mirror of
https://github.com/RPCS3/llvm.git
synced 2024-12-28 23:43:50 +00:00
becdf4d7cd
Relationship maps are represented as InstrMapping records which are parsed by TableGen and the information is used to construct mapping tables to represent appropriate relations between instructions. These tables are emitted into XXXGenInstrInfo.inc file along with the functions to query them. Patch by Jyotsna Verma <jverma@codeaurora.org>. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@166685 91177308-0d34-0410-b5e6-96231b3b80d8
2558 lines
91 KiB
HTML
2558 lines
91 KiB
HTML
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
|
|
"http://www.w3.org/TR/html4/strict.dtd">
|
|
<html>
|
|
<head>
|
|
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
|
|
<title>Writing an LLVM Compiler Backend</title>
|
|
<link rel="stylesheet" href="_static/llvm.css" type="text/css">
|
|
</head>
|
|
|
|
<body>
|
|
|
|
<h1>
|
|
Writing an LLVM Compiler Backend
|
|
</h1>
|
|
|
|
<ol>
|
|
<li><a href="#intro">Introduction</a>
|
|
<ul>
|
|
<li><a href="#Audience">Audience</a></li>
|
|
<li><a href="#Prerequisite">Prerequisite Reading</a></li>
|
|
<li><a href="#Basic">Basic Steps</a></li>
|
|
<li><a href="#Preliminaries">Preliminaries</a></li>
|
|
</ul>
|
|
<li><a href="#TargetMachine">Target Machine</a></li>
|
|
<li><a href="#TargetRegistration">Target Registration</a></li>
|
|
<li><a href="#RegisterSet">Register Set and Register Classes</a>
|
|
<ul>
|
|
<li><a href="#RegisterDef">Defining a Register</a></li>
|
|
<li><a href="#RegisterClassDef">Defining a Register Class</a></li>
|
|
<li><a href="#implementRegister">Implement a subclass of TargetRegisterInfo</a></li>
|
|
</ul></li>
|
|
<li><a href="#InstructionSet">Instruction Set</a>
|
|
<ul>
|
|
<li><a href="#operandMapping">Instruction Operand Mapping</a></li>
|
|
<li><a href="#relationMapping">Instruction Relation Mapping</a></li>
|
|
<li><a href="#implementInstr">Implement a subclass of TargetInstrInfo</a></li>
|
|
<li><a href="#branchFolding">Branch Folding and If Conversion</a></li>
|
|
</ul></li>
|
|
<li><a href="#InstructionSelector">Instruction Selector</a>
|
|
<ul>
|
|
<li><a href="#LegalizePhase">The SelectionDAG Legalize Phase</a>
|
|
<ul>
|
|
<li><a href="#promote">Promote</a></li>
|
|
<li><a href="#expand">Expand</a></li>
|
|
<li><a href="#custom">Custom</a></li>
|
|
<li><a href="#legal">Legal</a></li>
|
|
</ul></li>
|
|
<li><a href="#callingConventions">Calling Conventions</a></li>
|
|
</ul></li>
|
|
<li><a href="#assemblyPrinter">Assembly Printer</a></li>
|
|
<li><a href="#subtargetSupport">Subtarget Support</a></li>
|
|
<li><a href="#jitSupport">JIT Support</a>
|
|
<ul>
|
|
<li><a href="#mce">Machine Code Emitter</a></li>
|
|
<li><a href="#targetJITInfo">Target JIT Info</a></li>
|
|
</ul></li>
|
|
</ol>
|
|
|
|
<div class="doc_author">
|
|
<p>Written by <a href="http://www.woo.com">Mason Woo</a> and
|
|
<a href="http://misha.brukman.net">Misha Brukman</a></p>
|
|
</div>
|
|
|
|
<!-- *********************************************************************** -->
|
|
<h2>
|
|
<a name="intro">Introduction</a>
|
|
</h2>
|
|
<!-- *********************************************************************** -->
|
|
|
|
<div>
|
|
|
|
<p>
|
|
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).
|
|
</p>
|
|
|
|
<p>
|
|
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.
|
|
</p>
|
|
|
|
<p>
|
|
The document focuses on existing examples found in subdirectories
|
|
of <tt>llvm/lib/Target</tt> 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.
|
|
</p>
|
|
|
|
<h3>
|
|
<a name="Audience">Audience</a>
|
|
</h3>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
The audience for this document is anyone who needs to write an LLVM backend to
|
|
generate code for a specific hardware or software target.
|
|
</p>
|
|
|
|
</div>
|
|
|
|
<h3>
|
|
<a name="Prerequisite">Prerequisite Reading</a>
|
|
</h3>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
These essential documents must be read before reading this document:
|
|
</p>
|
|
|
|
<ul>
|
|
<li><i><a href="LangRef.html">LLVM Language Reference
|
|
Manual</a></i> — a reference manual for the LLVM assembly language.</li>
|
|
|
|
<li><i><a href="CodeGenerator.html">The LLVM
|
|
Target-Independent Code Generator</a></i> — 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.</li>
|
|
|
|
<li><i><a href="TableGenFundamentals.html">TableGen
|
|
Fundamentals</a></i> —a document that describes the TableGen
|
|
(<tt>tblgen</tt>) application that manages domain-specific information to
|
|
support LLVM code generation. TableGen processes input from a target
|
|
description file (<tt>.td</tt> suffix) and generates C++ code that can be
|
|
used for code generation.</li>
|
|
|
|
<li><i><a href="WritingAnLLVMPass.html">Writing an LLVM
|
|
Pass</a></i> — The assembly printer is a <tt>FunctionPass</tt>, as are
|
|
several SelectionDAG processing steps.</li>
|
|
</ul>
|
|
|
|
<p>
|
|
To follow the SPARC examples in this document, have a copy of
|
|
<i><a href="http://www.sparc.org/standards/V8.pdf">The SPARC Architecture
|
|
Manual, Version 8</a></i> for reference. For details about the ARM instruction
|
|
set, refer to the <i><a href="http://infocenter.arm.com/">ARM Architecture
|
|
Reference Manual</a></i>. For more about the GNU Assembler format
|
|
(<tt>GAS</tt>), see
|
|
<i><a href="http://sourceware.org/binutils/docs/as/index.html">Using As</a></i>,
|
|
especially for the assembly printer. <i>Using As</i> contains a list of target
|
|
machine dependent features.
|
|
</p>
|
|
|
|
</div>
|
|
|
|
<h3>
|
|
<a name="Basic">Basic Steps</a>
|
|
</h3>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
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:
|
|
</p>
|
|
|
|
<ul>
|
|
<li>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
|
|
<tt>SparcTargetMachine.cpp</tt> and <tt>SparcTargetMachine.h</tt>, but
|
|
change the file names for your target. Similarly, change code that
|
|
references "Sparc" to reference your target. </li>
|
|
|
|
<li>Describe the register set of the target. Use TableGen to generate code for
|
|
register definition, register aliases, and register classes from a
|
|
target-specific <tt>RegisterInfo.td</tt> 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.</li>
|
|
|
|
<li>Describe the instruction set of the target. Use TableGen to generate code
|
|
for target-specific instructions from target-specific versions of
|
|
<tt>TargetInstrFormats.td</tt> and <tt>TargetInstrInfo.td</tt>. You should
|
|
write additional code for a subclass of the TargetInstrInfo class to
|
|
represent machine instructions supported by the target machine. </li>
|
|
|
|
<li>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 <tt>TargetInstrInfo.td</tt>. Write code
|
|
for <tt>XXXISelDAGToDAG.cpp</tt>, where XXX identifies the specific target,
|
|
to perform pattern matching and DAG-to-DAG instruction selection. Also write
|
|
code in <tt>XXXISelLowering.cpp</tt> to replace or remove operations and
|
|
data types that are not supported natively in a SelectionDAG. </li>
|
|
|
|
<li>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 <tt>TargetInstrInfo.td</tt>. You
|
|
should also write code for a subclass of AsmPrinter that performs the
|
|
LLVM-to-assembly conversion and a trivial subclass of TargetAsmInfo.</li>
|
|
|
|
<li>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 <tt>-mcpu=</tt>
|
|
and <tt>-mattr=</tt> command-line options.</li>
|
|
|
|
<li>Optionally, add JIT support and create a machine code emitter (subclass of
|
|
TargetJITInfo) that is used to emit binary code directly into memory. </li>
|
|
</ul>
|
|
|
|
<p>
|
|
In the <tt>.cpp</tt> and <tt>.h</tt>. 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.
|
|
</p>
|
|
|
|
</div>
|
|
|
|
<h3>
|
|
<a name="Preliminaries">Preliminaries</a>
|
|
</h3>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
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 <a href="CodeGenerator.html">LLVM
|
|
Target-Independent Code Generator</a> document.
|
|
</p>
|
|
|
|
<p>
|
|
First, you should create a subdirectory under <tt>lib/Target</tt> to hold all
|
|
the files related to your target. If your target is called "Dummy," create the
|
|
directory <tt>lib/Target/Dummy</tt>.
|
|
</p>
|
|
|
|
<p>
|
|
In this new
|
|
directory, create a <tt>Makefile</tt>. It is easiest to copy a
|
|
<tt>Makefile</tt> of another target and modify it. It should at least contain
|
|
the <tt>LEVEL</tt>, <tt>LIBRARYNAME</tt> and <tt>TARGET</tt> variables, and then
|
|
include <tt>$(LEVEL)/Makefile.common</tt>. The library can be
|
|
named <tt>LLVMDummy</tt> (for example, see the MIPS target). Alternatively, you
|
|
can split the library into <tt>LLVMDummyCodeGen</tt>
|
|
and <tt>LLVMDummyAsmPrinter</tt>, the latter of which should be implemented in a
|
|
subdirectory below <tt>lib/Target/Dummy</tt> (for example, see the PowerPC
|
|
target).
