Martin Storsjo [Sat, 19 Aug 2017 19:47:48 +0000 (19:47 +0000)]
[ARM] Check the right order for halves of VZIP/VUZP if both parts are used
This is the exact same fix as in SVN r247254. In that commit, the fix was
applied only for isVTRNMask and isVTRN_v_undef_Mask, but the same issue
is present for VZIP/VUZP as well.
Jatin Bhateja [Sat, 19 Aug 2017 18:08:59 +0000 (18:08 +0000)]
[DAGCombiner] Extending pattern detection for vector shuffle.
Summary:
If all the operands of a BUILD_VECTOR extract elements from same vector then split the
vector efficiently based on the maximum vector access index.
Teresa Johnson [Sat, 19 Aug 2017 18:04:25 +0000 (18:04 +0000)]
[ThinLTO] Fix ThinLTO crash
Summary:
Follow up to fix in r311023, which fixed the case where the combined
index is written to disk. The same samplePGO logic exists for the
in-memory index when computing imports, so we need to filter out
GlobalVariable summaries there too.
Chandler Carruth [Sat, 19 Aug 2017 06:56:11 +0000 (06:56 +0000)]
[Inliner] Fix a nasty bug when inlining a non-recursive trace of
a function into itself.
We tried to fix this before in r306495 but that got reverted as the
assert was actually hit.
This fixes the original bug (which we seem to have lost track of with
the revert) by blocking a second remapping when the function being
inlined is also the caller and the remapping could succeed but
erroneously.
The included test case would actually load from an inlined copy of the
alloca before this change, failing to load the stored value and
miscompiling.
Many thanks to Richard Smith for diagnosing a user miscompile to this
bug, and to Kyle for the first attempt and initial analysis and David Li
for remembering the issue and how to fix it and suggesting the patch.
I'm just stitching it together and landing it. =]
Chandler Carruth [Sat, 19 Aug 2017 05:01:19 +0000 (05:01 +0000)]
[x86] Teach the cmov converter to aggressively convert cmovs with memory
operands into control flow.
We have seen periodically performance problems with cmov where one
operand comes from memory. On modern x86 processors with strong branch
predictors and speculative execution, this tends to be much better done
with a branch than cmov. We routinely see cmov stalling while the load
is completed rather than continuing, and if there are subsequent
branches, they cannot be speculated in turn.
Also, in many (even simple) cases, macro fusion causes the control flow
version to be fewer uops.
Consider the IACA output for the initial sequence of code in a very hot
function in one of our internal benchmarks that motivates this, and notice the
micro-op reduction provided.
Before, SNB:
```
Throughput Analysis Report
--------------------------
Block Throughput: 2.20 Cycles Throughput Bottleneck: Port1
Note that this cannot be usefully restricted to inner loops. Much of the
hot code we see hitting this is not in an inner loop or not in a loop at
all. The optimization still remains effective and indeed critical for
some of our code.
I have run a suite of internal benchmarks with this change. I saw a few
very significant improvements and a very few minor regressions,
but overall this change rarely has a significant effect. However, the
improvements were very significant, and in quite important routines
responsible for a great deal of our C++ CPU cycles. The gains pretty
clealy outweigh the regressions for us.
I also ran the test-suite and SPEC2006. Only 11 binaries changed at all
and none of them showed any regressions.
Amjad Aboud at Intel also ran this over their benchmarks and saw no
regressions.
Chandler Carruth [Sat, 19 Aug 2017 04:28:20 +0000 (04:28 +0000)]
[x86] Refactor the CMOV conversion pass to be more flexible.
The primary thing that this accomplishes is to allow future re-use of
these routines in more contexts and clarify the behavior w.r.t. loops.
For example, if handling outer loops is desirable, doing so in
a inside-out order becomes straight forward because it walks the loop
nest itself (rather than walking the function's basic blocks) and
de-couples the CMOV rewriting from the loop structure as there isn't
actually anything loop-specific about this transformation.
This patch should be essentially a no-op. It potentially changes the
order in which we visit the inner loops, but otherwise should merely set
the stage for subsequent changes.
