Alexey Samsonov [Fri, 18 Apr 2014 20:30:27 +0000 (20:30 +0000)]
[DWARF parser] Respect address ranges specified in compile unit DIE.
When address ranges for compile unit are specified in compile unit DIE
itself, there is no need to collect ranges from children subprogram DIEs.
This change speeds up llvm-symbolizer on Clang-produced binaries with
full debug info. For instance, symbolizing a first address in a 1Gb binary
is now 2x faster (1s vs. 2s).
llvm-svn: 206641
Justin Bogner [Fri, 18 Apr 2014 20:27:24 +0000 (20:27 +0000)]
Teach users of OnDiskHashTable to define hash_value and offset types
This paves the way to making OnDiskHashTable work with hashes that are
not 32 bits wide and to making OnDiskHashTable work very large hash
tables. The LLVM change to use these types is upcoming.
llvm-svn: 206640
Rui Ueyama [Fri, 18 Apr 2014 20:06:00 +0000 (20:06 +0000)]
Remove redundant "explicit" keywords.
llvm-svn: 206639
Rui Ueyama [Fri, 18 Apr 2014 19:59:33 +0000 (19:59 +0000)]
Use switch for readability.
llvm-svn: 206638
Justin Bogner [Fri, 18 Apr 2014 19:57:06 +0000 (19:57 +0000)]
Remove OnDiskHashTable.h, since it's been moved to llvm
llvm-svn: 206637
Rui Ueyama [Fri, 18 Apr 2014 19:49:24 +0000 (19:49 +0000)]
[PECOFF] Ignore /ignore option.
/ignore:<number> is a linker option to disable warning specified by
the number. We ignore the option because it does not make sense for
LLD.
llvm-svn: 206636
Benjamin Kramer [Fri, 18 Apr 2014 19:48:03 +0000 (19:48 +0000)]
Remove a couple of redundant copies of SmallVector::operator==.
No functionality change.
llvm-svn: 206635
Adam Nemet [Fri, 18 Apr 2014 19:44:16 +0000 (19:44 +0000)]
[X86] Improve buildFromShuffleMostly for AVX
For a 256-bit BUILD_VECTOR consisting mostly of shuffles of 256-bit vectors,
both the BUILD_VECTOR and its operands may need to be legalized in multiple
steps. Consider:
(v8f32 (BUILD_VECTOR (extract_vector_elt (v8f32 %vreg0,) Constant<1>),
(extract_vector_elt %vreg0, Constant<2>),
(extract_vector_elt %vreg0, Constant<3>),
(extract_vector_elt %vreg0, Constant<4>),
(extract_vector_elt %vreg0, Constant<5>),
(extract_vector_elt %vreg0, Constant<6>),
(extract_vector_elt %vreg0, Constant<7>),
%vreg1))
a. We can't build a 256-bit vector efficiently so, we need to split it into
two 128-bit vecs and combine them with VINSERTX128.
b. Operands like (extract_vector_elt (v8f32 %vreg0), Constant<7>) needs to be
split into a VEXTRACTX128 and a further extract_vector_elt from the
resulting 128-bit vector.
c. The extract_vector_elt from b. is lowered into a shuffle to the first
element and a movss.
Depending on the order in which we legalize the BUILD_VECTOR and its
operands[1], buildFromShuffleMostly may be faced with:
(v4f32 (BUILD_VECTOR (extract_vector_elt
(vector_shuffle<1,u,u,u> (extract_subvector %vreg0, Constant<4>), undef),
Constant<0>),
(extract_vector_elt
(vector_shuffle<2,u,u,u> (extract_subvector %vreg0, Constant<4>), undef),
Constant<0>),
(extract_vector_elt
(vector_shuffle<3,u,u,u> (extract_subvector %vreg0, Constant<4>), undef),
Constant<0>),
%vreg1))
In order to figure out the underlying vector and their identity we need to see
through the shuffles.
[1] Note that the order in which operations and their operands are legalized is
only guaranteed in the first iteration of LegalizeDAG.
Fixes <rdar://problem/
16296956>
llvm-svn: 206634
Rui Ueyama [Fri, 18 Apr 2014 19:43:07 +0000 (19:43 +0000)]
[PECOFF] Support /manifestuac:NO.
