David Blaikie [Thu, 6 Jul 2017 19:00:12 +0000 (19:00 +0000)]
Prototype: Reduce llvm-profdata merge memory usage further
The InstrProfWriter already stores the name and hash of the record in
the nested maps it uses for lookup while merging - this data is
duplicated in the value within the maps.
Refactor the InstrProfRecord to use a nested struct for the counters
themselves so that InstrProfWriter can use this nested struct alone
without the name or hash duplicated there.
This work is incomplete, but enough to demonstrate the value (around a
50% decrease in memory usage for a large test case (10GB -> 5GB)).
Though most of that decrease is probably from removing the
SoftInstrProfError as well, but I haven't implemented a replacement for
it yet. (it needs to go with the counters, because the operations on the
counters - merging, etc, are where the failures are - unlike the
name/hash which are totally unused by those counter-related operations
and thus easy to split out)
Ongoing discussion about removing SoftInstrProfError as a field of the
InstrProfRecord is happening on the thread that added it - including
the possibility of moving back towards an earlier version of that
proposed patch that passed SoftInstrProfError through the various APIs,
rather than as a member of InstrProfRecord.
Leo Li [Thu, 6 Jul 2017 18:47:05 +0000 (18:47 +0000)]
Modify constraints in `llvm::canReplaceOperandWithVariable`
Summary:
`Instruction::Switch`: only first operand can be set to a non-constant value.
`Instruction::InsertValue` both the first and the second operand can be set to a non-constant value.
`Instruction::Alloca` return true for non-static allocation.
[Constants] If we already have a ConstantInt*, prefer to use isZero/isOne/isMinusOne instead of isNullValue/isOneValue/isAllOnesValue inherited from Constant. NFCI
Going through the Constant methods requires redetermining that the Constant is a ConstantInt and then calling isZero/isOne/isMinusOne.
Anna Thomas [Thu, 6 Jul 2017 18:39:26 +0000 (18:39 +0000)]
[LoopUnrollRuntime] Bailout when multiple exiting blocks to the unique latch exit block
Currently, we do not support multiple exiting blocks to the
latch exit block. However, this bailout wasn't triggered when we had a
unique exit block (which is the latch exit), with multiple exiting
blocks to that unique exit.
Moved the bailout so that it's triggered in both cases and added
testcase.
Adam Nemet [Thu, 6 Jul 2017 17:51:15 +0000 (17:51 +0000)]
[opt-viewer] Move under tools, install it
We weren't installing opt-viewer and co before, this fixes the omission. I am
also moving the tools from utils/ to tools/. I believe that this is more
appropriate since these tools have matured greatly in the past year through
contributions by multiple people (thanks!) so they are ready to become
external tools.
The tools are installed under <install>/share/opt-viewer/.
I am *not* adding the llvm- prefix. If people feel strongly about adding
that, this is probably a good time since the new location will require some
mental adjustment anyway.
[PDB] Fill in "Parent" and "End" fields of scope-like symbol records
Summary:
There are a variety of records that open scopes: function scopes, block
scopes, and inlined call site scopes. These symbol records contain
Parent and End fields with the offsets of other symbol records. The End
field contains the offset of the matching S_END or S_INLINESITE_END
record. The Parent field contains the offset of the parent record, or 0
if this is a top-level scope (i.e. a function).
With this change, `llvm-pdbutil pretty -all` no longer crashes on PDBs
produced by LLD. I haven't tried a real debugger yet.
[SimplifyCFG] Move a portion of an if statement that should already be implied to an assert
Summary: In this code we got to Dom by following the predecessor link of BB. So it stands to reason that BB should also show up as a successor of Dom's terminator right? There isn't a way to have the CFG connect in only one direction is there?
[InstCombine] Add single use checks to SimplifyBSwap to ensure we are really saving instructions
Bswap isn't a simple operation so we need to make sure we are really removing a call to it before doing these simplifications.
For the case when both LHS and RHS are bswaps I've allowed it to be moved if either LHS or RHS has a single use since that at least allows us to move it later where it might find another bswap to combine with and it decreases the use count on the other side so maybe the other user can be optimized.
