We were using an OpaqueValueExpr allocated on the stack to store
the size of a VLA. Because the VLASizeMap in CodegenFunction
uses the address of the expression to avoid recomputing VLAs,
we were accidentally reusing an earlier llvm::Value. This led to
invalid LLVM IR.
This is a temporary solution until VLASizeMap can be pushed and popped
based on the context.
Differential Revision: https://reviews.llvm.org/D107666
After D94315 we add the `NoInline` attribute to the outlined function to handle
data environments in the OpenMP if clause. This conflicted with the `AlwaysInline`
attribute added to the outlined function. for better performance in D106799.
The data environments should ideally not require NoInline, but for now this
fixes PR51349.
Reviewed By: mikerice
Differential Revision: https://reviews.llvm.org/D107649
The diagnostic texts for warning on attributes that don't appear on the
initial declaration is generally useful. We'd like to re-use it in
D106030, but first let's combine two that already are very similar so we
may re-use it a third time in that commit.
Also, fix a few places that were using notePreviousDefinition to point
to declarations, to instead use diag::note_previous_declaration.
Reviewed By: aaron.ballman
Differential Revision: https://reviews.llvm.org/D107613
Attempt to enable MemCpyOpt unconditionally in D104801 uncovered the fact that
there are users that do not expect LLVM to materialize `memset` intrinsic.
While other passes can do that, too, MemCpyOpt triggers it more frequently and
breaks sanitizers and some downstream users.
For now introduce a flag to force-enable the flag and opt-in only CUDA
compilation with NVPTX back-end.
Differential Revision: https://reviews.llvm.org/D106401
This reverts commit 5181be344a.
Break libcxx type_traits header which uses aligned storage with
alignments greater than 4096. Reverting untill we can fix the header.
%%%
This patch defines __HOS_AIX__ macro for AIX in case of a cross compiler implementation.
%%%
Tested with SPEC.
Reviewed By: cebowleratibm
Differential Revision: https://reviews.llvm.org/D107242
%%%
This patch defines the macro __THW_PPC__ for AIX.
%%%
Tested with SPEC.
Reviewed By: cebowleratibm
Differential Revision: https://reviews.llvm.org/D107243
%%%
This patch defines the macro __THW_BIG_ENDIAN__ for AIX.
%%%
Tested with SPEC.
Reviewed By: cebowleratibm
Differential Revision: https://reviews.llvm.org/D107241
This change provides a way to conveniently declare types that have
address space qualifiers removed.
Since OpenCL adds address spaces implicitly even when they are not
specified in source, it is useful to allow deriving address space
unqualified types.
Fixes llvm.org/PR45326
Differential Revision: https://reviews.llvm.org/D106785
This is recommit of the patch 16ff91ebcc,
reverted in 0c28a7c990 because it had
an error in call of getFastMathFlags (base type should be FPMathOperator
but not Instruction). The original commit message is duplicated below:
Clang has builtin function '__builtin_isnan', which implements C
library function 'isnan'. This function now is implemented entirely in
clang codegen, which expands the function into set of IR operations.
There are three mechanisms by which the expansion can be made.
* The most common mechanism is using an unordered comparison made by
instruction 'fcmp uno'. This simple solution is target-independent
and works well in most cases. It however is not suitable if floating
point exceptions are tracked. Corresponding IEEE 754 operation and C
function must never raise FP exception, even if the argument is a
signaling NaN. Compare instructions usually does not have such
property, they raise 'invalid' exception in such case. So this
mechanism is unsuitable when exception behavior is strict. In
particular it could result in unexpected trapping if argument is SNaN.
* Another solution was implemented in https://reviews.llvm.org/D95948.
It is used in the cases when raising FP exceptions by 'isnan' is not
allowed. This solution implements 'isnan' using integer operations.
It solves the problem of exceptions, but offers one solution for all
targets, however some can do the check in more efficient way.
* Solution implemented by https://reviews.llvm.org/D96568 introduced a
hook 'clang::TargetCodeGenInfo::testFPKind', which injects target
specific code into IR. Now only SystemZ implements this hook and it
generates a call to target specific intrinsic function.
Although these mechanisms allow to implement 'isnan' with enough
efficiency, expanding 'isnan' in clang has drawbacks:
* The operation 'isnan' is hidden behind generic integer operations or
target-specific intrinsics. It complicates analysis and can prevent
some optimizations.
