This reverts commit 80371c74ae.
Given the following source:
```
void a() {
for (;;)
;
}
```
It incorrectly enables vectorization (with vector width 1), as well as generating a warning that vectorization could not be performed.
Sometimes a global var is replaced by a different llvm value. clang use GetAddrOfGlobalVar to get the original llvm global variable.
For most targets, GetAddrOfGlobalVar returns either the llvm global variable or a bitcast of the llvm global variable.
However, for AMDGPU target, GetAddrOfGlobalVar returns the addrspace cast or addrspace cast plus bitcast of the llvm global variable.
To get the llvm global variable, these casts need to be stripped, otherwise there is assertion.
This patch fixes that.
Differential Revision: https://reviews.llvm.org/D69129
llvm-svn: 375362
This fixes an assertion failure in the case where an implicit conversion for a
function call involves an lvalue function conversion, and makes the AST for
initializations involving implicit lvalue function conversions more accurate.
Differential Revision: https://reviews.llvm.org/D66437
llvm-svn: 375313
This adds support for rewriting <, >, <=, and >= to a normal or reversed
call to operator<=>, for rewriting != to a normal or reversed call to
operator==, and for rewriting <=> and == to reversed forms of those same
operators.
Note that this is a breaking change for various C++17 code patterns,
including some in use in LLVM. The most common patterns (where an
operator== becomes ambiguous with a reversed form of itself) are still
accepted under this patch, as an extension (with a warning). I'm hopeful
that we can get the language rules fixed before C++20 ships, and the
extension warning is aimed primarily at providing data to inform that
decision.
llvm-svn: 375306
Remove dead virtual functions from vtables with
replaceNonMetadataUsesWith, so that CGProfile metadata gets cleaned up
correctly.
Original commit message:
Currently, it is hard for the compiler to remove unused C++ virtual
functions, because they are all referenced from vtables, which are referenced
by constructors. This means that if the constructor is called from any live
code, then we keep every virtual function in the final link, even if there
are no call sites which can use it.
This patch allows unused virtual functions to be removed during LTO (and
regular compilation in limited circumstances) by using type metadata to match
virtual function call sites to the vtable slots they might load from. This
information can then be used in the global dead code elimination pass instead
of the references from vtables to virtual functions, to more accurately
determine which functions are reachable.
To make this transformation safe, I have changed clang's code-generation to
always load virtual function pointers using the llvm.type.checked.load
intrinsic, instead of regular load instructions. I originally tried writing
this using clang's existing code-generation, which uses the llvm.type.test
and llvm.assume intrinsics after doing a normal load. However, it is possible
for optimisations to obscure the relationship between the GEP, load and
llvm.type.test, causing GlobalDCE to fail to find virtual function call
sites.
The existing linkage and visibility types don't accurately describe the scope
in which a virtual call could be made which uses a given vtable. This is
wider than the visibility of the type itself, because a virtual function call
could be made using a more-visible base class. I've added a new
!vcall_visibility metadata type to represent this, described in
TypeMetadata.rst. The internalization pass and libLTO have been updated to
change this metadata when linking is performed.
This doesn't currently work with ThinLTO, because it needs to see every call
to llvm.type.checked.load in the linkage unit. It might be possible to
extend this optimisation to be able to use the ThinLTO summary, as was done
for devirtualization, but until then that combination is rejected in the
clang driver.
To test this, I've written a fuzzer which generates random C++ programs with
complex class inheritance graphs, and virtual functions called through object
and function pointers of different types. The programs are spread across
multiple translation units and DSOs to test the different visibility
restrictions.
I've also tried doing bootstrap builds of LLVM to test this. This isn't
ideal, because only classes in anonymous namespaces can be optimised with
-fvisibility=default, and some parts of LLVM (plugins and bugpoint) do not
work correctly with -fvisibility=hidden. However, there are only 12 test
failures when building with -fvisibility=hidden (and an unmodified compiler),
and this change does not cause any new failures for either value of
-fvisibility.
On the 7 C++ sub-benchmarks of SPEC2006, this gives a geomean code-size
reduction of ~6%, over a baseline compiled with "-O2 -flto
-fvisibility=hidden -fwhole-program-vtables". The best cases are reductions
of ~14% in 450.soplex and 483.xalancbmk, and there are no code size
increases.
I've also run this on a set of 8 mbed-os examples compiled for Armv7M, which
show a geomean size reduction of ~3%, again with no size increases.
