The purpose of this patch is to eliminate a pass ordering dependence between LoopPredication and LICM. To understand the purpose, consider the following snippet of code inside some loop 'L' with IV 'i'
A = _a.length;
guard (i < A)
a = _a[i]
B = _b.length;
guard (i < B);
b = _b[i];
...
Z = _z.length;
guard (i < Z)
z = _z[i]
accum += a + b + ... + z;
Today, we need LICM to hoist the length loads, LoopPredication to make the guards loop invariant, and TrivialUnswitch to eliminate the loop invariant guard to establish must execute for the next length load. Today, if we can't prove speculation safety, we'd have to iterate these three passes 26 times to reduce this example down to the minimal form.
Using the fact that the array lengths are known to be invariant, we can short circuit this iteration. By forming the loop invariant form of all the guards at once, we remove the need for LoopPredication from the iterative cycle. At the moment, we'd still have to iterate LICM and TrivialUnswitch; we'll leave that part for later.
As a secondary benefit, this allows LoopPred to expose peeling oppurtunities in a much more obvious manner. See the udiv test changes as an example. If the udiv was not hoistable (i.e. we couldn't prove speculation safety) this would be an example where peeling becomes obviously profitable whereas it wasn't before.
A couple of subtleties in the implementation:
- SCEV's isSafeToExpand guarantees speculation safety (i.e. let's us expand at a new point). It is not a precondition for expansion if we know the SCEV corresponds to a Value which dominates the requested expansion point.
- SCEV's isLoopInvariant returns true for expressions which compute the same value across all iterations executed, regardless of where the original Value is located. (i.e. it can be in the loop) This implies we have a speculation burden to prove before expanding them outside loops.
- invariant_loads and AA->pointsToConstantMemory are two cases that SCEV currently does not handle, but meets the SCEV definition of invariance. I plan to sink this part into SCEV once this has baked for a bit.
Differential Revision: https://reviews.llvm.org/D60093
llvm-svn: 358684
Summary:
This patch removes the "//----" frames and "///" leading lines from
docstring comments. We already have code doing transformations like this in
modify-python-lldb.py, but that's a script I'd like to remove. Instead
of running these transformations everytime we run swig, we can just
perform equivalent on its input once.
This patch can be reproduced (e.g. for downstream merges) with the
following "sweet" perl command:
perl -i -p -e 'BEGIN{ $/ = undef;} s:(" *\n) *//-----*\n:\1:gs; s:^( *)/// ?:\1:gsm; s:^ *//------*\n( *\n)?( *"):\2:gsm; s: *$::gsm; s:\n *"\):"):gsm' scripts/interface/*.i
This command produces nearly equivalent python files to those produced
by the relevant code in modify-python-lldb.py. The only difference I
noticed is that here I am slightly more agressive in removing trailing
newlines from docstring comments (the python script seems to leave
newlines in class-level docstrings).
Reviewers: amccarth, clayborg, jingham, aprantl
Subscribers: lldb-commits
Differential Revision: https://reviews.llvm.org/D60498
llvm-svn: 358683
The change landed in r358657 broke some of the buildbots because they
use an older version of Python 2 that raises this error.
```
File "/Volumes/data/dev/llvm/upstream/master/src/projects/compiler-rt/lib/asan/scripts/asan_symbolize.py", line 509
exec(f.read(), globals_space, None)
SyntaxError: unqualified exec is not allowed in function 'load_plugin_from_file' it contains a nested function with free variables
```
I can reproduce this problem when using Python 2.7.6.
To workaround this some indirection has been added to prevent the broken
(the line at fault would never be executed) SyntaxError error in old
Python versions from being raised.
rdar://problem/49476995
llvm-svn: 358682
Summary:
The immediate post dominator of the loop header may be part of the divergent loop.
Since this /was/ the divergence propagation bound the SDA would not detect joins of divergent paths outside the loop.
