Roman Lebedev [Tue, 13 Dec 2022 13:13:34 +0000 (16:13 +0300)]
[NFC][CMake] Explicitly link to MC when needed
These targets use `MCInst`, but don't explicitly link
to the library providing it (MC), and just rely on it
being pulled transitively through e.g. MCDisassembler,
but that only pulls includes, and does not link to it.
Case in point, when i add explicit destructor to `MCInst`,
defined in `.cpp`, these targets were failing to link.
Matt Arsenault [Thu, 17 Nov 2022 05:26:29 +0000 (21:26 -0800)]
InstCombine: Fold negations of is_fpclass intrinsics
Can invert the result by inverting the test mask.
Matt Arsenault [Tue, 13 Dec 2022 13:01:48 +0000 (08:01 -0500)]
InstCombine: Regenerate checks
update_test_checks has changed what it emits for function labels,
so keep this noise out of a real output change.
Aliia Khasanova [Tue, 13 Dec 2022 12:35:24 +0000 (13:35 +0100)]
Remove sentinel argument from dispatchIndexOpFoldResults.
Post clean-up after merger of kDynamicSize and kDynamicStrideOrOffset.
Differential Revision: https://reviews.llvm.org/D139929
Guillaume Chatelet [Fri, 9 Dec 2022 14:30:40 +0000 (14:30 +0000)]
[libc] Add compiler, builtin and feature detection
This is a first step to support GCC. This patch adds support for builtin and feature detection.
Differential Revision: https://reviews.llvm.org/D139712
OCHyams [Mon, 12 Dec 2022 09:18:59 +0000 (09:18 +0000)]
Reapply [Assignment Tracking][13/*] Account for assignment tracking in SROA
The Assignment Tracking debug-info feature is outlined in this RFC:
https://discourse.llvm.org/t/
rfc-assignment-tracking-a-better-way-of-specifying-variable-locations-in-ir
Split dbg.assign intrinsics into fragments similarly to what SROA already does
for dbg.declares, except that there's many more intrinsics to split. The
function migrateDebugInfo generates new dbg.assigns intrinsic for each part of
a split store.
Reviewed By: jmorse
Differential Revision: https://reviews.llvm.org/D133296
Alexey Baturo [Tue, 13 Dec 2022 12:43:49 +0000 (15:43 +0300)]
re-land [RISC-V][HWASAN] Support tagging global variables for RISC-V HWASAN
Now with fix to limit added tagged-globals.ll to risc-v platform
--
[RISC-V][HWASAN] Support tagging global variables for RISC-V HWASAN
Reviewed by: luismarques
Differential Revision: https://reviews.llvm.org/D132995
Roman Lebedev [Tue, 13 Dec 2022 12:22:00 +0000 (15:22 +0300)]
[exegesis] Benchmark: provide optional progress meter / ETA
Now that `--opcode-index=-1` is mostly stable,
and i can migrate off of my custom tooling that emulated it,
there comes a bit of confusion as to the status of the run.
It is normal for the single all-opcode run to take ~3 minutes,
and it's a bit more than one can be comfortable with,
without having some sort of visual indication of the progress.
Thus, i present:
```
$ ./bin/llvm-exegesis -mode=inverse_throughput --opcode-index=-1 --benchmarks-file=/dev/null --dump-object-to-disk=0 --measurements-print-progress --skip-measurements
<...>
XAM_Fp80: unsupported opcode: pseudo instruction
XBEGIN: Unsupported opcode: isPseudo/usesCustomInserter
XBEGIN_2: Unsupported opcode: isBranch/isIndirectBranch
XBEGIN_4: Unsupported opcode: isBranch/isIndirectBranch
XCH_F: unsupported second-form X87 instruction
Processing... 1%, ETA 02:10
Processing... 2%, ETA 02:03
Processing... 3%, ETA 02:00
Processing... 4%, ETA 01:57
Processing... 5%, ETA 01:54
Processing... 6%, ETA 01:53
Processing... 7%, ETA 01:51
Processing... 8%, ETA 01:50
Processing... 9%, ETA 01:49
Processing... 10%, ETA 01:48
Processing... 11%, ETA 01:46
Processing... 12%, ETA 01:45
Processing... 13%, ETA 01:44
Processing... 14%, ETA 01:43
Processing... 15%, ETA 01:42
Processing... 16%, ETA 01:42
Processing... 17%, ETA 01:41
```
As usual, the ETA estimation is statically-insignificant,
and is a lie/does not converge at least until 50% through.
It would be nice to have an actual progress indicator like in LIT,
but i'm not sure we have such a luxury in C++ form in LLVM codebase already.