If the value for /manifestuac is "NO", LLD will create a manifest XM
file but won't emit <trustinfo> element.
llvm-svn: 206633
Benjamin Kramer [Fri, 18 Apr 2014 19:01:53 +0000 (19:01 +0000)]
DebugInfo: Remove some initializer lists to make MSVC happy again.
llvm-svn: 206632
David Blaikie [Fri, 18 Apr 2014 18:24:25 +0000 (18:24 +0000)]
Add range access to MCAssembler's symbol collection.
llvm-svn: 206631
Reid Kleckner [Fri, 18 Apr 2014 18:19:18 +0000 (18:19 +0000)]
Update comment in LLVMBitCodes.h to reflect the actual bitcode record
llvm-svn: 206630
Matt Arsenault [Fri, 18 Apr 2014 18:08:31 +0000 (18:08 +0000)]
Fix uint -> size_t conversion warning.
This warning is disabled for the LLVM build,
but external users of the header can still
run into this.
Patch by Ke Bai
llvm-svn: 206629
Duncan P. N. Exon Smith [Fri, 18 Apr 2014 17:56:08 +0000 (17:56 +0000)]
Revert "blockfreq: Rewrite BlockFrequencyInfoImpl" (#2)
This reverts commit r206622 and the MSVC fixup in r206626.
Apparently the remotely failing tests are still failing, despite my
attempt to fix the nondeterminism in r206621.
llvm-svn: 206628
Greg Fitzgerald [Fri, 18 Apr 2014 17:39:50 +0000 (17:39 +0000)]
Fixed llvm-build when no targets are enabled
llvm-svn: 206627
Duncan P. N. Exon Smith [Fri, 18 Apr 2014 17:38:01 +0000 (17:38 +0000)]
Fixing MSVC after r206622?
llvm-svn: 206626
Andrew Trick [Fri, 18 Apr 2014 17:35:08 +0000 (17:35 +0000)]
Better comments to explain buffered/unbuffered processor resources.
llvm-svn: 206625
Alexey Samsonov [Fri, 18 Apr 2014 17:25:46 +0000 (17:25 +0000)]
[DWARF parser] Refactor fetching DIE address ranges.
Add a helper method to get address ranges specified in a DIE
(either by DW_AT_low_pc/DW_AT_high_pc, or by DW_AT_ranges). Use it
to untangle and simplify the code.
No functionality change.
llvm-svn: 206624
Marshall Clow [Fri, 18 Apr 2014 17:23:36 +0000 (17:23 +0000)]
Bug #19473. If you pass an allocator to std::function, we should use that allocator, not construct one from scratch. Add a test to make sure
llvm-svn: 206623
Duncan P. N. Exon Smith [Fri, 18 Apr 2014 17:22:25 +0000 (17:22 +0000)]
Reapply "blockfreq: Rewrite BlockFrequencyInfoImpl"
This reverts commit r206556, effectively reapplying commit r206548 and
its fixups in r206549 and r206550.
In an intervening commit I've added target triples to the tests that
were failing remotely [1] (but passing locally). I'm hoping the mystery
is solved? I'll revert this again if the tests are still failing
remotely.
[1]: http://bb.pgr.jp/builders/ninja-x64-msvc-RA-centos6/builds/1816
llvm-svn: 206622
Duncan P. N. Exon Smith [Fri, 18 Apr 2014 17:22:19 +0000 (17:22 +0000)]
Add some target triples for better determinism
These tests were failing on some buildbots after r206548 (reverted in
r206556), but passing locally.
They were missing target triples, so maybe that's the problem?
llvm-svn: 206621
Todd Fiala [Fri, 18 Apr 2014 17:07:09 +0000 (17:07 +0000)]
Marked TestAttachResume.process_attach_continue_interrupt_detach() as XFAIL on Linux per pr19478.
llvm-svn: 206620
Todd Fiala [Fri, 18 Apr 2014 17:04:21 +0000 (17:04 +0000)]
Fix TestPtrRef2Typedef with new const adornment on expected ref type.
llvm-svn: 206619
Todd Fiala [Fri, 18 Apr 2014 17:01:01 +0000 (17:01 +0000)]
Address hung tests in Linux.
Follow-up patch coming to address test failures exposed by this change.
llvm-svn: 206618
Benjamin Kramer [Fri, 18 Apr 2014 16:57:01 +0000 (16:57 +0000)]
LineIterator: Add DataTypes.h for int64_t on MSVC.
llvm-svn: 206617
Benjamin Kramer [Fri, 18 Apr 2014 16:46:29 +0000 (16:46 +0000)]
Add some missing includes for various standard library implementations.
llvm-svn: 206616
Benjamin Kramer [Fri, 18 Apr 2014 16:36:15 +0000 (16:36 +0000)]
Make the copy member of StringRef/ArrayRef generic wrt allocators.