Reviewers: nhaehnle
Reviewed By: nhaehnle
Subscribers: mmasten, arsenm, jvesely, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D59042
llvm-svn: 358681
isSafeToExpand was making a common, but dangerously wrong, mistake in assuming that if any instruction within a basic block executes, that all instructions within that block must execute. This can be trivially shown to be false by considering the following small example:
bb:
add x, y <-- InsertionPoint
call @throws()
udiv x, y <-- SCEV* S
br ...
It's clearly not legal to expand S above the throwing call, but the previous logic would do so since S dominates (but not properlyDominates) the block containing the InsertionPoint.
Since iterating instructions w/in a block is expensive, this change special cases two cases: 1) S is an operand of InsertionPoint, and 2) InsertionPoint is the terminator of it's block. These two together are enough to keep all current optimizations triggering while fixing the latent correctness issue.
As best I can tell, this is a silent bug in current ToT. Given that, there's no tests with this change. It was noticed in an upcoming optimization change (D60093), and was reviewed as part of that. That change will include the test which caused me to notice the issue. I'm submitting this seperately so that anyone bisecting a problem gets a clear explanation.
llvm-svn: 358680
This used to be guarded on whether the deployment target was greater
than macosx10.8, however testing against the dylibs for 10.8 and earlier
with the function enabled works too. The revision that introduced
__pad_and_output is r164241 and it does not mention a reason for the
guard.
llvm-svn: 358677
There are barely any lines I haven't changed in these files, so I think I could
might as well leave it in an LLVM coding style conforming state. I also renamed
2 functions and moved addDependency out of line to ease on followup patches.
Differential Revision: https://reviews.llvm.org/D59457
llvm-svn: 358676
Summary: We already have the structure internally, we just need to expose it.
Reviewers: ilya-biryukov
Subscribers: ioeric, MaskRay, jkorous, arphaman, kadircet, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D60267
llvm-svn: 358675
Sort the elements of Sema::OpenCLTypeExtMap and Sema::OpenCLDeclExtMap
by TypeIDs and DeclIDs to guarantee a stable serialization order.
Differential Revision: https://reviews.llvm.org/D60835
Reviewed By: Anastasia
Reviewers: Anastasia, lebedev.ri
llvm-svn: 358674
Summary:
This patch adds support for yaml (de)serialization of the minidump
ModuleList stream. It's a fairly straight forward-application of the
existing patterns to the ModuleList structures defined in previous
patches.
One thing, which may be interesting to call out explicitly is the
addition of "new" allocation functions to the helper BlobAllocator
class. The reason for this was, that there was an emerging pattern of a
need to allocate space for entities, which do not have a suitable
lifetime for use with the existing allocation functions. A typical
example of that was the "size" of various lists, which is only available
as a temporary returned by the .size() method of some container. For
these cases, one can use the new set of allocation functions, which
will take a temporary object, and store it in an allocator-managed
buffer until it is written to disk.
Reviewers: amccarth, jhenderson, clayborg, zturner
Subscribers: lldb-commits, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D60405
llvm-svn: 358672
Summary:
Let's not try to be clever and detect it based on the libc++abi setting.
The only build that puts new/delete in libc++abi is Apple's and we set
this CMake option explicitly in both libc++ and libc++abi. Complicated
dependent options hurt, let's avoid them when possible.
Reviewers: phosek, EricWF
Subscribers: mgorny, christof, jkorous, dexonsmith, libcxx-commits
Tags: #libc
Differential Revision: https://reviews.llvm.org/D60797
llvm-svn: 358671
Summary:
Also add a test to verify clang-tidy only apply the first alternative
fix.
Reviewers: alexfh
Subscribers: xazax.hun, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D60857
llvm-svn: 358666
If accessing the MSVC installation root directly on a case sensitive
filesystem, these details matter.
Differential Revision: https://reviews.llvm.org/D60627
llvm-svn: 358662
Summary:
Currently we emit an unfriendly "clang diagnostic" message when rename fails. This
patch makes clangd to emit a detailed diagnostic message.
Reviewers: sammccall
Subscribers: ilya-biryukov, ioeric, MaskRay, jkorous, arphaman, kadircet, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D60821
llvm-svn: 358658
Summary:
This change adds a simple plugin architecture to `asan_symbolize.py`.
