Juneyoung Lee [Sat, 14 Mar 2020 16:34:33 +0000 (01:34 +0900)]
[CodeGenPrepare] Freeze condition when transforming select to br
Summary:
This is a simple fix for CodeGenPrepare that freezes branch condition when transforming select to branch.
If it is not freezed, instsimplify or the later pipeline can potentially exploit undefined behavior.
The diff shows optimized form becase D75859 and D76048 already made a few changes to CodeGenPrepare for optimizing freeze(cmp).
Reviewers: jdoerfert, spatel, lebedev.ri, efriedma
Reviewed By: lebedev.ri
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D76179
Lang Hames [Sun, 15 Mar 2020 01:55:49 +0000 (18:55 -0700)]
[ORC] Remove an undefined static method from LLJIT.
Fixes http://llvm.org/PR44255. Thanks to Raoul Gough for finding the bug!
Lang Hames [Sun, 15 Mar 2020 01:54:40 +0000 (18:54 -0700)]
[JITLink][AArch64] Fix incorrect capitalization in a testcase name.
Lang Hames [Sun, 15 Mar 2020 01:52:10 +0000 (18:52 -0700)]
[ORC] Print symbol flags and materializer name in ExecutionSession::dump.
The extra information can be helpful in diagnosing JIT bugs.
Lang Hames [Sat, 14 Mar 2020 23:53:05 +0000 (16:53 -0700)]
[JITLink][MachO] Re-apply
b64afadf306, MachO linker-private support, with fixes.
Global symbols with linker-private prefixes should be resolvable across object
boundaries, but internal symbols with linker-private prefixes should not.
Sriraman Tallam [Fri, 13 Mar 2020 22:58:57 +0000 (15:58 -0700)]
Basic Block Sections Support.
This is the first in a series of patches to enable Basic Block Sections
in LLVM.
We introduce a new compiler option, -fbasicblock-sections=, which places every
basic block in a unique ELF text section in the object file along with a
symbol labeling the basic block. The linker can then order the basic block
sections in any arbitrary sequence which when done correctly can encapsulate
block layout, function layout and function splitting optimizations. However,
there are a couple of challenges to be addressed for this to be feasible:
1) The compiler must not allow any implicit fall-through between any two
adjacent basic blocks as they could be reordered at link time to be
non-adjacent. In other words, the compiler must make a fall-through
between adjacent basic blocks explicit by retaining the direct jump
instruction that jumps to the next basic block. These branches can only
be removed later by the linker after the blocks have been reordered.
2) All inter-basic block branch targets would now need to be resolved by
the linker as they cannot be calculated during compile time. This is
done using static relocations which bloats the size of the object files.
Further, the compiler tries to use short branch instructions on some ISAs
for branch offsets that can be accommodated in one byte. This is not
possible with basic block sections as the offset is not determined at
compile time, and long branch instructions have to be used everywhere.
3) Each additional section bloats object file sizes by tens of bytes. The
number of basic blocks can be potentially very large compared to the
size of functions and can bloat object sizes significantly. Option
fbasicblock-sections= also takes a file path which can be used to
specify a subset of basic blocks that needs unique sections to keep
the bloats small.
4) Debug Info and CFI need special handling and will be presented as
separate patches.
Basic Block Labels
With -fbasicblock-sections=labels, or when a basic block is placed in a
unique section, it is labelled with a symbol. This allows easy mapping of
virtual addresses from PMU profiles back to the corresponding basic blocks.
Since the number of basic blocks is large, the labeling bloats the symbol
table sizes and the string table sizes significantly. While the binary size
does increase, it does not affect performance as the symbol table is not
loaded in memory during run-time. The string table size bloat is kept very
minimal using a unary naming scheme that uses string suffix compression.
The basic blocks for function foo are named "a.BB.foo", "aa.BB.foo", ...
This turns out to be very good for string table sizes and the bloat in the
string table size for a very large binary is ~8 %. The naming also allows
using the --symbol-ordering-file option in LLD to arbitrarily reorder the
sections.
Differential Revision: https://reviews.llvm.org/D68063
Sriraman Tallam [Sun, 15 Mar 2020 01:08:26 +0000 (18:08 -0700)]
Test commit.
Lang Hames [Sat, 14 Mar 2020 23:52:25 +0000 (16:52 -0700)]
Revert "[JITLink][MachO] Treat linker private symbols as hidden rather than private."
This reverts commit
b64afadf306f284a684ee656c6eefbd43c192c8d.
Reverting while I investigate bot failures.
Craig Topper [Sat, 14 Mar 2020 23:28:28 +0000 (16:28 -0700)]
[X86] Add avx512f only command lines to the vector add/sub saturation tests. NFC
Gives us coverage of splitting the v32i16/v64i8 when we have
avx512f and not avx512bw.
Considering making v32i16/v64i8 a legal type on avx512f which
needs this test coverage.
Lang Hames [Sat, 14 Mar 2020 23:22:48 +0000 (16:22 -0700)]
[JITLink][MachO] Treat linker private symbols as hidden rather than private.
Linker-private symbols should be resolvable across object file boundaries.