Doesn't make sense to restrict this to BumpPtrAllocator. While there
replace an explicit loop with std::equal. Some standard libraries know
how to compile this down to a ::memcmp call if possible.
llvm-svn: 206615
Timur Iskhodzhanov [Fri, 18 Apr 2014 15:10:05 +0000 (15:10 +0000)]
Split out the no-thunk multiple inheritance tests
Also, intentionally duplicate base class definitions per test, so it's easier to copy tests while debugging failures
llvm-svn: 206614
Tim Northover [Fri, 18 Apr 2014 14:54:53 +0000 (14:54 +0000)]
AArch64/ARM64: add more NEON tests.
Mostly no testing this time, since they were just wrangling
target-specific intrinsics.
llvm-svn: 206613
Benjamin Kramer [Fri, 18 Apr 2014 14:54:51 +0000 (14:54 +0000)]
Allocator: Remove ReferenceAdder hack.
This was a workaround for compilers that had issues with reference
collapsing.
llvm-svn: 206612
Tim Northover [Fri, 18 Apr 2014 14:54:46 +0000 (14:54 +0000)]
ARM64: disable generation of .loh directives outside MachO.
Part of PR19455.
llvm-svn: 206611
Tim Northover [Fri, 18 Apr 2014 14:54:41 +0000 (14:54 +0000)]
ARM64: don't emit .subsections_via_symbols on ELF.
Part of PR19455.
llvm-svn: 206610
Tim Northover [Fri, 18 Apr 2014 14:54:35 +0000 (14:54 +0000)]
ARM64: add extra NEG pattern.
llvm-svn: 206609
Dmitri Gribenko [Fri, 18 Apr 2014 14:36:51 +0000 (14:36 +0000)]
Add more constness to module-related APIs
llvm-svn: 206595
Tim Northover [Fri, 18 Apr 2014 13:46:08 +0000 (13:46 +0000)]
ARM64: make sure the caller is expected to extend in AAPCS.
This is one of those DarwinPCS differences. It'd been caught in
arguments, but not return values.
llvm-svn: 206594
Evgeniy Stepanov [Fri, 18 Apr 2014 13:24:03 +0000 (13:24 +0000)]
[asan] Reenable tests that should pass since PR19207 is fixed.
llvm-svn: 206593
Tim Northover [Fri, 18 Apr 2014 13:16:55 +0000 (13:16 +0000)]
AArch64/ARM64: port more AArch64 tests to ARM64.
llvm-svn: 206592
Tim Northover [Fri, 18 Apr 2014 13:16:42 +0000 (13:16 +0000)]
AArch64/ARM64: add non-scalar lowering for more FCVT operations.
llvm-svn: 206591
Aaron Ballman [Fri, 18 Apr 2014 13:13:15 +0000 (13:13 +0000)]
Updating to use more range-based for loops, nullptr and auto. No functional changes.
llvm-svn: 206590
Evgeniy Stepanov [Fri, 18 Apr 2014 13:03:54 +0000 (13:03 +0000)]
[msan] Add missing quotes.
llvm-svn: 206589
Tim Northover [Fri, 18 Apr 2014 12:50:58 +0000 (12:50 +0000)]
AArch64/ARM64: improve spotting of EXT instructions from VECTOR_SHUFFLE.
We couldn't cope if the first mask element was UNDEF before, which
isn't ideal.
llvm-svn: 206588
Evgeniy Stepanov [Fri, 18 Apr 2014 12:19:28 +0000 (12:19 +0000)]
[msan] Run msan_test in the new with-calls mode.
llvm-svn: 206587
Evgeniy Stepanov [Fri, 18 Apr 2014 12:18:00 +0000 (12:18 +0000)]
[msan] Missing declarations for the new interface functions.
llvm-svn: 206586
Evgeniy Stepanov [Fri, 18 Apr 2014 12:17:20 +0000 (12:17 +0000)]
[msan] Add -msan-instrumentation-with-call-threshold.
This flag replaces inline instrumentation for checks and origin stores with
calls into MSan runtime library. This is a workaround for PR17409.
Disabled by default.
llvm-svn: 206585
Evgeniy Stepanov [Fri, 18 Apr 2014 12:15:24 +0000 (12:15 +0000)]
[msan] Add new MSan callbacks for instrumentation-with-calls mode.
llvm-svn: 206584
Chandler Carruth [Fri, 18 Apr 2014 11:02:33 +0000 (11:02 +0000)]
[LCG] Remove all of the complexity stemming from supporting copying.
