platform/upstream/llvm.git
2 years ago[AMDGPU] Remove now unused variable HasLdsModifier. NFC.
Stanislav Mekhanoshin [Wed, 27 Apr 2022 00:45:41 +0000 (17:45 -0700)]
[AMDGPU] Remove now unused variable HasLdsModifier. NFC.

2 years ago[CUDA][HIP] Fix linkage of __clang_gpu_used_external
Yaxun (Sam) Liu [Tue, 26 Apr 2022 17:06:53 +0000 (13:06 -0400)]
[CUDA][HIP] Fix linkage of __clang_gpu_used_external

Different TU's may have this globl var. appending linkage can
only be used with lld recognized special variables.

Change it to internal linkage.

Reviewed by: Artem Belevich

Differential Revision: https://reviews.llvm.org/D124466

2 years ago[AMDGPU] Add both mayLoad and mayStore to MUBUF LDS opcodes
Stanislav Mekhanoshin [Tue, 26 Apr 2022 22:00:42 +0000 (15:00 -0700)]
[AMDGPU] Add both mayLoad and mayStore to MUBUF LDS opcodes

Differential Revision: https://reviews.llvm.org/D124483

2 years ago[AMDGPU] Remove vdata from buffer to lds load
Stanislav Mekhanoshin [Tue, 26 Apr 2022 21:58:42 +0000 (14:58 -0700)]
[AMDGPU] Remove vdata from buffer to lds load

Differential Revision: https://reviews.llvm.org/D124485

2 years agoFix [libc] Minor test signedness fixes
Dominic Chen [Wed, 27 Apr 2022 00:14:39 +0000 (17:14 -0700)]
Fix [libc] Minor test signedness fixes

Add namespace specifier

Differential Revision: https://reviews.llvm.org/D124468

2 years ago[libc] Minor test signedness fixes
Dominic Chen [Tue, 26 Apr 2022 17:25:05 +0000 (10:25 -0700)]
[libc] Minor test signedness fixes

Resolve compiler warnings about signed conversion

Differential Revision: https://reviews.llvm.org/D124468

2 years agoFix missing import for test
Jeffrey Tan [Tue, 26 Apr 2022 21:29:10 +0000 (14:29 -0700)]
Fix missing import for test

The last fix missed an import in one test file causing skipIfWindows attribute
can't be recognized.
I feel embarrassed to miss it. I have run all tests on Mac to make sure them
passing in this patch.

Differential Revision: https://reviews.llvm.org/D124479

2 years ago[NFC] Fix argument types in doxygen comment
Warren Ristow [Tue, 26 Apr 2022 23:29:33 +0000 (16:29 -0700)]
[NFC] Fix argument types in doxygen comment

The argument types for _mm_extract_epi16 and _mm_insert_epi16 were
incorrectly identified as '__m256i' instead of '__m128i'.  Introduced
by df08b3493869540bad5d4b040dae814e078b411d, and fixed here.

2 years ago[AIX][XCOFF] error on emit symbol visibility for XCOFF object file
David Tenty [Fri, 22 Apr 2022 18:37:17 +0000 (14:37 -0400)]
[AIX][XCOFF] error on emit symbol visibility for XCOFF object file

This is a follow on to the revert of D84265 to add an error if we'd need
to write a non-zero visibility type in the xcoff object file. We can't
currently do that because we lack the auxilary header to interpret the
bits in XCOFF32. This is important because visibility is being enabled
in the assembly writing path, and without this error the visibility
could be silently ignored.

Differential Revision: https://reviews.llvm.org/D124392

2 years ago[SimplifyLibCalls] avoid slicing 64-bit integers in an ILP32 build (PR #54739)
Martin Sebor [Tue, 26 Apr 2022 22:26:09 +0000 (16:26 -0600)]
[SimplifyLibCalls] avoid slicing 64-bit integers in an ILP32 build (PR #54739)

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D123472

2 years ago[BOLT][NFC] Fix warning for unqualified call to std::move
Paul Kirth [Tue, 26 Apr 2022 23:17:26 +0000 (23:17 +0000)]
[BOLT][NFC] Fix warning for unqualified call to std::move

Fixes warning from RetpolineInsertion.cpp:171:44:
warning: unqualified call to std::move [-Wunqualified-std-cast-call]

Reviewed By: maksfb

Differential Revision: https://reviews.llvm.org/D124482

2 years ago[mlir][sparse] support pattern-only matrices from Matrix Market
Aart Bik [Tue, 26 Apr 2022 20:30:57 +0000 (13:30 -0700)]
[mlir][sparse] support pattern-only matrices from Matrix Market

We simply set nonzero entries to the value "1" in this case.

Reviewed By: bixia

Differential Revision: https://reviews.llvm.org/D124475

2 years ago[InstCombine] Fold strnlen of constant strings.
Martin Sebor [Tue, 26 Apr 2022 22:12:57 +0000 (16:12 -0600)]
[InstCombine] Fold strnlen of constant strings.

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D123817

2 years agotry to fix check-llvm on windows after 0a27622a1d62
Nico Weber [Tue, 26 Apr 2022 22:00:37 +0000 (18:00 -0400)]
try to fix check-llvm on windows after 0a27622a1d62

2 years ago[NFC] More synching of sources for upstreaming.
Eric Schweitz [Tue, 26 Apr 2022 20:44:34 +0000 (13:44 -0700)]
[NFC] More synching of sources for upstreaming.

Reviewed By: clementval

Differential Revision: https://reviews.llvm.org/D124476

2 years ago[NFC] Cleanup miscellaneous header items
Warren Ristow [Tue, 26 Apr 2022 21:33:14 +0000 (14:33 -0700)]
[NFC] Cleanup miscellaneous header items

- Explain the use of the _MM_SHUFFLE and _MM_SHUFFLE2 macros
- Update some doxygen parameter descriptions to match the implementations
- Add "see also" doxygen tags to some intrinsics
- Minor clang-format changes

Reviewers: RKSimon

Differential Revision: https://reviews.llvm.org/D124469

2 years ago[InstCombine] Update predicate when canonicalizing comparisons in canonicalizeClampLike.
Ricky Zhou [Tue, 26 Apr 2022 21:26:54 +0000 (17:26 -0400)]
[InstCombine] Update predicate when canonicalizing comparisons in canonicalizeClampLike.

canonicalizeClampLike canonicalizes the ule/ugt comparisons to ult/uge,
respectively. However, it does not update the variable holding the
comparison predicate type after doing this. Later code fails to handle
the non-canonical predicate type (specifically, the swap of
ThresholdLowIncl and ThresholdHighExcl when Pred0 has been canonicalized
from ugt to uge). This leads to the miscompile reported in PR53252. Fix
this by updating the comparison predicate after canonicalizing.

