scale.common
Class Machine

java.lang.Object
  scale.common.Machine

Direct Known Subclasses:

AlphaMachine, MipsMachine, PPCMachine, SparcMachine, Trips2Machine, X86Machine, XyzMachine

public abstract class Machine
extends java.lang.Object

This is the base class for all machine specific information.

$Id: Machine.java,v 1.61 2007-11-01 16:52:31 burrill Exp $

Copyright 2008 by the Scale Compiler Group,
Department of Computer Science
University of Massachusetts,
Amherst MA. 01003, USA
All Rights Reserved.

Field Summary

static int	ALL_CONDITIONAL_MOVES Capability: All the conditional move possibilities.
protected int	cacheLineSize The size in addressable units of an L1 cache line.
static Machine	currentMachine The current machine specification.
protected FloatType	doubleType The type used for the C double type.
protected FloatType	floatCalcType The real type best used for floating point calculations.
protected FloatType	floatType The type used for the C float type.
static int	HAS_EXPENSIVE_SUBROUTINE_CALLS Capability: bad to flatten/unroll loops that contain a function call.
static int	HAS_FF_CONDITIONAL_MOVE Capability: Has a conditional move using a floating point test and floating point values.
static int	HAS_FI_CONDITIONAL_MOVE Capability: Has a conditional move using a floating point test and integer values.
static int	HAS_IF_CONDITIONAL_MOVE Capability: Has a conditional move using an integer test and floating point values.
static int	HAS_II_CONDITIONAL_MOVE Capability: Has a conditional move using an integer test and integer values.
static int	HAS_INT_FROM_FP_CMP Capability: Has a floating point compare that returns an integer value.
static int	HAS_NO_FP_DIVIDE Capability: Floating point divides are not performed by an instruction but by a subroutine call.
static int	HAS_NO_INT_DIVIDE Capability: Integer divides are not performed by an instruction but by a subroutine call.
static int	HAS_NON_VOLATILE_FP_REGS Capability: Has a floating point registers that are saved over function calls.
static int	HAS_PREDICATION Capability: Has predicated instructions.
static int	HAS_SIMPLE_FLOOR Capability: Has a simple inline floor capability.
static int	HAS_WHILE_LOOP_UNROLL_BENEFIT Capability: Has benefit from unrolling loops with no induction variable.
protected IntegerType	intCalcType The integer type best used for integer calculations.
protected FloatType	longDoubleType The type used for the C long double type.
protected IntegerType	ptrdifftType The type used for the C ptrdiff_t type.
protected IntegerType	signedCharType The type used for the C signed char type.
protected IntegerType	signedIntType The type used for the C signed int type.
protected IntegerType	signedLongLongType The type used for the C signed long long type.
protected IntegerType	signedLongType The type used for the C signed long type.
protected IntegerType	signedShortType The type used for the C signed short type.
protected IntegerType	sizetType The type used for the C size_t type.
protected IntegerType	smallestAddressableUnitType The type for the smallest addressable unit (e.g., char).
protected IntegerType	unsignedCharType The type used for the C unsigned char type.
protected IntegerType	unsignedIntType The type used for the C unsigned int type.
protected IntegerType	unsignedLongLongType The type used for the C unsigned long long type.
protected IntegerType	unsignedLongType The type used for the C unsigned long type.
protected IntegerType	unsignedShortType The type used for the C unsigned short type.
protected Type	vaListType The type to be used for va_list.
protected PointerType	voidStarType The type used for the C void* type.
protected IntegerType	wchartType The type used for the C char_t type.

Constructor Summary

Machine(int capabilities)
Create a Machine instance.

