Commit Graph

369279 Commits

Author SHA1 Message Date
Jameson Nash ac2def2d8d make the AsmPrinterHandler array public
This lets external consumers customize the output, similar to how
AssemblyAnnotationWriter lets the caller define callbacks when printing
IR. The array of handlers already existed, this just cleans up the code
so that it can be exposed publically.

Differential Revision: https://reviews.llvm.org/D74158
2020-10-16 16:27:31 -04:00
Richard Smith 7e801ca0ef Treat constant contexts as being in the default rounding mode.
This addresses a regression where pretty much all C++ compilations using
-frounding-math now fail, due to rounding being performed in constexpr
function definitions in the standard library.

This follows the "manifestly constant evaluated" approach described in
https://reviews.llvm.org/D87528#2270676 -- evaluations that are required
to succeed at compile time are permitted even in regions with dynamic
rounding modes, as are (unfortunately) the evaluation of the
initializers of local variables of const integral types.

Differential Revision: https://reviews.llvm.org/D89360
2020-10-16 13:26:15 -07:00
Richard Smith 48c70c1664 Extend memset-to-zero optimization to C++11 aggregate functional casts
Aggr{...}.

We previously missed these cases due to not stepping over the additional
AST nodes representing their syntactic form.
2020-10-16 13:21:08 -07:00
Artem Belevich c36c0fabd1 [VectorCombine] Avoid crossing address space boundaries.
We can not bitcast pointers across different address spaces, and VectorCombine
should be careful when it attempts to find the original source of the loaded
data.

Differential Revision: https://reviews.llvm.org/D89577
2020-10-16 13:19:31 -07:00
Stanislav Mekhanoshin 874524ab88 [AMDGPU] Drop array size in AMDGCNGPUs and R600GPUs
Differential Revision: https://reviews.llvm.org/D89568
2020-10-16 12:37:22 -07:00
Nikita Popov 74c8c2d903 Revert "Recommit "[SCEV] Use nw flag and symbolic iteration count to sharpen ranges of AddRecs""
This reverts commit 32b72c3165.

While better than before, this change still introduces a large
compile-time regression (>3% on mafft):
https://llvm-compile-time-tracker.com/compare.php?from=fbd62fe60fb2281ca33da35dc25ca3c87ec0bb51&to=32b72c3165bf65cca2e8e6197b59eb4c4b60392a&stat=instructions

Additionally, the logic here doesn't look quite right to me,
I will comment in more detail on the differential revision.
2020-10-16 21:36:33 +02:00
Florian Hahn f085b7cbc1 [SCEV] Add additional tests where the max BTC is limited by wrapping. 2020-10-16 20:36:02 +01:00
Arthur Eubanks faf5210420 [CGSCC] Add -abort-on-max-devirt-iterations-reached option
Aborts if we hit the max devirtualization iteration.
Will be useful for testing that changes to devirtualization don't cause
devirtualization to repeat passes more times than necessary.

Reviewed By: rnk

Differential Revision: https://reviews.llvm.org/D89519
2020-10-16 12:34:52 -07:00
Peter Steinfeld f2897b8f45 [flang] Disallow pointer constants
None of the other Fortran compilers allow them.

Differential Revision: https://reviews.llvm.org/D89581
2020-10-16 12:31:42 -07:00
Austin Kerbow 978fbd8268 [AMDGPU] Run hazard recognizer pass later
If instructions were removed in peephole passes after the hazard recognizer was
run it is possible that new hazards could be introduced.

Fixes: SWDEV-253090

Reviewed By: rampitec, arsenm

Differential Revision: https://reviews.llvm.org/D89077
2020-10-16 12:15:51 -07:00
River Riddle a5ea60456c [mlir] Update SCCP and the Inliner to use SymbolTableCollection for symbol lookups
This transforms the symbol lookups to O(1) from O(NM), greatly speeding up both passes. For a large MLIR module this shaved seconds off of the compilation time.

Differential Revision: https://reviews.llvm.org/D89522
2020-10-16 12:08:48 -07:00
River Riddle 71eeb5ec4d [mlir] Add a new SymbolUserOpInterface class
The initial goal of this interface is to fix the current problems with verifying symbol user operations, but can extend beyond that in the future. The current problems with the verification of symbol uses are:
* Extremely inefficient:
Most current symbol users perform the symbol lookup using the slow O(N) string compare methods, which can lead to extremely long verification times in large modules.
* Invalid/break the constraints of verification pass
If the symbol reference is not-flat(and even if it is flat in some cases) a verifier for an operation is not permitted to touch the referenced operation because it may be in the process of being mutated by a different thread within the pass manager.