Reviewed By: courbet
Differential Revision: https://reviews.llvm.org/D139797
Guillaume Chatelet [Tue, 13 Dec 2022 10:31:00 +0000 (10:31 +0000)]
[rereland][Alignment][NFC] Remove access to deprecated GlobalObject::getAlignment from llvm
Differential Revision: https://reviews.llvm.org/D139836
Alexey Baturo [Tue, 13 Dec 2022 12:17:27 +0000 (15:17 +0300)]
Revert "[RISC-V][HWASAN] Support tagging global variables for RISC-V HWASAN"
This reverts commit
11937ca5642216a67e021e69fc824f709267bada.
gonglingqin [Tue, 13 Dec 2022 11:44:06 +0000 (19:44 +0800)]
[OpenMP] Skip extra blank line when parsing /proc/cpuinfo on LoongArch64
This fixes the following test cases:
* affinity/kmp-affinity.c
* affinity/kmp-hw-subset.c
* affinity/omp-places.c
Differential Revision: https://reviews.llvm.org/D139802
Nikita Popov [Tue, 13 Dec 2022 12:06:09 +0000 (13:06 +0100)]
[Tests] Convert Other tests to opaque pointers (NFC)
Nikita Popov [Tue, 13 Dec 2022 11:36:48 +0000 (12:36 +0100)]
[Bitcode] Convert test to opaque pointers (NFC)
Alexey Baturo [Tue, 30 Aug 2022 18:05:24 +0000 (21:05 +0300)]
[RISC-V][HWASAN] Support tagging global variables for RISC-V HWASAN
Reviewed by: luismarques
Differential Revision: https://reviews.llvm.org/D132995
Dmitry Preobrazhensky [Tue, 13 Dec 2022 11:48:36 +0000 (14:48 +0300)]
[AMDGPU][GFX1030][DOC][NFC] Update assembler syntax description
Summary of changes:
- Enable register tuples with 9, 10, 11 and 12 registers (https://reviews.llvm.org/D138205).
- Enable abs and neg modifiers for v_cndmask_b32_dpp (https://reviews.llvm.org/D135900).
- Enable literal operands for permlane16/permlanex16 (https://reviews.llvm.org/D137332).
- Enable omod modifiers for v_max3_f16, v_min3_f16, etc. (https://reviews.llvm.org/D139469).
- Correct v_mov_b32_sdwa (it does not support abs and neg input modifiers yet).
- Enable tfe modifier for MUBUF loads (https://reviews.llvm.org/D137783).
- Enable image_gather4h (https://reviews.llvm.org/D130764).
- Minor corrections and improvements.
Nikita Popov [Tue, 13 Dec 2022 11:46:09 +0000 (12:46 +0100)]
[AsmWriter] Fix printing of atomicrmw xchg with pointer op
atomicrmw always needs to print all types, even if the xchg value
type happens to be the same as the pointer operand type. This
couldn't occur prior to opaque pointers.
Dmitry Preobrazhensky [Tue, 13 Dec 2022 11:44:02 +0000 (14:44 +0300)]
[AMDGPU][GFX1013][DOC][NFC] Update assembler syntax description
Summary of changes:
- Enable register tuples with 9, 10, 11 and 12 registers (https://reviews.llvm.org/D138205).
Dmitry Preobrazhensky [Tue, 13 Dec 2022 11:36:43 +0000 (14:36 +0300)]
[AMDGPU][GFX10][DOC][NFC] Update assembler syntax description
Summary of changes:
- Enable register tuples with 9, 10, 11 and 12 registers (https://reviews.llvm.org/D138205).
- Enable abs and neg modifiers for v_cndmask_b32_dpp (https://reviews.llvm.org/D135900).
- Enable literal operands for permlane16/permlanex16 (https://reviews.llvm.org/D137332).
- Enable omod modifiers for v_max3_f16, v_min3_f16, etc. (https://reviews.llvm.org/D139469).
- Correct v_mov_b32_sdwa (it does not support abs and neg input modifiers yet).
- Enable tfe modifier for MUBUF loads (https://reviews.llvm.org/D137783).
- Enable image_gather4h (https://reviews.llvm.org/D130764).
- Minor corrections and improvements.
Dmitry Preobrazhensky [Tue, 13 Dec 2022 11:26:34 +0000 (14:26 +0300)]
[AMDGPU][GFX940][DOC][NFC] Update assembler syntax description
Summary of changes:
- Enable VOP3 variants of dot2c/dot4c/dot8c instructions (https://reviews.llvm.org/D138494).
- Enable abs and neg modifiers for v_cndmask_b32 (https://reviews.llvm.org/D135900).
- Enable omod modifiers for v_max3_f16, v_min3_f16, etc. (https://reviews.llvm.org/D139469).
- Correct v_mov_b32_sdwa (it does not support abs and neg input modifiers yet).
- Enable abs and neg modifiers for v_dot2c_f32_f16_dpp.
- Minor corrections and improvements.
Nikita Popov [Tue, 13 Dec 2022 11:19:35 +0000 (12:19 +0100)]
[Bitcode] Convert test to opaque pointers (NFC)
Dmitry Preobrazhensky [Tue, 13 Dec 2022 11:18:20 +0000 (14:18 +0300)]
[AMDGPU][GFX90A][DOC][NFC] Update assembler syntax description
Summary of changes:
- Enable register tuples with 9, 10, 11 and 12 registers (https://reviews.llvm.org/D138205).
- Enable VOP3 variants of dot2c/dot4c/dot8c instructions (https://reviews.llvm.org/D138494).
- Enable omod modifiers for v_max3_f16, v_min3_f16, etc. (https://reviews.llvm.org/D139469).