I had hoped that this would have no effect on performance, which would allow
it to awlays be enabled (when using -fwhole-program-vtables). However, the
changes in clang to use the llvm.type.checked.load intrinsic are causing ~1%
performance regression in the C++ parts of SPEC2006. It should be possible to
recover some of this perf loss by teaching optimisations about the
llvm.type.checked.load intrinsic, which would make it worth turning this on
by default (though it's still dependent on -fwhole-program-vtables).
Differential revision: https://reviews.llvm.org/D63932
llvm-svn: 375094
Implement mangling for CSEs to match regular template-ids.
Reviewed as part of D41569 <https://reviews.llvm.org/D41569>.
Re-commit fixing failing test.
llvm-svn: 375063
Summary:
We don't know what context to use until the classification result is
consumed by the parser, which could happen in a different semantic
context. So don't build the expression that results from name
classification until we get to that point and can handle it properly.
This covers everything except C++ implicit class member access, which
is a little awkward to handle properly in the face of the protected
member access check. But it at least fixes all the currently-filed
instances of PR43080.
Reviewers: efriedma
Subscribers: cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D68896
llvm-svn: 374826
MS name mangling supports cache for first 10 distinct function
arguments. The error was when non cached template type occurred twice
(e.g. 11th and 12th). For such case in code there is another cache
table TemplateArgStrings (for performance reasons). Then one '@'
character at the end of the mangled name taken from this table was
missing. For other cases the missing '@' character was added in
the call to mangleSourceName(TemplateMangling) in the cache miss code,
but the cache hit code didn't add it.
This fixes a regression from r362560.
Patch by Adam Folwarczny <adamf88@gmail.com>!
Differential Revision: https://reviews.llvm.org/D68099
llvm-svn: 374543
Currently, it is hard for the compiler to remove unused C++ virtual
functions, because they are all referenced from vtables, which are referenced
by constructors. This means that if the constructor is called from any live
code, then we keep every virtual function in the final link, even if there
are no call sites which can use it.
This patch allows unused virtual functions to be removed during LTO (and
regular compilation in limited circumstances) by using type metadata to match
virtual function call sites to the vtable slots they might load from. This
information can then be used in the global dead code elimination pass instead
of the references from vtables to virtual functions, to more accurately
determine which functions are reachable.
To make this transformation safe, I have changed clang's code-generation to
always load virtual function pointers using the llvm.type.checked.load
intrinsic, instead of regular load instructions. I originally tried writing
this using clang's existing code-generation, which uses the llvm.type.test
and llvm.assume intrinsics after doing a normal load. However, it is possible
for optimisations to obscure the relationship between the GEP, load and
llvm.type.test, causing GlobalDCE to fail to find virtual function call
sites.
The existing linkage and visibility types don't accurately describe the scope
in which a virtual call could be made which uses a given vtable. This is
wider than the visibility of the type itself, because a virtual function call
could be made using a more-visible base class. I've added a new
!vcall_visibility metadata type to represent this, described in
TypeMetadata.rst. The internalization pass and libLTO have been updated to
change this metadata when linking is performed.
This doesn't currently work with ThinLTO, because it needs to see every call
to llvm.type.checked.load in the linkage unit. It might be possible to
extend this optimisation to be able to use the ThinLTO summary, as was done
for devirtualization, but until then that combination is rejected in the
clang driver.
To test this, I've written a fuzzer which generates random C++ programs with
complex class inheritance graphs, and virtual functions called through object
and function pointers of different types. The programs are spread across
multiple translation units and DSOs to test the different visibility
restrictions.
I've also tried doing bootstrap builds of LLVM to test this. This isn't
ideal, because only classes in anonymous namespaces can be optimised with
-fvisibility=default, and some parts of LLVM (plugins and bugpoint) do not
work correctly with -fvisibility=hidden. However, there are only 12 test
failures when building with -fvisibility=hidden (and an unmodified compiler),
and this change does not cause any new failures for either value of
-fvisibility.
On the 7 C++ sub-benchmarks of SPEC2006, this gives a geomean code-size
reduction of ~6%, over a baseline compiled with "-O2 -flto
-fvisibility=hidden -fwhole-program-vtables". The best cases are reductions
of ~14% in 450.soplex and 483.xalancbmk, and there are no code size
increases.
I've also run this on a set of 8 mbed-os examples compiled for Armv7M, which
show a geomean size reduction of ~3%, again with no size increases.
I had hoped that this would have no effect on performance, which would allow
it to awlays be enabled (when using -fwhole-program-vtables). However, the
changes in clang to use the llvm.type.checked.load intrinsic are causing ~1%
performance regression in the C++ parts of SPEC2006. It should be possible to
recover some of this perf loss by teaching optimisations about the
llvm.type.checked.load intrinsic, which would make it worth turning this on
by default (though it's still dependent on -fwhole-program-vtables).