Lang Hames [Sat, 14 Mar 2020 23:07:46 +0000 (16:07 -0700)]
[llvm-jitlink] Add -show-init-es option to dump initial ExecutionSession state.
Inspecting this state can be helpful when debugging jit-linking testcases.
Lang Hames [Sat, 14 Mar 2020 22:10:59 +0000 (15:10 -0700)]
[Orc][examples] Actually return MainResult from main
LLVM GN Syncbot [Sat, 14 Mar 2020 21:50:50 +0000 (21:50 +0000)]
[gn build] Port
633ea07200e
Lang Hames [Sat, 14 Mar 2020 21:16:42 +0000 (14:16 -0700)]
[Orc] Add basic OrcV2 C bindings and example.
Renames the llvm/examples/LLJITExamples directory to llvm/examples/OrcV2Examples
since it is becoming a home for all OrcV2 examples, not just LLJIT.
See http://llvm.org/PR31103.
Aaron Ballman [Sat, 14 Mar 2020 21:15:59 +0000 (17:15 -0400)]
Remove an unnecessary explicit 'WarnDiag'; NFC
Aaron Ballman [Sat, 14 Mar 2020 21:00:45 +0000 (17:00 -0400)]
Remove some explicit calls to getName() when printing diagnostics; NFC
River Riddle [Sat, 14 Mar 2020 20:50:01 +0000 (13:50 -0700)]
[mlir] Only treat "Alloc" effects as dead if they are for operation results.
Allocate could be used for an "output" of an operation in the case of buffer-style operations.
River Riddle [Sat, 14 Mar 2020 20:36:42 +0000 (13:36 -0700)]
[mlir][NFC] Move the LoopLike interface out of Transforms/ and into Interfaces/
Differential Revision: https://reviews.llvm.org/D76155
Aaron Ballman [Sat, 14 Mar 2020 19:57:28 +0000 (15:57 -0400)]
Improve the attribute language option interface somewhat; NFCi.
The name field is optional if the custom code is supplied, so this updates the
documentation for LangOpt and introduces a tablegen warning if both custom code
and a language option name are supplied.
Florian Hahn [Sat, 14 Mar 2020 17:54:38 +0000 (17:54 +0000)]
[ValueLattice] Go to overdefined in getRange() for full ranges.
This is was split off
4878aa36d4aa27df644430139fab2734fde4a000,
as it can go in separately.
Krzysztof Parzyszek [Fri, 13 Mar 2020 13:37:57 +0000 (08:37 -0500)]
[Hexagon] Only allow single HVX vector loads/stores in lowering
This will prevent store widening from forming vector pair stores,
which eventually end up broken up into single stores.
Isuru Fernando [Sat, 14 Mar 2020 18:41:12 +0000 (18:41 +0000)]
[mlir] Fix cross compiling MLIR
Setting MLIR_TABLEGEN_EXE would prevent building the native tool which is used in cross-compiling
Differential Revision: https://reviews.llvm.org/D75299
Tamás Zolnai [Sat, 14 Mar 2020 16:57:02 +0000 (17:57 +0100)]
[clang-tidy] extend bugprone-signed-char-misuse check.
Summary:
Cover a new use case when using a 'signed char' as an integer
might lead to issue with non-ASCII characters. Comparing
a 'signed char' with an 'unsigned char' using equality / unequality
operator produces an unexpected result for non-ASCII characters.
Reviewers: aaron.ballman, alexfh, hokein, njames93
Reviewed By: njames93
Subscribers: xazax.hun, cfe-commits
Tags: #clang, #clang-tools-extra
Differential Revision: https://reviews.llvm.org/D75749
Simon Pilgrim [Sat, 14 Mar 2020 18:42:27 +0000 (18:42 +0000)]
Fix signed/unsigned comparison warning.
Simon Pilgrim [Sat, 14 Mar 2020 17:35:16 +0000 (17:35 +0000)]
[X86] getFauxShuffleMask - pull out repeated byte sizes varaibles. NFC.
Florian Hahn [Sat, 14 Mar 2020 16:50:09 +0000 (16:50 +0000)]
[ValueLattice] Add new state for undef constants.
This patch adds a new undef lattice state, which is used to represent
UndefValue constants or instructions producing undef.
The main difference to the unknown state is that merging undef values
with constants (or single element constant ranges) produces the
constant/constant range, assuming all uses of the merge result will be
replaced by the found constant.
Contrary, merging non-single element ranges with undef needs to go to
overdefined. Using unknown for UndefValues currently causes mis-compiles
in CVP/LVI (PR44949) and will become problematic once we use
ValueLatticeElement for SCCP.
Reviewers: efriedma, reames, davide, nikic
Reviewed By: efriedma
Differential Revision: https://reviews.llvm.org/D75120
Georgii Rymar [Fri, 13 Mar 2020 13:31:16 +0000 (16:31 +0300)]
[yaml2obj] - Set a default value for `PAddr` property of a program header to a value of `VAddr`
`PAddr` corresponds to `p_paddr` of a program header, which is the segment's physical
address for systems in which physical addressing is relevant. `p_paddr` is often equal
to `p_vaddr`, which is the virtual address of a segment.
This patch changes the default for `PAddr` from 0 to a value of `VAddr`.
