This fixes a problem where __va_list_tag was not correctly imported,
possibly leading to multiple definitions with different types.
This adds __va_list_tag to it's proper scope, so that the ASTImporter
can find it.
Signed-off-by: Matheus Izvekov <mizvekov@gmail.com>
Differential Revision: https://reviews.llvm.org/D136886
This patch implements P0634r3 that removes the need for 'typename' in certain contexts.
For example,
```
template <typename T>
using foo = T::type; // ok
```
This is also allowed in previous language versions as an extension, because I think it's pretty useful. :)
Reviewed By: #clang-language-wg, erichkeane
Differential Revision: https://reviews.llvm.org/D53847
This speeds up clangd's buildAST() (i.e. parsing with a preamble) by 5% on
clangd/AST.cpp, by avoiding filling up the diagnostic state map with entries for
all the files where templates are being instantiated from.
(I would assume it has a similar effect on PCH and modules compiles).
This approach is obviously pretty fragile, and we should find ways to make
isIgnored() cheaper instead. But these changes in particular don't seem to make
the code worse in any case.
Differential Revision: https://reviews.llvm.org/D129683
This change refactors the MuiltiplexExternalSemaSource to take ownership
of the underlying sources. As a result it makes a larger cleanup of
external source ownership in Sema and the ChainedIncludesSource.
Reviewed By: aaron.ballman, aprantl
Differential Revision: https://reviews.llvm.org/D133158
This adds a check for exported inline functions, that there is a definition in
the definition domain (which, in practice, can only be the module purview but
before any PMF starts) since the PMF definition domain cannot contain exports.
This is:
[dcl.inline]/7
If an inline function or variable that is attached to a named module is declared in
a definition domain, it shall be defined in that domain.
The patch also amends diagnostic output by excluding the PMF sub-module from the
set considered as sources of missing decls. There is no point in telling the user
that the import of a PMF object is missing - since such objects are never reachable
to an importer. We still show the definition (as unreachable), to help point out
this.
Differential Revision: https://reviews.llvm.org/D128328
Leverage the method OpenCL uses that adds C intrinsics when the lookup
failed. There is no need to define C intrinsics in the header file any
more. It could help to avoid the large header file to speed up the
compilation of RVV source code. Besides that, only the C intrinsics used
by the users will be added into the declaration table.
This patch is based on https://reviews.llvm.org/D103228 and inspired by
OpenCL implementation.
### Experimental Results
#### TL;DR:
- Binary size of clang increase ~200k, which is +0.07% for debug build and +0.13% for release build.
- Single file compilation speed up ~33x for debug build and ~8.5x for release build
- Regression time reduce ~10% (`ninja check-all`, enable all targets)
#### Header size change
```
| size | LoC |
------------------------------
Before | 4,434,725 | 69,749 |
After | 6,140 | 162 |
```
#### Single File Compilation Time
Testcase:
```
#include <riscv_vector.h>
vint32m1_t test_vadd_vv_vfloat32m1_t(vint32m1_t op1, vint32m1_t op2, size_t vl) {
return vadd(op1, op2, vl);
}
```
##### Debug build:
Before:
```
real 0m19.352s
user 0m19.252s
sys 0m0.092s
```
After:
```
real 0m0.576s
user 0m0.552s
sys 0m0.024s
```
~33x speed up for debug build
##### Release build:
Before:
```
real 0m0.773s
user 0m0.741s
sys 0m0.032s
```
After:
```
real 0m0.092s
user 0m0.080s
sys 0m0.012s
```
~8.5x speed up for release build
#### Regression time
Note: the failed case is `tools/llvm-debuginfod-find/debuginfod.test` which is unrelated to this patch.
