We currently support them only in AArch64. The NEON Reference,
however, says they are 'ARMv7, ARMv8' intrinsics.
Differential Revision: https://reviews.llvm.org/D47121
llvm-svn: 333829
I had to tweak the i32 tests so we check both reg-reg and reg-mem cases.
I also added i64 load tests.
Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible.
llvm-svn: 333827
We currently support them only in AArch64. The NEON Reference,
however, says they are 'ARMv7, ARMv8' intrinsics.
Differential Revision: https://reviews.llvm.org/D47120
llvm-svn: 333825
Object FIle Representation
At codegen time this is emitted into the ELF file a pair of symbol indices and a weight. In assembly it looks like:
.cg_profile a, b, 32
.cg_profile freq, a, 11
.cg_profile freq, b, 20
When writing an ELF file these are put into a SHT_LLVM_CALL_GRAPH_PROFILE (0x6fff4c02) section as (uint32_t, uint32_t, uint64_t) tuples as (from symbol index, to symbol index, weight).
Differential Revision: https://reviews.llvm.org/D44965
llvm-svn: 333823
As noted in the review thread for rL333782, we could have
made a bug harder to hit if we were simplifying instructions
before trying other folds.
The shuffle transform in question isn't ever a simplification;
it's just a canonicalization. So I've renamed that to make that
clearer.
This is NFCI at this point, but I've regenerated the test file
to show the cosmetic value naming difference of using
instcombine's RAUW vs. the builder.
Possible follow-ups:
1. Move reassociation folds after simplifies too.
2. Refactor common code; we shouldn't have so much repetition.
llvm-svn: 333820
We currently support them only in AArch64. The NEON Reference,
however, says they are 'ARMv7, ARMv8' intrinsics.
Differential Revision: https://reviews.llvm.org/D47121
llvm-svn: 333819
As noted in the review thread for rL333782, we're lacking coverage
for this transform, so add tests for each binop opcode with constant
operand.
llvm-svn: 333818
```
// Primary fixed point types
signed short _Accum s_short_accum;
signed _Accum s_accum;
signed long _Accum s_long_accum;
unsigned short _Accum u_short_accum;
unsigned _Accum u_accum;
unsigned long _Accum u_long_accum;
// Aliased fixed point types
short _Accum short_accum;
_Accum accum;
long _Accum long_accum;
```
This diff only allows for declaration of the fixed point types. Assignment and other operations done on fixed point types according to http://www.open-std.org/jtc1/sc22/wg14/www/docs/n1169.pdf will be added in future patches. The saturated versions of these types and the equivalent `_Fract` types will also be added in future patches.
The tests included are for asserting that we can declare these types.
Differential Revision: https://reviews.llvm.org/D46084
llvm-svn: 333814
On Darwin, the binary's symbol table points to debug info in object
files -- potentially object files within a static library. Such a
library may have multiple entries with the same name, distinguished
only by timestamp.
The code was already _attempting_ to handle this case (see the code in
ObjectContainerBSDArchive::Archive::FindObject which disambiguates via
timestamp). But, unfortunately, while the timestamp was taken into
account on the _first_ lookup, the result was then cached in a map
keyed only off of the path.
Added the timestamp to the cache, and added a test case.
Differential Revision: https://reviews.llvm.org/D47660
llvm-svn: 333813
These instructions are unusual in that they operate on 4 consecutive registers so supporting them in codegen will be more difficult than normal.
Includes an assembler check to warn if the source register is not the first register of a 4 register group.
llvm-svn: 333812
Summary:
I noticed this issue because we didn't put the primary cloned loop into
the `NonChildClonedLoops` vector and so never iterated on it. Once
I fixed that, it made it clear why I had to do a really complicated and
unnecesasry dance when updating the loops to remain in canonical form --
I was unwittingly working around the fact that the primary cloned loop
wasn't in the expected list of cloned loops. Doh!
Now that we include it in this vector, we don't need to return it and we
can consolidate the update logic as we correctly have a single place
where it can be handled.
I've just added a test for the iteration order aspect as every time
I changed the update logic partially or incorrectly here, an existing
test failed and caught it so that seems well covered (which is also
evidenced by the extensive working around of this missing update).
Reviewers: asbirlea, sanjoy
Subscribers: mcrosier, hiraditya, llvm-commits
Differential Revision: https://reviews.llvm.org/D47647
llvm-svn: 333811
LLD also supports order files using the `--symbol-ordering-file` option.
As the name would suggest, the order file format is slightly different
from gold; gold's order files specify section names, whereas LLD's
specify symbol names. Assuming you have an order file in the correct
format though, we should support using it with LLD.