- Enable abs and neg modifiers for v_cndmask_b32 (https://reviews.llvm.org/D135900).
- Correct v_mov_b32_sdwa (it does not support abs and neg input modifiers yet).
- Enable abs and neg modifiers for v_dot2c_f32_f16_dpp.
- Minor corrections and improvements.
Haojian Wu [Tue, 13 Dec 2022 11:15:39 +0000 (12:15 +0100)]
[clangd] Fix a semantic highlighting crash on dependent code.
Daniel Grumberg [Wed, 26 Oct 2022 17:23:37 +0000 (18:23 +0100)]
[clang][ExtractAPI] Add support for single symbol SGF and libclang support
This is mainly adding an entry point to `SymbolGraphSerializer` at
`serializeSingleSymbolSGF` and exposing the necessary data to make this
possible. Additionaly there are some changes to how symbol kinds and
path components are serialized to make the usage more ergonomic in
`serializeSingleSymbolSGF`.
On the libclang side this introduces APIs to:
- create an APISet from a TU
- dispose of an APISet
- query an APISet for a single symbol SGF for a given USR.
- generate a single symbol SGF for a given CXCursor, this only traverses
the necessary AST nodes to construct the result as oppposed as going
through the entire AST.
Differential Revision: https://reviews.llvm.org/D139115
Nikita Popov [Tue, 13 Dec 2022 11:15:55 +0000 (12:15 +0100)]
[Bitcode] Update test to use opaque pointers (NFC)
Dmitry Preobrazhensky [Tue, 13 Dec 2022 11:12:09 +0000 (14:12 +0300)]
[AMDGPU][GFX908][DOC][NFC] Update assembler syntax description
Summary of changes:
- Enable VOP3 variants of dot2c/dot4c/dot8c instructions (https://reviews.llvm.org/D138494).
- Enable abs and neg modifiers for v_dot2c_f32_f16_dpp.
- Minor corrections and improvements.
Dmitry Preobrazhensky [Tue, 13 Dec 2022 11:01:17 +0000 (14:01 +0300)]
[AMDGPU][GFX9][DOC][NFC] Update assembler syntax description
Summary of changes:
- Enable register tuples with 9, 10, 11 and 12 registers (https://reviews.llvm.org/D138205).
- Enable omod modifiers for v_max3_f16, v_min3_f16, etc. (https://reviews.llvm.org/D139469).
- Enable abs and neg modifiers for v_cndmask_b32 (https://reviews.llvm.org/D135900).
- Correct v_mov_b32_sdwa (it does not support abs and neg input modifiers yet).
- Enable tfe modifier for MUBUF loads (https://reviews.llvm.org/D137783).
- Enable image_gather4h (https://reviews.llvm.org/D130764).
- Minor corrections and improvements.
Dmitry Preobrazhensky [Tue, 13 Dec 2022 10:54:28 +0000 (13:54 +0300)]
[AMDGPU][GFX8][DOC][NFC] Update assembler syntax description
Summary of changes:
- Enable register tuples with 9, 10, 11 and 12 registers (https://reviews.llvm.org/D138205).
- Enable abs and neg modifiers for v_cndmask_b32 (https://reviews.llvm.org/D135900).
- Correct v_mov_b32_sdwa (it does not support abs and neg input modifiers yet).
- Enable tfe modifier for MUBUF loads (https://reviews.llvm.org/D137783).
- Minor corrections and improvements.
Dmitry Preobrazhensky [Tue, 13 Dec 2022 10:46:56 +0000 (13:46 +0300)]
[AMDGPU][GFX7][DOC][NFC] Update assembler syntax description
Summary of changes:
- Enable register tuples with 9, 10, 11 and 12 registers (https://reviews.llvm.org/D138205).
- Enable tfe modifier for MUBUF loads (https://reviews.llvm.org/D137783).
- Enable abs and neg modifiers for v_cndmask_b32_e64.
- Minor corrections and improvements.
Nikita Popov [Tue, 13 Dec 2022 10:48:14 +0000 (11:48 +0100)]
[Clang] Update some CUDA tests to opaque pointers (NFC)
Denis Antrushin [Mon, 28 Nov 2022 18:58:47 +0000 (01:58 +0700)]
[RS4GC] Add few tests for derived pointer rematerialization. NFC.
Precommit few tests for the upcoming 'rematerialize derived pointers
at uses' feature.
Reviewed By: skatkov
Differential Revision: https://reviews.llvm.org/D138911
Joshua Cao [Sat, 19 Nov 2022 18:25:30 +0000 (13:25 -0500)]
[CVP] Eliminate urem when LHS < RHS
Fol `X % Y -> X` when we can determine `X < Y` based on constant
range information.
Fixes https://github.com/llvm/llvm-project/issues/58408.
Differential Revision: https://reviews.llvm.org/D138360
Daniel Woodworth [Tue, 13 Dec 2022 10:35:17 +0000 (11:35 +0100)]
[IRBuilder] Fix CreateFDivFMF ignoring source FMF
This change fixes a small bug in IRBuilderBase::CreateFDivFMF introduced
by accident in rGf34dcf27637f which caused the builder's default fast math
flags to be used for the new fdiv instruction instead of the ones from
the source instruction. It also adds unit testing for the CreateF*FMF
family of functions to more easily catch similar bugs in the future.
