dependent context and can't be used in a constant expression.
Per C++ [temp.inst]p2, "the instantiation of a static data member does not
occur unless the static data member is used in a way that requires the
definition to exist".
This doesn't /quite/ match that, as we still instantiate static data members
that are usable in constant expressions even if the use doesn't require a
definition. A followup patch will fix that for both variables and functions.
llvm-svn: 291295
Aleksey Shlypanikov pointed out my mistake in migrating an explicit
unique_ptr to auto - I was expecting the function returned a unique_ptr,
but instead it returned a raw pointer - introducing a leak.
Thanks Aleksey!
This reapplies r291184, reverted in r291249.
llvm-svn: 291270
FunctionDecls.
This commit silences an incorrect warning that is issued when a function
pointer is cast to another function pointer type. The warning gets
issued because alignments of the source and destination do not match in
Sema::CheckCastAlign, which happens because ASTContext::getTypeInfoImpl
and ASTContext::getDeclAlign return different values for functions (the
former returns 4 while the latter returns 1).
This should fix PR31558.
rdar://problem/29533528
Differential Revision: https://reviews.llvm.org/D27478
llvm-svn: 291253
Summary:
This patch makes the type_mismatch static data 7 bytes smaller (and it
ends up being 16 bytes smaller due to alignment restrictions, at least
on some x86-64 environments).
It revs up the type_mismatch handler version since we're breaking binary
compatibility. I will soon post a patch for the compiler-rt side.
Reviewers: rsmith, kcc, vitalybuka, pgousseau, gbedwell
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D28242
llvm-svn: 291236
Now when completing blocks properties that return void the block call completion
result shows up before the setter, otherwise the setter completion shows up
before the block call completion. We normally want to use the result of the
block call, so one typically wouldn't call a block that returns a non-void type
in a standalone statement.
rdar://28846153
Differential Revision: https://reviews.llvm.org/D26034
llvm-svn: 291232
This change avoids the -Wstrict-prototypes warning for block literals with an
empty argument list or without argument lists.
rdar://15060615
Differential Revision: https://reviews.llvm.org/D28296
llvm-svn: 291231
Add builtins for the functions and custom codegen mapping the builtins to their
corresponding intrinsics and handling the endian related swapping.
https://reviews.llvm.org/D26546
llvm-svn: 291179
Summary:
MSVC seems to use "__in" and "__out" for its own purposes, so we have to
pick different names in this macro.
Reviewers: tra
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D28325
llvm-svn: 291138
Summary:
When compiling device code, we may still see host code with explicit
calling conventions. NVPTX needs to claim that it supports these CCs,
so that (a) we don't raise noisy warnings, and (b) we don't break
existing code which relies on the existence of these CCs when
specializing templates. (If a CC doesn't exist, clang ignores it, so
two template specializations which are different only insofar as one
specifies a CC are considered identical and therefore are an error if
that CC is not supported.)
Reviewers: tra
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D28323
llvm-svn: 291136
Summary:
CUDA lets users share structs between the host and device, so for that
and other reasons, primitive types such as ptrdiff_t should be the same
on both sides of the compilation.
Our code to do this wasn't entirely successful. In particular, we did a
bunch of work during the NVPTXTargetInfo constructor, only to override
it in the NVPTX{32,64}TargetInfo constructors. It worked well enough on
Linux and Mac, but Windows is LLP64, which is different enough to break
it.
This patch removes the NVPTX{32,64}TargetInfo classes entirely and fixes
the bug described above.
Reviewers: tra
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D28322
llvm-svn: 291135
Summary:
For the most part this is straightforward: Just add a CudaInstallation
object to the MSVC and MinGW toolchains.
CudaToolChain has to override computeMSVCVersion so that
Clang::constructJob passes the right version flag to cc1. We have to
modify IsWindowsMSVC and friends in Clang::constructJob to be true when
compiling CUDA device code on Windows for the same reason.
