Benjamin Kramer [Tue, 8 Dec 2020 14:37:32 +0000 (15:37 +0100)]
[mlir][Shape] Canonicalize assume_all with one input and tensor_cast of constant_shape
This allows simplifying some more complicated shape expressions
Differential Revision: https://reviews.llvm.org/D92843
Gabor Marton [Mon, 7 Dec 2020 17:35:26 +0000 (18:35 +0100)]
[analyzer][StdLibraryFunctionsChecker] Add more return value contraints
This time, we add contraints to functions that either return with [0, -1] or
with a file descriptor.
Differential Revision: https://reviews.llvm.org/D92771
Gabor Marton [Mon, 7 Dec 2020 15:10:40 +0000 (16:10 +0100)]
[analyzer][StdLibraryFunctionsChecker] Make close and mmap to accept -1 as fd
close:
It is quite often that users chose to call close even if the fd is
negative. Theoretically, it would be nicer to close only valid fds, but
in practice the implementations of close just returns with EBADF in case
of a non-valid fd param. So, we can eliminate many false positives if we
let close to take -1 as an fd. Other negative values are very unlikely,
because open and other fd factories return with -1 in case of failure.
mmap:
In the case of MAP_ANONYMOUS flag (which is supported e.g. in Linux) the
mapping is not backed by any file; its contents are initialized to zero.
The fd argument is ignored; however, some implementations require fd to
be -1 if MAP_ANONYMOUS (or MAP_ANON) is specified, and portable
applications should ensure this.
Consequently, we must allow -1 as the 4th arg.
Differential Revision: https://reviews.llvm.org/D92764
Valentin Clement [Tue, 8 Dec 2020 15:36:34 +0000 (10:36 -0500)]
[flang][openacc] Add clause validity tests for the update directive
Add couple of clause validity tests for the update directive and check for
the restriction where at least self, host or device clause must appear on the directive.
Reviewed By: sameeranjoshi
Differential Revision: https://reviews.llvm.org/D92447
Teresa Johnson [Tue, 8 Dec 2020 01:13:49 +0000 (17:13 -0800)]
[ICP] Don't promote when target not defined in module
This guards against cases where the symbol was dead code eliminated in
the binary by ThinLTO, and we have a sample profile collected for one
binary but used to optimize another.
Most of the benefit from ICP comes from inlining the target, which we
can't do with only a declaration anyway. If this is in the pre-ThinLTO
link step (e.g. for instrumentation based PGO), we will attempt the
promotion again in the ThinLTO backend after importing anyway, and we
don't need the early promotion to facilitate that.
Differential Revision: https://reviews.llvm.org/D92804
Christian Sigg [Fri, 4 Dec 2020 18:12:02 +0000 (19:12 +0100)]
[mlir] Tighten access of RewritePattern methods.
In RewritePattern, only expose `matchAndRewrite` as a public function. `match` can be protected (but needs to be protected because we want to call it from an override of `matchAndRewrite`). `rewrite` can be private.
For classes deriving from RewritePattern, all 3 functions can be private.
Side note: I didn't understand the need for the `using RewritePattern::matchAndRewrite` in derived classes, and started poking around. They are gone now, and I think the result is (only very slightly) cleaner.
Reviewed By: ftynse
Differential Revision: https://reviews.llvm.org/D92670
Stefan Pintilie [Tue, 8 Dec 2020 14:57:37 +0000 (08:57 -0600)]
[PowerPC] Fix missing nop after call to weak callee.
Weak functions can be replaced by other functions at link time. Previously it
was assumed that no matter what the weak callee function was replaced with it
would still share the same TOC as the caller. This is no longer true as a weak
callee with a TOC setup can be replaced by another function that was compiled
with PC Relative and does not have a TOC at all.
This patch makes sure that all calls to functions defined as weak from a caller
that has a valid TOC have a nop after the call to allow a place for the linker
to restore the TOC.
Reviewed By: NeHuang
Differential Revision: https://reviews.llvm.org/D91983
Simon Pilgrim [Tue, 8 Dec 2020 15:24:34 +0000 (15:24 +0000)]
[X86] Regenerate vector-shift-*.ll tests
Replace X32 check prefixes with X86 - X32 is generally used for gnux triple tests
Simon Pilgrim [Tue, 8 Dec 2020 14:48:24 +0000 (14:48 +0000)]
[X86] Regenerate store-narrow.ll tests
Replace X32 check prefixes with X86 - X32 is generally used for gnux triple tests
Simon Pilgrim [Tue, 8 Dec 2020 14:47:00 +0000 (14:47 +0000)]
[X86] Regenerate bmi-intrinsics-fast-isel.ll tests
Replace X32 check prefixes with X86 - X32 is generally used for gnux triple tests
Simon Pilgrim [Tue, 8 Dec 2020 14:45:59 +0000 (14:45 +0000)]
[X86] Regenerate addcarry2.ll tests
Replace X32 check prefixes with X86 - X32 is generally used for gnux triple tests
Simon Pilgrim [Tue, 8 Dec 2020 14:44:49 +0000 (14:44 +0000)]
[X86] Regenerate sttni.ll tests
Replace X32 check prefixes with X86 - X32 is generally used for gnux triple tests
Simon Pilgrim [Tue, 8 Dec 2020 14:43:52 +0000 (14:43 +0000)]
[X86] Regenerate clzero.ll tests
Replace X32 check prefixes with X86 - X32 is generally used for gnux triple tests
Chris Kennelly [Thu, 29 Oct 2020 02:45:09 +0000 (22:45 -0400)]
[clang-tidy] Omit std::make_unique/make_shared for default initialization.
This extends the check for default initialization in arrays added in
547f89d6070 to include scalar types and exclude them from the suggested fix for
make_unique/make_shared.
Rewriting std::unique_ptr<int>(new int) as std::make_unique<int>() (or for
other, similar trivial T) switches from default initialization to value
initialization, a performance regression for trivial T. For these use cases,
std::make_unique_for_overwrite is more suitable alternative.