|
|
</p>
|
|
|
|
<p>
|
|
Note that these two naming schemes are hardcoded into <tt>llvm-config</tt>.
|
|
Using any other naming scheme will confuse <tt>llvm-config</tt> and produce a
|
|
lot of (seemingly unrelated) linker errors when linking <tt>llc</tt>.
|
|
</p>
|
|
|
|
<p>
|
|
To make your target actually do something, you need to implement a subclass of
|
|
<tt>TargetMachine</tt>. This implementation should typically be in the file
|
|
<tt>lib/Target/DummyTargetMachine.cpp</tt>, but any file in
|
|
the <tt>lib/Target</tt> 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 <tt>LLVMTargetMachine</tt>. (To create a
|
|
target from scratch, create a subclass of <tt>TargetMachine</tt>.)
|
|
</p>
|
|
|
|
<p>
|
|
To get LLVM to actually build and link your target, you need to add it to
|
|
the <tt>TARGETS_TO_BUILD</tt> variable. To do this, you modify the configure
|
|
script to know about your target when parsing the <tt>--enable-targets</tt>
|
|
option. Search the configure script for <tt>TARGETS_TO_BUILD</tt>, add your
|
|
target to the lists there (some creativity required), and then
|
|
reconfigure. Alternatively, you can change <tt>autotools/configure.ac</tt> and
|
|
regenerate configure by running <tt>./autoconf/AutoRegen.sh</tt>.
|
|
</p>
|
|
|
|
</div>
|
|
|
|
</div>
|
|
|
|
<!-- *********************************************************************** -->
|
|
<h2>
|
|
<a name="TargetMachine">Target Machine</a>
|
|
</h2>
|
|
<!-- *********************************************************************** -->
|
|
|
|
<div>
|
|
|
|
<p>
|
|
<tt>LLVMTargetMachine</tt> is designed as a base class for targets implemented
|
|
with the LLVM target-independent code generator. The <tt>LLVMTargetMachine</tt>
|
|
class should be specialized by a concrete target class that implements the
|
|
various virtual methods. <tt>LLVMTargetMachine</tt> is defined as a subclass of
|
|
<tt>TargetMachine</tt> in <tt>include/llvm/Target/TargetMachine.h</tt>. The
|
|
<tt>TargetMachine</tt> class implementation (<tt>TargetMachine.cpp</tt>) also
|
|
processes numerous command-line options.
|
|
</p>
|
|
|
|
<p>
|
|
To create a concrete target-specific subclass of <tt>LLVMTargetMachine</tt>,
|
|
start by copying an existing <tt>TargetMachine</tt> 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 <tt>SparcTargetMachine.h</tt> and
|
|
<tt>SparcTargetMachine.cpp</tt>.
|
|
</p>
|
|
|
|
<p>
|
|
For a target machine <tt>XXX</tt>, the implementation of
|
|
<tt>XXXTargetMachine</tt> must have access methods to obtain objects that
|
|
represent target components. These methods are named <tt>get*Info</tt>, and are
|
|
intended to obtain the instruction set (<tt>getInstrInfo</tt>), register set
|
|
(<tt>getRegisterInfo</tt>), stack frame layout (<tt>getFrameInfo</tt>), and
|
|
similar information. <tt>XXXTargetMachine</tt> must also implement the
|
|
<tt>getDataLayout</tt> method to access an object with target-specific data
|
|
characteristics, such as data type size and alignment requirements.
|
|
</p>
|
|
|
|
<p>
|
|
For instance, for the SPARC target, the header file
|
|
<tt>SparcTargetMachine.h</tt> declares prototypes for several <tt>get*Info</tt>
|
|
and <tt>getDataLayout</tt> methods that simply return a class member.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
namespace llvm {
|
|
|
|
class Module;
|
|
|
|
class SparcTargetMachine : public LLVMTargetMachine {
|
|
const DataLayout DataLayout; // Calculates type size & alignment
|
|
SparcSubtarget Subtarget;
|
|
SparcInstrInfo InstrInfo;
|
|
TargetFrameInfo FrameInfo;
|
|
|
|
protected:
|
|
virtual const TargetAsmInfo *createTargetAsmInfo() const;
|
|
|
|
public:
|
|
SparcTargetMachine(const Module &M, const std::string &FS);
|
|
|
|
virtual const SparcInstrInfo *getInstrInfo() const {return &InstrInfo; }
|
|
virtual const TargetFrameInfo *getFrameInfo() const {return &FrameInfo; }
|
|
virtual const TargetSubtarget *getSubtargetImpl() const{return &Subtarget; }
|
|
virtual const TargetRegisterInfo *getRegisterInfo() const {
|
|
return &InstrInfo.getRegisterInfo();
|
|
}
|
|
virtual const DataLayout *getDataLayout() const { return &DataLayout; }
|
|
static unsigned getModuleMatchQuality(const Module &M);
|
|
|
|
// Pass Pipeline Configuration
|
|
virtual bool addInstSelector(PassManagerBase &PM, bool Fast);
|
|
virtual bool addPreEmitPass(PassManagerBase &PM, bool Fast);
|
|
};
|
|
|
|
} // end namespace llvm
|
|
</pre>
|
|
</div>
|
|
|
|
<ul>
|
|
<li><tt>getInstrInfo()</tt></li>
|
|
<li><tt>getRegisterInfo()</tt></li>
|
|
<li><tt>getFrameInfo()</tt></li>
|
|
<li><tt>getDataLayout()</tt></li>
|
|
<li><tt>getSubtargetImpl()</tt></li>
|
|
</ul>
|
|
|
|
<p>For some targets, you also need to support the following methods:</p>
|
|
|
|
<ul>
|
|
<li><tt>getTargetLowering()</tt></li>
|
|
<li><tt>getJITInfo()</tt></li>
|
|
</ul>
|
|
|
|
<p>
|
|
In addition, the <tt>XXXTargetMachine</tt> constructor should specify a
|
|
<tt>TargetDescription</tt> 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:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
SparcTargetMachine::SparcTargetMachine(const Module &M, const std::string &FS)
|
|
: DataLayout("E-p:32:32-f128:128:128"),
|
|
Subtarget(M, FS), InstrInfo(Subtarget),
|
|
FrameInfo(TargetFrameInfo::StackGrowsDown, 8, 0) {
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>Hyphens separate portions of the <tt>TargetDescription</tt> string.</p>
|
|
|
|
<ul>
|
|
<li>An upper-case "<tt>E</tt>" in the string indicates a big-endian target data
|
|
model. a lower-case "<tt>e</tt>" indicates little-endian.</li>
|
|
|
|
<li>"<tt>p:</tt>" is followed by pointer information: size, ABI alignment, and
|
|
preferred alignment. If only two figures follow "<tt>p:</tt>", then the
|
|
first value is pointer size, and the second value is both ABI and preferred
|
|
alignment.</li>
|
|
|
|
<li>Then a letter for numeric type alignment: "<tt>i</tt>", "<tt>f</tt>",
|
|
"<tt>v</tt>", or "<tt>a</tt>" (corresponding to integer, floating point,
|
|
vector, or aggregate). "<tt>i</tt>", "<tt>v</tt>", or "<tt>a</tt>" are
|
|
followed by ABI alignment and preferred alignment. "<tt>f</tt>" is followed
|
|
by three values: the first indicates the size of a long double, then ABI
|
|
alignment, and then ABI preferred alignment.</li>
|
|
</ul>
|
|
|
|
</div>
|
|
|
|
<!-- *********************************************************************** -->
|
|
<h2>
|
|
<a name="TargetRegistration">Target Registration</a>
|
|
</h2>
|
|
<!-- *********************************************************************** -->
|
|
|
|
<div>
|
|
|
|
<p>
|
|
You must also register your target with the <tt>TargetRegistry</tt>, which is
|
|
what other LLVM tools use to be able to lookup and use your target at
|
|
runtime. The <tt>TargetRegistry</tt> can be used directly, but for most targets
|
|
there are helper templates which should take care of the work for you.</p>
|
|
|
|
<p>
|
|
All targets should declare a global <tt>Target</tt> 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 <tt>RegisterTarget</tt> template to register the target. For example, the Sparc registration code looks like this:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
Target llvm::TheSparcTarget;
|
|
|
|
extern "C" void LLVMInitializeSparcTargetInfo() {
|
|
RegisterTarget<Triple::sparc, /*HasJIT=*/false>
|
|
X(TheSparcTarget, "sparc", "Sparc");
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
This allows the <tt>TargetRegistry</tt> 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:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
extern "C" void LLVMInitializeSparcAsmPrinter() {
|
|
RegisterAsmPrinter<SparcAsmPrinter> X(TheSparcTarget);
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
For more information, see
|
|
"<a href="/doxygen/TargetRegistry_8h-source.html">llvm/Target/TargetRegistry.h</a>".
|
|
</p>
|
|
|
|
</div>
|
|
|
|
<!-- *********************************************************************** -->
|
|
<h2>
|
|
<a name="RegisterSet">Register Set and Register Classes</a>
|
|
</h2>
|
|
<!-- *********************************************************************** -->
|
|
|
|
<div>
|
|
|
|
<p>
|
|
You should describe a concrete target-specific class that represents the
|
|
register file of a target machine. This class is called <tt>XXXRegisterInfo</tt>
|
|
(where <tt>XXX</tt> identifies the target) and represents the class register
|
|
file data that is used for register allocation. It also describes the
|
|
interactions between registers.
|
|
</p>
|
|
|
|
<p>
|
|
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.