Fixes #53252

Differential Revision: https://reviews.llvm.org/D119690

2 years ago[OpenMPIRBuilder] Remove ContinuationBB argument from Body callback.
Michael Kruse [Tue, 26 Apr 2022 18:35:13 +0000 (13:35 -0500)]
[OpenMPIRBuilder] Remove ContinuationBB argument from Body callback.

The callback is expected to create a branch to the ContinuationBB (sometimes called FiniBB in some lambdas) argument when finishing. This creates problems:

 1. The InsertPoint used for CodeGenIP does not need to be the end of a block. If it is not, a naive callback will insert a branch instruction into the middle of the block.

 2. The BasicBlock the CodeGenIP is pointing to may or may not have a terminator. There is an conflict where to branch to if the block already has a terminator.

 3. Some API functions work only with block having a terminator. Some workarounds have been used to insert a temporary terminator that is removed again.

 4. Some callbacks are sensitive to whether the BasicBlock has a terminator or not. This creates a callback ordering problem where different callback may have different behaviour depending on whether a previous callback created a terminator or not. The problem also exists for FinalizeCallbackTy where some callbacks do create branch to another "continue" block, but unlike BodyGenCallbackTy does not receive the target as argument. This is not addressed in this patch.

With this patch, the callback receives an CodeGenIP into a BasicBlock where to insert instructions. If it has to insert control flow, it can split the block at that position as needed but otherwise no separate ContinuationBB is needed. In particular, a callback can be empty without breaking the emitted IR. If the caller needs the control flow to branch to a specific target, it can insert the branch instruction itself and pass an InsertPoint before the terminator to the callback.

Certain frontends such as Clang may expect the current IRBuilder position to be at the end of a basic block. In this case its callbacks must split the block at CodeGenIP before setting the IRBuilder position such that the instructions after CodeGenIP are moved to another basic block and before returning create a new branch instruction to the split block.

Some utility functions such as `splitBB` are supporting correct splitting of BasicBlocks, independent of whether they have a terminator or not, returning/setting the InsertPoint of an IRBuilder to the end of split predecessor block, and optionally omitting creating a branch to the split successor block to be added later.

Reviewed By: kiranchandramohan

Differential Revision: https://reviews.llvm.org/D118409

2 years ago[InstCombine] Add tests reproducing PR53252 (NFC)
Ricky Zhou [Tue, 26 Apr 2022 21:21:27 +0000 (17:21 -0400)]
[InstCombine] Add tests reproducing PR53252 (NFC)

This are baseline tests for D119690 ( #53252 )

Differential Revision: https://reviews.llvm.org/D119689

2 years agoRevert "[compiler-rt] Use C_STANDARD instead of custom logic for adding -std=c11"
Martin Storsjö [Tue, 26 Apr 2022 21:19:14 +0000 (00:19 +0300)]
Revert "[compiler-rt] Use C_STANDARD instead of custom logic for adding -std=c11"

This reverts commit 2f251925db060f83894a4cf5448259b7a1255552.

This was reported to break the Builtins-powerpc64le-linux::qdiv_test.c
test on ppc64le buildbots, e.g. these:

https://lab.llvm.org/buildbot/#/builders/105/builds/24538
https://lab.llvm.org/buildbot/#/builders/121/builds/18755

2 years ago[AMDGPU] Properly mark MUBUF and FLAT LDS DMA instructions. NFC.
Stanislav Mekhanoshin [Tue, 26 Apr 2022 19:34:27 +0000 (12:34 -0700)]
[AMDGPU] Properly mark MUBUF and FLAT LDS DMA instructions. NFC.

Add these bits to the MUBUF and FLAT LDS DMA instructions:

- LGKM_CNT - these operate on LDS;
- VALU - SPG 3.9.8: This instruction acts as both a MUBUF and
VALU instruction;

Codegen currently does not produce any of this, so the change is NFC.

Differential Revision: https://reviews.llvm.org/D124472

2 years agoRecommit "[SLP][TTI] Refactoring of `getShuffleCost` `Args` to work like `getArithmet...
Vasileios Porpodas [Tue, 26 Apr 2022 18:32:24 +0000 (11:32 -0700)]
Recommit "[SLP][TTI] Refactoring of `getShuffleCost` `Args` to work like `getArithmeticInstrCost`"

This reverts commit 6a9bbd9f20dcd700e28738788bb63a160c6c088c.

Code review: https://reviews.llvm.org/D124202

2 years ago[PS5] Add debug option testing for the PS5 target
Paul Robinson [Tue, 26 Apr 2022 20:36:22 +0000 (13:36 -0700)]
[PS5] Add debug option testing for the PS5 target

The functionality was added in 7726ad0, catching up on the tests now.

This test mostly runs clang and checks one thing each time, which isn't
very efficient.  I added some -DAG suffixes and combined some checks so
adding PS5 checking didn't actually increase the number of clang runs.

2 years agoDisable symbol on-demand feature for Windows
Jeffrey Tan [Tue, 26 Apr 2022 19:13:49 +0000 (12:13 -0700)]
Disable symbol on-demand feature for Windows

Symbol on-demand feature is never tested on Windows so it is not a surprise
that we are getting Buildbot failure from Windows:
https://lab.llvm.org/buildbot/#/builders/83/builds/18228

This patch disables symbol on-demand feature on Windows. I will find a Windows
machine to test and re-enable symbol on-demand feature as follow-up.

Differential Revision: https://reviews.llvm.org/D124471

2 years agoRevert "[demangler] Simplify OutputBuffer initialization"
Kirill Stoimenov [Tue, 26 Apr 2022 20:24:06 +0000 (20:24 +0000)]
Revert "[demangler] Simplify OutputBuffer initialization"

Reverting due to a bot failure:
https://lab.llvm.org/buildbot/#/builders/5/builds/22738

This reverts commit 5b3ca24a35e91bf9c19af856e7f92c69b17f989e.