* IR can be created by tools other than clang, in this case treatment
of 'isnan' has to be duplicated in that tool.
Another issue with the current implementation of 'isnan' comes from the
use of options '-ffast-math' or '-fno-honor-nans'. If such option is
specified, 'fcmp uno' may be optimized to 'false'. It is valid
optimization in general, but it results in 'isnan' always returning
'false'. For example, in some libc++ implementations the following code
returns 'false':
std::isnan(std::numeric_limits<float>::quiet_NaN())
The options '-ffast-math' and '-fno-honor-nans' imply that FP operation
operands are never NaNs. This assumption however should not be applied
to the functions that check FP number properties, including 'isnan'. If
such function returns expected result instead of actually making
checks, it becomes useless in many cases. The option '-ffast-math' is
often used for performance critical code, as it can speed up execution
by the expense of manual treatment of corner cases. If 'isnan' returns
assumed result, a user cannot use it in the manual treatment of NaNs
and has to invent replacements, like making the check using integer
operations. There is a discussion in https://reviews.llvm.org/D18513#387418,
which also expresses the opinion, that limitations imposed by
'-ffast-math' should be applied only to 'math' functions but not to
'tests'.
To overcome these drawbacks, this change introduces a new IR intrinsic
function 'llvm.isnan', which realizes the check as specified by IEEE-754
and C standards in target-agnostic way. During IR transformations it
does not undergo undesirable optimizations. It reaches instruction
selection, where is lowered in target-dependent way. The lowering can
vary depending on options like '-ffast-math' or '-ffp-model' so the
resulting code satisfies requested semantics.
Differential Revision: https://reviews.llvm.org/D104854
On AVR, '.ctors' is used, not '.init_array'. Make this the default
unless specifically overridden by driver argument.
This matches gcc, and it matches the behavior in (e.g.) the NetBSD
driver (for certain OS variants).
Reviewed by: MaskRay
Differential Revision: https://reviews.llvm.org/D107610
`__alignof__(x)` always returns `ABIAlign` if the "x" is marked `__attribute__((aligned()))`. However, the "aligned" attribute should only increase the alignment of a struct, or struct member, unless it's used together with the "packed" attribute, or used as a part of a typedef, in which case, the "aligned" attribute can both increase and decrease alignment.
Reviewed By: sfertile
Differential Revision: https://reviews.llvm.org/D107598
GCC supports multiple forms of -falign-loops=.
-falign-loops= is currently ignored in Clang.
This patch implements the simplest but the most useful form where N is a
power of 2.
The underlying implementation uses a `llvm::TargetOptions` option for now.
Bitcode generation ignores this option.
Differential Revision: https://reviews.llvm.org/D106701
Asm is a gnu extension for C, so at present -fopenmp -std=c99
and similar fail to compile on nvptx, bug 51344
Changing to `__asm__` or `__asm` works for openmp, all three appear to work
for cuda. Suggesting `__asm__` here as `__asm` is used by MSVC with different
syntax, so this should make for better error diagnostics if the header is
passed to a compiler other than clang.
Reviewed By: tra, emankov
Differential Revision: https://reviews.llvm.org/D107492
The root problem is a null pointer is accessed during the call to
checkOpenMPLoop, because loop up bound expr is an error expression
due to error diagnostic was emit early.
To fix this, in setLCDeclAndLB, setUB and setStep instead return false,
return true when LB, UB or Step contains Error, so that the checking is
stopped in checkOpenMPLoop.
Differential Revision: https://reviews.llvm.org/D107385
On AIX an aligned attribute cannot decrease the alignment of a variable
when placed on a variable declaration of vector type.
Differential Revision: https://reviews.llvm.org/D107522
Limit the maximum alignment for attribute aligned to 4096 to match
the limit of the .align pseudo op in the system assembler.
Differential Revision: https://reviews.llvm.org/D107497
Clang diagnostics should not start with a capital letter or use
trailing punctuation (https://clang.llvm.org/docs/InternalsManual.html#the-format-string),
but quite a few driver diagnostics were not following this advice. This
corrects the grammar and punctuation to improve consistency, but does
not change the circumstances under which the diagnostics are produced.
