Writing an LLVM Compiler Backend

This document describes techniques for writing compiler backends -that convert the LLVM IR (intermediate representation) to code for a specified -machine or other languages. Code intended for a specific machine can take the -form of either assembly code or binary code (usable for a JIT compiler).

- -

The backend of LLVM features a target-independent code generator -that may create output for several types of target CPUs, including X86, -PowerPC, Alpha, and SPARC. The backend may also be used to generate code -targeted at SPUs of the Cell processor or GPUs to support the execution of -compute kernels.

- -

The document focuses on existing examples found in subdirectories -of llvm/lib/Target in a downloaded LLVM release. In particular, this document -focuses on the example of creating a static compiler (one that emits text -assembly) for a SPARC target, because SPARC has fairly standard +

+ +

+This document describes techniques for writing compiler backends that convert +the LLVM Intermediate Representation (IR) to code for a specified machine or +other languages. Code intended for a specific machine can take the form of +either assembly code or binary code (usable for a JIT compiler). +

+ +

+The backend of LLVM features a target-independent code generator that may create +output for several types of target CPUs — including X86, PowerPC, ARM, +and SPARC. The backend may also be used to generate code targeted at SPUs of the +Cell processor or GPUs to support the execution of compute kernels. +

+ +

+The document focuses on existing examples found in subdirectories +of llvm/lib/Target in a downloaded LLVM release. In particular, this +document focuses on the example of creating a static compiler (one that emits +text assembly) for a SPARC target, because SPARC has fairly standard characteristics, such as a RISC instruction set and straightforward calling -conventions.

+conventions. +

Audience -

+ + +

+ +

+The audience for this document is anyone who needs to write an LLVM backend to +generate code for a specific hardware or software target. +

The audience for this document is anyone who needs to write an -LLVM backend to generate code for a specific hardware or software target.

Prerequisite Reading -

+ + +

+ +

+These essential documents must be read before reading this document: +

-These essential documents must be read before reading this document:

-LLVM Language Reference Manual - -a reference manual for the LLVM assembly language -
-The LLVM Target-Independent Code Generator - -a guide to the components (classes and code generation algorithms) for translating -the LLVM internal representation to the machine code for a specified target. -Pay particular attention to the descriptions of code generation stages: -Instruction Selection, Scheduling and Formation, SSA-based Optimization, -Register Allocation, Prolog/Epilog Code Insertion, Late Machine Code Optimizations, -and Code Emission. -
-TableGen Fundamentals - -a document that describes the TableGen (tblgen) application that manages domain-specific -information to support LLVM code generation. TableGen processes input from a -target description file (.td suffix) and generates C++ code that can be used -for code generation. -
-Writing an LLVM Pass - -The assembly printer is a FunctionPass, as are several SelectionDAG processing steps. -
LLVM Language Reference + Manual — a reference manual for the LLVM assembly language.
The LLVM + Target-Independent Code Generator — a guide to the components + (classes and code generation algorithms) for translating the LLVM internal + representation into machine code for a specified target. Pay particular + attention to the descriptions of code generation stages: Instruction + Selection, Scheduling and Formation, SSA-based Optimization, Register + Allocation, Prolog/Epilog Code Insertion, Late Machine Code Optimizations, + and Code Emission.
TableGen + Fundamentals —a document that describes the TableGen + (tblgen) application that manages domain-specific information to + support LLVM code generation. TableGen processes input from a target + description file (.td suffix) and generates C++ code that can be + used for code generation.
Writing an LLVM + Pass — The assembly printer is a FunctionPass, as are + several SelectionDAG processing steps.

-To follow the SPARC examples in this document, have a copy of -The SPARC Architecture Manual, Version 8 -for reference. For details about the ARM instruction set, refer to the -ARM Architecture Reference Manual -For more about the GNU Assembler format (GAS), see -Using As -especially for the assembly printer. Using As contains lists of target machine dependent features. + +

+To follow the SPARC examples in this document, have a copy of +The SPARC Architecture +Manual, Version 8 for reference. For details about the ARM instruction +set, refer to the ARM Architecture +Reference Manual. For more about the GNU Assembler format +(GAS), see +Using As, +especially for the assembly printer. Using As contains a list of target +machine dependent features. +

Basic Steps -

To write a compiler -backend for LLVM that converts the LLVM IR (intermediate representation) -to code for a specified target (machine or other language), follow these steps:

+ + +

+ +

+To write a compiler backend for LLVM that converts the LLVM IR to code for a +specified target (machine or other language), follow these steps: +

-Create a subclass of the TargetMachine class that describes -characteristics of your target machine. Copy existing examples of specific -TargetMachine class and header files; for example, start with SparcTargetMachine.cpp -and SparcTargetMachine.h, but change the file names for your target. Similarly, -change code that references "Sparc" to reference your target.
Describe the register set of the target. Use TableGen to generate -code for register definition, register aliases, and register classes from a -target-specific RegisterInfo.td input file. You should also write additional -code for a subclass of TargetRegisterInfo class that represents the class -register file data used for register allocation and also describes the -interactions between registers.
Describe the instruction set of the target. Use TableGen to -generate code for target-specific instructions from target-specific versions of -TargetInstrFormats.td and TargetInstrInfo.td. You should write additional code -for a subclass of the TargetInstrInfo -class to represent machine -instructions supported by the target machine.
Describe the selection and conversion of the LLVM IR from a DAG (directed -acyclic graph) representation of instructions to native target-specific -instructions. Use TableGen to generate code that matches patterns and selects -instructions based on additional information in a target-specific version of -TargetInstrInfo.td. Write code for XXXISelDAGToDAG.cpp -(where XXX identifies the specific target) to perform pattern -matching and DAG-to-DAG instruction selection. Also write code in XXXISelLowering.cpp -to replace or remove operations and data types that are not supported natively -in a SelectionDAG.
Write code for an -assembly printer that converts LLVM IR to a GAS format for your target machine. -You should add assembly strings to the instructions defined in your -target-specific version of TargetInstrInfo.td. You should also write code for a -subclass of AsmPrinter that performs the LLVM-to-assembly conversion and a -trivial subclass of TargetAsmInfo.
Optionally, add support for subtargets (that is, variants with -different capabilities). You should also write code for a subclass of the -TargetSubtarget class, which allows you to use the -mcpu= -and -mattr= command-line options.
Optionally, add JIT support and create a machine code emitter (subclass -of TargetJITInfo) that is used to emit binary code directly into memory.
Create a subclass of the TargetMachine class that describes characteristics + of your target machine. Copy existing examples of specific TargetMachine + class and header files; for example, start with + SparcTargetMachine.cpp and SparcTargetMachine.h, but + change the file names for your target. Similarly, change code that + references "Sparc" to reference your target.
Describe the register set of the target. Use TableGen to generate code for + register definition, register aliases, and register classes from a + target-specific RegisterInfo.td input file. You should also write + additional code for a subclass of the TargetRegisterInfo class that + represents the class register file data used for register allocation and + also describes the interactions between registers.
Describe the instruction set of the target. Use TableGen to generate code + for target-specific instructions from target-specific versions of + TargetInstrFormats.td and TargetInstrInfo.td. You should + write additional code for a subclass of the TargetInstrInfo class to + represent machine instructions supported by the target machine.
Describe the selection and conversion of the LLVM IR from a Directed Acyclic + Graph (DAG) representation of instructions to native target-specific + instructions. Use TableGen to generate code that matches patterns and + selects instructions based on additional information in a target-specific + version of TargetInstrInfo.td. Write code + for XXXISelDAGToDAG.cpp, where XXX identifies the specific target, + to perform pattern matching and DAG-to-DAG instruction selection. Also write + code in XXXISelLowering.cpp to replace or remove operations and + data types that are not supported natively in a SelectionDAG.
Write code for an assembly printer that converts LLVM IR to a GAS format for + your target machine. You should add assembly strings to the instructions + defined in your target-specific version of TargetInstrInfo.td. You + should also write code for a subclass of AsmPrinter that performs the + LLVM-to-assembly conversion and a trivial subclass of TargetAsmInfo.
Optionally, add support for subtargets (i.e., variants with different + capabilities). You should also write code for a subclass of the + TargetSubtarget class, which allows you to use the -mcpu= + and -mattr= command-line options.
Optionally, add JIT support and create a machine code emitter (subclass of + TargetJITInfo) that is used to emit binary code directly into memory.

In the .cpp and .h files, initially stub up these methods and +

+In the .cpp and .h. files, initially stub up these methods and then implement them later. Initially, you may not know which private members -that the class will need and which components will need to be subclassed.

+that the class will need and which components will need to be subclassed. +

Preliminaries -

To actually create -your compiler backend, you need to create and modify a few files. The absolute -minimum is discussed here, but to actually use the LLVM target-independent code -generator, you must perform the steps described in the LLVM -Target-Independent Code Generator document.

- -

First, you should -create a subdirectory under lib/Target to hold all the files related to your -target. If your target is called "Dummy", create the directory -lib/Target/Dummy.

- -

In this new -directory, create a Makefile. It is easiest to copy a Makefile of another -target and modify it. It should at least contain the LEVEL, LIBRARYNAME and -TARGET variables, and then include $(LEVEL)/Makefile.common. The library can be -named LLVMDummy (for example, see the MIPS target). Alternatively, you can -split the library into LLVMDummyCodeGen and LLVMDummyAsmPrinter, the latter of -which should be implemented in a subdirectory below lib/Target/Dummy (for -example, see the PowerPC target).

- -

Note that these two -naming schemes are hardcoded into llvm-config. Using any other naming scheme -will confuse llvm-config and produce lots of (seemingly unrelated) linker -errors when linking llc.

- -

To make your target -actually do something, you need to implement a subclass of TargetMachine. This -implementation should typically be in the file -lib/Target/DummyTargetMachine.cpp, but any file in the lib/Target directory will -be built and should work. To use LLVM's target -independent code generator, you should do what all current machine backends do: create a subclass -of LLVMTargetMachine. (To create a target from scratch, create a subclass of -TargetMachine.)

- -

To get LLVM to -actually build and link your target, you need to add it to the TARGETS_TO_BUILD -variable. To do this, you modify the configure script to know about your target -when parsing the --enable-targets option. Search the configure script for TARGETS_TO_BUILD, -add your target to the lists there (some creativity required) and then + + +

+ +

+To actually create your compiler backend, you need to create and modify a few +files. The absolute minimum is discussed here. But to actually use the LLVM +target-independent code generator, you must perform the steps described in +the LLVM +Target-Independent Code Generator document. +

+ +

+First, you should create a subdirectory under lib/Target to hold all +the files related to your target. If your target is called "Dummy," create the +directory lib/Target/Dummy. +

+ +

+In this new +directory, create a Makefile. It is easiest to copy a +Makefile of another target and modify it. It should at least contain +the LEVEL, LIBRARYNAME and TARGET variables, and then +include $(LEVEL)/Makefile.common. The library can be +named LLVMDummy (for example, see the MIPS target). Alternatively, you +can split the library into LLVMDummyCodeGen +and LLVMDummyAsmPrinter, the latter of which should be implemented in a +subdirectory below lib/Target/Dummy (for example, see the PowerPC +target). +

+ +

+Note that these two naming schemes are hardcoded into llvm-config. +Using any other naming scheme will confuse llvm-config and produce a +lot of (seemingly unrelated) linker errors when linking llc. +

+ +

+To make your target actually do something, you need to implement a subclass of +TargetMachine. This implementation should typically be in the file +lib/Target/DummyTargetMachine.cpp, but any file in +the lib/Target directory will be built and should work. To use LLVM's +target independent code generator, you should do what all current machine +backends do: create a subclass of LLVMTargetMachine. (To create a +target from scratch, create a subclass of TargetMachine.) +

+ +

+To get LLVM to actually build and link your target, you need to add it to +the TARGETS_TO_BUILD variable. To do this, you modify the configure +script to know about your target when parsing the --enable-targets +option. Search the configure script for TARGETS_TO_BUILD, add your +target to the lists there (some creativity required), and then reconfigure. Alternatively, you can change autotools/configure.ac and -regenerate configure by running ./autoconf/AutoRegen.sh

+regenerate configure by running ./autoconf/AutoRegen.sh. +

+ +

Target Machine -

+ -

LLVMTargetMachine is designed as a base class for targets -implemented with the LLVM target-independent code generator. The -LLVMTargetMachine class should be specialized by a concrete target class that -implements the various virtual methods. LLVMTargetMachine is defined as a -subclass of TargetMachine in include/llvm/Target/TargetMachine.h. The -TargetMachine class implementation (TargetMachine.cpp) also processes numerous -command-line options.

- -

To create a concrete target-specific subclass of -LLVMTargetMachine, start by copying an existing TargetMachine class and header. -You should name the files that you create to reflect your specific target. For + +

+ +

+LLVMTargetMachine is designed as a base class for targets implemented +with the LLVM target-independent code generator. The LLVMTargetMachine +class should be specialized by a concrete target class that implements the +various virtual methods. LLVMTargetMachine is defined as a subclass of +TargetMachine in include/llvm/Target/TargetMachine.h. The +TargetMachine class implementation (TargetMachine.cpp) also +processes numerous command-line options. +

+ +

+To create a concrete target-specific subclass of LLVMTargetMachine, +start by copying an existing TargetMachine class and header. You +should name the files that you create to reflect your specific target. For instance, for the SPARC target, name the files SparcTargetMachine.h and -SparcTargetMachine.cpp

+SparcTargetMachine.cpp. +

For a target machine XXX, the implementation of XXXTargetMachine -must have access methods to obtain objects that represent target components. -These methods are named get*Info and are intended to obtain the instruction set -(getInstrInfo), register set (getRegisterInfo), stack frame layout -(getFrameInfo), and similar information. XXXTargetMachine must also implement -the getTargetData method to access an object with target-specific data -characteristics, such as data type size and alignment requirements.

+For a target machine XXX, the implementation of +XXXTargetMachine must have access methods to obtain objects that +represent target components. These methods are named get*Info, and are +intended to obtain the instruction set (getInstrInfo), register set +(getRegisterInfo), stack frame layout (getFrameInfo), and +similar information. XXXTargetMachine must also implement the +getTargetData method to access an object with target-specific data +characteristics, such as data type size and alignment requirements. +

For instance, for the SPARC target, the header file SparcTargetMachine.h -declares prototypes for several get*Info and getTargetData methods that simply -return a class member.