2 years ago[InstCombine] Fold memrchr calls with a constant character.
Martin Sebor [Tue, 26 Apr 2022 19:15:03 +0000 (13:15 -0600)]
[InstCombine] Fold memrchr calls with a constant character.

Reviewed By: nikic

Differential Revision: //reviews.llvm.org/D123629

2 years ago[InstCombine] Fold memrchr calls with constant size, bail on excessive.
Martin Sebor [Tue, 26 Apr 2022 17:52:15 +0000 (11:52 -0600)]
[InstCombine] Fold memrchr calls with constant size, bail on excessive.

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D123626
Differential Revision: https://reviews.llvm.org/D123628

2 years ago[InstCombine] Fold strnlen with a bound of zero and one.
Martin Sebor [Tue, 26 Apr 2022 17:48:43 +0000 (11:48 -0600)]
[InstCombine] Fold strnlen with a bound of zero and one.

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D123816

2 years ago[InstCombine] add a strnlen handler stub.
Martin Sebor [Tue, 26 Apr 2022 17:35:59 +0000 (11:35 -0600)]
[InstCombine] add a strnlen handler stub.

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D123815

2 years ago[NVPTX] Temporary disable dwarf-file-dir.ll on Windows
Andrew Savonichev [Tue, 26 Apr 2022 19:22:38 +0000 (22:22 +0300)]
[NVPTX] Temporary disable dwarf-file-dir.ll on Windows

The test was added in D121299 and it fails on Windows:

    error: CHECK-NODIR: expected string
    not found in input
    ; CHECK-NODIR: .file {{[0-9]+}} "/tmp/dbginfo/a/a.cpp"
   ^
    <stdin>:1:1: note: scanning from here
    //
    ^
    <stdin>:25:2: note: possible intended match here
     .file 1 "/tmp/dbginfo/a\\a.cpp"

2 years ago[NFC] Fix order of initialization in MCTargetOptions
Andrew Savonichev [Tue, 26 Apr 2022 19:05:32 +0000 (22:05 +0300)]
[NFC] Fix order of initialization in MCTargetOptions

This patch fixes a compiler warning after D121299:

field 'MCUseDwarfDirectory' will be initialized after field
'MCIncrementalLinkerCompatible'

2 years ago[EarlyCSE] Add tests demonstrating missed opportunitites
Artur Pilipenko [Tue, 26 Apr 2022 18:58:08 +0000 (11:58 -0700)]
[EarlyCSE] Add tests demonstrating missed opportunitites

Add tests demonstrating missed opportunitites around
invariant.start intrinsic.

NFC.

2 years ago[NVPTX] Disable DWARF .file directory for PTX
Andrew Savonichev [Tue, 26 Apr 2022 15:45:09 +0000 (18:45 +0300)]
[NVPTX] Disable DWARF .file directory for PTX

Default behavior for .file directory was changed in D105856, but
ptxas (CUDA 11.5 release) refuses to parse it:

    $ llc -march=nvptx64 llvm/test/DebugInfo/NVPTX/debug-file-loc.ll
    $ ptxas debug-file-loc.s
    ptxas debug-file-loc.s, line 42; fatal   : Parsing error near
    '"foo.h"': syntax error

Added a new field to MCAsmInfo to control default value of
UseDwarfDirectory. This value is used if -dwarf-directory command line
option is not specified.

Differential Revision: https://reviews.llvm.org/D121299

2 years ago[InstCombine] try to fold icmp with mismatched extended operands
Sanjay Patel [Tue, 26 Apr 2022 18:22:16 +0000 (14:22 -0400)]
[InstCombine] try to fold icmp with mismatched extended operands

If a value is known to be non-negative and zexted,
that's the same thing as sexted.

So for the purpose of looking past the casts with
an icmp, treat it as if it was a sext:
https://alive2.llvm.org/ce/z/_BDsGV

This is necessary, but not enough to solve the
motivating problem:
https://github.com/llvm/llvm-project/issues/55013

Differential Revision: https://reviews.llvm.org/D124419

2 years agoRevert "[SLP][TTI] Refactoring of `getShuffleCost` `Args` to work like `getArithmetic...
Vasileios Porpodas [Tue, 26 Apr 2022 18:24:48 +0000 (11:24 -0700)]
Revert "[SLP][TTI] Refactoring of `getShuffleCost` `Args` to work like `getArithmeticInstrCost`"

This reverts commit 55ce296d6f217fd0defed2592ff7b74b79b2c1f0.

2 years ago[InstCombine] fold freeze of partial undef/poison vector constants
Sanjay Patel [Tue, 26 Apr 2022 17:59:32 +0000 (13:59 -0400)]
[InstCombine] fold freeze of partial undef/poison vector constants

We can always replace the undef elements in a vector constant
with regular constants to get rid of the freeze:
https://alive2.llvm.org/ce/z/nfRb4F

The select diffs show that we might do better by adjusting the
logic for a frozen select condition. We may also want to refine
the vector constant replacement to consider forming a splat.

Differential Revision: https://reviews.llvm.org/D123962

2 years ago[RISCV] Add more out of range immediate tests to rv64zbp-invalid.s. NFC
Craig Topper [Tue, 26 Apr 2022 18:02:34 +0000 (11:02 -0700)]
[RISCV] Add more out of range immediate tests to rv64zbp-invalid.s. NFC

This tests the Zbp instructions that change immediate range between
riscv32 and riscv64. We were only testing them for riscv32 before.

2 years ago[SLP][TTI] Refactoring of `getShuffleCost` `Args` to work like `getArithmeticInstrCost`
Vasileios Porpodas [Thu, 21 Apr 2022 17:39:08 +0000 (10:39 -0700)]
[SLP][TTI] Refactoring of `getShuffleCost` `Args` to work like `getArithmeticInstrCost`

Before this patch `Args` was used to pass a broadcat's arguments by SLP.
This patch changes this. `Args` is now used for passing the operands of
the shuffle.

Differential Revision: https://reviews.llvm.org/D124202

2 years agoAttributes: add a new `allocptr` attribute
Augie Fackler [Thu, 24 Feb 2022 15:40:20 +0000 (10:40 -0500)]
Attributes: add a new `allocptr` attribute

This continues the push away from hard-coded knowledge about functions
towards attributes. We'll use this to annotate free(), realloc() and
cousins and obviate the hard-coded list of free functions.