Differential revision: https://reviews.llvm.org/D63932
llvm-svn: 374539
Summary:
Quote from http://eel.is/c++draft/expr.add#4:
```
4 When an expression J that has integral type is added to or subtracted
from an expression P of pointer type, the result has the type of P.
(4.1) If P evaluates to a null pointer value and J evaluates to 0,
the result is a null pointer value.
(4.2) Otherwise, if P points to an array element i of an array object x with n
elements ([dcl.array]), the expressions P + J and J + P
(where J has the value j) point to the (possibly-hypothetical) array
element i+j of x if 0≤i+j≤n and the expression P - J points to the
(possibly-hypothetical) array element i−j of x if 0≤i−j≤n.
(4.3) Otherwise, the behavior is undefined.
```
Therefore, as per the standard, applying non-zero offset to `nullptr`
(or making non-`nullptr` a `nullptr`, by subtracting pointer's integral value
from the pointer itself) is undefined behavior. (*if* `nullptr` is not defined,
i.e. e.g. `-fno-delete-null-pointer-checks` was *not* specified.)
To make things more fun, in C (6.5.6p8), applying *any* offset to null pointer
is undefined, although Clang front-end pessimizes the code by not lowering
that info, so this UB is "harmless".
Since rL369789 (D66608 `[InstCombine] icmp eq/ne (gep inbounds P, Idx..), null -> icmp eq/ne P, null`)
LLVM middle-end uses those guarantees for transformations.
If the source contains such UB's, said code may now be miscompiled.
Such miscompilations were already observed:
* https://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20190826/687838.html
* https://github.com/google/filament/pull/1566
Surprisingly, UBSan does not catch those issues
... until now. This diff teaches UBSan about these UB's.
`getelementpointer inbounds` is a pretty frequent instruction,
so this does have a measurable impact on performance;
I've addressed most of the obvious missing folds (and thus decreased the performance impact by ~5%),
and then re-performed some performance measurements using my [[ https://github.com/darktable-org/rawspeed | RawSpeed ]] benchmark:
(all measurements done with LLVM ToT, the sanitizer never fired.)
* no sanitization vs. existing check: average `+21.62%` slowdown
* existing check vs. check after this patch: average `22.04%` slowdown
* no sanitization vs. this patch: average `48.42%` slowdown
Reviewers: vsk, filcab, rsmith, aaron.ballman, vitalybuka, rjmccall, #sanitizers
Reviewed By: rsmith
Subscribers: kristof.beyls, nickdesaulniers, nikic, ychen, dtzWill, xbolva00, dberris, arphaman, rupprecht, reames, regehr, llvm-commits, cfe-commits
Tags: #clang, #sanitizers, #llvm
Differential Revision: https://reviews.llvm.org/D67122
llvm-svn: 374293
This was further discussed at the llvm dev list:
http://lists.llvm.org/pipermail/llvm-dev/2019-October/135602.html
I think the brief summary of that is that this change is an improvement,
this is the behaviour that we expect and promise in ours docs, and also
as a result there are cases where we now emit diagnostics whereas before
pragmas were silently ignored. Two areas where we can improve: 1) the
diagnostic message itself, and 2) and in some cases (e.g. -Os and -Oz)
the vectoriser is (quite understandably) not triggering.
Original commit message:
Specifying the vectorization width was supposed to implicitly enable
vectorization, except that it wasn't really doing this. It was only
setting the vectorize.width metadata, but not vectorize.enable.
This should fix PR27643.
llvm-svn: 374288
This reverts r374268 (git commit c34385d07c)
I think I reverted this by mistake, so I'm relanding it. While my bisect
found this revision, I think the crashes I'm seeing locally must be
environmental. Maybe the version of clang I'm using miscompiles tot
clang.
llvm-svn: 374269
Summary:
- [Itanium C++ ABI][1], for certain contexts like default parameter and
etc., mangling numbering will be local to the particular argument in
which it appears.
- However, for these cases, the mangle numbering context is allocated per
expression evaluation stack entry. That causes, for example, two
lambdas defined/used understand the same default parameter are
numbered as the same value and, in turn, one of them is not generated
at all.
- In this patch, an extra mangle numbering context map is maintained in
the AST context to map taht extra declaration context to its numbering
context. So that, 2 different lambdas defined/used in the same default
parameter are numbered differently.