Reviewed By: hokein
Differential Revision: https://reviews.llvm.org/D90392
David Green [Tue, 8 Dec 2020 15:22:46 +0000 (15:22 +0000)]
[ARM] Turn pred_cast(xor(x, -1)) into xor(pred_cast(x), -1)
This folds a not (an xor -1) though a predicate_cast, so that it can be
turned into a VPNOT and potentially be folded away as an else predicate
inside a VPT block.
Differential Revision: https://reviews.llvm.org/D92235
Kevin P. Neal [Tue, 8 Dec 2020 15:15:08 +0000 (10:15 -0500)]
[FPEnv] Correct constrained metadata in fp16-ops-strict.c
This test shows we're in some cases not getting strictfp information from
the AST. Correct that.
Differential Revision: https://reviews.llvm.org/D92596
Chris Kennelly [Sun, 6 Dec 2020 16:26:31 +0000 (11:26 -0500)]
[clang-tidy] Recognize single character needles for absl::StrContains.
Commit
fbdff6f3ae0b in the Abseil tree adds an overload for
absl::StrContains to accept a single character needle for optimized
lookups.
Reviewed By: hokein
Differential Revision: https://reviews.llvm.org/D92810
Adam Czachorowski [Tue, 1 Dec 2020 18:04:42 +0000 (19:04 +0100)]
[clangd] ExtractFunction: disable on regions that sometimes, but not always return.
apply() will fail in those cases, so it's better to detect it in
prepare() already and hide code action from the user.
This was especially annoying on code bases that use a lot of
RETURN_IF_ERROR-like macros.
Differential Revision: https://reviews.llvm.org/D92408
Adhemerval Zanella [Tue, 8 Dec 2020 13:26:18 +0000 (10:26 -0300)]
[compiler-rt] [builtins] Support conversion between fp16 and fp128
This patch adds both extendhftf2 and trunctfhf2 to support
conversion between half-precision and quad-precision floating-point
values. They are built iff the compiler supports _Float16.
Some notes on ARM plaforms: while fp16 is supported on all
architectures, _Float16 is supported only for 32-bit ARM, 64-bit ARM,
and SPIR (as indicated by clang/docs/LanguageExtensions.rst). Also,
fp16 is a storage format and 64-bit ARM supports floating-point
convert precision to half as base armv8-a instruction.
This patch does not change the ABI for 32-bit ARM, it will continue
to pass _Float16 as uint16.
This re-enabled revert done by https://reviews.llvm.org/rGb534beabeed3ba1777cd0ff9ce552d077e496726
Differential Revision: https://reviews.llvm.org/D92242
David Sherwood [Tue, 8 Dec 2020 14:38:12 +0000 (14:38 +0000)]
[SVE] Remove duplicate assert in DAGTypeLegalizer::SplitVecOp_EXTRACT_SUBVECTOR
Jay Foad [Tue, 8 Dec 2020 13:13:13 +0000 (13:13 +0000)]
[AMDGPU] Add occupancy level tests for GFX10.3. NFC.
getMaxWavesPerEU and getVGPRAllocGranule both changed in GFX10.3 and
they both affect the occupancy calculation.
Differential Revision: https://reviews.llvm.org/D92839
ergawy [Tue, 8 Dec 2020 14:02:02 +0000 (09:02 -0500)]
[MLIR][SPIRV] Add initial support for OpSpecConstantOp.
This commit adds initial support for SPIR-V OpSpecConstantOp
instruction. The following is introdcued:
- A new `spv.specConstantOperation` operation consisting of a single
region and of 2 operations within that regions (more details in the
docs of the op itself).
- A new `spv.yield` instruction that acts a terminator for
`spv.specConstantOperation`.
For now, the generic form of the new op is supported (i.e. no custom
parsing or printing). This will be done in a follow up patch.
Reviewed By: antiagainst
Differential Revision: https://reviews.llvm.org/D92232
David Green [Tue, 8 Dec 2020 14:05:07 +0000 (14:05 +0000)]
[ARM] Remove dead instructions before creating VPT block bundles
We remove VPNOT instructions in VPT blocks as we create them, turning
them into else predicates. We don't remove the dead instructions until
after the block has been created though. Because the VPNOT will have
killed the vpr register it used, this makes finalizeBundle add internal
flags to the vpr uses of any instructions after the VPNOT. These
incorrect flags can then confuse what is alive and what is not, leading
to machine verifier problems.
This patch removes them earlier instead, before the bundle is finalized
so that kill flags remain valid.
Differential Revision: https://reviews.llvm.org/D92227
David Sherwood [Tue, 1 Dec 2020 13:15:57 +0000 (13:15 +0000)]
[SVE] Fix crashes with inline assembly
All the crashes found compiling inline assembly are fixed in this
patch by changing AArch64TargetLowering::getRegForInlineAsmConstraint
to be more resilient to mismatched value and register types. For
example, it makes no sense to request a predicate register for
a nxv2i64 type and so on.
Tests have been added here:
test/CodeGen/AArch64/inline-asm-constraints-bad-sve.ll
Differential Revision: https://reviews.llvm.org/D92554
Frederik Gossen [Tue, 8 Dec 2020 12:57:54 +0000 (13:57 +0100)]
Add rsqrt lowering from standard to NVVM
Differential Revision: https://reviews.llvm.org/D92838
Jan Svoboda [Mon, 7 Dec 2020 13:22:23 +0000 (14:22 +0100)]
[clang][cli] Unify boolean marshalling
Use lambdas with captures to replace the redundant infrastructure for marshalling of two boolean flags that control the same keypath.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D92773
Simon Pilgrim [Tue, 8 Dec 2020 12:19:43 +0000 (12:19 +0000)]
[SLP][X86] Extend PR46983 tests to include SSE2,SSE42,AVX512BW test coverage
Noticed while reviewing D92824
Raphael Isemann [Tue, 8 Dec 2020 12:33:14 +0000 (13:33 +0100)]
[lldb][import-std-module] Add a test for typedef'd std types
Nigel Perks [Tue, 8 Dec 2020 12:19:20 +0000 (12:19 +0000)]
[XCore][docs] Fix XCore compiler writer documentation links.