Differential Revision: https://reviews.llvm.org/D123083

2 years ago[RISCV] Add out of range immediate tests for shifts and rotates on rv64.
Craig Topper [Tue, 26 Apr 2022 17:42:58 +0000 (10:42 -0700)]
[RISCV] Add out of range immediate tests for shifts and rotates on rv64.

We were only testing the shared instruction mnemonics in rv32 tests.

Reviewed By: reames

Differential Revision: https://reviews.llvm.org/D124460

2 years ago[flang] Fix crash from PDT component init in module file
Peter Klausler [Mon, 25 Apr 2022 23:00:01 +0000 (16:00 -0700)]
[flang] Fix crash from PDT component init in module file

Semantics now needs to preserve the parse trees from module files,
in case they contain parameterized derived type definitions with
component initializers that may require re-analysis during PDT
instantiation.  Save them in the SemanticsContext.

Differential Revision: https://reviews.llvm.org/D124467

2 years ago[AMDGPU] Precommit a test case for D124450
Jay Foad [Tue, 26 Apr 2022 16:01:02 +0000 (17:01 +0100)]
[AMDGPU] Precommit a test case for D124450

2 years ago[gn build] Port 7b81192d462b
LLVM GN Syncbot [Tue, 26 Apr 2022 17:42:25 +0000 (17:42 +0000)]
[gn build] Port 7b81192d462b

2 years agoIntroduce new symbol on-demand for debug info
Jeffrey Tan [Wed, 20 Apr 2022 14:30:53 +0000 (07:30 -0700)]
Introduce new symbol on-demand for debug info

This diff introduces a new symbol on-demand which skips
loading a module's debug info unless explicitly asked on
demand. This provides significant performance improvement
for application with dynamic linking mode which has large
number of modules.
The feature can be turned on with:
"settings set symbols.load-on-demand true"

The feature works by creating a new SymbolFileOnDemand class for
each module which wraps the actual SymbolFIle subclass as member
variable. By default, most virtual methods on SymbolFileOnDemand are
skipped so that it looks like there is no debug info for that module.
But once the module's debug info is explicitly requested to
be enabled (in the conditions mentioned below) SymbolFileOnDemand
will allow all methods to pass through and forward to the actual SymbolFile
which would hydrate module's debug info on-demand.

In an internal benchmark, we are seeing more than 95% improvement
for a 3000 modules application.

Currently we are providing several ways to on demand hydrate
a module's debug info:
* Source line breakpoint: matching in supported files
* Stack trace: resolving symbol context for an address
* Symbolic breakpoint: symbol table match guided promotion
* Global variable: symbol table match guided promotion

In all above situations the module's debug info will be on-demand
parsed and indexed.

Some follow-ups for this feature:
* Add a command that allows users to load debug info explicitly while using a
  new or existing command when this feature is enabled
* Add settings for "never load any of these executables in Symbols On Demand"
  that takes a list of globs
* Add settings for "always load the the debug info for executables in Symbols
  On Demand" that takes a list of globs
* Add a new column in "image list" that shows up by default when Symbols On
  Demand is enable to show the status for each shlib like "not enabled for
  this", "debug info off" and "debug info on" (with a single character to
  short string, not the ones I just typed)

Differential Revision: https://reviews.llvm.org/D121631

2 years ago[AArch64][NFC] Deleted llvm/test/Analysis/CostModel/AArch64/splat-load.ll test
Vasileios Porpodas [Tue, 26 Apr 2022 14:33:42 +0000 (07:33 -0700)]
[AArch64][NFC] Deleted llvm/test/Analysis/CostModel/AArch64/splat-load.ll test

This test is no longer necessary as it is a subset of:
llvm/test/Analysis/CostModel/AArch64/shuffle-load.ll

Differential Revision: https://reviews.llvm.org/D124456

2 years agollvm-reduce: Simplify virtual register cloning
Matt Arsenault [Sat, 16 Apr 2022 23:42:23 +0000 (19:42 -0400)]
llvm-reduce: Simplify virtual register cloning

Just clone all the virtual registers instead of looking for def
operands. This preserves the register values used, simplifying the
rest of the code. This avoids needing to expose the register map to
target code.

2 years agollvm-reduce: Preserve frame index values when cloning function
Matt Arsenault [Sat, 16 Apr 2022 21:17:41 +0000 (17:17 -0400)]
llvm-reduce: Preserve frame index values when cloning function

Previously the specific values used for fixed frame indexes was in
reverse order in the cloned function from the original, and a map was
used to adjust all frame indexes to the potentially new values. Insert
the fixed objects in reverse to avoid this. This simplifies other
code, since now we don't need to track down all frame indexes. This
will allow targets that store frame indexes in MachineFunctionInfo to
simply copy the values.

Note this isn't directly observable in the test since the resulting
MIR print/parse can shuffle the IDs around (in particular the final
serialization implicitly strips out dead objects).

2 years agoSupport non identity layout map for reshape ops in MemRefToLLVM lowering
Yi Zhang [Tue, 19 Apr 2022 00:50:30 +0000 (20:50 -0400)]
Support non identity layout map for reshape ops in MemRefToLLVM lowering

This change borrows the ideas from `computeExpanded/CollapsedLayoutMap`
and computes the dynamic strides at runtime for the memref descriptors.

Differential Revision: https://reviews.llvm.org/D124001

2 years ago[flang] Get ppc64le build bot back up
Peter Klausler [Tue, 26 Apr 2022 00:36:05 +0000 (17:36 -0700)]
[flang] Get ppc64le build bot back up

A recent change assumed that the native C++ "long double" maps to
a Fortran data type; but this turns out to not be true for ppc64le,
which uses "double-double" for "long double".

This is a quick patch to get the ppc64le flang build bot back up.
A better fix that either uses HostTypeExists<> or replaces "long double"
with "ieee128_t" (or some other solution) is expected to follow soon.

Differential Revision: https://reviews.llvm.org/D124423

2 years agoConsistently use the same apostrophe in these docs.
Tanya Lattner [Tue, 26 Apr 2022 16:29:44 +0000 (09:29 -0700)]
Consistently use the same apostrophe in these docs.