Matthias Braun [Sat, 19 Aug 2017 01:21:11 +0000 (01:21 +0000)]
ARMRegsiterInfo: Define more ssub indexes; NFC
This doesn't really change anything as Tablegen would have inferred
those indices anyway; defining them gives us shorter names that are
easier to read while debugging (i.e. "ssub_4" rather than
"dsub2_then_ssub_0")
Eric Beckmann [Sat, 19 Aug 2017 00:37:41 +0000 (00:37 +0000)]
llvm-mt: Merge manifest namespaces.
mt.exe performs a tree merge where certain element nodes are combined
into one. This introduces the possibility of xml namespaces conflicting
with each other. The original mt.exe has a hierarchy whereby certain
namespace names can override others, and nodes that would then end up in
ambigious namespaces have their namespaces explicitly defined. This
namespace handles this merging process.
Max Kazantsev [Fri, 18 Aug 2017 22:50:29 +0000 (22:50 +0000)]
[IRCE] Fix buggy behavior in Clamp
Clamp function was too optimistic when choosing signed or unsigned min/max function for calculations.
In fact, `!IsSignedPredicate` guarantees us that `Smallest` and `Greatest` can be compared safely using unsigned
predicates, but we did not check this for `S` which can in theory be negative.
This patch makes Clamp use signed min/max for cases when it fails to prove `S` being non-negative,
and it adds a test where such situation may lead to incorrect conditions calculation.
Justin Bogner [Fri, 18 Aug 2017 21:38:03 +0000 (21:38 +0000)]
IR: Make stripDebugInfo robust against (invalid) empty basic blocks
Since stripDebugInfo runs before the verifier when reading IR, we can
end up in a situation where we read some invalid IR but don't know its
invalid yet. Before this patch we would crash in stripDebugInfo when
given IR with a completely empty basic block, and after we get a nice
error from the verifier instead.
Summary:
Augment SanitizerCoverage to insert maximum stack depth tracing for
use by libFuzzer. The new instrumentation is enabled by the flag
-fsanitize-coverage=stack-depth and is compatible with the existing
trace-pc-guard coverage. The user must also declare the following
global variable in their code:
thread_local uintptr_t __sancov_lowest_stack
Jakub Kuderski [Fri, 18 Aug 2017 17:06:37 +0000 (17:06 +0000)]
[Dominators] Don't print the whole tree when running with -debug
As the incremental API is now used in several transforms, printing
the whole dominator tree creates a lot of noise when running with
the `-debug` flag. This patch fixes that.
Brian Gesiak [Fri, 18 Aug 2017 15:35:53 +0000 (15:35 +0000)]
[Lexicon] Add "GEP"
Summary:
`getelementptr` is frequently abbreviated as "GEP", often in source files that
do not ever reference the full name of the instruction. Add it to the Lexicon,
in case readers go to look for what it means there.
Test plan:
1. `ninja sphinx`
2. Confirm that the rendered docs HTML contains the new "GEP" entry
Sam Parker [Fri, 18 Aug 2017 14:27:51 +0000 (14:27 +0000)]
[ARM] Add PostRAScheduler option
This patch adds the option to allow also using the PostRA scheduler,
which brings the ARM backend inline with AArch64 targets. The
SchedModel can also set 'PostRAScheduler', as the R52 does, so also
query this property in the overridden function.
Sjoerd Meijer [Fri, 18 Aug 2017 10:51:14 +0000 (10:51 +0000)]
[AArch64] Do not promote f16 when subtarget HasFullFP16
Armv8.2-A adds FP16 support, i.e. f16 is not only a storage-only type, but it
also supports performing data processing on 16-bit floating-point quantities.
All the necessary (tablegen) groundwork of adding the ARMv8.2-A FP16 (scalar)
instructions was done in D15014. To take advantage of this, this patch avoids
promotion of f16 to f32 types when the subtarget supports FullFP16, which
enables instruction selection of these FP16 instructions.
Renato Golin [Fri, 18 Aug 2017 10:35:42 +0000 (10:35 +0000)]
[Triple] Define OS Check for Haiku
This adds the OS check for the Haiku operating system, as it was
missing in the Triple class. Tests for x86_64-unknown-haiku and
i586-pc-haiku were also added.
These patches only affect Haiku and are completely harmless for
other platforms.