Fix links to XMOS website. Add link for XS2 architecture.
Reviewed By: jryans
Differential Revision: https://reviews.llvm.org/D92019
Sjoerd Meijer [Tue, 8 Dec 2020 11:39:46 +0000 (11:39 +0000)]
[LICM][docs] Document that LICM is also a canonicalization transform. NFC.
This documents that LICM is a canonicalization transform, which we discussed
recently in:
http://lists.llvm.org/pipermail/llvm-dev/2020-December/147184.html
but which was also discused earlier, e.g. in:
http://lists.llvm.org/pipermail/llvm-dev/2019-September/135058.html
David Green [Tue, 8 Dec 2020 11:54:39 +0000 (11:54 +0000)]
[NFC] Chec[^k] -> Check
Some test updates all appearing to use the wrong spelling of CHECK.
Hafiz Abid Qadeer [Tue, 8 Dec 2020 11:46:36 +0000 (11:46 +0000)]
[compiler-rt] Allow appending to 'target_cflags' value from lit_config.
This patch is similar to D84708. When testing compiler-rt on different
baremetal targets, it helps to have the ability to pass some more parameters
at test time that allows you to build the test executable for a
given target. For an example, you may need a different linker command
file for different targets.
This patch will allows to do things like
$ llvm-lit --param=append_target_cflags="-T simulator.ld"
or
$ llvm-lit --param=append_target_cflags="-T hardware.ld"
In this way, you can run tests on different targets without having to run
cmake again.
Reviewed By: delcypher
Differential Revision: https://reviews.llvm.org/D91783
Jeremy Morse [Tue, 8 Dec 2020 10:55:47 +0000 (10:55 +0000)]
[DebugInfo][Docs] Document MIR language debug-info constructs
This patch documents the MIR syntax for a number of things relevant to
debugging information:
* Trailing 'debug-location' metadata that becomes a DebugLoc,
* Variable location metadata for stack slots,
* Syntax for DBG_VALUE metainstructions,
* Syntax for DBG_INSTR_REF, including trailing instruction numbers
attached to MIR instructions.
Differential Revision: https://reviews.llvm.org/D89337
Pan, Tao [Tue, 8 Dec 2020 09:57:03 +0000 (17:57 +0800)]
[CodeGen] Add text section prefix for COFF object file
Text section prefix is created in CodeGenPrepare, it's file format independent implementation, text section name is written into object file in TargetLoweringObjectFile, it's file format dependent implementation, port code of adding text section prefix to text section name from ELF to COFF.
Different with ELF that use '.' as concatenation character, COFF use '$' as concatenation character. That is, concatenation character is variable, so split concatenation character from text section prefix.
Text section prefix is existing feature of ELF, it can help to reduce icache and itlb misses, it's also make possible aggregate other compilers e.g. v8 created same prefix sections. Furthermore, the recent feature Machine Function Splitter (basic block level text prefix section) is based on text section prefix.
Reviewed By: pengfei, rnk
Differential Revision: https://reviews.llvm.org/D92073
Cullen Rhodes [Thu, 3 Dec 2020 11:05:51 +0000 (11:05 +0000)]
[IR] Remove CastInst::isCastable since it is not used
It was removed back in 2013 (f63dfbb) by Matt Arsenault but then
reverted since DragonEgg used it, but that project is no longer
maintained.
Reviewed By: ldionne, dexonsmith
Differential Revision: https://reviews.llvm.org/D92571
Tim Northover [Wed, 21 Oct 2020 09:11:25 +0000 (10:11 +0100)]
UBSAN: emit distinctive traps
Sometimes people get minimal crash reports after a UBSAN incident. This change
tags each trap with an integer representing the kind of failure encountered,
which can aid in tracking down the root cause of the problem.
Stefan Gränitz [Tue, 8 Dec 2020 09:56:50 +0000 (10:56 +0100)]
[Orc] Two small fixes in TPCDynamicLibrarySearchGenerator
There is one result per lookup symbol, so we have to advance the result iterator no matter whether it's NULL or not.
MissingSymbols variable is unused.
Reviewed By: lhames
Differential Revision: https://reviews.llvm.org/D91707
Stefan Gränitz [Tue, 8 Dec 2020 09:53:50 +0000 (10:53 +0100)]
[JITLink][ELF] Route objects to their matching linker backends based on header info
Distinguish objects by target properties address size, endian and machine architecture. So far we only
support x86-64 (ELFCLASS64, ELFDATA2LSB, EM_X86_64).
Reviewed By: lhames
Differential Revision: https://reviews.llvm.org/D90860
Evgeniy Brevnov [Fri, 4 Dec 2020 09:53:17 +0000 (16:53 +0700)]
[DSE][NFC] Need to be carefull mixing signed and unsigned types
Currently in some places we use signed type to represent size of an access and put explicit casts from unsigned to signed.
For example: int64_t EarlierSize = int64_t(Loc.Size.getValue());
Even though it doesn't loos bits (immidiatly) it may overflow and we end up with negative size. Potentially that cause later code to work incorrectly. A simple expample is a check that size is not negative.
I think it would be safer and clearer if we use unsigned type for the size and handle it appropriately.
Reviewed By: fhahn
Differential Revision: https://reviews.llvm.org/D92648
Alex Zinenko [Mon, 7 Dec 2020 15:18:32 +0000 (16:18 +0100)]
[mlir] Add an option to control the number of loops in affine parallelizer
Add a pass option to control the number of nested parallel loops produced by
the parallelization passes. This is useful to build end-to-end passes targeting
systems that don't need multiple parallel dimensions (e.g., CPUs typically need
only one).