Differential Revision: https://reviews.llvm.org/D139583
Nikita Popov [Tue, 13 Dec 2022 10:28:20 +0000 (11:28 +0100)]
[InstCombine] Support logical ops in foldAndOrOfICmpsWithConstEq() (NFC)
This is largely just for the sake of completeness. For logical ops,
this is mostly subsumed by foldSelectValueEquivalence() in a more
generic way. The only exception is vector support, as select
value equivalence folding currently doesn't support this for the
case where the replacement does not simplify.
Fangrui Song [Tue, 13 Dec 2022 10:14:09 +0000 (10:14 +0000)]
[DWARFLinker] llvm::Optional => std::optional
Fangrui Song [Tue, 13 Dec 2022 10:06:57 +0000 (10:06 +0000)]
[WindowsDriver] llvm::Optional => std::optional
Fangrui Song [Tue, 13 Dec 2022 10:04:35 +0000 (10:04 +0000)]
LLParser: llvm::Optional => std::optional
Rainer Orth [Tue, 13 Dec 2022 09:58:58 +0000 (10:58 +0100)]
Reland "[compiler-rt][test] Heed COMPILER_RT_DEBUG when compiling unittests"
When trying to debug some `compiler-rt` unittests, I initially had a hard
time because
- even in a `Debug` build one needs to set `COMPILER_RT_DEBUG` to get
debugging info for some of the code and
- even so the unittests used a hardcoded `-O2` which often makes debugging
impossible.
This patch addresses this by instead using `-O0` if `COMPILER_RT_DEBUG`.
Changes relative to the previous commit:
- Use `string(APPEND)` for `COMPILER_RT_TEST_COMPILER_CFLAGS`.
- Omit `-O3` from `COMPILER_RT_TEST_COMPILER_CFLAGS` in non-debug builds for now.
- Provide `__sanitizer::integral_constant<bool, true>::value` instantiation
for `sanitizer_type_traits_test.cpp` in debug builds.
- Disable subtests of `tsan/tests/unit/tsan_trace_test.cpp` that deadlock
in debug builds.
- `XFAIL` `tsan/Linux/check_memcpy.c` in debug builds.
Tested on `sparcv9-sun-solaris2.11`, `amd64-pc-solaris2.11`, and
`x86_64-pc-linux-gnu`.
Differential Revision: https://reviews.llvm.org/D91620
Nikita Popov [Tue, 13 Dec 2022 09:56:00 +0000 (10:56 +0100)]
[InstCombine] Add additional foldAndOrOfICmpsWithConstEq() tests (NFC)
Add logical variants of vector tests.
Krasimir Georgiev [Tue, 13 Dec 2022 09:54:26 +0000 (09:54 +0000)]
Revert "[Assignment Tracking][13/*] Account for assignment tracking in SROA"
This reverts commit
3bfba672afd52dfd5bde54dc8b67ec96275f9e15.
Temporary revert since this potentially causes
https://github.com/llvm/llvm-project/issues/59490.
Mehdi Amini [Tue, 13 Dec 2022 09:41:13 +0000 (09:41 +0000)]
[mlir][scf] Fixes IndexSwitchOp verifier crash
Fixes #59460
Martin Storsjö [Tue, 13 Dec 2022 09:36:12 +0000 (11:36 +0200)]
Revert "[AArch64][GlobalISel] Lower formal arguments of AAPCS & ms_abi variadic functions."
This reverts commit
56fd846f370adf16bea333b12637038ea2f3c225.
This commit regressed handling of functions with floats as arguments,
reproducible e.g. like this:
$ cat test.c
double func(double f) {
return f;
}
$ clang -target aarch64-windows -S -o - test.c -fno-asynchronous-unwind-tables
func:
sub sp, sp, #16
str x0, [sp, #8]
ldr d0, [sp, #8]
add sp, sp, #16
ret
Nikita Popov [Tue, 13 Dec 2022 09:02:46 +0000 (10:02 +0100)]
[InstCombine] Support logical ops in foldAndOrOfICmpEqZeroAndICmp()
If the and/or is logical and one of the operands only occurs on the
RHS, we need to freeze it: https://alive2.llvm.org/ce/z/vuMuE_
Denis Antrushin [Wed, 23 Nov 2022 15:31:41 +0000 (22:31 +0700)]
[RS4GC] Turn lambda into static function. NFC.
Extract `rematerializeChain()` lambda into static function.
We'll need it in upcoming patch to RS4GC pass.
There is small interface change: now reversal of `ChainToBase` is
performed within this function, not outside.
Still this is non-functional change.
Reviewed By: skatkov
Differential Revision: https://reviews.llvm.org/D138910
Nikita Popov [Tue, 13 Dec 2022 09:07:54 +0000 (10:07 +0100)]
[InstCombine] Add additional tests for foldAndOrOfICmpEqZeroAndICmp (NFC)
Adds logical variants of the commuted tests.
