David Blaikie [Tue, 5 Feb 2019 23:38:55 +0000 (23:38 +0000)]
Orc: Simplify RPC naming system by using function-local statics
The existing scheme of class template static members for Name and
NameMutex is a bit verbose, involves global ctors (even if they're cheap
for string and mutex, still not entirely free), and (importantly/my
immediate motivation here) trips over a bug in LLVM's modules
implementation that's a bit involved (hmm, sounds like Mr. Smith has a
fix for the modules thing - but I'm still inclined to commit this patch
as general goodness).
Lang Hames [Tue, 5 Feb 2019 23:17:11 +0000 (23:17 +0000)]
[ADT] Add a fallible_iterator wrapper.
A fallible iterator is one whose increment or decrement operations may fail.
This would usually be supported by replacing the ++ and -- operators with
methods that return error:
The downside of this style is that it no longer conforms to the C++ iterator
concept, and can not make use of standard algorithms and features such as
range-based for loops.
The fallible_iterator wrapper takes an iterator written in the style above
and adapts it to (mostly) conform with the C++ iterator concept. It does this
by providing standard ++ and -- operator implementations, returning any errors
generated via a side channel (an Error reference passed into the wrapper at
construction time), and immediately jumping the iterator to a known 'end'
value upon error. It also marks the Error as checked any time an iterator is
compared with a known end value and found to be inequal, allowing early exit
from loops without redundant error checking*.
Usage looks like:
MyFallibleIterator I = ..., E = ...;
Error Err = Error::success();
for (auto &Elem : make_fallible_range(I, E, Err)) {
// Loop body is only entered when safe.
// Early exits from loop body permitted without checking Err.
if (SomeCondition)
return;
}
if (Err)
// Handle error.
* Since failure causes a fallible iterator to jump to end, testing that a
fallible iterator is not an end value implicitly verifies that the error is a
success value, and so is equivalent to an error check.
Sanjay Patel [Tue, 5 Feb 2019 22:58:45 +0000 (22:58 +0000)]
[InstCombine] limit extracting shuffle transform based on uses
As discussed in D53037, this can lead to worse codegen, and we
don't generally expect the backend to be able to optimize
arbitrary shuffles. If there's only one use of the 1st shuffle,
that means it's getting removed, so that should always be
safe.
Heejin Ahn [Tue, 5 Feb 2019 22:47:29 +0000 (22:47 +0000)]
[WebAssembly] Disable a v128.const test line temporarily
r353131 caused failures in v128.const test for clang-ppc64be-linux-lnt
and clang-s390x-linux bots. This temporarily disables that line until
it is fixed.
Petar Jovanovic [Tue, 5 Feb 2019 22:23:46 +0000 (22:23 +0000)]
[elfabi] Fix the type of the variable formated for error output
Change the format type of Dyn.SONameOffset to PRIx64 since it is a uint64_t.
The problem was detected on mips builds, where it was printing junk values
and causing test failure.
[NFC][GlobalISel]: Add a convenience method to MachineInstrBuilder to simplify getOperand(i).getReg()
https://reviews.llvm.org/D57608
It's a common pattern in GISel to have a MachineInstrBuilder from which we get various regs
(commonly MIB->getOperand(0).getReg()). This adds a helper method and the above can be
replaced with MIB.getReg(0).
Reid Kleckner [Tue, 5 Feb 2019 21:14:09 +0000 (21:14 +0000)]
[MC] Don't error on numberless .file directives on MachO
Summary:
Before r349976, MC ignored such directives when producing an object file
and asserted when re-producing textual assembly output. I turned this
assertion into a hard error in both cases in r349976, but this makes it
unnecessarily difficult to write a single assembly file that supports
both MachO and other object formats that support .file. A user reported
this as PR40578, and we decided to go back to ignoring the directive.
Matt Davis [Tue, 5 Feb 2019 21:01:01 +0000 (21:01 +0000)]
[llvm-readobj] Display sections that do not belong to a segment in the section-mapping
Summary:
The following patch adds the "None" line to the section to segment mapping dump.
That line lists the sections that do not belong to any segment.
I realize that this change differs from GNU readelf which does not display the latter information.
I'd rather not add this "feature" under a command line option. I think that might introduce confusion, since users would have to
make an additional decision as to if they want to see all of the section-to-segment map or just a subset of it.
