llvm-svn: 175385

MachineBasicBlock::SplitCriticalEdge. Since this is an iterator rather than
an instr_iterator, the isBundled() check only passes if getFirstTerminator()
returned end() and the garbage memory happens to lean that way.

Multiple successors can be present without any terminator instructions in the
case of exception handling with a fallthrough.

llvm-svn: 175383

terminators that actually have register uses when splitting critical edges.

This commit also introduces a method repairIntervalsInRange() on LiveIntervals,
which allows for repairing LiveIntervals in a small range after an arbitrary
target hook modifies, inserts, and removes instructions. It's pretty limited
right now, but I hope to extend it to support all of the things that are done
by the convertToThreeAddress() target hooks.

llvm-svn: 175382

    (or (bool?A:B),(bool?C:D)) --> (bool?(or A,C):(or B,D))

By the time the OR is visited, both the SELECTs have been visited and not
optimized and the OR itself hasn't been transformed so we do this transform in
the hopes that the new ORs will be optimized.

The transform is explicitly disabled for vector-selects until "codegen matures
to handle them better".

Patch by Muhammad Tauqir!

llvm-svn: 175380

llvm-svn: 175379

llvm-svn: 175375

llvm-svn: 175373

llvm-svn: 175372

llvm-svn: 175371

Avoids malloc and is a lot denser. We lose iteration over target independent
attributes, but that's a strange interface anyways and didn't have any users
outside of AttrBuilder.

llvm-svn: 175370

as well as 16/32 bit variants to do and so I want this to look nice
when I do it. I've been experimenting with this. No new test cases
are needed.

llvm-svn: 175369

This avoids unnecessary copies. No functionality change.

llvm-svn: 175367

No functionality change.

llvm-svn: 175366

llvm-svn: 175365

No functionality change.

llvm-svn: 175364

llvm-svn: 175363

updateScheduledPressure method.

llvm-svn: 175362

llvm-svn: 175361

GNU as rejects them and there are configure scripts in the wild that check if
the assembler rejects ".align 3" to determine whether the alignment is in bytes
or powers of two.

llvm-svn: 175360

llvm-svn: 175359

llvm-svn: 175358

Also fix one test by changing "vpermilps" to "vpshufd".

llvm-svn: 175357

This is a candidate for the stable branch.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 175356

It's completely unnecessary and can be replace with proper
SReg_64 handling instead.

This actually fixes a piglit test on SI.

v2: use correct register class in addRegisterClass,
    set special classes as not allocatable
v3: revert setting special classes as not allocateable

This is a candidate for the stable branch.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 175355

Seems to be allot simpler, and also paves the
way for further improvements.

v2: rebased on master, use 0 in BUFFER_LOAD_FORMAT_XYZW,
    use VGPR0 in dummy EXP, avoid compiler warning, break
    after encoding the first literal.
v3: correctly use V_ADD_F32_e64

This is a candidate for the stable branch.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 175354

Mark all the operands that can also have an immediate.

v2: SOFFSET is also an SSrc_32 operand

This is a candidate for the stable branch.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 175353

Previously it only worked because of coincident.

v2: fix 64bit versions, use 0x80 (inline 0) instead of SGPR0
    for the unused SRC2

This is a candidate for the stable branch.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 175352

This is a candidate for the stable branch.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 175351

This is a candidate for the stable branch.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 175350

Stop adding more instructions than necessary.

This is a candidate for the stable branch.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 175349

Generate more than one loop if it seems to make sense.

This is a candidate for the stable branch.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 175348

Using the new NearestCommonDominator class.

This is a candidate for the stable branch.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 175347

Using the new NearestCommonDominator class.

This is a candidate for the stable branch.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 175346

This is a candidate for the stable branch.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 175345

llvm-svn: 175344

llvm-svn: 175343

consistent with how BuildMI works. No new tests needed. All should work
the same as before.

llvm-svn: 175342

llvm-svn: 175338

llvm-svn: 175336

If the frame pointer is omitted, and any stack changes occur in the inline
assembly, e.g.: "pusha", then any C local variable or C argument references
will be incorrect.  

I pass no judgement on anyone who would do such a thing. ;)
rdar://13218191

llvm-svn: 175334