+For instance, for the SPARC target, the header file +SparcTargetMachine.h declares prototypes for several get*Info +and getTargetData methods that simply return a class member. +

namespace llvm {
++namespace llvm {
 
 class Module;
 
@@ -286,8 +337,7 @@ class SparcTargetMachine : public LLVMTargetMachine {
   TargetFrameInfo FrameInfo;
   
 protected:
-  virtual const TargetAsmInfo *createTargetAsmInfo()
-const;
+  virtual const TargetAsmInfo *createTargetAsmInfo() const;
   
 public:
   SparcTargetMachine(const Module &M, const std::string &FS);
@@ -304,118 +354,177 @@ public:
   // Pass Pipeline Configuration
   virtual bool addInstSelector(PassManagerBase &PM, bool Fast);
   virtual bool addPreEmitPass(PassManagerBase &PM, bool Fast);
-  virtual bool addAssemblyEmitter(PassManagerBase &PM, bool Fast, 
-                                  std::ostream &Out);
 };
 
 } // end namespace llvm

getInstrInfo
getRegisterInfo
getFrameInfo
getTargetData
getSubtargetImpl
getInstrInfo()
getRegisterInfo()
getFrameInfo()
getTargetData()
getSubtargetImpl()

For some targets, you also need to support the following methods: -

+ +

For some targets, you also need to support the following methods:

getTargetLowering
getJITInfo
getTargetLowering()
getJITInfo()

In addition, the XXXTargetMachine constructor should specify a -TargetDescription string that determines the data layout for the target machine, -including characteristics such as pointer size, alignment, and endianness. For -example, the constructor for SparcTargetMachine contains the following:

+ +

+In addition, the XXXTargetMachine constructor should specify a +TargetDescription string that determines the data layout for the target +machine, including characteristics such as pointer size, alignment, and +endianness. For example, the constructor for SparcTargetMachine contains the +following: +

 SparcTargetMachine::SparcTargetMachine(const Module &M, const std::string &FS)
-  : DataLayout("E-p:32:32-f128:128:128"),
+  : DataLayout("E-p:32:32-f128:128:128"),
     Subtarget(M, FS), InstrInfo(Subtarget),
     FrameInfo(TargetFrameInfo::StackGrowsDown, 8, 0) {
 }

Hyphens separate portions of the TargetDescription string.

Hyphens separate portions of the TargetDescription string.

The "E" in the string indicates a big-endian target data model; a -lower-case "e" would indicate little-endian.
"p:" is followed by pointer information: size, ABI alignment, and -preferred alignment. If only two figures follow "p:", then the first value is -pointer size, and the second value is both ABI and preferred alignment.
then a letter for numeric type alignment: "i", "f", "v", or "a" -(corresponding to integer, floating point, vector, or aggregate). "i", "v", or -"a" are followed by ABI alignment and preferred alignment. "f" is followed by -three values, the first indicates the size of a long double, then ABI alignment -and preferred alignment.
An upper-case "E" in the string indicates a big-endian target data + model. a lower-case "e" indicates little-endian.
"p:" is followed by pointer information: size, ABI alignment, and + preferred alignment. If only two figures follow "p:", then the + first value is pointer size, and the second value is both ABI and preferred + alignment.
Then a letter for numeric type alignment: "i", "f", + "v", or "a" (corresponding to integer, floating point, + vector, or aggregate). "i", "v", or "a" are + followed by ABI alignment and preferred alignment. "f" is followed + by three values: the first indicates the size of a long double, then ABI + alignment, and then ABI preferred alignment.

You must also register your target using the RegisterTarget -template. (See the TargetMachineRegistry class.) For example, in SparcTargetMachine.cpp, -the target is registered with:

+ +

+ + +

+ Target Registration +

+ + +

+ +

+You must also register your target with the TargetRegistry, which is +what other LLVM tools use to be able to lookup and use your target at +runtime. The TargetRegistry can be used directly, but for most targets +there are helper templates which should take care of the work for you.

+ +

+All targets should declare a global Target object which is used to +represent the target during registration. Then, in the target's TargetInfo +library, the target should define that object and use +the RegisterTarget template to register the target. For example, the Sparc registration code looks like this: +

+ +

+Target llvm::TheSparcTarget;
+
+extern "C" void LLVMInitializeSparcTargetInfo() { 
+  RegisterTarget<Triple::sparc, /*HasJIT=*/false>
+    X(TheSparcTarget, "sparc", "Sparc");
+}
+

+This allows the TargetRegistry to look up the target by name or by +target triple. In addition, most targets will also register additional features +which are available in separate libraries. These registration steps are +separate, because some clients may wish to only link in some parts of the target +-- the JIT code generator does not require the use of the assembler printer, for +example. Here is an example of registering the Sparc assembly printer: +

-namespace {
-  // Register the target.
-  RegisterTarget<SparcTargetMachine>X("sparc", "SPARC");
+extern "C" void LLVMInitializeSparcAsmPrinter() { 
+  RegisterAsmPrinter<SparcAsmPrinter> X(TheSparcTarget);
 }

+For more information, see +"llvm/Target/TargetRegistry.h". +

+ +

+ -

Register Set and Register Classes -

+ -

You should describe -a concrete target-specific class -that represents the register file of a target machine. This class is -called XXXRegisterInfo (where XXX identifies the target) and represents the -class register file data that is used for register allocation and also -describes the interactions between registers.

- -

You also need to -define register classes to categorize related registers. A register class -should be added for groups of registers that are all treated the same way for -some instruction. Typical examples are register classes that include integer, -floating-point, or vector registers. A register allocator allows an + +

+ +

+You should describe a concrete target-specific class that represents the +register file of a target machine. This class is called XXXRegisterInfo +(where XXX identifies the target) and represents the class register +file data that is used for register allocation. It also describes the +interactions between registers. +

+ +

+You also need to define register classes to categorize related registers. A +register class should be added for groups of registers that are all treated the +same way for some instruction. Typical examples are register classes for +integer, floating-point, or vector registers. A register allocator allows an instruction to use any register in a specified register class to perform the instruction in a similar manner. Register classes allocate virtual registers to instructions from these sets, and register classes let the target-independent -register allocator automatically choose the actual registers.

+register allocator automatically choose the actual registers. +

Much of the code for registers, including register definition, -register aliases, and register classes, is generated by TableGen from -XXXRegisterInfo.td input files and placed in XXXGenRegisterInfo.h.inc and -XXXGenRegisterInfo.inc output files. Some of the code in the implementation of -XXXRegisterInfo requires hand-coding.

+Much of the code for registers, including register definition, register aliases, +and register classes, is generated by TableGen from XXXRegisterInfo.td +input files and placed in XXXGenRegisterInfo.h.inc and +XXXGenRegisterInfo.inc output files. Some of the code in the +implementation of XXXRegisterInfo requires hand-coding. +

Defining a Register -

The XXXRegisterInfo.td file typically starts with register definitions -for a target machine. The Register class (specified in Target.td) is used to -define an object for each register. The specified string n becomes the Name of -the register. The basic Register object does not have any subregisters and does -not specify any aliases.

+ + +

+ +

+The XXXRegisterInfo.td file typically starts with register definitions +for a target machine. The Register class (specified +in Target.td) is used to define an object for each register. The +specified string n becomes the Name of the register. The +basic Register object does not have any subregisters and does not +specify any aliases. +

 class Register<string n> {
-  string Namespace = "";
+  string Namespace = "";
   string AsmName = n;
   string Name = n;
   int SpillSize = 0;
@@ -427,47 +536,50 @@ class Register<string n> {

For example, in the X86RegisterInfo.td file, there are register -definitions that utilize the Register class, such as:

+For example, in the X86RegisterInfo.td file, there are register +definitions that utilize the Register class, such as: +

-def AL : Register<"AL">,
-DwarfRegNum<[0, 0, 0]>;
+def AL : Register<"AL">, DwarfRegNum<[0, 0, 0]>;

This defines the register AL and assigns it values (with -DwarfRegNum) that are used by gcc, gdb, or a debug information writer (such as -DwarfWriter in llvm/lib/CodeGen) to identify a register. For register AL, -DwarfRegNum takes an array of 3 values, representing 3 different modes: the -first element is for X86-64, the second for EH (exception handling) on X86-32, -and the third is generic. -1 is a special Dwarf number that indicates the gcc -number is undefined, and -2 indicates the register number is invalid for this -mode.

+This defines the register AL and assigns it values (with +DwarfRegNum) that are used by gcc, gdb, or a debug +information writer to identify a register. For register +AL, DwarfRegNum takes an array of 3 values representing 3 +different modes: the first element is for X86-64, the second for exception +handling (EH) on X86-32, and the third is generic. -1 is a special Dwarf number +that indicates the gcc number is undefined, and -2 indicates the register number +is invalid for this mode. +

+ +

+From the previously described line in the X86RegisterInfo.td file, +TableGen generates this code in the X86GenRegisterInfo.inc file: +

From the previously described line in the X86RegisterInfo.td -file, TableGen generates this code in the X86GenRegisterInfo.inc file:

-  static const unsigned GR8[] = { X86::AL, ... };
- 
-  const unsigned AL_AliasSet[] = { X86::AX, X86::EAX, X86::RAX, 0 };
- 
-  const TargetRegisterDesc RegisterDescriptors[] = { 
-    ...
-    { "AL", "AL", AL_AliasSet, Empty_SubRegsSet, Empty_SubRegsSet, AL_SuperRegsSet }, ...
+static const unsigned GR8[] = { X86::AL, ... };
+
+const unsigned AL_AliasSet[] = { X86::AX, X86::EAX, X86::RAX, 0 };
+
+const TargetRegisterDesc RegisterDescriptors[] = { 
+  ...
+{ "AL", "AL", AL_AliasSet, Empty_SubRegsSet, Empty_SubRegsSet, AL_SuperRegsSet }, ...

From the register info file, TableGen generates a -TargetRegisterDesc object for each register. TargetRegisterDesc is defined in -include/llvm/Target/TargetRegisterInfo.h with the following fields:

+From the register info file, TableGen generates a TargetRegisterDesc +object for each register. TargetRegisterDesc is defined in +include/llvm/Target/TargetRegisterInfo.h with the following fields: +

@@ -481,41 +593,47 @@ struct TargetRegisterDesc {
 };

TableGen uses the entire target description file (.td) to -determine text names for the register (in the AsmName and Name fields of -TargetRegisterDesc) and the relationships of other registers to the defined -register (in the other TargetRegisterDesc fields). In this example, other -definitions establish the registers "AX", "EAX", and "RAX" as aliases for one -another, so TableGen generates a null-terminated array (AL_AliasSet) for this -register alias set.

+TableGen uses the entire target description file (.td) to determine +text names for the register (in the AsmName and Name fields of +TargetRegisterDesc) and the relationships of other registers to the +defined register (in the other TargetRegisterDesc fields). In this +example, other definitions establish the registers "AX", +"EAX", and "RAX" as aliases for one another, so TableGen +generates a null-terminated array (AL_AliasSet) for this register alias +set. +

+ +

+The Register class is commonly used as a base class for more complex +classes. In Target.td, the Register class is the base for the +RegisterWithSubRegs class that is used to define registers that need to +specify subregisters in the SubRegs list, as shown here: +

The Register class is commonly used as a base class for more -complex classes. In Target.td, the Register class is the base for the -RegisterWithSubRegs class that is used to define registers that need to specify -subregisters in the SubRegs list, as shown here:

 class RegisterWithSubRegs<string n,
 list<Register> subregs> : Register<n> {
   let SubRegs = subregs;
-}

+} +

In SparcRegisterInfo.td, additional register classes are defined -for SPARC: a Register subclass, SparcReg, and further subclasses: Ri, Rf, and -Rd. SPARC registers are identified by 5-bit ID numbers, which is a feature -common to these subclasses. Note the use of ‘let’ expressions to override values -that are initially defined in a superclass (such as SubRegs field in the Rd -class).

+In SparcRegisterInfo.td, additional register classes are defined for +SPARC: a Register subclass, SparcReg, and further subclasses: Ri, +Rf, and Rd. SPARC registers are identified by 5-bit ID +numbers, which is a feature common to these subclasses. Note the use of +'let' expressions to override values that are initially defined in a +superclass (such as SubRegs field in the Rd class). +

 class SparcReg<string n> : Register<n> {
   field bits<5> Num;
-  let Namespace = "SP";
+  let Namespace = "SP";
 }
 // Ri - 32-bit integer registers
 class Ri<bits<5> num, string n> :
@@ -533,315 +651,344 @@ class Rd<bits<5> num, string n,
 list<Register> subregs> : SparcReg<n> {
   let Num = num;
   let SubRegs = subregs;
-}

In the SparcRegisterInfo.td file, there are register definitions -that utilize these subclasses of Register, such as:

+} +

+ +

+In the SparcRegisterInfo.td file, there are register definitions that +utilize these subclasses of Register, such as: +

-def G0 : Ri< 0, "G0">,
+def G0 : Ri< 0, "G0">,
 DwarfRegNum<[0]>;
-def G1 : Ri< 1, "G1">, DwarfRegNum<[1]>;
+def G1 : Ri< 1, "G1">, DwarfRegNum<[1]>;
 ...
-def F0 : Rf< 0, "F0">,
+def F0 : Rf< 0, "F0">,
 DwarfRegNum<[32]>;
-def F1 : Rf< 1, "F1">,
+def F1 : Rf< 1, "F1">,
 DwarfRegNum<[33]>;
 ...
-def D0 : Rd< 0, "F0", [F0, F1]>,
+def D0 : Rd< 0, "F0", [F0, F1]>,
 DwarfRegNum<[32]>;
-def D1 : Rd< 2, "F2", [F2, F3]>,
+def D1 : Rd< 2, "F2", [F2, F3]>,
 DwarfRegNum<[34]>;

The last two registers shown above (D0 and D1) are double-precision -floating-point registers that are aliases for pairs of single-precision -floating-point sub-registers. In addition to aliases, the sub-register and -super-register relationships of the defined register are in fields of a -register’s TargetRegisterDesc.