Reviewed By: wsmoses, chelini
Differential Revision: https://reviews.llvm.org/D92765
Alex Zinenko [Mon, 7 Dec 2020 14:45:39 +0000 (15:45 +0100)]
[mlir] properly support min/max in affine parallelization
The existing implementation of the affine parallelization silently copies over
the lower and upper bound maps from affine.for to affine.parallel. However, the
semantics of these maps differ between these two ops: in affine.for, a max(min)
of results is taken for the lower(upper) bound; in affine.parallel, multiple
induction variables can be defined an each result corresponds to one induction
variable. Thus the existing implementation could generate invalid IR or IR that
passes the verifier but has different semantics than the original code. Fix the
parallelization utility to emit dedicated min/max operations before the
affine.parallel in such cases. Disallow parallelization if min/max would have
been in an operation without the AffineScope trait, e.g., in another loop,
since the result of these operations is not considered a valid affine dimension
identifier and may not be properly handled by the affine analyses.
Reviewed By: wsmoses
Differential Revision: https://reviews.llvm.org/D92763
Luís Marques [Mon, 7 Dec 2020 23:50:43 +0000 (23:50 +0000)]
[Clang][CodeGen][RISCV] Fix hard float ABI for struct with empty struct and complex
Fixes bug 44904.
Differential Revision: https://reviews.llvm.org/D91278
Luís Marques [Mon, 7 Dec 2020 23:50:42 +0000 (23:50 +0000)]
[Clang][CodeGen][RISCV] Fix hard float ABI test cases with empty struct
The code seemed not to account for the field 1 offset.
Differential Revision: https://reviews.llvm.org/D91270
Luís Marques [Mon, 7 Dec 2020 23:50:35 +0000 (23:50 +0000)]
[Clang][CodeGen][RISCV] Add hard float ABI tests with empty struct
This patch adds tests that showcase a behavior that is currently buggy.
Fix in a follow-up patch.
Differential Revision: https://reviews.llvm.org/D91269
Anton Afanasyev [Tue, 8 Dec 2020 08:54:42 +0000 (11:54 +0300)]
[SLP][Test] Differentiate SSE/AVX512 test coverage (NFC)
Add test coverage for SSE/AVX512 for insert-after-bundle.ll test.
Prepare this test for accurate showing of PR46983 fix.
Richard Smith [Tue, 8 Dec 2020 08:42:26 +0000 (00:42 -0800)]
Revert "Add new 'preferred_name' attribute."
This change exposed a pre-existing issue with deserialization cycles
caused by a combination of attributes and template instantiations
violating the deserialization ordering restrictions; see PR48434 for
details.
A previous commit attempted to work around PR48434, but appears to have
only been a partial fix, and fixing this properly seems non-trivial.
Backing out for now to unblock things.
This reverts commit
98f76adf4e941738c0b9fe3b9965fa63603e9c89 and
commit
a64c26a47a81b1b44e36d235ff3bc6a74a0bad9f.
David Blaikie [Tue, 8 Dec 2020 07:49:37 +0000 (23:49 -0800)]
Fix shtest-timeout test harder
Don't produce or expect any output from the infinite looping test -
doing so is a recipe for racey flakyness without a longer timeout to
ensure the output is received first, even though that doesn't seem
integral/important to the test. Instead have a plain, no output infinite
loop and check that that is caught and handled.
If for some reason the output is valuable for test coverage - the
timeout should be increased from 1 second to give the process time to
output the text, flush, and for that text to be received and buffered
before the test is timed out.
Marek Kurdej [Tue, 8 Dec 2020 08:00:45 +0000 (09:00 +0100)]
[libc++] Mark LWG3200 as Nothing To Do. NFC.
This is only a wording change, because it is currently impossible to constrain the overload set on whether the type is complete or not.
Qiu Chaofan [Tue, 8 Dec 2020 07:20:15 +0000 (15:20 +0800)]
[NFC] [PowerPC] Move i1-to-fp tests and use script
Arthur Eubanks [Tue, 8 Dec 2020 07:08:02 +0000 (23:08 -0800)]
[test] Pin provenance.ll to legacy PM
It doesn't seem right to port -pa-eval just for one test, punting
decision for how to handle this.
Arthur Eubanks [Tue, 8 Dec 2020 07:04:33 +0000 (23:04 -0800)]
[test] Fix loop-micro-op-buffer-size-t99.ll under NPM
The NPM runs loop passes on loops in forward program order instead of
the legacy PM's reverse program order, causing some debug output to be
in a different order.
wlei [Tue, 8 Dec 2020 06:01:17 +0000 (22:01 -0800)]
[llvm-profgen][NFC] Fix test failure by making unwinder's output deterministic
Don't know why under Sanitizer build(asan/msan/ubsan), the `std::unordered_map<string, ...>`'s output order is reversed, make the regression test failed.
This change creates a workaround by using sorted container to make the output deterministic.
Reviewed By: hoy, wenlei
Differential Revision: https://reviews.llvm.org/D92816
Qiu Chaofan [Tue, 8 Dec 2020 06:08:52 +0000 (14:08 +0800)]
[PowerPC] Implement intrinsic for DARN instruction
Instruction darn was introduced in ISA 3.0. It means 'Deliver A Random
Number'. The immediate number L means:
- L=0, the number is 32-bit (higher 32-bits are all-zero)
- L=1, the number is 'conditioned' (processed by hardware to reduce bias)
- L=2, the number is not conditioned, directly from noise source
GCC implements them in three separate intrinsics: __builtin_darn,
__builtin_darn_32 and __builtin_darn_raw. This patch implements the
same intrinsics. And this change also addresses Bugzilla PR39800.
Reviewed By: steven.zhang
Differential Revision: https://reviews.llvm.org/D92465
Arthur Eubanks [Tue, 8 Dec 2020 05:54:38 +0000 (21:54 -0800)]
[test] Fix Transforms/LoopVersioningLICM under NPM
There were already both legacy and new PM RUN lines.
Also make the NPM RUN line actually match the legacy PM RUN line.
