | ~Y
(zext i1 X) + (sext i1 Y)) s> -1 --> X | ~Y
(zext i1 X) + (sext i1 Y)) u< -1 --> X | ~Y
(zext i1 X) + (sext i1 Y)) s> 0 --> X & ~Y
(zext i1 X) + (sext i1 Y)) s< 0 --> ~X & Y
(zext i1 X) + (sext i1 Y)) != 1 --> ~X | Y
(zext i1 X) + (sext i1 Y)) s< 1 --> ~X | Y
(zext i1 X) + (sext i1 Y)) u> 1 --> ~X & Y
```
All alive proofs:
https://alive2.llvm.org/ce/z/KmgDpF
https://alive2.llvm.org/ce/z/fLwWa9
https://alive2.llvm.org/ce/z/ZKQn2P
Fix: https://github.com/llvm/llvm-project/issues/59666
Reviewed By: spatel
Differential Revision: https://reviews.llvm.org/D143373
Liren Peng [Wed, 15 Feb 2023 02:18:41 +0000 (02:18 +0000)]
[SeparateConstOffsetFromGEP] Fix: `b - a` matched `a - b` during reuniteExts
During the SeparateConstOffsetFromGEP pass, a - b and b - a will be
considered equivalent in some instances.
An example- the IR contains:
BB1:
%add = add %a, 511
br label %BB2
BB2:
%sub2 = sub %b, %a
br label %BB3
BB3:
%sub1 = sub %add, %b
%gep = getelementptr float, ptr %p, %sub1
Step 1 in the SeparateConstOffsetFromGEP pass, after split constant index:
BB1:
%add = add %a, 511
br label %BB2
BB2:
%sub2 = sub %b, %a
br label %BB3
BB3:
%sub.t = sub %a, %b
%gep.base = getelementptr float, ptr %p, %sub.t
%gep = getelementptr float, ptr %gep.base, 511
Step 2, after reuniteExts:
BB1:
br label %BB2
BB2:
%sub2 = sub %b, %a
br label %BB3
BB3:
%gep.base = getelementptr float, ptr %p, %sub2
%gep = getelementptr float, ptr %gep.base, 511
Obviously, reuniteExts treated a - b and b - a as equivalent.
This patch fixes that.
Reviewed By: nikic, spatel
Differential Revision: https://reviews.llvm.org/D143542
chenglin.bi [Wed, 15 Feb 2023 02:31:27 +0000 (10:31 +0800)]
[Instcombine] Precommit tests for icmp range; NFC
Liren Peng [Wed, 15 Feb 2023 01:22:31 +0000 (01:22 +0000)]
[NFC][SeparateConstOffsetFromGEP] Added flag `lower-gep`
We need such a flag to check whether the transformation is correct if
LowerGEP was enabled.
Reviewed By: nikic, arsenm, spatel
Differential Revision: https://reviews.llvm.org/D143980
Fangrui Song [Wed, 15 Feb 2023 01:42:48 +0000 (17:42 -0800)]
eopXD [Thu, 9 Feb 2023 18:14:59 +0000 (10:14 -0800)]
[Clang][RISCV] Guard vector int64, float32, float64 with semantic analysis
Depends on D143657
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D143665
Jie Fu [Wed, 15 Feb 2023 01:31:16 +0000 (09:31 +0800)]
[mlir] Fix two build warnings in VectorToGPU (NFC)
In file included from /data/llvm-project/mlir/lib/Conversion/VectorToGPU/VectorToGPU.cpp:13:
/data/llvm-project/mlir/include/mlir/Conversion/VectorToGPU/VectorToGPU.h:15:1: error: class 'LogicalResult' was previously declared as a struct; this is valid, but may result in linker errors under the Microsoft C++ ABI [-Werror,-Wmismatched-tags]
class LogicalResult;
^
/data/llvm-project/mlir/include/mlir/Support/LogicalResult.h:26:22: note: previous use is here
struct [[nodiscard]] LogicalResult {
^
/data/llvm-project/mlir/include/mlir/Conversion/VectorToGPU/VectorToGPU.h:15:1: note: did you mean struct here?
class LogicalResult;
^~~~~
struct
/data/llvm-project/mlir/lib/Conversion/VectorToGPU/VectorToGPU.cpp:724:5: error: ignoring return value of function declared with 'nodiscard' attribute [-Werror,-Wunused-result]
rewriter.notifyMatchFailure(
^~~~~~~~~~~~~~~~~~~~~~~~~~~
2 errors generated.
Arthur Eubanks [Wed, 15 Feb 2023 01:33:15 +0000 (17:33 -0800)]
[gn build] Add missing count dependency for check-asan
Chia-hung Duan [Wed, 15 Feb 2023 01:23:13 +0000 (01:23 +0000)]
[scudo] Update ring buffer test to make it accept zero size
allocation ring buffer is allowed to be zero. Update the logic in the
test so that on the platform that disables it won't fail this case.
Reviewed By: fmayer
Differential Revision: https://reviews.llvm.org/D144055
Fangrui Song [Wed, 15 Feb 2023 01:23:15 +0000 (17:23 -0800)]
[LoopUnrollAndJam] Remove legacy pass
Following recent changes to remove non-core features of the legacy PM/optimization pipeline.