Differential revision: https://reviews.llvm.org/D76131
Simon Pilgrim [Sat, 14 Mar 2020 13:11:03 +0000 (13:11 +0000)]
[X86] getFauxShuffleMask - merge insertelement paths
Merge the INSERT_VECTOR_ELT/SCALAR_TO_VECTOR and PINSRW/PINSRB shuffle mask paths - they both do the same thing (find source vector + handle implicit zero extension). The PINSRW/PINSRB path also handled in the insertion of zero case which needed to be added to the general case as well.
Martin Storsjö [Wed, 11 Mar 2020 08:54:37 +0000 (10:54 +0200)]
[llvm-dlltool] Add a testcase to show the kind of weak external used for import library aliases. NFC.
Shengchen Kan [Thu, 12 Mar 2020 09:13:09 +0000 (17:13 +0800)]
[X86] Disable nop padding before instruction following a prefix
Reviewers: reames, MaskRay, craig.topper, LuoYuanke, jyknight
Reviewed By: LuoYuanke
Subscribers: hiraditya, llvm-commits, annita.zhang
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D76052
Uday Bondhugula [Sat, 14 Mar 2020 04:15:07 +0000 (21:15 -0700)]
[mlir][NFC] modernize / clean up some loop transform utils, affine analysis utils
Summary:
- remove stale declarations on flat affine constraints
- avoid allocating small vectors where possible
- clean up code comments, rename some variables
Differential Revision: https://reviews.llvm.org/D76117
Richard Smith [Sat, 14 Mar 2020 02:29:44 +0000 (19:29 -0700)]
PR44992 Don't crash when a defaulted <=> is in a class declared in a
transparent context.
(The same crash would happen if a class template with a friend was
declared in an 'export' block, but there are more issues with that
case.)
Diogo Sampaio [Fri, 13 Mar 2020 17:32:31 +0000 (17:32 +0000)]
[AArch64][Fix] LdSt optimization generate premature stack-popping
Summary:
When moving add and sub to memory operand instructions,
aarch64-ldst-opt would prematurally pop the stack pointer,
before memory instructions that do access the stack using
indirect loads.
e.g.
```
int foo(int offset){
int local[4] = {0};
return local[offset];
}
```
would generate:
```
sub sp, sp, #16 ; Push the stack
mov x8, sp ; Save stack in register
stp xzr, xzr, [sp], #16 ; Zero initialize stack, and post-increment, making it invalid
------ If an exception goes here, the stack value might be corrupted
ldr w0, [x8, w0, sxtw #2] ; Access correct position, but it is not guarded by SP
```
Reviewers: fhahn, foad, thegameg, eli.friedman, efriedma
Reviewed By: efriedma
Subscribers: efriedma, kristof.beyls, hiraditya, danielkiss, llvm-commits, simon_tatham
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D75755
Richard Smith [Sat, 14 Mar 2020 01:53:08 +0000 (18:53 -0700)]
PR45063: Fix crash on invalid processing an elaborated class template-id
with an invalid scope specifier.
Craig Topper [Sat, 14 Mar 2020 01:02:48 +0000 (18:02 -0700)]
[X86] Remove isel patterns for X86VBroadcast+trunc+extload. Replace with DAG combines.
This is a little more complicated than I'd like it to be. We have
to manually match a trunc+srl+load pattern that generic DAG
combine won't do for us due to isTypeDesirableForOp.
Michael Liao [Sat, 14 Mar 2020 00:50:35 +0000 (20:50 -0400)]
Fix `-Wunused-variable`. NFC.
Dan Albert [Fri, 13 Mar 2020 20:55:26 +0000 (13:55 -0700)]
Revert "Revert "Move more tests to globalMemCounter and reset.""
Test regressions not included this time :)
This reverts commit
1ed671082ef4b13d44e2c0f42ddedf9d450258a7.
Whitney Tsang [Sat, 14 Mar 2020 00:04:10 +0000 (00:04 +0000)]
[NFC][LoopUnrollAndJam] clang-format.
I am currently working on this file.
Philip Reames [Fri, 13 Mar 2020 22:54:43 +0000 (15:54 -0700)]
Adjust debug output for MCRelaxableFragment to include the size so that sanity checking relaxation offsets from -debug output is easier
Eli Friedman [Fri, 13 Mar 2020 23:12:45 +0000 (16:12 -0700)]
[SCEV] Add support for GEPs over scalable vectors.
Because we have to use a ConstantExpr at some point, the canonical form
isn't set in stone, but this seems reasonable.
The pretty sizeof(<vscale x 4 x i32>) dumping is a relic of ancient
LLVM; I didn't have to touch that code. :)
Differential Revision: https://reviews.llvm.org/D75887
Brian Cain [Fri, 13 Mar 2020 20:24:19 +0000 (15:24 -0500)]
Initialize IsFast* values
We must initialize these values in case some targets do not assign to
them in allowsMemoryAccess().
Jan Korous [Fri, 13 Mar 2020 22:21:12 +0000 (15:21 -0700)]
[LLJIT] Add std::move() as a workaround for older compilers
Clang 3.8 isn't able to bind the variable to rvalue-ref which breaks the build.