+ +

+The last two registers shown above (D0 and D1) are +double-precision floating-point registers that are aliases for pairs of +single-precision floating-point sub-registers. In addition to aliases, the +sub-register and super-register relationships of the defined register are in +fields of a register's TargetRegisterDesc. +

Defining a Register Class -

The RegisterClass class (specified in Target.td) is used to + + +

+ +

+The RegisterClass class (specified in Target.td) is used to define an object that represents a group of related registers and also defines the default allocation order of the registers. A target description file -XXXRegisterInfo.td that uses Target.td can construct register classes using the -following class:

+XXXRegisterInfo.td that uses Target.td can construct register +classes using the following class: +

 class RegisterClass<string namespace,
-list<ValueType> regTypes, int alignment,
-                    list<Register> regList> {
+list<ValueType> regTypes, int alignment, dag regList> {
   string Namespace = namespace;
   list<ValueType> RegTypes = regTypes;
   int Size = 0;  // spill size, in bits; zero lets tblgen pick the size
   int Alignment = alignment;
- 
+
   // CopyCost is the cost of copying a value between two registers
   // default value 1 means a single instruction
   // A negative value means copying is extremely expensive or impossible
   int CopyCost = 1;  
-  list<Register> MemberList = regList;
+  dag MemberList = regList;
   
   // for register classes that are subregisters of this class
   list<RegisterClass> SubRegClassList = [];  
   
   code MethodProtos = [{}];  // to insert arbitrary code
   code MethodBodies = [{}];
-}

+} +

To define a RegisterClass, use the following 4 arguments:

The first argument of the definition is the name of the -namespace.
The second argument is a list of ValueType register type values -that are defined in include/llvm/CodeGen/ValueTypes.td. Defined values include -integer types (such as i16, i32, and i1 for Boolean), floating-point types -(f32, f64), and vector types (for example, v8i16 for an 8 x i16 vector). All -registers in a RegisterClass must have the same ValueType, but some registers -may store vector data in different configurations. For example a register that -can process a 128-bit vector may be able to handle 16 8-bit integer elements, 8 -16-bit integers, 4 32-bit integers, and so on.
The third argument of the RegisterClass definition specifies the -alignment required of the registers when they are stored or loaded to memory.
The final argument, regList, specifies which registers are in -this class. If an allocation_order_* method is not specified, then regList also -defines the order of allocation used by the register allocator.
The first argument of the definition is the name of the namespace.
The second argument is a list of ValueType register type values + that are defined in include/llvm/CodeGen/ValueTypes.td. Defined + values include integer types (such as i16, i32, + and i1 for Boolean), floating-point types + (f32, f64), and vector types (for example, v8i16 + for an 8 x i16 vector). All registers in a RegisterClass + must have the same ValueType, but some registers may store vector + data in different configurations. For example a register that can process a + 128-bit vector may be able to handle 16 8-bit integer elements, 8 16-bit + integers, 4 32-bit integers, and so on.
The third argument of the RegisterClass definition specifies the + alignment required of the registers when they are stored or loaded to + memory.
The final argument, regList, specifies which registers are in this + class. If an alternative allocation order method is not specified, then + regList also defines the order of allocation used by the register + allocator. Besides simply listing registers with (add R0, R1, ...), + more advanced set operators are available. See + include/llvm/Target/Target.td for more information.

In SparcRegisterInfo.td, three RegisterClass objects are defined: -FPRegs, DFPRegs, and IntRegs. For all three register classes, the first -argument defines the namespace with the string “SP”. FPRegs defines a group of 32 -single-precision floating-point registers (F0 to F31); DFPRegs defines a group -of 16 double-precision registers (D0-D15). For IntRegs, the MethodProtos and -MethodBodies methods are used by TableGen to insert the specified code into generated -output.

+In SparcRegisterInfo.td, three RegisterClass objects are defined: +FPRegs, DFPRegs, and IntRegs. For all three register +classes, the first argument defines the namespace with the string +'SP'. FPRegs defines a group of 32 single-precision +floating-point registers (F0 to F31); DFPRegs defines +a group of 16 double-precision registers +(D0-D15). +

-def FPRegs : RegisterClass<"SP", [f32], 32, [F0, F1, F2, F3, F4, F5, F6, F7,   
-  F8, F9, F10, F11, F12, F13, F14, F15, F16, F17, F18, F19, F20, F21, F22,
-  F23, F24, F25, F26, F27, F28, F29, F30, F31]>;
- 
-def DFPRegs : RegisterClass<"SP", [f64], 64, [D0, D1, D2, D3, D4, D5, D6, D7,
-  D8, D9, D10, D11, D12, D13, D14, D15]>;
+// F0, F1, F2, ..., F31
+def FPRegs : RegisterClass<"SP", [f32], 32, (sequence "F%u", 0, 31)>;
+
+def DFPRegs : RegisterClass<"SP", [f64], 64,
+                            (add D0, D1, D2, D3, D4, D5, D6, D7, D8,
+                                 D9, D10, D11, D12, D13, D14, D15)>;
  
-def IntRegs : RegisterClass<"SP", [i32], 32, [L0, L1, L2, L3, L4, L5, L6, L7,
-                                     I0, I1, I2, I3, I4, I5,
-                                     O0, O1, O2, O3, O4, O5, O7,
-                                     G1,
-                                     // Non-allocatable regs:
-                                     G2, G3, G4, 
-                                     O6, // stack ptr
-                                     I6, // frame ptr
-                                     I7, // return address
-                                     G0, // constant zero
-                                     G5, G6, G7 // reserved for kernel
-                                     ]> {
-  let MethodProtos = [{
-    iterator allocation_order_end(const MachineFunction &MF) const;
-  }];
-  let MethodBodies = [{
-    IntRegsClass::iterator
-    IntRegsClass::allocation_order_end(const MachineFunction &MF) const {
-      return end()-10  // Don't allocate special registers
-         -1;  
-    }
-  }];
-}
+def IntRegs : RegisterClass<"SP", [i32], 32,
+    (add L0, L1, L2, L3, L4, L5, L6, L7,
+         I0, I1, I2, I3, I4, I5,
+         O0, O1, O2, O3, O4, O5, O7,
+         G1,
+         // Non-allocatable regs:
+         G2, G3, G4,
+         O6,        // stack ptr
+         I6,        // frame ptr
+         I7,        // return address
+         G0,        // constant zero
+         G5, G6, G7 // reserved for kernel
+    )>;

Using SparcRegisterInfo.td with TableGen generates several output -files that are intended for inclusion in other source code that you write. -SparcRegisterInfo.td generates SparcGenRegisterInfo.h.inc, which should be -included in the header file for the implementation of the SPARC register -implementation that you write (SparcRegisterInfo.h). In +

+Using SparcRegisterInfo.td with TableGen generates several output files +that are intended for inclusion in other source code that you write. +SparcRegisterInfo.td generates SparcGenRegisterInfo.h.inc, +which should be included in the header file for the implementation of the SPARC +register implementation that you write (SparcRegisterInfo.h). In SparcGenRegisterInfo.h.inc a new structure is defined called -SparcGenRegisterInfo that uses TargetRegisterInfo as its base. It also -specifies types, based upon the defined register classes: DFPRegsClass, FPRegsClass, -and IntRegsClass.

+SparcGenRegisterInfo that uses TargetRegisterInfo as its +base. It also specifies types, based upon the defined register +classes: DFPRegsClass, FPRegsClass, and IntRegsClass. +

SparcRegisterInfo.td also generates SparcGenRegisterInfo.inc, -which is included at the bottom of SparcRegisterInfo.cpp, the SPARC register -implementation. The code below shows only the generated integer registers and -associated register classes. The order of registers in IntRegs reflects the -order in the definition of IntRegs in the target description file. Take special -note of the use of MethodBodies in SparcRegisterInfo.td to create code in -SparcGenRegisterInfo.inc. MethodProtos generates similar code in -SparcGenRegisterInfo.h.inc.

+SparcRegisterInfo.td also generates SparcGenRegisterInfo.inc, +which is included at the bottom of SparcRegisterInfo.cpp, the SPARC +register implementation. The code below shows only the generated integer +registers and associated register classes. The order of registers +in IntRegs reflects the order in the definition of IntRegs in +the target description file. +

  // IntRegs Register Class...
   static const unsigned IntRegs[] = {
     SP::L0, SP::L1, SP::L2, SP::L3, SP::L4, SP::L5,
-SP::L6, SP::L7, SP::I0, SP::I1, SP::I2, SP::I3, SP::I4, SP::I5, SP::O0, SP::O1,
-SP::O2, SP::O3, SP::O4, SP::O5, SP::O7, SP::G1, SP::G2, SP::G3, SP::G4, SP::O6,
-SP::I6, SP::I7, SP::G0, SP::G5, SP::G6, SP::G7, 
+    SP::L6, SP::L7, SP::I0, SP::I1, SP::I2, SP::I3,
+    SP::I4, SP::I5, SP::O0, SP::O1, SP::O2, SP::O3,
+    SP::O4, SP::O5, SP::O7, SP::G1, SP::G2, SP::G3,
+    SP::G4, SP::O6, SP::I6, SP::I7, SP::G0, SP::G5,
+    SP::G6, SP::G7, 
   };
- 
+
   // IntRegsVTs Register Class Value Types...
   static const MVT::ValueType IntRegsVTs[] = {
     MVT::i32, MVT::Other
   };
+
 namespace SP {   // Register class instances
   DFPRegsClass    DFPRegsRegClass;
   FPRegsClass     FPRegsRegClass;
   IntRegsClass    IntRegsRegClass;
 ...
- 
-// IntRegs Sub-register Classess...
+  // IntRegs Sub-register Classess...
   static const TargetRegisterClass* const IntRegsSubRegClasses [] = {
     NULL
   };
 ...
-// IntRegs Super-register Classess...
+  // IntRegs Super-register Classess...
   static const TargetRegisterClass* const IntRegsSuperRegClasses [] = {
     NULL
   };
- 
-// IntRegs Register Class sub-classes...
+...
+  // IntRegs Register Class sub-classes...
   static const TargetRegisterClass* const IntRegsSubclasses [] = {
     NULL
   };
 ...
- 
-// IntRegs Register Class super-classes...
+  // IntRegs Register Class super-classes...
   static const TargetRegisterClass* const IntRegsSuperclasses [] = {
     NULL
   };
-...
- 
-  IntRegsClass::iterator
-  IntRegsClass::allocation_order_end(const MachineFunction &MF) const {
 
-     return end()-10  // Don't allocate special registers
-         -1; 
-  }
-  
-IntRegsClass::IntRegsClass() : TargetRegisterClass(IntRegsRegClassID, 
-   IntRegsVTs, IntRegsSubclasses, IntRegsSuperclasses, IntRegsSubRegClasses, 
-   IntRegsSuperRegClasses, 4, 4, 1, IntRegs, IntRegs + 32) {}
+  IntRegsClass::IntRegsClass() : TargetRegisterClass(IntRegsRegClassID, 
+    IntRegsVTs, IntRegsSubclasses, IntRegsSuperclasses, IntRegsSubRegClasses, 
+    IntRegsSuperRegClasses, 4, 4, 1, IntRegs, IntRegs + 32) {}
 }

+ +

+The register allocators will avoid using reserved registers, and callee saved +registers are not used until all the volatile registers have been used. That +is usually good enough, but in some cases it may be necessary to provide custom +allocation orders. +

+ +

+ -

- Implement a subclass of - TargetRegisterInfo -

The final step is to hand code portions of XXXRegisterInfo, which -implements the interface described in TargetRegisterInfo.h. These functions -return 0, NULL, or false, unless overridden. Here’s a list of functions that -are overridden for the SPARC implementation in SparcRegisterInfo.cpp:

getCalleeSavedRegs (returns a list of callee-saved registers in -the order of the desired callee-save stack frame offset)

+ Implement a subclass of + TargetRegisterInfo +

+ +

+The final step is to hand code portions of XXXRegisterInfo, which +implements the interface described in TargetRegisterInfo.h. These +functions return 0, NULL, or false, unless +overridden. Here is a list of functions that are overridden for the SPARC +implementation in SparcRegisterInfo.cpp: +

getCalleeSavedRegClasses (returns a list of preferred register -classes with which to spill each callee saved register)

getCalleeSavedRegs — Returns a list of callee-saved registers + in the order of the desired callee-save stack frame offset.
getReservedRegs (returns a bitset indexed by physical register -numbers, indicating if a particular register is unavailable)
getReservedRegs — Returns a bitset indexed by physical + register numbers, indicating if a particular register is unavailable.
hasFP (return a Boolean indicating if a function should have a -dedicated frame pointer register)
hasFP — Return a Boolean indicating if a function should have + a dedicated frame pointer register.
eliminateCallFramePseudoInstr (if call frame setup or destroy -pseudo instructions are used, this can be called to eliminate them)
eliminateCallFramePseudoInstr — If call frame setup or + destroy pseudo instructions are used, this can be called to eliminate + them.
eliminateFrameIndex (eliminate abstract frame indices from -instructions that may use them)
eliminateFrameIndex — Eliminate abstract frame indices from + instructions that may use them.
emitPrologue (insert prologue code into the function)
emitPrologue — Insert prologue code into the function.
emitEpilogue (insert epilogue code into the function)
emitEpilogue — Insert epilogue code into the function.

+ +

Instruction Set -

+ + -

During the early stages of code generation, the LLVM IR code is -converted to a SelectionDAG with nodes that are instances of the SDNode class -containing target instructions. An SDNode has an opcode, operands, type -requirements, and operation properties (for example, is an operation -commutative, does an operation load from memory). The various operation node -types are described in the include/llvm/CodeGen/SelectionDAGNodes.h file (values -of the NodeType enum in the ISD namespace).

- -

TableGen uses the following target description (.td) input files -to generate much of the code for instruction definition:

+ +

+During the early stages of code generation, the LLVM IR code is converted to a +SelectionDAG with nodes that are instances of the SDNode class +containing target instructions. An SDNode has an opcode, operands, type +requirements, and operation properties. For example, is an operation +commutative, does an operation load from memory. The various operation node +types are described in the include/llvm/CodeGen/SelectionDAGNodes.h +file (values of the NodeType enum in the ISD namespace). +

+ +

+TableGen uses the following target description (.td) input files to +generate much of the code for instruction definition: +

Target.td, where the Instruction, Operand, InstrInfo, and other -fundamental classes are defined
TargetSelectionDAG.td, used by SelectionDAG instruction selection -generators, contains SDTC* classes (selection DAG type constraint), definitions -of SelectionDAG nodes (such as imm, cond, bb, add, fadd, sub), and pattern -support (Pattern, Pat, PatFrag, PatLeaf, ComplexPattern)
XXXInstrFormats.td, patterns for definitions of target-specific -instructions
XXXInstrInfo.td, target-specific definitions of instruction -templates, condition codes, and instructions of an instruction set. (For architecture -modifications, a different file name may be used. For example, for Pentium with -SSE instruction, this file is X86InstrSSE.td, and for Pentium with MMX, this -file is X86InstrMMX.td.)
Target.td — Where the Instruction, Operand, + InstrInfo, and other fundamental classes are defined.
TargetSelectionDAG.td— Used by SelectionDAG + instruction selection generators, contains SDTC* classes (selection + DAG type constraint), definitions of SelectionDAG nodes (such as + imm, cond, bb, add, fadd, + sub), and pattern support (Pattern, Pat, + PatFrag, PatLeaf, ComplexPattern.
XXXInstrFormats.td — Patterns for definitions of + target-specific instructions.
XXXInstrInfo.td — Target-specific definitions of instruction + templates, condition codes, and instructions of an instruction set. For + architecture modifications, a different file name may be used. For example, + for Pentium with SSE instruction, this file is X86InstrSSE.td, and + for Pentium with MMX, this file is X86InstrMMX.td.