Fangrui Song [Tue, 13 Dec 2022 09:06:36 +0000 (09:06 +0000)]
[CodeGen] llvm::Optional => std::optional
Haojian Wu [Tue, 13 Dec 2022 08:55:58 +0000 (09:55 +0100)]
[clangd] Fix some header guard names, NFC
Per the LLVM code style, there should be no trailing `_` on the header
guard name.
Corentin Jabot [Sun, 23 Oct 2022 15:32:58 +0000 (17:32 +0200)]
Implement CWG2631
Implement https://cplusplus.github.io/CWG/issues/2631.html.
Immediate calls in default arguments and defaults members
are not evaluated.
Instead, we evaluate them when constructing a
`CXXDefaultArgExpr`/`BuildCXXDefaultInitExpr`.
The immediate calls are executed by doing a
transform on the initializing expression.
Note that lambdas are not considering subexpressions so
we do not need to transform them.
As a result of this patch, unused default member
initializers are not considered odr-used, and
errors about members binding to local variables
in an outer scope only surface at the point
where a constructor is defined.
Reviewed By: aaron.ballman, #clang-language-wg
Differential Revision: https://reviews.llvm.org/D136554
Reed [Tue, 13 Dec 2022 08:50:02 +0000 (09:50 +0100)]
Fix APFloat::toString on Float8E5M2 values.
Before, an APInt with value 10 was created, whose width was the significand width. But 10 cannot fit in Float8E5M2's significand.
Differential Revision: https://reviews.llvm.org/D138540
Nikita Popov [Tue, 13 Dec 2022 08:41:28 +0000 (09:41 +0100)]
[InstCombine] Handle logical op in simplifyRangeCheck() (PR59484)
We need to freeze to avoid propagating a potentially poison
upper bound (https://alive2.llvm.org/ce/z/MsD38k).
This resolves the existing TODO in the code.
Fixes https://github.com/llvm/llvm-project/issues/59484.
Mehdi Amini [Tue, 13 Dec 2022 08:48:32 +0000 (08:48 +0000)]
Add LLVMDialect as dependent for "llvm-legalize-for-export" pass
Fixes #59462
Valentin Clement [Tue, 13 Dec 2022 08:44:01 +0000 (09:44 +0100)]
[flang] Use input type when emboxing/reboxing polymorphic entities
When emboxing an entity to a polymorphic box, use the input type to
compute the type code and element size as the box type is too generic.
When reboxing a polymorphic box, get this information from the input
box.
Reviewed By: jeanPerier
Differential Revision: https://reviews.llvm.org/D139916
Fangrui Song [Tue, 13 Dec 2022 08:32:44 +0000 (08:32 +0000)]
[Transforms/Coroutines] llvm::Optional => std::optional
Fangrui Song [Tue, 13 Dec 2022 08:26:08 +0000 (08:26 +0000)]
[Transforms/InstCombine] llvm::Optional => std::optional
Fangrui Song [Tue, 13 Dec 2022 08:15:56 +0000 (08:15 +0000)]
[MemProf] llvm::Optional => std::optional
Nikita Popov [Tue, 13 Dec 2022 08:13:39 +0000 (09:13 +0100)]
Revert "[UpdateTestChecks] Match define for labels"
This reverts commit
a888825aeef8d6592c6cf5f4e5854cc39af49633.
This changes the default output of UTC, and as such introduces
spurious changes whenever existing tests are regenerated.
I've indicated in https://reviews.llvm.org/D139006#3989954 how
this can be implemented without causing test churn.
Fangrui Song [Tue, 13 Dec 2022 08:05:14 +0000 (08:05 +0000)]
[Transforms/Scalar] llvm::Optional => std::optional
Corentin Jabot [Mon, 28 Nov 2022 21:45:15 +0000 (22:45 +0100)]
[Clang] Implement CWG2640 Allow more characters in an n-char sequence
Reviewed By: #clang-language-wg, aaron.ballman, tahonermann
Differential Revision: https://reviews.llvm.org/D138861
esmeyi [Tue, 13 Dec 2022 07:44:49 +0000 (02:44 -0500)]
[NFC][PowerPC] Add tests for 64-bit constants that require 5 instructions to materialize.
Differential Revision: https://reviews.llvm.org/D139914
Mehdi Amini [Sat, 10 Dec 2022 10:28:26 +0000 (10:28 +0000)]
Apply clang-tidy fixes for readability-identifier-naming in AsyncToAsyncRuntime.cpp (NFC)
Mehdi Amini [Sat, 10 Dec 2022 10:27:47 +0000 (10:27 +0000)]
Apply clang-tidy fixes for performance-unnecessary-value-param in AsyncToAsyncRuntime.cpp (NFC)
Johannes Doerfert [Tue, 13 Dec 2022 06:08:28 +0000 (22:08 -0800)]
Revert "[OpenMP][NFCI] Remove effectively dead code in clang and the runtime"
This reverts commit
c1c8cbbf5f29257d084a23a2f6c4236c40b7afb9. One of the
tests seems to be flaky/non-deterministic.