Alexey Bataev [Mon, 9 Mar 2020 12:55:57 +0000 (08:55 -0400)]
[OPENMP]Fix PR45047: Do not copy firstprivates in tasks twice.
Avoid copying of the orignal variable if it is going to be marked as
firstprivate in task regions. For taskloops, still need to copy the
non-trvially copyable variables to correctly construct them upon task
creation.
Sean Silva [Wed, 11 Mar 2020 22:15:06 +0000 (15:15 -0700)]
Add Builder::getI64VectorAttr.
This matches Builder::getI32VectorAttr.
Differential Revision: https://reviews.llvm.org/D75883
Ahmed Taei [Fri, 13 Mar 2020 20:41:14 +0000 (13:41 -0700)]
[mlir][LLVMIR] Add a support for boolean type arguments conversion
Summary:
Otherwise this will fail translating ops with boolean arguments and
added test will fail.
Reviewers: nicolasvasilache, rriddle
Subscribers: mehdi_amini, jpienaar, burmako, shauheen, antiagainst, arpith-jacob, mgester, lucyrfox, aartbik, liufengdb, Joonsoo, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D76159
Hanhan Wang [Fri, 13 Mar 2020 21:33:27 +0000 (14:33 -0700)]
[mlir][Linalg] Implement padding for linalg.conv and lowering to loops.
Summary:
To enable this, two changes are needed:
1) Add an optional attribute `padding` to linalg.conv.
2) Compute if the indices accessing is out of bound in the loops. If so, use the
padding value `0`. Otherwise, use the value derived from load.
In the patch, the padding only works for lowering without other transformations,
e.g., tiling, fusion, etc.
Differential Revision: https://reviews.llvm.org/D75722
Jacques Pienaar [Fri, 13 Mar 2020 17:09:02 +0000 (10:09 -0700)]
[mlir] Automatically add DerivedAttribute op interface
Summary: When an operation has derived attributes, add the DerivedAttribute op interface.
Differential Revision: https://reviews.llvm.org/D76144
Shoaib Meenai [Fri, 13 Mar 2020 21:15:58 +0000 (14:15 -0700)]
[ELF] Add test for freebsd bfdname
Make sure the output error contains the full bfdname. Suggested by
George Rimar in https://reviews.llvm.org/D76109#inline-693972.
Jan Korous [Thu, 12 Mar 2020 22:27:08 +0000 (15:27 -0700)]
[clangd] Add json::Object->Value conversion workaround for older compilers
The build was broken for clang-3.8 which we still support.
Craig Topper [Fri, 13 Mar 2020 19:57:20 +0000 (12:57 -0700)]
[SelectionDAGBuilder] Simplify the struct type handling in getUniformBase.
Craig Topper [Fri, 13 Mar 2020 19:19:36 +0000 (12:19 -0700)]
[IR] Fix formatting. NFC
Lang Hames [Fri, 13 Mar 2020 16:24:09 +0000 (09:24 -0700)]
[MCJIT] Check for RuntimeDyld errors in MCJIT::finalizeLoadedModules.
Patch based on https://reviews.llvm.org/D75912 by Alexander Shishkin. Thanks
Alexander!
To minimize disruption to existing clients, who may be relying on the fact that
unused references to unresolved symbols do not generate an error, this patch
makes error checking opt-in: Clients can call ExecutionEngine::hasError or
LLVMExecutionEngineGetError to check whether and error has occurred.
Differential revision: https://reviews.llvm.org/D75912
Richard Smith [Fri, 13 Mar 2020 20:56:57 +0000 (13:56 -0700)]
Fix "unused variable" warning in NDEBUG builds.
Amy Huang [Thu, 12 Mar 2020 22:51:40 +0000 (15:51 -0700)]
CMake: Turn LLVM_ENABLE_ZLIB into a tri-state option
Summary:
Add FORCE_ON option to LLVM_ENABLE_ZLIB, which causes a configuration
error if zlib is not found.
Similar to https://reviews.llvm.org/D40050.
Reviewers: hans, thakis, rnk
Subscribers: mgorny, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D76110
Akira Hatanaka [Fri, 13 Mar 2020 19:15:17 +0000 (12:15 -0700)]
[ObjC][ARC] Don't remove autoreleaseRV/retainRV pairs if the call isn't
a tail call
This reapplies the patch in https://reviews.llvm.org/rG1f5b471b8bf4,
which was reverted because it was causing crashes.
https://bugs.chromium.org/p/chromium/issues/detail?id=1061289#c2
Check that HasSafePathToCall is true before checking the call is a tail
call.
Original commit message:
Previosly ARC optimizer removed the autoreleaseRV/retainRV pair in the
following code, which caused the object returned by @something to be
placed in the autorelease pool because the call to @something isn't a
tail call:
```
%call = call i8* @something(...)
%2 = call i8* @objc_retainAutoreleasedReturnValue(i8* %call)
%3 = call i8* @objc_autoreleaseReturnValue(i8* %2)
ret i8* %3
```
Fix the bug by checking whether @something is a tail call.
rdar://problem/
59275894
Stanislav Mekhanoshin [Fri, 13 Mar 2020 19:44:59 +0000 (12:44 -0700)]
[AMDGPU] Fix endcf collapse
Only collapse inner endcf if the outer one belongs to SI_IF.