The motivation here is that sometimes it's necessary to perform extra
work to figure out where binaries with debug symbols can actually be
found. For example it might be the case that a remote service needs
to be queried for binaries and then copied to the local system.
This "extra work" can be extremely site-specific such that adding the
code directly into the `asan_symbolize.py` would just clutter the code
for a very niche use case. To avoid this, the `asan_symbolize.py` can
now load external code via a new `--plugins` command line option.
These plugins are loaded before main command line argument parsing so
that they can add their own command line options.
Right now the only hook into the behaviour of symbolization is the
`filter_binary_path()` function which assumes a very similar role
to the `binary_name_filter` function that was previously in the code.
We can add more hooks as necessary.
Code in the `asan_symbolize.py` script does not call plugin code
directly. Instead it uses a `AsanSymbolizerPlugInProxy` object.
This object
* Loads plugins from files.
* Manages the lifetime of the plugins.
* Provides an interface for calling into plugin functions and handles
calling into multiple plugins.
To unify the way binary paths are filtered the old `sysroot_path_filter`
function (and associated code) has been turned into a simple plugin
(`SysRootFilterPlugIn`) that is always loaded. The plugin unloads
itself if the `-s` option is not present on the command line. Users
should not see any functional change relating to this command line
option.
Some simple tests are provided to illustrate what plugin code looks
like and also to check the functionality continues to work.
rdar://problem/49476995
Reviewers: kubamracek, yln, samsonov, dvyukov, vitalybuka
Subscribers: srhines, #sanitizers, llvm-commits
Tags: #llvm, #sanitizers
Differential Revision: https://reviews.llvm.org/D60529
llvm-svn: 358657
.rela.dyn is a section that has sh_info normally
set to zero. And Info is an optional field in the description
of the relocation section in YAML.
But currently, yaml2obj would fail to produce the object when
Info is not explicitly listed.
The patch fixes the issue.
Differential revision: https://reviews.llvm.org/D60820
llvm-svn: 358656
This is https://bugs.llvm.org//show_bug.cgi?id=39857.
I added the comment with much more details to the bug page,
the short version is below.
The following script and code demonstrates the issue:
aliasto__text = __text;
SECTIONS {
.text 0x1000 : { __text = . ; *(.text) }
}
...
call aliasto__text
LLD fails with "cannot refer to absolute symbol: aliasto__text" error.
It happens because at the moment of scanning the relocations
we do not yet assign the correct/final/any section value for the symbol aliasto__text.
I made a change to Relocations.cpp to fix that.
Also, I had to remove the symbol-location.s test case completely, because now it does not
trigger any error. Since now all linker scripts symbols are resolved to constants, no
errors can be triggered at all it seems. I checked that it is consistent with the behavior
of bfd and gold (they do not trigger errors for the case from symbol-location.s), so it should
be OK. I.e. at least it is probably not the best possible, but natural behavior we obtained.
Differential revision: https://reviews.llvm.org/D55423
llvm-svn: 358652
This replaces the MOVMSK combine introduced at D52121/rL342326
(movmsk (setne (and X, (1 << C)), 0)) -> (movmsk (X << C))
with the more general icmp lowering so it can pick up more cases through bitcasts - notably vXi8 cases which use vXi16 shifts+masks, this patch can remove the mask and use pcmpgtb(0,x) for the sra.
Differential Revision: https://reviews.llvm.org/D60625
llvm-svn: 358651
Summary:
If the output section contains only symbol assignments, we copy flags
from the previous sections. Don't set SHF_ALLOC if NonAlloc is true.
We also have to change the type from SHT_NOBITS to SHT_PROGBITS.
In ld.bfd, bfd_elf_get_default_section_type maps non-alloctable sections to SHT_PROGBITS.
Non-alloctable SHT_NOBITS sections do not make sense.