Depends on: D28319
Reviewers: tra
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D28320
llvm-svn: 291131
Summary:
Previously it was taking the true target triple, which is not really
what it needs: The location of the CUDA installation depends on the host
OS.
Reviewers: tra
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D28319
llvm-svn: 291130
Summary:
These duplicate declarations cause a problem for CUDA compiles on
Windows. All implicitly-defined functions are host+device, and this
applies to the declarations in Builtin.def. But then when we see the
declarations in intrin.h, they have no attributes, so are host-only
functions. This is an error.
(A better fix might be to make these builtins host-only, but that is a
much bigger change.)
Reviewers: rnk
Subscribers: cfe-commits, echristo
Differential Revision: https://reviews.llvm.org/D28317
llvm-svn: 291128
Summary:
This patch adds two fields to the offload entry descriptor. One field is meant to signal Ctors/Dtors and `link` global variables, and the other is reserved for runtime library use.
Currently, these fields are only filled with zeros in the current code generation, but that will change when `declare target` is added.
The reason, we are adding these fields now is to make the code generation consistent with the runtime library proposal under review in https://reviews.llvm.org/D14031.
Reviewers: ABataev, hfinkel, carlo.bertolli, kkwli0, arpith-jacob, Hahnfeld
Subscribers: cfe-commits, caomhin, jholewinski
Differential Revision: https://reviews.llvm.org/D28298
llvm-svn: 291124
inline assembly may use the `.include` directive to include other
content into the file. Without the integrated assembler, the `-I` group
gets passed to the assembler. Emulate this by collecting the header
search paths and passing them to the IAS.
Resolves PR24811!
llvm-svn: 291123
This patch includes updates for codegen of the target region for the NVPTX
device. It moves initializers from the compiler to the runtime and updates
the worker loop to assume parallel work is retrieved from the runtime. A
subsequent patch will update the codegen to retrieve the parallel work using
calls to the runtime. It includes the removal of the inline attribute
for the worker loop and disabling debug info in it.
This allows codegen for a target directive and serial execution on the
NVPTX device.
Reviewers: ABataev
Differential Revision: https://reviews.llvm.org/D28125
llvm-svn: 291121
Summary:
Authored by Senthil Kumar Selvaraj
This patch adds barebones support in Clang for the (experimental) AVR target. It uses the integrated assembler for assembly, and the GNU linker for linking, as lld doesn't know about the target yet.
The DataLayout string is the same as the one in AVRTargetMachine.cpp. The alignment specs look wrong to me, as it's an 8 bit target and all types only need 8 bit alignment. Clang failed with a datalayout mismatch error when I tried to change it, so I left it that way for now.
Reviewers: rsmith, dylanmckay, cfe-commits, rengolin
Subscribers: rengolin, jroelofs, wdng
Differential Revision: https://reviews.llvm.org/D27123
llvm-svn: 291082
We were previously incorrectly using TDK_TooFewArguments to report a template
argument list that's too short, but it actually means that the number of
arguments in a top-level function call was insufficient. When diagnosing the
problem, SemaOverload would (rightly) assert that the failure kind didn't make
any sense.
llvm-svn: 291064
Summary:
Replace some old code that probably pre-dated the change to delay
emission of dllexported code until after the closing brace of the
outermost record type. Only uninstantiated default argument expressions
need to be handled now. It is enough to instantiate default argument
expressions when instantiating dllexported default ctors. This also
fixes some double-diagnostic issues in this area.
Fixes PR31500
Reviewers: rsmith
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D28274
llvm-svn: 291045
Front end component (back end changes are D27392). The vectorcall
calling convention was broken subtly in two cases. First,
it didn't properly handle homogeneous vector aggregates (HVAs).
Second, the vectorcall specification requires that only the
first 6 parameters be eligible for register assignment.
This patch fixes both issues.