If it does belong to SI_ELSE then mask being restored in fact
a partial inverse of what we need.
Differential Revision: https://reviews.llvm.org/D76154
Martin Storsjö [Tue, 10 Mar 2020 21:44:03 +0000 (23:44 +0200)]
[COFF] Assign unique names to autogenerated .weak.<name>.default symbols
These symbols need to be external (MSVC tools error out if a weak
external points at a symbol that isn't external; this was tried before
but had to be reverted in
bc5b7217dceecd3eec69593026a9e38dfbfd6908,
and this was originally explicitly fixed in
732eeaf2a930ad2755cb4eb5d99a3deae0de4a72).
If multiple object files have weak symbols with defaults, their
defaults could cause linker errors due to duplicate definitions,
unless the names of the defaults are unique.
GNU binutils handles this by appending the name of another symbol
from the same object file to the name of the default symbol. Try
to implement something similar; before writing the object file,
locate a symbol that should have a unique name and use the name of
that one for making the weak defaults unique.
Differential Revision: https://reviews.llvm.org/D75989
Nicolas Vasilache [Fri, 13 Mar 2020 20:32:27 +0000 (16:32 -0400)]
[mlir][Vector]Lower vector.contract to llvm.intr.matrix_multiply
Summary:
This revision adds lowering of vector.contract to llvm.intr.matrix_multiply.
Note that there is currently a mismatch between the MLIR vector dialect which
expects row-major layout and the LLVM matrix intrinsics which expect column
major layout.
As a consequence, we currently only match a vector.contract with indexing maps
that express column-major matrix multiplication.
Other cases would require additional transposes and it is better to wait for
LLVM intrinsics to provide a per-operation attribute that would specify which
layout is expected.
A separate integration test, not submitted to MLIR core, has independently
verified that correct execution occurs on a 2x2x2 matrix multiplication.
Differential Revision: https://reviews.llvm.org/D76014
Matt Arsenault [Wed, 11 Mar 2020 20:13:52 +0000 (16:13 -0400)]
AMDGPU: Add flag to used fixed function ABI
Pass all arguments to every function, rather than only passing the
minimum set of inputs needed for the call graph.
Dan Albert [Fri, 13 Mar 2020 20:26:35 +0000 (13:26 -0700)]
Revert "Move more tests to globalMemCounter and reset."
Not all of these changes were correct. Will reland appropriate parts
in a follow up.
This reverts commit
4a792965de08fb69652370625f1aec2e87111313.
Akira Hatanaka [Fri, 13 Mar 2020 20:08:48 +0000 (13:08 -0700)]
[Sema] Use the canonical type in function isVector
This reapplies the following patch, which was reverted because it caused
neon CodeGen tests to fail:
https://reviews.llvm.org/rGa6150b48cea00ab31e9335cc73770327acc4cb3a
I've added checks to detect half precision neon vectors and avoid
promiting them to vectors of floats.
See the discussion here: https://reviews.llvm.org/rG825235c140e7
Original commit message:
This fixes an assertion in Sema::CreateBuiltinBinOp that fails when one
of the vector operand's element type is a typedef of __fp16.
rdar://problem/
55983556
Alexey Zhikhartsev [Fri, 13 Mar 2020 19:15:11 +0000 (15:15 -0400)]
[LoopInterchange] Fix interchanging contents of preheader BBs
Summary:
Previously LCSSA was getting broken by placing instructions into the
(newly) inner *header* instead of the *pre*header.
Fixes PR43474
Reviewers: fhahn
Reviewed By: fhahn
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D75943
Matt Arsenault [Thu, 12 Mar 2020 18:35:22 +0000 (14:35 -0400)]
AMDGPU: Don't handle kernarg.segment.ptr in functions
Just lower this to null. Pass implicitarg.ptr in its place in the
argument list.
Nico Weber [Fri, 13 Mar 2020 19:37:44 +0000 (15:37 -0400)]
Revert "Reland "[DebugInfo] Enable the debug entry values feature by default""
This reverts commit
5aa5c943f7da155b95564058cd5d50a93eabfc89.
Causes clang to assert, see
https://bugs.chromium.org/p/chromium/issues/detail?id=1061533#c4
for a repro.
Stanislav Mekhanoshin [Fri, 13 Mar 2020 17:58:19 +0000 (10:58 -0700)]
[AMDGPU] Disable endcf collapse
There are some functional regressions and I suspect our
scopes are not as perfectly enclosed as I expected.
Disable it for now.
Differential Revision: https://reviews.llvm.org/D76148
Richard Sandiford [Thu, 27 Feb 2020 10:25:31 +0000 (10:25 +0000)]
[Sema][SVE] Reject arrays of sizeless types
The SVE ACLE doesn't allow arrays of sizeless types. At the moment
clang accepts the TU:
__SVInt8_t x[2];
but trying to code-generate it triggers the LLVM assertion:
llvm/lib/IR/Type.cpp:588: static llvm::ArrayType* llvm::ArrayType::get(llvm::Type*, uint64_t): Assertion `isValidElementType(ElementType) && "Invalid type for array element!"' failed.
This patch reports an appropriate error instead.
The rules are slightly more restrictive than for general incomplete types.