Wei Mi [Thu, 6 Jul 2017 15:52:14 +0000 (15:52 +0000)]
[LSR] Narrow search space by filtering non-optimal formulae with the same ScaledReg and Scale.
When the formulae search space is huge, LSR uses a series of heuristic to keep
pruning the search space until the number of possible solutions are within
certain limit.
The big hammer of the series of heuristics is NarrowSearchSpaceByPickingWinnerRegs,
which picks the register which is used by the most LSRUses and deletes the other
formulae which don't use the register. This is a effective way to prune the search
space, but quite often not a good way to keep the best solution. We saw cases before
that the heuristic pruned the best formula candidate out of search space.
To relieve the problem, we introduce a new heuristic called
NarrowSearchSpaceByFilterFormulaWithSameScaledReg. The basic idea is in order to
reduce the search space while keeping the best formula, we want to keep as many
formulae with different Scale and ScaledReg as possible. That is because the central
idea of LSR is to choose a group of loop induction variables and use those induction
variables to represent LSRUses. An induction variable candidate is often represented
by the Scale and ScaledReg in a formula. If we have more formulae with different
ScaledReg and Scale to choose, we have better opportunity to find the best solution.
That is why we believe pruning search space by only keeping the best formula with the
same Scale and ScaledReg should be more effective than PickingWinnerReg. And we use
two criteria to choose the best formula with the same Scale and ScaledReg. The first
criteria is to select the formula using less non shared registers, and the second
criteria is to select the formula with less cost got from RateFormula. The patch
implements the heuristic before NarrowSearchSpaceByPickingWinnerRegs, which is the
last resort.
Testing shows we get 1.8% and 2% on two internal benchmarks on x86. llvm nightly
testsuite performance is neutral. We also tried lsr-exp-narrow and it didn't help
on the two improved internal cases we saw.
Mikael Holmen [Thu, 6 Jul 2017 13:18:21 +0000 (13:18 +0000)]
[MachineVerifier] Add check that tied physregs aren't different.
Summary: Added MachineVerifier code to check register ties more thoroughly, especially so that physical registers that are tied are the same. This may help e.g. when creating MIR files.
- Put buildfiles into /tmp/clang-build/build, instead of /tmp/clang-build.
We checkout the sources to /tmp/clang-build/src and running
cmake in /tmp/clang-build was done by mistake.
- Don't add an extra ';' at the start of enabled projects list.
It worked either way, but looked strange.
- Minor comment update.
Made a script to build docker images easier to use.
Summary:
- Removed double indirection via command-line args (i.e. two `--`
options of `build_docker_image.sh`).
- Added a comment on how to build 2-stage clang install into the
`build_docker_image.sh`, it used to be only in the `docs/Docker.rst`.
Daniel Sanders [Thu, 6 Jul 2017 10:37:17 +0000 (10:37 +0000)]
[globalisel][tablegen] Rename and re-comment render functions to match the new MatchTables. NFC.
The conversion to MatchTable left the function names and comments referring to
C++ statements and expressions. Updated the names and comments to account for
the fact that they're no longer unconstrained statements/expressions.
Daniel Sanders [Thu, 6 Jul 2017 10:06:12 +0000 (10:06 +0000)]
[globalisel][tablegen] Rename and re-comment to match the new MatchTables. NFC.
The conversion to MatchTable left the function names and comments referring to
C++ statements and expressions. Updated the names and comments to account for
the fact that they're no longer unconstrained statements/expressions.
Max Kazantsev [Thu, 6 Jul 2017 09:57:41 +0000 (09:57 +0000)]
Revert "Revert "[IndVars] Canonicalize comparisons between non-negative values and indvars""
It seems that the patch was reverted by mistake. Clang testing showed failure of the
MathExtras.SaturatingMultiply test, however I was unable to reproduce the issue on the
fresh code base and was able to confirm that the transformation introduced by the change
does not happen in the said test. This gives a strong confidence that the actual reason of
the failure of the initial patch was somewhere else, and that problem now seems to be
fixed. Re-submitting the change to confirm that.
Soft float is more involved, because there are several different ways to
handle it based on the predicate: one and ueq need not only one, but two
libcalls to get a result. Furthermore, we have large differences between
the values returned by the AEABI and GNU functions.
