llvm-6502/include/llvm/Transforms/Utils/BypassSlowDivision.h
Evan Cheng 911908dcb8 Stylistic and 80-col fixes
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@163940 91177308-0d34-0410-b5e6-96231b3b80d8
2012-09-14 21:25:34 +00:00

34 lines
1.2 KiB
C++

//===- llvm/Transforms/Utils/BypassSlowDivision.h --------------*- C++ -*-===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file contains an optimization for div and rem on architectures that
// execute short instructions significantly faster than longer instructions.
// For example, on Intel Atom 32-bit divides are slow enough that during
// runtime it is profitable to check the value of the operands, and if they are
// positive and less than 256 use an unsigned 8-bit divide.
//
//===----------------------------------------------------------------------===//
#ifndef TRANSFORMS_UTILS_BYPASSSLOWDIVISION_H
#define TRANSFORMS_UTILS_BYPASSSLOWDIVISION_H
#include "llvm/Function.h"
namespace llvm {
/// This optimization identifies DIV instructions that can be
/// profitably bypassed and carried out with a shorter, faster divide.
bool bypassSlowDivision(Function &F,
Function::iterator &I,
const DenseMap<Type*, Type*> &BypassTypeMap);
} // End llvm namespace
#endif