Ilya Biryukov [Fri, 18 Aug 2017 09:37:23 +0000 (09:37 +0000)]
Addressed some security issues in Dockerfiles.
Summary:
- Removed --trust-server-cert from `svn checkout` invocations.
Installing 'ca-certificates' package on ubuntu adds required CAs to
the system and svn can do proper checkout using https.
- Added checksum verification when installing cmake from cmake.org.
Diana Picus [Fri, 18 Aug 2017 09:31:21 +0000 (09:31 +0000)]
Revert "GlobalISel (AArch64): fix ABI at border between GPRs and SP."
This reverts commit e8fd20964798ca6d46d2729dd3a789707a6416da in an
attempt to appease the GlobalISel buildbot, which fails in the
test-suite with errors like
fpcmp: files differ without tolerance allowance
Sam Parker [Fri, 18 Aug 2017 08:39:54 +0000 (08:39 +0000)]
[AArch64] Remove DecodeAuthLoadWriteback
The BaseAuthLoad instruction class was incorrectly passing an empty
constraint string to its parent, so I have corrected this. This makes
the DecodeAuthLoadWriteback function redundant, so I've also removed
it.
Alex Bradbury [Fri, 18 Aug 2017 06:45:34 +0000 (06:45 +0000)]
Refine report_fatal_error guidance after post-commit review
Use text suggested by Justin Bogner in post-commit review of r311146
<http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20170814/479898.html>,
which makes it clear that report_fatal_error shouldn't be used when there is a
practicable alternative. Also make this clearer in CodingStandards.
Alex Bradbury [Fri, 18 Aug 2017 05:29:21 +0000 (05:29 +0000)]
Give guidance on report_fatal_error in CodingStandards.rst and ProgrammersManual.rst
The current ProgrammersManual.rst document has a lot of well-written
documentation on error handling thanks to @lhames. It suggests errors can be
split cleanly into "programmatic" and "recoverable" errors. However, the
reality in current LLVM seems to be there are a number of cases where a
non-programmatic error is not easily recoverable. Therefore, add a note to
indicate the existence of report_fatal_error for these cases. I've also added
a reminder to CodingStandards.rst in the section on assertions, to indicate
that llvm_unreachable and assertions should not be relied upon to report
errors triggered by user input.
The ProgrammersManual is also silent on the use of LLVMContext::diagnose,
which is used in BPF+WebAssembly+AMDGPU to report some errors during
instruction selection. I don't address that in this patch, as it's not quite
clear how to fit in to the current error handling story
Craig Topper [Thu, 17 Aug 2017 23:20:57 +0000 (23:20 +0000)]
[X86] Remove SSE/AVX patterns for AND/XOR/OR/ANDN that checked for the inputs being bitcasted from floating point types.
There's really no reason to do this we should just let isel pick the integer version and let the execution dependency fixing pass take care of moving to FP if necessary.
It's not very reliable to look for bitcasts at the edges of patterns. If for some reason one input was bitcasted and the other wasn't, or if one was a v4f32 bitcast and one was a v2f64 bitcast, we would have fallen back to the integer pattern anyway.
Tim Northover [Thu, 17 Aug 2017 23:14:01 +0000 (23:14 +0000)]
GlobalISel (AArch64): fix ABI at border between GPRs and SP.
If a struct would end up half in GPRs and half on SP the ABI says it should
actually go entirely on the stack. We were getting this wrong in GlobalISel
before, causing compatibility issues.
Geoff Berry [Thu, 17 Aug 2017 23:06:55 +0000 (23:06 +0000)]
Re-enable "[MachineCopyPropagation] Extend pass to do COPY source forwarding"
Two issues identified by buildbots were addressed:
- The pass no longer forwards COPYs to physical register uses, since
doing so can break code that implicitly relies on the physical
register number of the use.
- The pass no longer forwards COPYs to undef uses, since doing so
can break the machine verifier by creating LiveRanges that don't
end on a use (since the undef operand is not considered a use).
[MachineCopyPropagation] Extend pass to do COPY source forwarding
This change extends MachineCopyPropagation to do COPY source forwarding.
