llvm-svn: 136432

llvm-svn: 136431

This generates the correct SDNodes for the landingpad instruction. It makes an
assumption that the result of the landingpad instruction has at least two
values. And that the first value is a pointer to the exception object and the
second value is the "selector."

llvm-svn: 136430

AddLandingPadInfo takes a landingpad instruction and grabs all of the
information from it that it needs for EH table generation.

llvm-svn: 136429

LBH_TAKEN_WEIGHT + LBH_NONTAKEN_WEIGHT = 128 which in _most_ cases reduce
number of rounding errors.

llvm-svn: 136428

llvm-svn: 136427

llvm-svn: 136408

Add parsing support for BLX (immediate). Since the register operand version is
predicated and the label operand version is not, we have to use some special
handling to get the operand list right for matching.

llvm-svn: 136406

llvm-svn: 136405

'atomicrmw' instructions, which allow representing all the current atomic
rmw intrinsics.

The allowed operands for these instructions are heavily restricted at the
moment; we can probably loosen it a bit, but supporting general
first-class types (where it makes sense) might get a bit complicated,
given how SelectionDAG works.

As an initial cut, these operations do not support specifying an alignment,
but it would be possible to add if we think it's useful. Specifying an
alignment lower than the natural alignment would be essentially
impossible to support on anything other than x86, but specifying a greater
alignment would be possible.  I can't think of any useful optimizations which
would use that information, but maybe someone else has ideas.

Optimizer/codegen support coming soon.

llvm-svn: 136404

llvm-svn: 136403

llvm-svn: 136402

Code like that would only be produced by bugpoint, but we should still
handle it correctly.

When a register is defined by a REG_SEQUENCE of undefs, the register
itself is undef. Previously, we would create a register with uses but no
defs.

Fixes part of PR10520.

llvm-svn: 136401

llvm-svn: 136400

Add parsing support that handles converting the lsb+width source into the
odd way we represent the instruction (an inverted bitfield mask).

llvm-svn: 136399

there is no frequency difference whether condition is in the header or in
the latch.

llvm-svn: 136398

llvm-svn: 136396

llvm-svn: 136392

llvm-svn: 136390

There are two conflicting strategies in play:

- Under high register pressure, we want to assign large live ranges
  first. Smaller live ranges are easier to place afterwards.

- Live range splitting is guided by interference, so splitting should be
  deferred until interference is as realistic as possible.

With the recent changes to the live range stages, and with compact
regions enabled, it is less traumatic to split a live range too early.
If some of the split products were too big, they can often be split
again.

By reversing the RS_Split order, we get this queue order:

1. Normal live ranges, large to small.
2. RS_Split live ranges, large to small.

The large-to-small order improves RAGreedy's puzzle solving skills under
high register pressure. It may cause a bit more iterated splitting, but
we handle that better now.

With this change, -compact-regions is mostly an improvement on SPEC.

llvm-svn: 136388

This should be the only code necessary for DWARF EH prepare.

llvm-svn: 136387

llvm-svn: 136384

llvm-svn: 136381

llvm-svn: 136375

llvm-svn: 136369

llvm-svn: 136367

llvm-svn: 136366

Revert r136295.  It broke nightly testers because some parts of codegen weren't aware of the changes to operand ordering.  I hope to revive this sometime in the future, but it's not strictly necessary for now.

llvm-svn: 136362

The label does not have a '#' prefix. Add parsing and encoding tests.

llvm-svn: 136360

llvm-svn: 136358

identical.

llvm-svn: 136355

more graphs, like all graphs with 5 nodes or less.  With a 32 bit
unsigned type, the maximum is graphs with 6 nodes or less, but that
would take a while to test - 5 nodes or less already requires a few
seconds.

llvm-svn: 136354

iterates over SCC's.

llvm-svn: 136353

This computes every graph with 4 or fewer nodes, and checks that the SCC
class indeed returns exactly the simply connected components reachable
from the initial node.

llvm-svn: 136351

bitcasts in this test rather than getelementptr instructions;
llvm-gcc produces two bitcasts, clang produces one.

llvm-svn: 136349

llvm-svn: 136344

llvm-svn: 136341

The new EH is more simple in many respects. Mainly, we don't have to worry about
the "llvm.eh.exception" and "llvm.eh.selector" calls being in weird places.

llvm-svn: 136339

llvm-svn: 136338

of the empty key for U. This shouldn't really matter because the tombstone key
for the pair was still distinct from every other key, but it is odd. Patch by
Michael Ilseman!

llvm-svn: 136336