Method Summary

abstract void	addCPPFlags(Vector<java.lang.String> v) Add the flags necessary for the C preprocessor.
abstract int	addressableMemoryUnits(int bitSize) Return the number of addressable units required
abstract int	alignData(int dataSize) Return the integer value alignmentthat satisfies (0 == address % alignment) for the data size specified.
static long	alignTo(long value, int to) Round up a value so that it is evenly divisible by the second value.
abstract java.lang.String	determineArchitecture(java.lang.String architecture, java.lang.String extension) Determine the architecture sub-type.
abstract int	executionCostEstimate(double value) Return an estimate of the execution cost to provide this value.
abstract int	executionCostEstimate(long value) Return an estimate of the execution cost to provide this value.
abstract int	generalAlignment() Return the most general purpose alignment in memory units.
abstract java.lang.String	getArchitectureName() Return the name of the specific target architecture.
java.lang.String	getAsmFileExtension() Return the file extension to use for an assembler source file.
java.lang.String	getAssemblerCommand(int backendFeatures) Return the assembler command appropriate to the architecture.
static java.lang.String	getAssemblerCommand(java.lang.String user, int backendFeatures) Return the assembler command to use.
int	getCacheSize(int typeSize) Return the number of elements of the specified size in a cache line.
static Generator	getCodeGenerator(CallGraph cg, int backendFeatures) Return a backend code generator instance for the specified target architecture.
FloatType	getDoubleType() Return the type used for the C double type.
FloatType	getFloatType() Return the type used for the C float type.
byte[]	getFunctionalUnitDescriptions() Return the functional unit information.
abstract java.lang.String	getGenericArchitectureName() Return the name of the generic target architecture.
ICEstimator	getInstructionCountEstimator() Return the proper instruction count estimator for the target architecture.
IntegerType	getIntegerCalcType() Return the type best used for integer calculations.
FloatType	getLongDoubleType() Return the type used for the C long double type.
int	getMaxBlockSize() For architectures that block instructions, this is the maximum number of instructions that can be in a block.
IntegerType	getPtrdifftType() Return the type used for the C ptrdiff_t type.
FloatType	getRealCalcType() Return the type best used for floating point calculations.
IntegerType	getSignedCharType() Return the type used for the C signed char type.
IntegerType	getSignedIntType() Return the type used for the C signed int type.
IntegerType	getSignedLongLongType() Return the type used for the C signed long long type.
IntegerType	getSignedLongType() Return the type used for the C signed long type.
IntegerType	getSignedShortType() Return the type used for the C signed short type.
IntegerType	getSizetType() Return the type used for the C size_t type.
IntegerType	getSmallestAddressableUnitType() Return the type of the smallest addressable unit for this machine.
IntegerType	getUnsignedCharType() Return the type used for the C unsigned char type.
IntegerType	getUnsignedIntType() Return the type used for the C unsigned int type.
IntegerType	getUnsignedLongLongType() Return the type used for the C unsigned long long type.
IntegerType	getUnsignedLongType() Return the type used for the C unsigned long type.
IntegerType	getUnsignedShortType() Return the type used for the C unsigned short type.
Type	getVaListType() Return the type to be used for va_list.
PointerType	getVoidStarType() Return the type used for the C void* type.
IntegerType	getWchartType() Return the type used for the C wchar_t type.
boolean	hasCapabilities(int capabilities) Return true if this architecture has all of the specified capabilities.
boolean	hasCapability(int capability) Return true if this architecture has one of the specified capabilities.
boolean	keepTypeInRegister(Type type, boolean temporary) Return true if a value of the type should be allocated to a register.
abstract boolean	littleEndian() Return true if the machine is little-endian.
int	maxBitFieldSize() Return the maximum bit field size in bits.
protected void	setup() Setup this instance of the machine.
static java.lang.String	setup(java.lang.String architecture) Setup for the specified machine.
static ICEstimator	sGetInstructionCountEstimator() Return the proper instruction count estimator for the target architecture.
abstract int	stackAlignment(Type type) Return the alignment of things stored on the stack.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

HAS_II_CONDITIONAL_MOVE

public static final int HAS_II_CONDITIONAL_MOVE

Capability: Has a conditional move using an integer test and integer values.

See Also:

Constant Field Values

HAS_IF_CONDITIONAL_MOVE

public static final int HAS_IF_CONDITIONAL_MOVE

Capability: Has a conditional move using an integer test and floating point values.