Arthur Eubanks [Tue, 8 Dec 2020 05:48:21 +0000 (21:48 -0800)]
[test] Fix Transforms/LoopVectorize under NPM
The -enable-new-pm=1 translation caused loop-vectorize to run on all
functions, then instcombine, rather than all passes on one function then
the next. This caused the output of -debug-only and -print-after to be
interleaved in an unexpected way.
Arthur Eubanks [Tue, 8 Dec 2020 05:17:07 +0000 (21:17 -0800)]
[test] Fix store_cost.ll under NPM
The NPM processes loops in forward program order, whereas the legacy PM
processes them in reverse program order. No reason to test both PMs
here, so just stick to the NPM.
Jonas Devlieghere [Tue, 8 Dec 2020 04:31:07 +0000 (20:31 -0800)]
[lldb] Include thread id in the reproducer trace (NFC)
Include the current thread ID in the reproducer trace during
capture/recording.
Vitaly Buka [Sat, 5 Dec 2020 07:46:58 +0000 (23:46 -0800)]
[NFC][MSan] Round up OffsetPtr in PoisonMembers
getFieldOffset(layoutStartOffset) is expected to point to the first trivial
field or the one which follows non-trivial. So it must be byte aligned already.
However this is not obvious without assumptions about callers.
This patch will avoid the need in such assumptions.
Depends on D92727.
Differential Revision: https://reviews.llvm.org/D92728
Arthur Eubanks [Tue, 8 Dec 2020 03:19:37 +0000 (19:19 -0800)]
[test] Fix widen-iv.ll under NPM
The -loop-flatten legacy pass preserves loop analyses. The legacy PM
will check all passes that preserve loop analyses that they preserve
LCSSA. This implicitly involves running -loop-simplify. The test
shouldn't depend on verify flags being set in order to run
-loop-simplify, so explicitly add it. The new PM ends up not running it
otherwise.
Kai Luo [Tue, 8 Dec 2020 03:22:32 +0000 (03:22 +0000)]
[DAGCombine][PowerPC] Simplify nabs by using legal `smin` operation
Convert `0 - abs(x)` to `smin (x, -x)` if `smin` is a legal operation.
Verification: https://alive2.llvm.org/ce/z/vpquFR
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D92637
Esme-Yi [Tue, 8 Dec 2020 03:20:12 +0000 (03:20 +0000)]
[PowerPC] Correct the bit-width definition for some imm operand in td.
Summary: The imm operands of some instructions are not defined accurately in td.
This is a small patch to correct these definitions.
Reviewed By: steven.zhang
Differential Revision: https://reviews.llvm.org/D91603
Richard Smith [Tue, 8 Dec 2020 02:41:43 +0000 (18:41 -0800)]
Fix assertion failure due to incorrect dependence bits on a DeclRefExpr
that can only be set correctly after instantiating the initializer for a
variable.
Fangrui Song [Tue, 8 Dec 2020 02:40:31 +0000 (18:40 -0800)]
[test] Rewrite split-debug.c
Use generic ELF target triples.
Add missing coverage: -gsplit-dwarf=split -g -fsplit-dwarf-inlining
Reorganize and add comments.
Test -gno-pubnames
Arthur Eubanks [Tue, 8 Dec 2020 01:36:18 +0000 (17:36 -0800)]
[test] Fix LoopFusion tests under NewPM
The legacy pass depended on -loop-simplify running. The NPM does not
allow for a non-analysis pass to depend on another non-analysis pass.
Jessica Paquette [Tue, 1 Dec 2020 19:58:19 +0000 (11:58 -0800)]
[AArch64][GlobalISel] Refactor G_BRCOND selection
`selectCompareBranch` was hard to understand.
Also, it was being needlessly pessimistic with the `ProduceNonFlagSettingCondBr`
case. It assumed that everything in `selectCompareBranch` would emit a TB(N)Z
or C(B)NZ. That's not true; the G_FCMP + G_BRCOND case would never emit those
instructions, and the G_ICMP + G_BRCOND case was capable of emitting an integer
compare + Bcc.
- Refactor `selectCompareBranch` into separate functions based off of what is
feeding the G_BRCOND's condition.
- Move G_BRCOND selection code from `select` to `selectCompareBranch`.
- Remove duplicated constraint code from the code originally in `select`;
`emitTestBit` already handles that, so no need to constrain twice.
- Factor out the G_FCMP + G_BRCOND case into `selectCompareBranchFedByFCmp`.
- Split the G_ICMP + G_BRCOND case into an optimization function,
`tryOptCompareBranchFedByICmp` and a general selection function,
`selectCompareBranchFedByICmp`.
- Reduce the number of things passed to `tryOptAndIntoCompareBranch`.
- Improve documentation.
- Give some variables more descriptive names.
Other than improving the code generation for functions with
speculative_load_hardening by getting the logic correct, this is NFC.
Differential Revision: https://reviews.llvm.org/D92582
Valentin Churavy [Mon, 7 Dec 2020 21:13:57 +0000 (16:13 -0500)]
[VNCoercion] Disallow coercion between different ni addrspaces
I'm not sure if it would be legal by the IR reference to introduce
an addrspacecast here, since the IR reference is a bit vague on
the exact semantics, but at least for our usage of it (and I
suspect for many other's usage) it is not. For us, addrspacecasts
between non-integral address spaces carry frontend information that the
optimizer cannot deduce afterwards in a generic way (though we
have frontend specific passes in our pipline that do propagate
these). In any case, I'm sure nobody is using it this way at
the moment, since it would have introduced inttoptrs, which
are definitely illegal.
Fixes PR38375
Co-authored-by: Keno Fischer <keno@alumni.harvard.edu>
Reviewed By: reames
Differential Revision: https://reviews.llvm.org/D50010
Yaxun (Sam) Liu [Tue, 8 Dec 2020 00:48:59 +0000 (19:48 -0500)]
Fix lit test failure due to 0b81d9
These lit tests now requires amdgpu-registered-target since they
use clang driver and clang driver passes an LLVM option which
is available only if amdgpu target is registered.