Sameer Sahasrabuddhe [Tue, 13 Dec 2022 05:45:15 +0000 (11:15 +0530)]
[Clang][NFC] Prevent lit tests from matching substrings in current path
Alex Brachet [Tue, 13 Dec 2022 05:38:09 +0000 (05:38 +0000)]
[libc] Use correct type for atol
Johannes Doerfert [Tue, 13 Dec 2022 03:38:19 +0000 (19:38 -0800)]
[OpenMP][FIX] Ensure combing accesses does not violate invariants
Johannes Doerfert [Thu, 27 Oct 2022 23:53:17 +0000 (16:53 -0700)]
[OpenMP][NFCI] Remove effectively dead code in clang and the runtime
Johannes Doerfert [Tue, 4 Oct 2022 14:39:45 +0000 (07:39 -0700)]
[Attributor] Make non-side-effect inline asm be "no-call"
If we have inline asm with side effects we assume any function might be
called. For non-side-effect asm we now assume no function is called.
Blue Gaston [Thu, 8 Dec 2022 20:02:14 +0000 (12:02 -0800)]
[Sanitizers][CFG][arm64e] Fix test because -fsanitize-coverage=control-flow does not sign BB entry
-fsanitize-coverage=control-flow does not sign entries into basic blocks on arm64e. This test compares a local pointer to a function [signed] with the basic block pointer. Because the entry into the
basic block is unsigned the addresses being compared are signed and unsigned, causing the path never to be taken.
This is a "bandaid" to get this test passing. We strip the signed bits from the pointer to the local functions so that the comparisons pass.
Filed radar: rdar://
103042879 to note the behavior.
context: https://github.com/llvm/llvm-project/blob/main/llvm/lib/Transforms/Instrumentation/SanitizerCoverage.cpp#L1068
// blockaddress can not be used on function's entry block.
if (&BB == &F.getEntryBlock())
CFs.push_back((Constant *)IRB.CreatePointerCast(&F, IntptrPtrTy));
else
CFs.push_back((Constant *)IRB.CreatePointerCast(BlockAddress::get(&BB),
IntptrPtrTy));
BlockAddress::get is responsible for signing the pointer.
Because of:
https://reviews.llvm.org/D133157
rdar://
103042879
Differential Revision: https://reviews.llvm.org/D139661
Vasileios Porpodas [Mon, 12 Dec 2022 22:57:47 +0000 (14:57 -0800)]
[IR][NFC] Adds Function::insertBasicBlockAt() to replace things like F->getBasicBlockList().insert()
This is part of a series of patches that aim at making Function::getBasicBlockList() private.
Differential Revision: https://reviews.llvm.org/D139906
Matt Arsenault [Sun, 4 Dec 2022 16:36:11 +0000 (11:36 -0500)]
InstCombine: Fix metadata arguments blocking freeze combining
These are used for special arguments to intrinsics and don't make any
sense to consider for poisonness. Fixes not pushing freeze through
llvm.fptrunc.round.
92106641ae297c24877085e0357e8095aa7b43c9 made
isGuaranteedNotToBeUndefOrPoison return false for metadata arguments,
which doesn't entirely make sense. An alternate patch could switch
that to true, and try to filter out adding some pointless noundefs on
metadata arguments (I tried that, attributor breaks one case with a
llvm.dbg.value in it).
Matt Arsenault [Mon, 5 Dec 2022 03:55:57 +0000 (22:55 -0500)]
ValueTracking: Teach canCreateUndefOrPoison about FP ops
Probably could replace the switch by marking the intrinsic definitions
with NoUndef<RetIndex>.
Matt Arsenault [Sun, 4 Dec 2022 00:55:15 +0000 (19:55 -0500)]
InstCombine: Add baseline tests for freeze with some FP ops
wanglei [Tue, 13 Dec 2022 03:32:10 +0000 (11:32 +0800)]
[LoongArch] Add custom parser for atomic instructions' memory operand
In order to be compatible with the form of the atomic instruction in
GAS that accepts the fourth operand as 0 (i.e. `am* $rd, $rk, $rj, 0`),
we need to treat `$rj, 0` as one operand, but only print `$rj`.
For this, the number of result operands of inline assembly memory
operand `ZB` constraint is modified to 2 (reg + 0).
Restrictions on register usage in `am*` instructions have also been
adjusted. When `$rd` is equal to `$r0`, the instruction must be
considered legal, because of some special usage like `PseudoUNIMP`.
Reviewed By: SixWeining, xen0n
Differential Revision: https://reviews.llvm.org/D139303
Matt Arsenault [Sat, 26 Nov 2022 18:02:48 +0000 (13:02 -0500)]
ObjCARC: Update tests to use opaque pointers
escape.ll needed a simple manual check line update.
contract-storestrong.ll:test12 is kind of contrived now. The comment
says it's for testing bitcasts of pointers, which don't really matter
anymore. Leaves identity ptr to ptr bitcasts (which I thought were
illegal).
Matt Arsenault [Sat, 26 Nov 2022 19:07:43 +0000 (14:07 -0500)]
ObjCARC: Try to fix faulty tests
These were trying to check if there was not an llvm.objc call before a
closing "}", which presumably was intended to match the end of the
function. Really this was matching the closing } in "bitcast {}* %self
to i8*", which can't be what anyone intended. This broke after
converting the test to opaque pointer deleted this bitcast.