Another option is to only print the "None" line if the `--section-mapping` option is passed; however,
that might also introduce some confusion, because the section-to-segment map would be different between`--program-headers`
and the `--section-mapping` output. While the difference is just the "None" line, it seems that if we choose to display
the segment-to-section mapping, then we should always display the whole map including the sections
that do not belong to segments.
Alexey Bataev [Tue, 5 Feb 2019 19:33:47 +0000 (19:33 +0000)]
[DEBUG_INFO][NVPTX] Generate DW_AT_address_class to get the values in debugger.
Summary:
According to
https://docs.nvidia.com/cuda/archive/10.0/ptx-writers-guide-to-interoperability/index.html#cuda-specific-dwarf,
the compiler should emit the DW_AT_address_class attribute for all
variable and parameter. It means, that DW_AT_address_class attribute
should be used in the non-standard way to support compatibility with the
cuda-gdb debugger.
Clang is able to generate the information about the variable address
class. This information is emitted as the expression sequence
`DW_OP_constu <DWARF Address Space> DW_OP_swap DW_OP_xderef`. The patch
tries to find all such expressions and transform them into
`DW_AT_address_class <DWARF Address Space>` if target is NVPTX and the debugger is gdb.
If the expression is not found, then default values are used. For the
local variables <DWARF Address Space> is set to ADDR_local_space(6), for
the globals <DWARF Address Space> is set to ADDR_global_space(5). The
values are taken from the table in the same section 5.2. CUDA-Specific
DWARF Definitions.
Matt Arsenault [Tue, 5 Feb 2019 19:23:57 +0000 (19:23 +0000)]
AMDGPU: Fix assert on trunc from bitcast of build_vector
The v2i64 argument is lowered to a bitcast of v4i32 build_vector.
This would then attempt to use the i32-element as the source of the
vector truncate. This really would need to collect 2 elements from the
build_vector to produce the intended truncate.
Simon Pilgrim [Tue, 5 Feb 2019 19:15:48 +0000 (19:15 +0000)]
[X86][SSE] Disable ZERO_EXTEND shuffle combining
rL352997 enabled ZERO_EXTEND from non-shuffle-able value types. I've disabled it for now to fix a regression identified by @asbirlea until I can fix this properly.
Petar Jovanovic [Tue, 5 Feb 2019 18:09:28 +0000 (18:09 +0000)]
[PGO] Fix the type of the formated variable
Change the format type of Value to PRIu64 since it is a uint64_t.
The problem was detected on mips boards building 32-bit binaries,
where it was printing junk values and causing test failure.
We can't outline BTI instructions, because they need to be the very first
instruction executed after an indirect call or branch. If we outline them, then
an indirect call might go to the branch to the outlined function, which will
fault.
Summary:
While the backend code of FileCheck relies on definition of variable
from the command-line to have an equal sign '=' and a variable name
before that, the frontend does not actually enforce it. This leads to
FileCheck crashing when invoked with invalid syntax for the -D option.
This patch adds the missing validation in the frontend. It also makes
the -D option an AlwaysPrefix option to be able to detect -D=FOO as
being a define without variable and -D as missing its value.
Copyright:
- Linaro (changes in version 2 of revision D55940)
- GraphCore (changes in later versions)
Summary:
Add support for options that always prefix their value, giving an error
if the value is in the next argument or if the option is given a value
assignment (ie. opt=val). This is the desired behavior for the -D option
of FileCheck for instance.
Copyright:
- Linaro (changes in version 2 of revision D55940)
- GraphCore (changes in later versions and introduced when creating
D56549)
[MCA] Moved the logic that updates register dependencies from DispatchStage to RegisterFile. NFC
DispatchStage should always delegate to an object of class RegisterFile the task
of updating data dependencies. ReadState and WriteState objects should not be
modified directly by DispatchStage.
This patch also renames stage IS_AVAILABLE to IS_DISPATCHED.
[MCA] Simplify the logic in method WriteState::addUser. NFCI
In some cases, it is faster to just grow the set of 'Users' rather than
performing a llvm::find_if every time a new user is added to
the set. No functional change intended.
Jeremy Morse [Tue, 5 Feb 2019 11:11:28 +0000 (11:11 +0000)]
[DebugInfo][NFCI] Split salvageDebugInfo into helper functions
Some use cases are appearing where salvaging is needed that does not
correspond to an instruction being deleted -- for example an instruction
being sunk, or a Value not being available in a block being isel'd.