Change-Id: I2df31967409f1627fc6d342d1ab5cc8aa17c9c0c
Douglas Yung [Tue, 8 Dec 2020 00:36:58 +0000 (16:36 -0800)]
Fixup test in path to use C:\ instead of D:\ which may be mapped to a removable.
Our internal build bot hit a failure in llvm/test/tools/llvm-symbolizer/pdb/missing_pdb.test
because the test was checking for an error message that is emitted when a pdb file is
missing. But when the drive is mapped to a removalable drive (such as a DVD drive) in
Windows, you get a different error message which causes the test to fail.
This fixes the test by changing the drive the missing pdb is expected to be on to C:\
instead of D:\ as that is the drive historically used to install Windows and thus
if present should be a hard drive.
Reviewed By: rnk
Differential Revision: https://reviews.llvm.org/D92787
Richard Smith [Mon, 7 Dec 2020 23:50:18 +0000 (15:50 -0800)]
Fix deserialization cycle in preferred_name attribute.
This is really just a workaround for a more fundamental issue in the way
we deserialize attributes. See PR48434 for details.
Also fix tablegen code generator to produce more correct indentation to
resolve buildbot issues with -Werror=misleading-indentation firing
inside the generated code.
Yaxun (Sam) Liu [Tue, 27 Oct 2020 18:10:02 +0000 (14:10 -0400)]
[AMDGPU] add -mcode-object-version=n
Add option -mcode-object-version=n to control code object version for
AMDGPU.
Differential Revision: https://reviews.llvm.org/D91310
Yaxun (Sam) Liu [Thu, 17 Sep 2020 18:53:24 +0000 (14:53 -0400)]
[clang][AMDGPU] remove mxnack and msramecc options
Remove mxnack and msramecc options since they
are deprecated by --offload-arch.
This is part of https://reviews.llvm.org/D60620
Yaxun (Sam) Liu [Thu, 13 Aug 2020 20:31:51 +0000 (16:31 -0400)]
[HIP] fix bundle entry ID for --
Canonicalize triple used in fat binary. Change from
amdgcn-amd-amdhsa to amdgcn-amd-amdhsa-.
This is part of https://reviews.llvm.org/D60620
Mehdi Amini [Sat, 5 Dec 2020 02:08:38 +0000 (02:08 +0000)]
Add Python binding for MLIR Type Attribute
Differential Revision: https://reviews.llvm.org/D92711
Mehdi Amini [Sat, 5 Dec 2020 01:28:41 +0000 (01:28 +0000)]
Customize exception thrown from mlir.Operation.create() python bindings
The default exception handling isn't very user friendly and does not
point accurately to the issue. Instead we can indicate which of the
operands isn't valid and provide contextual information in the error
message.
Differential Revision: https://reviews.llvm.org/D92710
Yaxun (Sam) Liu [Mon, 7 Dec 2020 22:31:28 +0000 (17:31 -0500)]
[clang][AMDGPU] rename sram-ecc as sramecc
As backend renamed sram-ecc to sramecc, this patch makes
corresponding change in clang.
Differential Revision: https://reviews.llvm.org/D86217
Jessica Paquette [Fri, 4 Dec 2020 23:51:44 +0000 (15:51 -0800)]
[AArch64][GlobalISel] Narrow 128-bit regs to 64-bit regs in emitTestBit
When we have a 128-bit register, emitTestBit would incorrectly narrow to 32
bits always. If the bit number was > 32, then we would need a TB(N)ZX. This
would cause a crash, as we'd have the wrong register class. (PR48379)
This generalizes `narrowExtReg` into `moveScalarRegClass`.
This also allows us to remove `widenGPRBankRegIfNeeded` entirely, since
`selectCopy` correctly handles SUBREG_TO_REG etc.
This does create some codegen changes (since `selectCopy` uses the `all`
regclass variants). However, I think that these will likely be optimized away,
and we can always improve the `selectCopy` code. It looks like we should
revisit `selectCopy` at this point, and possibly refactor it into at least one
`emit` function.
Differential Revision: https://reviews.llvm.org/D92707
Philip Reames [Mon, 7 Dec 2020 22:36:19 +0000 (14:36 -0800)]
Teach isKnownNonEqual how to recurse through invertible multiplies
Build on the work started in 8f07629, and add the multiply case. In the process, more clearly describe the requirement for the operation we're looking through.
Differential Revision: https://reviews.llvm.org/D92726
Jann Horn [Mon, 7 Dec 2020 22:48:41 +0000 (14:48 -0800)]
[clang] Fix noderef for AddrOf on MemberExpr
Committing on behalf of thejh (Jann Horn).
As part of this change, one existing test case has to be adjusted
because it accidentally stripped the NoDeref attribute without
getting caught.
Depends on D92140
Differential Review: https://reviews.llvm.org/D92141
peter klausler [Mon, 7 Dec 2020 20:08:58 +0000 (12:08 -0800)]
[flang] Improve initializer semantics, esp. for component default values
This patch plugs many holes in static initializer semantics, improves error
messages for default initial values and other component properties in
parameterized derived type instantiations, and cleans up several small
issues noticed during development. We now do proper scalar expansion,
folding, and type, rank, and shape conformance checking for component
default initializers in derived types and PDT instantiations.
The initial values of named constants are now guaranteed to have been folded
when installed in the symbol table, and are no longer folded or
scalar-expanded at each use in expression folding. Semantics documentation
was extended with information about the various kinds of initializations
in Fortran and when each of them are processed in the compiler.
