[PATCH][AArch64] updated patches with initial implementation of Neon scalar instructions

[Ana Pazos]

Hi folks,

I updated the patches for Neon scalar instructions.

1) Removed OP_SCALAR_ALIAS operator.
   The AArch64 scalar intrinsics that are alias to ARM legacy intrinsics are
no longer implemented as a call to the ARM legacy intrinsics.
   The AArch64 scalar intrinsics use builtins that are translated to LLVM IR
intrinsics with v1ix and v1fx operand types to guarantee the expected Neon
operation is generated.
   Otherwise we are vulnerable to the compiler optimizations and integer
code might result.
   Defined some instruction selection patterns to allow code generation of
Neon scalar instructions without using ACLE and  LLVM intrinsics if one
writes code that uses v1ix and v1fx types directly.

2) Fixed the failure reported by Kevin.
  I realized I was not building LLVM with assertions turned on.  I could
reproduce the error reported and it is now fixed.

Let me know what you think,

Thanks,
Ana.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: latest-clang-scalar-arith-scalar-reduce-pairwise-version3
Type: application/octet-stream
Size: 24502 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130918/aeb64a16/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: latest-llvm-scalar-arith-scalar-reduce-pairwise-version3
Type: application/octet-stream
Size: 102883 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130918/aeb64a16/attachment-0001.obj>