Reality is that we're never going to copy one of these. Supporting this
was becoming a nightmare because nothing even causes it to compile most
of the time. Lots of subtle errors built up that wouldn't have been
caught by any "normal" testing.
Also, make the move assignment actually work rather than the bogus swap
implementation that would just infloop if used. As part of that, factor
out the graph pointer updates into a helper to share between move
construction and move assignment.
llvm-svn: 206583
Chandler Carruth [Fri, 18 Apr 2014 11:02:29 +0000 (11:02 +0000)]
[Allocator] Fix an obvious think-o with the move assignment
implementation of the SpecificBumpPtrAllocator -- we have to actually
move the subobject. =] Noticed when using this code more directly.
llvm-svn: 206582
Chandler Carruth [Fri, 18 Apr 2014 10:50:32 +0000 (10:50 +0000)]
[LCG] Add support for building persistent and connected SCCs to the
LazyCallGraph. This is the start of the whole point of this different
abstraction, but it is just the initial bits. Here is a run-down of
what's going on here. I'm planning to incorporate some (or all) of this
into comments going forward, hopefully with better editing and wording.
=]
The crux of the problem with the traditional way of building SCCs is
that they are ephemeral. The new pass manager however really needs the
ability to associate analysis passes and results of analysis passes with
SCCs in order to expose these analysis passes to the SCC passes. Making
this work is kind-of the whole point of the new pass manager. =]
So, when we're building SCCs for the call graph, we actually want to
build persistent nodes that stick around and can be reasoned about
later. We'd also like the ability to walk the SCC graph in more complex
ways than just the traditional postorder traversal of the current CGSCC
walk. That means that in addition to being persistent, the SCCs need to
be connected into a useful graph structure.
However, we still want the SCCs to be formed lazily where possible.
These constraints are quite hard to satisfy with the SCC iterator. Also,
using that would bypass our ability to actually add data to the nodes of
the call graph to facilite implementing the Tarjan walk. So I've
re-implemented things in a more direct and embedded way. This
immediately makes it easy to get the persistence and connectivity
correct, and it also allows leveraging the existing nodes to simplify
the algorithm. I've worked somewhat to make this implementation more
closely follow the traditional paper's nomenclature and strategy,
although it is still a bit obtuse because it isn't recursive, using
an explicit stack and a tail call instead, and it is interruptable,
resuming each time we need another SCC.
The other tricky bit here, and what actually took almost all the time
and trials and errors I spent building this, is exactly *what* graph
structure to build for the SCCs. The naive thing to build is the call
graph in its newly acyclic form. I wrote about 4 versions of this which
did precisely this. Inevitably, when I experimented with them across
various use cases, they became incredibly awkward. It was all
implementable, but it felt like a complete wrong fit. Square peg, round
hole. There were two overriding aspects that pushed me in a different
direction:
1) We want to discover the SCC graph in a postorder fashion. That means
the root node will be the *last* node we find. Using the call-SCC DAG
as the graph structure of the SCCs results in an orphaned graph until
we discover a root.
2) We will eventually want to walk the SCC graph in parallel, exploring
distinct sub-graphs independently, and synchronizing at merge points.
This again is not helped by the call-SCC DAG structure.
The structure which, quite surprisingly, ended up being completely
natural to use is the *inverse* of the call-SCC DAG. We add the leaf
SCCs to the graph as "roots", and have edges to the caller SCCs. Once
I switched to building this structure, everything just fell into place
elegantly.
Aside from general cleanups (there are FIXMEs and too few comments
overall) that are still needed, the other missing piece of this is
support for iterating across levels of the SCC graph. These will become
useful for implementing #2, but they aren't an immediate priority.
Once SCCs are in good shape, I'll be working on adding mutation support
for incremental updates and adding the pass manager that this analysis
enables.
llvm-svn: 206581
Tim Northover [Fri, 18 Apr 2014 10:47:44 +0000 (10:47 +0000)]
ARM64: make sure HFAs on the stack get properly aligned.
Another AAPCS bug, part of PR19432.
llvm-svn: 206580
Benjamin Kramer [Fri, 18 Apr 2014 10:45:33 +0000 (10:45 +0000)]
X86: Pattern match scalar loads + vcvtph2ps into just vcvtph2ps.
vcvtph2ps only reads the lower 64 bits of the address passed to the
intrinsic.
llvm-svn: 206579
Dmitry Vyukov [Fri, 18 Apr 2014 10:37:39 +0000 (10:37 +0000)]
tsan: add benchmark that allows to investigate shadow memory consumption
llvm-svn: 206578
Tobias Grosser [Fri, 18 Apr 2014 09:46:35 +0000 (09:46 +0000)]
Really fix the load case.