There is also a target-specific XXX.td file, where XXX is the -name of the target. The XXX.td file includes the other .td input files, but its -contents are only directly important for subtargets.

- -

You should describe -a concrete target-specific class -XXXInstrInfo that represents machine -instructions supported by a target machine. XXXInstrInfo contains an array of -XXXInstrDescriptor objects, each of which describes one instruction. An -instruction descriptor defines:

+ +

+There is also a target-specific XXX.td file, where XXX is the +name of the target. The XXX.td file includes the other .td +input files, but its contents are only directly important for subtargets. +

+ +

+You should describe a concrete target-specific class XXXInstrInfo that +represents machine instructions supported by a target machine. +XXXInstrInfo contains an array of XXXInstrDescriptor objects, +each of which describes one instruction. An instruction descriptor defines:

opcode mnemonic
Opcode mnemonic
number of operands
Number of operands
list of implicit register definitions and uses
List of implicit register definitions and uses
target-independent properties (such as memory access, is -commutable)
Target-independent properties (such as memory access, is commutable)
target-specific flags
Target-specific flags

The Instruction class (defined in Target.td) is mostly used as a -base for more complex instruction classes.

+The Instruction class (defined in Target.td) is mostly used as a base +for more complex instruction classes. +

class Instruction {
-  string Namespace = "";
+  string Namespace = "";
   dag OutOperandList;       // An dag containing the MI def operand list.
   dag InOperandList;        // An dag containing the MI use operand list.
-  string AsmString = "";    // The .s format to print the instruction with.
+  string AsmString = "";    // The .s format to print the instruction with.
   list<dag> Pattern;  // Set to the DAG pattern for this instruction
   list<Register> Uses = []; 
   list<Register> Defs = [];
@@ -850,135 +997,157 @@ base for more complex instruction classes.
 }

A SelectionDAG node (SDNode) should contain an object -representing a target-specific instruction that is defined in XXXInstrInfo.td. The -instruction objects should represent instructions from the architecture manual -of the target machine (such as the -SPARC Architecture Manual for the SPARC target).

A single -instruction from the architecture manual is often modeled as multiple target -instructions, depending upon its operands. For example, a manual might +

+A SelectionDAG node (SDNode) should contain an object +representing a target-specific instruction that is defined +in XXXInstrInfo.td. The instruction objects should represent +instructions from the architecture manual of the target machine (such as the +SPARC Architecture Manual for the SPARC target). +

+ +

+A single instruction from the architecture manual is often modeled as multiple +target instructions, depending upon its operands. For example, a manual might describe an add instruction that takes a register or an immediate operand. An -LLVM target could model this with two instructions named ADDri and ADDrr.

- -

You should define a -class for each instruction category and define each opcode as a subclass of the -category with appropriate parameters such as the fixed binary encoding of -opcodes and extended opcodes. You should map the register bits to the bits of -the instruction in which they are encoded (for the JIT). Also you should specify -how the instruction should be printed when the automatic assembly printer is -used.

- -

As is described in -the SPARC Architecture Manual, Version 8, there are three major 32-bit formats -for instructions. Format 1 is only for the CALL instruction. Format 2 is for -branch on condition codes and SETHI (set high bits of a register) instructions. -Format 3 is for other instructions.

- -

Each of these -formats has corresponding classes in SparcInstrFormat.td. InstSP is a base -class for other instruction classes. Additional base classes are specified for -more precise formats: for example in SparcInstrFormat.td, F2_1 is for SETHI, -and F2_2 is for branches. There are three other base classes: F3_1 for -register/register operations, F3_2 for register/immediate operations, and F3_3 for -floating-point operations. SparcInstrInfo.td also adds the base class Pseudo for -synthetic SPARC instructions.

- -

SparcInstrInfo.td -largely consists of operand and instruction definitions for the SPARC target. In -SparcInstrInfo.td, the following target description file entry, LDrr, defines -the Load Integer instruction for a Word (the LD SPARC opcode) from a memory -address to a register. The first parameter, the value 3 (11₂), is -the operation value for this category of operation. The second parameter -(000000₂) is the specific operation value for LD/Load Word. The -third parameter is the output destination, which is a register operand and -defined in the Register target description file (IntRegs).

+LLVM target could model this with two instructions named ADDri and +ADDrr. +

+ +

+You should define a class for each instruction category and define each opcode +as a subclass of the category with appropriate parameters such as the fixed +binary encoding of opcodes and extended opcodes. You should map the register +bits to the bits of the instruction in which they are encoded (for the +JIT). Also you should specify how the instruction should be printed when the +automatic assembly printer is used. +

+ +

+As is described in the SPARC Architecture Manual, Version 8, there are three +major 32-bit formats for instructions. Format 1 is only for the CALL +instruction. Format 2 is for branch on condition codes and SETHI (set +high bits of a register) instructions. Format 3 is for other instructions. +

+ +

+Each of these formats has corresponding classes in SparcInstrFormat.td. +InstSP is a base class for other instruction classes. Additional base +classes are specified for more precise formats: for example +in SparcInstrFormat.td, F2_1 is for SETHI, +and F2_2 is for branches. There are three other base +classes: F3_1 for register/register operations, F3_2 for +register/immediate operations, and F3_3 for floating-point +operations. SparcInstrInfo.td also adds the base class Pseudo for +synthetic SPARC instructions. +

+ +

+SparcInstrInfo.td largely consists of operand and instruction +definitions for the SPARC target. In SparcInstrInfo.td, the following +target description file entry, LDrr, defines the Load Integer +instruction for a Word (the LD SPARC opcode) from a memory address to a +register. The first parameter, the value 3 (11₂), is the +operation value for this category of operation. The second parameter +(000000₂) is the specific operation value +for LD/Load Word. The third parameter is the output destination, which +is a register operand and defined in the Register target description +file (IntRegs). +

def LDrr : F3_1 <3, 0b000000, (outs IntRegs:$dst), (ins MEMrr:$addr),
-                 "ld [$addr], $dst",
+                 "ld [$addr], $dst",
                  [(set IntRegs:$dst, (load ADDRrr:$addr))]>;

The fourth -parameter is the input source, which uses the address operand MEMrr that is -defined earlier in SparcInstrInfo.td:

+The fourth parameter is the input source, which uses the address +operand MEMrr that is defined earlier in SparcInstrInfo.td: +

def MEMrr : Operand<i32> {
-  let PrintMethod = "printMemOperand";
+  let PrintMethod = "printMemOperand";
   let MIOperandInfo = (ops IntRegs, IntRegs);
 }

The fifth parameter is a string that is used by the assembly -printer and can be left as an empty string until the assembly printer interface -is implemented. The sixth and final parameter is the pattern used to match the -instruction during the SelectionDAG Select Phase described in -(The LLVM Target-Independent Code Generator). -This parameter is detailed in the next section, Instruction Selector.

- -

Instruction class definitions are not overloaded for different -operand types, so separate versions of instructions are needed for register, -memory, or immediate value operands. For example, to perform a -Load Integer instruction for a Word + +

+The fifth parameter is a string that is used by the assembly printer and can be +left as an empty string until the assembly printer interface is implemented. The +sixth and final parameter is the pattern used to match the instruction during +the SelectionDAG Select Phase described in +(The LLVM +Target-Independent Code Generator). This parameter is detailed in the next +section, Instruction Selector. +

+ +

+Instruction class definitions are not overloaded for different operand types, so +separate versions of instructions are needed for register, memory, or immediate +value operands. For example, to perform a Load Integer instruction for a Word from an immediate operand to a register, the following instruction class is -defined:

+defined: +

def LDri : F3_2 <3, 0b000000, (outs IntRegs:$dst), (ins MEMri:$addr),
-                 "ld [$addr], $dst",
+                 "ld [$addr], $dst",
                  [(set IntRegs:$dst, (load ADDRri:$addr))]>;

Writing these definitions for so many similar instructions can -involve a lot of cut and paste. In td files, the multiclass directive enables -the creation of templates to define several instruction classes at once (using -the defm directive). For example in -SparcInstrInfo.td, the multiclass pattern F3_12 is defined to create 2 -instruction classes each time F3_12 is invoked:

+ +

+Writing these definitions for so many similar instructions can involve a lot of +cut and paste. In td files, the multiclass directive enables the +creation of templates to define several instruction classes at once (using +the defm directive). For example in SparcInstrInfo.td, the +multiclass pattern F3_12 is defined to create 2 instruction +classes each time F3_12 is invoked: +

multiclass F3_12 <string OpcStr, bits<6> Op3Val, SDNode OpNode> {
   def rr  : F3_1 <2, Op3Val, 
                  (outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
-                 !strconcat(OpcStr, " $b, $c, $dst"),
+                 !strconcat(OpcStr, " $b, $c, $dst"),
                  [(set IntRegs:$dst, (OpNode IntRegs:$b, IntRegs:$c))]>;
   def ri  : F3_2 <2, Op3Val,
                  (outs IntRegs:$dst), (ins IntRegs:$b, i32imm:$c),
-                 !strconcat(OpcStr, " $b, $c, $dst"),
+                 !strconcat(OpcStr, " $b, $c, $dst"),
                  [(set IntRegs:$dst, (OpNode IntRegs:$b, simm13:$c))]>;
 }

So when the defm directive is used for the XOR and ADD -instructions, as seen below, it creates four instruction objects: XORrr, XORri, -ADDrr, and ADDri.

+ +

+So when the defm directive is used for the XOR +and ADD instructions, as seen below, it creates four instruction +objects: XORrr, XORri, ADDrr, and ADDri. +

defm XOR   : F3_12<"xor", 0b000011, xor>;
-defm ADD   : F3_12<"add", 0b000000, add>;
++defm XOR   : F3_12<"xor", 0b000011, xor>;
+defm ADD   : F3_12<"add", 0b000000, add>;

SparcInstrInfo.td -also includes definitions for condition codes that are referenced by branch -instructions. The following definitions in SparcInstrInfo.td indicate the bit location -of the SPARC condition code; for example, the 10^th bit represents -the ‘greater than’ condition for integers, and the 22^nd bit -represents the ‘greater than’ condition for floats.

+SparcInstrInfo.td also includes definitions for condition codes that +are referenced by branch instructions. The following definitions +in SparcInstrInfo.td indicate the bit location of the SPARC condition +code. For example, the 10^th bit represents the 'greater than' +condition for integers, and the 22^nd bit represents the 'greater +than' condition for floats. +

def ICC_NE  : ICC_VAL< 9>;  // Not Equal
++def ICC_NE  : ICC_VAL< 9>;  // Not Equal
 def ICC_E   : ICC_VAL< 1>;  // Equal
 def ICC_G   : ICC_VAL<10>;  // Greater
 ...
@@ -989,93 +1158,199 @@ def FCC_UG  : FCC_VAL<21>;  // Unordered or Greater

(Note that Sparc.h -also defines enums that correspond to the same SPARC condition codes. Care must -be taken to ensure the values in Sparc.h correspond to the values in -SparcInstrInfo.td; that is, SPCC::ICC_NE = 9, SPCC::FCC_U = 23 and so on.)

+(Note that Sparc.h also defines enums that correspond to the same SPARC +condition codes. Care must be taken to ensure the values in Sparc.h +correspond to the values in SparcInstrInfo.td. I.e., +SPCC::ICC_NE = 9, SPCC::FCC_U = 23 and so on.) +

- Implement a subclass of - TargetInstrInfo +

+ Instruction Operand Mapping +

+ +

+The code generator backend maps instruction operands to fields in the +instruction. Operands are assigned to unbound fields in the instruction in the +order they are defined. Fields are bound when they are assigned a value. For +example, the Sparc target defines the XNORrr instruction as +a F3_1 format instruction having three operands. +

+ +

+def XNORrr  : F3_1<2, 0b000111,
+                   (outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
+                   "xnor $b, $c, $dst",
+                   [(set IntRegs:$dst, (not (xor IntRegs:$b, IntRegs:$c)))]>;
+

The final step is to hand code portions of XXXInstrInfo, which -implements the interface described in TargetInstrInfo.h. These functions return -0 or a Boolean or they assert, unless overridden. Here's a list of functions -that are overridden for the SPARC implementation in SparcInstrInfo.cpp:

isMoveInstr (return true if the instruction is a register to -register move; false, otherwise)

+The instruction templates in SparcInstrFormats.td show the base class +for F3_1 is InstSP. +

+class InstSP<dag outs, dag ins, string asmstr, list<dag> pattern> : Instruction {
+  field bits<32> Inst;
+  let Namespace = "SP";
+  bits<2> op;
+  let Inst{31-30} = op;       
+  dag OutOperandList = outs;
+  dag InOperandList = ins;
+  let AsmString   = asmstr;
+  let Pattern = pattern;
+}
+

InstSP leaves the op field unbound.