Enable more fine grained control over how salavging occurs by splitting
the logic into helper functions, separating things that are specific to
working on DbgVariableIntrinsics from those specific to interpreting IR
and building DIExpressions.
Hans Wennborg [Tue, 5 Feb 2019 11:01:54 +0000 (11:01 +0000)]
Fix format string in bindings/go/llvm/ir_test.go (PR40561)
The test started failing for me recently. I don't see any changes around
this code, so maybe it's my local go version that changed or something.
The error seems real to me: we're trying to print an Attribute with %d.
The test talks about "attribute masks" I'm not sure what that refers to,
but I suppose we could print the raw pointer value, since that's
what the test seems to be comparing.
Florian Hahn [Tue, 5 Feb 2019 10:27:40 +0000 (10:27 +0000)]
[CGP] Add support for sinking operands to their users, if they are free.
This patch improves code generation for some AArch64 ACLE intrinsics. It adds
support to CGP to duplicate and sink operands to their user, if they can be
folded into a target instruction, like zexts and sub into usubl. It adds a
TargetLowering hook shouldSinkOperands, which looks at the operands of
instructions to see if sinking is profitable.
I decided to add a new target hook, as for the sinking to be profitable,
at least on AArch64, we have to look at multiple operands of an
instruction, instead of looking at the users of a zext for example.
The sinking is done in CGP, because it works around an instruction
selection limitation. If instruction selection is not limited to a
single basic block, this patch should not be needed any longer.
Alternatively this could be done in the LoopSink pass, which tries to
undo LICM for instructions in blocks that are not executed frequently.
Note that we do not force the operands to sink to have a single user,
because we duplicate them before sinking. Therefore this is only
desirable if they really can be done for free. Additionally we could
consider the impact on live ranges later on.
This should fix https://bugs.llvm.org/show_bug.cgi?id=40025.
As for performance, we have internal code that uses intrinsics and can
be speed up by 10% by this change.
Dan Liew [Tue, 5 Feb 2019 08:47:28 +0000 (08:47 +0000)]
Previously if the user configured their build but then changed
LLVM_ENABLED_PROJECT and reconfigured it had no effect on what
projects were actually built. This was very confusing behaviour. The
reason for this is that the value of the `LLVM_TOOL_<PROJECT>_BUILD`
variables are already set.
The problem here is that we have two sources of truth:
* The projects listed in LLVM_ENABLE_PROJECTS.
* The projects enabled/disabled with LLVM_TOOL_<PROJECT>_BUILD.
At configure time we have no real way of knowing which source of truth
the user wants so we apply the following heuristic:
If the user ever sets `LLVM_ENABLE_PROJECTS` in the CMakeCache then that
is used as the single source of truth and we force the
`LLVM_TOOL_<PROJECT>_BUILD` CMake cache variables to have the
appropriate values that match the contents of the
`LLVM_ENABLE_PROJECTS`. If the user never sets `LLVM_ENABLE_PROJECTS`
then they can continue to use and set the `LLVM_TOOL_<PROJECT>_BUILD`
variables as the "source of truth".
The problem with this approach is that if the user ever tries to use
both `LLVM_ENABLE_PROJECTS` and `LLVM_TOOL_<PROJECT>_BUILD` for the same
build directory then any user set value for `LLVM_TOOL_<PROJECT>_BUILD`
variables will get overwriten, likely without the user noticing.
Hopefully the above shouldn't matter in practice because the
LLVM_TOOL_<PROJECT>_BUILD variables are not documented, but
LLVM_ENABLE_PROJECTS is.
We should probably deprecate the `LLVM_TOOL_<PROJECT>_BUILD`
variables at some point by turning them into to regular CMake
variables that don't live in the CMake cache.
Craig Topper [Tue, 5 Feb 2019 06:13:06 +0000 (06:13 +0000)]
[X86] Connect the default fpsr and dirflag clobbers in inline assembly to the registers we have defined for them.
Summary:
We don't currently map these constraints to physical register numbers so they don't make it to the MachineIR representation of inline assembly.
This could have problems for proper dependency tracking in the machine schedulers though I don't have a test case that shows that.
NDK r19 includes a sysroot that can be used directly by the compiler
without creating a standalone toolchain, so we just need a handful
of flags to point Clang there.