There are in fact @llvm.obj calls remaining in the function, so this
may indicate the transform this was intended to check is actually
broken. In @"\01-[A z]" (great test name), the first retain call seems
to move down to the printf. The second case, @"\01-[Top0 _getX]", has
no change.
Change the checks to what's produced and add FIXMES. Also change the }
checks to match only at the start of the line for the function end.
Matt Arsenault [Tue, 29 Nov 2022 15:26:00 +0000 (10:26 -0500)]
AMDGPU: Add sanity test if amdgcn.device.{init|fini} already exists
Matt Arsenault [Fri, 11 Nov 2022 00:15:34 +0000 (16:15 -0800)]
InstSimplify: Add basic folding of llvm.is.fpclass intrinsic
Copied from the existing llvm.amdgcn.class handling; eventually I will
fold that to the generic intrinsic when legal. The tests should
probably move into an instsimplify only test.
Matt Arsenault [Sat, 10 Dec 2022 02:18:29 +0000 (21:18 -0500)]
Verifier: Enforce value of llvm.is.fpclass test mask
As requested in D137811
Chuanqi Xu [Tue, 13 Dec 2022 02:09:59 +0000 (10:09 +0800)]
[Coroutines] Don't mark the parameter attribute of resume function as noalias
Close https://github.com/llvm/llvm-project/issues/59221.
The root cause for the problem is that we marked the parameter of the
resume/destroy functions as noalias previously. But this is not true.
See https://github.com/llvm/llvm-project/issues/59221 for the details.
Long story short, for this C++ program
(https://compiler-explorer.com/z/6qGcozG93), the optimized frame will be
something like:
```
struct test_frame {
void (*__resume_)(), // a function pointer points to the
`test.resume` function, which can be imaged as the test() function in
the example.
....
struct a_frame {
...
void **caller; // may points to test_frame at runtime.
};
};
```
And the function a and function test looks just like:
```
define i32 @a(ptr noalias %alloc_8) {
%alloc_8_16 = getelementptr ptr, ptr %alloc_8, i64 16
store i32 42, ptr %alloc_8_16, align 8
%alloc_8_8 = getelementptr ptr, ptr %alloc_8, i64 8
%alloc = load ptr, ptr %alloc_8_8, align 8
%p = load ptr, ptr %alloc, align 8
%r = call i32 %p(ptr %alloc)
ret i32 %r
}
define i32 @b(ptr %p) {
entry:
%alloc = alloca [128 x i8], align 8
%alloc_8 = getelementptr ptr, ptr %alloc, i64 8
%alloc_8_8 = getelementptr ptr, ptr %alloc_8, i64 8
store ptr %alloc, ptr %alloc_8_8, align 8
store ptr %p, ptr %alloc, align 8
%r = call i32 @a(ptr nonnull %alloc_8)
ret i32 %r
}
```
Here inside the function `a`, we can access the parameter `%alloc_8` by
`%alloc` and we pass `%alloc` to an unknown function. So it breaks the
assumption of `noalias` parameter.
Note that although only CoroElide optimization can put a frame inside
another frame directly, the following case is not valid too:
```
struct test_frame {
....
void **a_frame; // may points to a_frame at runtime.
};
struct a_frame {
void **caller; // may points to test_frame at runtime.
};
```
Since the C++ language allows the programmer to get the address of
coroutine frames, we can't assume the above case wouldn't happen in the
source codes. So we can't set the parameter as noalias no matter if
CoroElide applies or not. And for other languages, it may be safe if
they don't allow the programmers to get the address of coroutine frames.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D139295
Vasileios Porpodas [Mon, 28 Nov 2022 22:53:24 +0000 (14:53 -0800)]
[NFC] Remove the instruction list from the arguments of llvm::ReplaceInstWithValue().
This is part of a series of cleanup patches towards making BasicBlock::getInstList() private.
Differential Revision: https://reviews.llvm.org/D139153
Richard Smith [Tue, 13 Dec 2022 01:38:28 +0000 (17:38 -0800)]
Fix test on 32-bit targets.
Diego Caballero [Thu, 24 Nov 2022 02:16:46 +0000 (02:16 +0000)]
[mlir][Vector] Initial masking support in Linalg vectorizer
This patch introduces the initial bits to support vector masking
using the `vector.mask` operation. Vectorization changes should be
NFC for non-masked cases. We can't test masked cases directly until
we extend the Transform dialect to support masking.
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D137690
Roman Lebedev [Tue, 13 Dec 2022 01:04:42 +0000 (04:04 +0300)]
[NFC][Codegen][X86] Revisit interleaved store codegen tests
This matches the coverage with the Costmodel tests,
adds stride 5/7/8, and improves AVX512 checks.
I *think* i've compressed check prefixes
as much as possible, but it's a bit hard to tell.
But hey, at one no longer has to fight against FileCheck+UTC :).