Differential Revision: https://reviews.llvm.org/D27529
llvm-svn: 291041
Most code paths would already bail out in this case, but certain paths,
particularly overload resolution and typo correction, would not. Carrying on
with an invalid declaration could in some cases result in crashes due to
downstream code relying on declaration invariants that are not necessarily
met for invalid declarations, and in other cases just resulted in undesirable
follow-on diagnostics.
llvm-svn: 291030
Without this, we drop everything after the first late-parsed attribute
in a single __attribute__. (Where "drop" means "stuff everything into
LA->Toks.")
llvm-svn: 291020
The one use of CheckerManager (AnalysisConsumer, calling
createCheckerManager) keeps a strong reference to the AnalysisOptions
anyway, so this ownership wasn't necessary.
(I'm not even sure AnalysisOptions needs ref counting at all - but
that's more involved)
llvm-svn: 291017
Previously, if an overloaded function in a braced-init-list was encountered in
template argument deduction, and the overload set couldn't be resolved to a
particular function, we'd immediately produce a deduction failure. That's not
correct; this situation is supposed to result in that particular P/A pair being
treated as a non-deduced context, and deduction can still succeed if the type
can be deduced from elsewhere.
llvm-svn: 291014
This patch includes updates for codegen of the target region for the NVPTX
device. It moves initializers from the compiler to the runtime and updates
the worker loop to assume parallel work is retrieved from the runtime. A
subsequent patch will update the codegen to retrieve the parallel work using
calls to the runtime. It includes the removal of the inline attribute
for the worker loop and disabling debug info in it.
This allows codegen for a target directive and serial execution on the
NVPTX device.
Reviewers: ABataev
Differential Revision: https://reviews.llvm.org/D28125
llvm-svn: 290983
CUDA-8.0 comes with new headers which nvcc pre-includes via cuda_runtime.h
Clang now makes them available as well.
Differential Revision: https://reviews.llvm.org/D28301
llvm-svn: 290982
in non-void functions that fall off at the end without returning a value when
compiling C++.
Clang uses the new compiler flag to determine when it should treat control flow
paths that fall off the end of a non-void function as unreachable. If
-fno-strict-return is on, the code generator emits the ureachable and trap
IR only when the function returns either a record type with a non-trivial
destructor or another non-trivially copyable type.
The primary goal of this flag is to avoid treating falling off the end of a
non-void function as undefined behaviour. The burden of undefined behaviour
is placed on the caller instead: if the caller ignores the returned value then
the undefined behaviour is avoided. This kind of behaviour is useful in
several cases, e.g. when compiling C code in C++ mode.
rdar://13102603
Differential Revision: https://reviews.llvm.org/D27163
llvm-svn: 290960
Summary:
Before:
declare function foo();
let x = 1;
After:
declare function foo();
let x = 1;
The problem was that clang-format would unconditionally try to parse a child block, even though ambient function declarations do not have a body (similar to forward declarations).
Reviewers: djasper
Subscribers: cfe-commits, klimek
Differential Revision: https://reviews.llvm.org/D28246
llvm-svn: 290959
Previously, if the arguments for a parameter pack contained a braced-init-list,
we would abort deduction (keeping the pack deductions from prior arguments) at
the point when we reached the braced-init-list, resulting in wrong deductions
and rejects-valids. We now just leave a "hole" in the pack for such an argument,
which needs to be filled by another deduction of the same pack.
llvm-svn: 290933
Chromium is starting to use clang-format on more JavaScript.
In doing this, we discovered that our defaults were not doing a good job
differentiating between JS and C++.
This change moves some defaults to only apply to C++.
https://reviews.llvm.org/D28165
Patch from Dan Beam <dbeam@chromium.org>!
llvm-svn: 290930
When a parameter pack has multiple corresponding arguments, and some subset of
them are overloaded functions, it's possible that some subset of the parameters
are non-deduced contexts. In such a case, keep deducing from the remainder of
the arguments, and resolve the incomplete pack against whatever other
deductions we've performed for the pack.
GCC, MSVC, and ICC give three different bad behaviors for this case; what we do
now (and what we did before) don't exactly match any of them, sadly :( I'm
getting a core issue opened to specify more precisely how this should be
handled.
llvm-svn: 290923
Previously, -Wtautological-overlap-compare did not warn on cases where the
boolean expression was in an assignment or return statement. This patch
should cause all boolean statements to be passed to the tautological compare
checks in the CFG analysis.
