In reality, some unaligned memory accesses are legal for 32-bit types and
smaller too, but it all depends on the address space. Allowing
unaligned loads/stores for > 32-bit types is mainly to prevent the
legalizer from splitting one load into multiple loads of smaller types.
https://bugs.freedesktop.org/show_bug.cgi?id=65873
llvm-svn: 184822
This should only make a difference in programs that use a lot of the
vector ALU instructions like BFI_INT and BIT_ALIGN. There is a slight
improvement in the phatk bitcoin mining kernel with this patch on
Evergreen (vector size == 1):
Before:
1173 Instruction Groups / 9520 dwords
After:
1167 Instruction Groups / 9510 dwords
Reviewed-by: Reviewed-by: Vincent Lejeune<vljn at ovi.com>
llvm-svn: 184819
darwin.
After talking with Jim Grosbach pretty extensively, he was OK with me
punting on this to the Apple folks.
We agreed that the correct fix is either to extend the fake SDK headers
in compiler-rt to support the rest of libc needed by GCDAProfiling.c or
to teach the make build to test for the existence of a suitable SDK on
the system prior to building it. Both of these require someone with
access to the SDK which I don't have, and the latter (my preferred
solution) requires understanding the strange compiler-rt make build
system. Punting to Dan Dunbar or one of the others who support this
stuff on ios to write the appropriate fix, and we can let the build bots
proceed in the mean time.
llvm-svn: 184816
Re-apply r184511, reverted in r184561, with the trivial default constructor
fast path removed -- it turned out not to be necessary here.
Certain expressions can cause a constructor invocation to zero-initialize
its object even if the constructor itself does no initialization. The
analyzer now handles that before evaluating the call to the constructor,
using the same "default binding" mechanism that calloc() uses, rather
than simply ignoring the zero-initialization flag.
<rdar://problem/14212563>
llvm-svn: 184815
In order to make sure virtual base classes are always initialized once,
the AST contains initializers for the base class in /all/ of its
descendents, not just the immediate descendents. However, at runtime,
the most-derived object is responsible for initializing all the virtual
base classes; all the other initializers will be ignored.
The analyzer now checks to see if it's being called from another base
constructor, and if so does not perform virtual base initialization.
<rdar://problem/14236851>
llvm-svn: 184814
numbers as we deserialize class template partial specializations. We can't
assume that the old sequence numbers will work.
The sequence numbers are still deterministic, but are now a lot less
predictable for class template partial specializations in modules/PCH.
llvm-svn: 184811
when building llvm. This saves quite a bit of time and space when
linking. Please report any problems via bugzilla.
Caveats:
a) This will only work on linux
b) This requires a fairly new binutils
c) This requires a fairly new gdb
llvm-svn: 184808
runtime in the Makefile build system as well. Sorry for the temporary
fallout, it took me a while to find these bits.
Bill, I'm not at all confident of the change for Darwin and iOS, but as
discussed we're completely blocked on fixing this. Anyways, please
review and let me know if this will work for you guys. If necessary,
I can work with you to rig up an errno.h stub for Darwin, but I expect
that to be... moderately challenging.
llvm-svn: 184805
This is really ugly, but it is no worse than what we have in clang right now and
it is better to get it working first and clean/optimize it afterwards.
Will be tested from clang in the next patch.
llvm-svn: 184802
Use castAs<> where appropriate. Don't check conditionals which are
always true. Delete a bit of dead code. Reindent a bunch of code which
is no longer guarded by an if statement.
llvm-svn: 184801
through Bill's patch:
1) Correctly test the file descriptor after the sceond attempt at
creating the file.
2) Make the filename a global so that we can issue error messages from
other routines.
3) Check errno in several places and print it out so that errors are
easier to track down.
I don't really expect any of these to fix the current failures I'm
seeing, but I'm hopeful they'll at least let me debug them.
llvm-svn: 184799
We really need the C standard library to be available to implement the
profiling runtime library reasonably, and replicating everything in the
SDKs tree really isn't addressing any problems we have. Notably, all of
the sanitizer runtimes take the same approach, and this isn't a library
which could end up in a bootstrapping problem where the system headers
aren't even available.
This will hopefully prevent subsequent changes which start using various
other bits of C standard library to make things more debuggable.
llvm-svn: 184798
command line. Change the darwin universal binary options to
be TargetCommonOpts so that they'll be passed to the linker since
-arch at least is still needed.
Someone on darwin with a buildit based build should probably verify
that this doesn't break anything there.
llvm-svn: 184793
-performSelector: and friends return a value that is boxed as an Objective-C
pointer. Sometimes it is an Objective-C pointer, sometimes it isn't.
Some clients may wish to silence this warning based on calling
this method.
Fixes <rdar://problem/14147304>
llvm-svn: 184789
This is a large test and thus it will only run if you pass in --param
run_long_tests=trueto LIT. This is intended so that this test can run on
buildbots and not when one runs make check.
llvm-svn: 184787
This will prevent the tests from running on normal make check. You will need to
actually pass in --param run_long_tests=true to LIT in order to run these.
llvm-svn: 184784
The top-level hash is used to determine if we need to update the global code-completion results.
ImportDecls did not affect the hash so a newly introduced ImportDecl would not trigger an update of the global results.
rdar://14202797
llvm-svn: 184782
This changes the mangling of local static variables/etc. inside blocks
to do something simple and sane. This avoids depending on the way we mangle
blocks, which isn't really appropriate here.
John, please take a look at this to make sure the mangling I chose is sane.
Fixes <rdar://problem/14074423>.
llvm-svn: 184780
At this point, it's clear that the MSVC mangler uses the type-as-written
instead of the canonical type, so this should bring us closer to MSVC.
The main thrust of this change is to fix the way we mangle decayed array
parameters of function pointer parameters. With a DecayedType sugar
node, this code can now be much simpler.
Fixes PR16096.
This also fixes a separate issue that Richard spotted in review.
Because separate declarations of the same entity can be spelled and
mangled differently, MSVC always mangles the earliest declaration in an
attempt to avoid link errors. Clang now does the same.
Reviewers: rsmith
Differential Revision: http://llvm-reviews.chandlerc.com/D844
llvm-svn: 184777
Tom Stellard