This change also extends the MachineCopyPropagation pass to be able to
be run during register allocation, after physical registers have been
assigned, but before the virtual registers have been re-written, which
allows it to remove virtual register COPY LiveIntervals that become dead
through the forwarding of all of their uses.
Tom Stellard [Thu, 17 Aug 2017 22:20:04 +0000 (22:20 +0000)]
AMDGPU: Add R600InstPrinter class
Summary:
This is step towards separating the GCN and R600 tablegen'd code.
This is a little awkward for now, because the R600 functions won't have the
MCSubtargetInfo parameter, so we need to have AMDMGPUInstPrinter
delegate to R600InstPrinter, but once the tablegen'd code is split,
we will be able to drop the delegation and use R600InstPrinter directly.
Zachary Turner [Thu, 17 Aug 2017 20:04:51 +0000 (20:04 +0000)]
[llvm-pdbutil] Fix some dumping issues.
When dumping, we were treating the S_INLINESITESYM as referring
to a type record, when it actually refers to an id record. We
had this correct in TypeIndexDiscovery, so our merging algorithm
should be fine, but we had it wrong in the dumper, which means it
would appear to work most of the time, unless the index was out
of bounds in the type stream, when it would fail. Fixed this, and
audited a few other cases to make them match the behavior in
TypeIndexDiscovery.
Also, I've now observed a new symbol record with kind 0x1168 which
I have no clue what it is, so to avoid crashing we have to just
print "Unknown Symbol Kind".
Zachary Turner [Thu, 17 Aug 2017 20:04:31 +0000 (20:04 +0000)]
Fix a few minor issues when dumping symbols.
1) We weren't handling symbol types that weren't able to parse,
even if we knew what the leaf type was. This was triggering
when trying to dump /DEBUG:FASTLINK PDBs, where we expect a
certain symbol to show up, but we just don't know how to parse
it.
2) We lost the code for dumping record bytes, so this was added
back.
Lang Hames [Thu, 17 Aug 2017 18:21:53 +0000 (18:21 +0000)]
[docs] Tweak phrasing of the varargs explanation in the command section of the
CMake primer.
This moves the introduction of the ARGV/ARGN variables up to immmediately follow
the introduction of the concept of variable argument functions, and explicitly
connects this concept to C varargs functions.
Sanjay Patel [Thu, 17 Aug 2017 17:07:37 +0000 (17:07 +0000)]
[x86] add tests for vector select-of-constants; NFC
We've discussed canonicalizing to this form in IR, so the backend
should be prepared to lower these in ways better than what we see
here in most cases.
Adrian Prantl [Thu, 17 Aug 2017 16:57:13 +0000 (16:57 +0000)]
Improve line debug info when translating a CaseBlock to SDNodes.
The SelectionDAGBuilder translates various conditional branches into
CaseBlocks which are then translated into SDNodes. If a conditional
branch results in multiple CaseBlocks only the first CaseBlock is
translated into SDNodes immediately, the rest of the CaseBlocks are
put in a queue and processed when all LLVM IR instructions in the
basic block have been processed.
When a CaseBlock is transformed into SDNodes the SelectionDAGBuilder
is queried for the current LLVM IR instruction and the resulting
SDNodes are annotated with the debug info of the current
instruction (if it exists and has debug metadata).
When the deferred CaseBlocks are processed, the SelectionDAGBuilder
does not have a current LLVM IR instruction, and the resulting SDNodes
will not have any debuginfo. As DwarfDebug::beginInstruction() outputs
a .loc directive for the first instruction in a labeled
block (typically the case for something coming from a CaseBlock) this
tends to produce a line-0 directive.
This patch changes the handling of CaseBlocks to store the current
instruction's debug info into the CaseBlock when it is created (and the
SelectionDAGBuilder knows the current instruction) and to always use
the stored debug info when translating a CaseBlock to SDNodes.
Jakub Kuderski [Thu, 17 Aug 2017 16:45:35 +0000 (16:45 +0000)]
[Dominators] Teach LoopUnswitch to use the incremental API
Summary:
This patch makes LoopUnswitch use new incremental API for updating dominators.
It also updates SplitCriticalEdge, as it is called in LoopUnswitch.
There doesn't seem to be any noticeable performance difference when bootstrapping clang with this patch.
