where this would be bad as the backend shouldn't have a problem inlining small
memcpys.
rdar://10510150

llvm-svn: 145865

llvm-svn: 145863

llvm-svn: 145861

rdar://10529664

llvm-svn: 145860

llvm-svn: 145854

PerformANDCombine and PerformOrCombine aware of them. Test cases are included
too.

llvm-svn: 145853

them.

llvm-svn: 145852

rdar://10529348

llvm-svn: 145851

O32 with relocation-model=pic too.

llvm-svn: 145850

Finish up rdar://10522016.

llvm-svn: 145846

llvm-svn: 145844

llvm-svn: 145843

Combined destination and first source operand for f32 variant of the VMUL
(by scalar) instruction.

rdar://10522016

llvm-svn: 145842

This is a patch by Guoping Long!

As part of utilizing LLVM Dominator computation in Clang, made two changes to LLVM dominators tree implementation:

 - (1) Change the recalculate() template function to only rely on GraphTraits.
 - (2) Add a size() method to GraphTraits template class to query the number of nodes in the graph.

llvm-svn: 145837

llvm-svn: 145820

llvm-svn: 145819

llvm-svn: 145818

llvm-svn: 145817

llvm-svn: 145816

- Calling getUser in a loop is much more expensive than iterating over a few instructions.
- Use it instead of the open-coded loop in AddrModeMatcher.
- 5% speedup on ARMDisassembler.cpp Release builds.

llvm-svn: 145810

llvm-svn: 145805

llvm-svn: 145804

llvm-svn: 145803

Add support for vectors of pointers.

llvm-svn: 145801

llvm-svn: 145793

llvm-svn: 145792

not get there any other way.

llvm-svn: 145789

llvm-svn: 145783

Maybe some targets should use this as well.

Patch by Evgeniy Stepanov!

llvm-svn: 145781

AnalyzeBranch doesn't change the successor, just the order.

llvm-svn: 145779

-3% on ARMDissasembler.cpp.

llvm-svn: 145773

llvm-svn: 145771

Add a "seen blocks" cache to LVI to avoid a linear scan over the whole cache just to remove no blocks from the maps.

-15% on ARMDisassembler.cpp (Release build).  It's not that great to add another
layer of caching to the caching-heavy LVI but I don't see a better way.

llvm-svn: 145770

libgcc sets the stack limit field in TCB to 256 bytes above the actual
allocated stack limit.  This means if the function's stack frame needs
less than 256 bytes, we can just compare the stack pointer with the
stack limit.  This should result in lesser calls to __morestack.

llvm-svn: 145766

Currently LLVM pads the call to __morestack with a add and sub of 8
bytes to esp.  This isn't correct since __morestack expects the call
to be followed directly by a ret.

This commit also adjusts the relevant test-case.

llvm-svn: 145765

the same value) to this variable. This code could be refactored, but it doesn't
matter since the old JIT is going away. Add tsan annotations to ignore the
race.

llvm-svn: 145745

rdar://10510150

llvm-svn: 145742

Fixed deadstoreelimination bug where negative indices were incorrectly causing the optimisation to occur

Turns out long long + unsigned long long is unsigned.  Doh!

Fixes http://llvm.org/bugs/show_bug.cgi?id=11455

llvm-svn: 145731

rdar://10514247

llvm-svn: 145730

llvm-svn: 145726