This is one of the issues from PR13101
llvm-svn: 290920
The special case to widen the integer literal zero when passed to
variadic function calls should only apply to variadic functions, not
unprototyped functions. This is consistent with what MSVC does. In this
test case, MSVC uses a 4-byte store to pass the 5th argument to 'kr' and
an 8-byte store to pass the zero to 'v':
void v(int, ...);
void kr();
void f(void) {
v(1, 2, 3, 4, 0);
kr(1, 2, 3, 4, 0);
}
Aaron Ballman discovered this issue in https://reviews.llvm.org/D28166
llvm-svn: 290906
This patch cleans up private methods for NVPTX OpenMP codegen. It converts private
members to static functions to follow the coding style of CGOpenMPRuntime.cpp and
declutter the header file.
Reviewers: ABataev
Differential Revision: https://reviews.llvm.org/D28124
llvm-svn: 290904
https://reviews.llvm.org/D17840
This patch enables private, firstprivate, and lastprivate clauses for the OpenMP distribute directive.
Regression tests differ from the similar case of the same clauses on the for directive, by removing a reference to two global variables g and g1. This is necessary because: 1. a distribute pragma is only allowed inside a target region; 2. referring a global variable (e.g. g and g1) in a target region requires the program to enclose the variable in a "declare target" region; 3. declare target pragmas, which are used to define a declare target region, are currently unavailable in clang (patch being prepared).
For this reason, I moved the global declarations into local variables.
llvm-svn: 290898
Add a field indicating the associated check for every replacement to the YAML
report generated with the '-export-fixes' option. Update
clang-apply-replacements to handle the new format.
Patch by Alpha Abdoulaye!
Differential revision: https://reviews.llvm.org/D26137
llvm-svn: 290892
This commit fixes a crash that occurs when -print-decl-contexts AST consumer
tries to print an unhandled declaration.
rdar://19467234
Differential Revision: https://reviews.llvm.org/D26964
llvm-svn: 290887
This commit fixes a crash that occurs when -print-decl-contexts AST consumer
tries to print an unhandled declaration.
rdar://19467234
Differential Revision: https://reviews.llvm.org/D26964
llvm-svn: 290886
This commit fixes a crash that occurs when -print-decl-contexts AST consumer
tries to print an unhandled declaration.
rdar://19467234
Differential Revision: https://reviews.llvm.org/D26964
llvm-svn: 290885
This commit fixes a crash that occurs when -print-decl-contexts AST consumer
tries to print an unhandled declaration.
rdar://19467234
Differential Revision: https://reviews.llvm.org/D26964
llvm-svn: 290884
This commit fixes a crash that occurs when -print-decl-contexts AST consumer
tries to print an unhandled declaration.
rdar://19467234
Differential Revision: https://reviews.llvm.org/D26964
llvm-svn: 290882
This commit fixes a crash that occurs when -print-decl-contexts AST consumer
tries to print an unhandled declaration.
rdar://19467234
Differential Revision: https://reviews.llvm.org/D26964
llvm-svn: 290881
This commit fixes a crash that occurs when -print-decl-contexts AST consumer
tries to print an unhandled declaration.
rdar://19467234
Differential Revision: https://reviews.llvm.org/D26964
llvm-svn: 290880
This patch is to implement sema and parsing for 'target teams distribute parallel for simd’ pragma.
Differential Revision: https://reviews.llvm.org/D28202
llvm-svn: 290862
In many translation units I have tried, the calls to isIgnored() removed
in this patch are more expensive than doing the analysis that is behind
it. The speed-up in translation units I have tried is between 10 and
20%.