Some necessary concomitant changes have bulked this patch out a bit:
* contextual messages attachments, which are now produced for parameterized
derived type instantiations so that the user can figure out which
instance caused a problem with a component, have been added as part
of ContextualMessages, and their implementation was debugged
* several APIs in evaluate::characteristics was changed so that a FoldingContext
is passed as an argument rather than just its intrinsic procedure table;
this affected client call sites in many files
* new tools in Evaluate/check-expression.cpp to determine when an Expr
actually is a single constant value and to validate a non-pointer
variable initializer or object component default value
* shape conformance checking has additional arguments that control
whether scalar expansion is allowed
* several now-unused functions and data members noticed and removed
* several crashes and bogus errors exposed by testing this new code
were fixed
* a -fdebug-stack-trace option to enable LLVM's stack tracing on
a crash, which might be useful in the future
TL;DR: Initialization processing does more and takes place at the right
times for all of the various kinds of things that can be initialized.
Differential Review: https://reviews.llvm.org/D92783
Leonard Chan [Mon, 7 Dec 2020 22:39:42 +0000 (14:39 -0800)]
[clang] Fix noderef for array member of deref expr
Committing on behalf of thejh (Jann Horn).
Given an attribute((noderef)) pointer "p" to the struct
struct s { int a[2]; };
ensure that the following expressions are treated the same way by the
noderef logic:
p->a
(*p).a
Until now, the first expression would be treated correctly (nothing is
added to PossibleDerefs because CheckMemberAccessOfNoDeref() bails out
on array members), but the second expression would incorrectly warn
because "*p" creates a PossibleDerefs entry.
Handle this case the same way as for the AddrOf operator.
Differential Revision: https://reviews.llvm.org/D92140
Mitch Phillips [Mon, 7 Dec 2020 22:30:53 +0000 (14:30 -0800)]
Revert "[test] Fix asan/TestCases/Linux/globals-gc-sections-lld.cpp with -fsanitize-address-globals-dead-stripping"
This reverts commit
140808768d3e5c0f4e52dd42094650f5d282e34a.
Reason: Broke the upstream bots - discussed offline.
Erik Pilkington [Sun, 6 Dec 2020 20:08:12 +0000 (15:08 -0500)]
[clang] Add support for attribute 'swift_async'
This attributes specifies how (or if) a given function or method will be
imported into a swift async method. rdar://
70111252
Differential revision: https://reviews.llvm.org/D92742
Erik Pilkington [Mon, 7 Dec 2020 14:14:25 +0000 (09:14 -0500)]
[clang] Add a new nullability annotation for swift async: _Nullable_result
_Nullable_result generally like _Nullable, except when being imported into a
swift async method. rdar://
70106409
Differential revision: https://reviews.llvm.org/D92495
Mehdi Amini [Mon, 7 Dec 2020 21:57:32 +0000 (21:57 +0000)]
Set the target branch for `arc land` to main
wlei [Mon, 19 Oct 2020 19:55:59 +0000 (12:55 -0700)]
[CSSPGO][llvm-profgen] Context-sensitive profile data generation
This stack of changes introduces `llvm-profgen` utility which generates a profile data file from given perf script data files for sample-based PGO. It’s part of(not only) the CSSPGO work. Specifically to support context-sensitive with/without pseudo probe profile, it implements a series of functionalities including perf trace parsing, instruction symbolization, LBR stack/call frame stack unwinding, pseudo probe decoding, etc. Also high throughput is achieved by multiple levels of sample aggregation and compatible format with one stop is generated at the end. Please refer to: https://groups.google.com/g/llvm-dev/c/1p1rdYbL93s for the CSSPGO RFC.
This change supports context-sensitive profile data generation into llvm-profgen. With simultaneous sampling for LBR and call stack, we can identify leaf of LBR sample with calling context from stack sample . During the process of deriving fall through path from LBR entries, we unwind LBR by replaying all the calls and returns (including implicit calls/returns due to inlining) backwards on top of the sampled call stack. Then the state of call stack as we unwind through LBR always represents the calling context of current fall through path.
we have two types of virtual unwinding 1) LBR unwinding and 2) linear range unwinding.
Specifically, for each LBR entry which can be classified into call, return, regular branch, LBR unwinding will replay the operation by pushing, popping or switching leaf frame towards the call stack and since the initial call stack is most recently sampled, the replay should be in anti-execution order, i.e. for the regular case, pop the call stack when LBR is call, push frame on call stack when LBR is return. After each LBR processed, it also needs to align with the next LBR by going through instructions from previous LBR's target to current LBR's source, which we named linear unwinding. As instruction from linear range can come from different function by inlining, linear unwinding will do the range splitting and record counters through the range with same inline context.
With each fall through path from LBR unwinding, we aggregate each sample into counters by the calling context and eventually generate full context sensitive profile (without relying on inlining) to driver compiler's PGO/FDO.
A breakdown of noteworthy changes:
- Added `HybridSample` class as the abstraction perf sample including LBR stack and call stack
* Extended `PerfReader` to implement auto-detect whether input perf script output contains CS profile, then do the parsing. Multiple `HybridSample` are extracted
* Speed up by aggregating `HybridSample` into `AggregatedSamples`
* Added VirtualUnwinder that consumes aggregated `HybridSample` and implements unwinding of calls, returns, and linear path that contains implicit call/return from inlining. Ranges and branches counters are aggregated by the calling context.
Here calling context is string type, each context is a pair of function name and callsite location info, the whole context is like `main:1 @ foo:2 @ bar`.
* Added PorfileGenerater that accumulates counters by ranges unfolding or branch target mapping, then generates context-sensitive function profile including function body, inferring callee's head sample, callsite target samples, eventually records into ProfileMap.
* Leveraged LLVM build-in(`SampleProfWriter`) writer to support different serialization format with no stop
- `getCanonicalFnName` for callee name and name from ELF section
- Added regression test for both unwinding and profile generation
Test Plan:
ninja & ninja check-llvm
Reviewed By: hoy, wenlei, wmi
Differential Revision: https://reviews.llvm.org/D89723
Vitaly Buka [Sat, 5 Dec 2020 07:47:11 +0000 (23:47 -0800)]
[CodeGen][MSan] Don't use offsets of zero-sized fields
Such fields will likely have offset zero making
__sanitizer_dtor_callback poisoning wrong regions.
E.g. it can poison base class member from derived class constructor.