AEABI functions return a nice 1 or 0 representing true and respectively
false. GNU functions generally return a value that needs to be compared
against 0 (e.g. for ogt, the value returned by the libcall is > 0 for
true). We could introduce redundant comparisons for AEABI as well, but
they don't seem easy to remove afterwards, so we do different processing
based on whether or not the result really needs to be compared against
something (and just truncate if it doesn't).
David L. Jones [Thu, 6 Jul 2017 03:23:18 +0000 (03:23 +0000)]
[lit] Fix unit test discovery for Visual Studio builds.
Fix by Andrew Ng!
The Visual Studio build can contain output for multiple configuration types (
e.g. Debug, Release & RelWithDebInfo) within the same build output
directory. Therefore when discovering unit tests, the "build mode" sub directory
containing the appropriate configuration is included in the search. This sub
directory may not always be present, so a test for its existence is required.
Avoid constructing GlobalExtensions only to find out it is empty.
Summary:
GlobalExtensions is dereferenced twice, once for iteration and then a check if it is empty.
As a ManagedStatic this dereference forces it's construction which is unnecessary.
The initial revert was done in order to prevent ongoing errors on
chromium bots such as CrWinClangLLD. However, this was done haphazardly
and I didn't realize there were test and compilation failures, so this
revert was reverted. Now that those have been fixed, we can revert the
revert of the revert.
The initial revert was done in order to prevent ongoing errors on
chromium bots such as CrWinClangLLD. However, this was done haphazardly
and I didn't realize there were test and compilation failures, so this
revert was reverted. Now that those have been fixed, we can revert the
revert of the revert.
Davide Italiano [Wed, 5 Jul 2017 22:28:28 +0000 (22:28 +0000)]
[GlobalOpt] Remove unreachable blocks before optimizing a function.
LLVM's definition of dominance allows instructions that are cyclic
in unreachable blocks, e.g.:
%pat = select i1 %condition, @global, i16* %pat
because any instruction dominates an instruction in a block that's
not reachable from entry.
So, remove unreachable blocks from the function, because a) there's
no point in analyzing them and b) GlobalOpt should otherwise grow
some more complicated logic to break these cycles.
[DependenceAnalysis] Make sure base objects are the same when comparing GEPs
The dependence analysis was returning incorrect information when using the GEPs
to compute dependences. The analysis uses the GEP indices under certain
conditions, but was doing it incorrectly when the base objects of the GEP are
aliases, but pointing to different locations in the same array.
This patch adds another check for the base objects. If the base pointer SCEVs
are not equal, then the dependence analysis should fall back on the path
that uses the whole SCEV for the dependence check. This fixes PR33567.
If a method / function returns a StringRef but the
variable is of type const std::string& a temporary string is
created (StringRef has a cast operator to std::string),
which is a suboptimal behavior.
[PDB] Add a test that verifies every known type record.
We had a lot of one-off tests for this type and that type,
or "every type that happens to be generated by this program
I built". Eventually I got a bug report filed where we were
crashing on a type that was not covered by any of these tests.
So this test carefully constructs a minimal C++ program that
will cause every type we support to be emitted. This ensures
full coverage for type records.
Sean Fertile [Wed, 5 Jul 2017 18:37:10 +0000 (18:37 +0000)]
[Power9] Disable removing extra swaps on P9.
On power 8 we sometimes insert swaps to deal with the difference between
Little-Endian and Big-Endian. The swap removal pass is supposed to clean up
these swaps. On power 9 we don't need this pass since we do not need to insert
the swaps in the first place.
Commiting on behalf of Stefan Pintilie.
Differential Revision: https://reviews.llvm.org/D34627
Tony Jiang [Wed, 5 Jul 2017 16:55:00 +0000 (16:55 +0000)]
[Power9] Exploit vector extract with variable index.
This patch adds the exploitation for new power 9 instructions which extract
variable elements from vectors:
VEXTUBLX
VEXTUBRX
VEXTUHLX
VEXTUHRX
VEXTUWLX
VEXTUWRX
Differential Revision: https://reviews.llvm.org/D34032
Commit on behalf of Zaara Syeda (syzaara@ca.ibm.com)
Tony Jiang [Wed, 5 Jul 2017 16:00:38 +0000 (16:00 +0000)]
[Power9] Exploit vector integer extend instructions when indices aren't correct.