Review: https://reviews.llvm.org/D28208
llvm-svn: 290842
This reverts commit r290808, as it broken all ARM and AArch64 test-suite
test: MultiSource/UnitTests/C++11/frame_layout
Also, please, next time, try to write a commit message in according to
our guidelines:
http://llvm.org/docs/DeveloperPolicy.html#commit-messages
llvm-svn: 290811
template parameters between deduction and substitution. The idea is to accept
as many cases as possible, on the basis that substitution failure outside
the immediate context is much more common during substitution than during
implicit conversion sequence formation.
This does not implement the partial ordering portion of DR1391, which so
far appears to be misguided.
llvm-svn: 290808
The MS ABI RTTI has a reserved field which is used as a cache for the
demangled name. It must be zero-initialized, which is used as a hint by
the runtime to say that the cache has not been populated. Since this
field is populated at runtime, the RTTI structures must be placed in the
.data section rather than .rdata. NFC
llvm-svn: 290799
to be specified for a template template parameter whenever the parameter is at
least as specialized as the argument (when there's an obvious and correct
mapping from uses of the parameter to uses of the argument). For example, a
template with more parameters can be passed to a template template parameter
with fewer, if those trailing parameters have default arguments.
This is disabled by default, despite being a DR resolution, as it's fairly
broken in its current state: there are no partial ordering rules to cope with
template template parameters that have different parameter lists, meaning that
code that attempts to decompose template-ids based on arity can hit unavoidable
ambiguity issues.
The diagnostics produced on a non-matching argument are also pretty bad right
now, but I aim to improve them in a subsequent commit.
llvm-svn: 290792
This patch is to implement sema and parsing for 'target teams distribute parallel for’ pragma.
Differential Revision: https://reviews.llvm.org/D28160
llvm-svn: 290725
Summary:
We can simply import all external values with summaries included in
the individual index file created for the distributed backend job,
as only those are added to the individual index file created by the
WriteIndexesThinBackend (in addition to summaries for the original
module, which are skipped here).
While computing the cross module imports on this index would come to
the same conclusion as the original thin link import logic, it is
unnecessary work. And when tuning, it avoids the need to pass the
same function importing parameters (e.g. -import-instr-limit) to
both the thin link and the backends (otherwise they won't make the
same decisions).
Reviewers: mehdi_amini, pcc
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D28139
llvm-svn: 290674
Our newly aggressive constant folding logic makes it possible for
CGExprConstant to see the same CompoundLiteralExpr more than once. So,
emitting a new GlobalVariable every time we see a CompoundLiteral is no
longer correct.
We had a similar issue with BlockExprs that was caught while testing
said aggressive folding, so I applied the same style of fix (see D26410)
here. If we find yet another case where this needs to happen, we should
probably refactor this so we don't have a third DenseMap+getter+setter.
As a design note: getAddrOfConstantCompoundLiteralIfEmitted is really
only intended to be called by ConstExprEmitter::EmitLValue. So,
returning a GlobalVariable* instead of a ConstantAddress costs us
effectively nothing, and saves us either a few bytes per entry in our
map or a bit of code duplication.
llvm-svn: 290661
to make reference to template parameters. This is only a partial
implementation; we retain the restriction that the argument must not be
type-dependent, since it's unclear how that would work given the existence of
other language rules requiring an exact type match in this context, even for
type-dependent cases (a question has been raised on the core reflector).
llvm-svn: 290647
This seems like it must've been a leftover by accident - no tests were
backing it up & it doesn't make much sense to include size/alignment on
class declarations (it'd only be on those declarations for which the
definition was available - otherwise the size/alignment would not be
known).
llvm-svn: 290631
Added \n commands to insert a line breaks where necessary, since one long line of documentation is nearly unreadable.
Formatted comments to fit into 80 chars.