|
|
</p>
|
|
|
|
<p>
|
|
Much of the code for registers, including register definition, register aliases,
|
|
and register classes, is generated by TableGen from <tt>XXXRegisterInfo.td</tt>
|
|
input files and placed in <tt>XXXGenRegisterInfo.h.inc</tt> and
|
|
<tt>XXXGenRegisterInfo.inc</tt> output files. Some of the code in the
|
|
implementation of <tt>XXXRegisterInfo</tt> requires hand-coding.
|
|
</p>
|
|
|
|
<!-- ======================================================================= -->
|
|
<h3>
|
|
<a name="RegisterDef">Defining a Register</a>
|
|
</h3>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
The <tt>XXXRegisterInfo.td</tt> file typically starts with register definitions
|
|
for a target machine. The <tt>Register</tt> class (specified
|
|
in <tt>Target.td</tt>) is used to define an object for each register. The
|
|
specified string <tt>n</tt> becomes the <tt>Name</tt> of the register. The
|
|
basic <tt>Register</tt> object does not have any subregisters and does not
|
|
specify any aliases.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
class Register<string n> {
|
|
string Namespace = "";
|
|
string AsmName = n;
|
|
string Name = n;
|
|
int SpillSize = 0;
|
|
int SpillAlignment = 0;
|
|
list<Register> Aliases = [];
|
|
list<Register> SubRegs = [];
|
|
list<int> DwarfNumbers = [];
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
For example, in the <tt>X86RegisterInfo.td</tt> file, there are register
|
|
definitions that utilize the Register class, such as:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
def AL : Register<"AL">, DwarfRegNum<[0, 0, 0]>;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
This defines the register <tt>AL</tt> and assigns it values (with
|
|
<tt>DwarfRegNum</tt>) that are used by <tt>gcc</tt>, <tt>gdb</tt>, or a debug
|
|
information writer to identify a register. For register
|
|
<tt>AL</tt>, <tt>DwarfRegNum</tt> 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.
|
|
</p>
|
|
|
|
<p>
|
|
From the previously described line in the <tt>X86RegisterInfo.td</tt> file,
|
|
TableGen generates this code in the <tt>X86GenRegisterInfo.inc</tt> file:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
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 }, ...
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
From the register info file, TableGen generates a <tt>TargetRegisterDesc</tt>
|
|
object for each register. <tt>TargetRegisterDesc</tt> is defined in
|
|
<tt>include/llvm/Target/TargetRegisterInfo.h</tt> with the following fields:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
struct TargetRegisterDesc {
|
|
const char *AsmName; // Assembly language name for the register
|
|
const char *Name; // Printable name for the reg (for debugging)
|
|
const unsigned *AliasSet; // Register Alias Set
|
|
const unsigned *SubRegs; // Sub-register set
|
|
const unsigned *ImmSubRegs; // Immediate sub-register set
|
|
const unsigned *SuperRegs; // Super-register set
|
|
};</pre>
|
|
</div>
|
|
|
|
<p>
|
|
TableGen uses the entire target description file (<tt>.td</tt>) to determine
|
|
text names for the register (in the <tt>AsmName</tt> and <tt>Name</tt> fields of
|
|
<tt>TargetRegisterDesc</tt>) and the relationships of other registers to the
|
|
defined register (in the other <tt>TargetRegisterDesc</tt> fields). In this
|
|
example, other definitions establish the registers "<tt>AX</tt>",
|
|
"<tt>EAX</tt>", and "<tt>RAX</tt>" as aliases for one another, so TableGen
|
|
generates a null-terminated array (<tt>AL_AliasSet</tt>) for this register alias
|
|
set.
|
|
</p>
|
|
|
|
<p>
|
|
The <tt>Register</tt> class is commonly used as a base class for more complex
|
|
classes. In <tt>Target.td</tt>, the <tt>Register</tt> class is the base for the
|
|
<tt>RegisterWithSubRegs</tt> class that is used to define registers that need to
|
|
specify subregisters in the <tt>SubRegs</tt> list, as shown here:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
class RegisterWithSubRegs<string n,
|
|
list<Register> subregs> : Register<n> {
|
|
let SubRegs = subregs;
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
In <tt>SparcRegisterInfo.td</tt>, additional register classes are defined for
|
|
SPARC: a Register subclass, SparcReg, and further subclasses: <tt>Ri</tt>,
|
|
<tt>Rf</tt>, and <tt>Rd</tt>. SPARC registers are identified by 5-bit ID
|
|
numbers, which is a feature common to these subclasses. Note the use of
|
|
'<tt>let</tt>' expressions to override values that are initially defined in a
|
|
superclass (such as <tt>SubRegs</tt> field in the <tt>Rd</tt> class).
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
class SparcReg<string n> : Register<n> {
|
|
field bits<5> Num;
|
|
let Namespace = "SP";
|
|
}
|
|
// Ri - 32-bit integer registers
|
|
class Ri<bits<5> num, string n> :
|
|
SparcReg<n> {
|
|
let Num = num;
|
|
}
|
|
// Rf - 32-bit floating-point registers
|
|
class Rf<bits<5> num, string n> :
|
|
SparcReg<n> {
|
|
let Num = num;
|
|
}
|
|
// Rd - Slots in the FP register file for 64-bit
|
|
floating-point values.
|
|
class Rd<bits<5> num, string n,
|
|
list<Register> subregs> : SparcReg<n> {
|
|
let Num = num;
|
|
let SubRegs = subregs;
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
In the <tt>SparcRegisterInfo.td</tt> file, there are register definitions that
|
|
utilize these subclasses of <tt>Register</tt>, such as:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
def G0 : Ri< 0, "G0">,
|
|
DwarfRegNum<[0]>;
|
|
def G1 : Ri< 1, "G1">, DwarfRegNum<[1]>;
|
|
...
|
|
def F0 : Rf< 0, "F0">,
|
|
DwarfRegNum<[32]>;
|
|
def F1 : Rf< 1, "F1">,
|
|
DwarfRegNum<[33]>;
|
|
...
|
|
def D0 : Rd< 0, "F0", [F0, F1]>,
|
|
DwarfRegNum<[32]>;
|
|
def D1 : Rd< 2, "F2", [F2, F3]>,
|
|
DwarfRegNum<[34]>;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
The last two registers shown above (<tt>D0</tt> and <tt>D1</tt>) 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.
|
|
</p>
|
|
|
|
</div>
|
|
|
|
<!-- ======================================================================= -->
|
|
<h3>
|
|
<a name="RegisterClassDef">Defining a Register Class</a>
|
|
</h3>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
The <tt>RegisterClass</tt> class (specified in <tt>Target.td</tt>) 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
|
|
<tt>XXXRegisterInfo.td</tt> that uses <tt>Target.td</tt> can construct register
|
|
classes using the following class:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
class RegisterClass<string namespace,
|
|
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;
|
|
dag MemberList = regList;
|
|
|
|
// for register classes that are subregisters of this class
|
|
list<RegisterClass> SubRegClassList = [];
|
|
|
|
code MethodProtos = [{}]; // to insert arbitrary code
|
|
code MethodBodies = [{}];
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>To define a RegisterClass, use the following 4 arguments:</p>
|
|
|
|
<ul>
|
|
<li>The first argument of the definition is the name of the namespace.</li>
|
|
|
|
<li>The second argument is a list of <tt>ValueType</tt> register type values
|
|
that are defined in <tt>include/llvm/CodeGen/ValueTypes.td</tt>. Defined
|
|
values include integer types (such as <tt>i16</tt>, <tt>i32</tt>,
|
|
and <tt>i1</tt> for Boolean), floating-point types
|
|
(<tt>f32</tt>, <tt>f64</tt>), and vector types (for example, <tt>v8i16</tt>
|
|
for an <tt>8 x i16</tt> vector). All registers in a <tt>RegisterClass</tt>
|
|
must have the same <tt>ValueType</tt>, 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. </li>
|
|
|
|
<li>The third argument of the <tt>RegisterClass</tt> definition specifies the
|
|
alignment required of the registers when they are stored or loaded to
|
|
memory.</li>
|
|
|
|
<li>The final argument, <tt>regList</tt>, specifies which registers are in this
|
|
class. If an alternative allocation order method is not specified, then
|
|
<tt>regList</tt> also defines the order of allocation used by the register
|
|
allocator. Besides simply listing registers with <tt>(add R0, R1, ...)</tt>,
|
|
more advanced set operators are available. See
|
|
<tt>include/llvm/Target/Target.td</tt> for more information.</li>
|
|
</ul>
|
|
|
|
<p>
|
|
In <tt>SparcRegisterInfo.td</tt>, three RegisterClass objects are defined:
|
|
<tt>FPRegs</tt>, <tt>DFPRegs</tt>, and <tt>IntRegs</tt>. For all three register
|
|
classes, the first argument defines the namespace with the string
|
|
'<tt>SP</tt>'. <tt>FPRegs</tt> defines a group of 32 single-precision
|
|
floating-point registers (<tt>F0</tt> to <tt>F31</tt>); <tt>DFPRegs</tt> defines
|
|
a group of 16 double-precision registers
|
|
(<tt>D0-D15</tt>).