Switch the check to actually use LLVM's linker detection rather than
just checking for the presence of the gold executable, since we might
have a gold executable present but be using LLD (or bfd for that matter)
as our linker.
Differential Revision: https://reviews.llvm.org/D47669
llvm-svn: 333810
In r333801 I added a test for a dump method that, for reasons I don't
understand, fails on an msvc bot:
http://lab.llvm.org:8011/builders/clang-x86-windows-msvc2015/builds/12306/
I'll remove the test for now to unblock the bot and try to look into why
there's a discrepancy on this platform later.
llvm-svn: 333807
and using the latter in DIBuilder::createArtificialType and
DIBuilder::createObjectPointerType methods as well as introducing
mirroring DISubprogram::cloneWithFlags and
DIBuilder::createArtificialSubprogram methods.
The primary goal here is to add createArtificialSubprogram to support
a pass downstream while keeping the method consistent with the
existing ones and making sure we don't encourage changing already
created DI-nodes.
Reviewed By: aprantl
Differential Revision: https://reviews.llvm.org/D47615
llvm-svn: 333806
Previously we just returned undef, but really we should be returning the pass thru input. We also need to make sure we preserve the chain output that the original intrinsic node had to maintain connectivity in the DAG. So we should just return the incoming chain as the output chain.
llvm-svn: 333804
This re-lands r333797 with a fix for big endian systems.
Original commit message:
This isn't encountered anywhere inside LLVM, so I wrote a test case to expose the issue and verify that it is fixed.
The basic problem is that the macho_load_command union contains all load comamnd structs. Load command structs in 32-bit macho files can be 32-bit aligned instead of 64-bit aligned.
There are some strange circumstances in which this can be exposed in a 64-bit macho if the load commands are invalid or if a 32-bit aligned load command is used. In the past we've worked around this type of problem with changes like r264232.
llvm-svn: 333803
The idea behind WindowsSupport.h is that it's in the source directory so
that windows.h'isms don't leak out into the larger LLVM project. To that
end, any symbol that references a symbol from windows.h must be in this
private header, and not in a public header.
However, we had some useful utility functions in WindowsSupport.h which
have no dependency on the Windows API, but still only make sense on
Windows. Those functions should be usable outside of Support since there
is no risk of causing a windows.h leak. Although this introduces some
preprocessor logic in some header files, It's not too egregious and it's
better than the alternative of duplicating a ton of code.
Differential Revision: https://reviews.llvm.org/D47662
llvm-svn: 333798
This isn't encountered anywhere inside LLVM, so I wrote a test case to expose the issue and verify that it is fixed.
The basic problem is that the macho_load_command union contains all load comamnd structs. Load command structs in 32-bit macho files can be 32-bit aligned instead of 64-bit aligned.
There are some strange circumstances in which this can be exposed in a 64-bit macho if the load commands are invalid or if a 32-bit aligned load command is used. In the past we've worked around this type of problem with changes like r264232.
llvm-svn: 333797
Summary:
This allows to build and link the code with e.g.
-fsanitize=dataflow -fsanitize-coverage=trace-pc-guard,pc-table,func,trace-cmp
w/o providing (all) the definitions of trace-cmp hooks.
This is similar to dummy hooks provided by asan/ubsan/msan for the same purpose,
except that some of the hooks need to have the __dfsw_ prefix
since we need dfsan to replace them.
Reviewers: pcc
Reviewed By: pcc
Subscribers: delcypher, #sanitizers, llvm-commits
Differential Revision: https://reviews.llvm.org/D47605
llvm-svn: 333796
The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking.
This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly.
llvm-svn: 333795
Clang passes --plugin /path/to/LLVMgold.so to the linker when -flto is
passed. After r333607 we only ignore --plugin as a joined argument,
which means that the following argument (/path/to/LLVMgold.so) is
interpreted as an input file. This means that either every LTO'd
program ends up being linked with the gold plugin or we error out
if the plugin does not exist. The fix is to use Eq to ignore both
--plugin=foo and --plugin foo as before.
Differential Revision: https://reviews.llvm.org/D47657
llvm-svn: 333793
This fixes two major problems:
- We were not capping vector alignment as desired on 32-bit ARM.
- We were using different alignments based on the AVX settings on
Intel, so we did not have a consistent ABI.
This is an ABI break, but we think we can get away with it because
vectors tend to be used mostly in inline code (which is why not having
a consistent ABI has not proven disastrous on Intel).
Intel's AVX types are specified as having 32-byte / 64-byte alignment,
so align them explicitly instead of relying on the base ABI rule.
