An archive containing device code object files can be passed to
clang command line for linking. For each given offload target
it creates a device specific archives which is either passed to llvm-link
if the target is amdgpu, or to clang-nvlink-wrapper if the target is
nvptx. -L/-l flags are used to specify these fat archives on the command
line. E.g.
clang++ -fopenmp -fopenmp-targets=nvptx64 main.cpp -L. -lmylib
It currently doesn't support linking an archive directly, like:
clang++ -fopenmp -fopenmp-targets=nvptx64 main.cpp libmylib.a
Linking with x86 offload also does not work.
Reviewed By: ye-luo
Differential Revision: https://reviews.llvm.org/D105191
This reverts c7f16ab3e3f27d944db72908c9c1b1b7366f5515 / r109694 - which
suggested this was done to improve consistency with the gdb test suite.
Possible that at the time GCC did not canonicalize integer types, and so
matching types was important for cross-compiler validity, or that it was
only a case of over-constrained test cases that printed out/tested the
exact names of integer types.
In any case neither issue seems to exist today based on my limited
testing - both gdb and lldb canonicalize integer types (in a way that
happens to match Clang's preferred naming, incidentally) and so never
print the original text name produced in the DWARF by GCC or Clang.
This canonicalization appears to be in `integer_types_same_name_p` for
GDB and in `TypeSystemClang::GetBasicTypeEnumeration` for lldb.
(I tested this with one translation unit defining 3 variables - `long`,
`long (*)()`, and `int (*)()`, and another translation unit that had
main, and a function that took `long (*)()` as a parameter - then
compiled them with mismatched compilers (either GCC+Clang, or
Clang+(Clang with this patch applied)) and no matter the combination,
despite the debug info for one CU naming the type "long int" and the
other naming it "long", both debuggers printed out the name as "long"
and were able to correctly perform overload resolution and pass the
`long int (*)()` variable to the `long (*)()` function parameter)
Did find one hiccup, identified by the lldb test suite - that CodeView
was relying on these names to map them to builtin types in that format.
So added some handling for that in LLVM. (these could be split out into
separate patches, but seems small enough to not warrant it - will do
that if there ends up needing any reverti/revisiting)
Differential Revision: https://reviews.llvm.org/D110455
The patch implements header-only support for testure lookups.
The patch has been tested on a source file with all possible combinations of
argument types supported by CUDA headers, compiled and verified that the
generated instructions and their parameters match the code generated by NVCC.
Unfortunately, compiling texture code requires CUDA headers and can't be tested
in clang itself. The test will need to be added to the test-suite later.
While generated code compiles and seems to match NVCC, I do not have any code
that uses textures that I could test correctness of the implementation. Hence
the experimental status.
Differential Revision: https://reviews.llvm.org/D110089
declaration.
Names starting with an underscore are reserved at the global scope, so
cannot be used as the name of an extern "C" symbol in any scope because
such usages conflict with a name at global scope.
Also do not warn on `#define _foo` or `#undef _foo`.
Only global scope names starting with _[a-z] are reserved, not the use
of such an identifier in any other context.
adjustMemberOfForLambdaCaptures.
The problem is happening when user passes lambda function with reference
type in the map clause.
The natural of the problem when processing generateInfoForCapture,
the BasePointer is generated with new load for a lambda variable with
reference type. It is not expected in adjustMemberOfForLambdaCaptures.
One way to fix this is to skipping call to generateInfoForCapture for
map(to:lambda). The map info will be generated later in the call to
generateDefaultMapInfo samiler as firsprivate clase.
This to fix https://bugs.llvm.org/show_bug.cgi?id=52071
Differential Revision:https://reviews.llvm.org/D111115
This is to save memory for Clang compiles.
Measuring building PassBuilder.cpp under /usr/bin/time, max rss goes from 0.93GB to 0.7GB.
This does not turn it by default yet.
I've turned on the option locally and run it over a good amount of files without any issues.
For more background, see
https://lists.llvm.org/pipermail/cfe-dev/2021-September/068930.html.