##### Debug build
Before:
```
Testing Time: 1358.38s
Skipped : 11
Unsupported : 446
Passed : 75767
Expectedly Failed: 190
Failed : 1
```
After
```
Testing Time: 1220.29s
Skipped : 11
Unsupported : 446
Passed : 75767
Expectedly Failed: 190
Failed : 1
```
##### Release build
Before:
```
Testing Time: 381.98s
Skipped : 12
Unsupported : 1407
Passed : 74765
Expectedly Failed: 176
Failed : 1
```
After:
```
Testing Time: 346.25s
Skipped : 12
Unsupported : 1407
Passed : 74765
Expectedly Failed: 176
Failed : 1
```
#### Binary size of clang
##### Debug build
Before
```
text data bss dec hex filename
335261851 12726004 552812 348540667 14c64efb bin/clang
```
After
```
text data bss dec hex filename
335442803 12798708 552940 348794451 14ca2e53 bin/clang
```
+253K, +0.07% code size
##### Release build
Before
```
text data bss dec hex filename
144123975 8374648 483140 152981763 91e5103 bin/clang
```
After
```
text data bss dec hex filename
144255762 8447296 483268 153186326 9217016 bin/clang
```
+204K, +0.13%
Authored-by: Kito Cheng <kito.cheng@sifive.com>
Co-Authored-by: Hsiangkai Wang <kai.wang@sifive.com>
Reviewed By: khchen, aaron.ballman
Differential Revision: https://reviews.llvm.org/D111617
This is a recommit of b822efc7404bf09ccfdc1ab7657475026966c3b2,
reverted in dc34d8df4c48b3a8f474360970cae8a58e6c84f0. The commit caused
fails because the test ast-print-fp-pragmas.c did not specify particular
target, and it failed on targets which do not support constrained
intrinsics. The original commit message is below.
AST does not have special nodes for pragmas. Instead a pragma modifies
some state variables of Sema, which in turn results in modified
attributes of AST nodes. This technique applies to floating point
operations as well. Every AST node that can depend on FP options keeps
current set of them.
This technique works well for options like exception behavior or fast
math options. They represent instructions to the compiler how to modify
code generation for the affected nodes. However treatment of FP control
modes has problems with this technique. Modifying FP control mode
(like rounding direction) usually requires operations on hardware, like
writing to control registers. It must be done prior to the first
operation that depends on the control mode. In particular, such
operations are required for implementation of `pragma STDC FENV_ROUND`,
compiler should set up necessary rounding direction at the beginning of
compound statement where the pragma occurs. As there is no representation
for pragmas in AST, the code generation becomes a complicated task in
this case.
To solve this issue FP options are kept inside CompoundStmt. Unlike to FP
options in expressions, these does not affect any operation on FP values,
but only inform the codegen about the FP options that act in the body of
the statement. As all pragmas that modify FP environment may occurs only
at the start of compound statement or at global level, such solution
works for all relevant pragmas. The options are kept as a difference
from the options in the enclosing compound statement or default options,
it helps codegen to set only changed control modes.
Differential Revision: https://reviews.llvm.org/D123952
On some buildbots test `ast-print-fp-pragmas.c` fails, need to investigate it.
This reverts commit 0401fd12d4aa0553347fe34d666fb236d8719173.
This reverts commit b822efc7404bf09ccfdc1ab7657475026966c3b2.
AST does not have special nodes for pragmas. Instead a pragma modifies
some state variables of Sema, which in turn results in modified
attributes of AST nodes. This technique applies to floating point
operations as well. Every AST node that can depend on FP options keeps
current set of them.
This technique works well for options like exception behavior or fast
math options. They represent instructions to the compiler how to modify
code generation for the affected nodes. However treatment of FP control
modes has problems with this technique. Modifying FP control mode
(like rounding direction) usually requires operations on hardware, like
writing to control registers. It must be done prior to the first
operation that depends on the control mode. In particular, such
operations are required for implementation of `pragma STDC FENV_ROUND`,
compiler should set up necessary rounding direction at the beginning of
compound statement where the pragma occurs. As there is no representation
for pragmas in AST, the code generation becomes a complicated task in
this case.
To solve this issue FP options are kept inside CompoundStmt. Unlike to FP
options in expressions, these does not affect any operation on FP values,
but only inform the codegen about the FP options that act in the body of
the statement. As all pragmas that modify FP environment may occurs only
at the start of compound statement or at global level, such solution
works for all relevant pragmas. The options are kept as a difference
from the options in the enclosing compound statement or default options,
it helps codegen to set only changed control modes.
Differential Revision: https://reviews.llvm.org/D123952
When running `clang -E -Ofast` on macOS, the `__FLT_EVAL_METHOD__` macro is `0`, which causes the following typedef to be emitted into the preprocessed source: `typedef float float_t`.
However, when running `clang -c -Ofast`, `__FLT_EVAL_METHOD__` is `-1`, and `typedef long double float_t` is emitted.