For example:
struct s;
typedef struct s arr[2];
is valid as far as it goes, whereas arrays of sizeless types are
invalid in all contexts. BuildArrayType therefore needs a specific
check for isSizelessType in addition to the usual handling of
incomplete types.
Differential Revision: https://reviews.llvm.org/D76082
Richard Sandiford [Fri, 21 Feb 2020 17:12:44 +0000 (17:12 +0000)]
[Sema][SVE] Reject by-copy capture of sizeless types
Since fields can't have sizeless type, it also doesn't make sense
to capture sizeless types by value in lambda expressions. This patch
makes sure that we diagnose that and that we use "sizeless type" rather
"incomplete type" in the associated message. (Both are correct, but
"sizeless type" is more specific and hopefully more user-friendly.)
Differential Revision: https://reviews.llvm.org/D75738
Dan Albert [Fri, 13 Mar 2020 19:23:45 +0000 (12:23 -0700)]
Allow site-specific test_exec_root.
Reviewers: EricWF, mclow.lists, #libc, ldionne
Reviewed By: #libc, ldionne
Subscribers: dexonsmith, ldionne, libcxx-commits
Tags: #libc
Differential Revision: https://reviews.llvm.org/D76092
Richard Sandiford [Fri, 21 Feb 2020 16:14:55 +0000 (16:14 +0000)]
[Sema][SVE] Don't allow fields to have sizeless type
The SVE ACLE doesn't allow fields to have sizeless type. At the moment
clang accepts things like:
struct s { __SVInt8_t x; } y;
but trying to code-generate it leads to LLVM asserts like:
llvm/include/llvm/Support/TypeSize.h:126: uint64_t llvm::TypeSize::getFixedSize() const: Assertion `!IsScalable && "Request for a fixed size on a scalable object"' failed.
This patch adds an associated clang diagnostic.
Differential Revision: https://reviews.llvm.org/D75737
Dan Albert [Fri, 13 Mar 2020 19:01:59 +0000 (12:01 -0700)]
Revert "Revert "Update system_error tests for more platforms.""
This time using old fashioned starts_with.
This reverts commit
d4a8c3f2511f12e21e3c9adf58e162db25538c16.
Reid Kleckner [Tue, 18 Feb 2020 22:51:32 +0000 (14:51 -0800)]
Revert "[ObjC][ARC] Check the basic block size before calling DominatorTree::dominate"
This reverts commit
5c3117b0a98dd11717eaffd7fb583985d39544b2
This should not be necessary after
7593a480dbce4e26f7dda4aa8f15bffd03acbfdb, and Florian Hahn has confirmed
that the problem no longer reproduces with this patch.
I happened to notice this code because the FIXME talks about
OrderedBasicBlock.
Reviewed By: fhahn, dexonsmith
Differential Revision: https://reviews.llvm.org/D76075
Simon Pilgrim [Fri, 13 Mar 2020 18:42:43 +0000 (18:42 +0000)]
[X86][SSE] Prefer trunc(movd(x)) to pextrb(x,0)
If we're extracting the 0'th index of a v16i8 vector we're better off using MOVD than PEXTRB, unless we're storing the value or we require the implicit zero extension of PEXTRB.
The biggest perf diff is on SLM targets where MOVD (uops=1, lat=3 tp=1) is notably faster than PEXTRB (uops=2, lat=5, tp=4).
This matches what we already do for PEXTRW.
Differential Revision: https://reviews.llvm.org/D76138
aartbik [Fri, 13 Mar 2020 16:35:29 +0000 (09:35 -0700)]
[mlir] [VectorOps,LinAlg] Remove direct LLVM lowering for vector.broadcast
Summary:
The direct lowering of vector.broadcast into LLVM has been replaced by
progressive lowering into elementary vector ops. This also required a
small refactoring of a llvm.mlir test that used a direct vector.broadcast
operator (just to define a matmul).
Reviewers: nicolasvasilache, andydavis1, rriddle
Reviewed By: nicolasvasilache
Subscribers: mehdi_amini, rriddle, jpienaar, burmako, shauheen, antiagainst, nicolasvasilache, arpith-jacob, mgester, lucyrfox, liufengdb, Joonsoo, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D76143
Sanjay Patel [Fri, 13 Mar 2020 18:36:26 +0000 (14:36 -0400)]
[SimplifyCFG] add test for chain of empty block conditional branches; NFC
Huihui Zhang [Fri, 13 Mar 2020 18:23:01 +0000 (11:23 -0700)]
[SLPVectorizer][SVE] Bail out early for scalable vector.
Summary:
SLPVectorizer try to vectorize list of scalar instructions of the same type,
instructions already vectorized are rejected through isValidElementType().
Without this patch, tryToVectorizeList() will first try to determine vectorization
factor of a list of Instructions before checking whether each instruction has unsupported
type or not. For instructions already vectorized for SVE, it will crash at getVectorElementSize(),
where it try to return a fixed size.
This patch make sure invalid element types are rejected before trying to get vectorization
factor. This make sure we are not trying to vectorize instructions already vectorized.