2 years ago[NFC][RISCV][CodeGen] Use ArrayRef in TargetLowering functions
Shao-Ce SUN [Tue, 26 Apr 2022 15:36:42 +0000 (23:36 +0800)]
[NFC][RISCV][CodeGen] Use ArrayRef in TargetLowering functions

Based on D123467.

Reviewed By: reames

Differential Revision: https://reviews.llvm.org/D123653

2 years ago[NFC][libc++] Fixes some tests on Linux.
Mark de Wever [Sat, 23 Apr 2022 16:19:58 +0000 (18:19 +0200)]
[NFC][libc++] Fixes some tests on Linux.

The tests are adapted to the output produced on Linux.

Reviewed By: #libc, philnik, ldionne

Differential Revision: https://reviews.llvm.org/D124331

2 years ago[InstCombine] Add extra use tests for GEP of GEP fold (NFC)
Nikita Popov [Tue, 26 Apr 2022 15:36:39 +0000 (17:36 +0200)]
[InstCombine] Add extra use tests for GEP of GEP fold (NFC)

2 years ago[TableGen] Dump RC.AllocationPriority with -register-info-debug
Jay Foad [Tue, 26 Apr 2022 15:29:06 +0000 (16:29 +0100)]
[TableGen] Dump RC.AllocationPriority with -register-info-debug

2 years ago[SPIRV][NFC] Remove unused variable
Chris Bieneman [Tue, 26 Apr 2022 14:58:16 +0000 (09:58 -0500)]
[SPIRV][NFC] Remove unused variable

This removes an unused local variable that was causing a warning to be
emitted.

2 years ago[SystemZ] Fix C++ ABI for passing args of structs containing zero width bitfield.
Jonas Paulsson [Thu, 24 Mar 2022 11:41:34 +0000 (12:41 +0100)]
[SystemZ] Fix C++ ABI for passing args of structs containing zero width bitfield.

A struct like { float a; int :0; } should per the SystemZ ABI be passed in a
GPR, but to match a bug in GCC it has been passed in an FPR (see 759449c).

GCC has now corrected the C++ ABI for this case, and this patch for clang
follows suit.

Reviewed By: Ulrich Weigand

Differential Revision: https://reviews.llvm.org/D122388

2 years ago[InstCombine] Add tests for GEP of GEP with opaque pointers (NFC)
Nikita Popov [Tue, 26 Apr 2022 15:04:25 +0000 (17:04 +0200)]
[InstCombine] Add tests for GEP of GEP with opaque pointers (NFC)

Test cases with element type mismatch, some of which can still
be combined.

2 years ago[SPIRV][NFC] Fix warnings for switch cases
Chris Bieneman [Tue, 26 Apr 2022 14:46:50 +0000 (09:46 -0500)]
[SPIRV][NFC] Fix warnings for switch cases

Switch statements that cover all cases should not have a `default`
case. When a switch covers all cases and includes a `default` case,
clang emits a diagnostic. Omitting the `default` case allows the
compiler to instead emit a diagnostic on unhandled enum values.

This change removes default cases from all the places that they
shouldn't be, and adds a missing enum case for one switch statement
that wasn't covering all values.

2 years ago[SPIRV][NFC] Fix warning on class/struct mismatch
Chris Bieneman [Tue, 26 Apr 2022 14:46:07 +0000 (09:46 -0500)]
[SPIRV][NFC] Fix warning on class/struct mismatch

Clang issues a warning on class/struct mismatch because the MSVC ABI
varries for classes and structs.

2 years ago[OpenMP] Fix incorrect path taken when searching for LLD for offloading
Joseph Huber [Tue, 26 Apr 2022 14:49:31 +0000 (10:49 -0400)]
[OpenMP] Fix incorrect path taken when searching for LLD for offloading

Summary:
A previous patch updated the path searching in the linker wrapper. I
made an error and caused `lld`, which is necessary to link AMDGPU
images, to not be found on some systems. This patch fixes this by
correctly searching that linker-wrapper's binary path first again.

2 years agoReapply D124184, [DebugInfo][InstrRef] Add a size operand to DBG_PHI
Jeremy Morse [Tue, 26 Apr 2022 10:08:05 +0000 (11:08 +0100)]
Reapply D124184, [DebugInfo][InstrRef] Add a size operand to DBG_PHI

This was reverted twice, in 987cd7c3ed75b and 13815e8cbf8d4. The latter
stemed from not accounting for rare register classes in a pre-allocated
array, and the former from an array not being completely initialized,
leading to asan complaining.

2 years ago[LoopPeel][NFCI] Simplify the code to calculate peel count for PGO
Igor Kudrin [Tue, 26 Apr 2022 14:44:24 +0000 (18:44 +0400)]
[LoopPeel][NFCI] Simplify the code to calculate peel count for PGO

This reorganizes the code as a preparation for D123865:

 * Use more descriptive names for variables
 * Simplify a condition by use an already calculated value
   for `MaxPeelCount`
 * Remove a duplicate log entry
 * Report basic values for loop costs

Differential Revision: https://reviews.llvm.org/D124388

2 years ago[LoopPeel][NFC] Exit early if there is no room for peeling
Igor Kudrin [Tue, 26 Apr 2022 14:43:56 +0000 (18:43 +0400)]
[LoopPeel][NFC] Exit early if there is no room for peeling

Differential Revision: https://reviews.llvm.org/D123864

2 years ago[HIP] Fix diag msg about sanitizer
Yaxun (Sam) Liu [Mon, 25 Apr 2022 15:34:08 +0000 (11:34 -0400)]
[HIP] Fix diag msg about sanitizer

Fix the misleading diag msg as the option is ignored for a particular
offload arch only.

Reviewed by: Artem Belevich

Differential Revision: https://reviews.llvm.org/D124396

2 years ago[NVPTX][tests] add "XFAIL: nvptx" for some tests
Igor Chebykin [Tue, 26 Apr 2022 14:23:17 +0000 (17:23 +0300)]
[NVPTX][tests] add "XFAIL: nvptx" for some tests

Some tests failed for NVPTX target, but it seems that NVPTX will be
fixed and the tests will pass. So, just mark the tests as XFAIL

Differential Revision: https://reviews.llvm.org/D124125

2 years ago[NVPTX][tests] Do not run the tests which are not supported by nvptx
Igor Chebykin [Tue, 26 Apr 2022 14:15:50 +0000 (17:15 +0300)]
[NVPTX][tests] Do not run the tests which are not supported by nvptx

Some generic tests are not supported by the nvptx now.  Moreover, they
are no plans to fix the tested features in nvptx. So, suggest to mark
them as UNSUPPORTED

Differential Revision: https://reviews.llvm.org/D123928

2 years ago[clang-format][docs] Fix incorrect 'clang-format 9' option marker
Krystian Kuzniarek [Thu, 7 Apr 2022 11:11:02 +0000 (13:11 +0200)]
[clang-format][docs] Fix incorrect 'clang-format 9' option marker

Introduced by 23a5090c6, this style option marker indicated
'clang-format 9', though its respective option was available in
an earlier release.