Note that this sort of attribute is stripped from template arguments
in template substitution, so there's a possibility that code templated
over vectors will produce inadequately-aligned objects. The right
long-term solution for this is for alignment attributes to be
interpreted as true qualifiers and thus preserved in the canonical type.
llvm-svn: 333791
Summary: One of the tests is failing to build because it needs GS-, the second test does not correctly match all the expected function names because newer DIA SDKs annotate the function names with their return type and inputs (e.g. "static long `anonymous namespace'::StaticFunction(int)")
Reviewers: asmith, zturner
Reviewed By: zturner
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D47653
llvm-svn: 333790
Summary: Skip the new break-insert test on Windows because it hangs and so the test suite never completes. All other lldb-mi tests in the test suite are also skipped on windows
Reviewers: asmith, aprantl, polyakov.alex
Reviewed By: aprantl
Subscribers: ki.stfu, llvm-commits
Differential Revision: https://reviews.llvm.org/D47651
llvm-svn: 333789
In r331719, I changed Module::FindTypes not to limit the amount
of types returned by the Symbol provider, because we want all
possible matches to be able to filter them. In one code path,
the filtering was applied to the TypeList without changing the
number of types that gets returned. This is turn could cause
consumers to access beyond the end of the TypeList.
This patch fixes this case and also adds an assertion to
TypeList::GetTypeAtIndex to catch those obvious programming
mistakes.
Triggering the condition in which we performed the incorrect
access was not easy. It happened a lot in mixed Swift/ObjectiveC
code, but I was able to trigger it in pure Objective C++ although
in a contrieved way.
rdar://problem/40254997
llvm-svn: 333786
It's been pointed out in https://reviews.llvm.org/D47646 that lldb-test
fails to create a usable process on Windows when running this test.
llvm-svn: 333785
Myriad only uses the platform interceptors for memory allocation
routines. Configure them properly.
Also add a missing guard around aligned alloc interceptor.
Differential Revision: https://reviews.llvm.org/D47641
llvm-svn: 333784
Summary:
Getelementptr returns a vector of pointers, instead of a single address,
when one or more of its arguments is a vector. In such case it is not
possible to simplify the expression by inserting a bitcast of operand(0)
into the destination type, as it will create a bitcast between different
sizes.
Reviewers: majnemer, mkuper, mssimpso, spatel
Reviewed By: spatel
Subscribers: lebedev.ri, llvm-commits
Differential Revision: https://reviews.llvm.org/D46379
llvm-svn: 333783
This bug:
https://bugs.llvm.org/show_bug.cgi?id=37648
...was created with the enhancement to this transform with rL332479.
The urem test shows the disaster potential: any undef divisor lane makes
the whole op undef.
The test diffs show that vector demanded elements turns some of the potential,
but not all, unused binop operands back into undef already.
llvm-svn: 333782
Summary:
Occasionally, when launching a process in lldb (especially on windows, but not limited to), lldb will hang before the process is launched and it will never recover. This happens because the timing of the processing of the state changes can be slightly different. The state changes that are issued are:
1) SetPublicState(eStateLaunching)
2) SetPrivateState(eStateLaunching)
3) SetPublicState(eStateStopped)
4) SetPrivateState(eStateStopped)
What we expect to see is:
public state: launching -> launching -> stopped
private state: launching -> stopped
What we see is:
public state: launching -> stopped -> launching
private state: launching -> stopped
The second launching change to the public state is issued when WaitForProcessStopPrivate calls HandlePrivateEvent on the event which was created when the private state was set to launching. HandlePrivateEvent has logic to determine whether to broadcase the event and a launching event is *always* broadcast. At the same time, when the stopped event is processed by WaitForProcessStopPrivate next, the function exists and that event is never broadcast, so the public state remains as launching.
HandlePrivateEvent does two things: determine whether there's a next action as well as determine whether to broadcast the event that was processed. There's only ever a next action set if we are trying to attach to a process, but WaitForProcessStopPrivate is only ever called when we are launching a process or connecting remotely, so the first part of HandlePrivateEvent (handling the next action) is irrelevant for WaitForProcessStopPrivate. As far as broadcasting the event is concerned, since we are handling state changes that already occurred to the public state (and are now duplicated in the private state), I believe the broadcast step is unnecessary also (and in fact, it causes the hang).
This change removes the call to HandlePrivateEvent from inside WaitForProcessStopPrivate.
Incidentally, there was also a bug filed recently that is the same issue: https://bugs.llvm.org/show_bug.cgi?id=37496
Reviewers: asmith, labath, zturner, jingham
Reviewed By: zturner, jingham
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D47609
llvm-svn: 333781
Simon Pilgrim