Reviewers: sdesmalen, efriedma, spatel, RKSimon, ABataev, apazos, rengolin
Reviewed By: efriedma
Subscribers: tschuett, hiraditya, rkruppe, psnobl, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D76017
Adrian Prantl [Thu, 12 Mar 2020 21:19:04 +0000 (14:19 -0700)]
Debug Info: Store the SDK in the DICompileUnit.
This is another intermediate step for PR44213
(https://bugs.llvm.org/show_bug.cgi?id=44213).
This stores the SDK *name* in the debug info, to make it possible to
`-fdebug-prefix-map`-replace the sysroot with a recognizable string
and allowing the debugger to find a fitting SDK relative to itself,
not the machine the executable was compiled on.
rdar://problem/
51645582
Sanjay Patel [Fri, 13 Mar 2020 18:12:04 +0000 (14:12 -0400)]
[SimplifyCFG] regenerate complete test checks; NFC
Sanjay Patel [Fri, 13 Mar 2020 18:08:54 +0000 (14:08 -0400)]
[SimplifyCFG] regenerate test checks; NFC
Sanjay Patel [Fri, 13 Mar 2020 17:40:53 +0000 (13:40 -0400)]
[SimplifyCFG] fix formatting; NFC
Sanjay Patel [Fri, 13 Mar 2020 17:14:15 +0000 (13:14 -0400)]
[SimplifyCFG] fix debug print formatting; NFC
Florian Hahn [Mon, 24 Feb 2020 19:26:18 +0000 (19:26 +0000)]
[CVP,SCCP] Precommit test for D75055.
Test case for PR44949.
Philip Reames [Fri, 13 Mar 2020 17:49:38 +0000 (10:49 -0700)]
Use 15 byte long nops on modern Intel processors
Back in D42616, we switched our default nop length from 15 to 10 bytes because some platforms have painful decode stalls when encountering multiple instruction prefixes. (10 byte long nops come from the fact that prefixes are used to pad after 8 bytes, and some platforms have issues w/more than two prefixes.)
Based on Agner's guides, it appears to be the case that modern Intel (SandyBridge and later) can decode an arbitrary number of prefixes without issue. Intel's guide only provides up to 9 bytes; I read that as providing a safe default for all their chips. Older chips and Atom series have serious decode stalls. I can't find a conclusive reference beyond those two.
Differential Revision: https://reviews.llvm.org/D75945
Simon Cook [Fri, 13 Mar 2020 17:13:51 +0000 (17:13 +0000)]
[TableGen] Support combining AssemblerPredicates with ORs
For context, the proposed RISC-V bit manipulation extension has a subset
of instructions which require one of two SubtargetFeatures to be
enabled, 'zbb' or 'zbp', and there is no defined feature which both of
these can imply to use as a constraint either (see comments in D65649).
AssemblerPredicates allow multiple SubtargetFeatures to be declared in
the "AssemblerCondString" field, separated by commas, and this means
that the two features must both be enabled. There is no equivalent to
say that _either_ feature X or feature Y must be enabled, short of
creating a dummy SubtargetFeature for this purpose and having features X
and Y imply the new feature.
To solve the case where X or Y is needed without adding a new feature,
and to better match a typical TableGen style, this replaces the existing
"AssemblerCondString" with a dag "AssemblerCondDag" which represents the
same information. Two operators are defined for use with
AssemblerCondDag, "all_of", which matches the current behaviour, and
"any_of", which adds the new proposed ORing features functionality.
This was originally proposed in the RFC at
http://lists.llvm.org/pipermail/llvm-dev/2020-February/139138.html
Changes to all current backends are mechanical to support the replaced
functionality, and are NFCI.
At this stage, it is illegal to combine features with ands and ors in a
single AssemblerCondDag. I suspect this case is sufficiently rare that
adding more complex changes to support it are unnecessary.
Differential Revision: https://reviews.llvm.org/D74338
Jonas Devlieghere [Fri, 13 Mar 2020 17:06:45 +0000 (10:06 -0700)]
[lldb/Host] s/FindProcesses/FindProcessesImpl/ in windows/Host.cpp
Fix the Windows build.
Jonas Devlieghere [Fri, 13 Mar 2020 17:03:52 +0000 (10:03 -0700)]
[lldb/Test] Temporarily skip TestReproducerAttach on Linux
The test is failing with an unexpected packet during replay. Temporarily
disabling the test while I setup and environment to investigate.
Florian Hahn [Fri, 13 Mar 2020 16:40:03 +0000 (16:40 +0000)]
Recommit "[SCCP] Use ValueLatticeElement instead of LatticeVal (NFCI)"
This patch should fix the cause of the stage2 failures and
PR45185.
This reverts the revert commit
c52f839e723ee288db2a3e21860b011f6a9d707e.
Hyrum Wright [Tue, 3 Mar 2020 20:08:03 +0000 (15:08 -0500)]
[clang-tidy] Update Abseil Duration Conversion check to find more cases.
This change improves the check to handle cases with internal scalar
multiplication.