Differential Revision: https://reviews.llvm.org/D123299

2 years ago[PS5] Correctly test for analyzer defaults
Paul Robinson [Tue, 26 Apr 2022 14:14:27 +0000 (07:14 -0700)]
[PS5] Correctly test for analyzer defaults

Commit 5c90eca added some analyzer option checking, but a typo meant
it was redundantly checking PS4 and not adding checking for PS5.
With the test corrected, it identified the necessary driver updates,
added in this commit.

2 years ago[Index] [clangd] Support for concept declarations and requires expressions
Ilya Biryukov [Tue, 26 Apr 2022 13:38:28 +0000 (13:38 +0000)]
[Index] [clangd] Support for concept declarations and requires expressions

Add support for concepts and requires expression in the clang index.
Genarate USRs for concepts.

Also change how `RecursiveASTVisitor` handles return type requirement in
requires expressions. The new code unpacks the synthetic template parameter
list used for storing the actual expression. This simplifies
implementation of the indexing. No code seems to depend on the original
traversal anyway and the synthesized template parameter list is easily
accessible from inside the requires expression if needed.

Add tests in the clangd codebase.

Fixes https://github.com/clangd/clangd/issues/1103.

Reviewed By: sammccall

Differential Revision: https://reviews.llvm.org/D124441

2 years agoFix test from 5ff992 to work with win 32 bit, fix a typo.
Erich Keane [Tue, 26 Apr 2022 12:50:33 +0000 (05:50 -0700)]
Fix test from 5ff992 to work with win 32 bit, fix a typo.

See the discussion on D123319.  Default calling convention on Windows32
bit is to have 'this-call' for member functions, so add a wildcard to
take care of that.

Also, one of the FileCheck 'match' flags was incorrect, so fix that too.

2 years ago[ConstraintElimination] Check if const. is small enough before using it
Florian Hahn [Tue, 26 Apr 2022 12:16:02 +0000 (13:16 +0100)]
[ConstraintElimination] Check if const. is small enough before using it

Check if the value of a ConstantInt is small enough to be used for
solving before calling getSExtValue.

Fixes #55085

2 years ago[llvm][Inline] Remove PriorityInlineOrder in SCC inliner
Liqiang Tao [Tue, 26 Apr 2022 12:16:37 +0000 (20:16 +0800)]
[llvm][Inline] Remove PriorityInlineOrder in SCC inliner

Since the size of most of SCC's is 1, the PriorityInlineOrder would not change the inline
order in SCC inliner.

Reviewed By: kazu

Differential Revision: https://reviews.llvm.org/D123608

2 years ago[mlir] Fix -Wunused-private-field in the Transform dialect
Alex Zinenko [Tue, 26 Apr 2022 11:57:03 +0000 (13:57 +0200)]
[mlir] Fix -Wunused-private-field in the Transform dialect

Classes derived from TransformState::Extension may need access to the
parent state.

2 years ago[IPSCCP] Support unfeasible default dests for switch.
Florian Hahn [Tue, 26 Apr 2022 11:41:41 +0000 (12:41 +0100)]
[IPSCCP] Support unfeasible default dests for switch.

At the moment, unfeasible default destinations are not handled properly
in removeNonFeasibleEdges. So far, only unfeasible cases are removed,
but later code expects unreachable blocks to have no predecessors.

This is causing the crash reported in PR49573.

If the default destination is unfeasible it won't be executed. Create
a new unreachable block on demand and use that as default
destination.

Note that at the moment this only is relevant for cases where
resolvedUndefsIn marks the first case as executable. Regular switch
handling has a FIXME/TODO to support determining whether the default
case is feasible or not.

Fixes #48917.

Differential Revision: https://reviews.llvm.org/D113497

2 years ago[demangler] Simplify OutputBuffer initialization
Nathan Sidwell [Mon, 28 Feb 2022 18:13:44 +0000 (10:13 -0800)]
[demangler] Simplify OutputBuffer initialization

Every non-testcase use of OutputBuffer contains code to allocate an
initial buffer (using either 128 or 1024 as initial guesses). There's
now no need to do that, given recent changes to the buffer extension
heuristics -- it allocates a 1k(ish) buffer on first need.

Just pass in a buffer (if any) to the constructor.  Thus the
OutputBuffer's ownership of the buffer starts at its own lifetime
start. We can reduce the lifetime of this object in several cases.

That new constructor takes a 'size_t *' for the size argument, as all
uses with a non-null buffer are passing through a malloc'd buffer from
their own caller in this manner.

The buffer reset member function is never used, and is deleted.

The original buffer initialization code would return a failure code if
that first malloc failed.  Existing code either ignored that, called
std::terminate with a FIXME, or returned an error code.

But that's not foolproof anyway, as a subsequent buffer extension
failure ends up calling std::terminate. I am working on addressing
that unfortunate failure mode in a manner more consistent with the C++
ABI design.

Reviewed By: dblaikie

Differential Revision: https://reviews.llvm.org/D122604

2 years ago[mlir] Add a title to the Transform Dialect doc
Alex Zinenko [Tue, 26 Apr 2022 11:04:41 +0000 (13:04 +0200)]
[mlir] Add a title to the Transform Dialect doc

2 years ago[llvm-objcopy][NFC] rename variable.
Alexey Lapshin [Tue, 26 Apr 2022 09:45:28 +0000 (12:45 +0300)]
[llvm-objcopy][NFC] rename variable.

2 years ago[compiler-rt] Use C_STANDARD instead of custom logic for adding -std=c11
Martin Storsjö [Thu, 24 Mar 2022 11:17:28 +0000 (13:17 +0200)]
[compiler-rt] Use C_STANDARD instead of custom logic for adding -std=c11

Apply this in add_compiler_rt_runtime instead of manually adding it
to the individual projects. This applies the option on more
parts of compiler-rt than before, but should ideally not make any
difference assuming the other runtimes that lacked the option
also were C11 compatible.