In some cases added \a command in front of the parameter names to display them in italics.
llvm-svn: 290619
specialized than the primary template. (Put another way, if we imagine there
were a partial specialization matching the primary template, we should never
select it if some other partial specialization also matches.)
llvm-svn: 290593
template parameters of reference type basically doesn't work, because we're
always deducing from an argument expression of non-reference type, so the type
of the deduced expression never matches. Instead, compare the type of an
expression naming the parameter to the type of the argument.
llvm-svn: 290586
Windows uses PE/COFF which is inherently position independent. The use
of the PIC model is unnecessary. In fact, we would generate invalid
code using the ELF PIC model when PIC was enabled previously. Now that
we no longer accept -fPIC and -fpoc, this switches the internal
representation to the static model to permit us to make PIC modules
invalid when targeting Windows. This should not change the code
generation, only the internal state management.
llvm-svn: 290569
dependent contexts when processing the template in C++11 and C++14, just like
we do in C++98 and C++1z. This allows us to diagnose invalid templates earlier.
llvm-svn: 290567
Improved doxygen comments for the following intrinsics headers: __wmmintrin_pclmul.h, bmiintrin.h, emmintrin.h, f16cintrin.h, immintrin.h, mmintrin.h, pmmintrin.h, tmmintrin.h
Added \n commands to insert a line breaks where necessary, since one long line of documentation is nearly unreadable.
Formatted comments to fit into 80 chars.
In some cases added \a command in front of the parameter names to display them in italics.
llvm-svn: 290561
This is kind of funny because I specifically did work to make this easy
and then it didn't actually get implemented.
I've also ported a set of tests that rely on this functionality to run
with the new PM as well as the old PM so that we don't mess this up in
the future.
llvm-svn: 290558
According to extended asm syntax, a case where the clobber list includes a variable from the inputs or outputs should be an error - conflict.
for example:
const long double a = 0.0;
int main()
{
char b;
double t1 = a;
__asm__ ("fucompp": "=a" (b) : "u" (t1), "t" (t1) : "cc", "st", "st(1)");
return 0;
}
This should conflict with the output - t1 which is st, and st which is st aswell.
The patch fixes it.
Commit on behald of Ziv Izhar.
Differential Revision: https://reviews.llvm.org/D15075
llvm-svn: 290539
Use of these flags would result in the use of ELF-style PIE/PIC code
which is incorrect on Windows. Windows is inherently PIC by means of
the DLL slide that occurs at load. This also mirrors the behaviour on
GCC for MinGW. Currently, the Windows x86_64 forces the relocation
model to PIC (Level 2). This is unchanged for now, though we should
remove any assumptions on that and change it to a static relocation
model.
llvm-svn: 290533
non-type template parameters.
During partial ordering, when checking the substituted deduced template
arguments match the original, check the types of non-type template arguments
match even if they're dependent. The only way we get dependent types here is if
they really represent types of the other template (which are supposed to be
modeled as being substituted for unique, non-dependent types).
In order to make this work for auto-typed non-type template arguments, we need
to be able to perform auto deduction even when the initializer and
(potentially) the auto type are dependent, support for which is the bulk of
this patch. (Note that this requires the ability to deduce only a single level
of a multi-level dependent type.)
llvm-svn: 290511
This patch is to implement sema and parsing for 'target teams distribute' pragma.
Differential Revision: https://reviews.llvm.org/D28015
llvm-svn: 290508
The patch also simplifies an assume of a constraint of the form: "(exp comparison_op expr) != 0" to true into an assume of "exp comparison_op expr" to true. (And similarly, an assume of the form "(exp comparison_op expr) == 0" to true as an assume of exp comparison_op expr to false.) which improves precision overall.
https://reviews.llvm.org/D22862
llvm-svn: 290505
Summary:
I needed to know whether a FieldDecl had an in-class
initializer for D26453. I used a narrowing matcher there, but a
traversal matcher might be generally useful.
Reviewers: sbenza, bkramer, klimek, aaron.ballman
Subscribers: aaron.ballman, Prazek, cfe-commits
Differential Revision: https://reviews.llvm.org/D28034
llvm-svn: 290492
template arguments as written rather than the canonical template arguments,
so we print more user-friendly names for template parameters.
llvm-svn: 290483
Added \n commands to insert a line breaks where necessary to make the documentation more readable.
Formatted comments to fit into 80 chars.
llvm-svn: 290458
Tagged parameter names with \a doxygen command to display parameters in italics.
