a switch or goto somewhere in the function. Indirect gotos trigger the
jump-checker regardless, because the conditions there are slightly more
elaborate and it's too marginal a case to be worth optimizing.
Turns off the jump-checker in a lot of cases in C++. rdar://problem/7702918
llvm-svn: 109962
have a "coerce to" type which often matches the default lowering of Clang
type to LLVM IR type, but the coerce case can be handled by making them
not be the same.
This simplifies things and fixes issues where X86-64 abi lowering would
return coerce after making preferred types exactly match up. This caused
us to compile:
typedef float v4f32 __attribute__((__vector_size__(16)));
v4f32 foo(v4f32 X) {
return X+X;
}
into this code at -O0:
define <4 x float> @foo(<4 x float> %X.coerce) nounwind {
entry:
%retval = alloca <4 x float>, align 16 ; <<4 x float>*> [#uses=2]
%coerce = alloca <4 x float>, align 16 ; <<4 x float>*> [#uses=2]
%X.addr = alloca <4 x float>, align 16 ; <<4 x float>*> [#uses=3]
store <4 x float> %X.coerce, <4 x float>* %coerce
%X = load <4 x float>* %coerce ; <<4 x float>> [#uses=1]
store <4 x float> %X, <4 x float>* %X.addr
%tmp = load <4 x float>* %X.addr ; <<4 x float>> [#uses=1]
%tmp1 = load <4 x float>* %X.addr ; <<4 x float>> [#uses=1]
%add = fadd <4 x float> %tmp, %tmp1 ; <<4 x float>> [#uses=1]
store <4 x float> %add, <4 x float>* %retval
%0 = load <4 x float>* %retval ; <<4 x float>> [#uses=1]
ret <4 x float> %0
}
Now we get:
define <4 x float> @foo(<4 x float> %X) nounwind {
entry:
%X.addr = alloca <4 x float>, align 16 ; <<4 x float>*> [#uses=3]
store <4 x float> %X, <4 x float>* %X.addr
%tmp = load <4 x float>* %X.addr ; <<4 x float>> [#uses=1]
%tmp1 = load <4 x float>* %X.addr ; <<4 x float>> [#uses=1]
%add = fadd <4 x float> %tmp, %tmp1 ; <<4 x float>> [#uses=1]
ret <4 x float> %add
}
This implements rdar://8248065
llvm-svn: 109733
Before we'd compile the example into something like:
%coerce.dive2 = getelementptr %struct.v4f32wrapper* %retval, i32 0, i32 0 ; <<4 x float>*> [#uses=1]
%1 = bitcast <4 x float>* %coerce.dive2 to <2 x double>* ; <<2 x double>*> [#uses=1]
%2 = load <2 x double>* %1, align 1 ; <<2 x double>> [#uses=1]
ret <2 x double> %2
Now we produce:
%coerce.dive2 = getelementptr %struct.v4f32wrapper* %retval, i32 0, i32 0 ; <<4 x float>*> [#uses=1]
%0 = load <4 x float>* %coerce.dive2, align 1 ; <<4 x float>> [#uses=1]
ret <4 x float> %0
llvm-svn: 109732
possible. This improves the example to pass <4 x float> instead of
<2 x double> but we still get awful code, and still don't get the
return value right.
llvm-svn: 109700
their call expressions synthetically have the "deduced" types based on their
first argument. We only insert conversions in the AST for arguments whose
values require conversion to match the value type expected. This keeps PR7600
closed by maintaining the return type, but avoids assertions due to unexpected
implicit casts making the type unsigned (test case added from Daniel).
The magic is moved into the codegen for the atomic builtin which inserts the
casts as needed at the IR level to raise the type to an integer suitable for
the LLVM intrinsic. This shouldn't cause any real change in functionality, but
now we can make the builtin be more truly polymorphic.
llvm-svn: 108638
- This issue here is that /usr/include/Blocks.h wants to define some of the
block runtime globals as weak, depending on the target. This doesn't work in
Clang because we aren't using the AST decl for these globals.