This causes build errors for certain projects, which are not reproducible when compiling from preprocessed source.
The issue is that `__FLT_EVAL_METHOD__` is configured in `Sema::Sema` which is not executed when running in `-E` mode.
This change moves that logic into the preprocessor initialization method, which is invoked correctly in `-E` mode.
rdar://96134605
rdar://92748429
Differential Revision: https://reviews.llvm.org/D128814
Warns when end-of-file is reached without seeing all matching
'omp end declare target' directives. The diagnostic shows the
location of the related begin directive.
Differential Revision: https://reviews.llvm.org/D126331
This reverts commit 69dd89fdcbd846375a45e2fe3a88710887236d7a.
This reverts commit 04000c2f928a7adc32138a664d167f01b642bef3.
The current states breaks libstdc++ usage (https://reviews.llvm.org/D119136#3455423).
The fixup has been reverted as it caused other valid code to be disallowed.
I think we should start from the clean state by reverting all relevant commits.
Implement P2036R3.
Captured variables by copy (explicitely or not), are deduced
correctly at the point we know whether the lambda is mutable,
and ill-formed before that.
Up until now, the entire lambda declaration up to the start of the body would be parsed in the parent scope, such that capture would not be available to look up.
The scoping is changed to have an outer lambda scope, followed by the lambda prototype and body.
The lambda scope is necessary because there may be a template scope between the start of the lambda (to which we want to attach the captured variable) and the prototype scope.
We also need to introduce a declaration context to attach the captured variable to (and several parts of clang assume captures are handled from the call operator context), before we know the type of the call operator.
The order of operations is as follow:
* Parse the init capture in the lambda's parent scope
* Introduce a lambda scope
* Create the lambda class and call operator
* Add the init captures to the call operator context and the lambda scope. But the variables are not capured yet (because we don't know their type).
Instead, explicit captures are stored in a temporary map that conserves the order of capture (for the purpose of having a stable order in the ast dumps).
* A flag is set on LambdaScopeInfo to indicate that we have not yet injected the captures.
* The parameters are parsed (in the parent context, as lambda mangling recurses in the parent context, we couldn't mangle a lambda that is attached to the context of a lambda whose type is not yet known).
* The lambda qualifiers are parsed, at this point We can switch (for the second time) inside the lambda context, unset the flag indicating that we have not parsed the lambda qualifiers,
record the lambda is mutable and capture the explicit variables.
* We can parse the rest of the lambda type, transform the lambda and call operator's types and also transform the call operator to a template function decl where necessary.
At this point, both captures and parameters can be injected in the body's scope. When trying to capture an implicit variable, if we are before the qualifiers of a lambda, we need to remember that the variables are still in the parent's context (rather than in the call operator's).
Reviewed By: aaron.ballman, #clang-language-wg, ChuanqiXu
Differential Revision: https://reviews.llvm.org/D119136
Implement P2036R3.
Captured variables by copy (explicitely or not), are deduced
correctly at the point we know whether the lambda is mutable,
and ill-formed before that.
Up until now, the entire lambda declaration up to the start
of the body would be parsed in the parent scope, such that
captures would not be available to look up.
The scoping is changed to have an outer lambda scope,
followed by the lambda prototype and body.
The lambda scope is necessary because there may be a template scope
between the start of the lambda (to which we want to attach
the captured variable) and the prototype scope.
We also need to introduce a declaration context to attach the captured
variable to (and several parts of clang assume captures are handled from
the call operator context), before we know the type of the call operator.
The order of operations is as follow:
* Parse the init capture in the lambda's parent scope
* Introduce a lambda scope
* Create the lambda class and call operator
* Add the init captures to the call operator context and the lambda scope.
But the variables are not capured yet (because we don't know their type).
Instead, explicit captures are stored in a temporary map that
conserves the order of capture (for the purpose of having a stable order in the ast dumps).
* A flag is set on LambdaScopeInfo to indicate that we have not yet injected the captures.
* The parameters are parsed (in the parent context, as lambda mangling recurses in the parent context,
we couldn't mangle a lambda that is attached to the context of a lambda whose type is not yet known).
* The lambda qualifiers are parsed, at this point,
we can switch (for the second time) inside the lambda context,
unset the flag indicating that we have not parsed the lambda qualifiers,
record the lambda is mutable and capture the explicit variables.