See Also:

Constant Field Values

HAS_FI_CONDITIONAL_MOVE

public static final int HAS_FI_CONDITIONAL_MOVE

Capability: Has a conditional move using a floating point test and integer values.

See Also:

Constant Field Values

HAS_FF_CONDITIONAL_MOVE

public static final int HAS_FF_CONDITIONAL_MOVE

Capability: Has a conditional move using a floating point test and floating point values.

See Also:

Constant Field Values

HAS_INT_FROM_FP_CMP

public static final int HAS_INT_FROM_FP_CMP

Capability: Has a floating point compare that returns an integer value.

See Also:

Constant Field Values

HAS_SIMPLE_FLOOR

public static final int HAS_SIMPLE_FLOOR

Capability: Has a simple inline floor capability.

See Also:

Constant Field Values

HAS_NON_VOLATILE_FP_REGS

public static final int HAS_NON_VOLATILE_FP_REGS

Capability: Has a floating point registers that are saved over function calls.

See Also:

Constant Field Values

HAS_PREDICATION

public static final int HAS_PREDICATION

Capability: Has predicated instructions.

See Also:

Constant Field Values

HAS_WHILE_LOOP_UNROLL_BENEFIT

public static final int HAS_WHILE_LOOP_UNROLL_BENEFIT

Capability: Has benefit from unrolling loops with no induction variable.

See Also:

Constant Field Values

HAS_EXPENSIVE_SUBROUTINE_CALLS

public static final int HAS_EXPENSIVE_SUBROUTINE_CALLS

Capability: bad to flatten/unroll loops that contain a function call.

See Also:

Constant Field Values

HAS_NO_INT_DIVIDE

public static final int HAS_NO_INT_DIVIDE

Capability: Integer divides are not performed by an instruction but by a subroutine call.

See Also:

Constant Field Values

HAS_NO_FP_DIVIDE

public static final int HAS_NO_FP_DIVIDE

Capability: Floating point divides are not performed by an instruction but by a subroutine call.

See Also:

Constant Field Values

ALL_CONDITIONAL_MOVES

public static final int ALL_CONDITIONAL_MOVES

Capability: All the conditional move possibilities.

See Also:

Constant Field Values

currentMachine

public static Machine currentMachine

The current machine specification.

cacheLineSize

protected int cacheLineSize

The size in addressable units of an L1 cache line.

smallestAddressableUnitType

protected IntegerType smallestAddressableUnitType

The type for the smallest addressable unit (e.g., char).

intCalcType

protected IntegerType intCalcType

The integer type best used for integer calculations.

floatCalcType

protected FloatType floatCalcType

The real type best used for floating point calculations.

vaListType

protected Type vaListType

The type to be used for va_list.

signedCharType

protected IntegerType signedCharType

The type used for the C signed char type.

unsignedCharType

protected IntegerType unsignedCharType

The type used for the C unsigned char type.

signedShortType

protected IntegerType signedShortType

The type used for the C signed short type.

unsignedShortType

protected IntegerType unsignedShortType

The type used for the C unsigned short type.

signedIntType

protected IntegerType signedIntType

The type used for the C signed int type.

unsignedIntType

protected IntegerType unsignedIntType

The type used for the C unsigned int type.

signedLongType

protected IntegerType signedLongType

The type used for the C signed long type.

unsignedLongType

protected IntegerType unsignedLongType

The type used for the C unsigned long type.

signedLongLongType

protected IntegerType signedLongLongType

The type used for the C signed long long type.

unsignedLongLongType

protected IntegerType unsignedLongLongType

The type used for the C unsigned long long type.

sizetType

protected IntegerType sizetType

The type used for the C size_t type.

ptrdifftType

protected IntegerType ptrdifftType

The type used for the C ptrdiff_t type.

wchartType

protected IntegerType wchartType

The type used for the C char_t type.

voidStarType

protected PointerType voidStarType

The type used for the C void* type.

floatType

protected FloatType floatType

The type used for the C float type.

doubleType

protected FloatType doubleType

The type used for the C double type.

longDoubleType

protected FloatType longDoubleType

The type used for the C long double type.