Roman Lebedev [Tue, 13 Dec 2022 01:23:34 +0000 (04:23 +0300)]
[NFC][Codegen][X86] Add higher-VF interleaved load codegen tests that got lost
Richard Smith [Tue, 13 Dec 2022 00:39:14 +0000 (16:39 -0800)]
Add missing check for constant evaluation of a comparison of a pointer
to member naming a weak member to nullptr.
This fixes a miscompile where constant evaluation would incorrectly
determine that a weak member function pointer is never null.
In passing, also improve the diagnostics for constant evaluation of some
nearby cases.
Ting Wang [Tue, 13 Dec 2022 01:07:23 +0000 (20:07 -0500)]
[PowerPC][NFC] Add test case for memset tail store
Add test case to show something can be improved.
Reviewed By: shchenz
Differential Revision: https://reviews.llvm.org/D138881
wren romano [Mon, 12 Dec 2022 23:17:01 +0000 (15:17 -0800)]
[mlir][sparse] Simplifying SparseTensorEncodingAttr function arguments
Since STEA isa Attribute, and that's just (a wrapper around) a pointer, the extra `const` and `&` aren't necessary for function arguments.
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D139886
bixia1 [Wed, 7 Dec 2022 23:39:17 +0000 (15:39 -0800)]
[mlir][sparse] Replace vector.print with printMemref for some tests.
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D139489
Craig Topper [Tue, 13 Dec 2022 00:41:28 +0000 (16:41 -0800)]
[RISCV] Set ShouldSignExtI32Param in TargetLibraryInfo for riscv64.
riscv64 sign extends signed and unsigned i32 arguments like Mips.
Based on discussion here
https://discourse.llvm.org/t/can-we-preserve-signext-return-attribute-when-converting-memcmp-to-bcmp/67126
I'll work on returns next.
Differential Revision: https://reviews.llvm.org/D139893
Roman Lebedev [Mon, 12 Dec 2022 23:32:46 +0000 (02:32 +0300)]
[NFC][Codegen][X86] Revisit interleaved load codegen tests
This matches the coverage with the Costmodel tests,
adds stride 5/7/8, and improves AVX512 checks.
I *think* i've compressed check prefixes
as much as possible, but it's a bit hard to tell.
But hey, at one no longer has to fight against FileCheck+UTC :).
Jordan Rupprecht [Tue, 13 Dec 2022 00:30:51 +0000 (16:30 -0800)]
[NFC][bazel] Run buildifier on all bzl/BUILD.bazel files
Vasileios Porpodas [Tue, 29 Nov 2022 01:35:34 +0000 (17:35 -0800)]
[NFC] Replaces: BB->getInstList().erase(I) with I->eraseFromParent().
Differential Revision: https://reviews.llvm.org/D138976
Dominic Chen [Sat, 10 Dec 2022 01:22:38 +0000 (17:22 -0800)]
[scudo] Optimize scudo test string allocation
When the underlying vector becomes full, it resizes, remaps, and then copies over the old data. To avoid thes excess allocations, allow reservation from the backing vector.
Differential Revision: https://reviews.llvm.org/D135119
Peter Rong [Thu, 8 Dec 2022 01:27:10 +0000 (17:27 -0800)]
[FuzzMutate] InstModificationStrategy, add FastMath flags and exact flags to instructions.
I think there are more attributes, flags we can add to `call`, functions declarations and global variables. Let's start with these two flags.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D139594
Peter Rong [Thu, 1 Dec 2022 01:27:01 +0000 (17:27 -0800)]
[FuzzMutate] New InsertCFGStrategy
Mutating CFG is hard as we have to maintain dominator relations.
We avoid this problem by inserting a CFG into a splitted block.
switch, ret, and br instructions are generated.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D139067
Terry Wilmarth [Thu, 1 Dec 2022 15:13:48 +0000 (09:13 -0600)]
[OpenMP] Refactoring: Move teams forking and serial region forking to separate functions.
Code for serial parallel regions and teams construct have been moved
out of __kmp_fork_call and into separate functions. This is to reduce
the size of the __kmp_fork_call function, and aid in debugging.
Differential Revision: https://reviews.llvm.org/D139116
Sanjay Patel [Mon, 12 Dec 2022 22:48:22 +0000 (17:48 -0500)]
[InstSimplify] try harder to propagate existing NaN values through FP folds
Any undef element in a vector would trigger the whole constant
to be replaced with a canonical NaN. This propagates each
element when possible.
issue #59122
Sanjay Patel [Mon, 12 Dec 2022 22:19:43 +0000 (17:19 -0500)]
[InstSimplify] add tests for vectors with NaN + partial undef; NFC
issue #59122
Craig Topper [Mon, 12 Dec 2022 22:49:26 +0000 (14:49 -0800)]
[RISCV] Make DemandedFields::usedVTYPE() const. NFC
Noticed while reviewing D139877.
Reviewed By: reames
Differential Revision: https://reviews.llvm.org/D139879
Philip Reames [Mon, 12 Dec 2022 22:41:49 +0000 (14:41 -0800)]
[llvm-stress] Minor code improvements around vector types [nfc]
Philip Reames [Mon, 12 Dec 2022 21:48:05 +0000 (13:48 -0800)]
[RISCV][InsertVSETVLI] Reorder code to reduce a future diff [nfc]