Not marking this as required, to match the existing behaviour (where
`-std=c11` was added only if supported by the compiler).

This was suggested during the review of D110005.

Differential Revision: https://reviews.llvm.org/D124343

2 years ago[gn build] Port 854c33946fd4
LLVM GN Syncbot [Tue, 26 Apr 2022 09:06:40 +0000 (09:06 +0000)]
[gn build] Port 854c33946fd4

2 years ago[RS4GC] Prune inputs of BDV if they are BDV themselves
Dmitry Makogon [Fri, 15 Apr 2022 07:59:40 +0000 (14:59 +0700)]
[RS4GC] Prune inputs of BDV if they are BDV themselves

Don't check whether an input of BDV can be pruned if the input
is the BDV itself. BDV is present in the states map, so in case
the input is the BDV itself, we'd return false. So explicitly check this case.

Differential Revision: https://reviews.llvm.org/D123846

2 years ago[llvm-gsymutil][NFC] refactor AddressRange&AddresRanges structures.
Alexey Lapshin [Fri, 22 Apr 2022 09:58:13 +0000 (12:58 +0300)]
[llvm-gsymutil][NFC] refactor AddressRange&AddresRanges structures.

llvm-gsymutil has an implementation of AddressRange and AddressRanges
classes. That implementation might be reused in other parts of llvm.
This patch moves AddressRange and AddressRanges classes into llvm/ADT.

Differential Revision: https://reviews.llvm.org/D124350

2 years ago[pseudo] NFC, fix some code-style naming violations.
Haojian Wu [Tue, 26 Apr 2022 08:35:53 +0000 (10:35 +0200)]
[pseudo] NFC, fix some code-style naming violations.

2 years ago[Clang] Use std::move in GlobalModuleIndex::readIndex. NFC
Jun Zhang [Sat, 9 Apr 2022 02:42:25 +0000 (10:42 +0800)]
[Clang] Use std::move in GlobalModuleIndex::readIndex. NFC

BitstreamCursors are heavy-weight objects that should not be passed by value.

Differential Revision: https://reviews.llvm.org/D123436

2 years ago[X86] Add use condition for combineSetCCMOVMSK
Xiang1 Zhang [Fri, 22 Apr 2022 06:59:53 +0000 (14:59 +0800)]
[X86] Add use condition for combineSetCCMOVMSK

Reviewed by RKSimon, LuoYuanke

Differential Revision: https://reviews.llvm.org/D123652

2 years ago[clang] Fix a constant evaluator crash on a NULL-type expr.
Haojian Wu [Mon, 25 Apr 2022 13:42:35 +0000 (15:42 +0200)]
[clang] Fix a constant evaluator crash on a NULL-type expr.

Differential Revision: https://reviews.llvm.org/D124384

2 years ago[AMDGPU] Vector register spill test cleanup (NFC)
Christudasan Devadasan [Tue, 26 Apr 2022 06:45:08 +0000 (12:15 +0530)]
[AMDGPU] Vector register spill test cleanup (NFC)

The vector register spills have no dependency with
SILowerSGPRSpills pass anymore. The entire handling
has been moved to PrologEpilogInserter with D55301.

2 years ago[libc] Add stdio entrypoints to aarch64 fullbuild.
Siva Chandra [Tue, 26 Apr 2022 07:25:12 +0000 (00:25 -0700)]
[libc] Add stdio entrypoints to aarch64 fullbuild.

2 years ago[libc][Obvious] Add deps of fopencookie_test only if it is enabled.
Siva Chandra [Tue, 26 Apr 2022 07:09:35 +0000 (00:09 -0700)]
[libc][Obvious] Add deps of fopencookie_test only if it is enabled.

2 years ago[X86][AMX] Report error when shapes are not pre-defined.
Luo, Yuanke [Sun, 24 Apr 2022 08:31:43 +0000 (16:31 +0800)]
[X86][AMX] Report error when shapes are not pre-defined.

Instead of report fatal error, this patch emit error message and exit
when shapes are not pre-defined. This would cause the compiling fail but
not crash.

Differential Revision: https://reviews.llvm.org/D124342

2 years ago[libc++] Remove <functional> includes
Nikolas Klauser [Sat, 23 Apr 2022 08:07:24 +0000 (10:07 +0200)]
[libc++] Remove <functional> includes

Reviewed By: var-const, #libc, ldionne

Spies: #libc_vendors, ldionne, libcxx-commits, miyuki

Differential Revision: https://reviews.llvm.org/D124123

2 years ago[analyzer] Fix ValistChecker false-positive involving symbolic pointers
Balazs Benics [Tue, 26 Apr 2022 06:49:05 +0000 (08:49 +0200)]
[analyzer] Fix ValistChecker false-positive involving symbolic pointers

In the following example:

  int va_list_get_int(va_list *va) {
    return va_arg(*va, int); // FP
  }

The `*va` expression will be something like `Element{SymRegion{va}, 0, va_list}`.
We use `ElementRegions` for representing the result of the dereference.
In this case, the `IsSymbolic` was set to `false` in the
`getVAListAsRegion()`.

Hence, before checking if the memregion is a SymRegion, we should take
the base of that region.

Analogously to the previous example, one can craft other cases:

  struct MyVaList {
    va_list l;
  };
  int va_list_get_int(struct MyVaList va) {
    return va_arg(va.l, int); // FP
  }

But it would also work if the `va_list` would be in the base or derived
part of a class. `ObjCIvarRegions` are likely also susceptible.
I'm not explicitly demonstrating these cases.

PS: Check the `MemRegion::getBaseRegion()` definition.

Fixes #55009

Reviewed By: xazax.hun

Differential Revision: https://reviews.llvm.org/D124239

2 years agoIntrinsic for checking floating point class
Serge Pavlov [Sun, 10 Oct 2021 15:46:58 +0000 (22:46 +0700)]
Intrinsic for checking floating point class

This change introduces a new intrinsic, `llvm.is.fpclass`, which checks
if the provided floating-point number belongs to any of the the specified
value classes. The intrinsic implements the checks made by C standard
library functions `isnan`, `isinf`, `isfinite`, `isnormal`, `issubnormal`,
`issignaling` and corresponding IEEE-754 operations.