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
// 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,
|
|
(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
|
|
)>;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
Using <tt>SparcRegisterInfo.td</tt> with TableGen generates several output files
|
|
that are intended for inclusion in other source code that you write.
|
|
<tt>SparcRegisterInfo.td</tt> generates <tt>SparcGenRegisterInfo.h.inc</tt>,
|
|
which should be included in the header file for the implementation of the SPARC
|
|
register implementation that you write (<tt>SparcRegisterInfo.h</tt>). In
|
|
<tt>SparcGenRegisterInfo.h.inc</tt> a new structure is defined called
|
|
<tt>SparcGenRegisterInfo</tt> that uses <tt>TargetRegisterInfo</tt> as its
|
|
base. It also specifies types, based upon the defined register
|
|
classes: <tt>DFPRegsClass</tt>, <tt>FPRegsClass</tt>, and <tt>IntRegsClass</tt>.
|
|
</p>
|
|
|
|
<p>
|
|
<tt>SparcRegisterInfo.td</tt> also generates <tt>SparcGenRegisterInfo.inc</tt>,
|
|
which is included at the bottom of <tt>SparcRegisterInfo.cpp</tt>, the SPARC
|
|
register implementation. The code below shows only the generated integer
|
|
registers and associated register classes. The order of registers
|
|
in <tt>IntRegs</tt> reflects the order in the definition of <tt>IntRegs</tt> in
|
|
the target description file.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre> // 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,
|
|
};
|
|
|
|
// 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...
|
|
static const TargetRegisterClass* const IntRegsSubRegClasses [] = {
|
|
NULL
|
|
};
|
|
...
|
|
// IntRegs Super-register Classess...
|
|
static const TargetRegisterClass* const IntRegsSuperRegClasses [] = {
|
|
NULL
|
|
};
|
|
...
|
|
// IntRegs Register Class sub-classes...
|
|
static const TargetRegisterClass* const IntRegsSubclasses [] = {
|
|
NULL
|
|
};
|
|
...
|
|
// IntRegs Register Class super-classes...
|
|
static const TargetRegisterClass* const IntRegsSuperclasses [] = {
|
|
NULL
|
|
};
|
|
|
|
IntRegsClass::IntRegsClass() : TargetRegisterClass(IntRegsRegClassID,
|
|
IntRegsVTs, IntRegsSubclasses, IntRegsSuperclasses, IntRegsSubRegClasses,
|
|
IntRegsSuperRegClasses, 4, 4, 1, IntRegs, IntRegs + 32) {}
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
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.
|
|
</p>
|
|
|
|
</div>
|
|
|
|
<!-- ======================================================================= -->
|
|
<h3>
|
|
<a name="implementRegister">Implement a subclass of</a>
|
|
<a href="CodeGenerator.html#targetregisterinfo">TargetRegisterInfo</a>
|
|
</h3>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
The final step is to hand code portions of <tt>XXXRegisterInfo</tt>, which
|
|
implements the interface described in <tt>TargetRegisterInfo.h</tt>. These
|
|
functions return <tt>0</tt>, <tt>NULL</tt>, or <tt>false</tt>, unless
|
|
overridden. Here is a list of functions that are overridden for the SPARC
|
|
implementation in <tt>SparcRegisterInfo.cpp</tt>:
|
|
</p>
|
|
|
|
<ul>
|
|
<li><tt>getCalleeSavedRegs</tt> — Returns a list of callee-saved registers
|
|
in the order of the desired callee-save stack frame offset.</li>
|
|
|
|
<li><tt>getReservedRegs</tt> — Returns a bitset indexed by physical
|
|
register numbers, indicating if a particular register is unavailable.</li>
|
|
|
|
<li><tt>hasFP</tt> — Return a Boolean indicating if a function should have
|
|
a dedicated frame pointer register.</li>
|
|
|
|
<li><tt>eliminateCallFramePseudoInstr</tt> — If call frame setup or
|
|
destroy pseudo instructions are used, this can be called to eliminate
|
|
them.</li>
|
|
|
|
<li><tt>eliminateFrameIndex</tt> — Eliminate abstract frame indices from
|
|
instructions that may use them.</li>
|
|
|
|
<li><tt>emitPrologue</tt> — Insert prologue code into the function.</li>
|
|
|
|
<li><tt>emitEpilogue</tt> — Insert epilogue code into the function.</li>
|
|
</ul>
|
|
|
|
</div>
|
|
|
|
</div>
|
|
|
|
<!-- *********************************************************************** -->
|
|
<h2>
|
|
<a name="InstructionSet">Instruction Set</a>
|
|
</h2>
|
|
|
|
<!-- *********************************************************************** -->
|
|
<div>
|
|
|
|
<p>
|
|
During the early stages of code generation, the LLVM IR code is converted to a
|
|
<tt>SelectionDAG</tt> with nodes that are instances of the <tt>SDNode</tt> class
|
|
containing target instructions. An <tt>SDNode</tt> 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 <tt>include/llvm/CodeGen/SelectionDAGNodes.h</tt>
|
|
file (values of the <tt>NodeType</tt> enum in the <tt>ISD</tt> namespace).
|
|
</p>
|
|
|
|
<p>
|
|
TableGen uses the following target description (<tt>.td</tt>) input files to
|
|
generate much of the code for instruction definition:
|
|
</p>
|
|
|
|
<ul>
|
|
<li><tt>Target.td</tt> — Where the <tt>Instruction</tt>, <tt>Operand</tt>,
|
|
<tt>InstrInfo</tt>, and other fundamental classes are defined.</li>
|
|
|
|
<li><tt>TargetSelectionDAG.td</tt>— Used by <tt>SelectionDAG</tt>
|
|
instruction selection generators, contains <tt>SDTC*</tt> classes (selection
|
|
DAG type constraint), definitions of <tt>SelectionDAG</tt> nodes (such as
|
|
<tt>imm</tt>, <tt>cond</tt>, <tt>bb</tt>, <tt>add</tt>, <tt>fadd</tt>,
|
|
<tt>sub</tt>), and pattern support (<tt>Pattern</tt>, <tt>Pat</tt>,
|
|
<tt>PatFrag</tt>, <tt>PatLeaf</tt>, <tt>ComplexPattern</tt>.</li>
|
|
|
|
<li><tt>XXXInstrFormats.td</tt> — Patterns for definitions of
|
|
target-specific instructions.</li>
|
|
|
|
<li><tt>XXXInstrInfo.td</tt> — 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 <tt>X86InstrSSE.td</tt>, and
|
|
for Pentium with MMX, this file is <tt>X86InstrMMX.td</tt>.</li>
|
|
</ul>
|
|
|
|
<p>
|
|
There is also a target-specific <tt>XXX.td</tt> file, where <tt>XXX</tt> is the
|
|
name of the target. The <tt>XXX.td</tt> file includes the other <tt>.td</tt>
|
|
input files, but its contents are only directly important for subtargets.
|
|
</p>
|
|
|
|
<p>
|
|
You should describe a concrete target-specific class <tt>XXXInstrInfo</tt> that
|
|
represents machine instructions supported by a target machine.
|
|
<tt>XXXInstrInfo</tt> contains an array of <tt>XXXInstrDescriptor</tt> objects,
|
|
each of which describes one instruction. An instruction descriptor defines:</p>
|
|
|
|
<ul>
|
|
<li>Opcode mnemonic</li>
|
|
|
|
<li>Number of operands</li>
|
|
|
|
<li>List of implicit register definitions and uses</li>
|
|
|
|
<li>Target-independent properties (such as memory access, is commutable)</li>
|
|
|
|
<li>Target-specific flags </li>
|
|
</ul>
|
|
|
|
<p>
|
|
The Instruction class (defined in <tt>Target.td</tt>) is mostly used as a base
|
|
for more complex instruction classes.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>class Instruction {
|
|
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.
|
|
list<dag> Pattern; // Set to the DAG pattern for this instruction
|
|
list<Register> Uses = [];
|
|
list<Register> Defs = [];
|
|
list<Predicate> Predicates = []; // predicates turned into isel match code
|
|
... remainder not shown for space ...
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
A <tt>SelectionDAG</tt> node (<tt>SDNode</tt>) should contain an object
|
|
representing a target-specific instruction that is defined
|
|
in <tt>XXXInstrInfo.td</tt>. The instruction objects should represent
|
|
instructions from the architecture manual of the target machine (such as the
|
|
SPARC Architecture Manual for the SPARC target).
|
|
</p>
|
|
|
|
<p>
|
|
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 <tt>ADDri</tt> and
|
|
<tt>ADDrr</tt>.
|
|
</p>
|
|
|
|
<p>
|
|
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.
|
|
</p>
|
|
|
|
<p>
|
|
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 <tt>CALL</tt>
|
|
instruction. Format 2 is for branch on condition codes and <tt>SETHI</tt> (set
|
|
high bits of a register) instructions. Format 3 is for other instructions.
|
|
</p>
|
|
|
|
<p>
|
|
Each of these formats has corresponding classes in <tt>SparcInstrFormat.td</tt>.
|
|
<tt>InstSP</tt> is a base class for other instruction classes. Additional base
|
|
classes are specified for more precise formats: for example
|
|
in <tt>SparcInstrFormat.td</tt>, <tt>F2_1</tt> is for <tt>SETHI</tt>,
|
|
and <tt>F2_2</tt> is for branches. There are three other base
|
|
classes: <tt>F3_1</tt> for register/register operations, <tt>F3_2</tt> for
|
|
register/immediate operations, and <tt>F3_3</tt> for floating-point
|
|
operations. <tt>SparcInstrInfo.td</tt> also adds the base class Pseudo for
|
|
synthetic SPARC instructions.
|
|
</p>
|
|
|
|
<p>
|
|
<tt>SparcInstrInfo.td</tt> largely consists of operand and instruction
|
|
definitions for the SPARC target. In <tt>SparcInstrInfo.td</tt>, the following
|
|
target description file entry, <tt>LDrr</tt>, defines the Load Integer
|
|
instruction for a Word (the <tt>LD</tt> SPARC opcode) from a memory address to a
|
|
register. The first parameter, the value 3 (<tt>11<sub>2</sub></tt>), is the
|
|
operation value for this category of operation. The second parameter
|
|
(<tt>000000<sub>2</sub></tt>) is the specific operation value
|
|
for <tt>LD</tt>/Load Word. The third parameter is the output destination, which
|
|
is a register operand and defined in the <tt>Register</tt> target description
|
|
file (<tt>IntRegs</tt>).