* We can parse the rest of the lambda type, transform the lambda and call operator's types and also
transform the call operator to a template function decl where necessary.
At this point, both captures and parameters can be injected in the body's scope.
When trying to capture an implicit variable, if we are before the qualifiers of a lambda,
we need to remember that the variables are still in the parent's context (rather than in the call operator's).
This is a recommit of adff142dc2 after a fix in d8d793f29b4
Reviewed By: aaron.ballman, #clang-language-wg, ChuanqiXu
Differential Revision: https://reviews.llvm.org/D119136
This reverts commit adff142dc253d65b6560e420bba6b858d88d4a98.
This broke clang bootstrap: it made existing C++ code in LLVM invalid:
llvm/include/llvm/CodeGen/LiveInterval.h:630:53: error: captured variable 'Idx' cannot appear here
[=](std::remove_reference_t<decltype(*Idx)> V,
^
Implement P2036R3.
Captured variables by copy (explicitely or not), are deduced
correctly at the point we know whether the lambda is mutable,
and ill-formed before that.
Up until now, the entire lambda declaration up to the start of the body would be parsed in the parent scope, such that capture would not be available to look up.
The scoping is changed to have an outer lambda scope, followed by the lambda prototype and body.
The lambda scope is necessary because there may be a template scope between the start of the lambda (to which we want to attach the captured variable) and the prototype scope.
We also need to introduce a declaration context to attach the captured variable to (and several parts of clang assume captures are handled from the call operator context), before we know the type of the call operator.
The order of operations is as follow:
* Parse the init capture in the lambda's parent scope
* Introduce a lambda scope
* Create the lambda class and call operator
* Add the init captures to the call operator context and the lambda scope. But the variables are not capured yet (because we don't know their type).
Instead, explicit captures are stored in a temporary map that conserves the order of capture (for the purpose of having a stable order in the ast dumps).
* A flag is set on LambdaScopeInfo to indicate that we have not yet injected the captures.
* The parameters are parsed (in the parent context, as lambda mangling recurses in the parent context, we couldn't mangle a lambda that is attached to the context of a lambda whose type is not yet known).
* The lambda qualifiers are parsed, at this point We can switch (for the second time) inside the lambda context, unset the flag indicating that we have not parsed the lambda qualifiers,
record the lambda is mutable and capture the explicit variables.
* We can parse the rest of the lambda type, transform the lambda and call operator's types and also transform the call operator to a template function decl where necessary.
At this point, both captures and parameters can be injected in the body's scope. When trying to capture an implicit variable, if we are before the qualifiers of a lambda, we need to remember that the variables are still in the parent's context (rather than in the call operator's).
Reviewed By: aaron.ballman, #clang-language-wg, ChuanqiXu
Differential Revision: https://reviews.llvm.org/D119136
This builtin returns the address of a global instance of the
`std::source_location::__impl` type, which must be defined (with an
appropriate shape) before calling the builtin.
It will be used to implement std::source_location in libc++ in a
future change. The builtin is compatible with GCC's implementation,
and libstdc++'s usage. An intentional divergence is that GCC declares
the builtin's return type to be `const void*` (for
ease-of-implementation reasons), while Clang uses the actual type,
`const std::source_location::__impl*`.
In order to support this new functionality, I've also added a new
'UnnamedGlobalConstantDecl'. This artificial Decl is modeled after
MSGuidDecl, and is used to represent a generic concept of an lvalue
constant with global scope, deduplicated by its value. It's possible
that MSGuidDecl itself, or some of the other similar sorts of things
in Clang might be able to be refactored onto this more-generic
concept, but there's enough special-case weirdness in MSGuidDecl that
I gave up attempting to share code there, at least for now.
Finally, for compatibility with libstdc++'s <source_location> header,
I've added a second exception to the "cannot cast from void* to T* in
constant evaluation" rule. This seems a bit distasteful, but feels
like the best available option.
Reviewers: aaron.ballman, erichkeane
Differential Revision: https://reviews.llvm.org/D120159
This is the first in a series of patches that introduce C++20 importable
header units.
These differ from clang header modules in that:
(a) they are identifiable by an internal name
(b) they represent the top level source for a single header - although
that might include or import other headers.
We name importable header units with the path by which they are specified
(although that need not be the absolute path for the file).
