mirror of
https://github.com/c64scene-ar/llvm-6502.git
synced 2024-10-17 19:25:48 +00:00
f2d8190b81
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@163160 91177308-0d34-0410-b5e6-96231b3b80d8
30 lines
1.2 KiB
C++
30 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"
|
|
|
|
/// This optimization identifies DIV instructions that can be
|
|
/// profitably bypassed and carried out with a shorter, faster divide.
|
|
bool bypassSlowDivision(llvm::Function &F,
|
|
llvm::Function::iterator &I,
|
|
const llvm::DenseMap<llvm::Type *, llvm::Type *> &BypassTypeMap);
|
|
|
|
#endif
|