llvm-svn: 79664

This preserves the existing behavior much more closely than my previous attempt.

llvm-svn: 79663

llvm-svn: 79659

llvm-svn: 79657

llvm-svn: 79656

llvm-svn: 79648

  warning: comparison is always true due to limited range of data type.

llvm-svn: 79642

llvm-svn: 79636

llvm-svn: 79634

leaf functions. This pass will be extended to color other nodes of the call tree 
as well in future.

llvm-svn: 79631

llvm-svn: 79626

llvm-svn: 79625

llvm-svn: 79624

llvm-svn: 79622

llvm-svn: 79621

llvm-svn: 79620

Use masks not nodes for vector shuffle predicates. Provide set of 'legal' masks, so legalizer won't infinite cycle

llvm-svn: 79619

llvm-svn: 79603

vector shuffles.  Temporarily remove the tests for these operations until the
new implementation is working.

llvm-svn: 79579

llvm-svn: 79578

llvm-svn: 79572

U    include/llvm/Target/TargetData.h
U    lib/Target/TargetData.cpp

Temporarily revert 79555. It was causing hangs and test failures.

llvm-svn: 79568

llvm-svn: 79563

Clone functions that are shared between the Main thread and Interrupt thread.
CallSites are changed in AsmPrinter currently. A better solution would have been to modify the legalizer (SoftenFloat) to allow targets to change the name of libcalls for float operations. But that currently breaks other targets.
Also, cloing of automatic variables is done AsmPrinter, a better approach would
be to use the ValueMap in CloneFunction itself.

llvm-svn: 79562

llvm-svn: 79555

Intel documentation.

llvm-svn: 79554

over absolute addressing even in non-PIC mode (unless the address
has an index or something else incompatible), because it has a
smaller encoding.

llvm-svn: 79553

llvm-svn: 79535

llvm-svn: 79532

llvm-svn: 79495

llvm-svn: 79494

llvm-svn: 79484

llvm-svn: 79483

llvm-svn: 79443

a smaller encoding. These kinds of patterns are very frequent in
sqlite3, for example.

llvm-svn: 79439

llvm-svn: 79436

This is derived from a patch by Anton Korzh.  I modified it to recognize
the VEXT shuffles during legalization and lower them to a target-specific
DAG node.

llvm-svn: 79428

talk to the MCStreamer directly instead.

llvm-svn: 79405

Add patterns and instruction encoding information.
Add custom lowering to deal with hardwired return register of
uncertain type (xmm0).

llvm-svn: 79377

- Drop the Candidates argument and fix all callers. Now that RegScavenger
  tracks available registers accurately, there is no need to restict the
  search.
- Make sure that no aliases of the found register are in use. This was a potential bug.

llvm-svn: 79369