Differential Revision: https://reviews.llvm.org/D92727
Alex Zinenko [Wed, 2 Dec 2020 11:53:12 +0000 (12:53 +0100)]
[OpenMPIRBuilder] introduce createStaticWorkshareLoop
Introduce a function that creates a statically-scheduled workshare loop
out of a canonical loop created earlier by the OpenMPIRBuilder. This
basically amounts to injecting runtime calls to the preheader and the
after block and updating the trip count. Static scheduling kind is
currently hardcoded and needs to be extracted from the runtime library
into common TableGen definitions.
Differential Revision: https://reviews.llvm.org/D92476
Michael Kruse [Mon, 7 Dec 2020 20:39:32 +0000 (14:39 -0600)]
[Polly][CodeGen] Remove use of ScalarEvolution.
ScalarEvolution::getSCEV cannot be used during codegen. ScalarEvolution
assumes a stable IR and control flow which is under construction during
Polly's CodeGen. In particular, it uses DominatorTree for compute the
backedge taken count. However the DominatorTree is not updated during
codegen.
In this case, SCEV was used to determine the base pointer of an array
access. Replace it by our own function. Polly generates only GEP and
BitCasts for array acceses, i.e. it is sufficient to handle these to to
find the base pointer.
Fixes llvm.org/PR48422
Amy Huang [Thu, 3 Dec 2020 17:03:55 +0000 (09:03 -0800)]
[CodeView] Fix inline sites that are missing code offsets.
When an inline site has a starting code offset of 0, we sometimes
don't emit the starting offset.
Bug: https://bugs.llvm.org/show_bug.cgi?id=48377
Differential Revision: https://reviews.llvm.org/D92590
Nico Weber [Mon, 7 Dec 2020 16:29:25 +0000 (11:29 -0500)]
docs: Add pointer to cmake caches for PGO
Also add a link to end-user PGO documentation.
Differential Revision: https://reviews.llvm.org/D92768
Richard Smith [Thu, 12 Nov 2020 01:12:18 +0000 (17:12 -0800)]
Add new 'preferred_name' attribute.
This attribute permits a typedef to be associated with a class template
specialization as a preferred way of naming that class template
specialization. This permits us to specify that (for example) the
preferred way to express 'std::basic_string<char>' is as 'std::string'.
The attribute is applied to the various class templates in libc++ that have
corresponding well-known typedef names.
Differential Revision: https://reviews.llvm.org/D91311
Amara Emerson [Mon, 7 Dec 2020 20:47:28 +0000 (12:47 -0800)]
[AArch64] Fix some minor coding style issues in AArch64CompressJumpTables
Nathan James [Mon, 7 Dec 2020 20:23:11 +0000 (20:23 +0000)]
[llvm][NFC] Made RefCountBase constructors protected
Matches ThreadSafeRefCountBase and forces the class to be inherited.
Nathan James [Mon, 7 Dec 2020 20:20:08 +0000 (20:20 +0000)]
[llvm] Add asserts in (ThreadSafe)?RefCountedBase destructors
Added a trivial destructor in release mode and in debug mode a destructor that asserts RefCount is indeed zero.
This ensure people aren't manually (maybe accidentally) destroying these objects like in this contrived example.
```lang=c++
{
std::unique_ptr<SomethingRefCounted> Object;
holdIntrusiveOwnership(Object.get());
// Object Destructor called here will assert.
}
```
Reviewed By: dblaikie
Differential Revision: https://reviews.llvm.org/D92480
Derek Schuff [Fri, 4 Dec 2020 21:45:42 +0000 (13:45 -0800)]
[WebAssembly] Add Object and ObjectWriter support for wasm COMDAT sections
Allow sections to be placed into COMDAT groups, in addtion to functions and data
segments.
Also make section symbols unnamed, which allows sections with identical names
(section names are independent of their section symbols, but previously we
gave the symbols the same name as their sections, which results in collisions
when sections are identically-named).
Differential Revision: https://reviews.llvm.org/D92691
Sanjay Patel [Mon, 7 Dec 2020 20:09:21 +0000 (15:09 -0500)]
[SLP] fix typo in debug string; NFC
Aart Bik [Mon, 7 Dec 2020 19:54:58 +0000 (11:54 -0800)]
[mlir][sparse] hoist loop invariant tensor loads in sparse compiler
After bufferization, the backend has much more trouble hoisting loop invariant
loads from the loops generated by the sparse compiler. Therefore, this is done
during sparse code generation. Note that we don't bother hoisting derived
invariant expressions on SSA values, since the backend does that very well.
Still TBD: scalarize reductions to avoid load-add-store cycles
Reviewed By: penpornk
Differential Revision: https://reviews.llvm.org/D92534
Erich Keane [Mon, 7 Dec 2020 19:27:01 +0000 (11:27 -0800)]
Stop ExtractTypeForDeductionGuide from recursing on TypeSourceInfo
As reported in PR48177, the type-deduction extraction ends up going into
an infinite loop when the type referred to has a recursive definition.
This stops recursing and just substitutes the type-source-info the
TypeLocBuilder identified when transforming the base.
Bardia Mahjour [Mon, 7 Dec 2020 19:23:52 +0000 (14:23 -0500)]
[LV] Epilogue Vectorization with Optimal Control Flow - Default Enablement
This patch enables epilogue vectorization by default per reviewer requests.
Differential Revision: https://reviews.llvm.org/D89566
Yu Shan [Mon, 7 Dec 2020 18:58:44 +0000 (10:58 -0800)]
[analyzer] Ignore annotations if func is inlined.
When we annotating a function header so that it could be used by other
TU, we also need to make sure the function is parsed correctly within
the same TU. So if we can find the function's implementation,
ignore the annotations, otherwise, false positive would occur.
Move the escape by value case to post call and do not escape the handle
if the function is inlined and we have analyzed the handle.
Differential Revision: https://reviews.llvm.org/D91902