llvm-svn: 33695

llvm-svn: 33694

llvm-svn: 33693

This allows us to eliminate many temporary vectors, and theirassociated malloc/free pairs.

llvm-svn: 33692

vectors.  This allows us to eliminate many temporary vectors, and their
associated malloc/free pairs.

llvm-svn: 33691

- Islands are inserted right after the user MI since thumb LDR cannot encode
  negative offset.

llvm-svn: 33690

llvm-svn: 33689

llvm-svn: 33688

it can be used by multiple clients.  This specifically allows the inliner
to constant fold symbolically.

llvm-svn: 33687

llvm-svn: 33686

llvm-svn: 33685

ConstantFoldInstOperands.  Switch to ConstantFoldInstOperands and remove
ConstantFold.

llvm-svn: 33683

  1. Remove using namespace llvm;
  2. Remove unneeded #includes.
  3. Add #include guard.

Also add prototypes for new code moved from transformutils to libanalysis.

llvm-svn: 33682

llvm-svn: 33681

llvm-svn: 33680

libanalysis/ConstantFolding.cpp.

llvm-svn: 33679

llvm-svn: 33678

llvm-svn: 33677

used by constant folding.

llvm-svn: 33676

llvm-svn: 33675

llvm-svn: 33674

llvm-svn: 33673

llvm-svn: 33672

llvm-svn: 33671

ConstantFoldInstOperands/ConstantFoldCall to take a pointer to an array
of operands + size, instead of an std::vector.

In some cases, switch to using a SmallVector instead of a vector.
This allows us to get rid of some special case gross code that was there
to avoid the cost of constructing a vector.

llvm-svn: 33670

ConstantFoldInstOperands/ConstantFoldCall to take a pointer to an array
of operands + size, instead of an std::vector.

llvm-svn: 33669

llvm-svn: 33668

that would require > 3 instructions to materialize), load the immediate from a
constpool entry.

llvm-svn: 33667

llvm-svn: 33666

This occurs in C++ code like:

#include <iostream>
#include <iterator>
int a[] = { 1, 2, 3, 4, 5 };
int main() {
  using namespace std;
  copy(a, a + sizeof(a)/sizeof(a[0]), ostream_iterator<int>(cout, "\n"));
  return 0;
}

Before we would decide the loop trip count is:
sdiv (i32 sub (i32 ptrtoint (i32* getelementptr ([5 x i32]* @a, i32 0, i32 5) to i32), i32 ptrtoint ([5 x i32]* @a to i32)), i32 4)

Now we decide it is "5".  Amazing.

This code will need to be refactored, but I'm doing that as a separate
commit.

llvm-svn: 33665

        .set PCRELV0, (LJTI1_0_0-(LPCRELL0+4))
LPCRELL0:
        add r1, pc, #PCRELV0
This is not legal since add r1, pc, #c requires the constant be a multiple of 4.
Do the following instead:
        .set PCRELV0, (LJTI1_0_0-(LPCRELL0+4))
LPCRELL0:
        mov r1, #PCRELV0
        add r1, pc

- In thumb mode, it's not possible to use .set generate a pc relative stub
  address. The stub is ARM code which is in a different section from the thumb
  code. Load the value from a constpool instead.
- Some asm printing clean up.

llvm-svn: 33664

confusion with external linkage types.

llvm-svn: 33663

llvm-svn: 33662

outweight its computational costs. This patch removes all compaction
table handling from the bcreader and bcwriter. For the record, here's the
difference betweeen having and not having compaction tables for some tests:

Test             With       Without   Size Chg
Olden/mst       5,602         5,598      +0.1%
viterbi        18,026        17,795      +1.3%
obsequi       162,133       166,663      -2.8%
burg          224,090       228,148      -1.8%
kimwitu++   4,933,263     5,121,159      -3.8%
176.gcc     8,470,424     9,141,539      -7.3%

It seems that it is more beneficial to larger files, but even on the largest
test case we have (176.gcc) it only amounts ot an I/O saving of 7.3%.

llvm-svn: 33661

Update these tests to not use the same name even though the type of the
value differs. After PR411 hits, type planes will be gone and it will be
illegal for a name to be used twice, regardless of type.

llvm-svn: 33660

llvm-svn: 33659

llvm-svn: 33658

llvm-svn: 33657

llvm-svn: 33656

llvm-svn: 33655