Nikolai Bozhenov [Fri, 20 Oct 2017 10:08:47 +0000 (10:08 +0000)]
[ValueTracking] Enabling ValueTracking patch by default
(recommit #2 after checking for timeout issue).
The original patch was an improvement to IR ValueTracking on
non-negative integers. It has been checked in to trunk (D18777,
r284022). But was disabled by default due to performance regressions.
Perf impact has improved. The patch would be enabled by default.
Alex Bradbury [Thu, 19 Oct 2017 21:37:38 +0000 (21:37 +0000)]
[RISCV] Initial codegen support for ALU operations
This adds the minimum necessary to support codegen for simple ALU operations
on RV32. Prolog and epilog insertion, support for memory operations etc etc
follow in future patches.
Leave guessInstructionProperties=1 until https://reviews.llvm.org/D37065 is
reviewed and lands.
Graham Yiu [Thu, 19 Oct 2017 18:16:31 +0000 (18:16 +0000)]
The cost of splitting a large vector instruction is not being taken into account by the getUserCost function. This was leading to some loops being over unrolled. The cost of a vector instruction is now being multiplied by the cost of the type legalization. This will return a more accurate cost.
Committing on behalf on Brad Nemanich (brad.nemanich@ibm.com)
Serguei Katkov [Thu, 19 Oct 2017 11:16:03 +0000 (11:16 +0000)]
Fix APFloat from string conversion for Inf
The method IEEEFloat::convertFromStringSpecials() does not recognize
the "+Inf" and "-Inf" strings but these strings are printed for
the double Infinities by the IEEEFloat::toString().
This patch adds the "+Inf" and "-Inf" strings to the list of recognized
patterns in IEEEFloat::convertFromStringSpecials().
whitequark [Thu, 19 Oct 2017 04:47:48 +0000 (04:47 +0000)]
[MergeFunctions] Don't blindly RAUW a GlobalValue with a ConstantExpr.
MergeFunctions uses (through FunctionComparator) a map of GlobalValues
to identifiers because it needs to compare functions and globals
do not have an inherent total order. Thus, FunctionComparator
(through GlobalNumberState) has a ValueMap<GlobalValue *>.
r315852 added a RAUW on globals that may have been previously
encountered by the FunctionComparator, which would replace
a GlobalValue * key with a ConstantExpr *, which is illegal.
This commit adjusts that code path to remove the function being
replaced from the ValueMap as well.
Chris Bieneman [Thu, 19 Oct 2017 00:43:48 +0000 (00:43 +0000)]
[CMake] Allow parent projects to use in-source builds
LLVM checks if it is performing an in-source build and then stop the
build. However, this check is also triggered if LLVM is being build as
part of a parent project, which prevents the parent project itself from
using in-source builds. For example, CMake allows a parent project to
specify the output of its subproject:
add_subdirectory(llvm llvm_build)
This tells CMake to conduct an out-tree build of LLVM, which without
this patch will still fails because what is being tested is the parent
project, not LLVM. This is fixed by using the "CURRENT" variable, which
is only concerned by the CMakeLists that is actually bein processed at
the moment.
Tests:
Ran `make check-llvm`.
Patch by Henrique Jung <henriquenj_AT_gmail_DOT_com>
Vedant Kumar [Wed, 18 Oct 2017 23:58:28 +0000 (23:58 +0000)]
[llvm-cov] Move LineCoverageIterator to libCoverage. NFC.
LineCoverageIterator makes it easy for clients of coverage data to
determine line execution counts for a file or function. The coverage
iteration logic is tricky enough that it really pays not to have
multiple copies of it. Hopefully having just one implementation in LLVM
will make the iteration logic easier to test, reuse, and update.
This commit is NFC but I've added a unit test to go along with it just
because it's easy to do now.