Fixes PR38626
Differential Revision: https://reviews.llvm.org/D59986
llvm-svn: 358650
llvm-objcopy currently emits an error if a section to be removed is
referenced by another section. This is a reasonable thing to do, but is
different to GNU objcopy. We should allow users who know what they are
doing to have a way to produce the invalid ELF. This change adds a new
switch --allow-broken-links to both llvm-strip and llvm-objcopy to do
precisely that. The corresponding sh_link field is then set to 0 instead
of an error being emitted.
I cannot use llvm-readelf/readobj to test the link fields because they
emit an error if any sections, like the .dynsym, cannot be properly
loaded.
Reviewed by: rupprecht, grimar
Differential Revision: https://reviews.llvm.org/D60324
llvm-svn: 358649
This generalizes code and also fixes the broken behavior shown in
one of our test cases for some targets, like x86-64.
The issue occurs when the forward declarations are used in the script.
One of the samples is:
SECTIONS {
foo = ADDR(.text) - ABSOLUTE(ADDR(.text));
};
In that case, we have a broken output when output target does
not use thunks. That happens because thunks creating code
(called from maybeAddThunks)
calls Script->assignAddresses() at least one more time,
what fixups the values. As a result final symbols values can
be different on AArch64 and x86, for example.
In this patch, I generalize and rename maybeAddThunks to
finalizeAddressDependentContent and now it is used and called
by all targets.
Differential revision: https://reviews.llvm.org/D55550
llvm-svn: 358646
Summary:
We access Live and OutputOff (which may share the same memory location)
concurrently in 2 parallelForEachN loops. Separating them avoids subtle
data races like D41884/PR35788. This patch places Live and Hash
together.
2 reasons this is appealing:
1) Hash is immutable. Live is almost read-only - only written once in MarkLive.cpp where
Hash is not accessed
2) we already discard low bits of Hash to decide ShardID. It doesn't
matter much if we make 32-bit Hash to 31-bit.
For a huge internal clang -O3 executable (1.6GiB),
`Strings` in StringTableBuilder::finalizeStringTable contains at most 310253 elements.
The expected number of pair-wise collisions 2^(-31) * C(310253,2) ~= 22.41 is too small to have a negative impact on performance.
Actually, my benchmark shows there is actually a minor performance improvement.
Differential Revision: https://reviews.llvm.org/D60765
llvm-svn: 358645
Summary:
This issue from the bugzilla: https://bugs.llvm.org/show_bug.cgi?id=41177
When the two operands for BUILD_VECTOR are same, we will get assert error.
llvm::SDValue combineBVOfConsecutiveLoads(llvm::SDNode*, llvm::SelectionDAG&):
Assertion `!(InputsAreConsecutiveLoads && InputsAreReverseConsecutive) &&
"The loads cannot be both consecutive and reverse consecutive."' failed.
This error caused by the wrong ElemSIze when calling isConsecutiveLS(). We
should use `getScalarType().getStoreSize();` to get the ElemSize instread of
`getScalarSizeInBits() / 8`.
Reviewed By: jsji
Differential Revision: https://reviews.llvm.org/D60811
llvm-svn: 358644
fneg combining attempts to turn it into fadd(fneg(A), fneg(0)), but
creating the new fadd folds to just fneg(A). When A has multiple uses,
this confuses it and you get an assert. Fixed.
Differential Revision: https://reviews.llvm.org/D60633
Change-Id: I0ddc9b7286abe78edc0cd8d734fdeb05ff09821c
llvm-svn: 358640
Patch by Gabriel Smith.
The address for a section would be evaluated before the region was
switched to. Because of this, the position within the region would not
be updated. After the region is swapped to the dot would be set to the
out of date position within the region, undoing the section address
evaluation.
To fix this, the region is swapped to before the section's address is
evaluated. As part of the fallout of this, expandMemoryRegions needed
to be gated in setDot on the condition that the evaluated address is
less than the dot. This is for the case where sections are not listed
from lowest address to highest address.
Finally, a test for the case where sections are listed "out of order"
was added.
Differential Revision: https://reviews.llvm.org/D60744
llvm-svn: 358638
Reverse the checking of the domiance order so that when a self compare happens,
it returns false. This makes compare function have strict weak ordering.
llvm-svn: 358636
Philip Reames