Builtin definitions with pointer arguments were duplicated to provide
overloads differing in the pointer argument's address space.
Reduce this duplication by capturing the definitions in multiclasses.
This still results in the same number of builtins in the generated
tables, but the description is more concise now.
Differential Revision: https://reviews.llvm.org/D107151
Implement target builtins for gfx90a including fadd64, fadd32, add2h,
max and min on various global, flat and ds address spaces for which
intrinsics are implemented.
Differential Revision: https://reviews.llvm.org/D106909
This matches the behavior of GCC.
Patch does not change remapping logic itself, so adding one simple smoke test should be enough.
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D107393
For fixed SVE types, predicates are represented using vectors of i8,
where as for scalable types they are represented using vectors of i1. We
can avoid going through memory for casts between these by bitcasting the
i1 scalable vectors to/from a scalable i8 vector of matching size, which
can then use the existing vector insert/extract logic.
Differential Revision: https://reviews.llvm.org/D106860
Zero-width bitfields on AIX pad out to the natral alignment boundary but
do not change the containing records alignment.
Differential Revision: https://reviews.llvm.org/D106900
For some use-cases, it might be useful to be able to turn off modules for C++ in `-cc1`. (The feature is implied by `-std=C++20`.)
This patch exposes the `-fno-cxx-modules` option in `-cc1`.
Reviewed By: arphaman
Differential Revision: https://reviews.llvm.org/D106864
Clang has builtin function '__builtin_isnan', which implements C
library function 'isnan'. This function now is implemented entirely in
clang codegen, which expands the function into set of IR operations.
There are three mechanisms by which the expansion can be made.
* The most common mechanism is using an unordered comparison made by
instruction 'fcmp uno'. This simple solution is target-independent
and works well in most cases. It however is not suitable if floating
point exceptions are tracked. Corresponding IEEE 754 operation and C
function must never raise FP exception, even if the argument is a
signaling NaN. Compare instructions usually does not have such
property, they raise 'invalid' exception in such case. So this
mechanism is unsuitable when exception behavior is strict. In
particular it could result in unexpected trapping if argument is SNaN.
* Another solution was implemented in https://reviews.llvm.org/D95948.
It is used in the cases when raising FP exceptions by 'isnan' is not
allowed. This solution implements 'isnan' using integer operations.
It solves the problem of exceptions, but offers one solution for all
targets, however some can do the check in more efficient way.
* Solution implemented by https://reviews.llvm.org/D96568 introduced a
hook 'clang::TargetCodeGenInfo::testFPKind', which injects target
specific code into IR. Now only SystemZ implements this hook and it
generates a call to target specific intrinsic function.
Although these mechanisms allow to implement 'isnan' with enough
efficiency, expanding 'isnan' in clang has drawbacks:
* The operation 'isnan' is hidden behind generic integer operations or
target-specific intrinsics. It complicates analysis and can prevent
some optimizations.
* IR can be created by tools other than clang, in this case treatment
of 'isnan' has to be duplicated in that tool.
Another issue with the current implementation of 'isnan' comes from the
use of options '-ffast-math' or '-fno-honor-nans'. If such option is
specified, 'fcmp uno' may be optimized to 'false'. It is valid
optimization in general, but it results in 'isnan' always returning
'false'. For example, in some libc++ implementations the following code
returns 'false':
std::isnan(std::numeric_limits<float>::quiet_NaN())
The options '-ffast-math' and '-fno-honor-nans' imply that FP operation
operands are never NaNs. This assumption however should not be applied
to the functions that check FP number properties, including 'isnan'. If
such function returns expected result instead of actually making
checks, it becomes useless in many cases. The option '-ffast-math' is
often used for performance critical code, as it can speed up execution
by the expense of manual treatment of corner cases. If 'isnan' returns
assumed result, a user cannot use it in the manual treatment of NaNs
and has to invent replacements, like making the check using integer
operations. There is a discussion in https://reviews.llvm.org/D18513#387418,
which also expresses the opinion, that limitations imposed by
'-ffast-math' should be applied only to 'math' functions but not to
'tests'.
To overcome these drawbacks, this change introduces a new IR intrinsic
function 'llvm.isnan', which realizes the check as specified by IEEE-754
and C standards in target-agnostic way. During IR transformations it
does not undergo undesirable optimizations. It reaches instruction
selection, where is lowered in target-dependent way. The lowering can
vary depending on options like '-ffast-math' or '-ffp-model' so the
resulting code satisfies requested semantics.