The new SymbolUserOpInterface exposes a method `verifySymbolUses` that will be invoked from the parent symbol table to allow for verifying the constraints of any referenced symbols. This method is passed a `SymbolTableCollection` to allow for O(1) lookups of any necessary symbol operation.

Differential Revision: https://reviews.llvm.org/D89512
2020-10-16 12:08:48 -07:00
River Riddle 7bc7d0ac7a [mlir] Optimize symbol related checks in SymbolDCE
This revision contains two optimizations related to symbol checking:
* Optimize SymbolOpInterface to only check for a name attribute if the operation is an optional symbol.
This removes an otherwise unnecessary attribute lookup from a majority of symbols.
* Add a new SymbolTableCollection class to represent a collection of SymbolTables.
This allows for perfoming non-flat symbol lookups in O(1) time by caching SymbolTables for symbol table operations. This class is very useful for algorithms that operate on multiple symbol tables, either recursively or not.

Differential Revision: https://reviews.llvm.org/D89505
2020-10-16 12:08:48 -07:00
River Riddle f3df3b58e7 [mlir] Add a utility class, ThreadLocalCache, for storing non static thread local objects.
(Note: This is a reland of D82597)

This class allows for defining thread local objects that have a set non-static lifetime. This internals of the cache use a static thread_local map between the various different non-static objects and the desired value type. When a non-static object destructs, it simply nulls out the entry in the static map. This will leave an entry in the map, but erase any of the data for the associated value. The current use cases for this are in the MLIRContext, meaning that the number of items in the static map is ~1-2 which aren't particularly costly enough to warrant the complexity of pruning. If a use case arises that requires pruning of the map, the functionality can be added.

This is especially useful in the context of MLIR for implementing thread-local caching of context level objects that would otherwise have very high lock contention. This revision adds a thread local cache in the MLIRContext for attributes, identifiers, and types to reduce some of the locking burden. This led to a speedup of several seconds when compiling a somewhat large mlir module.

Differential Revision: https://reviews.llvm.org/D89504
2020-10-16 12:08:48 -07:00
Richard Smith dd4e8a54b2 [docs] Fix some out-of-date / inaccurate text and missing formatting in the User's Manual. 2020-10-16 12:00:13 -07:00
Fangrui Song 08d6650b7b [llvm-objdump][test] Unsupport system-windows for source-interleave-prefix.test 2020-10-16 11:44:33 -07:00
Adrian Prantl cf24508651 Add support for more OS types to AddClangModuleCompilationOptionsForSDKType()
This patch also avoids hardcoding the clang options, which makes it
less likely for them to become out-of-date.

rdar://problem/63791367+66927829

Differential Revision: https://reviews.llvm.org/D89428
2020-10-16 11:43:22 -07:00
Fangrui Song 0b1b2a180c [llvm-objdump][test] Delete source-interleave-prefix-windows.test
It needs more work: http://45.33.8.238/win/25999/step_11.txt
2020-10-16 11:42:25 -07:00
Amara Emerson 39c05a1a71 [AArch64][GlobalISel] Add selection support for v2s32 and v2s64 reductions for FADD/ADD.
We'll need legalizer lower() support for the other types to work.

Differential Revision: https://reviews.llvm.org/D89159
2020-10-16 11:41:57 -07:00
Arthur Eubanks 53b69820f4 [test] Pin null-function.ll to legacy PM
The NPM prints CGSCCs in a different way.
2020-10-16 11:24:28 -07:00
LLVM GN Syncbot 1c54a91d11 [gn build] Port 3d4bba302d 2020-10-16 18:13:18 +00:00
Nico Weber 98e134f5d0 [gn build] (manually) port 3d4bba302d 2020-10-16 14:12:38 -04:00
Martin Storsjö 3784bdf217 [libcxx] [test] Fix string type handling in a few fairly trivial class.path tests
Use string() for convenience for testing where possible, but keep using
native() for move tests where we want to check that no allocations are
made, constructing a reference fs::path::string_type instead.

Use the right value_type in a few places.

Make the synop test check for the right types and for the expected
preferred separator.

Differential Revision: https://reviews.llvm.org/D89537
2020-10-16 21:04:23 +03:00
Benjamin Kramer b740899c50 [Indvars][NFCI] Simplify assertion.
This should be semantically identical. Also avoids unused variable
warnings in Release builds.
2020-10-16 19:58:55 +02:00
Scott Linder c4d10e7e9b [AMDGPU][HIP] Switch default DWARF version to 5
Another attempt at this, see D59008 for previous attempt.