Commit r206510 falsely advertised to fix the load cases, even though it only
fixed the store case. This commit adds the same fix for the load case including
the missing test coverage.
llvm-svn: 206577
Chandler Carruth [Fri, 18 Apr 2014 09:35:51 +0000 (09:35 +0000)]
Revert r206565 (and r206566 which updated tests).
This commit was attributed to a different person from the person who
posted the patch to the list, and the person who posted it the list
claimed when they did that they were not the author, but that the author
was yet a third person. I don't know what is going on here, but
reverting until the attribution is clear and the author has explicitly
contributed the patch.
Also, the review hasn't really involved any of the MC maintainers and
that seems questionable too.
llvm-svn: 206576
Tim Northover [Fri, 18 Apr 2014 09:31:31 +0000 (09:31 +0000)]
AArch64/ARM64: port atomics test to ARM64.
Covers quite a few extra instructions (like any of the max/min ones
which were broken until recently on ARM64).
llvm-svn: 206575
Tim Northover [Fri, 18 Apr 2014 09:31:27 +0000 (09:31 +0000)]
AArch64/ARM64: spot a greater variety of concat_vector operations.
Code mostly copied from AArch64, just tidied up a trifle and plumbed
into the ARM64 way of doing things.
This also enables the AArch64 tests which inspired the previous
untested commits.
llvm-svn: 206574
Tim Northover [Fri, 18 Apr 2014 09:31:20 +0000 (09:31 +0000)]
ARM64: implement cunning optimisation from AArch64
A vector extract followed by a dup can become a single instruction even if the
types don't match. AArch64 handled this in ISelLowering, but a few reasonably
simple patterns can take care of it in TableGen, so that's where I've put it.
llvm-svn: 206573
Tim Northover [Fri, 18 Apr 2014 09:31:15 +0000 (09:31 +0000)]
ARM64: spot a vector_shuffle that maps to INS and expand.
Tests will be coming very shortly when all the optimisations needed to
support AArch64's neon-copy.ll file are committed.
llvm-svn: 206572
Tim Northover [Fri, 18 Apr 2014 09:31:11 +0000 (09:31 +0000)]
ARM64: nick some AArch64 patterns for extract/insert -> INS.
Tests will be committed shortly when all optimisations needed to
support AArch64's neon-copy.ll file are supported.
llvm-svn: 206571
Tim Northover [Fri, 18 Apr 2014 09:31:07 +0000 (09:31 +0000)]
AArch64/ARM64: emit all vector FP comparisons as such.
ARM64 was scalarizing some vector comparisons which don't quite map to
AArch64's compare and mask instructions. AArch64's approach of sacrificing a
little efficiency to emulate them with the limited set available was better, so
I ported it across.
More "inspired by" than copy/paste since the backend's internal expectations
were a bit different, but the tests were invaluable.
llvm-svn: 206570
Tim Northover [Fri, 18 Apr 2014 09:31:01 +0000 (09:31 +0000)]
AArch64/ARM64: port BSL logic from AArch64 & enable test.
I enhanced it a little in the process. The decision shouldn't really be beased
on whether a BUILD_VECTOR is a splat: any set of constants will do the job
provided they're related in the correct way.
Also, the BUILD_VECTOR could be any operand of the incoming AND nodes, so it's
best to check for all 4 possibilities rather than assuming it'll be the RHS.
llvm-svn: 206569
Tim Northover [Fri, 18 Apr 2014 09:30:52 +0000 (09:30 +0000)]
AArch64/ARM64: copy byval implementation from AArch64.
It's not actually used to handle C or C++ ABI rules on ARM64, but could well be
emitted by other language front-ends, so it's as well to have a sensible
implementation.
llvm-svn: 206568
Jiangning Liu [Fri, 18 Apr 2014 09:05:50 +0000 (09:05 +0000)]
Add missing config file for newly added test case introduced by r206563.
llvm-svn: 206567
Yaron Keren [Fri, 18 Apr 2014 08:50:09 +0000 (08:50 +0000)]
Updated test with register names following r206565.
llvm-svn: 206566
Yaron Keren [Fri, 18 Apr 2014 08:03:38 +0000 (08:03 +0000)]
Patch by Ray Donnelly.