The primary motivation for this intrinsic is the support of strict FP
mode. In this mode using compare instructions or other FP operations is
not possible, because if the value is a signaling NaN, floating-point
exception `Invalid` is raised, but the aforementioned functions must
never raise exceptions.

Currently there are two solutions for this problem, both are
implemented partially. One of them is using integer operations to
implement the check. It was implemented in https://reviews.llvm.org/D95948
for `isnan`. It solves the problem of exceptions, but offers one
solution for all targets, although some can do the check in more
efficient way.

The other, implemented in https://reviews.llvm.org/D96568, introduced a
hook 'clang::TargetCodeGenInfo::testFPKind', which injects a target
specific code into IR to implement `isnan` and some other functions. It is
convenient for targets that have dedicated instruction to determine FP data
class. However using target-specific intrinsic complicates analysis and can
prevent some optimizations.

A special intrinsic for value class checks allows representing data class
tests with enough flexibility. During IR transformations it represents the
check in target-independent way and saves it from undesired transformations.
In the instruction selector it allows efficient lowering depending on the
used target and mode.

This implementation is an extended variant of `llvm.isnan` introduced
in https://reviews.llvm.org/D104854. It is limited to minimal intrinsic
support. Target-specific treatment will be implemented in separate
patches.

Differential Revision: https://reviews.llvm.org/D112025

2 years agoProvide the complete response and reporting Code of Conduct documentation. Remove...
Tanya Lattner [Tue, 26 Apr 2022 05:32:56 +0000 (22:32 -0700)]
Provide the complete response and reporting Code of Conduct documentation. Remove the word draft from all documents, add information about the CoC committee expectations and add a place for transparency reports.

This patch provides the response and reporting guides for Code of Conduct reports. It also removes the word draft from all the documents,
adds information about the CoC committee, and a place to put transparency reports. A post will also be made on Discourse to provide more details about the history of the code of conduct, this patch, and next steps. Please see that post for full details. I will put a link below once I have it.

Reviewed By: aaron.ballman, hubert.reinterpretcast

Differential Revision: https://reviews.llvm.org/D122937

2 years ago[NFC][trace] simplify the instruction dumper
Walter Erquinigo [Wed, 20 Apr 2022 04:57:31 +0000 (21:57 -0700)]
[NFC][trace] simplify the instruction dumper

TraceInstructionDumper::DumpInstructions was becoming too big, so I'm
refactoring it into smaller functions. I also made some static methods proper
instance methods to simplify calls. Other minor improvements are also done.

Differential Revision: https://reviews.llvm.org/D124064

2 years ago[RISCV][SelectionDAG] Support VP_ADD/VP_MUL/VP_SUB mask operations
Lian Wang [Thu, 21 Apr 2022 03:17:31 +0000 (03:17 +0000)]
[RISCV][SelectionDAG] Support VP_ADD/VP_MUL/VP_SUB mask operations

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D124144

2 years ago[trace][intel pt] Support events
Walter Erquinigo [Mon, 18 Apr 2022 16:28:10 +0000 (09:28 -0700)]
[trace][intel pt] Support events

A trace might contain events traced during the target's execution. For
example, a thread might be paused for some period of time due to context
switches or breakpoints, which actually force a context switch. Not only
that, a trace might be paused because the CPU decides to trace only a
specific part of the target, like the address filtering provided by
intel pt, which will cause pause events. Besides this case, other kinds
of events might exist.

This patch adds the method `TraceCursor::GetEvents()`` that returns the
list of events that happened right before the instruction being pointed
at by the cursor. Some refactors were done to make this change simpler.

Besides this new API, the instruction dumper now supports the -e flag
which shows pause events, like in the following example, where pauses
happened due to breakpoints.

```
thread #1: tid = 2717361
  a.out`main + 20 at main.cpp:27:20
    0: 0x00000000004023d9    leaq   -0x1200(%rbp), %rax
  [paused]
    1: 0x00000000004023e0    movq   %rax, %rdi
  [paused]
    2: 0x00000000004023e3    callq  0x403a62                  ; std::vector<int, std::allocator<int> >::vector at stl_vector.h:391:7
  a.out`std::vector<int, std::allocator<int> >::vector() at stl_vector.h:391:7
    3: 0x0000000000403a62    pushq  %rbp
    4: 0x0000000000403a63    movq   %rsp, %rbp
```

The `dump info` command has also been updated and now it shows the
number of instructions that have associated events.

Differential Revision: https://reviews.llvm.org/D123982

2 years ago[gn build] Port b1fa5ac3ba34
LLVM GN Syncbot [Tue, 26 Apr 2022 01:46:12 +0000 (01:46 +0000)]
[gn build] Port b1fa5ac3ba34

2 years ago[mlgo] Factor out TensorSpec
Mircea Trofin [Mon, 25 Apr 2022 20:49:27 +0000 (13:49 -0700)]
[mlgo] Factor out TensorSpec

This is a simple datatype with a few JSON utilities, and is independent
of the underlying executor. The main motivation is to allow taking a
dependency on it on the AOT side, and allow us build a correctly-sized
buffer in the cases when the requested feature isn't supported by the
model. This, in turn, allows us to grow the feature set supported by the
compiler in a backward-compatible way; and also collect traces exposing
the new features, but starting off the older model, and continue
training from those new traces.

Differential Revision: https://reviews.llvm.org/D124417

2 years agoRefactor protected virtual functions from SymbolFile into new SymbolFileCommon class.
Jeffrey Tan [Wed, 20 Apr 2022 14:30:53 +0000 (07:30 -0700)]
Refactor protected virtual functions from SymbolFile into new SymbolFileCommon class.

This is a preparatory patch for https://reviews.llvm.org/D121631.
It refactors protected virtual members of SymbolFile
into a new SymbolFileCommon class per suggestion in:
https://reviews.llvm.org/D121631

This will avoid the friendship declaration in that patch.

Differential Revision: https://reviews.llvm.org/D124110

2 years agoNFC. Add missing DXILPointerTyID case
Chris Bieneman [Tue, 26 Apr 2022 01:07:56 +0000 (20:07 -0500)]
NFC. Add missing DXILPointerTyID case

This resolves -Werror hexigon build failures.