[1]: https://itanium-cxx-abi.github.io/cxx-abi/abi.html
Reviewers: rsmith, eli.friedman
Subscribers: cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D68715
llvm-svn: 374200
"non-constant" value.
If the constant evaluator evaluates part of a variable initializer,
including the initializer for some lifetime-extended temporary, but
fails to fully evaluate the initializer, it can leave behind wrong
values for temporaries encountered in that initialization. Don't try to
emit those from CodeGen! Instead, look at the values that constant
evaluation produced if (and only if) it actually succeeds and we're
emitting the lifetime-extending declaration's initializer as a constant.
llvm-svn: 374119
r369697 changed the behavior of stripPointerCasts to no longer include
aliases. However, the code in CGDeclCXX.cpp's createAtExitStub counted
on the looking through aliases to properly set the calling convention of
a call.
The result of the change was that the calling convention mismatch of the
call would be replaced with a llvm.trap, causing a runtime crash.
Differential Revision: https://reviews.llvm.org/D68584
llvm-svn: 373929
same.
We were missing the lvalue-to-rvalue conversion entirely in this case,
and in fact still need the full CK_LValueToRValueBitCast conversion to
perform a load with no TBAA.
llvm-svn: 373874
We previously failed to treat an array with an instantiation-dependent
but not value-dependent bound as being an instantiation-dependent type.
We now track the array bound expression as part of a constant array type
if it's an instantiation-dependent expression.
llvm-svn: 373685
variable with non-trivial destruction.
We still need to invoke the thread wrapper to trigger registration of
the destructor call on thread shutdown.
llvm-svn: 373289
has a constexpr destructor.
For constexpr variables, reject if the variable does not have constant
destruction. In all cases, do not emit runtime calls to the destructor
for variables with constant destruction.
llvm-svn: 373159
This work-around was necessary to handle standard library headers in
Visual Studio 2019 16.2. Now that 16.3 has shipped to stable, we can
remove it.
> Re-commit r363191 "[MS] Pretend constexpr variable template specializations are inline"
>
> While the next Visual Studio update (16.3) will fix this issue, that hasn't
> shipped yet. Until then Clang wouldn't work with MSVC's headers which seems
> unfortunate. Let's keep this in until VS 16.3 ships. (See also PR42843.)
>
>> Fixes link errors with clang and the latest Visual C++ 14.21.27702
>> headers, which was reported as PR42027.
>>
>> I chose to intentionally make these things linkonce_odr, i.e.
>> discardable, so that we don't emit definitions of these things in every
>> translation unit that includes STL headers.
>>
>> Note that this is *not* what MSVC does: MSVC has not yet implemented C++
>> DR2387, so they emit fully specialized constexpr variable templates with
>> static / internal linkage.
>>
>> Reviewers: rsmith
>>
>> Differential Revision: https://reviews.llvm.org/D63175
llvm-svn: 372844
This broke the Chromium build. Consider the following code:
float ScaleSumSamples_C(const float* src, float* dst, float scale, int width) {
float fsum = 0.f;
int i;
#if defined(__clang__)
#pragma clang loop vectorize_width(4)
#endif
for (i = 0; i < width; ++i) {
float v = *src++;
fsum += v * v;
*dst++ = v * scale;
}
return fsum;
}
Compiling at -Oz, Clang now warns:
$ clang++ -target x86_64 -Oz -c /tmp/a.cc
/tmp/a.cc:1:7: warning: loop not vectorized: the optimizer was unable to
perform the requested transformation; the transformation might be disabled or
specified as part of an unsupported transformation ordering
[-Wpass-failed=transform-warning]
this suggests it's not actually enabling vectorization hard enough.
At -Os it asserts instead:
$ build.release/bin/clang++ -target x86_64 -Os -c /tmp/a.cc
clang-10: /work/llvm.monorepo/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp:2734: void
llvm::InnerLoopVectorizer::emitMemRuntimeChecks(llvm::Loop*, llvm::BasicBlock*): Assertion `
!BB->getParent()->hasOptSize() && "Cannot emit memory checks when optimizing for size"' failed.
Of course neither of these are what the developer expected from the pragma.
> Specifying the vectorization width was supposed to implicitly enable
> vectorization, except that it wasn't really doing this. It was only
> setting the vectorize.width metadata, but not vectorize.enable.
>
> This should fix PR27643.
>
> Differential Revision: https://reviews.llvm.org/D66290
llvm-svn: 372225
Specifying the vectorization width was supposed to implicitly enable
vectorization, except that it wasn't really doing this. It was only
setting the vectorize.width metadata, but not vectorize.enable.
