llvm-svn: 90567

llvm-svn: 90566

architectures that LLVM targets, because they don't use this code.

llvm-svn: 90564

llvm-svn: 90557

llvm-svn: 90553

llvm-svn: 90548

llvm-svn: 90547

folding a load from constant.

llvm-svn: 90545

llvm-svn: 90539

that I'm working on.  This is manifesting as a miscompile of 255.vortex
on some targets.  No check lines yet because it fails.

llvm-svn: 90520

gvn (this is just a skeleton so far).  This will ultimately be used
to fix a nasty miscompilation with GVN.

llvm-svn: 90518

llvm-svn: 90514

Teach AsmWriter to write inline (not via a global metadata slot) metadata that contains an instruction

llvm-svn: 90512

llvm-svn: 90511

llvm-svn: 90508

a compile time perf optimization to avoid a load.

llvm-svn: 90507

The coalescer is supposed to clean these up, but when setting up parameters
for a function call, there may be copies to physregs. If the defining
instruction has been LICM'ed far away, the coalescer won't touch it.

The register allocation hint does not always work - when the register
allocator is backtracking, it clears the hints.

This patch takes care of a few more cases that r90163 missed.

llvm-svn: 90502

- If the reaching definition is an undef and the use is a PHI, add the implicit_def to the end of the source block.
- When reaching value is replaced with another, update the cache as well.

llvm-svn: 90501

Insert composite type DIE into the map before processing type fields. This allows fields to find their context DIE from the map.

llvm-svn: 90498

Add ParseInlineMetadata() which can parses metadata that refers to an instruction.  Extend ParseParameterList() to use this new function so that calls to llvm.dbg.declare can pass inline metadata

llvm-svn: 90497

llvm-svn: 90490

llvm-svn: 90489

llvm-svn: 90488

llvm-svn: 90487

Pointed out by Javier Martinez (who also provided a patch).  Since
this logic is not used on (for example) x86, I guess nobody noticed.
Tested by generating SHL, SRL, SRA on various choices of i64 for all
possible shift amounts, and comparing with gcc.  Since I did this on
x86-32, I had to force the use of ExpandShiftWithUnknownAmountBit.
What I'm saying here is that I don't have a testcase I can add to the
repository.

llvm-svn: 90482

llvm-svn: 90481

llvm-svn: 90474

end up on the same line.

llvm-svn: 90473

llvm-svn: 90454

llvm-svn: 90451

Use integer values for weights to prevent rounding errors.
Make ProfileEstimator more robust in general CFGs.

llvm-svn: 90449

 - This is a pretty slow / memory intensive implementation, and I will likely
   change it to an iterative model, but it works.

llvm-svn: 90447

llvm-svn: 90446

llvm-svn: 90445

llvm-svn: 90432

llvm-svn: 90429

llvm-svn: 90428

Don't pull vector sext through both hands of a logical operation, since doing so prevents the fusion of vector sext and setcc into vsetcc.
Add a testcase for the above transformation.
Fix a bogus use of APInt noticed while tracking this down.

llvm-svn: 90423

llvm-svn: 90419

both source operands.  In the canonical form, the 2nd operand is changed to an
undef and the shuffle mask is adjusted to only reference elements from the 1st
operand.  Radar 7434842.

llvm-svn: 90417