Constructor Detail

Machine

public Machine(int capabilities)

Create a Machine instance.

Parameters:

capabilities - specifies special abilities of this architecture that the compiler can use

Method Detail

hasCapability

public final boolean hasCapability(int capability)

Return true if this architecture has one of the specified capabilities.

hasCapabilities

public final boolean hasCapabilities(int capabilities)

Return true if this architecture has all of the specified capabilities.

getSmallestAddressableUnitType

public final IntegerType getSmallestAddressableUnitType()

Return the type of the smallest addressable unit for this machine. For most machines this is type char.

getSignedCharType

public final IntegerType getSignedCharType()

Return the type used for the C signed char type.

getUnsignedCharType

public final IntegerType getUnsignedCharType()

Return the type used for the C unsigned char type.

getSignedShortType

public final IntegerType getSignedShortType()

Return the type used for the C signed short type.

getUnsignedShortType

public final IntegerType getUnsignedShortType()

Return the type used for the C unsigned short type.

getSignedIntType

public final IntegerType getSignedIntType()

Return the type used for the C signed int type.

getUnsignedIntType

public final IntegerType getUnsignedIntType()

Return the type used for the C unsigned int type.

getSignedLongType

public final IntegerType getSignedLongType()

Return the type used for the C signed long type.

getUnsignedLongType

public final IntegerType getUnsignedLongType()

Return the type used for the C unsigned long type.

getSignedLongLongType

public final IntegerType getSignedLongLongType()

Return the type used for the C signed long long type.

getUnsignedLongLongType

public final IntegerType getUnsignedLongLongType()

Return the type used for the C unsigned long long type.

getSizetType

public final IntegerType getSizetType()

Return the type used for the C size_t type.

getPtrdifftType

public final IntegerType getPtrdifftType()

Return the type used for the C ptrdiff_t type.

getWchartType

public final IntegerType getWchartType()

Return the type used for the C wchar_t type.

getVoidStarType

public final PointerType getVoidStarType()

Return the type used for the C void* type.

getFloatType

public final FloatType getFloatType()

Return the type used for the C float type.

getDoubleType

public final FloatType getDoubleType()

Return the type used for the C double type.

getLongDoubleType

public final FloatType getLongDoubleType()

Return the type used for the C long double type.

getIntegerCalcType

public final IntegerType getIntegerCalcType()

Return the type best used for integer calculations. On the Alpha this is a signed 64-bit integer type while on the Sparc V8 it is a signed 32-bit integer type.

getRealCalcType

public final FloatType getRealCalcType()

Return the type best used for floating point calculations. On Trips this is a signed 64-bit real type.

getVaListType

public final Type getVaListType()

Return the type to be used for va_list.

setup

protected void setup()

Setup this instance of the machine. Each code generator should implement this method. For example

   protected void setup()
   {
     super.setup(); // Initialize type, etc that are not set here.
 
     // Specify the types that are different.
 
     PointerType.setMinBitSize(64);
     intCalcType = scale.clef.type.SignedIntegerType.create(64);
     ....
   }
 

alignData

public abstract int alignData(int dataSize)

Return the integer value alignmentthat satisfies (0 == address % alignment) for the data size specified.

Parameters:

dataSize - is the size of the data in addressable units.

addressableMemoryUnits

public abstract int addressableMemoryUnits(int bitSize)

Return the number of addressable units required

Parameters:

bitSize - is the number of bits required for the data

executionCostEstimate

public abstract int executionCostEstimate(long value)

Return an estimate of the execution cost to provide this value. The cost should be zero if the value can be represented as a hard-wired register or in the immediate field of an instruction. The cost is relative to the cost of other operations; a good metric is the number of cycles needed to generate the value or load it from memory.

executionCostEstimate

public abstract int executionCostEstimate(double value)

Return an estimate of the execution cost to provide this value. The cost should be zero if the value can be represented as a hard-wired register or in the immediate field of an instruction. The cost is relative to the cost of other operations; a good metric is the number of cycles needed to generate the value or load it from memory.

maxBitFieldSize

public int maxBitFieldSize()

Return the maximum bit field size in bits. Also, bit fields can not be split over this boundary.

alignTo

public static long alignTo(long value,
                           int to)

Round up a value so that it is evenly divisible by the second value.