This should fix PR27643.
Differential Revision: https://reviews.llvm.org/D66290
llvm-svn: 372082
levels:
-- none: no lax vector conversions [new GCC default]
-- integer: only conversions between integer vectors [old GCC default]
-- all: all conversions between same-size vectors [Clang default]
For now, Clang still defaults to "all" mode, but per my proposal on
cfe-dev (2019-04-10) the default will be changed to "integer" as soon as
that doesn't break lots of testcases. (Eventually I'd like to change the
default to "none" to match GCC and general sanity.)
Following GCC's behavior, the driver flag -flax-vector-conversions is
translated to -flax-vector-conversions=integer.
This reinstates r371805, reverted in r371813, with an additional fix for
lldb.
llvm-svn: 371817
levels:
-- none: no lax vector conversions [new GCC default]
-- integer: only conversions between integer vectors [old GCC default]
-- all: all conversions between same-size vectors [Clang default]
For now, Clang still defaults to "all" mode, but per my proposal on
cfe-dev (2019-04-10) the default will be changed to "integer" as soon as
that doesn't break lots of testcases. (Eventually I'd like to change the
default to "none" to match GCC and general sanity.)
Following GCC's behavior, the driver flag -flax-vector-conversions is
translated to -flax-vector-conversions=integer.
llvm-svn: 371805
Summary:
* Don't bother using a thread wrapper when the variable is known to
have constant initialization.
* Emit the thread wrapper as discardable-if-unused in TUs that don't
contain a definition of the thread_local variable.
* Don't emit the thread wrapper at all if the thread_local variable
is unused and discardable; it will be emitted by all TUs that need
it.
Reviewers: rjmccall, jdoerfert
Subscribers: cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D67429
llvm-svn: 371767
D67208 added a new test line to wasm-eh.cpp that invokes the LLVM
backend and this test fails on bots that don't have WebAssembly target.
This makes wasm-eh.cpp explicitly require WebAssembly so this will be
skipped on those targets.
llvm-svn: 371711
Summary:
This adds `-fwasm-exceptions` (in similar fashion with
`-fdwarf-exceptions` or `-fsjlj-exceptions`) that turns on everything
with wasm exception handling from the frontend to the backend.
We currently have `-mexception-handling` in clang frontend, but this is
only about the architecture capability and does not turn on other
necessary options such as the exception model in the backend. (This can
be turned on with `llc -exception-model=wasm`, but llc is not invoked
separately as a command line tool, so this option has to be transferred
from clang.)
Turning on `-fwasm-exceptions` in clang also turns on
`-mexception-handling` if not specified, and will error out if
`-mno-exception-handling` is specified.
Reviewers: dschuff, tlively, sbc100
Subscribers: aprantl, jgravelle-google, sunfish, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D67208
llvm-svn: 371708
Summary:
Microsoft seems to do this regardless of the language mode, so we must
also do it in order to be ABI compatible.
Fixes PR36125
Reviewers: thakis
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D47956
llvm-svn: 371642
Multi-versioned functions defined by cpu_dispatch and implemented with IFunc
can not be called outside the translation units where they are defined due to
lack of symbols. This patch add function aliases for these functions and thus
make them visible outside.
Differential Revision: https://reviews.llvm.org/D67058
Patch by Senran Zhang
llvm-svn: 371586
The anon namespace id is a hash of the main input path to the compiler,
which varies in the test suite because the input path is absolute.
llvm-svn: 371277
This avoids cloning variadic virtual methods when the target supports
musttail and the return type is not covariant. I think we never
implemented this previously because it doesn't handle the covariant
case. But, in the MS ABI, there are some cases where vtable thunks must
be emitted even when the variadic method defintion is not available, so
it looks like we need to implement this. Do it for both ABIs, since it's
a nice size improvement and simplification for Itanium.
Emit an error when emitting thunks for variadic methods with a covariant
return type. This case is essentially not implementable unless the ABI
provides a way to perfectly forward variadic arguments without a tail
call.
Fixes PR43173.
Differential Revision: https://reviews.llvm.org/D67028
llvm-svn: 371269
Match cl.exe's mangling for decomposition declarations.
Decomposition declarations are considered to be anonymous structs,
and use the same convention as for anonymous struct/union declarations.
Naming confirmed to match https://godbolt.org/z/K2osJa
Patch from Eric Astor <epastor@google.com>!
Differential Revision: https://reviews.llvm.org/D67202
llvm-svn: 371124
Jordan Rupprecht