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>def LDrr : F3_1 <3, 0b000000, (outs IntRegs:$dst), (ins MEMrr:$addr),
|
|
"ld [$addr], $dst",
|
|
[(set IntRegs:$dst, (load ADDRrr:$addr))]>;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
The fourth parameter is the input source, which uses the address
|
|
operand <tt>MEMrr</tt> that is defined earlier in <tt>SparcInstrInfo.td</tt>:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>def MEMrr : Operand<i32> {
|
|
let PrintMethod = "printMemOperand";
|
|
let MIOperandInfo = (ops IntRegs, IntRegs);
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
The fifth parameter is a string that is used by the assembly printer and can be
|
|
left as an empty string until the assembly printer interface is implemented. The
|
|
sixth and final parameter is the pattern used to match the instruction during
|
|
the SelectionDAG Select Phase described in
|
|
(<a href="CodeGenerator.html">The LLVM
|
|
Target-Independent Code Generator</a>). This parameter is detailed in the next
|
|
section, <a href="#InstructionSelector">Instruction Selector</a>.
|
|
</p>
|
|
|
|
<p>
|
|
Instruction class definitions are not overloaded for different operand types, so
|
|
separate versions of instructions are needed for register, memory, or immediate
|
|
value operands. For example, to perform a Load Integer instruction for a Word
|
|
from an immediate operand to a register, the following instruction class is
|
|
defined:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>def LDri : F3_2 <3, 0b000000, (outs IntRegs:$dst), (ins MEMri:$addr),
|
|
"ld [$addr], $dst",
|
|
[(set IntRegs:$dst, (load ADDRri:$addr))]>;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
Writing these definitions for so many similar instructions can involve a lot of
|
|
cut and paste. In td files, the <tt>multiclass</tt> directive enables the
|
|
creation of templates to define several instruction classes at once (using
|
|
the <tt>defm</tt> directive). For example in <tt>SparcInstrInfo.td</tt>, the
|
|
<tt>multiclass</tt> pattern <tt>F3_12</tt> is defined to create 2 instruction
|
|
classes each time <tt>F3_12</tt> is invoked:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>multiclass F3_12 <string OpcStr, bits<6> Op3Val, SDNode OpNode> {
|
|
def rr : F3_1 <2, Op3Val,
|
|
(outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
|
|
!strconcat(OpcStr, " $b, $c, $dst"),
|
|
[(set IntRegs:$dst, (OpNode IntRegs:$b, IntRegs:$c))]>;
|
|
def ri : F3_2 <2, Op3Val,
|
|
(outs IntRegs:$dst), (ins IntRegs:$b, i32imm:$c),
|
|
!strconcat(OpcStr, " $b, $c, $dst"),
|
|
[(set IntRegs:$dst, (OpNode IntRegs:$b, simm13:$c))]>;
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
So when the <tt>defm</tt> directive is used for the <tt>XOR</tt>
|
|
and <tt>ADD</tt> instructions, as seen below, it creates four instruction
|
|
objects: <tt>XORrr</tt>, <tt>XORri</tt>, <tt>ADDrr</tt>, and <tt>ADDri</tt>.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
defm XOR : F3_12<"xor", 0b000011, xor>;
|
|
defm ADD : F3_12<"add", 0b000000, add>;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
<tt>SparcInstrInfo.td</tt> also includes definitions for condition codes that
|
|
are referenced by branch instructions. The following definitions
|
|
in <tt>SparcInstrInfo.td</tt> indicate the bit location of the SPARC condition
|
|
code. For example, the 10<sup>th</sup> bit represents the 'greater than'
|
|
condition for integers, and the 22<sup>nd</sup> bit represents the 'greater
|
|
than' condition for floats.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
def ICC_NE : ICC_VAL< 9>; // Not Equal
|
|
def ICC_E : ICC_VAL< 1>; // Equal
|
|
def ICC_G : ICC_VAL<10>; // Greater
|
|
...
|
|
def FCC_U : FCC_VAL<23>; // Unordered
|
|
def FCC_G : FCC_VAL<22>; // Greater
|
|
def FCC_UG : FCC_VAL<21>; // Unordered or Greater
|
|
...
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
(Note that <tt>Sparc.h</tt> also defines enums that correspond to the same SPARC
|
|
condition codes. Care must be taken to ensure the values in <tt>Sparc.h</tt>
|
|
correspond to the values in <tt>SparcInstrInfo.td</tt>. I.e.,
|
|
<tt>SPCC::ICC_NE = 9</tt>, <tt>SPCC::FCC_U = 23</tt> and so on.)
|
|
</p>
|
|
|
|
<!-- ======================================================================= -->
|
|
<h3>
|
|
<a name="operandMapping">Instruction Operand Mapping</a>
|
|
</h3>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
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 <tt>XNORrr</tt> instruction as
|
|
a <tt>F3_1</tt> format instruction having three operands.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
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)))]>;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
The instruction templates in <tt>SparcInstrFormats.td</tt> show the base class
|
|
for <tt>F3_1</tt> is <tt>InstSP</tt>.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
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;
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p><tt>InstSP</tt> leaves the <tt>op</tt> field unbound.</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
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;
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
<tt>F3</tt> binds the <tt>op</tt> field and defines the <tt>rd</tt>,
|
|
<tt>op3</tt>, and <tt>rs1</tt> fields. <tt>F3</tt> format instructions will
|
|
bind the operands <tt>rd</tt>, <tt>op3</tt>, and <tt>rs1</tt> fields.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
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;
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
<tt>F3_1</tt> binds the <tt>op3</tt> field and defines the <tt>rs2</tt>
|
|
fields. <tt>F3_1</tt> format instructions will bind the operands to the <tt>rd</tt>,
|
|
<tt>rs1</tt>, and <tt>rs2</tt> fields. This results in the <tt>XNORrr</tt>
|
|
instruction binding <tt>$dst</tt>, <tt>$b</tt>, and <tt>$c</tt> operands to
|
|
the <tt>rd</tt>, <tt>rs1</tt>, and <tt>rs2</tt> fields respectively.
|
|
</p>
|
|
|
|
</div>
|
|
|
|
<!-- ======================================================================= -->
|
|
<h3>
|
|
<a name="relationMapping">Instruction Relation Mapping</a>
|
|
</h3>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
This TableGen feature is used to relate instructions with each other. It is
|
|
particularly useful when you have multiple instruction formats and need to
|
|
switch between them after instruction selection. This entire feature is driven
|
|
by relation models which can be defined in <tt>XXXInstrInfo.td</tt> files
|
|
according to the target-specific instruction set. Relation models are defined
|
|
using <tt>InstrMapping</tt> class as a base. TableGen parses all the models
|
|
and generates instruction relation maps using the specified information.
|
|
Relation maps are emitted as tables in the <tt>XXXGenInstrInfo.inc</tt> file
|
|
along with the functions to query them. For the detailed information on how to
|
|
use this feature, please refer to
|
|
<a href="HowToUseInstrMappings.html">How to add Instruction Mappings</a>
|
|
document.
|
|
</p>
|
|
</div>
|
|
|
|
<!-- ======================================================================= -->
|
|
<h3>
|
|
<a name="implementInstr">Implement a subclass of </a>
|
|
<a href="CodeGenerator.html#targetinstrinfo">TargetInstrInfo</a>
|
|
</h3>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
The final step is to hand code portions of <tt>XXXInstrInfo</tt>, which
|
|
implements the interface described in <tt>TargetInstrInfo.h</tt>. These
|
|
functions return <tt>0</tt> or a Boolean or they assert, unless
|
|
overridden. Here's a list of functions that are overridden for the SPARC
|
|
implementation in <tt>SparcInstrInfo.cpp</tt>:
|
|
</p>
|
|
|
|
<ul>
|
|
<li><tt>isLoadFromStackSlot</tt> — If the specified machine instruction is
|
|
a direct load from a stack slot, return the register number of the
|
|
destination and the <tt>FrameIndex</tt> of the stack slot.</li>
|
|
|
|
<li><tt>isStoreToStackSlot</tt> — If the specified machine instruction is
|
|
a direct store to a stack slot, return the register number of the
|
|
destination and the <tt>FrameIndex</tt> of the stack slot.</li>
|
|
|
|
<li><tt>copyPhysReg</tt> — Copy values between a pair of physical
|
|
registers.</li>
|
|
|
|
<li><tt>storeRegToStackSlot</tt> — Store a register value to a stack
|
|
slot.</li>
|
|
|
|
<li><tt>loadRegFromStackSlot</tt> — Load a register value from a stack
|
|
slot.</li>
|
|
|
|
<li><tt>storeRegToAddr</tt> — Store a register value to memory.</li>
|
|
|
|
<li><tt>loadRegFromAddr</tt> — Load a register value from memory.</li>
|
|
|
|
<li><tt>foldMemoryOperand</tt> — Attempt to combine instructions of any
|
|
load or store instruction for the specified operand(s).</li>
|
|
</ul>
|
|
|
|
</div>
|
|
|
|
<!-- ======================================================================= -->
|
|
<h3>
|
|
<a name="branchFolding">Branch Folding and If Conversion</a>
|
|
</h3>
|
|
<div>
|
|
|
|
<p>
|
|
Performance can be improved by combining instructions or by eliminating
|
|
instructions that are never reached. The <tt>AnalyzeBranch</tt> method
|
|
in <tt>XXXInstrInfo</tt> may be implemented to examine conditional instructions
|
|
and remove unnecessary instructions. <tt>AnalyzeBranch</tt> looks at the end of
|
|
a machine basic block (MBB) for opportunities for improvement, such as branch
|
|
folding and if conversion. The <tt>BranchFolder</tt> and <tt>IfConverter</tt>
|
|
machine function passes (see the source files <tt>BranchFolding.cpp</tt> and
|
|
<tt>IfConversion.cpp</tt> in the <tt>lib/CodeGen</tt> directory) call
|
|
<tt>AnalyzeBranch</tt> to improve the control flow graph that represents the
|
|
instructions.