- The fix is a pretty gross hack which just watches all the decls for the
specific blocks globals we need to know about; if we see one we use it,
otherwise we use the hand coded type.
In time, I would like to clean this up by changing IRgen to ask Sema/AST for
the decl, which would then be lazily loaded from the builtin table if
necessary. This could be used in a whole host of places in IRgen and would
get rid of a lot of grotty hand coding of LLVM IR; however, we need some
extra Sema support for this as well as support for builtin global variables.
llvm-svn: 108482
strip cv-qualifiers from the expression's type when the language calls
for it: in C, that's all the time, while C++ only does it for
non-class types.
Centralized the computation of the call expression type in
QualType::getCallResultType() and some helper functions in other nodes
(FunctionDecl, ObjCMethodDecl, FunctionType), and updated all relevant
callers of getResultType() to getCallResultType().
Fixes PR7598 and PR7463, along with a bunch of getResultType() call
sites that weren't stripping references off the result type (nothing
stripped cv-qualifiers properly before this change).
llvm-svn: 108234
at -O0. The only change from the previous patch is that we don't try
to generate virtual method thunks for an available_externally
function.
llvm-svn: 108230
-O0, since we won't be using the definitions for anything anyway. For
lib/System/Path.o when built in Debug+Asserts mode, this leads to a 4%
improvement in compile time (and suppresses 440 function bodies).
<rdar://problem/7987644>
llvm-svn: 108156
was not producing a memcpy with the right address
spaces because of two places in it doing casts of
the arguments to i8, one of which that didn't
preserve the address space.
There is also an optimizer bug here.
llvm-svn: 107842
alloca for an argument. Make sure the argument gets the proper
decl alignment, which may be different than the type alignment.
This fixes PR7567
llvm-svn: 107627
r107173, "fix PR7519: after thrashing around and remembering how all this stuff"
r107216, "fix PR7523, which was caused by the ABI code calling ConvertType instead"
This includes a fix to make ConvertTypeForMem handle the "recursive" case, and call
it as such when lowering function types which have an indirect result.
llvm-svn: 107310
would trigger an extra method call).
- While in the area, I also changed Clang to not emit an unnecessary load from
'x' in cases like 'y = (x = 1)'.
llvm-svn: 107210
works, the fix is quite simple: just make sure to call ConvertTypeRecursive
when the function type being lowered is in the midst of ConvertType.
llvm-svn: 107173
load/store nonsense in the epilog. For example, for:
int foo(int X) {
int A[100];
return A[X];
}
we used to generate:
%arrayidx = getelementptr inbounds [100 x i32]* %A, i32 0, i64 %idxprom ; <i32*> [#uses=1]
%tmp1 = load i32* %arrayidx ; <i32> [#uses=1]
store i32 %tmp1, i32* %retval
%0 = load i32* %retval ; <i32> [#uses=1]
ret i32 %0
}
which codegen'd to this code:
_foo: ## @foo
## BB#0: ## %entry
subq $408, %rsp ## imm = 0x198
movl %edi, 400(%rsp)
movl 400(%rsp), %edi
movslq %edi, %rax
movl (%rsp,%rax,4), %edi
movl %edi, 404(%rsp)
movl 404(%rsp), %eax
addq $408, %rsp ## imm = 0x198
ret
Now we generate:
%arrayidx = getelementptr inbounds [100 x i32]* %A, i32 0, i64 %idxprom ; <i32*> [#uses=1]
%tmp1 = load i32* %arrayidx ; <i32> [#uses=1]
ret i32 %tmp1
}
and:
_foo: ## @foo
## BB#0: ## %entry
subq $408, %rsp ## imm = 0x198
movl %edi, 404(%rsp)
movl 404(%rsp), %edi
movslq %edi, %rax
movl (%rsp,%rax,4), %eax
addq $408, %rsp ## imm = 0x198
ret
This actually does matter, cutting out 2000 lines of IR from CGStmt.ll
for example.