So "foo/bar.h" would have a name "foo/bar.h". Header units are made a
separate module type so that we can deal with diagnosing places where they
are permitted but a named module is not.
Differential Revision: https://reviews.llvm.org/D121095
Update `WeakUndeclaredIdentifiers` to hold a collection of weak
aliases per identifier instead of only one.
This also allows the "used" state to be removed from `WeakInfo`
because it is really only there as an alternative to removing
processed map entries, and we can represent that using an empty set
now. The serialization code is updated for the removal of the field.
Additionally, a PCH test is added for the new functionality.
The records are grouped by the "target" identifier, which was already
being used as a key for lookup purposes. We also store only one record
per alias name; combined, this means that diagnostics are grouped by
the "target" and limited to one per alias (which should be acceptable).
Fixes PR28611.
Fixesllvm/llvm-project#28985.
Reviewed By: aaron.ballman, cebowleratibm
Differential Revision: https://reviews.llvm.org/D121927
Co-authored-by: Rachel Craik <rcraik@ca.ibm.com>
Co-authored-by: Jamie Schmeiser <schmeise@ca.ibm.com>
CUDA/HIP determines whether a function can be called based on
the device/host attributes of callee and caller. Clang assumes the
caller is CurContext. This is correct in most cases, however, it is
not correct in OpenMP parallel region when CUDA/HIP program
is compiled with -fopenmp. This causes incorrect overloading
resolution and missed diagnostics.
To get the correct caller, clang needs to chase the parent chain
of DeclContext starting from CurContext until a function decl
or a lambda decl is reached. Sema API is adapted to achieve that
and used to determine the caller in hostness check.
Reviewed by: Artem Belevich, Richard Smith
Differential Revision: https://reviews.llvm.org/D121765
FLT_EVAL_METHOD tells the user the precision at which, temporary results
are evaluated but when fast-math is enabled, the numeric values are not
guaranteed to match the source semantics, so the eval-method is
meaningless.
For example, the expression `x + y + z` has as source semantics `(x + y)
+ z`. FLT_EVAL_METHOD is telling the user at which precision `(x + y)`
is evaluated. With fast-math enable the compiler can choose to
evaluate the expression as `(y + z) + x`.
The correct behavior is to set the FLT_EVAL_METHOD to `-1` to tell the
user that the precision of the intermediate values is unknow. This
patch is doing that.
Differential Revision: https://reviews.llvm.org/D121122
We were not creating an evaluation context for the TU scope, so we
never popped an evaluation context for it. Popping the evaluation
context triggers a number of diagnostics, including warnings about
immediate invocations that we were previously missing.
Note: I think we have an additional issue that we should solve, but not
as part of this patch. I don't think Clang is properly modeling static
initialization as happening before constant expression evaluation. I
think structure members members are zero initialized per
http://eel.is/c++draft/basic.start.static#1,
https://eel.is/c++draft/basic.start.static#2.sentence-2, and
http://eel.is/c++draft/dcl.init#general-6.2 and the new test case
actually should be accepted. However, it's also worth noting that other
compilers behave the way this patch makes Clang behave:
https://godbolt.org/z/T7noqhdPr
C++20 non-type template parameter prints `MyType<{{116, 104, 105, 115}}>` when the code is as simple as `MyType<"this">`. This patch prints `MyType<{"this"}>`, with one layer of braces preserved for the intermediate structural type to trigger CTAD.
`StringLiteral` handles this case, but `StringLiteral` inside `APValue` code looks like a circular dependency. The proposed patch implements a cheap strategy to emit string literals in diagnostic messages only when they are readable and fall back to integer sequences.
Reviewed By: aaron.ballman
Differential Revision: https://reviews.llvm.org/D115031
If this is a SFINAE context, then continuing to look up names
(in particular, to treat a non-function as a function, and then
do ADL) might too-eagerly complete a type that it's not safe to
complete right now. We should just say "okay, that's a substitution
failure" and not do any more work than absolutely required.
Fixes#52970.
Differential Revision: https://reviews.llvm.org/D117603
If this is a SFINAE context, then continuing to look up names
(in particular, to treat a non-function as a function, and then
do ADL) might too-eagerly complete a type that it's not safe to
complete right now. We should just say "okay, that's a substitution
failure" and not do any more work than absolutely required.
Fixes#52970.