Differential Revision: https://reviews.llvm.org/D75558
Simon Pilgrim [Fri, 13 Mar 2020 16:50:57 +0000 (16:50 +0000)]
[CostModel][X86] Improve ISD::CTTZ costs accounting for BSF/TZCNT implementations
Simon Pilgrim [Fri, 13 Mar 2020 16:19:39 +0000 (16:19 +0000)]
[X86] Add cttz/ctlz tests for i686 with CMOV target
Jonas Devlieghere [Fri, 13 Mar 2020 16:49:00 +0000 (09:49 -0700)]
[lldb/Test] Convert stdout to str by calling decode('utf-8') on it.
Make sure both arguments to assertIn are of type str. This should fix
the following error:
TypeError: a bytes-like object is required, not 'str'.
Ehud Katz [Fri, 13 Mar 2020 16:31:47 +0000 (18:31 +0200)]
[SCEV] Fix usage of invalid IP with FoldingSet
Fix the use of invalid Insertion Point pointer with the UniqueSCEVs FoldingSet,
which caused memory corruption.
Tyker [Fri, 13 Mar 2020 13:35:26 +0000 (14:35 +0100)]
[AssumeBundles] filter usefull attriutes to preserve
Summary:
This patch will filter attributes to only preserve those that are usefull.
In the case of NoAlias it is filtered out not because it isn't usefull
but because it is incorrect to preserve it as it is only valdi for the
duration of the function.
Reviewers: jdoerfert
Reviewed By: jdoerfert
Subscribers: jdoerfert, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D75828
Tyker [Fri, 13 Mar 2020 13:14:55 +0000 (14:14 +0100)]
[AssumeBundles] Preserve Information in the inliner
Summary:
during inling Create and insert an llvm.assume with attributes to preserve them.
to prevent any changes for now generation of llvm.assume is under a flag disabled by default.
Reviewers: jdoerfert
Reviewed By: jdoerfert
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D75825
Jonas Devlieghere [Fri, 13 Mar 2020 15:49:15 +0000 (08:49 -0700)]
[lldb/Reproducers] Intercept the FindProcesses API
This patch extends the reproducers to intercept calls to FindProcesses.
During capture it serializes the ProcessInstanceInfoList returned by the
API. During replay, it returns the serialized data instead of querying
the host.
The motivation for this patch is supporting the process attach workflow
during replay. Without this change it would incorrectly look for the
inferior on the host during replay and failing if no matching process
was found.
Differential revision: https://reviews.llvm.org/D75877
Alexandre Ganea [Fri, 13 Mar 2020 16:22:09 +0000 (12:22 -0400)]
[CodeView] Align type records on 4-bytes when emitting PDBs
When emitting PDBs, the TypeStreamMerger class is used to merge .debug$T records from the input .OBJ files into the output .PDB stream.
Records in .OBJs are not required to be aligned on 4-bytes, and "The Netwide Assembler 2.14" generates non-aligned records.
When compiling with -DLLVM_ENABLE_ASSERTIONS=ON, an assert was triggered in MergingTypeTableBuilder when non-ghash merging was used.
With ghash merging there was no assert.
As a result, LLD could potentially generate a non-aligned TPI stream.
We now align records on 4-bytes when record indices are remapped, in TypeStreamMerger::remapIndices().
Differential Revision: https://reviews.llvm.org/D75081
omarahmed1111 [Fri, 13 Mar 2020 15:30:36 +0000 (10:30 -0500)]
[Attributor] Detect possibly unbounded cycles in functions
This patch add mayContainUnboundedCycle helper function which checks whether a function has any cycle which we don't know if it is bounded or not.
Loops with maximum trip count are considered bounded, any other cycle not.
It also contains some fixed tests and some added tests contain bounded and
unbounded loops and non-loop cycles.
Reviewed By: jdoerfert, uenoku, baziotis
Differential Revision: https://reviews.llvm.org/D74691
Sid Manning [Mon, 10 Feb 2020 23:27:53 +0000 (17:27 -0600)]
[LLD][ELF][Hexagon] Support GDPLT transforms
Hexagon ABI specifies that call x@gdplt is transformed to call __tls_get_addr.
Example:
call x@gdplt
is changed to
call __tls_get_addr
When x is an external tls variable.
Differential Revision: https://reviews.llvm.org/D74443
Pankaj Gode [Fri, 13 Mar 2020 15:39:08 +0000 (21:09 +0530)]
[Attributor] Improve noalias preservation using reachability
Resolution for below fixme:
(ii) Check whether the value is captured in the scope using AANoCapture.
FIXME: This is conservative though, it is better to look at CFG and
check only uses possibly executed before this callsite.
Propagates caller argument's noalias attribute to callee.
Reviewed by: jdoerfert, uenoku
Reviewers: jdoerfert, sstefan1, uenoku
Subscribers: uenoku, sstefan1, hiraditya, llvm-commits
Differential Revision: https://reviews.llvm.org/D71617
Simon Pilgrim [Fri, 13 Mar 2020 15:35:13 +0000 (15:35 +0000)]
[X86] combineExtractWithShuffle - pull out repeated getSizeInBits() call. NFC.
Simon Pilgrim [Fri, 13 Mar 2020 14:58:16 +0000 (14:58 +0000)]
[X86] LowerEXTRACT_VECTOR_ELT - pull out repeated getOperand() calls. NFC.
Also, cleanup LowerEXTRACT_VECTOR_ELT_SSE4 comments which had references to non-constant extraction indices.