Another interesting effect is that altivec.h functions which are dead
now get dce'd by the inliner. Hence all the changes to
builtins-ppc-altivec.c to ensure the calls aren't dead.
llvm-svn: 106970
As part of this, pull together trapv handling into the same enum.
This also add support for NSW multiplies.
This also makes PCH disagreement on overflow behavior silent, since it
really doesn't matter except for warnings and codegen (no macros get
defined etc).
llvm-svn: 106956
but whose operand isn't a float: specifically, __real__ and __imag__. Instead
of filtering these out, just implement them.
Fixes <rdar://problem/7958272>.
llvm-svn: 103307
they're unreachable. This matters because (if they're POD, or if this is C)
the scope containing the variable might be reachable even if the variable
isn't. Fixes PR7044.
llvm-svn: 103052
input and output types when the smaller value isn't mentioned in the
asm string. Extend this support from integers to also allowing
fp values to be mismatched (if not mentioned in the asm string).
llvm-svn: 102188
- This fixes the last known ABI issues with ARM/APCS.
- I've run the first 1k ABITests with '--no-unsigned --no-vector --no-complex'
on {armv6, armv7} x {-mno-thumb, -mthumb}, and the first 10k tests for armv7
-mthumb, for both function return types and single argument calls. These all
pass now (they failed horribly before without --no-bitfield).
llvm-svn: 102070
- This fixes some pedantic bugs with packed structures, as well as major problems with -fno-bitfield-type-align.
- Fixes PR5591, PR5567, and all known -fno-bitfield-type-align issues.
- Review appreciated.
llvm-svn: 102045
This mirror's Dan's patch for llvm-gcc in r97989, and
fixes the miscompilation in PR6525. There is some contention
over whether this is the right thing to do, but it is the
conservative answer and demonstrably fixes a miscompilation.
llvm-svn: 101877
struct may cause it to shrink more than one byte. Before
my recent changes we compiled the new test into:
%0 = type { [6 x i8] }
@x = global %0 { [6 x i8] undef }, align 2 ; <%0*> [#uses=0]
which is obviously bogus. Now we compile it into:
%0 = type <{ i32, i8, i8 }>
@x = global %0 zeroinitializer, align 2 ; <%0*> [#uses=0]
Where the last byte only is tail padding.
llvm-svn: 101536
- Sadly, this doesn't seem to give any .ll size win so far. It is possible to make this routine significantly smarter & avoid various shifting, masking, and zext/sext, but I'm not really convinced it is worth it. It is tricky, and this is really instcombine's job.
- No intended functionality change; the test case is just to increase coverage & serves as a demo file, it worked before this commit.
The new fixes from r101222 are:
1. The shift to the target position needs to occur after the value is extended to the correct size. This broke Clang bootstrap, among other things no doubt.
2. Swap the order of arguments to OR, to get a tad more constant folding.
llvm-svn: 101339
Stop multiplying constant by 8 accordingly in the header and change
intrinsic definition for what types we expect.
Add to existing palignr test to check that we're emitting the correct things.
llvm-svn: 101332
- Sadly, this doesn't seem to give any .ll size win so far. It is possible to make this routine significantly smarter & avoid various shifting, masking, and zext/sext, but I'm not really convinced it is worth it. It is tricky, and this is really instcombine's job.
- No intended functionality change; the test case is just to increase coverage & serves as a demo file, it worked before this commit.
llvm-svn: 101222
elements with explicit zero values instead of with tail padding.