Max Kazantsev [Tue, 5 Feb 2019 04:30:37 +0000 (04:30 +0000)]
[LSR] Check SCEV on isZero() after extend. PR40514
When LSR first adds SCEVs to BaseRegs, it only does it if `isZero()` has
returned false. In the end, in invocation of `InsertFormula`, it asserts that
all values there are still not zero constants. However between these two
points, it makes some transformations, in particular extends them to wider
type.
SCEV does not give us guarantee that if `S` is not a constant zero, then
`sext(S)` is also not a constant zero. It might have missed some optimizing
transforms when it was calculating `S` and then made them when it took `sext`.
For example, it may happen if previously optimizing transforms were limited
by depth or somehow else.
This patch adds a bailout when we may end up with a zero SCEV after extension.
Teresa Johnson [Tue, 5 Feb 2019 04:09:19 +0000 (04:09 +0000)]
[SamplePGO] More pipeline changes when flattened profile used in ThinLTO postlink
Summary:
Follow on to D54819/r351476.
We also don't need to perform extra InstCombine pass when we aren't
loading the sample profile in the ThinLTO backend because we have a
flattened sample profile.
Additionally, for consistency and clarity, when we aren't reloading the
sample profile, perform ICP in the same location as non-sample PGO
backends. To this end I have moved the ICP invocation for non-SamplePGO
ThinLTO down into buildModuleSimplificationPipeline (partly addresses
the FIXME where we were previously setting this up).
[WebAssembly] Make disassembler always emit most canonical name.
Summary:
There are a few instructions that all map to the same opcode, so
when disassembling, we have to pick one. That was just the first one
before (the except_ref variant in the case of "call"), now it is the
one marked as IsCanonical in tablegen, or failing that, the shortest
name (which is typically the "canonical" one).
Also introduced a canonical "end" instruction for this purpose.
Matt Arsenault [Tue, 5 Feb 2019 00:26:12 +0000 (00:26 +0000)]
GlobalISel: Consolidate load/store legalization
The fewerElementsVectors implementation for load/stores
handles the scalar reduction case just as well, so drop
the redundant code in narrowScalar. This also introduces
support for narrowing irregular size breakdowns for
scalars.
Craig Topper [Tue, 5 Feb 2019 00:22:23 +0000 (00:22 +0000)]
[DAGCombiner] Discard pointer info when combining extract_vector_elt of a vector load when the index isn't constant
Summary:
If the index isn't constant, this transform inserts a multiply and an add on the index to calculating the base pointer for a scalar load. But we still create a memory operand with an offset of 0 and the size of the scalar access. But the access is really to an unknown offset within the original access size.
This can cause the machine scheduler to incorrectly calculate dependencies between this load and other accesses. In the case we saw, there was a 32 byte vector store that was split into two 16 byte stores, one with offset 0 and one with offset 16. The size of the memory operand for both was 16. The scheduler correctly detected the alias with the offset 0 store, but not the offset 16 store.
This patch discards the pointer info so we don't incorrectly detect aliasing. I wasn't sure if we could keep using the original offset and size without risking some other transform on the load changing the size.
I tried to reduce a test case, but there's still a lot of memory operations needed to get the scheduler to do the bad reordering. So it looked pretty fragile to maintain.
Teresa Johnson [Tue, 5 Feb 2019 00:18:38 +0000 (00:18 +0000)]
[SamplePGO] Minor efficiency improvement in samplePGO ICP
Summary:
When attaching prof metadata to promoted direct calls in SamplePGO
mode, no need to construct and use a SmallVector to pass a single count
to the ArrayRef parameter, we can simply use a brace-enclosed init list.
This made a small but consistent improvement for a ThinLTO backend
compile I was measuring.
Matt Arsenault [Mon, 4 Feb 2019 23:41:59 +0000 (23:41 +0000)]
GlobalISel: Combine g_extract with g_merge_values
Try to use the underlying source registers.
This enables legalization in more cases where some irregular
operations are widened and others narrowed.
This seems to make the test_combines_2 AArch64 test worse, since the
MERGE_VALUES has multiple uses. Since this should be required for
legalization, a hasOneUse check is probably inappropriate (or maybe
should only be used if the merge is legal?).
Evandro Menezes [Mon, 4 Feb 2019 23:29:41 +0000 (23:29 +0000)]
[PATCH] [TargetLibraryInfo] Update run time support for Windows
It seems that the run time for Windows has changed and supports more math
functions than before. Since LLVM requires at least VS2015, I assume that
this is the run time that would be redistributed with programs built with
Clang. Thus, I based this update on the header file `math.h` that
accompanies it.