Justin Bogner [Wed, 18 Oct 2017 23:33:31 +0000 (23:33 +0000)]
GISel: Canonicalize select tests using update_mir_test_checks
This runs `udpate_mir_test_checks --add-vreg-checks` on the tests taht
are already more or less in the format that generates, so that there
will be less churn in some upcoming changes.
Justin Bogner [Wed, 18 Oct 2017 23:18:12 +0000 (23:18 +0000)]
Canonicalize a large number of mir tests using update_mir_test_checks
This converts a large and somewhat arbitrary set of tests to use
update_mir_test_checks. I ran the script on all of the tests I expect
to need to modify for an upcoming mir syntax change and kept the ones
that obviously didn't change the tests in ways that might make it
harder to understand.
Justin Bogner [Wed, 18 Oct 2017 22:39:55 +0000 (22:39 +0000)]
update_mir_test_checks: Support adding checks for vreg classes
This is a temporary hack to support adding checks for the "registers:"
block of mir functions. This is necessary to convert a number of tests
so that there's less churn when we change the MIR printer to put the
vreg classes on defs instead of in their own block.
Sanjoy Das [Wed, 18 Oct 2017 22:00:57 +0000 (22:00 +0000)]
Revert "[ScalarEvolution] Handling for ICmp occuring in the evolution chain."
This reverts commit r316054. There was some confusion over the review process:
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20171016/495884.html
Vedant Kumar [Wed, 18 Oct 2017 18:52:29 +0000 (18:52 +0000)]
[llvm-cov] Suppress sub-line highlights in simple cases
llvm-cov tends to highlight too many regions because its policy is to
highlight all region entry segments. This can look confusing to users:
not all region entry segments are interesting and deserve highlighting.
Emitting these highlights only when the region count differs from the
line count is a more user-friendly policy.
Vedant Kumar [Wed, 18 Oct 2017 18:52:28 +0000 (18:52 +0000)]
[llvm-cov] Pass LineCoverageStats in SourceCoverageView. NFC.
Instead of copying around the wrapped segment and the list of line
segments, just pass a reference to a LineCoverageStats object. This
simplifies the interface. It also makes an upcoming change to suppress
distracting highlights possible.
This patch lets the llvm tools handle the new HVX target features that
are added by frontend (clang). The target-features are of the form
"hvx-length64b" for 64 Byte HVX mode, "hvx-length128b" for 128 Byte mode HVX.
"hvx-double" is an alias to "hvx-length128b" and is soon will be deprecated.
The hvx version target feature is upgated form "+hvx" to "+hvxv{version_number}.
Eg: "+hvxv62"
For the correct HVX code generation, the user must use the following
target features.
For 64B mode: "+hvxv62" "+hvx-length64b"
For 128B mode: "+hvxv62" "+hvx-length128b"
Clang picks a default length if none is specified. If for some reason,
no hvx-length is specified to llvm, the compilation will bail out.
There is a corresponding clang patch.
[Hexagon] Mark vector loads as predicable, update instruction mappings
All loads of form V6_vL32b_{,cur,nt,tmp,nt_cur,nt_tmp}_{ai,pi,ppu} are
predicable on v62 (but not on v60). Mark them all as predicable in the
instruction definitions, and handle the v60 case in HII::isPredicable.
Justin Bogner [Wed, 18 Oct 2017 15:37:09 +0000 (15:37 +0000)]
update_mir_test_checks: Do a better job of disambiguating names
Matching prefixes isn't good enough, because it leads to things like
calling the first constant C3 just because there were two copies
before it. Tighten up the check to match more precisely, but also be
careful about ambiguity when dealing with target opcodes that end in a
number.
Justin Bogner [Wed, 18 Oct 2017 15:34:33 +0000 (15:34 +0000)]
AArch64/GISel: Fix a couple of tests that were testing the wrong thing
Fix a couple of tests that were extending the wrong vreg, and
regenerate their checks with update_mir_test_checks. This looks like
it was a copy-paste or test update error.