+class F3<dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstSP<outs, ins, asmstr, pattern> {
+  bits<5> rd;
+  bits<6> op3;
+  bits<5> rs1;
+  let op{1} = 1;   // Op = 2 or 3
+  let Inst{29-25} = rd;
+  let Inst{24-19} = op3;
+  let Inst{18-14} = rs1;
+}
+

+F3 binds the op field and defines the rd, +op3, and rs1 fields. F3 format instructions will +bind the operands rd, op3, and rs1 fields. +

+class F3_1<bits<2> opVal, bits<6> op3val, dag outs, dag ins,
+           string asmstr, list<dag> pattern> : F3<outs, ins, asmstr, pattern> {
+  bits<8> asi = 0; // asi not currently used
+  bits<5> rs2;
+  let op         = opVal;
+  let op3        = op3val;
+  let Inst{13}   = 0;     // i field = 0
+  let Inst{12-5} = asi;   // address space identifier
+  let Inst{4-0}  = rs2;
+}
+

isLoadFromStackSlot (if the specified machine instruction is a -direct load from a stack slot, return the register number of the destination -and the FrameIndex of the stack slot)

+F3_1 binds the op3 field and defines the rs2 +fields. F3_1 format instructions will bind the operands to the rd, +rs1, and rs2 fields. This results in the XNORrr +instruction binding $dst, $b, and $c operands to +the rd, rs1, and rs2 fields respectively. +

isStoreToStackSlot (if the specified machine instruction is a -direct store to a stack slot, return the register number of the destination and -the FrameIndex of the stack slot)

copyRegToReg (copy values between a pair of registers)

+ +

+ Implement a subclass of + TargetInstrInfo +

+ +

+The final step is to hand code portions of XXXInstrInfo, which +implements the interface described in TargetInstrInfo.h. These +functions return 0 or a Boolean or they assert, unless +overridden. Here's a list of functions that are overridden for the SPARC +implementation in SparcInstrInfo.cpp: +

storeRegToStackSlot (store a register value to a stack slot)

isLoadFromStackSlot — If the specified machine instruction is + a direct load from a stack slot, return the register number of the + destination and the FrameIndex of the stack slot.
loadRegFromStackSlot (load a register value from a stack slot)
isStoreToStackSlot — If the specified machine instruction is + a direct store to a stack slot, return the register number of the + destination and the FrameIndex of the stack slot.
storeRegToAddr (store a register value to memory)
copyPhysReg — Copy values between a pair of physical + registers.
loadRegFromAddr (load a register value from memory)
storeRegToStackSlot — Store a register value to a stack + slot.
foldMemoryOperand (attempt to combine instructions of any load or -store instruction for the specified operand(s))
loadRegFromStackSlot — Load a register value from a stack + slot.
storeRegToAddr — Store a register value to memory.
loadRegFromAddr — Load a register value from memory.
foldMemoryOperand — Attempt to combine instructions of any + load or store instruction for the specified operand(s).

Branch Folding and If Conversion -

Performance can be improved by combining instructions or by eliminating -instructions that are never reached. The AnalyzeBranch method in XXXInstrInfo may -be implemented to examine conditional instructions and remove unnecessary -instructions. AnalyzeBranch looks at the end of a machine basic block (MBB) for -opportunities for improvement, such as branch folding and if conversion. The -BranchFolder and IfConverter machine function passes (see the source files -BranchFolding.cpp and IfConversion.cpp in the lib/CodeGen directory) call + +

+ +

+Performance can be improved by combining instructions or by eliminating +instructions that are never reached. The AnalyzeBranch method +in XXXInstrInfo may be implemented to examine conditional instructions +and remove unnecessary instructions. AnalyzeBranch looks at the end of +a machine basic block (MBB) for opportunities for improvement, such as branch +folding and if conversion. The BranchFolder and IfConverter +machine function passes (see the source files BranchFolding.cpp and +IfConversion.cpp in the lib/CodeGen directory) call AnalyzeBranch to improve the control flow graph that represents the -instructions.

+instructions. +

Several implementations of AnalyzeBranch (for ARM, Alpha, and -X86) can be examined as models for your own AnalyzeBranch implementation. Since -SPARC does not implement a useful AnalyzeBranch, the ARM target implementation -is shown below.

+Several implementations of AnalyzeBranch (for ARM, Alpha, and X86) can +be examined as models for your own AnalyzeBranch implementation. Since +SPARC does not implement a useful AnalyzeBranch, the ARM target +implementation is shown below. +

AnalyzeBranch returns a Boolean value and takes four parameters:

MachineBasicBlock &MBB – the incoming block to be -examined
MachineBasicBlock &MBB — The incoming block to be + examined.
MachineBasicBlock *&TBB – a destination block that is -returned; for a conditional branch that evaluates to true, TBB is the -destination
MachineBasicBlock *&TBB — A destination block that is + returned. For a conditional branch that evaluates to true, TBB is + the destination.
MachineBasicBlock *&FBB – for a conditional branch that -evaluates to false, FBB is returned as the destination
MachineBasicBlock *&FBB — For a conditional branch that + evaluates to false, FBB is returned as the destination.
std::vector<MachineOperand> &Cond – list of -operands to evaluate a condition for a conditional branch
std::vector<MachineOperand> &Cond — List of + operands to evaluate a condition for a conditional branch.

In the simplest case, if a block ends without a branch, then it -falls through to the successor block. No destination blocks are specified for -either TBB or FBB, so both parameters return NULL. The start of the AnalyzeBranch -(see code below for the ARM target) shows the function parameters and the code -for the simplest case.

+In the simplest case, if a block ends without a branch, then it falls through to +the successor block. No destination blocks are specified for either TBB +or FBB, so both parameters return NULL. The start of +the AnalyzeBranch (see code below for the ARM target) shows the +function parameters and the code for the simplest case. +

bool ARMInstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,
@@ -1088,28 +1363,30 @@ for the simplest case.

If a block ends with a single unconditional branch instruction, -then AnalyzeBranch (shown below) should return the destination of that branch -in the TBB parameter.

+If a block ends with a single unconditional branch instruction, then +AnalyzeBranch (shown below) should return the destination of that +branch in the TBB parameter. +

if (LastOpc == ARM::B || LastOpc == ARM::tB) {
-      TBB = LastInst->getOperand(0).getMBB();
-      return false;
-    }
++  if (LastOpc == ARM::B || LastOpc == ARM::tB) {
+    TBB = LastInst->getOperand(0).getMBB();
+    return false;
+  }

If a block ends with two unconditional branches, then the second -branch is never reached. In that situation, as shown below, remove the last -branch instruction and return the penultimate branch in the TBB parameter.

+If a block ends with two unconditional branches, then the second branch is never +reached. In that situation, as shown below, remove the last branch instruction +and return the penultimate branch in the TBB parameter. +

if ((SecondLastOpc == ARM::B || SecondLastOpc==ARM::tB) &&
++  if ((SecondLastOpc == ARM::B || SecondLastOpc==ARM::tB) &&
       (LastOpc == ARM::B || LastOpc == ARM::tB)) {
     TBB = SecondLastInst->getOperand(0).getMBB();
     I = LastInst;
@@ -1118,37 +1395,41 @@ branch instruction and return the penultimate branch in the TBB parameter. 
   }

A block may end with a single conditional branch instruction that -falls through to successor block if the condition evaluates to false. In that -case, AnalyzeBranch (shown below) should return the destination of that -conditional branch in the TBB parameter and a list of operands in the Cond -parameter to evaluate the condition.

+ +

+A block may end with a single conditional branch instruction that falls through +to successor block if the condition evaluates to false. In that case, +AnalyzeBranch (shown below) should return the destination of that +conditional branch in the TBB parameter and a list of operands in +the Cond parameter to evaluate the condition. +

if (LastOpc == ARM::Bcc || LastOpc == ARM::tBcc) {
-      // Block ends with fall-through condbranch.
-      TBB = LastInst->getOperand(0).getMBB();
-      Cond.push_back(LastInst->getOperand(1));
-      Cond.push_back(LastInst->getOperand(2));
-      return false;
-    }
++  if (LastOpc == ARM::Bcc || LastOpc == ARM::tBcc) {
+    // Block ends with fall-through condbranch.
+    TBB = LastInst->getOperand(0).getMBB();
+    Cond.push_back(LastInst->getOperand(1));
+    Cond.push_back(LastInst->getOperand(2));
+    return false;
+  }

If a block ends with both a conditional branch and an ensuing -unconditional branch, then AnalyzeBranch (shown below) should return the -conditional branch destination (assuming it corresponds to a conditional -evaluation of ‘true’) in the TBB parameter and the unconditional branch -destination in the FBB (corresponding to a conditional evaluation of ‘false’). -A list of operands to evaluate the condition should be returned in the Cond -parameter.

+If a block ends with both a conditional branch and an ensuing unconditional +branch, then AnalyzeBranch (shown below) should return the conditional +branch destination (assuming it corresponds to a conditional evaluation of +'true') in the TBB parameter and the unconditional branch +destination in the FBB (corresponding to a conditional evaluation of +'false'). A list of operands to evaluate the condition should be +returned in the Cond parameter. +

unsigned SecondLastOpc = SecondLastInst->getOpcode();
++  unsigned SecondLastOpc = SecondLastInst->getOpcode();
+
   if ((SecondLastOpc == ARM::Bcc && LastOpc == ARM::B) ||
       (SecondLastOpc == ARM::tBcc && LastOpc == ARM::tB)) {
     TBB =  SecondLastInst->getOperand(0).getMBB();
@@ -1160,96 +1441,119 @@ parameter.

For the last two cases (ending with a single conditional branch or -ending with one conditional and one unconditional branch), the operands returned -in the Cond parameter can be passed to methods of other instructions to create -new branches or perform other operations. An implementation of AnalyzeBranch -requires the helper methods RemoveBranch and InsertBranch to manage subsequent -operations.

+For the last two cases (ending with a single conditional branch or ending with +one conditional and one unconditional branch), the operands returned in +the Cond parameter can be passed to methods of other instructions to +create new branches or perform other operations. An implementation +of AnalyzeBranch requires the helper methods RemoveBranch +and InsertBranch to manage subsequent operations. +

AnalyzeBranch should return false indicating success in most circumstances. +

+AnalyzeBranch should return false indicating success in most circumstances. AnalyzeBranch should only return true when the method is stumped about what to do, for example, if a block has three terminating branches. AnalyzeBranch may return true if it encounters a terminator it cannot handle, such as an indirect -branch.

+branch. +

+ +

Instruction Selector -

+ -

LLVM uses a SelectionDAG to represent LLVM IR instructions, and nodes -of the SelectionDAG ideally represent native target instructions. During code -generation, instruction selection passes are performed to convert non-native -DAG instructions into native target-specific instructions. The pass described -in XXXISelDAGToDAG.cpp is used to match patterns and perform DAG-to-DAG -instruction selection. Optionally, a pass may be defined (in -XXXBranchSelector.cpp) to perform similar DAG-to-DAG operations for branch -instructions. Later, -the code in XXXISelLowering.cpp replaces or removes operations and data types -not supported natively (legalizes) in a Selection DAG.

- -

TableGen generates code for instruction selection using the -following target description input files:

XXXInstrInfo.td contains definitions of instructions in a -target-specific instruction set, generates XXXGenDAGISel.inc, which is included -in XXXISelDAGToDAG.cpp.

+ +

+LLVM uses a SelectionDAG to represent LLVM IR instructions, and nodes +of the SelectionDAG ideally represent native target +instructions. During code generation, instruction selection passes are performed +to convert non-native DAG instructions into native target-specific +instructions. The pass described in XXXISelDAGToDAG.cpp is used to +match patterns and perform DAG-to-DAG instruction selection. Optionally, a pass +may be defined (in XXXBranchSelector.cpp) to perform similar DAG-to-DAG +operations for branch instructions. Later, the code in +XXXISelLowering.cpp replaces or removes operations and data types not +supported natively (legalizes) in a SelectionDAG. +

+ +

+TableGen generates code for instruction selection using the following target +description input files: +

XXXCallingConv.td contains the calling and return value conventions -for the target architecture, and it generates XXXGenCallingConv.inc, which is -included in XXXISelLowering.cpp.

XXXInstrInfo.td — Contains definitions of instructions in a + target-specific instruction set, generates XXXGenDAGISel.inc, which + is included in XXXISelDAGToDAG.cpp.
XXXCallingConv.td — Contains the calling and return value + conventions for the target architecture, and it generates + XXXGenCallingConv.inc, which is included in + XXXISelLowering.cpp.

The implementation of an instruction selection pass must include -a header that declares the FunctionPass class or a subclass of FunctionPass. In -XXXTargetMachine.cpp, a Pass Manager (PM) should add each instruction selection -pass into the queue of passes to run.

+The implementation of an instruction selection pass must include a header that +declares the FunctionPass class or a subclass of FunctionPass. In +XXXTargetMachine.cpp, a Pass Manager (PM) should add each instruction +selection pass into the queue of passes to run. +

The LLVM static -compiler (llc) is an excellent tool for visualizing the contents of DAGs. To display -the SelectionDAG before or after specific processing phases, use the command -line options for llc, described at +

+The LLVM static compiler (llc) is an excellent tool for visualizing the +contents of DAGs. To display the SelectionDAG before or after specific +processing phases, use the command line options for llc, described +at SelectionDAG Instruction Selection Process.

To describe instruction selector behavior, you should add -patterns for lowering LLVM code into a SelectionDAG as the last parameter of -the instruction definitions in XXXInstrInfo.td. For example, in -SparcInstrInfo.td, this entry defines a register store operation, and the last -parameter describes a pattern with the store DAG operator.

+To describe instruction selector behavior, you should add patterns for lowering +LLVM code into a SelectionDAG as the last parameter of the instruction +definitions in XXXInstrInfo.td. For example, in +SparcInstrInfo.td, this entry defines a register store operation, and +the last parameter describes a pattern with the store DAG operator. +

def STrr  : F3_1< 3, 0b000100, (outs), (ins MEMrr:$addr, IntRegs:$src),
-                 "st $src, [$addr]", [(store IntRegs:$src, ADDRrr:$addr)]>;
++def STrr  : F3_1< 3, 0b000100, (outs), (ins MEMrr:$addr, IntRegs:$src),
+                 "st $src, [$addr]", [(store IntRegs:$src, ADDRrr:$addr)]>;

ADDRrr is a memory mode that is also defined in SparcInstrInfo.td:

+ADDRrr is a memory mode that is also defined in +SparcInstrInfo.td: +

def ADDRrr : ComplexPattern<i32, 2, "SelectADDRrr", [], []>;
++def ADDRrr : ComplexPattern<i32, 2, "SelectADDRrr", [], []>;

The definition of ADDRrr refers to SelectADDRrr, which is a function defined in an -implementation of the Instructor Selector (such as SparcISelDAGToDAG.cpp).

+The definition of ADDRrr refers to SelectADDRrr, which is a +function defined in an implementation of the Instructor Selector (such +as SparcISelDAGToDAG.cpp). +

In lib/Target/TargetSelectionDAG.td, the DAG operator for store -is defined below:

+In lib/Target/TargetSelectionDAG.td, the DAG operator for store is +defined below: +

def store : PatFrag<(ops node:$val, node:$ptr),
++def store : PatFrag<(ops node:$val, node:$ptr),
                     (st node:$val, node:$ptr), [{
   if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N))
     return !ST->isTruncatingStore() && 
@@ -1258,17 +1562,19 @@ is defined below:
 }]>;

XXXInstrInfo.td also generates (in XXXGenDAGISel.inc) the -SelectCode method that is used to call the appropriate processing method for an -instruction. In this example, SelectCode calls Select_ISD_STORE for the -ISD::STORE opcode.