Simon Dardis [Thu, 17 Aug 2017 14:14:25 +0000 (14:14 +0000)]
[dfsan] Add explicit zero extensions for shadow parameters in function wrappers.
In the case where dfsan provides a custom wrapper for a function,
shadow parameters are added for each parameter of the function.
These parameters are i16s. For targets which do not consider this
a legal type, the lack of sign extension information would cause
LLVM to generate anyexts around their usage with phi variables
and calling convention logic.
Address this by introducing zero exts for each shadow parameter.
We currently visit them twice.
Once, through `visitMDNode()` -> (the code generated by)
`../include/llvm/IR/Metadata.def:109` -> `visitDIGlobalVariable()`
Then, through `visitMDNode()` -> `visitDIGlobalVariableExpression()`
-> `visitDIGlobalVariable()`
This results in verification failures printed twice, e.g.:
$ ./opt -verify ../../test/DebugInfo/pr34186.ll
missing global variable type
!4 = distinct !DIGlobalVariable(name: "pat", scope: !0,
file: !1, line: 27, isLocal: true, isDefinition: true)
missing global variable type
!4 = distinct !DIGlobalVariable(name: "pat", scope: !0,
file: !1, line: 27, isLocal: true, isDefinition: true)
./opt: ../../test/DebugInfo/pr34186.ll: error: input module is broken!
The patch removes one call so we ensure each GV is visited exactly once.
Ayal Zaks [Thu, 17 Aug 2017 09:29:59 +0000 (09:29 +0000)]
[LV] Using VPlan to model the vectorized code and drive its transformation
VPlan is an ongoing effort to refactor and extend the Loop Vectorizer. This
patch introduces the VPlan model into LV and uses it to represent the vectorized
code and drive the generation of vectorized IR.
In this patch VPlan models the vectorized loop body: the vectorized control-flow
is represented using VPlan's Hierarchical CFG, with predication refactored from
being a post-vectorization-step into a vectorization planning step modeling
if-then VPRegionBlocks, and generating code inline with non-predicated code. The
vectorized code within each VPBasicBlock is represented as a sequence of
Recipes, each responsible for modelling and generating a sequence of IR
instructions. To keep the size of this commit manageable the Recipes in this
patch are coarse-grained and capture large chunks of LV's code-generation logic.
The constructed VPlans are dumped in dot format under -debug.
This commit retains current vectorizer output, except for minor instruction
reorderings; see associated modifications to lit tests.
For further details on the VPlan model see docs/Proposals/VectorizationPlan.rst
and its references.
Daniel Sanders [Thu, 17 Aug 2017 09:26:14 +0000 (09:26 +0000)]
Re-commit: [globalisel][tablegen] Support zero-instruction emission.
Summary:
Support the case where an operand of a pattern is also the whole of the
result pattern. In this case the original result and all its uses must be
replaced by the operand. However, register class restrictions can require
a COPY. This patch handles both cases by always emitting the copy and
leaving it for the register allocator to optimize.
The previous commit failed on Windows machines due to a flaw in the sort
predicate which allowed both A < B < C and B == C to be satisfied
simultaneously. The cause of this was some sloppiness in the priority order of
G_CONSTANT instructions compared to other instructions. These had equal priority
because it makes no difference, however there were operands had higher priority
than G_CONSTANT but lower priority than any other instruction. As a result, a
priority order between G_CONSTANT and other instructions must be enforced to
ensure the predicate defines a strict weak order.
The idea of this patch is to continue the scheduler state over an MBB boundary
in the case where the successor block has only one predecessor. This means
that the scheduler will continue in the successor block (after emitting any
branch instructions) with e.g. maintained processor resource counters.
Benchmarks have been confirmed to benefit from this.
The algorithm in MachineScheduler.cpp that extracts scheduling regions of an
MBB has been extended so that the strategy may optionally reverse the order
of processing the regions themselves. This is controlled by a new method
doMBBSchedRegionsTopDown(), which defaults to false.
Handling the top-most region of an MBB first also means that a top-down
scheduler can continue the scheduler state across any scheduling boundary
between to regions inside MBB.
Review: Ulrich Weigand, Matthias Braun, Andy Trick.
https://reviews.llvm.org/D35053
projects / llvm / log