Reviewed By: kzhuravl, t-tye

Differential Revision: https://reviews.llvm.org/D89484
2020-10-16 17:53:27 +00:00
Teresa Johnson 3ed77ecd0a [MemProf] Don't build memprof if sanitizer not being built
This should fix a couple of debian bot failures:
http://lab.llvm.org:8011/#builders/14/builds/380
http://lab.llvm.org:8011/#builders/109/builds/533

They disable building of the sanitizer, which memprof relies on. So
disable building of memprof in that case.
2020-10-16 10:47:37 -07:00
Amara Emerson 32f77eea2d [AArch64][GlobalISel] Regbankselect reductions to use FPR bank for scalars.
Differential Revision: https://reviews.llvm.org/D89075
2020-10-16 10:42:15 -07:00
Amara Emerson 9190411fcf [AArch64][GlobalISel] Add basic legalizer rules for supported add/fadd reductions.
NEON is pretty limited in it's reduction support. As a first step add some
basic rules for the legal types we can select.

Differential Revision: https://reviews.llvm.org/D89070
2020-10-16 10:35:46 -07:00
Amara Emerson 6042c25b0a [GlobalISel] Add translation support for vector reduction intrinsics.
In order to prevent the ExpandReductions pass from expanding some intrinsics
before they get to codegen, I had to add a -disable-expand-reductions flag
for testing purposes.

Differential Revision: https://reviews.llvm.org/D89028
2020-10-16 10:17:53 -07:00
Louis Dionne 2408fc2a1e [libc++] Avoid relying on `realpath` being installed on the system
It doesn't appear to be a standard utility.
2020-10-16 13:07:18 -04:00
Vinicius Tinti e95f9a23fa [llvm-objdump] Implement --prefix option
The prefix given to --prefix will be added to GNU absolute paths when
used with --source option (source interleaved with the disassembly).

This matches GNU's objdump behavior.

GNU and C++17 rules for absolute paths are different.

Differential Revision: https://reviews.llvm.org/D85024

Fixes PR46368.

Differential Revision: https://reviews.llvm.org/D85024
2020-10-16 17:50:42 +01:00
Teresa Johnson 3d4bba302d [MemProf] Memory profiling runtime support
See RFC for background:
http://lists.llvm.org/pipermail/llvm-dev/2020-June/142744.html

Follow on companion to the clang/llvm instrumentation support in D85948
and committed earlier.

This patch adds the compiler-rt runtime support for the memory
profiling.

Note that much of this support was cloned from asan (and then greatly
simplified and renamed). For example the interactions with the
sanitizer_common allocators, error handling, interception, etc.

The bulk of the memory profiling specific code can be found in the
MemInfoBlock, MemInfoBlockCache, and related classes defined and used
in memprof_allocator.cpp.

For now, the memory profile is dumped to text (stderr by default, but
honors the sanitizer_common log_path flag). It is dumped in either a
default verbose format, or an optional terse format.

This patch also adds a set of tests for the core functionality.

Differential Revision: https://reviews.llvm.org/D87120
2020-10-16 09:47:02 -07:00
Louis Dionne 880fc4d581 [libc++] Move the oss-fuzz script to libcxx/utils/ci 2020-10-16 12:40:43 -04:00
Louis Dionne e1612c3866 [libc++] Move the oss-fuzz script to libc++
Instead of having this script be part of the OSS-Fuzz repository, I think
it makes more sense to have it alongside the rest of the fuzzing targets
in libc++.
2020-10-16 12:23:03 -04:00
Jay Foad 1417abe54c [AMDGPU] Add new llvm.amdgcn.fma.legacy intrinsic
Differential Revision: https://reviews.llvm.org/D89558
2020-10-16 17:10:21 +01:00
Jay Foad 0c1381d795 [llc] Use -filetype=null to disable MIR printing
If you use -stop-after or similar options, llc will normally print MIR.
This patch checks for -filetype=null as a special case to disable MIR
printing. As the comment says, "The Null output is intended for use for
performance analysis ...", and I found this useful for timing a subset
of the passes that llc runs without the significant overhead of printing
MIR just to send it to /dev/null.

Differential Revision: https://reviews.llvm.org/D89476
2020-10-16 16:51:56 +01:00
ahmedsabie 7dff6b818b [MLIR] Add idempotent trait folding
This trait simply adds a fold of f(f(x)) = f(x) when an operation is labelled as idempotent

Reviewed By: rriddle, andyly

Differential Revision: https://reviews.llvm.org/D89421
2020-10-16 15:51:04 +00:00
Matt Arsenault 0a7cd99a70 Reapply "OpaquePtr: Add type to sret attribute"
This reverts commit eb9f7c28e5.