|
|
</p>
|
|
|
|
<p>
|
|
Several implementations of <tt>AnalyzeBranch</tt> (for ARM, Alpha, and X86) can
|
|
be examined as models for your own <tt>AnalyzeBranch</tt> implementation. Since
|
|
SPARC does not implement a useful <tt>AnalyzeBranch</tt>, the ARM target
|
|
implementation is shown below.
|
|
</p>
|
|
|
|
<p><tt>AnalyzeBranch</tt> returns a Boolean value and takes four parameters:</p>
|
|
|
|
<ul>
|
|
<li><tt>MachineBasicBlock &MBB</tt> — The incoming block to be
|
|
examined.</li>
|
|
|
|
<li><tt>MachineBasicBlock *&TBB</tt> — A destination block that is
|
|
returned. For a conditional branch that evaluates to true, <tt>TBB</tt> is
|
|
the destination.</li>
|
|
|
|
<li><tt>MachineBasicBlock *&FBB</tt> — For a conditional branch that
|
|
evaluates to false, <tt>FBB</tt> is returned as the destination.</li>
|
|
|
|
<li><tt>std::vector<MachineOperand> &Cond</tt> — List of
|
|
operands to evaluate a condition for a conditional branch.</li>
|
|
</ul>
|
|
|
|
<p>
|
|
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 <tt>TBB</tt>
|
|
or <tt>FBB</tt>, so both parameters return <tt>NULL</tt>. The start of
|
|
the <tt>AnalyzeBranch</tt> (see code below for the ARM target) shows the
|
|
function parameters and the code for the simplest case.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>bool ARMInstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,
|
|
MachineBasicBlock *&TBB, MachineBasicBlock *&FBB,
|
|
std::vector<MachineOperand> &Cond) const
|
|
{
|
|
MachineBasicBlock::iterator I = MBB.end();
|
|
if (I == MBB.begin() || !isUnpredicatedTerminator(--I))
|
|
return false;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
If a block ends with a single unconditional branch instruction, then
|
|
<tt>AnalyzeBranch</tt> (shown below) should return the destination of that
|
|
branch in the <tt>TBB</tt> parameter.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
if (LastOpc == ARM::B || LastOpc == ARM::tB) {
|
|
TBB = LastInst->getOperand(0).getMBB();
|
|
return false;
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
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 <tt>TBB</tt> parameter.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
if ((SecondLastOpc == ARM::B || SecondLastOpc==ARM::tB) &&
|
|
(LastOpc == ARM::B || LastOpc == ARM::tB)) {
|
|
TBB = SecondLastInst->getOperand(0).getMBB();
|
|
I = LastInst;
|
|
I->eraseFromParent();
|
|
return false;
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
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,
|
|
<tt>AnalyzeBranch</tt> (shown below) should return the destination of that
|
|
conditional branch in the <tt>TBB</tt> parameter and a list of operands in
|
|
the <tt>Cond</tt> parameter to evaluate the condition.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
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;
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
If a block ends with both a conditional branch and an ensuing unconditional
|
|
branch, then <tt>AnalyzeBranch</tt> (shown below) should return the conditional
|
|
branch destination (assuming it corresponds to a conditional evaluation of
|
|
'<tt>true</tt>') in the <tt>TBB</tt> parameter and the unconditional branch
|
|
destination in the <tt>FBB</tt> (corresponding to a conditional evaluation of
|
|
'<tt>false</tt>'). A list of operands to evaluate the condition should be
|
|
returned in the <tt>Cond</tt> parameter.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
unsigned SecondLastOpc = SecondLastInst->getOpcode();
|
|
|
|
if ((SecondLastOpc == ARM::Bcc && LastOpc == ARM::B) ||
|
|
(SecondLastOpc == ARM::tBcc && LastOpc == ARM::tB)) {
|
|
TBB = SecondLastInst->getOperand(0).getMBB();
|
|
Cond.push_back(SecondLastInst->getOperand(1));
|
|
Cond.push_back(SecondLastInst->getOperand(2));
|
|
FBB = LastInst->getOperand(0).getMBB();
|
|
return false;
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
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 <tt>Cond</tt> parameter can be passed to methods of other instructions to
|
|
create new branches or perform other operations. An implementation
|
|
of <tt>AnalyzeBranch</tt> requires the helper methods <tt>RemoveBranch</tt>
|
|
and <tt>InsertBranch</tt> to manage subsequent operations.
|
|
</p>
|
|
|
|
<p>
|
|
<tt>AnalyzeBranch</tt> should return false indicating success in most circumstances.
|
|
<tt>AnalyzeBranch</tt> should only return true when the method is stumped about what to
|
|
do, for example, if a block has three terminating branches. <tt>AnalyzeBranch</tt> may
|
|
return true if it encounters a terminator it cannot handle, such as an indirect
|
|
branch.
|
|
</p>
|
|
|
|
</div>
|
|
|
|
</div>
|
|
|
|
<!-- *********************************************************************** -->
|
|
<h2>
|
|
<a name="InstructionSelector">Instruction Selector</a>
|
|
</h2>
|
|
<!-- *********************************************************************** -->
|
|
|
|
<div>
|
|
|
|
<p>
|
|
LLVM uses a <tt>SelectionDAG</tt> to represent LLVM IR instructions, and nodes
|
|
of the <tt>SelectionDAG</tt> 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 <tt>XXXISelDAGToDAG.cpp</tt> is used to
|
|
match patterns and perform DAG-to-DAG instruction selection. Optionally, a pass
|
|
may be defined (in <tt>XXXBranchSelector.cpp</tt>) to perform similar DAG-to-DAG
|
|
operations for branch instructions. Later, the code in
|
|
<tt>XXXISelLowering.cpp</tt> replaces or removes operations and data types not
|
|
supported natively (legalizes) in a <tt>SelectionDAG</tt>.
|
|
</p>
|
|
|
|
<p>
|
|
TableGen generates code for instruction selection using the following target
|
|
description input files:
|
|
</p>
|
|
|
|
<ul>
|
|
<li><tt>XXXInstrInfo.td</tt> — Contains definitions of instructions in a
|
|
target-specific instruction set, generates <tt>XXXGenDAGISel.inc</tt>, which
|
|
is included in <tt>XXXISelDAGToDAG.cpp</tt>.</li>
|
|
|
|
<li><tt>XXXCallingConv.td</tt> — Contains the calling and return value
|
|
conventions for the target architecture, and it generates
|
|
<tt>XXXGenCallingConv.inc</tt>, which is included in
|
|
<tt>XXXISelLowering.cpp</tt>.</li>
|
|
</ul>
|
|
|
|
<p>
|
|
The implementation of an instruction selection pass must include a header that
|
|
declares the <tt>FunctionPass</tt> class or a subclass of <tt>FunctionPass</tt>. In
|
|
<tt>XXXTargetMachine.cpp</tt>, a Pass Manager (PM) should add each instruction
|
|
selection pass into the queue of passes to run.
|
|
</p>
|
|
|
|
<p>
|
|
The LLVM static compiler (<tt>llc</tt>) is an excellent tool for visualizing the
|
|
contents of DAGs. To display the <tt>SelectionDAG</tt> before or after specific
|
|
processing phases, use the command line options for <tt>llc</tt>, described
|
|
at <a href="CodeGenerator.html#selectiondag_process">
|
|
SelectionDAG Instruction Selection Process</a>.
|
|
</p>
|
|
|
|
<p>
|
|
To describe instruction selector behavior, you should add patterns for lowering
|
|
LLVM code into a <tt>SelectionDAG</tt> as the last parameter of the instruction
|
|
definitions in <tt>XXXInstrInfo.td</tt>. For example, in
|
|
<tt>SparcInstrInfo.td</tt>, this entry defines a register store operation, and
|
|
the last parameter describes a pattern with the store DAG operator.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
def STrr : F3_1< 3, 0b000100, (outs), (ins MEMrr:$addr, IntRegs:$src),
|
|
"st $src, [$addr]", [(store IntRegs:$src, ADDRrr:$addr)]>;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
<tt>ADDRrr</tt> is a memory mode that is also defined in
|
|
<tt>SparcInstrInfo.td</tt>:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
def ADDRrr : ComplexPattern<i32, 2, "SelectADDRrr", [], []>;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
The definition of <tt>ADDRrr</tt> refers to <tt>SelectADDRrr</tt>, which is a
|
|
function defined in an implementation of the Instructor Selector (such
|
|
as <tt>SparcISelDAGToDAG.cpp</tt>).
|
|
</p>
|
|
|
|
<p>
|
|
In <tt>lib/Target/TargetSelectionDAG.td</tt>, the DAG operator for store is
|
|
defined below:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
def store : PatFrag<(ops node:$val, node:$ptr),
|
|
(st node:$val, node:$ptr), [{
|
|
if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N))
|
|
return !ST->isTruncatingStore() &&
|
|
ST->getAddressingMode() == ISD::UNINDEXED;
|
|
return false;
|
|
}]>;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
<tt>XXXInstrInfo.td</tt> also generates (in <tt>XXXGenDAGISel.inc</tt>) the
|
|
<tt>SelectCode</tt> method that is used to call the appropriate processing
|
|
method for an instruction. In this example, <tt>SelectCode</tt>
|
|
calls <tt>Select_ISD_STORE</tt> for the <tt>ISD::STORE</tt> opcode.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
SDNode *SelectCode(SDValue N) {
|
|
...
|
|
MVT::ValueType NVT = N.getNode()->getValueType(0);
|
|
switch (N.getOpcode()) {
|
|
case ISD::STORE: {
|
|
switch (NVT) {
|
|
default:
|
|
return Select_ISD_STORE(N);
|
|
break;
|
|
}
|
|
break;
|
|
}
|
|
...