Emit WIN64 SEH registers by name instead of just number.
llvm-svn: 206565
Kostya Serebryany [Fri, 18 Apr 2014 08:02:42 +0000 (08:02 +0000)]
[asan] one more workaround for PR17409: don't do BB-level coverage instrumentation if there are more than N (=1500) basic blocks. This makes ASanCoverage work on libjpeg_turbo/jchuff.c used by Chrome, which has 1824 BBs
llvm-svn: 206564
Jiangning Liu [Fri, 18 Apr 2014 07:57:54 +0000 (07:57 +0000)]
This commit allows vectorized loops to be unrolled by a factor of 2 for AArch64.
A new test case is also added for ARM64.
Patched by Z.Zheng
llvm-svn: 206563
Matt Arsenault [Fri, 18 Apr 2014 07:40:20 +0000 (07:40 +0000)]
R600: Minor cleanups.
Fix indentation, better line wrapping, unused includes.
llvm-svn: 206562
Lang Hames [Fri, 18 Apr 2014 06:48:23 +0000 (06:48 +0000)]
[ExecutionEngine] Allow JIT clients to enable/disable module verification.
Previously module verification was always enabled, with no way to turn it off.
As of this commit, module verification is on by default in Debug builds, and off
by default in release builds. The default behaviour can be overridden by calling
setVerifyModules(bool) on the JIT instance (this works for both the old JIT, and
MCJIT).
<rdar://problem/
16150008>
llvm-svn: 206561
Rui Ueyama [Fri, 18 Apr 2014 06:01:43 +0000 (06:01 +0000)]
[ELF] Fix GNU_RELRO section name.
llvm-svn: 206560
Jiangning Liu [Fri, 18 Apr 2014 05:58:09 +0000 (05:58 +0000)]
This is one of the optimizations ported from ARM64 to AArch64 to address the performance gap between these two back ends. The test case newly added for AArch64 already exists in ARM64.
Patched by Z.Zheng
llvm-svn: 206559
Matt Arsenault [Fri, 18 Apr 2014 05:19:26 +0000 (05:19 +0000)]
R600/SI: Try to use scalar BFE.
Use scalar BFE with constant shift and offset when possible.
This is complicated by the fact that the scalar version packs
the two operands of the vector version into one.
llvm-svn: 206558
Jiangning Liu [Fri, 18 Apr 2014 03:58:38 +0000 (03:58 +0000)]
This commit enables unaligned memory accesses of vector types on AArch64 back end. This should boost vectorized code performance.
Patched by Z. Zheng
llvm-svn: 206557
Duncan P. N. Exon Smith [Fri, 18 Apr 2014 02:17:43 +0000 (02:17 +0000)]
Revert "blockfreq: Rewrite BlockFrequencyInfoImpl"
This reverts commits r206548, r206549 and r206549.
There are some unit tests failing that aren't failing locally [1], so
reverting until I have time to investigate.
[1]: http://bb.pgr.jp/builders/ninja-x64-msvc-RA-centos6/builds/1816
llvm-svn: 206556
Justin Bogner [Fri, 18 Apr 2014 02:10:26 +0000 (02:10 +0000)]
OnDiskHashTable: Provide iterator_range for keys and data
llvm-svn: 206555
Duncan P. N. Exon Smith [Fri, 18 Apr 2014 02:10:09 +0000 (02:10 +0000)]
blockfreq: Really fix r206548 (and r206549)
Turns out this code is dead.
llvm-svn: 206554
Jim Grosbach [Fri, 18 Apr 2014 02:09:07 +0000 (02:09 +0000)]
c++11: Tidy up tblgen w/ range loops.
IntrInfoEmitter cleanup.
llvm-svn: 206553
Jim Grosbach [Fri, 18 Apr 2014 02:09:04 +0000 (02:09 +0000)]
iterator access to scheduling classes
llvm-svn: 206552
Jim Grosbach [Fri, 18 Apr 2014 02:09:02 +0000 (02:09 +0000)]
iterator_range accessor for CodeGenTarget instruction list.
llvm-svn: 206551
Jim Grosbach [Fri, 18 Apr 2014 02:08:58 +0000 (02:08 +0000)]
iterator based accessors for CodeGenInstruction operand list.
llvm-svn: 206550
Duncan P. N. Exon Smith [Fri, 18 Apr 2014 02:06:24 +0000 (02:06 +0000)]
blockfreq: Fixing MSVC after r206548?
llvm-svn: 206549
Duncan P. N. Exon Smith [Fri, 18 Apr 2014 01:57:45 +0000 (01:57 +0000)]
blockfreq: Rewrite BlockFrequencyInfoImpl
Rewrite the shared implementation of BlockFrequencyInfo and
MachineBlockFrequencyInfo entirely.