Reviewed By: dblaikie
Differential Revision: https://reviews.llvm.org/D111105
Currently the max alignment representable is 1GB, see D108661.
Setting the align of an object to 4GB is desirable in some cases to make sure the lower 32 bits are clear which can be used for some optimizations, e.g. https://crbug.com/1016945.
This uses an extra bit in instructions that carry an alignment. We can store 15 bits of "free" information, and with this change some instructions (e.g. AtomicCmpXchgInst) use 14 bits.
We can increase the max alignment representable above 4GB (up to 2^62) since we're only using 33 of the 64 values, but I've just limited it to 4GB for now.
The one place we have to update the bitcode format is for the alloca instruction. It stores its alignment into 5 bits of a 32 bit bitfield. I've added another field which is 8 bits and should be future proof for a while. For backward compatibility, we check if the old field has a value and use that, otherwise use the new field.
Updating clang's max allowed alignment will come in a future patch.
Reviewed By: hans
Differential Revision: https://reviews.llvm.org/D110451
Currently the max alignment representable is 1GB, see D108661.
Setting the align of an object to 4GB is desirable in some cases to make sure the lower 32 bits are clear which can be used for some optimizations, e.g. https://crbug.com/1016945.
This uses an extra bit in instructions that carry an alignment. We can store 15 bits of "free" information, and with this change some instructions (e.g. AtomicCmpXchgInst) use 14 bits.
We can increase the max alignment representable above 4GB (up to 2^62) since we're only using 33 of the 64 values, but I've just limited it to 4GB for now.
The one place we have to update the bitcode format is for the alloca instruction. It stores its alignment into 5 bits of a 32 bit bitfield. I've added another field which is 8 bits and should be future proof for a while. For backward compatibility, we check if the old field has a value and use that, otherwise use the new field.
Updating clang's max allowed alignment will come in a future patch.
Reviewed By: hans
Differential Revision: https://reviews.llvm.org/D110451
Clang would reject
#pragma omp for
#pragma omp tile sizes(P)
for (int i = 0; i < 128; ++i) {}
where P is a template parameter, but the loop itself is not
template-dependent. Because P context-dependent, the TransformedStmt
cannot be generated and therefore is nullptr (until the template is
instantiated by TreeTransform). The OMPForDirective would still expect
the a loop is the dependent context and trigger an error.
Fix by introducing a NumGeneratedLoops field to OMPLoopTransformation.
This is used to distinguish the case where no TransformedStmt will be
generated at all (e.g. #pragma omp unroll full) and template
instantiation is needed. In the latter case, delay resolving the
iteration space like when the for-loop itself is template-dependent
until the template instatiation.
A more radical solution would always delay the iteration space analysis
until template instantiation, but would also break many test cases.
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D111124
There is an error in the implementation of the logic of reaching the `Unknonw` tristate in CmpOpTable.
```
void cmp_op_table_unknownX2(int x, int y, int z) {
if (x >= y) {
// x >= y [1, 1]
if (x + z < y)
return;
// x + z < y [0, 0]
if (z != 0)
return;
// x < y [0, 0]
clang_analyzer_eval(x > y); // expected-warning{{TRUE}} expected-warning{{FALSE}}
}
}
```
We miss the `FALSE` warning because the false branch is infeasible.
We have to exploit simplification to discover the bug. If we had `x < y`
as the second condition then the analyzer would return the parent state
on the false path and the new constraint would not be part of the State.
But adding `z` to the condition makes both paths feasible.
The root cause of the bug is that we reach the `Unknown` tristate
twice, but in both occasions we reach the same `Op` that is `>=` in the
test case. So, we reached `>=` twice, but we never reached `!=`, thus
querying the `Unknonw2x` column with `getCmpOpStateForUnknownX2` is
wrong.
The solution is to ensure that we reached both **different** `Op`s once.