Added \n commands to insert a line break to make the documentation more readable.
Formatted comments to fit into 80 chars.
llvm-svn: 290455
manager, and a code path to use it.
The option is actually a top-level option but does contain
'experimental' in the name. This is the compromise suggested by Richard
in discussions. We expect this option will be around long enough and
have enough users towards the end that it merits not being relegated to
CC1, but it still needs to be clear that this option will go away at
some point.
The backend code is a fresh codepath dedicated to handling the flow with
the new pass manager. This was also Richard's suggested code structuring
to essentially leave a clean path for development rather than carrying
complexity or idiosyncracies of how we do things just to share code with
the parts of this in common with the legacy pass manager. And it turns
out, not much is really in common even though we use the legacy pass
manager for codegen at this point.
I've switched a couple of tests to run with the new pass manager, and
they appear to work. There are still plenty of bugs that need squashing
(just with basic experiments I've found two already!) but they aren't in
this code, and the whole point is to expose the necessary hooks to start
experimenting with the pass manager in more realistic scenarios.
That said, I want to *strongly caution* anyone itching to play with
this: it is still *very shaky*. Several large components have not yet
been shaken down. For example I have bugs in both the always inliner and
inliner that I have already spotted and will be fixing independently.
Still, this is a fun milestone. =D
One thing not in this patch (but that might be very reasonable to add)
is some level of support for raw textual pass pipelines such as what
Sean had a patch for some time ago. I'm mostly interested in the more
traditional flow of getting the IR out of Clang and then running it
through opt, but I can see other use cases so someone may want to add
it.
And of course, *many* features are not yet supported!
- O1 is currently more like O2
- None of the sanitizers are wired up
- ObjC ARC optimizer isn't wired up
- ...
So plenty of stuff still lef to do!
Differential Revision: https://reviews.llvm.org/D28077
llvm-svn: 290450
Summary:
We compile user opencl kernel code with spir triple. But built-ins are written in OpenCL and we compile it with triple x86_64 to be able to use x86 intrinsics. And we need address spaces to match in both cases. So, we change fake address space map in OpenCL for matching with spir.
On CPU address spaces are not really important but we'd like to preserve address space information in order to perform optimizations relying on this info like enhanced alias analysis.
Reviewers: pekka.jaaskelainen, Anastasia
Subscribers: pekka.jaaskelainen, yaxunl, bader, cfe-commits
Differential Revision: https://reviews.llvm.org/D28048
llvm-svn: 290436
diagnostics and fix one such diagnostic.
Sadly, this assert doesn't catch this bug because we have no tests that
emit this diagnostic! Doh! I'm following up on the commit that
introduces it to get that fixed. Then this assert will help in a more
direct way.
llvm-svn: 290417
The patch also simplifies an assume of a constraint of the form: "(exp comparison_op expr) != 0" to true into an assume of "exp comparison_op expr" to true. (And similarly, an assume of the form "(exp comparison_op expr) == 0" to true as an assume of exp comparison_op expr to false.) which improves precision overall.
https://reviews.llvm.org/D22862
llvm-svn: 290413
twice, in finalization of template argument deduction.
This is a re-commit of r290310 (reverted in r290329); the bug found by the
buildbots was fixed in r290399 (we would sometimes build a deduced template
argument with a bogus type).
llvm-svn: 290403
fail the merge if the arguments have different types (except if one of them was
deduced from an array bound, in which case take the type from the other).
This is correct because (except in the array bound case) the type of the
template argument in each deduction must match the type of the parameter, so at
least one of the two deduced arguments must have a mismatched type.
This is necessary because we would otherwise lose the type information for the
discarded template argument in the merge, and fail to diagnose the mismatch.
In order to power this, we now properly retain the type of a deduced non-type
template argument deduced from a declaration, rather than giving it the type of
the template parameter; we'll convert it to the template parameter type when
checking the deduced arguments.
llvm-svn: 290399
-fno-inline-functions, -O0, and optnone.