Previously this was incorrectly handling linking of the contained
type, so this merges the fixes from D88973.
2020-10-16 11:05:02 -04:00
Krzysztof Parzyszek 97533b10b2 [Hexagon] Fix license headers in some .td files, NFC 2020-10-16 10:03:05 -05:00
Hans Wennborg 95fb3542e8 Disable DynamicLibraryTests when using LLVM_INTEGRATED_CRT_ALLOC (PR47881) 2020-10-16 16:49:58 +02:00
Simon Pilgrim 83ae625f0c [InstCombine] visitAnd - pull out repeated I.getType() calls. NFCI. 2020-10-16 15:43:11 +01:00
Simon Pilgrim fe8281e2d0 [InstCombine] visitAnd - add some ((val OP C1) & C2) vector test coverage 2020-10-16 15:43:11 +01:00
Simon Pilgrim 253f24cf4c [InstCombine] Remove custom and(trunc(and(x,c1)),c2) fold
This is more correctly handled by canEvaluateTruncated (one use checks etc.) and covers all the tests cases that were added for this fold.
2020-10-16 15:43:10 +01:00
Martin Storsjö ddb4693a92 [libcxx] [test] Add (void) return casts on functions marked nodiscard in MS STL
Differential Revision: https://reviews.llvm.org/D89533
2020-10-16 17:36:56 +03:00
Martin Storsjö 66427d7359 [libcxx] [test] Mark tests for libcxx specific implementation details with LIBCPP_ONLY()
This matches an existing marking in enum.path.format.pass.cpp.

Differential Revision: https://reviews.llvm.org/D89534
2020-10-16 17:36:56 +03:00
Martin Storsjö 701fa703fc [libcxx] [test] Use fs::path::string_type instead of std::string in fs.op.relative
Differential Revision: https://reviews.llvm.org/D89538
2020-10-16 17:36:56 +03:00
Martin Storsjö fdbfff8fd4 [libcxx] [test] Use string() instead of native() as parameter to std::ifstream/ofstream in copy_file_large.pass.cpp
This fixes building with libstdc++ for windows. MS STL has got
ifstream/ofstream overloads that taken wide strings though.

Differential Revision: https://reviews.llvm.org/D89539
2020-10-16 17:36:56 +03:00
Duncan P. N. Exon Smith 59a3b1afb2 clang-format: Assert in-memory file created in createInMemoryFile, NFC
`SourceManager::createFileID` asserts that the given `FileEntry` is not
null, so remove the logic that passed in `nullptr`. Since we just added
the file to an in-memory FS via an API that cannot fail, use
`llvm_unreachable` on the error path. Didn't use an `assert` since it
seems cleaner semantically to check the error (and better,
hypothetically, for updating the API to use `Expected` instead of
`ErrorOr`).

I noticed this incidentally while auditing calls to `createFileID`.
2020-10-16 10:20:32 -04:00
Peter Steinfeld c757418869 [flang] Failed call to CHECK() for call to ASSOCIATED(NULL())
Calling "ASSOCATED(NULL()) was causing an internal check of the compiler to
fail.

I fixed this by changing the entry for "ASSOCIATED" in the intrinsics table to
accept "AnyPointer" which contains a new "KindCode" of "pointerType".  I also
changed the function "FromActual()" to return a typeless intrinsic when called
on a pointer, which duplicates its behavior for BOZ literals.  This required
changing the analysis of procedure arguments.  While testing processing for
procedure arguments, I found another bad call to `CHECK()` which I fixed.

I made several other changes:
  -- I implemented constant folding for ASSOCIATED().
  -- I fixed handling of NULL() in relational operations.
  -- I implemented semantic analysis for ASSOCIATED().
    -- I noticed that the semantics for ASSOCIATED() are similar to those for
       pointer assignment.  So I extracted the code that pointer assignment uses
       for procedure pointer compatibility to a place where it could be used by
       the semantic analysis for ASSOCIATED().
    -- I couldn't figure out how to make the general semantic analysis for
       procedure arguments work with ASSOCIATED()'s second argument, which can
       be either a pointer or a target.  So I stopped using normal semantic
       analysis for arguments for ASSOCIATED().
  -- I added tests for all of this.

Differential Revision: https://reviews.llvm.org/D88313
2020-10-16 07:12:57 -07:00
Matt Arsenault ee6e25e439 llvm-reduce: Don't replace intrinsic calls with undef
These don't really have function bodies to try to eliminate. This also
has a good chance of just producing invalid IR since intrinsics can
have special operand constraints (e.g. metadata arguments aren't valid
for an arbitrary call). This was wasting quite a bit of time producing
and failing on invalid IR when replacing dbg.values with undefs.
2020-10-16 10:10:53 -04:00