+XXXInstrInfo.td also generates (in XXXGenDAGISel.inc) the +SelectCode method that is used to call the appropriate processing +method for an instruction. In this example, SelectCode +calls Select_ISD_STORE for the ISD::STORE opcode. +

SDNode *SelectCode(SDOperand N) {
++SDNode *SelectCode(SDValue N) {
   ... 
-  MVT::ValueType NVT = N.Val->getValueType(0);
+  MVT::ValueType NVT = N.getNode()->getValueType(0);
   switch (N.getOpcode()) {
   case ISD::STORE: {
     switch (NVT) {
@@ -1281,29 +1587,32 @@ ISD::STORE opcode.
   ...

The pattern for STrr is matched, so elsewhere in -XXXGenDAGISel.inc, code for STrr is created for Select_ISD_STORE. The Emit_22 method -is also generated in XXXGenDAGISel.inc to complete the processing of this -instruction.

+The pattern for STrr is matched, so elsewhere in +XXXGenDAGISel.inc, code for STrr is created for +Select_ISD_STORE. The Emit_22 method is also generated +in XXXGenDAGISel.inc to complete the processing of this +instruction. +

SDNode *Select_ISD_STORE(const SDOperand &N) {
-  SDOperand Chain = N.getOperand(0);
-  if (Predicate_store(N.Val)) {
-    SDOperand N1 = N.getOperand(1);
-    SDOperand N2 = N.getOperand(2);
-    SDOperand CPTmp0;
-    SDOperand CPTmp1;
- 
++SDNode *Select_ISD_STORE(const SDValue &N) {
+  SDValue Chain = N.getOperand(0);
+  if (Predicate_store(N.getNode())) {
+    SDValue N1 = N.getOperand(1);
+    SDValue N2 = N.getOperand(2);
+    SDValue CPTmp0;
+    SDValue CPTmp1;
+
     // Pattern: (st:void IntRegs:i32:$src, 
     //           ADDRrr:i32:$addr)<<P:Predicate_store>>
     // Emits: (STrr:void ADDRrr:i32:$addr, IntRegs:i32:$src)
     // Pattern complexity = 13  cost = 1  size = 0
     if (SelectADDRrr(N, N2, CPTmp0, CPTmp1) &&
-        N1.Val->getValueType(0) == MVT::i32 &&
-        N2.Val->getValueType(0) == MVT::i32) {
+        N1.getNode()->getValueType(0) == MVT::i32 &&
+        N2.getNode()->getValueType(0) == MVT::i32) {
       return Emit_22(N, SP::STrr, CPTmp0, CPTmp1);
     }
 ...
@@ -1311,169 +1620,210 @@ instruction.

The SelectionDAG Legalize Phase -

The Legalize phase converts a DAG to use types and operations -that are natively supported by the target. For natively unsupported types and -operations, you need to add code to the target-specific XXXTargetLowering implementation -to convert unsupported types and operations to supported ones.

+ -

In the constructor for the XXXTargetLowering class, first use the -addRegisterClass method to specify which types are supports and which register -classes are associated with them. The code for the register classes are generated -by TableGen from XXXRegisterInfo.td and placed in XXXGenRegisterInfo.h.inc. For -example, the implementation of the constructor for the SparcTargetLowering -class (in SparcISelLowering.cpp) starts with the following code:

+ +

+The Legalize phase converts a DAG to use types and operations that are natively +supported by the target. For natively unsupported types and operations, you need +to add code to the target-specific XXXTargetLowering implementation to convert +unsupported types and operations to supported ones. +

+ +

+In the constructor for the XXXTargetLowering class, first use the +addRegisterClass method to specify which types are supports and which +register classes are associated with them. The code for the register classes are +generated by TableGen from XXXRegisterInfo.td and placed +in XXXGenRegisterInfo.h.inc. For example, the implementation of the +constructor for the SparcTargetLowering class (in +SparcISelLowering.cpp) starts with the following code: +

addRegisterClass(MVT::i32, SP::IntRegsRegisterClass);
++addRegisterClass(MVT::i32, SP::IntRegsRegisterClass);
 addRegisterClass(MVT::f32, SP::FPRegsRegisterClass);
 addRegisterClass(MVT::f64, SP::DFPRegsRegisterClass);

You should examine the node types in the ISD namespace -(include/llvm/CodeGen/SelectionDAGNodes.h) -and determine which operations the target natively supports. For operations -that do not have native support, add a callback to the constructor for -the XXXTargetLowering class, so the instruction selection process knows what to -do. The TargetLowering class callback methods (declared in -llvm/Target/TargetLowering.h) are:

+You should examine the node types in the ISD namespace +(include/llvm/CodeGen/SelectionDAGNodes.h) and determine which +operations the target natively supports. For operations that do not have +native support, add a callback to the constructor for the XXXTargetLowering +class, so the instruction selection process knows what to do. The TargetLowering +class callback methods (declared in llvm/Target/TargetLowering.h) are: +

setOperationAction (general operation)
setOperationAction — General operation.
setLoadExtAction (load with extension)
setLoadExtAction — Load with extension.
setTruncStoreAction (truncating store)
setTruncStoreAction — Truncating store.
setIndexedLoadAction (indexed load)
setIndexedLoadAction — Indexed load.
setIndexedStoreAction (indexed store)
setIndexedStoreAction — Indexed store.
setConvertAction (type conversion)
setConvertAction — Type conversion.
setCondCodeAction (support for a given condition code)
setCondCodeAction — Support for a given condition code.

Note: on older releases, setLoadXAction is used instead of setLoadExtAction. -Also, on older releases, setCondCodeAction may not be supported. Examine your -release to see what methods are specifically supported.

+Note: on older releases, setLoadXAction is used instead +of setLoadExtAction. Also, on older releases, +setCondCodeAction may not be supported. Examine your release +to see what methods are specifically supported. +

These callbacks are used to determine that an operation does or -does not work with a specified type (or types). And in all cases, the third -parameter is a LegalAction type enum value: Promote, Expand, +

+These callbacks are used to determine that an operation does or does not work +with a specified type (or types). And in all cases, the third parameter is +a LegalAction type enum value: Promote, Expand, Custom, or Legal. SparcISelLowering.cpp -contains examples of all four LegalAction values.

+contains examples of all four LegalAction values. +

Promote -

+ -

For an operation without native support for a given type, the -specified type may be promoted to a larger type that is supported. For example, -SPARC does not support a sign-extending load for Boolean values (i1 type), so -in SparcISelLowering.cpp the third -parameter below, Promote, changes i1 type -values to a large type before loading.

+ +

+For an operation without native support for a given type, the specified type may +be promoted to a larger type that is supported. For example, SPARC does not +support a sign-extending load for Boolean values (i1 type), so +in SparcISelLowering.cpp the third parameter below, Promote, +changes i1 type values to a large type before loading. +

setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
++setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);

+ -

Expand -

For a type without native support, a value may need to be broken -down further, rather than promoted. For an operation without native support, a -combination of other operations may be used to similar effect. In SPARC, the -floating-point sine and cosine trig operations are supported by expansion to -other operations, as indicated by the third parameter, Expand, to -setOperationAction:

+ + +

+ +

+For a type without native support, a value may need to be broken down further, +rather than promoted. For an operation without native support, a combination of +other operations may be used to similar effect. In SPARC, the floating-point +sine and cosine trig operations are supported by expansion to other operations, +as indicated by the third parameter, Expand, to +setOperationAction: +

setOperationAction(ISD::FSIN, MVT::f32, Expand);
++setOperationAction(ISD::FSIN, MVT::f32, Expand);
 setOperationAction(ISD::FCOS, MVT::f32, Expand);

+ -

Custom -

For some operations, simple type promotion or operation expansion -may be insufficient. In some cases, a special intrinsic function must be -implemented.

+ -

For example, a constant value may require special treatment, or -an operation may require spilling and restoring registers in the stack and -working with register allocators.

As seen in SparcISelLowering.cpp code below, to perform a type +

+For some operations, simple type promotion or operation expansion may be +insufficient. In some cases, a special intrinsic function must be implemented. +

+ +

+For example, a constant value may require special treatment, or an operation may +require spilling and restoring registers in the stack and working with register +allocators. +

+ +

+As seen in SparcISelLowering.cpp code below, to perform a type conversion from a floating point value to a signed integer, first the -setOperationAction should be called with Custom as the third parameter:

+setOperationAction should be called with Custom as the third +parameter: +

setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
++setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);

In the LowerOperation method, for each Custom operation, a case -statement should be added to indicate what function to call. In the following -code, an FP_TO_SINT opcode will call the LowerFP_TO_SINT method:

+ +

+In the LowerOperation method, for each Custom operation, a +case statement should be added to indicate what function to call. In the +following code, an FP_TO_SINT opcode will call +the LowerFP_TO_SINT method: +

SDOperand SparcTargetLowering::LowerOperation(
-                               SDOperand Op, SelectionDAG &DAG) {
++SDValue SparcTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) {
   switch (Op.getOpcode()) {
   case ISD::FP_TO_SINT: return LowerFP_TO_SINT(Op, DAG);
   ...
   }
 }
 
-

Finally, the LowerFP_TO_SINT method is implemented, using an FP -register to convert the floating-point value to an integer.

+Finally, the LowerFP_TO_SINT method is implemented, using an FP +register to convert the floating-point value to an integer. +

static SDOperand LowerFP_TO_SINT(SDOperand Op, SelectionDAG &DAG) {
-assert(Op.getValueType() == MVT::i32);
++static SDValue LowerFP_TO_SINT(SDValue Op, SelectionDAG &DAG) {
+  assert(Op.getValueType() == MVT::i32);
   Op = DAG.getNode(SPISD::FTOI, MVT::f32, Op.getOperand(0));
-  return DAG.getNode(ISD::BIT_CONVERT, MVT::i32, Op);
+  return DAG.getNode(ISD::BITCAST, MVT::i32, Op);
 }

+ +

+ -

Legal -

The Legal LegalizeAction enum value simply indicates that an -operation is natively supported. Legal represents the default condition, -so it is rarely used. In SparcISelLowering.cpp, the action for CTPOP (an -operation to count the bits set in an integer) is natively supported only for -SPARC v9. The following code enables the Expand conversion technique for non-v9 -SPARC implementations.

+ + +

+ +

+The Legal LegalizeAction enum value simply indicates that an +operation is natively supported. Legal represents the default +condition, so it is rarely used. In SparcISelLowering.cpp, the action +for CTPOP (an operation to count the bits set in an integer) is +natively supported only for SPARC v9. The following code enables +the Expand conversion technique for non-v9 SPARC implementations. +

setOperationAction(ISD::CTPOP, MVT::i32, Expand);
++setOperationAction(ISD::CTPOP, MVT::i32, Expand);
 ...
 if (TM.getSubtarget<SparcSubtarget>().isV9())
   setOperationAction(ISD::CTPOP, MVT::i32, Legal);
@@ -1485,227 +1835,278 @@ if (TM.getSubtarget<SparcSubtarget>().isV9())
 }

+ +

Calling Conventions -

To support target-specific calling conventions, XXXGenCallingConv.td + + +

+ +

+To support target-specific calling conventions, XXXGenCallingConv.td uses interfaces (such as CCIfType and CCAssignToReg) that are defined in -lib/Target/TargetCallingConv.td. TableGen can take the target descriptor file -XXXGenCallingConv.td and generate the header file XXXGenCallingConv.inc, which -is typically included in XXXISelLowering.cpp. You can use the interfaces in -TargetCallingConv.td to specify:

+lib/Target/TargetCallingConv.td. TableGen can take the target +descriptor file XXXGenCallingConv.td and generate the header +file XXXGenCallingConv.inc, which is typically included +in XXXISelLowering.cpp. You can use the interfaces in +TargetCallingConv.td to specify: +

the order of parameter allocation
The order of parameter allocation.
where parameters and return values are placed (that is, on the -stack or in registers)
Where parameters and return values are placed (that is, on the stack or in + registers).
which registers may be used
Which registers may be used.
whether the caller or callee unwinds the stack
Whether the caller or callee unwinds the stack.

The following example demonstrates the use of the CCIfType and -CCAssignToReg interfaces. If the CCIfType predicate is true (that is, if the -current argument is of type f32 or f64), then the action is performed. In this -case, the CCAssignToReg action assigns the argument value to the first -available register: either R0 or R1.

+The following example demonstrates the use of the CCIfType and +CCAssignToReg interfaces. If the CCIfType predicate is true +(that is, if the current argument is of type f32 or f64), then +the action is performed. In this case, the CCAssignToReg action assigns +the argument value to the first available register: either R0 +or R1. +

CCIfType<[f32,f64], CCAssignToReg<[R0, R1]>>
++CCIfType<[f32,f64], CCAssignToReg<[R0, R1]>>

SparcCallingConv.td contains definitions for a target-specific return-value -calling convention (RetCC_Sparc32) and a basic 32-bit C calling convention -(CC_Sparc32). The definition of RetCC_Sparc32 (shown below) indicates which -registers are used for specified scalar return types. A single-precision float -is returned to register F0, and a double-precision float goes to register D0. A -32-bit integer is returned in register I0 or I1.

+ +

+SparcCallingConv.td contains definitions for a target-specific +return-value calling convention (RetCC_Sparc32) and a basic 32-bit C calling +convention (CC_Sparc32). The definition of RetCC_Sparc32 +(shown below) indicates which registers are used for specified scalar return +types. A single-precision float is returned to register F0, and a +double-precision float goes to register D0. A 32-bit integer is +returned in register I0 or I1. +

def RetCC_Sparc32 : CallingConv<[
++def RetCC_Sparc32 : CallingConv<[
   CCIfType<[i32], CCAssignToReg<[I0, I1]>>,
   CCIfType<[f32], CCAssignToReg<[F0]>>,
   CCIfType<[f64], CCAssignToReg<[D0]>>
 ]>;

The definition of CC_Sparc32 in SparcCallingConv.td introduces -CCAssignToStack, which assigns the value to a stack slot with the specified size -and alignment. In the example below, the first parameter, 4, indicates the size -of the slot, and the second parameter, also 4, indicates the stack alignment -along 4-byte units. (Special cases: if size is zero, then the ABI size is used; -if alignment is zero, then the ABI alignment is used.)