Simon Dardis [Wed, 18 Oct 2017 14:35:29 +0000 (14:35 +0000)]
[mips] Fix analyzeBranch to handle debug data
In the case where there was a conditional branch followed by a unconditional
branch with debug instruction separating them, MipsInstrInfo::analyzeBranch
would not skip past debug instruction when searching for the second branch
which give erroneous results about the control flow of the block.
This could lead to the branch folder to merge the non-fall through case
into it's predecessor, leaving the conditional branch with a dangling
basic block operand.
This resolves PR34975.
Thanks to Alexander Richardson for reporting the issue!
Nikolai Bozhenov [Wed, 18 Oct 2017 14:24:50 +0000 (14:24 +0000)]
Fixup patch for revision rL316070.
Added check that type of CmpConst and source type of trunc are equal
for correct matching of the case when we can set widened C constant
equal to CmpConstant.
%cond = cmp iN %x, CmpConst
%tr = trunc iN %x to iK
%narrowsel = select i1 %cond, iK %t, iK C
Patch by: Gainullin, Artur <artur.gainullin@intel.com>
[AVX512][AVX2]Cost calculation for interleave load/store patterns {v8i8,v16i8,v32i8,v64i8}
This patch adds accurate instructions cost.
The formula presents two cases(stride 3 and stride 4) and calculates the cost according to the VF and stride.
Hiroshi Inoue [Wed, 18 Oct 2017 10:31:19 +0000 (10:31 +0000)]
[PowerPC] Use helper functions to check sign-/zero-extended value
Helper functions to identify sign- and zero-extending machine instruction is introduced in rL315888.
This patch makes PPCInstrInfo::optimizeCompareInstr use the helper functions. It simplifies the code and also makes possible more optimizations since the helper can do more analysis than the original check code; I observed about 5000 more compare instructions are eliminated while building LLVM.
Also, this patch fixes a bug in helpers on ANDIo instruction handling due to the order of checks. This bug causes a failure in an existing test case for optimizeCompareInstr.
Nikolai Bozhenov [Wed, 18 Oct 2017 09:28:09 +0000 (09:28 +0000)]
Improve lookThroughCast function.
Summary:
When we have the following case:
%cond = cmp iN %x, CmpConst
%tr = trunc iN %x to iK
%narrowsel = select i1 %cond, iK %t, iK C
We could possibly match only min/max pattern after looking through cast.
So it is more profitable if widened C constant will be equal CmpConst.
That is why just set widened C constant equal to CmpConst, because there
is a further check in this function that trunc CmpConst == C.
Also description for lookTroughCast function was added.
Serguei Katkov [Wed, 18 Oct 2017 06:26:39 +0000 (06:26 +0000)]
Fix the incorrect detection of ICONV_LIBRARY_PATH
This is introduced in rL308711.
Check for c library is incorrect here just because libc will be found always
and it does not mean that iconv is presented.
Thank to Andrew Krasny for narrowing down the root cause.
Vlad Tsyrklevich [Wed, 18 Oct 2017 04:27:53 +0000 (04:27 +0000)]
Statically link llvm-cfi-verify's libraries.
Summary:
llvm-cfi-verify (D38379) introduced a potential build failure when compiling with `-DLLVM_BUILD_LLVM_DYLIB=ON -DLLVM_LINK_LLVM_DYLIB=ON`. Specific versions of cmake seem to treat the `add_subdirectory()` rule differently. It seems as if old versions of cmake BFS these rules, adding them to the fringe for expansion later. Newer versions of cmake seem to immediately execute CMakeFiles that are present in this subdirectory.
If the subdirectory is expanded through the fringe, the globbing resultant from `llvm_add_implicit_projects()` from `cmake/modules/AddLLVM.cmake:1012` means that `tools/llvm-shlib/CMakeFile.txt` gets executed before `tools/llvm-cfi-verify/lib/CMakeFile.txt`. As the latter CMakeFile adds a new library, this expansion order means that the library files required the unit tests in `unittests/tools/llvm-cfi-verify/` are not present in the dynamic library. This causes unit tests to fail as the required functions can't be found.
