mirror of
https://github.com/c64scene-ar/llvm-6502.git
synced 2025-01-04 05:31:06 +00:00
911908dcb8
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@163940 91177308-0d34-0410-b5e6-96231b3b80d8
34 lines
1.2 KiB
C++
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
|