cases. First, omit all builtin overloads when no non-record type is in the set
of candidate types. Second, avoid arithmetic type overloads for non-arithmetic
or enumeral types (counting vector types as arithmetic due to Clang
extensions). When heavily using constructs such as STL's '<<' based stream
logging, this can have a significant impact. One logging-heavy test case's
compile time dropped by 10% with this. Self-host shows 1-2% improvement in
compile time, but that's likely in the noise.
llvm-svn: 121665
a translation unit to the ActOnEndOfTranslationUnit function instead of doing
it at the start of DefineUsedVTables. The latter is now called *recursively*
during template instantiation, which causes an absolutely insane number of
walks of every record decl in the translation unit.
After this patch, an extremely template instantiation heavy test case's compile
time drops by 10x, and we see between 15% and 20% improvement in average
compile times across a project. This is just recovering a regression, it
doesn't make anything faster than it was several weeks ago.
llvm-svn: 121644
type. Localize all of the logic within a single function rather than spreading
it throughout the class.
Also fixes a buglet where we failed to check for a RHS arithmetic type wider
than the LHS and return its canonical type. I've yet to produce a test case
that breaks because of this, but it was spotted by inspection by folks on the
IRC channel and is obviously correct now.
llvm-svn: 121633
user-defined operator overloads on the same enumeral types to the one place
where it is used.
In theory this removes wasted computation from several paths through this code,
but I'm not aware of a case where it actually matters. This is mostly for
cleanliness.
llvm-svn: 121630
others have another ordering they would prefer, I'm all ears, but this one made
it much easier for me to find the group of operators I'm interested in.
llvm-svn: 121629
candidate set. This breaks apart a huge switch + goto system into distinct
methods on a class. It also places the current mess of tables and other static
state used in the process within that class.
This is still a work in progress. I did a few simplifications that jumped out
at me as I went, but I plan to iterate on this a bit before it's truly clean.
However, this is easily the most invasive chunk. I benchmarked it on
all-std-headers.cpp and an internal testcase that has a major hotspot in
overload resolution and saw no real performance impact.
llvm-svn: 121623
declared setter or getter in current class extension or one
of the other class extensions. Mark them as synthesized as
property will be synthesized when property with same name is
seen in the @implementation. This prevents bogus warning
about unimplemented methods to be issued for these methods.
Fixes // rdar://8747333
llvm-svn: 121597
Most Neon shift intrinsics do not have variants for polynomial types, but
vsri_n and vsli_n do support them, and we need to properly range-check the
shift immediates for them.
llvm-svn: 121509
We should not substitute template types if the template has a dependent
context because the template argument stack is not yet fully formed.
Instead, defer substitution until the template has a non-dependent
context (i.e. instantiation of an outer template).
llvm-svn: 121491
common base for ExtQuals and Type that stores the underlying type
pointer. This results in a 2% performance win for -emit-llvm on a
typical C file, with 1% memory growth in the AST.
Note that there is an API change in this optimization:
QualType::getTypePtr() can no longer be invoked on a NULL
QualType. If the QualType might be NULL, use
QualType::getTypePtrOrNull(). I've audited all uses of getTypePtr() in
the code base and changed the appropriate uses over to
getTypePtrOrNull().
A future optimization opportunity would be to distinguish between
cast/dyn_cast and cast_or_null/dyn_cast_or_null; for the former, we
could use getTypePtr() rather than getTypePtrOrNull(), to take another
branch out of the cast/dyn_cast implementation.
llvm-svn: 121489
space better. Remove this reference. To make that work, change some APIs
(most importantly, getDesugaredType()) to take an ASTContext& if they
need to return a QualType. Simultaneously, diminish the need to return a
QualType by introducing some useful APIs on SplitQualType, which is
just a std::pair<const Type *, Qualifiers>.
llvm-svn: 121478
increasing the required type alignment from 8 to 16. This provides a
2.5% speedup for -fsyntax-only on a token-cached Cocoa.h, while only
increasing memory consumption in the ASTContext by 0.8%.
llvm-svn: 121474
visibility. Fixes PR8713.
I've disabled a test which was testing that you can #pragma pop visibility
to get out of a namespace's visibility attribute. We should probably just
diagnose that as an error unless it's instrumental to someone's system
headers.
llvm-svn: 121459
global code completions are disabled (e.g., because they are
cached). Also, make sure that forward-declared protocols are visited
when we look for all visible names within a declaration context.
Previously, we would end up with duplicate completions for protocols.
llvm-svn: 121416
"inline", we weren't giving the definition weak linkage because the
"inline" bit wasn't propagated. This was a longstanding FIXME that,
somehow, hadn't triggered a bug in the wild. Fix this problem by
tracking whether any declaration was marked "inline", and clean up the
semantics of GNU's "extern inline" semantics calculation based on this
change.