This change now ensures that the libraries created by `llvm-cfi-verify` are statically linked into the unit tests. As `tools/llvm-cfi-verify/lib` no longer adds anything to `llvm-shlib`, there should be no concern about the order-of-compilation.
Justin Bogner [Wed, 18 Oct 2017 02:20:31 +0000 (02:20 +0000)]
Add a utility to update MIR checks, similar to update_llc_test_checks
This adds update_mir_test_checks, which updates the check lines in mir
tests. This can only update tests that start and end with .mir
currently (ie, -run-pass) but it should be sufficient for updating at
least some of the GlobalISel tests.
Jatin Bhateja [Wed, 18 Oct 2017 01:36:16 +0000 (01:36 +0000)]
[ScalarEvolution] Handling for ICmp occuring in the evolution chain.
Summary:
If a compare instruction is same or inverse of the compare in the
branch of the loop latch, then return a constant evolution node.
Currently scope of evaluation is limited to SCEV computation for
PHI nodes.
This shall facilitate computations of loop exit counts in cases
where compare appears in the evolution chain of induction variables.
Will fix PR 34538
Reviewers: sanjoy, hfinkel, junryoungju
Adrian Prantl [Wed, 18 Oct 2017 01:11:01 +0000 (01:11 +0000)]
Verifier: Ignore CUs pulled in by ODR-uniqued types.
When more than one Module is imported into the same context, such as during
an LTO build before linking the modules, ODR type uniquing may cause types
to point to a different CU. This check does not make sense in this case.
Adrian Prantl [Wed, 18 Oct 2017 00:49:31 +0000 (00:49 +0000)]
Verifier: Ignore CUs pulled in by ODR-uniqued types.
When more than one Module is imported into the same context, such as during
an LTO build before linking the modules, ODR type uniquing may cause types
to point to a different CU. This check does not make sense in this case.
[GlobalDCE] Use DenseMap instead of unordered_multimap for GVDependencies.
Summary:
std::unordered_multimap happens to be very slow when the number of elements
grows large. On one of our internal applications we observed a 17x compile time
improvement from changing it to DenseMap.
Tim Northover [Tue, 17 Oct 2017 21:43:52 +0000 (21:43 +0000)]
AArch64: account for possible frame index operand in compares.
If the address of a local is used in a comparison, AArch64 can fold the
address-calculation into the comparison via "adds". Unfortunately, a couple of
places (both hit in this one test) are not ready to deal with that yet and just
assume the first source operand is a register.
Jessica Paquette [Tue, 17 Oct 2017 21:11:58 +0000 (21:11 +0000)]
[MachineOutliner][NFC] Clean up prune logic a bit
Move the prune logic in pruneOverlaps to a new function, prune. This lets us
reuse the prune functionality. Makes the code a bit more readable. It'll also
make it easier to emit remarks/debug statements for pruned functions.
Jessica Paquette [Tue, 17 Oct 2017 19:03:23 +0000 (19:03 +0000)]
[MachineOutliner][NFC] Move decrement logic to OutlinedFunction
This commit moves the decrement logic for outlined functions into the class,
and makes OccurrenceCount private. It can now be accessed via
getOccurrenceCount().
This makes it more difficult to accidentally introduce bugs by incorrectly
decrementing the occurrence count on OutlinedFunctions.
Jessica Paquette [Tue, 17 Oct 2017 18:43:15 +0000 (18:43 +0000)]
[MachineOutliner][NFC] Move end index calculation into Candidate
Cleanup to Candidate that moves all end index calculations into
Candidate.endIdx(). For the sake of consistency, StartIdx and Len are now
private members, and can be accessed with length() and startIdx() respectively.
projects / llvm / log