Differential Revision: https://reviews.llvm.org/D117603
This feature requires support of __opencl_c_generic_address_space and
__opencl_c_program_scope_global_variables so diagnostics for that is provided as well.
Reviewed By: Anastasia
Differential Revision: https://reviews.llvm.org/D115640
This makes the mapping between iOS & tvOS/watchOS versions more accurate. For example, iOS 9.3 now gets correctly mapped into tvOS 9.2 and not tvOS 9.3.
Before this change, the incorrect mapping could cause excessive or missing warnings for code that specifies availability for iOS, but not for tvOS/watchOS.
rdar://81491680
Differential Revision: https://reviews.llvm.org/D116822
This reverts commit 80e2c587498a7b2bf14dde47a33a058da6e88a9a.
The original patch causes a lot of warnings on gcc like:
llvm-project/clang/include/clang/Basic/Diagnostic.h:1329:3: warning:
base class ‘class clang::StreamingDiagnostic’ should be explicitly
initialized in the copy constructor [-Wextra]
Consider case where `__int128` type is supported by the host target but
not by a device target (e.g. spirv*). Clang emits an error message for
unsupported type even if the device code does not use it. This patch
fixes this issue by emitting the error message when the device code
attempts to use the unsupported type.
Reviewed By: tra
Differential Revision: https://reviews.llvm.org/D111047
Adds diagnosing on attempt to use zero length arrays, pointers, refs, arrays
of them and structs/classes containing all of it.
In case a struct/class with zero length array is used this emits a set
of notes pointing out how zero length array got into used struct, like
this:
```
struct ContainsArr {
int A[0]; // note: field of illegal type declared here
};
struct Wrapper {
ContainsArr F; // note: within field of type ContainsArr declared here
// ...
}
// Device code
Wrapper W;
W.use(); // error: zero-length arrays are not permitted
```
Total deep check of each used declaration may result in double
diagnosing at the same location.
Reviewed By: aaron.ballman
Differential Revision: https://reviews.llvm.org/D114080
WG14 adopted the _ExtInt feature from Clang for C23, but renamed the
type to be _BitInt. This patch does the vast majority of the work to
rename _ExtInt to _BitInt, which accounts for most of its size. The new
type is exposed in older C modes and all C++ modes as a conforming
extension. However, there are functional changes worth calling out:
* Deprecates _ExtInt with a fix-it to help users migrate to _BitInt.
* Updates the mangling for the type.
* Updates the documentation and adds a release note to warn users what
is going on.
* Adds new diagnostics for use of _BitInt to call out when it's used as
a Clang extension or as a pre-C23 compatibility concern.
* Adds new tests for the new diagnostic behaviors.
I want to call out the ABI break specifically. We do not believe that
this break will cause a significant imposition for early adopters of
the feature, and so this is being done as a full break. If it turns out
there are critical uses where recompilation is not an option for some
reason, we can consider using ABI tags to ease the transition.
This patch attempts to fix a compiler crash that occurs when long
double type is used with -mno-x87 compiler option.
The option disables x87 target feature, which in turn disables x87
registers, so CG cannot select them for x86_fp80 LLVM IR type. Long
double is lowered as x86_fp80 for some targets, so it leads to a
crash.
The option seems to contradict the SystemV ABI, which requires long
double to be represented as a 80-bit floating point, and it also
requires to use x87 registers.
To avoid that, `long double` type is disabled when -mno-x87 option is
set. In addition to that, `float` and `double` also use x87 registers
for return values on 32-bit x86, so they are disabled as well.
Differential Revision: https://reviews.llvm.org/D98895
This was committed as ec6c847179fd, but then reverted after a failure
in: https://lab.llvm.org/buildbot/#/builders/84/builds/13983
I was not able to reproduce the problem, but I added an extra check
for a NULL QualType just in case.
Original comit message:
The patch adds missing diagnostics for cases like:
float F3 = ((__float128)F1 * (__float128)F2) / 2.0f;
Sema::checkDeviceDecl (renamed to checkTypeSupport) is changed to work
with a type without the corresponding ValueDecl. It is also refactored
so that host diagnostics for unsupported types can be added here as
well.
Differential Revision: https://reviews.llvm.org/D109315
Add atomic_half types and builtins operating on the types from the
cl_ext_float_atomics extension.
Patch by Haonan Yang.
Differential Revision: https://reviews.llvm.org/D109740