Fixes <rdar://problem/8740363>.
llvm-svn: 121373
declaration that is a value in ill-formed code. Instead of crashing,
treat this as a dependent typename specifier and suggest that the
using add "typename" into the using declaration. Fixes <rdar://problem/8740998>.
llvm-svn: 121322
zextOrTrunc(), and APSInt methods extend(), extOrTrunc() and new method
trunc(), to be const and to return a new value instead of modifying the
object in place.
llvm-svn: 121121
cv-qualification conversions. More specifically, there's an implicit
cv-qualification conversion (even one that drops qualifiers) when
converting to 'id' or qualified 'id'. Fixes <rdar://problem/8734046>.
llvm-svn: 121047
FunctionArrayLvalueConversion but without the function/array decay. Generally
this is only appropriate for use sites that know the type of the expression
and thus that it can't be subject to the decays.
Also make sure we do lvalue-to-rvalue on the bases of ivar references.
llvm-svn: 121035
My previous attempt at solving the compile-time problem with many
redeclarations of the same entity cached both linkage and visibility,
while this patch only tackles linkage. There are several reasons for
this difference:
- Linkage is a language concept, and is evaluated many times during
semantic analysis and codegen, while visibility is only a
code-generation concept that is evaluated only once per (unique)
declaration. Hence, we *must* optimize linkage calculations but
don't need to optimize visibility computation.
- Once we know the linkage of a declaration, subsequent
redeclarations can't change that linkage. Hence, cache
invalidation is far simpler than for visibility, where a later
redeclaration can completely change the visibility.
- We have 3 spare bits in Decl to store the linkage cache, so the
cache doesn't increase the size of declarations. With the
visibility+linkage cache, NamedDecl got larger.
llvm-svn: 121023
and visibility of declarations, because it was extremely messy and it
increased the size of NamedDecl.
An improved implementation is forthcoming.
llvm-svn: 121012
so that's not a valid thing to do at all. Instead, switch to a ValueDecl
argument, the template isn't really necessary here.
When handling the types explicitly in the code, it becomes awkward to cerate
the CXXBaseOrMemberInitializer object in so many places. Re-flow the code to
calculate the Init expression first, and then create the initializer. If this
is too gross, we can factor the init expression logic into helper functions,
but it's not past my threshold yet.
llvm-svn: 120997
the LHS, or else the pointer might be invalid. This is kindof dumb, but
go ahead and make sure we're doing that for l-value scalar assignment,
which fixes a miscompile of obj-c++.dg/block-seq.mm.
Leave a FIXME for how to solve this problem for agg __blocks.
llvm-svn: 120992
build one when either of the operands calls itself type-dependent;
previously we were building when one of the operand types was dependent,
which is not always the same thing and which can lead to unfortunate
inconsistencies later. Fixes PR8739.
llvm-svn: 120990
Fix a bug in the emission of complex compound assignment l-values.
Introduce a method to emit an expression whose value isn't relevant.
Make that method evaluate its operand as an l-value if it is one.
Fixes our volatile compliance in C++.
llvm-svn: 120931
struct X {
X() : au_i1(123) {}
union {
int au_i1;
float au_f1;
};
};
clang will now deal with au_i1 explicitly as an IndirectFieldDecl.
llvm-svn: 120900
not actually frequently used, because ImpCastExprToType only creates a node
if the types differ. So explicitly create an ICE in the lvalue-to-rvalue
conversion code in DefaultFunctionArrayLvalueConversion() as well as several
other new places, and consistently deal with the consequences throughout the
compiler.
In addition, introduce a new cast kind for loading an ObjCProperty l-value,
and make sure we emit those nodes whenever an ObjCProperty l-value appears
that's not on the LHS of an assignment operator.
This breaks a couple of rewriter tests, which I've x-failed until future
development occurs on the rewriter.
Ted Kremenek kindly contributed the analyzer workarounds in this patch.
llvm-svn: 120890
Also, move the l-value emission code into CGObjC.cpp and teach it, for
completeness, to store away self for a super send.
Also, inline the super cases for property gets and sets and make them
use the correct result type for implicit getter/setter calls.
llvm-svn: 120887
This patch refactors the CompilerInvocation code to introduce a
CompilerInvocation::setLangDefaults function, which can set up a
LangOptions with the defaults for a given language and language
standard. This function is useful for non-command line based Clang
clients which need to set up a CompilerInvocation manually for a
specific language.
llvm-svn: 120874
declarations.