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
The pattern for <tt>STrr</tt> is matched, so elsewhere in
|
|
<tt>XXXGenDAGISel.inc</tt>, code for <tt>STrr</tt> is created for
|
|
<tt>Select_ISD_STORE</tt>. The <tt>Emit_22</tt> method is also generated
|
|
in <tt>XXXGenDAGISel.inc</tt> to complete the processing of this
|
|
instruction.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
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.getNode()->getValueType(0) == MVT::i32 &&
|
|
N2.getNode()->getValueType(0) == MVT::i32) {
|
|
return Emit_22(N, SP::STrr, CPTmp0, CPTmp1);
|
|
}
|
|
...
|
|
</pre>
|
|
</div>
|
|
|
|
<!-- ======================================================================= -->
|
|
<h3>
|
|
<a name="LegalizePhase">The SelectionDAG Legalize Phase</a>
|
|
</h3>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
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.
|
|
</p>
|
|
|
|
<p>
|
|
In the constructor for the <tt>XXXTargetLowering</tt> class, first use the
|
|
<tt>addRegisterClass</tt> 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 <tt>XXXRegisterInfo.td</tt> and placed
|
|
in <tt>XXXGenRegisterInfo.h.inc</tt>. For example, the implementation of the
|
|
constructor for the SparcTargetLowering class (in
|
|
<tt>SparcISelLowering.cpp</tt>) starts with the following code:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
addRegisterClass(MVT::i32, SP::IntRegsRegisterClass);
|
|
addRegisterClass(MVT::f32, SP::FPRegsRegisterClass);
|
|
addRegisterClass(MVT::f64, SP::DFPRegsRegisterClass);
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
You should examine the node types in the <tt>ISD</tt> namespace
|
|
(<tt>include/llvm/CodeGen/SelectionDAGNodes.h</tt>) and determine which
|
|
operations the target natively supports. For operations that do <b>not</b> 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 <tt>llvm/Target/TargetLowering.h</tt>) are:
|
|
</p>
|
|
|
|
<ul>
|
|
<li><tt>setOperationAction</tt> — General operation.</li>
|
|
|
|
<li><tt>setLoadExtAction</tt> — Load with extension.</li>
|
|
|
|
<li><tt>setTruncStoreAction</tt> — Truncating store.</li>
|
|
|
|
<li><tt>setIndexedLoadAction</tt> — Indexed load.</li>
|
|
|
|
<li><tt>setIndexedStoreAction</tt> — Indexed store.</li>
|
|
|
|
<li><tt>setConvertAction</tt> — Type conversion.</li>
|
|
|
|
<li><tt>setCondCodeAction</tt> — Support for a given condition code.</li>
|
|
</ul>
|
|
|
|
<p>
|
|
Note: on older releases, <tt>setLoadXAction</tt> is used instead
|
|
of <tt>setLoadExtAction</tt>. Also, on older releases,
|
|
<tt>setCondCodeAction</tt> may not be supported. Examine your release
|
|
to see what methods are specifically supported.
|
|
</p>
|
|
|
|
<p>
|
|
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 <tt>LegalAction</tt> type enum value: <tt>Promote</tt>, <tt>Expand</tt>,
|
|
<tt>Custom</tt>, or <tt>Legal</tt>. <tt>SparcISelLowering.cpp</tt>
|
|
contains examples of all four <tt>LegalAction</tt> values.
|
|
</p>
|
|
|
|
<!-- _______________________________________________________________________ -->
|
|
<h4>
|
|
<a name="promote">Promote</a>
|
|
</h4>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
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 (<tt>i1</tt> type), so
|
|
in <tt>SparcISelLowering.cpp</tt> the third parameter below, <tt>Promote</tt>,
|
|
changes <tt>i1</tt> type values to a large type before loading.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
|
|
</pre>
|
|
</div>
|
|
|
|
</div>
|
|
|
|
<!-- _______________________________________________________________________ -->
|
|
<h4>
|
|
<a name="expand">Expand</a>
|
|
</h4>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
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, <tt>Expand</tt>, to
|
|
<tt>setOperationAction</tt>:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
setOperationAction(ISD::FSIN, MVT::f32, Expand);
|
|
setOperationAction(ISD::FCOS, MVT::f32, Expand);
|
|
</pre>
|
|
</div>
|
|
|
|
</div>
|
|
|
|
<!-- _______________________________________________________________________ -->
|
|
<h4>
|
|
<a name="custom">Custom</a>
|
|
</h4>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
For some operations, simple type promotion or operation expansion may be
|
|
insufficient. In some cases, a special intrinsic function must be implemented.
|
|
</p>
|
|
|
|
<p>
|
|
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.
|
|
</p>
|
|
|
|
<p>
|
|
As seen in <tt>SparcISelLowering.cpp</tt> code below, to perform a type
|
|
conversion from a floating point value to a signed integer, first the
|
|
<tt>setOperationAction</tt> should be called with <tt>Custom</tt> as the third
|
|
parameter:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
In the <tt>LowerOperation</tt> method, for each <tt>Custom</tt> operation, a
|
|
case statement should be added to indicate what function to call. In the
|
|
following code, an <tt>FP_TO_SINT</tt> opcode will call
|
|
the <tt>LowerFP_TO_SINT</tt> method:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
SDValue SparcTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) {
|
|
switch (Op.getOpcode()) {
|
|
case ISD::FP_TO_SINT: return LowerFP_TO_SINT(Op, DAG);
|
|
...
|
|
}
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
Finally, the <tt>LowerFP_TO_SINT</tt> method is implemented, using an FP
|
|
register to convert the floating-point value to an integer.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
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::BITCAST, MVT::i32, Op);
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
</div>
|
|
|
|
<!-- _______________________________________________________________________ -->
|
|
<h4>
|
|
<a name="legal">Legal</a>
|
|
</h4>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
The <tt>Legal</tt> LegalizeAction enum value simply indicates that an
|
|
operation <b>is</b> natively supported. <tt>Legal</tt> represents the default
|
|
condition, so it is rarely used. In <tt>SparcISelLowering.cpp</tt>, the action
|
|
for <tt>CTPOP</tt> (an operation to count the bits set in an integer) is
|
|
natively supported only for SPARC v9. The following code enables
|
|
the <tt>Expand</tt> conversion technique for non-v9 SPARC implementations.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
setOperationAction(ISD::CTPOP, MVT::i32, Expand);
|
|
...
|
|
if (TM.getSubtarget<SparcSubtarget>().isV9())
|
|
setOperationAction(ISD::CTPOP, MVT::i32, Legal);
|
|
case ISD::SETULT: return SPCC::ICC_CS;
|
|
case ISD::SETULE: return SPCC::ICC_LEU;
|
|
case ISD::SETUGT: return SPCC::ICC_GU;
|
|
case ISD::SETUGE: return SPCC::ICC_CC;
|
|
}
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
</div>
|
|
|
|
</div>
|
|
|
|
<!-- ======================================================================= -->
|
|
<h3>
|
|
<a name="callingConventions">Calling Conventions</a>
|
|
</h3>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
To support target-specific calling conventions, <tt>XXXGenCallingConv.td</tt>
|
|
uses interfaces (such as CCIfType and CCAssignToReg) that are defined in
|
|
<tt>lib/Target/TargetCallingConv.td</tt>. TableGen can take the target
|
|
descriptor file <tt>XXXGenCallingConv.td</tt> and generate the header
|
|
file <tt>XXXGenCallingConv.inc</tt>, which is typically included
|
|
in <tt>XXXISelLowering.cpp</tt>. You can use the interfaces in
|
|
<tt>TargetCallingConv.td</tt> to specify:
|
|
</p>
|
|
|
|
<ul>
|
|
<li>The order of parameter allocation.</li>
|
|
|
|
<li>Where parameters and return values are placed (that is, on the stack or in
|
|
registers).</li>
|
|
|
|
<li>Which registers may be used.</li>
|
|
|
|
<li>Whether the caller or callee unwinds the stack.</li>
|
|
</ul>
|
|
|
|
<p>
|
|
The following example demonstrates the use of the <tt>CCIfType</tt> and
|
|
<tt>CCAssignToReg</tt> interfaces. If the <tt>CCIfType</tt> predicate is true
|
|
(that is, if the current argument is of type <tt>f32</tt> or <tt>f64</tt>), then
|
|
the action is performed. In this case, the <tt>CCAssignToReg</tt> action assigns
|
|
the argument value to the first available register: either <tt>R0</tt>
|
|
or <tt>R1</tt>.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
CCIfType<[f32,f64], CCAssignToReg<[R0, R1]>>
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
<tt>SparcCallingConv.td</tt> contains definitions for a target-specific
|
|
return-value calling convention (RetCC_Sparc32) and a basic 32-bit C calling
|
|
convention (<tt>CC_Sparc32</tt>). The definition of <tt>RetCC_Sparc32</tt>
|
|
(shown below) indicates which registers are used for specified scalar return
|
|
types. A single-precision float is returned to register <tt>F0</tt>, and a
|
|
double-precision float goes to register <tt>D0</tt>. A 32-bit integer is
|
|
returned in register <tt>I0</tt> or <tt>I1</tt>.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
def RetCC_Sparc32 : CallingConv<[
|
|
CCIfType<[i32], CCAssignToReg<[I0, I1]>>,
|
|
CCIfType<[f32], CCAssignToReg<[F0]>>,
|
|
CCIfType<[f64], CCAssignToReg<[D0]>>
|
|
]>;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
The definition of <tt>CC_Sparc32</tt> in <tt>SparcCallingConv.td</tt> introduces
|
|
<tt>CCAssignToStack</tt>, 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.)