Differential Revision: https://reviews.llvm.org/D104854
See PR48656.
The implementation of the template instantiation of requires expressions
was incorrectly trying to get the expression from an 'ExprRequirement'
before checking if it was an error state.
Signed-off-by: Matheus Izvekov <mizvekov@gmail.com>
Reviewed By: rsmith
Differential Revision: https://reviews.llvm.org/D107399
See PR47174.
When canonicalizing nested name specifiers of the type kind,
the prefix for 'DependentTemplateSpecialization' types was being
dropped, leading to malformed types which would cause failures
when rebuilding template names.
Signed-off-by: Matheus Izvekov <mizvekov@gmail.com>
Reviewed By: rsmith
Differential Revision: https://reviews.llvm.org/D107311
where should not.
Currently we are using QTy->isIncompleteType(&ND) to check incomplete
type. But before doing that, need to instantiate for a class template
specialization or a class member of a class template specialization,
or an array with known size of such..., so that we know it is really
incomplete type.
To fix this using RequireCompleteType instead.
The new test is added into "test/OpenMP/target_update_messages.cpp"
The different of using RequireCompleteType is when emit incomplete type,
an additional note is also emitted to point to where incomplete type
is declared. Because this change, many tests are needed to be fixed
by adding additional note.
This is to fix https://bugs.llvm.org/show_bug.cgi?id=50508
Differential Revision: https://reviews.llvm.org/D107200
https://reviews.llvm.org/D105964 updated the detection of function
definitions. It had the unfortunate effect to start marking object
definitions with attribute-like macros as function definitions.
This addresses this issue.
Reviewed By: owenpan
Differential Revision: https://reviews.llvm.org/D107269
This patch implements P2092
Simple requirements in requirement body shall not start with requires.
A warning was already in place so we just turn this warning into an error.
In addition, we add tests to make sure typename is optional in
requirement-parameter-list as per the same paper.
Previously we would show an error, but keep the member, and also the
CXXRrecordDecl, valid. This could lead to crashes when attempting to
access the record layout or size.
Differential Revision: https://reviews.llvm.org/D105478
The patch https://reviews.llvm.org/D105964 (58494c856a)
updated detection of function declaration names. It had the unfortunate
consequence that it started breaking between `function` and the function
name in some cases in JavaScript code.
This patch addresses this.
Reviewed By: MyDeveloperDay, owenpan
Differential Revision: https://reviews.llvm.org/D107267
The declaration for the global new function in C++ is generated in the compiler front-end. When examining exception propagation, we found that this is the largest root throw site propagator requiring unwind code to be generated for callers up the stack. Allowing this to be handled immediately with termination stops upward propagation and leads to significantly less landing pads generated. This in turns leads to a performance and .text size win.
With `-fnew-infallible` this annotates the declaration with `throw()` and `__attribute__((returns_nonnull))`. `throw()` allows the compiler to assume exceptions do not propagate out of new and eliminate it as a root throw site. Note that the definition of global new is user-replaceable so users should ensure that the one used follows these semantics.
Measuring internally, we're seeing at 0.5% CPU win in one of our large internal FB workload. Measuring on clang self-build (cd0a1226b5) we get:
thinlto/
"dwarfehprepare.NumCleanupLandingPadsRemaining": 153494,
"dwarfehprepare.NumNoUnwind": 26309,
thinlto_newinfallible/
"dwarfehprepare.NumCleanupLandingPadsRemaining": 143660,
"dwarfehprepare.NumNoUnwind": 28744,
a 1-143660/153494 = 6.4% reduction in landing pads and a 28744/26309 = 9.3% increase in the number of nounwind functions.
Testing:
ninja check-all
new test case to make sure these attributes are added correctly to global new.
Reviewed By: urnathan
Differential Revision: https://reviews.llvm.org/D105225
The new -mtargetos= option is a replacement for the existing, OS-specific options
like -miphoneos-version-min=. This allows us to introduce support for new darwin OSes
easier as they won't require the use of a new option. The older options will be
deprecated and the use of the new option will be encouraged instead.
Differential Revision: https://reviews.llvm.org/D106316
Craig Topper