The motivation for this patch is that linkage/visibility computations
are linear in the number of redeclarations of an entity, and we've run
into a case where a single translation unit has > 6500 redeclarations
of the same (unused!) external variable. Since each redeclaration
involves a linkage check, the resulting quadratic behavior makes Clang
slow to a crawl. With this change, a simple test with 512
redeclarations of a variable syntax-checks ~20x faster than
before.
That said, I hate this change, and will probably end up reverting it
in a few hours. Reasons to hate it:
- It makes NamedDecl larger, since we don't have enough free bits in
Decl to squeeze in the extra information about caching.
- There are way too many places where we need to invalidate this
cache, because the visibility of a declaration can change due to
redeclarations (!). Despite self-hosting and passing the testsuite,
I have no confidence that I've found all of places where this cache
needs to be invalidated.
llvm-svn: 120808
a node in the trimmed graph might not always
correctly map back to the original error node.
This could cause a crash in some cases when
flagging memory leaks.
llvm-svn: 120795
about deprecated Objective-C pointer conversions. Plus, make sure to
actually set an appropriate AssignmentAction when performing an
implicit conversion from an InitializationSequence. Fixes regressions
in the GCC DejaGNU testsuite.
llvm-svn: 120744
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
a global is larger than 32 bytes and has fewer than 6 non-zero values in the
initializer. Previously we'd turn something like this:
char test8(int X) {
char str[10000] = "abc";
into a 10K global variable which we then memcpy'd from. Now we generate:
%str = alloca [10000 x i8], align 16
%tmp = getelementptr inbounds [10000 x i8]* %str, i64 0, i64 0
call void @llvm.memset.p0i8.i64(i8* %tmp, i8 0, i64 10000, i32 16, i1 false)
store i8 97, i8* %tmp, align 16
%0 = getelementptr [10000 x i8]* %str, i64 0, i64 1
store i8 98, i8* %0, align 1
%1 = getelementptr [10000 x i8]* %str, i64 0, i64 2
store i8 99, i8* %1, align 2
Which is much smaller in space and also likely faster.
This is part of PR279
llvm-svn: 120645
Check for compatible gcc, Altivec and Neon vectors before handling the
lax-vector-conversions case. Otherwise there is no way to avoid the
warnings from -Wvector-conversions.
llvm-svn: 120633
definition of an Objective-C class. Unlike with C/C++ classes, we
don't have a well-defined point in Sema where Objective-C classes are
checked for completeness, nor do we need to involve Sema when
completing a class. Therefore, we take the appropriate of having the
external AST source mark a particular Objective-C class as having an
external declaration; when using one of the accessors of an
Objective-C class that has an external declaration, we request that
the external AST source fill in the Objective-C class definition.
llvm-svn: 120627
conversions. Previously, we would end up collapsing qualification
conversions into the Objective-C pointer conversion step, including
(possibly) stripping qualifiers that shouldn't be removed.
This generalizes BuildSimilarlyQualifiedPointerType() to also work on
Objective-C object pointers, then eliminates the (redundant, not
totally correct) BuildSimilarlyQualifiedObjCObjectPointerType()
function.
Fixes <rdar://problem/8714395>.
llvm-svn: 120607
instantiations, GCC also supports "inline" and "static" explicit
template instantiations. Parse and warn about such constructs, but
don't implement the semantics of either "inline" or "static". They
don't seem to be widely used.
llvm-svn: 120599
disambiguate between an expression (for a bit-field width) and a type
(for a fixed underlying type). Since the disambiguation can be
expensive (due to tentative parsing), we perform a simplistic
disambiguation based on one-token lookahead before going into the
full-blown tentative parsing. Based on a patch by Daniel Wallin.
llvm-svn: 120582
a specific language. We are adding such language info. by
extensing Builtins.def and via a language flag added
to LIBBUILTIN/BUILTIN and check for that when deciding
a name is builtin or not. Implements //rdar://8689273.
llvm-svn: 120429
trap the serialized preprocessing records (macro definitions, macro
instantiations, macro definitions) from the generation of the
precompiled preamble, then replay those when walking the list of
preprocessed entities. This eliminates a bug where clang_getCursor()
wasn't able to find preprocessed-entity cursors in the preamble.
llvm-svn: 120396
precompiled preamble as the "main" source file's file ID within the
source manager. This makes compiling with a precompiled preamble
produce the same source locations as when compiling without the
precompiled preamble; prior to this change, we ended up with different
file IDs for source locations within the precompiled preamble
vs. those after the precompiled preamble, even for entities (e.g.,
preprocessing entities) in the same file.
llvm-svn: 120390
Chandler Carruth