+ +

+The definition of CC_Sparc32 in SparcCallingConv.td introduces +CCAssignToStack, which assigns the value to a stack slot with the +specified size and alignment. In the example below, the first parameter, 4, +indicates the size of the slot, and the second parameter, also 4, indicates the +stack alignment along 4-byte units. (Special cases: if size is zero, then the +ABI size is used; if alignment is zero, then the ABI alignment is used.) +

def CC_Sparc32 : CallingConv<[
++def CC_Sparc32 : CallingConv<[
   // All arguments get passed in integer registers if there is space.
   CCIfType<[i32, f32, f64], CCAssignToReg<[I0, I1, I2, I3, I4, I5]>>,
   CCAssignToStack<4, 4>
 ]>;

CCDelegateTo is another commonly used interface, which tries to find -a specified sub-calling convention and, if a match is found, it is invoked. In -the following example (in X86CallingConv.td), the definition of RetCC_X86_32_C -ends with CCDelegateTo. After the current value is assigned to the register ST0 -or ST1, the RetCC_X86Common is invoked.

+ +

+CCDelegateTo is another commonly used interface, which tries to find a +specified sub-calling convention, and, if a match is found, it is invoked. In +the following example (in X86CallingConv.td), the definition of +RetCC_X86_32_C ends with CCDelegateTo. After the current value +is assigned to the register ST0 or ST1, +the RetCC_X86Common is invoked. +

def RetCC_X86_32_C : CallingConv<[
++def RetCC_X86_32_C : CallingConv<[
   CCIfType<[f32], CCAssignToReg<[ST0, ST1]>>,
   CCIfType<[f64], CCAssignToReg<[ST0, ST1]>>,
   CCDelegateTo<RetCC_X86Common>
 ]>;

CCIfCC is an interface that attempts to match the given name to -the current calling convention. If the name identifies the current calling + +

+CCIfCC is an interface that attempts to match the given name to the +current calling convention. If the name identifies the current calling convention, then a specified action is invoked. In the following example (in -X86CallingConv.td), if the Fast calling convention is in use, then RetCC_X86_32_Fast -is invoked. If the SSECall calling convention is in use, then RetCC_X86_32_SSE -is invoked.

+X86CallingConv.td), if the Fast calling convention is in use, +then RetCC_X86_32_Fast is invoked. If the SSECall calling +convention is in use, then RetCC_X86_32_SSE is invoked. +

def RetCC_X86_32 : CallingConv<[
-  CCIfCC<"CallingConv::Fast", CCDelegateTo<RetCC_X86_32_Fast>>,
-  CCIfCC<"CallingConv::X86_SSECall", CCDelegateTo<RetCC_X86_32_SSE>>,
++def RetCC_X86_32 : CallingConv<[
+  CCIfCC<"CallingConv::Fast", CCDelegateTo<RetCC_X86_32_Fast>>,
+  CCIfCC<"CallingConv::X86_SSECall", CCDelegateTo<RetCC_X86_32_SSE>>,
   CCDelegateTo<RetCC_X86_32_C>
 ]>;

Other calling convention interfaces include:

CCIf <predicate, action> - if the predicate matches, apply -the action
CCIf <predicate, action> — If the predicate matches, + apply the action.
CCIfInReg <action> - if the argument is marked with the -‘inreg’ attribute, then apply the action
CCIfInReg <action> — If the argument is marked with the + 'inreg' attribute, then apply the action.
CCIfNest <action> - if the argument is marked with the -‘nest’ attribute, then apply the action
CCIfNest <action> — Inf the argument is marked with the + 'nest' attribute, then apply the action.
CCIfNotVarArg <action> - if the current function does not -take a variable number of arguments, apply the action
CCIfNotVarArg <action> — If the current function does + not take a variable number of arguments, apply the action.
CCAssignToRegWithShadow <registerList, shadowList> - -similar to CCAssignToReg, but with a shadow list of registers
CCAssignToRegWithShadow <registerList, shadowList> — + similar to CCAssignToReg, but with a shadow list of registers.
CCPassByVal <size, align> - assign value to a stack slot -with the minimum specified size and alignment
CCPassByVal <size, align> — Assign value to a stack + slot with the minimum specified size and alignment.
CCPromoteToType <type> - promote the current value to the specified -type
CCPromoteToType <type> — Promote the current value to + the specified type.
CallingConv <[actions]> - define each calling convention -that is supported
CallingConv <[actions]> — Define each calling + convention that is supported.

+ +

Assembly Printer -

During the code -emission stage, the code generator may utilize an LLVM pass to produce assembly -output. To do this, you want to implement the code for a printer that converts -LLVM IR to a GAS-format assembly language for your target machine, using the -following steps:

+ +

+During the code emission stage, the code generator may utilize an LLVM pass to +produce assembly output. To do this, you want to implement the code for a +printer that converts LLVM IR to a GAS-format assembly language for your target +machine, using the following steps: +

Define all the assembly strings for your target, adding them to -the instructions defined in the XXXInstrInfo.td file. -(See Instruction Set.) -TableGen will produce an output file (XXXGenAsmWriter.inc) with an -implementation of the printInstruction method for the XXXAsmPrinter class.
Define all the assembly strings for your target, adding them to the + instructions defined in the XXXInstrInfo.td file. + (See Instruction Set.) TableGen will produce + an output file (XXXGenAsmWriter.inc) with an implementation of + the printInstruction method for the XXXAsmPrinter class.
Write XXXTargetAsmInfo.h, which contains the bare-bones -declaration of the XXXTargetAsmInfo class (a subclass of TargetAsmInfo).
Write XXXTargetAsmInfo.h, which contains the bare-bones declaration + of the XXXTargetAsmInfo class (a subclass + of TargetAsmInfo).
Write XXXTargetAsmInfo.cpp, which contains target-specific values -for TargetAsmInfo properties and sometimes new implementations for methods

TargetAsmInfo

Write XXXAsmPrinter.cpp, which implements the AsmPrinter class -that performs the LLVM-to-assembly conversion.
Write XXXAsmPrinter.cpp, which implements the AsmPrinter + class that performs the LLVM-to-assembly conversion.

The code in XXXTargetAsmInfo.h is usually a trivial declaration -of the XXXTargetAsmInfo class for use in XXXTargetAsmInfo.cpp. Similarly, -XXXTargetAsmInfo.cpp usually has a few declarations of XXXTargetAsmInfo replacement -values that override the default values in TargetAsmInfo.cpp. For example in -SparcTargetAsmInfo.cpp,

+The code in XXXTargetAsmInfo.h is usually a trivial declaration of the +XXXTargetAsmInfo class for use in XXXTargetAsmInfo.cpp. +Similarly, XXXTargetAsmInfo.cpp usually has a few declarations of +XXXTargetAsmInfo replacement values that override the default values +in TargetAsmInfo.cpp. For example in SparcTargetAsmInfo.cpp: +

SparcTargetAsmInfo::SparcTargetAsmInfo(const SparcTargetMachine &TM) {
-  Data16bitsDirective = "\t.half\t";
-  Data32bitsDirective = "\t.word\t";
++SparcTargetAsmInfo::SparcTargetAsmInfo(const SparcTargetMachine &TM) {
+  Data16bitsDirective = "\t.half\t";
+  Data32bitsDirective = "\t.word\t";
   Data64bitsDirective = 0;  // .xword is only supported by V9.
-  ZeroDirective = "\t.skip\t";
-  CommentString = "!";
-  ConstantPoolSection = "\t.section \".rodata\",#alloc\n";
+  ZeroDirective = "\t.skip\t";
+  CommentString = "!";
+  ConstantPoolSection = "\t.section \".rodata\",#alloc\n";
 }

The X86 assembly printer implementation (X86TargetAsmInfo) is an -example where the target specific TargetAsmInfo class uses overridden methods: -ExpandInlineAsm and PreferredEHDataFormat.

A target-specific implementation of AsmPrinter is written in -XXXAsmPrinter.cpp, which implements the AsmPrinter class that converts the LLVM -to printable assembly. The implementation must include the following headers -that have declarations for the AsmPrinter and MachineFunctionPass classes. The -MachineFunctionPass is a subclass of FunctionPass.

+The X86 assembly printer implementation (X86TargetAsmInfo) is an +example where the target specific TargetAsmInfo class uses an +overridden methods: ExpandInlineAsm. +

+ +

+A target-specific implementation of AsmPrinter is written in +XXXAsmPrinter.cpp, which implements the AsmPrinter class that +converts the LLVM to printable assembly. The implementation must include the +following headers that have declarations for the AsmPrinter and +MachineFunctionPass classes. The MachineFunctionPass is a +subclass of FunctionPass. +

#include "llvm/CodeGen/AsmPrinter.h"
-#include "llvm/CodeGen/MachineFunctionPass.h" 
++#include "llvm/CodeGen/AsmPrinter.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"

As a FunctionPass, AsmPrinter first calls doInitialization to set -up the AsmPrinter. In SparcAsmPrinter, a Mangler object is instantiated to -process variable names.

+As a FunctionPass, AsmPrinter first +calls doInitialization to set up the AsmPrinter. In +SparcAsmPrinter, a Mangler object is instantiated to process +variable names. +

+ +

+In XXXAsmPrinter.cpp, the runOnMachineFunction method +(declared in MachineFunctionPass) must be implemented +for XXXAsmPrinter. In MachineFunctionPass, +the runOnFunction method invokes runOnMachineFunction. +Target-specific implementations of runOnMachineFunction differ, but +generally do the following to process each machine function: +

In XXXAsmPrinter.cpp, the runOnMachineFunction method (declared -in MachineFunctionPass) must be implemented for XXXAsmPrinter. In -MachineFunctionPass, the runOnFunction method invokes runOnMachineFunction. -Target-specific implementations of runOnMachineFunction differ, but generally -do the following to process each machine function:

call SetupMachineFunction to perform initialization
Call SetupMachineFunction to perform initialization.
call EmitConstantPool to print out (to the output stream) -constants which have been spilled to memory
Call EmitConstantPool to print out (to the output stream) constants + which have been spilled to memory.
call EmitJumpTableInfo to print out jump tables used by the -current function
Call EmitJumpTableInfo to print out jump tables used by the current + function.
print out the label for the current function
Print out the label for the current function.
print out the code for the function, including basic block labels -and the assembly for the instruction (using printInstruction)
Print out the code for the function, including basic block labels and the + assembly for the instruction (using printInstruction)

The XXXAsmPrinter implementation must also include the code -generated by TableGen that is output in the XXXGenAsmWriter.inc file. The code -in XXXGenAsmWriter.inc contains an implementation of the printInstruction -method that may call these methods:

+ +

+The XXXAsmPrinter implementation must also include the code generated +by TableGen that is output in the XXXGenAsmWriter.inc file. The code +in XXXGenAsmWriter.inc contains an implementation of the +printInstruction method that may call these methods: +

printOperand

printImplicitDef
printInlineAsm
printLabel
printPICJumpTableEntry
printPICJumpTableSetLabel

The implementations of printDeclare, printImplicitDef, -printInlineAsm, and printLabel in AsmPrinter.cpp are generally adequate for -printing assembly and do not need to be overridden. (printBasicBlockLabel is -another method that is implemented in AsmPrinter.cpp that may be directly used -in an implementation of XXXAsmPrinter.)

+The implementations of printDeclare, printImplicitDef, +printInlineAsm, and printLabel in AsmPrinter.cpp are +generally adequate for printing assembly and do not need to be +overridden. +

The printOperand method is implemented with a long switch/case +

+The printOperand method is implemented with a long switch/case statement for the type of operand: register, immediate, basic block, external symbol, global address, constant pool index, or jump table index. For an -instruction with a memory address operand, the printMemOperand method should be -implemented to generate the proper output. Similarly, printCCOperand should be -used to print a conditional operand.

+instruction with a memory address operand, the printMemOperand method +should be implemented to generate the proper output. Similarly, +printCCOperand should be used to print a conditional operand. +

+ +

doFinalization should be overridden in XXXAsmPrinter, and +it should be called to shut down the assembly printer. During +doFinalization, global variables and constants are printed to +output. +

doFinalization should be overridden in XXXAsmPrinter, and -it should be called to shut down the assembly printer. During doFinalization, -global variables and constants are printed to output.

+ -

Subtarget Support -

+ -

Subtarget support is used to inform the code generation process -of instruction set variations for a given chip set. For example, the LLVM -SPARC implementation provided covers three major versions of the SPARC -microprocessor architecture: Version 8 (V8, which is a 32-bit architecture), -Version 9 (V9, a 64-bit architecture), and the UltraSPARC architecture. V8 has -16 double-precision floating-point registers that are also usable as either 32 -single-precision or 8 quad-precision registers. V8 is also purely big-endian. V9 -has 32 double-precision floating-point registers that are also usable as 16 +

+ +

+Subtarget support is used to inform the code generation process of instruction +set variations for a given chip set. For example, the LLVM SPARC implementation +provided covers three major versions of the SPARC microprocessor architecture: +Version 8 (V8, which is a 32-bit architecture), Version 9 (V9, a 64-bit +architecture), and the UltraSPARC architecture. V8 has 16 double-precision +floating-point registers that are also usable as either 32 single-precision or 8 +quad-precision registers. V8 is also purely big-endian. V9 has 32 +double-precision floating-point registers that are also usable as 16 quad-precision registers, but cannot be used as single-precision registers. The UltraSPARC architecture combines V9 with UltraSPARC Visual Instruction Set -extensions.

+extensions. +

If subtarget support is needed, you should implement a -target-specific XXXSubtarget class for your architecture. This class should -process the command-line options –mcpu= and –mattr=

+If subtarget support is needed, you should implement a target-specific +XXXSubtarget class for your architecture. This class should process the +command-line options -mcpu= and -mattr=. +

TableGen uses definitions in the Target.td and Sparc.td files to -generate code in SparcGenSubtarget.inc. In Target.td, shown below, the -SubtargetFeature interface is defined. The first 4 string parameters of the -SubtargetFeature interface are a feature name, an attribute set by the feature, -the value of the attribute, and a description of the feature. (The fifth -parameter is a list of features whose presence is implied, and its default -value is an empty array.)

+TableGen uses definitions in the Target.td and Sparc.td files +to generate code in SparcGenSubtarget.inc. In Target.td, shown +below, the SubtargetFeature interface is defined. The first 4 string +parameters of the SubtargetFeature interface are a feature name, an +attribute set by the feature, the value of the attribute, and a description of +the feature. (The fifth parameter is a list of features whose presence is +implied, and its default value is an empty array.) +

class SubtargetFeature<string n, string a,  string v, string d,
++class SubtargetFeature<string n, string a,  string v, string d,
                        list<SubtargetFeature> i = []> {
   string Name = n;
   string Attribute = a;
@@ -1788,60 +2197,64 @@ value is an empty array.)
 }

In the Sparc.td file, the SubtargetFeature is used to define the -following features.