These were really, really tangled together:
- We used the noinline LLVM attribute for -fno-inline
- But not for -fno-inline-functions (breaking LTO)
- But we did use it for -finline-hint-functions (yay, LTO is happy!)
- But we didn't for -O0 (LTO is sad yet again...)
- We had weird structuring of CodeGenOpts with both an inlining
enumeration and a boolean. They interacted in weird ways and
needlessly.
- A *lot* of set smashing went on with setting these, and then got worse
when we considered optnone and other inlining-effecting attributes.
- A bunch of inline affecting attributes were managed in a completely
different place from -fno-inline.
- Even with -fno-inline we failed to put the LLVM noinline attribute
onto many generated function definitions because they didn't show up
as AST-level functions.
- If you passed -O0 but -finline-functions we would run the normal
inliner pass in LLVM despite it being in the O0 pipeline, which really
doesn't make much sense.
- Lastly, we used things like '-fno-inline' to manipulate the pass
pipeline which forced the pass pipeline to be much more
parameterizable than it really needs to be. Instead we can *just* use
the optimization level to select a pipeline and control the rest via
attributes.
Sadly, this causes a bunch of churn in tests because we don't run the
optimizer in the tests and check the contents of attribute sets. It
would be awesome if attribute sets were a bit more FileCheck friendly,
but oh well.
I think this is a significant improvement and should remove the semantic
need to change what inliner pass we run in order to comply with the
requested inlining semantics by relying completely on attributes. It
also cleans up tho optnone and related handling a bit.
One unfortunate aspect of this is that for generating alwaysinline
routines like those in OpenMP we end up removing noinline and then
adding alwaysinline. I tried a bunch of other approaches, but because we
recompute function attributes from scratch and don't have a declaration
here I couldn't find anything substantially cleaner than this.
Differential Revision: https://reviews.llvm.org/D28053
llvm-svn: 290398
Much to my surprise, '-disable-llvm-optzns' which I thought was the
magical flag I wanted to get at the raw LLVM IR coming out of Clang
deosn't do that. It still runs some passes over the IR. I don't want
that, I really want the *raw* IR coming out of Clang and I strongly
suspect everyone else using it is in the same camp.
There is actually a flag that does what I want that I didn't know about
called '-disable-llvm-passes'. I suspect many others don't know about it
either. It both does what I want and is much simpler.
This removes the confusing version and makes that spelling of the flag
an alias for '-disable-llvm-passes'. I've also moved everything in Clang
to use the 'passes' spelling as it seems both more accurate (*all* LLVM
passes are disabled, not just optimizations) and much easier to remember
and spell correctly.
This is part of simplifying how Clang drives LLVM to make it cleaner to
wire up to the new pass manager.
Differential Revision: https://reviews.llvm.org/D28047
llvm-svn: 290392
program
Offload related code is not quite ready yet, but some simple examples
must not crash the compiler. Patch fixes the problem in offloading code
with exceptions.
llvm-svn: 290364
Update the GTestChecker to tighten up the API detection and make it
cleaner in response to post-commit feedback. Also add tests for when
temporary destructors are enabled to make sure we get the expected behavior
when inlining constructors for temporaries.
llvm-svn: 290352
Warnings with suppress-on-sink are discarded during FlushReports when
BugReporter notices that all paths in ExplodedGraph that pass through the
warning eventually run into a sink node.
However, suppress-on-sink fails to filter out false positives when the analysis
terminates too early - by running into analyzer limits, such as block count
limits or graph size limits - and the interruption hits the narrow window
between throwing the leak report and reaching the no-return function call. In
such case the report is there, however suppression-on-sink doesn't work, because
the sink node was never constructed in the incomplete ExplodedGraph.
This patch implements a very partial solution: also suppress reports thrown
against a statement-node that corresponds to a statement that belongs to a
no-return block of the CFG.
rdar://problem/28832541
Differential Revision: https://reviews.llvm.org/D28023
llvm-svn: 290341
Richard Smith