llvm-svn: 105628

llvm-svn: 105624

copy-in of c++ class objects into blocks.

llvm-svn: 105622

pointer. Original patch by Troy D. Straszheim; fixes PR7283.

llvm-svn: 105621

llvm-svn: 105620

llvm-svn: 105619

llvm-svn: 105618

llvm-svn: 105617

DeltaAlgorithm: Tweak split to split by first/second half instead of even/odd, since adjacent changes are more likely to be related.

llvm-svn: 105613

ADT: Add DAGDeltaAlgorithm, which is a DAG minimization algorithm built on top of the standard 'delta debugging' algorithm.
 - This can give substantial speedups in the delta process for inputs we can construct dependency information for.

llvm-svn: 105612

When checking whether we can place a base subobject at an offset, we don't need to go past the highest offset that's known to contain an empty base subobject.

llvm-svn: 105611

the llvm tests :-(
It was failing with

-- Testing: 5324 tests, 8 threads --
Fatal Python error: PyEval_AcquireThread: NULL new thread state

llvm-svn: 105610

llvm-svn: 105608

llvm-svn: 105607

llvm-svn: 105606

realloc implementation can try to expand the allocated memory block in-place,
avoiding the copy.

llvm-svn: 105605

llvm-svn: 105604

the function call is left where the local variable is declared.

llvm-svn: 105602

Discussed here:
http://lists.cs.uiuc.edu/pipermail/llvmdev/2010-June/032107.html

llvm-svn: 105601

llvm-svn: 105600

Fix NEON intrinsic argument passing, support vext.  Most now successfully make it through codegen to the .s file

llvm-svn: 105599

Refine BuiltinsARM.def types a bit, we should do a better job of this to save some c++ code in CGBuiltins.

llvm-svn: 105598

llvm-svn: 105597

llvm-svn: 105596

llvm-svn: 105595

llvm-svn: 105594

Since the enum values for each arch's builtins overlap, it is not appropriate to check them when compiling or other archs.  Fixes a problem where compiling for NEON would use x86 sema rules.

llvm-svn: 105593

ARM.

Fixes PR7310.

llvm-svn: 105592

llvm-svn: 105591

llvm-svn: 105590

Extend __builtin_shufflevector to expose the full power of the llvm shufflevector instruction.  This means it can now be used for vector truncation and concatenation.  This will be used for the ARM NEON implementation.

llvm-svn: 105589

fix vcvt naming
handle vdup, vcombine with generic vector code

llvm-svn: 105588

- change isShuffleMaskLegal to show that all shuffles with 32-bit and 64-bit
  elements are legal
- the Neon shuffle instructions do not support 64-bit elements, but we were
  not checking for that before lowering shuffles to use them
- remove some 64-bit element vduplane patterns that are no longer needed

llvm-svn: 105586

that it is an immediate before checking that the instruction is an
EXTRACT_SUBREG.

llvm-svn: 105585

 - We actually pretend that we have two separate types for LLVM assembly/bitcode because we need to use the standard suffixes with LTO ('clang -O4 -c t.c' should generate 't.o').

It is now possible to do something like:
  $ clang -emit-llvm -S t.c -o t.ll ... assorted other compile flags ...
  $ clang -c t.ll -o t.o ... assorted other compile flags ...
and expect that the output will be almost* identical to:
  $ clang -c t.c -o t.o ... assorted other compile flags ...
because all the target settings (default CPU, target features, etc.) will all be initialized properly by the driver/frontend.

*: This isn't perfect yet, because in practice we will end up running the optimization passes twice. It's possible to get something equivalent out with a well placed -mllvm -disable-llvm-optzns, but I'm still thinking about the cleanest way to solve this problem more generally.

llvm-svn: 105584

 - This magically enables using 'clang -cc1' as a replacement for most of 'llvm-as', 'llvm-dis', 'llc' and 'opt' functionality.

For example, 'llvm-as' is:
  $ clang -cc1 -emit-llvm-bc FOO.ll -o FOO.bc
and 'llvm-dis' is:
  $ clang -cc1 -emit-llvm    FOO.bc -o -
and 'opt' is, e.g.:
  $ clang -cc1 -emit-llvm -O3 -o FOO.opt.ll FOO.ll
and 'llc' is, e.g.:
  $ clang -cc1 -S -o - FOO.ll

The nice thing about using the backend tools this way is that they are guaranteed to exactly match how the compiler generates code (for example, setting the same backend options).

llvm-svn: 105583

 - These inputs follow an abbreviated execution path, but are still worth handling by FrontendAction so they reuse all the other clang -cc1 features.

llvm-svn: 105582

llvm-svn: 105581

llvm-svn: 105580

Frontend: Move some initialization from CompilerInstance to FrontendAction, to parallel what is done for AST inputs.

llvm-svn: 105579