+ +

+In the Sparc.td file, the SubtargetFeature is used to define the +following features. +

def FeatureV9 : SubtargetFeature<"v9", "IsV9", "true",
-                     "Enable SPARC-V9 instructions">;
-def FeatureV8Deprecated : SubtargetFeature<"deprecated-v8", 
-                     "V8DeprecatedInsts", "true",
-                     "Enable deprecated V8 instructions in V9 mode">;
-def FeatureVIS : SubtargetFeature<"vis", "IsVIS", "true",
-                     "Enable UltraSPARC Visual Instruction Set extensions">;
++def FeatureV9 : SubtargetFeature<"v9", "IsV9", "true",
+                     "Enable SPARC-V9 instructions">;
+def FeatureV8Deprecated : SubtargetFeature<"deprecated-v8", 
+                     "V8DeprecatedInsts", "true",
+                     "Enable deprecated V8 instructions in V9 mode">;
+def FeatureVIS : SubtargetFeature<"vis", "IsVIS", "true",
+                     "Enable UltraSPARC Visual Instruction Set extensions">;

Elsewhere in Sparc.td, the Proc class is defined and then is used -to define particular SPARC processor subtypes that may have the previously -described features.

+Elsewhere in Sparc.td, the Proc class is defined and then is used to +define particular SPARC processor subtypes that may have the previously +described features. +

class Proc<string Name, list<SubtargetFeature> Features>
- : Processor<Name, NoItineraries, Features>;
++class Proc<string Name, list<SubtargetFeature> Features>
+  : Processor<Name, NoItineraries, Features>;
  
-def : Proc<"generic",         []>;
-def : Proc<"v8",              []>;
-def : Proc<"supersparc",      []>;
-def : Proc<"sparclite",       []>;
-def : Proc<"f934",            []>;
-def : Proc<"hypersparc",      []>;
-def : Proc<"sparclite86x",    []>;
-def : Proc<"sparclet",        []>;
-def : Proc<"tsc701",          []>;
-def : Proc<"v9",              [FeatureV9]>;
-def : Proc<"ultrasparc",      [FeatureV9, FeatureV8Deprecated]>;
-def : Proc<"ultrasparc3",     [FeatureV9, FeatureV8Deprecated]>;
-def : Proc<"ultrasparc3-vis", [FeatureV9, FeatureV8Deprecated, FeatureVIS]>;
+def : Proc<"generic",         []>;
+def : Proc<"v8",              []>;
+def : Proc<"supersparc",      []>;
+def : Proc<"sparclite",       []>;
+def : Proc<"f934",            []>;
+def : Proc<"hypersparc",      []>;
+def : Proc<"sparclite86x",    []>;
+def : Proc<"sparclet",        []>;
+def : Proc<"tsc701",          []>;
+def : Proc<"v9",              [FeatureV9]>;
+def : Proc<"ultrasparc",      [FeatureV9, FeatureV8Deprecated]>;
+def : Proc<"ultrasparc3",     [FeatureV9, FeatureV8Deprecated]>;
+def : Proc<"ultrasparc3-vis", [FeatureV9, FeatureV8Deprecated, FeatureVIS]>;

From Target.td and Sparc.td files, the resulting +

+From Target.td and Sparc.td files, the resulting SparcGenSubtarget.inc specifies enum values to identify the features, arrays of constants to represent the CPU features and CPU subtypes, and the ParseSubtargetFeatures method that parses the features string that sets -specified subtarget options. The generated SparcGenSubtarget.inc file should be -included in the SparcSubtarget.cpp. The target-specific implementation of the XXXSubtarget -method should follow this pseudocode:

+specified subtarget options. The generated SparcGenSubtarget.inc file +should be included in the SparcSubtarget.cpp. The target-specific +implementation of the XXXSubtarget method should follow this pseudocode: +

XXXSubtarget::XXXSubtarget(const Module &M, const std::string &FS) {
++XXXSubtarget::XXXSubtarget(const Module &M, const std::string &FS) {
   // Set the default features
   // Determine default and user specified characteristics of the CPU
   // Call ParseSubtargetFeatures(FS, CPU) to parse the features string
@@ -1850,61 +2263,77 @@ method should follow this pseudocode:

+ -

JIT Support -

+ -

The implementation of a target machine optionally includes a Just-In-Time -(JIT) code generator that emits machine code and auxiliary structures as binary -output that can be written directly to memory. -To do this, implement JIT code generation by performing the following -steps:

+ +

+The implementation of a target machine optionally includes a Just-In-Time (JIT) +code generator that emits machine code and auxiliary structures as binary output +that can be written directly to memory. To do this, implement JIT code +generation by performing the following steps: +

Write an XXXCodeEmitter.cpp file that contains a machine function -pass that transforms target-machine instructions into relocatable machine code.
Write an XXXJITInfo.cpp file that implements the JIT interfaces -for target-specific code-generation -activities, such as emitting machine code and stubs.
Write an XXXJITInfo.cpp file that implements the JIT interfaces for + target-specific code-generation activities, such as emitting machine code + and stubs.
Modify XXXTargetMachine so that it provides a TargetJITInfo -object through its getJITInfo method.
Modify XXXTargetMachine so that it provides a + TargetJITInfo object through its getJITInfo method.

There are several different approaches to writing the JIT support -code. For instance, TableGen and target descriptor files may be used for -creating a JIT code generator, but are not mandatory. For the Alpha and PowerPC -target machines, TableGen is used to generate XXXGenCodeEmitter.inc, which +

+There are several different approaches to writing the JIT support code. For +instance, TableGen and target descriptor files may be used for creating a JIT +code generator, but are not mandatory. For the Alpha and PowerPC target +machines, TableGen is used to generate XXXGenCodeEmitter.inc, which contains the binary coding of machine instructions and the -getBinaryCodeForInstr method to access those codes. Other JIT implementations -do not.

+getBinaryCodeForInstr method to access those codes. Other JIT +implementations do not. +

+ +

+Both XXXJITInfo.cpp and XXXCodeEmitter.cpp must include the +llvm/CodeGen/MachineCodeEmitter.h header file that defines the +MachineCodeEmitter class containing code for several callback functions +that write data (in bytes, words, strings, etc.) to the output stream. +

Both XXXJITInfo.cpp and XXXCodeEmitter.cpp must include the -llvm/CodeGen/MachineCodeEmitter.h header file that defines the MachineCodeEmitter -class containing code for several callback functions that write data (in bytes, -words, strings, etc.) to the output stream.

Machine Code Emitter -

- -

In XXXCodeEmitter.cpp, a target-specific of the Emitter class is -implemented as a function pass (subclass of MachineFunctionPass). The -target-specific implementation of runOnMachineFunction (invoked by -runOnFunction in MachineFunctionPass) iterates through the MachineBasicBlock -calls emitInstruction to process each instruction and emit binary code. emitInstruction -is largely implemented with case statements on the instruction types defined in -XXXInstrInfo.h. For example, in X86CodeEmitter.cpp, the emitInstruction method -is built around the following switch/case statements:

+ + +

+ +

+In XXXCodeEmitter.cpp, a target-specific of the Emitter class +is implemented as a function pass (subclass +of MachineFunctionPass). The target-specific implementation +of runOnMachineFunction (invoked by +runOnFunction in MachineFunctionPass) iterates through the +MachineBasicBlock calls emitInstruction to process each +instruction and emit binary code. emitInstruction is largely +implemented with case statements on the instruction types defined in +XXXInstrInfo.h. For example, in X86CodeEmitter.cpp, +the emitInstruction method is built around the following switch/case +statements: +

switch (Desc->TSFlags & X86::FormMask) {
++switch (Desc->TSFlags & X86::FormMask) {
 case X86II::Pseudo:  // for not yet implemented instructions 
    ...               // or pseudo-instructions
    break;
@@ -1944,23 +2373,26 @@ case X86II::MRMInitReg: // for instructions whose source and
 }

The implementations of these case statements often first emit the -opcode and then get the operand(s). Then depending upon the operand, helper -methods may be called to process the operand(s). For example, in X86CodeEmitter.cpp, -for the X86II::AddRegFrm case, the first data emitted (by emitByte) is the -opcode added to the register operand. Then an object representing the machine -operand, MO1, is extracted. The helper methods such as isImmediate, + +

+The implementations of these case statements often first emit the opcode and +then get the operand(s). Then depending upon the operand, helper methods may be +called to process the operand(s). For example, in X86CodeEmitter.cpp, +for the X86II::AddRegFrm case, the first data emitted +(by emitByte) is the opcode added to the register operand. Then an +object representing the machine operand, MO1, is extracted. The helper +methods such as isImmediate, isGlobalAddress, isExternalSymbol, isConstantPoolIndex, and -isJumpTableIndex -determine the operand type. (X86CodeEmitter.cpp also has private methods such -as emitConstant, emitGlobalAddress, +isJumpTableIndex determine the operand +type. (X86CodeEmitter.cpp also has private methods such +as emitConstant, emitGlobalAddress, emitExternalSymbolAddress, emitConstPoolAddress, -and emitJumpTableAddress that emit the data into the output stream.)

+and emitJumpTableAddress that emit the data into the output stream.) +

case X86II::AddRegFrm:
++case X86II::AddRegFrm:
   MCE.emitByte(BaseOpcode + getX86RegNum(MI.getOperand(CurOp++).getReg()));
   
   if (CurOp != NumOps) {
@@ -1989,75 +2421,97 @@ and emitJumpTableAddress that emit the data into the output stream.)

In the previous example, XXXCodeEmitter.cpp uses the variable rt, -which is a RelocationType enum that may be used to relocate addresses (for -example, a global address with a PIC base offset). The RelocationType enum for -that target is defined in the short target-specific XXXRelocations.h file. The -RelocationType is used by the relocate method defined in XXXJITInfo.cpp to -rewrite addresses for referenced global symbols.

For example, X86Relocations.h specifies the following relocation -types for the X86 addresses. In all four cases, the relocated value is added to -the value already in memory. For reloc_pcrel_word and reloc_picrel_word, -there is an additional initial adjustment.

+In the previous example, XXXCodeEmitter.cpp uses the +variable rt, which is a RelocationType enum that may be used to +relocate addresses (for example, a global address with a PIC base offset). The +RelocationType enum for that target is defined in the short +target-specific XXXRelocations.h file. The RelocationType is used by +the relocate method defined in XXXJITInfo.cpp to rewrite +addresses for referenced global symbols. +

+ +

+For example, X86Relocations.h specifies the following relocation types +for the X86 addresses. In all four cases, the relocated value is added to the +value already in memory. For reloc_pcrel_word +and reloc_picrel_word, there is an additional initial adjustment. +

enum RelocationType {
-  reloc_pcrel_word = 0,  // add reloc value after adjusting for the PC loc
-  reloc_picrel_word = 1, // add reloc value after adjusting for the PIC base
++enum RelocationType {
+  reloc_pcrel_word = 0,    // add reloc value after adjusting for the PC loc
+  reloc_picrel_word = 1,   // add reloc value after adjusting for the PIC base
   reloc_absolute_word = 2, // absolute relocation; no additional adjustment 
   reloc_absolute_dword = 3 // absolute relocation; no additional adjustment
 };

+ +

+ -

Target JIT Info -

XXXJITInfo.cpp implements the JIT interfaces for target-specific code-generation -activities, such as emitting machine code and stubs. At minimum, -a target-specific version of XXXJITInfo implements the following:

+ + +

+ +

+XXXJITInfo.cpp implements the JIT interfaces for target-specific +code-generation activities, such as emitting machine code and stubs. At minimum, +a target-specific version of XXXJITInfo implements the following: +

getLazyResolverFunction – initializes the JIT, gives the -target a function that is used for compilation
getLazyResolverFunction — Initializes the JIT, gives the + target a function that is used for compilation.
emitFunctionStub – returns a native function with a -specified address for a callback function
emitFunctionStub — Returns a native function with a specified + address for a callback function.
relocate – changes the addresses of referenced globals, -based on relocation types
relocate — Changes the addresses of referenced globals, based + on relocation types.
callback function that are wrappers to a function stub that is -used when the real target is not initially known
Callback function that are wrappers to a function stub that is used when the + real target is not initially known.

getLazyResolverFunction is generally trivial to implement. It -makes the incoming parameter as the global JITCompilerFunction and returns the +

+getLazyResolverFunction is generally trivial to implement. It makes the +incoming parameter as the global JITCompilerFunction and returns the callback function that will be used a function wrapper. For the Alpha target -(in AlphaJITInfo.cpp), the getLazyResolverFunction implementation is simply:

+(in AlphaJITInfo.cpp), the getLazyResolverFunction +implementation is simply: +

TargetJITInfo::LazyResolverFn AlphaJITInfo::getLazyResolverFunction(  
-                                            JITCompilerFn F) 
-{
++TargetJITInfo::LazyResolverFn AlphaJITInfo::getLazyResolverFunction(  
+                                            JITCompilerFn F) {
   JITCompilerFunction = F;
   return AlphaCompilationCallback;
 }

For the X86 target, the getLazyResolverFunction implementation is -a little more complication, because it returns a different callback function -for processors with SSE instructions and XMM registers.

The callback function initially saves and later restores the -callee register values, incoming arguments, and frame and return address. The -callback function needs low-level access to the registers or stack, so it is typically -implemented with assembler.

+For the X86 target, the getLazyResolverFunction implementation is a +little more complication, because it returns a different callback function for +processors with SSE instructions and XMM registers. +

+ +

+The callback function initially saves and later restores the callee register +values, incoming arguments, and frame and return address. The callback function +needs low-level access to the registers or stack, so it is typically implemented +with assembler. +

+ +

@@ -2065,12 +2519,12 @@ implemented with assembler.

+ src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS">

+ src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"> Mason Woo and Misha Brukman
- The LLVM Compiler Infrastructure + The LLVM Compiler Infrastructure
Last modified: $Date$

Writing an LLVM Compiler Backend -

Introduction -

Audience -

Prerequisite Reading -

Basic Steps -

Preliminaries -

Target Machine -

+ Target Registration +

Register Set and Register Classes -

Defining a Register -

Defining a Register Class -

+ Implement a subclass of + TargetRegisterInfo +

Instruction Set -

+ Instruction Operand Mapping +

+ Implement a subclass of + TargetInstrInfo +

Branch Folding and If Conversion -

Instruction Selector -

The SelectionDAG Legalize Phase -

Promote -

Expand -

Custom -

Legal -

Calling Conventions -

Assembly Printer -

Subtarget Support -

JIT Support -

Machine Code Emitter -

Target JIT Info -