This patch addresses the PR40541. Unfortunately, I have no access to a
Windows development environment to validate it.
Matt Arsenault [Mon, 4 Feb 2019 23:29:31 +0000 (23:29 +0000)]
GlobalISel: Enforce operand types for constants
A number of of tests were using imm operands, not cimm. Since CSE
relies on the exact ConstantInt* pointer used, and implicit
conversions are generally evil, also enforce the bitsize of the types.
Sam Clegg [Mon, 4 Feb 2019 23:07:34 +0000 (23:07 +0000)]
[WebAssembly] MC: Mark more function aliases as functions
Aliases of functions are now marked as function symbols even if
they are bitcast to some other other non-function type.
This is important for WebAssembly where object and function
symbols can't alias each other.
Julian Lettner [Mon, 4 Feb 2019 22:06:30 +0000 (22:06 +0000)]
[SanitizerCoverage] Clang crashes if user declares `__sancov_lowest_stack` variable
Summary:
If the user declares or defines `__sancov_lowest_stack` with an
unexpected type, then `getOrInsertGlobal` inserts a bitcast and the
following cast fails:
```
Constant *SanCovLowestStackConstant =
M.getOrInsertGlobal(SanCovLowestStackName, IntptrTy);
SanCovLowestStack = cast<GlobalVariable>(SanCovLowestStackConstant);
```
This variable is a SanitizerCoverage implementation detail and the user
should generally never have a need to access it, so we emit an error
now.
David Major [Mon, 4 Feb 2019 21:27:38 +0000 (21:27 +0000)]
gn build: Windows: use a more standard format for PDB filenames
The current build was producing names like llvm-undname.exe.pdb, which looks unusual to me at least. This switches them to the more common llvm-undname.pdb style.
Nicolai Haehnle [Mon, 4 Feb 2019 21:24:19 +0000 (21:24 +0000)]
[InstCombine] Cleanup the TFE/LWE check in AMDGPU SimplifyDemanded
Summary:
The fix added in r352904 is not quite correct, or rather misleading:
1. When the texfailctrl (TFC) argument was non-constant, the fix assumed
non-TFE/LWE, which is incorrect.
2. Regardless, this code path cannot even be hit for correct
TFE/LWE-enabled calls, because those return a struct. Added
a test case for those for completeness.
Craig Topper [Mon, 4 Feb 2019 21:24:13 +0000 (21:24 +0000)]
[CodeGen][ARC][SystemZ][WebAssembly] Use MachineInstr::isInlineAsm in more places instead of just comparing opcode. NFCI
I'm looking at adding a second INLINEASM opcode for better modeling asm-goto
as a terminator. Using the existing predicate will reduce teh number of
places that will need to use the new opcode.
David Major [Mon, 4 Feb 2019 21:20:25 +0000 (21:20 +0000)]
gn build: Windows: use a more standard format for PDB filenames
The current build was producing names like llvm-undname.exe.pdb, which looks unusual to me at least. This switches them to the more common llvm-undname.pdb style.
[Tablegen][DAG]: Fix build breakage when LLVM_ENABLE_DAGISEL_COV=1
LLVM_ENABLE_DAGISEL_COV can be used to instrument DAGISel tablegen
selection code to show which patterns along with Complex patterns were
used when selecting instructions. Unfortunately this is turned off by
default and was broken but never tested.
This required a simple fix (missing new line) to get it to build again.
Wolfgang Pieb [Mon, 4 Feb 2019 20:42:45 +0000 (20:42 +0000)]
[DEBUGINFO] Reposting r352642: Handle restore instructions in LiveDebugValues
The LiveDebugValues pass recognizes spills but not restores, which can
cause large gaps in location information for some variables, depending
on control flow. This patch make LiveDebugValues recognize restores and
generate appropriate DBG_VALUE instructions.
This patch was posted previously with r352642 and reverted in r352666 due
to buildbot errors. A missing return statement was the cause for the
failures.
Michael Kruse [Mon, 4 Feb 2019 19:55:59 +0000 (19:55 +0000)]
[WarnMissedTransforms] Do not warn about already vectorized loops.
LoopVectorize adds llvm.loop.isvectorized, but leaves
llvm.loop.vectorize.enable. Do not consider such a loop for user-forced
vectorization since vectorization already happened -- by prioritizing
llvm.loop.isvectorized except for TM_SuppressedByUser.
projects / llvm / log