Differential Revision: https://reviews.llvm.org/D110910
The default wchar type is different on AIX vs. Linux. When this test is run on
AIX, WCHAR_T_TYPE ends up being set to int. This is incorrect as the default
wchar type on AIX is actually unsigned short, and setting the type incorrectly
causes the expected errors to not be found.
This patch sets the type correctly (to unsigned short) for AIX.
Differential Revision: https://reviews.llvm.org/D110428
This simple change addresses a special case of structure/pointer
aliasing that produced different symbolvals, leading to false positives
during analysis.
The reproducer is as simple as this.
```lang=C++
struct s {
int v;
};
void foo(struct s *ps) {
struct s ss = *ps;
clang_analyzer_dump(ss.v); // reg_$1<int Element{SymRegion{reg_$0<struct s *ps>},0 S64b,struct s}.v>
clang_analyzer_dump(ps->v); //reg_$3<int SymRegion{reg_$0<struct s *ps>}.v>
clang_analyzer_eval(ss.v == ps->v); // UNKNOWN
}
```
Acks: Many thanks to @steakhal and @martong for the group debug session.
Reviewed By: steakhal, martong
Differential Revision: https://reviews.llvm.org/D110625
The builtin for vec_orc has support for the following two signatures,
but currently the compiler marks it ambiguous:
vector float vec_orc(vector float, vector float)
vector double vec_orc(vector double, vector double)
This patch implements these two builtins.
Differential revision: https://reviews.llvm.org/D110858
This also removes the need to disable the mandatory inlining phase in
tests.
In a departure from the previous remark, we don't output a 'cost' in
this case, because there's no such thing. We just report that inlining
happened because of the attribute.
Differential Revision: https://reviews.llvm.org/D110891
This patch allows the use of __vector_quad and __vector_pair, PPC MMA builtin
types, on all PowerPC 64-bit compilation units. When these types are
made available the builtins that use them automatically become available
so semantic checking for mma and pair vector memop __builtins is also
expanded to ensure these builtin function call are only allowed on
Power10 and new architectures. All related test cases are updated to
ensure test coverage.
Reviewed By: #powerpc, nemanjai
Differential Revision: https://reviews.llvm.org/D109599
Modify the IfStmt node to suppoort constant evaluated expressions.
Add a new ExpressionEvaluationContext::ImmediateFunctionContext to
keep track of immediate function contexts.
This proved easier/better/probably more efficient than walking the AST
backward as it allows diagnosing nested if consteval statements.
We keep a map from function name to source location so we don't have to
do it via looking up a source location from the AST. However, since
function names can be long, we actually use a hash of the function name
as the key.
Additionally, we can't rely on Clang's printing of function names via
the AST, so we just demangle the name instead.
This is necessary to implement
https://lists.llvm.org/pipermail/cfe-dev/2021-September/068930.html.
Reviewed By: dblaikie
Differential Revision: https://reviews.llvm.org/D110665
Per the GCC info page:
If the function is declared 'extern', then this definition of the
function is used only for inlining. In no case is the function
compiled as a standalone function, not even if you take its address
explicitly. Such an address becomes an external reference, as if
you had only declared the function, and had not defined it.
Respect that behavior for inline builtins: keep the original definition, and
generate a copy of the declaration suffixed by '.inline' that's only referenced
in direct call.
This fixes holes in c3717b6858d32d64514a187ede1a77be8ba4e542.
Differential Revision: https://reviews.llvm.org/D111009
The builtins: `__compare_and_swaplp`, `__fetch_and_addlp`,
` __fetch_and_andlp`, `__fetch_and_orlp`, `__fetch_and_swaplp` are
64 bit only. This patch ensures the compiler produces an error in 32 bit mode.
Reviewed By: #powerpc, nemanjai
Differential Revision: https://reviews.llvm.org/D110824
by Raul Penacoba.
The size of kmp_depend_info and the number of dependencies are computed multiplying the iterator sizes, which not right.
Now size is computed as:
itersize1*numclausedeps1 + itersize2*numclausedeps2 + ... + itersizeN*numclausedepsN
where itersizeX is the size of the iterator and numclausedepsX the number of dependencies in that depend clause.
