PR1909: Tail merging pass ran wild. It makes no sense to merge blocks in order to save a single instruction since a branch will be inserted for each BB.

llvm-svn: 47301

llvm-svn: 47300

llvm-svn: 47297

llvm-svn: 47296

- When DAG combiner is folding a bit convert into a BUILD_VECTOR, it should check if it's essentially a SCALAR_TO_VECTOR. Avoid turning (v8i16) <10, u, u, u> to <10, 0, u, u, u, u, u, u>. Instead, simply convert it to a SCALAR_TO_VECTOR of the proper type.
- X86 now normalize SCALAR_TO_VECTOR to (BIT_CONVERT (v4i32 SCALAR_TO_VECTOR)). Get rid of X86ISD::S2VEC.

llvm-svn: 47290

llvm-svn: 47287

on x86-32 since i64 itself is not a Legal type. And, update some
comments.

llvm-svn: 47282

- For now, conservatively ignore copy MI whose source is a physical register. Commuting its def MI can cause a physical register live interval to be live through a loop (since we know it's live coming into the def MI).

llvm-svn: 47281

llvm-svn: 47280

llvm-svn: 47279

llvm-svn: 47278

has plain one-result scalar integer multiplication instructions.
This avoids expanding such instructions into MUL_LOHI sequences that
must be special-cased at isel time, and avoids the problem with that
code that provented memory operands from being folded.

This fixes PR1874, addressesing the most common case. The uncommon
cases of optimizing multiply-high operations will require work
in DAGCombiner.

llvm-svn: 47277

to Dale for noticing this!

llvm-svn: 47276

llvm-svn: 47275

llvm-svn: 47274

requested by Chris.  While there, do the same
for an existing function committed by someone
called "lattner" :)

llvm-svn: 47273

llvm-svn: 47272

llvm-svn: 47271

that Object is an Argument before casting it to one.

llvm-svn: 47268

New helper function getMBBFromIndex() that given an index in any instruction of an MBB returns a pointer the MBB. Reviewed by Evan.

llvm-svn: 47267

Add support to GVN for performing sret return slot optimization.  This means that, if an sret function tail calls
another sret function, it should pass its own sret parameter to the tail callee, allowing it to fill in the correct
return value.  llvm-gcc does not emit this by default.  Instead, it allocates space in the caller for the sret of
the tail call and then uses memcpy to copy the result into the caller's sret parameter.  This optimization detects
and optimizes that case.

llvm-svn: 47265

llvm-svn: 47264

llvm-svn: 47263

For now, avoid commuting def MI for copy MI's whose source is not killed. That simply trade a live interval for another and because only the non-two-address operands can be folded into loads, may end up pessimising code.

llvm-svn: 47262

this speeds it up 2.3% on eon.

llvm-svn: 47261

llvm-svn: 47258

llvm-svn: 47257

only applies to the first parameter.

llvm-svn: 47256

llvm-svn: 47255

a function must be the first parameter.

llvm-svn: 47254

Fix a comment, and a bug where we weren't applying the tail call logic in cases that failed the first test.

llvm-svn: 47253

llvm-svn: 47252

llvm-svn: 47251

If we see a memcpy of a pointer, make sure to check later
uses of the pointer as well.

llvm-svn: 47250

llvm-svn: 47249

llvm-svn: 47248

Also, noalias arguments are be considered "like" stack allocated ones for this purpose, because
the only way they can be modref'ed is if they escape somewhere in the current function.

llvm-svn: 47247

the second half of PR2047

llvm-svn: 47244

llvm-svn: 47242

that is also a subtract.  This implements PR2047 and Transforms/Reassociate/subtest2.ll 

llvm-svn: 47241