The old implementation had a fundamental flaw: precision losses from
nested loops (or very wide branches) compounded past loop exits (and
convergence points).
The @nested_loops testcase at the end of
test/Analysis/BlockFrequencyAnalysis/basic.ll is motivating. This
function has three nested loops, with branch weights in the loop headers
of 1:4000 (exit:continue). The old analysis gives non-sensical results:
Printing analysis 'Block Frequency Analysis' for function 'nested_loops':
---- Block Freqs ----
entry = 1.0
for.cond1.preheader = 1.00103
for.cond4.preheader = 5.5222
for.body6 = 18095.19995
for.inc8 = 4.52264
for.inc11 = 0.00109
for.end13 = 0.0
The new analysis gives correct results:
Printing analysis 'Block Frequency Analysis' for function 'nested_loops':
block-frequency-info: nested_loops
- entry: float = 1.0, int = 8
- for.cond1.preheader: float = 4001.0, int = 32007
- for.cond4.preheader: float =
16008001.0, int =
128064007
- for.body6: float =
64048012001.0, int =
512384096007
- for.inc8: float =
16008001.0, int =
128064007
- for.inc11: float = 4001.0, int = 32007
- for.end13: float = 1.0, int = 8
Most importantly, the frequency leaving each loop matches the frequency
entering it.
The new algorithm leverages BlockMass and PositiveFloat to maintain
precision, separates "probability mass distribution" from "loop
scaling", and uses dithering to eliminate probability mass loss. I have
unit tests for these types out of tree, but it was decided in the review
to make the classes private to BlockFrequencyInfoImpl, and try to shrink
them (or remove them entirely) in follow-up commits.
The new algorithm should generally have a complexity advantage over the
old. The previous algorithm was quadratic in the worst case. The new
algorithm is still worst-case quadratic in the presence of irreducible
control flow, but it's linear without it.
The key difference between the old algorithm and the new is that control
flow within a loop is evaluated separately from control flow outside,
limiting propagation of precision problems and allowing loop scale to be
calculated independently of mass distribution. Loops are visited
bottom-up, their loop scales are calculated, and they are replaced by
pseudo-nodes. Mass is then distributed through the function, which is
now a DAG. Finally, loops are revisited top-down to multiply through
the loop scales and the masses distributed to pseudo nodes.
There are some remaining flaws.
- Irreducible control flow isn't modelled correctly. LoopInfo and
MachineLoopInfo ignore irreducible edges, so this algorithm will
fail to scale accordingly. There's a note in the class
documentation about how to get closer. See also the comments in
test/Analysis/BlockFrequencyInfo/irreducible.ll.
- Loop scale is limited to 4096 per loop (2^12) to avoid exhausting
the 64-bit integer precision used downstream.
- The "bias" calculation proposed on llvmdev is *not* incorporated
here. This will be added in a follow-up commit, once comments from
this review have been handled.
llvm-svn: 206548
Matt Arsenault [Fri, 18 Apr 2014 01:53:18 +0000 (01:53 +0000)]
R600/SI: Match sign_extend_inreg to s_sext_i32_i8 and s_sext_i32_i16
llvm-svn: 206547
Reid Kleckner [Fri, 18 Apr 2014 01:21:55 +0000 (01:21 +0000)]
Fix a -Wmicrosoft warning on an unrepresentable enum
0x80000000 isn't representable as an int, which is the default enum
type.
llvm-svn: 206545
Paul Robinson [Fri, 18 Apr 2014 01:20:08 +0000 (01:20 +0000)]
Fix example for VS2012.
llvm-svn: 206544
Duncan P. N. Exon Smith [Fri, 18 Apr 2014 01:05:25 +0000 (01:05 +0000)]
BackendUtil: Pass through -mdisable-tail-calls
The frontend option -fno-optimize-sibling-calls resolves to -cc1's
-mdisable-tail-calls, which is passed to the TargetMachine in the
backend. PassManagerBuilder was adding the -tailcallelim pass anyway.
Use a new DisableTailCalls option in PassManagerBuilder to disable tail
calls harder.
Requires the matching commit in LLVM that adds DisableTailCalls.
<rdar://problem/
16050591>
llvm-svn: 206543
Duncan P. N. Exon Smith [Fri, 18 Apr 2014 01:05:15 +0000 (01:05 +0000)]
PMBuilder: Expose an option to disable tail calls
Adds API to allow frontends to disable tail calls in PassManagerBuilder.
