OSDN Git Service

This is a bulk commit that implements the following primary improvements:
authorChris Lattner <sabre@nondot.org>
Sat, 1 Jan 2005 16:22:27 +0000 (16:22 +0000)
committerChris Lattner <sabre@nondot.org>
Sat, 1 Jan 2005 16:22:27 +0000 (16:22 +0000)
commit6e7ba45460af5e6169f32cd6d595fca06365bb31
tree99744ed8651210ddbdd32095cc47a5fc7af7a393
parent0f75849116423f5e57776600cd47d80db624e496
This is a bulk commit that implements the following primary improvements:
  * We can now fold cast instructions into select instructions that
    have at least one constant operand.
  * We now optimize expressions more aggressively based on bits that are
    known to be zero.  These optimizations occur a lot in code that uses
    bitfields even in simple ways.
  * We now turn more cast-cast sequences into AND instructions.  Before we
    would only do this if it if all types were unsigned.  Now only the
    middle type needs to be unsigned (guaranteeing a zero extend).
  * We transform sign extensions into zero extensions in several cases.

This corresponds to these test/Regression/Transforms/InstCombine testcases:
  2004-11-22-Missed-and-fold.ll
  and.ll: test28-29
  cast.ll: test21-24
  and-or-and.ll
  cast-cast-to-and.ll
  zeroext-and-reduce.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19220 91177308-0d34-0410-b5e6-96231b3b80d8
lib/Transforms/Scalar/InstructionCombining.cpp