On an example like this:
struct foo { int a; int b; };
struct foo fooarray[] = {
{1, 2},
{4},
};
We now lay this out as:
@fooarray = global [2 x %struct.foo] [%struct.foo { i32 1, i32 2 }, %struct.foo { i32 4, i32 0 }]
instead of as:
@fooarray = global %0 <{ %struct.foo { i32 1, i32 2 }, %1 { i32 4, [4 x i8] zeroinitializer } }>
Preserving both the struct type of the second element, but also the array type of the entire thing.
llvm-svn: 101155
trailing fields may not be represented in initializer lists, they
are being handled as padding and those fields *must* be zero
initialized.
llvm-svn: 101067
This introduces FunctionType::ExtInfo to hold the calling convention and the
noreturn attribute. The next patch will extend it to include the regparm
attribute and fix the bug.
llvm-svn: 99920
gcc, and the common expectation seems to be that they are unused. If and when
someone cares we can add them back with well documented demantics.
llvm-svn: 99522
expression computation in the wrong bit-width, and end up generating a totally
bogus array reference (_g0+8589934546).
- This showed up on Prolangs/cdecl.
llvm-svn: 99042
Clang's support for weakref is now better than llvm-gcc's :-)
We don't introduce a new symbol and we correctly mark undefined references weak only if there is no
definition or regular undefined references in the same file.
llvm-svn: 97733
a common source of oddities and, in theory, removes some redundant ABI
computations. Also fixes a miscompile I introduced yesterday by refactoring
some code and causing a slightly different code path to be taken that
didn't perform *parameter* type canonicalization, just normal type
canonicalization; this in turn caused a bit of ABI code to misfire because
it was looking for 'double' or 'float' but received 'const float'.
llvm-svn: 97030
of the block descriptor field. This field is the ObjC style @encode
signature of the implementation function, and was to this point
conditionally provided in the block literal data structure. That
provisional support is removed.
Additionally, eliminate unused enumerations for the block literal flags field.
The first shipping ABI unconditionally set (1<<29) but this bit is unused
by the runtime, so the second ABI will unconditionally have (1<<30) set so
that the runtime can in fact distinguish whether the additional data is
present or not.
llvm-svn: 96989
- This fixes many many more places than the test case, but my feeling is we need to audit alignment systematically so I'm not inclined to try hard to test the individual fixes in this patch. If this bothers you, patches welcome!
PR6240.
llvm-svn: 95648
follows (as conservatively as possible) gcc's current behavior: attributes
written on return types that don't apply there are applied to the function
instead, etc. Only parse CC attributes as type attributes, not as decl attributes;
don't accepet noreturn as a decl attribute on ValueDecls, either (it still
needs to apply to other decls, like blocks). Consistently consume CC/noreturn
information throughout codegen; enforce this by removing their default values
in CodeGenTypes::getFunctionInfo().
llvm-svn: 95436
WHAT!?!
It turns out that Type::isPromotableIntegerType() was not considering
enumeration types to be promotable, so we would never do the
promotion despite having properly computed the promotion type when the
enum was defined. Various operations on values of enum type just
"worked" because we could still compute the integer rank of an enum
type; the oddity, however, is that operations such as "add an enum and
an unsigned" would often have an enum result type (!). The bug
actually showed up as a spurious -Wformat diagnostic
(<rdar://problem/7595366>), but in theory it could cause miscompiles.
In this commit:
- Enum types with a promotion type of "int" or "unsigned int" are
promotable.
- Tweaked the computation of promotable types for enums
- For all of the ABIs, treat enum types the same way as their
underlying types (*not* their promotion types) for argument passing
and return values
- Extend the ABI tester with support for enumeration types
llvm-svn: 95117
by setting the section of the generated global. This is an
optimization done by the code generator, and the code being
removed didn't handle the case when the string contained an
embedded nul (which the code generator does correctly
handle). This is rdar://7589850
llvm-svn: 95003
the ABI spec, this turns out to simplify the code. We still have some annoying
code which mismatches the spec with regard to empty structures.
llvm-svn: 94796
only takes a boolean second argument now. Update tests accordingly.
Currently the builtin still accepts the full range for compatibility.
llvm-svn: 91983
- This is designed to make it obvious that %clang_cc1 is a "test variable"
which is substituted. It is '%clang_cc1' instead of '%clang -cc1' because it
can be useful to redefine what gets run as 'clang -cc1' (for example, to set
a default target).
llvm-svn: 91446
John McCall