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D111045
This patch fixes the return value of the builtin __builtin_ppc_load2r to
correctly return short instead of int.
Reviewed By: nemanjai, #powerpc
Differential Revision: https://reviews.llvm.org/D110771
Looks like exceptions are off-by-default with the PS4 triple.
Since adding -fexceptions defeats the purpose of the test change
in 8dfbe9b0a, pass an explicit triple instead.
These builtins produce inefficient code for CPU's prior to Power8
due to vcmpequd being unavailable. The predicate forms can actually
leverage the available vcmpequw along with xxlxor to produce a better
sequence.
The signatures for the PowerPC builtins lharx and
lbarx are incorrect, and causes issues when used in a function
that requires the return of the builtin to be promoted.
This patch fixes these signatures.
Differential revision: https://reviews.llvm.org/D110273
This consists of 3 compressed instructions, c.not, c.neg, and c.zext.w.
I believe these have been picked up by the Zce effort using different
encodings. I don't think it makes sense to keep them in bitmanip. It
will eventually cause a conflict if/when Zce is implemented in llvm.
Differential Revision: https://reviews.llvm.org/D110871
A followup to D110201.
For example, we'd set OptimizationRemarkMissed's Regex to '.*' when
encountering -Rpass. Normally this doesn't actually affect remarks we
emit because in clang::ProcessWarningOptions() we'll separately look at
all -R arguments and turn on/off corresponding diagnostic groups.
However, this is reproducible with -round-trip-args.
Reviewed By: JamesNagurne
Differential Revision: https://reviews.llvm.org/D110673
When clang crashes, it writes a standalone source file and shell script
to reproduce the crash.
The Driver used to set `Mode = CPPMode` in generateCompilationDiagnostics()
to force preprocessing mode. This has the side effect of making
IsCLMode() return false, which in turn meant Clang::AddClangCLArgs()
didn't get called when creating the standalone source file, which meant
the stand-alone file was preprocessed with the gcc driver's defaults
In particular, exceptions default to on with the gcc driver, but to
off with the cl driver. The .sh script did use the original command
line, so in the reproducer for a clang-cl crash, the standalone source
file could contain exception-using code after preprocessing that the
compiler invocation in the shell script would then complain about.
This patch removes the `Mode = CPPMode;` line and instead additionally
checks for `CCGenDiagnostics` in most places that check `CCCIsCPP().
This also matches the strategy Clang::ConstructJob() uses to add
-frewrite-includes for creating the standalone source file for a crash
report.
Fixes PR52007.
Differential Revision: https://reviews.llvm.org/D110783
Try to address Windows flakes from d87bdc272ba47b7d9109ff5c7191454ab2ae6fcb
by adding "|| true" as suggested in D110276 so the whole test doesn't
fail when Windows thinks it can't remove the binary.
There is a special situation with templates in local classes,
as can be seen in this example with generic lambdas in function scope:
```
template<class T1> void foo() {
(void)[]<class T2>() {
struct S {
void bar() { (void)[]<class T3>(T2) {}; }
};
};
};
template void foo<int>();
```
As a consequence of the resolution of DR1484, bar is instantiated during the
substitution of foo, and in this context we would substitute the lambda within
it with it's own parameters "injected" (turned into arguments).
This can't be properly dealt with for at least a couple of reasons:
* The 'TemplateTypeParm' type itself can only deal with canonical replacement
types, which the injected arguments are not.
* If T3 were constrained in the example above, our (non-conforming) eager
substitution of type constraints would just leave that parameter dangling.
Instead of substituting with injected parameters, this patch just leaves those
inner levels unreplaced.
Since injected arguments appear to be unused within the users of
`getTemplateInstantiationArgs`, this patch just removes that support there and
leaves a couple of asserts in place.
Signed-off-by: Matheus Izvekov <mizvekov@gmail.com>
Reviewed By: rsmith
Differential Revision: https://reviews.llvm.org/D110727
