only run for x86 with fastisel. I've found it being very effective in
eliminating some obvious dead code as result of formal parameter lowering
especially when tail call optimization eliminated the need for some of the loads
from fixed frame objects. It also shrinks a number of the tests. A couple of
tests no longer make sense and are now eliminated.

llvm-svn: 95493

llvm-svn: 95488

llvm-svn: 95477

Do not emit callseq instructions around sibcalls. This eliminated some unnecessary stack adjustments.

llvm-svn: 95475

addresses in dbg.declare; ignore this for the
moment to prevent things from breaking.

llvm-svn: 95471

llvm-svn: 95468

Function-local metadata whose operands had been optimized to no longer refer to function-local IR were not getting written by BitcodeWriter; solution is for these metadata to be enumerated just like global metadata.

llvm-svn: 95467

This time it's for real! I am going to hook this up in the frontends as well.

The inliner has some experimental heuristics for dealing with the inline hint.
When given a -respect-inlinehint option, functions marked with the inline
keyword are given a threshold just above the default for -O3.

We need some experiments to determine if that is the right thing to do.

llvm-svn: 95466

llvm-svn: 95461

Radar 7614112.

llvm-svn: 95456

llvm-svn: 95454

llvm-svn: 95453

llvm-svn: 95452

llvm-svn: 95448

redundant with a correct one) pattern that was added for the disassembler.

llvm-svn: 95446

llvm-svn: 95445

	pslld	69, %mm3                ## encoding: [0x0f,0xf2,0x1c,0x25,0x45,0x00,0x00,0x00]

instead of like this:
	pslld	69, %mm3                ## encoding: [0x0f,0xf2,0x1c,0x25,0x45,0000,0000,0000]

this only affects 0.

llvm-svn: 95441

llvm-svn: 95440

Weird code sometimes uses pointer constants other than null. This patch
teaches SimplifyCFG to build switch instructions in those cases.

Code like this:

void f(const char *x) {
  if (!x)
    puts("null");
  else if ((uintptr_t)x == 1)
    puts("one");
  else if (x == (char*)2 || x == (char*)3)
    puts("two");
  else if ((intptr_t)x == 4)
    puts("four");
  else
    puts(x);
}

Now becomes a switch:

define void @f(i8* %x) nounwind ssp {
entry:
  %magicptr23 = ptrtoint i8* %x to i64            ; <i64> [#uses=1]
  switch i64 %magicptr23, label %if.else16 [
    i64 0, label %if.then
    i64 1, label %if.then2
    i64 2, label %if.then9
    i64 3, label %if.then9
    i64 4, label %if.then14
  ]

Note that LLVM's own DenseMap uses magic pointers.

llvm-svn: 95439

llvm-svn: 95438

llvm-svn: 95437

llvm-svn: 95435

pseudo instructions.

llvm-svn: 95433

llvm-svn: 95432

lower the SETB* instructions.

llvm-svn: 95431

xform it is checking to actually pass.  There is no need to match
m_SelectCst<0, -1> since instcombine canonicalizes that into not(sext).

Add matches for sext(not(x)) in addition to not(sext(x)).

llvm-svn: 95420

llvm-svn: 95414

llvm-svn: 95410

container data. This prevents it from holding onto dangling
pointers and potentially behaving unpredictably.

llvm-svn: 95409

llvm-svn: 95408

malloc caller in a profile.

llvm-svn: 95407

TSFlags directly instead of a TargetInstrDesc.

llvm-svn: 95405

llvm-svn: 95403

that in mind.

llvm-svn: 95402

llvm-svn: 95397

llc.cpp also defined these flags, meaning that when I linked all of LLVM's
libraries into a single shared library, llc crashed on startup with duplicate
flag definitions.  This patch passes them through the EngineBuilder into
JIT::selectTarget().

llvm-svn: 95390

llvm-svn: 95378

llvm-svn: 95368

llvm-svn: 95351

to start limping.

llvm-svn: 95350