Parameters:

value - - returned rounded up

to - - specifies the divisor

Returns:

value rounded up

addCPPFlags

public abstract void addCPPFlags(Vector<java.lang.String> v)

Add the flags necessary for the C preprocessor.

generalAlignment

public abstract int generalAlignment()

Return the most general purpose alignment in memory units.

stackAlignment

public abstract int stackAlignment(Type type)

Return the alignment of things stored on the stack.

littleEndian

public abstract boolean littleEndian()

Return true if the machine is little-endian.

getCacheSize

public int getCacheSize(int typeSize)

Return the number of elements of the specified size in a cache line. Using a value of 1 returns the cache line size in addressable units.

Parameters:

typeSize - is the number of addressable units per element

getGenericArchitectureName

public abstract java.lang.String getGenericArchitectureName()

Return the name of the generic target architecture.

getArchitectureName

public abstract java.lang.String getArchitectureName()

Return the name of the specific target architecture.

determineArchitecture

public abstract java.lang.String determineArchitecture(java.lang.String architecture,
                                                       java.lang.String extension)
                                                throws java.lang.Exception

Determine the architecture sub-type.

Parameters:

architecture - specifies the target architecture

extension - specifies an extension to the a target architecture

Returns:

the name of the specific target architecture generator class.

Throws:

java.lang.Exception - if the extension is not understood

getAsmFileExtension

public java.lang.String getAsmFileExtension()

Return the file extension to use for an assembler source file.

setup

public static java.lang.String setup(java.lang.String architecture)

Setup for the specified machine. The currentMachine is initialized and the assembly generator class is located.

getCodeGenerator

public static Generator getCodeGenerator(CallGraph cg,
                                         int backendFeatures)
                                  throws java.lang.Exception

Return a backend code generator instance for the specified target architecture.

Parameters:

cg - is the call graph containing the CFGs to be converted to target architecture instructions

backendFeatures - specifies certain flags set on the command line

Throws:

java.lang.Exception

getAssemblerCommand

public java.lang.String getAssemblerCommand(int backendFeatures)

Return the assembler command appropriate to the architecture.

Parameters:

backendFeatures - specifies code generation flags such as -g

getAssemblerCommand

public static java.lang.String getAssemblerCommand(java.lang.String user,
                                                   int backendFeatures)

Return the assembler command to use. If a user-specified value is available, use it. Otherwise generate one based on the architecture.

Parameters:

user - is the command specified by the user.

backendFeatures - specifies code generation flags such as -g

keepTypeInRegister

public boolean keepTypeInRegister(Type type,
                                  boolean temporary)

Return true if a value of the type should be allocated to a register. If it is volatile or not an atomic type, it probably should not be kept in a register.

Parameters:

type - is the type

temporary - is true if the duration will be short

getInstructionCountEstimator

public ICEstimator getInstructionCountEstimator()

Return the proper instruction count estimator for the target architecture.

sGetInstructionCountEstimator

public static ICEstimator sGetInstructionCountEstimator()

Return the proper instruction count estimator for the target architecture.

getMaxBlockSize

public int getMaxBlockSize()

For architectures that block instructions, this is the maximum number of instructions that can be in a block. For other architectures, the value 0 is returned.

getFunctionalUnitDescriptions

public byte[] getFunctionalUnitDescriptions()

Return the functional unit information. This information is an array where each element of the array is the number of instructions that can be issued to the corresponding functional unit. This method should be overridden in each architecture.

See Also:

Instruction.getFunctionalUnit()