This patch adds on to the exploitation added by https://reviews.llvm.org/D33510.
This now catches build vector nodes where the inputs are coming from sign
extended vector extract elements where the indices used by the vector extract
are not correct. We can still use the new hardware instructions by adding a
shuffle to move the elements to the correct indices. I introduced a new PPCISD
node here because adding a vector_shuffle and changing the elements of the
vector_extracts was getting undone by another DAG combine.
Commit on behalf of Zaara Syeda (syzaara@ca.ibm.com)
Differential Revision: https://reviews.llvm.org/D34009
David Blaikie [Wed, 5 Jul 2017 15:23:56 +0000 (15:23 +0000)]
DebugInfo: Generalize LoadedObjectInfoHelper from RuntimeDyld
Make it usable by any class derived (even indirectly) from
LoadedObjectInfo by allowing a custom base class to be specified and
perfect forwarding to the ctor.
Daniel Sanders [Wed, 5 Jul 2017 14:50:18 +0000 (14:50 +0000)]
[globalisel][tablegen] Finish fixing compile-time regressions by merging the matcher and emitter state machines.
Summary:
Also, made a few minor tweaks to shave off a little more cumulative memory consumption:
* All rules share a single NewMIs instead of constructing their own. Only one
will end up using it.
* Use MIs.resize(1) instead of MIs.clear();MIs.push_back(I) and prevent
GIM_RecordInsn from changing MIs[0].
David Green [Wed, 5 Jul 2017 13:25:58 +0000 (13:25 +0000)]
[IndVarSimplify] Add AShr exact flags using induction variables ranges.
This adds exact flags to AShr/LShr flags where we can statically
prove it is valid using the range of induction variables. This
allows further optimisations to remove extra loads.
[SystemZ] Simplify handling of 128-bit multiply/divide instruction
Several integer multiply/divide instructions require use of a
register pair as input and output. This patch moves setting
up the input register pair from C++ code to TableGen, simplifying
the whole process and making it more easily extensible.
Fixes a couple of whitespace errors, re-sorts the vector floating-point
instructions to make them more easily extensible, and adds a missing
pseudo instruction.
Tom Stellard [Wed, 5 Jul 2017 12:57:30 +0000 (12:57 +0000)]
CMake: Add LLVM_UTILS_INSTALL_DIR option
Summary:
This is like the LLVM_TOOLS_INSTALL_DIR option, but for the utils
that are installed when the LLVM_INSTALL_UTILS. This option
defaults to 'bin' to remain consistent with the current behavior, but
distros may want to install these to libexec/llvm.
[GlobalISel] Refactor Legalizer helpers for libcalls
We used to have a helper that replaced an instruction with a libcall.
That turns out to be too aggressive, since sometimes we need to replace
the instruction with at least two libcalls. Therefore, change our
existing helper to only create the libcall and leave the instruction
removal as a separate step. Also rename the helper accordingly.
This implements suggesting other mnemonics when an invalid one is specified,
for example:
$ echo "adXd r1,r2,#3" | llvm-mc -triple arm
<stdin>:1:1: error: invalid instruction, did you mean: add, qadd?
adXd r1,r2,#3
^
The implementation is target agnostic, but as a first step I have added it only
to the ARM backend; so the ARM backend is a good example if someone wants to
enable this too for another target.
Daniel Sanders [Wed, 5 Jul 2017 09:39:33 +0000 (09:39 +0000)]
[globalisel][tablegen] Added instruction emission to the state-machine-based matcher.
Summary:
This further improves the compile-time regressions that will be caused by a
re-commit of r303259.
Also added included preliminary work in preparation for the multi-insn emitter
since I needed to change the relevant part of the API for this patch anyway.
Max Kazantsev [Wed, 5 Jul 2017 06:38:49 +0000 (06:38 +0000)]
[IndVars] Canonicalize comparisons between non-negative values and indvars
-If there is a IndVar which is known to be non-negative, and there is a value which is also non-negative,
then signed and unsigned comparisons between them produce the same result. Both of those can be
seen in the same loop. To allow other optimizations to simplify them, we turn all instructions like
projects / llvm / log