<rdar://problem/
16050591>
llvm-svn: 206542
Tom Stellard [Fri, 18 Apr 2014 00:36:21 +0000 (00:36 +0000)]
R600/SI: Use SReg_64 instead of VSrc_64 when selecting BUILD_PAIR
llvm-svn: 206541
Jim Grosbach [Thu, 17 Apr 2014 23:41:57 +0000 (23:41 +0000)]
[ARM64,C++11] Range'ify another loop.
llvm-svn: 206539
Rui Ueyama [Thu, 17 Apr 2014 23:38:01 +0000 (23:38 +0000)]
[ELF] Fix typo that caused a test to fail on FreeBSD.
llvm-svn: 206538
Jason Molenda [Thu, 17 Apr 2014 23:29:19 +0000 (23:29 +0000)]
Add FreeBSDSignals.cpp to project file.
llvm-svn: 206524
Tobias Grosser [Thu, 17 Apr 2014 23:13:49 +0000 (23:13 +0000)]
Ensure a scalar pointer when issuing a vector load
Even tough we may want to generate a vector load, the address from which to load
still is a scalar. Make sure even if previous address computations may have been
vectorized, that the addresses are also available as scalars.
This fixes http://llvm.org/PR19469
Reported-by: Jeremy Huddleston Sequoia <jeremyhu@apple.com>
llvm-svn: 206510
Justin Bogner [Thu, 17 Apr 2014 22:49:06 +0000 (22:49 +0000)]
test: Use llvm-profdata merge in Profile tests
In preparation for using a binary format for instrumentation based
profiling, explicitly treat the test inputs as text and transform them
before running. This will allow us to leave the checked in files in
human readable format once the instrumentation format is binary.
No functional change.
llvm-svn: 206509
Reid Kleckner [Thu, 17 Apr 2014 22:47:52 +0000 (22:47 +0000)]
MS ABI: Don't append to vbtables that we shouldn't extend
This was probably a benign bug, since nobody would look at the vbtable
slots that we were filling in.
llvm-svn: 206508
Diego Novillo [Thu, 17 Apr 2014 22:33:50 +0000 (22:33 +0000)]
Fix bug 19437 - Only add discriminators for DWARF 4 and above.
Summary:
This prevents the discriminator generation pass from triggering if
the DWARF version being used in the module is prior to 4.
Reviewers: echristo, dblaikie
CC: llvm-commits
Differential Revision: http://reviews.llvm.org/D3413
llvm-svn: 206507
Nuno Lopes [Thu, 17 Apr 2014 22:26:44 +0000 (22:26 +0000)]
remove some dead code
lib/Analysis/IPA/InlineCost.cpp | 18 ------------------
lib/Analysis/RegionPass.cpp | 1 -
lib/Analysis/TypeBasedAliasAnalysis.cpp | 1 -
lib/Transforms/Scalar/LoopUnswitch.cpp | 21 ---------------------
lib/Transforms/Utils/LCSSA.cpp | 2 --
lib/Transforms/Utils/LoopSimplify.cpp | 6 ------
utils/TableGen/AsmWriterEmitter.cpp | 13 -------------
utils/TableGen/DFAPacketizerEmitter.cpp | 7 -------
utils/TableGen/IntrinsicEmitter.cpp | 2 --
9 files changed, 71 deletions(-)
llvm-svn: 206506
Reed Kotler [Thu, 17 Apr 2014 22:15:34 +0000 (22:15 +0000)]
Start pushing changes for Mips Fast-Isel
llvm-svn: 206505
Timur Iskhodzhanov [Thu, 17 Apr 2014 22:01:48 +0000 (22:01 +0000)]
Follow-up to r206457 -- fix static adjustments for some subtle virtual inheritance cases
Reviewed at http://reviews.llvm.org/D3410
llvm-svn: 206504
Aaron Ballman [Thu, 17 Apr 2014 21:44:08 +0000 (21:44 +0000)]
Making some public members into private members. This also introduces a bit more const-correctness, and now uses some range-based for loops. No functional changes intended.
llvm-svn: 206503
Louis Gerbarg [Thu, 17 Apr 2014 21:32:41 +0000 (21:32 +0000)]
Make test/CodeGen/ARM64/vector-insertion.ll explicitly select neon syntax
Change the command line vector-insertion.ll to explicitly set the neon syntax
to apple so that buildbots that default to other syntaxes won't fail.
llvm-svn: 206502
Tom Stellard [Thu, 17 Apr 2014 21:00:13 +0000 (21:00 +0000)]
R600: Add comment clariying use of sext for result of MUL_U24
llvm-svn: 206501