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
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>
|
|
]>;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
<tt>CCDelegateTo</tt> 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 <tt>X86CallingConv.td</tt>), the definition of
|
|
<tt>RetCC_X86_32_C</tt> ends with <tt>CCDelegateTo</tt>. After the current value
|
|
is assigned to the register <tt>ST0</tt> or <tt>ST1</tt>,
|
|
the <tt>RetCC_X86Common</tt> is invoked.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
def RetCC_X86_32_C : CallingConv<[
|
|
CCIfType<[f32], CCAssignToReg<[ST0, ST1]>>,
|
|
CCIfType<[f64], CCAssignToReg<[ST0, ST1]>>,
|
|
CCDelegateTo<RetCC_X86Common>
|
|
]>;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
<tt>CCIfCC</tt> 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
|
|
<tt>X86CallingConv.td</tt>), if the <tt>Fast</tt> calling convention is in use,
|
|
then <tt>RetCC_X86_32_Fast</tt> is invoked. If the <tt>SSECall</tt> calling
|
|
convention is in use, then <tt>RetCC_X86_32_SSE</tt> is invoked.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
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>
|
|
]>;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>Other calling convention interfaces include:</p>
|
|
|
|
<ul>
|
|
<li><tt>CCIf <predicate, action></tt> — If the predicate matches,
|
|
apply the action.</li>
|
|
|
|
<li><tt>CCIfInReg <action></tt> — If the argument is marked with the
|
|
'<tt>inreg</tt>' attribute, then apply the action.</li>
|
|
|
|
<li><tt>CCIfNest <action></tt> — Inf the argument is marked with the
|
|
'<tt>nest</tt>' attribute, then apply the action.</li>
|
|
|
|
<li><tt>CCIfNotVarArg <action></tt> — If the current function does
|
|
not take a variable number of arguments, apply the action.</li>
|
|
|
|
<li><tt>CCAssignToRegWithShadow <registerList, shadowList></tt> —
|
|
similar to <tt>CCAssignToReg</tt>, but with a shadow list of registers.</li>
|
|
|
|
<li><tt>CCPassByVal <size, align></tt> — Assign value to a stack
|
|
slot with the minimum specified size and alignment.</li>
|
|
|
|
<li><tt>CCPromoteToType <type></tt> — Promote the current value to
|
|
the specified type.</li>
|
|
|
|
<li><tt>CallingConv <[actions]></tt> — Define each calling
|
|
convention that is supported.</li>
|
|
</ul>
|
|
|
|
</div>
|
|
|
|
</div>
|
|
|
|
<!-- *********************************************************************** -->
|
|
<h2>
|
|
<a name="assemblyPrinter">Assembly Printer</a>
|
|
</h2>
|
|
<!-- *********************************************************************** -->
|
|
|
|
<div>
|
|
|
|
<p>
|
|
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:
|
|
</p>
|
|
|
|
<ul>
|
|
<li>Define all the assembly strings for your target, adding them to the
|
|
instructions defined in the <tt>XXXInstrInfo.td</tt> file.
|
|
(See <a href="#InstructionSet">Instruction Set</a>.) TableGen will produce
|
|
an output file (<tt>XXXGenAsmWriter.inc</tt>) with an implementation of
|
|
the <tt>printInstruction</tt> method for the XXXAsmPrinter class.</li>
|
|
|
|
<li>Write <tt>XXXTargetAsmInfo.h</tt>, which contains the bare-bones declaration
|
|
of the <tt>XXXTargetAsmInfo</tt> class (a subclass
|
|
of <tt>TargetAsmInfo</tt>).</li>
|
|
|
|
<li>Write <tt>XXXTargetAsmInfo.cpp</tt>, which contains target-specific values
|
|
for <tt>TargetAsmInfo</tt> properties and sometimes new implementations for
|
|
methods.</li>
|
|
|
|
<li>Write <tt>XXXAsmPrinter.cpp</tt>, which implements the <tt>AsmPrinter</tt>
|
|
class that performs the LLVM-to-assembly conversion.</li>
|
|
</ul>
|
|
|
|
<p>
|
|
The code in <tt>XXXTargetAsmInfo.h</tt> is usually a trivial declaration of the
|
|
<tt>XXXTargetAsmInfo</tt> class for use in <tt>XXXTargetAsmInfo.cpp</tt>.
|
|
Similarly, <tt>XXXTargetAsmInfo.cpp</tt> usually has a few declarations of
|
|
<tt>XXXTargetAsmInfo</tt> replacement values that override the default values
|
|
in <tt>TargetAsmInfo.cpp</tt>. For example in <tt>SparcTargetAsmInfo.cpp</tt>:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
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";
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
The X86 assembly printer implementation (<tt>X86TargetAsmInfo</tt>) is an
|
|
example where the target specific <tt>TargetAsmInfo</tt> class uses an
|
|
overridden methods: <tt>ExpandInlineAsm</tt>.
|
|
</p>
|
|
|
|
<p>
|
|
A target-specific implementation of AsmPrinter is written in
|
|
<tt>XXXAsmPrinter.cpp</tt>, which implements the <tt>AsmPrinter</tt> class that
|
|
converts the LLVM to printable assembly. The implementation must include the
|
|
following headers that have declarations for the <tt>AsmPrinter</tt> and
|
|
<tt>MachineFunctionPass</tt> classes. The <tt>MachineFunctionPass</tt> is a
|
|
subclass of <tt>FunctionPass</tt>.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
#include "llvm/CodeGen/AsmPrinter.h"
|
|
#include "llvm/CodeGen/MachineFunctionPass.h"
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
As a <tt>FunctionPass</tt>, <tt>AsmPrinter</tt> first
|
|
calls <tt>doInitialization</tt> to set up the <tt>AsmPrinter</tt>. In
|
|
<tt>SparcAsmPrinter</tt>, a <tt>Mangler</tt> object is instantiated to process
|
|
variable names.
|
|
</p>
|
|
|
|
<p>
|
|
In <tt>XXXAsmPrinter.cpp</tt>, the <tt>runOnMachineFunction</tt> method
|
|
(declared in <tt>MachineFunctionPass</tt>) must be implemented
|
|
for <tt>XXXAsmPrinter</tt>. In <tt>MachineFunctionPass</tt>,
|
|
the <tt>runOnFunction</tt> method invokes <tt>runOnMachineFunction</tt>.
|
|
Target-specific implementations of <tt>runOnMachineFunction</tt> differ, but
|
|
generally do the following to process each machine function:
|
|
</p>
|
|
|
|
<ul>
|
|
<li>Call <tt>SetupMachineFunction</tt> to perform initialization.</li>
|
|
|
|
<li>Call <tt>EmitConstantPool</tt> to print out (to the output stream) constants
|
|
which have been spilled to memory.</li>
|
|
|
|
<li>Call <tt>EmitJumpTableInfo</tt> to print out jump tables used by the current
|
|
function.</li>
|
|
|
|
<li>Print out the label for the current function.</li>
|
|
|
|
<li>Print out the code for the function, including basic block labels and the
|
|
assembly for the instruction (using <tt>printInstruction</tt>)</li>
|
|
</ul>
|
|
|
|
<p>
|
|
The <tt>XXXAsmPrinter</tt> implementation must also include the code generated
|
|
by TableGen that is output in the <tt>XXXGenAsmWriter.inc</tt> file. The code
|
|
in <tt>XXXGenAsmWriter.inc</tt> contains an implementation of the
|
|
<tt>printInstruction</tt> method that may call these methods:
|
|
</p>
|
|
|
|
<ul>
|
|
<li><tt>printOperand</tt></li>
|
|
|
|
<li><tt>printMemOperand</tt></li>
|
|
|
|
<li><tt>printCCOperand (for conditional statements)</tt></li>
|
|
|
|
<li><tt>printDataDirective</tt></li>
|
|
|
|
<li><tt>printDeclare</tt></li>
|
|
|
|
<li><tt>printImplicitDef</tt></li>
|
|
|
|
<li><tt>printInlineAsm</tt></li>
|
|
</ul>
|
|
|
|
<p>
|
|
The implementations of <tt>printDeclare</tt>, <tt>printImplicitDef</tt>,
|
|
<tt>printInlineAsm</tt>, and <tt>printLabel</tt> in <tt>AsmPrinter.cpp</tt> are
|
|
generally adequate for printing assembly and do not need to be
|
|
overridden.
|
|
</p>
|
|
|
|
<p>
|
|
The <tt>printOperand</tt> 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 <tt>printMemOperand</tt> method
|
|
should be implemented to generate the proper output. Similarly,
|
|
<tt>printCCOperand</tt> should be used to print a conditional operand.
|
|
</p>
|
|
|
|
<p><tt>doFinalization</tt> should be overridden in <tt>XXXAsmPrinter</tt>, and
|
|
it should be called to shut down the assembly printer. During
|
|
<tt>doFinalization</tt>, global variables and constants are printed to
|
|
output.
|
|
</p>
|
|
|
|
</div>
|
|
|
|
<!-- *********************************************************************** -->
|
|
<h2>
|
|
<a name="subtargetSupport">Subtarget Support</a>
|
|
</h2>
|
|
<!-- *********************************************************************** -->
|
|
|
|
<div>
|
|
|
|
<p>
|
|
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.
|
|
</p>
|
|
|
|
<p>
|
|
If subtarget support is needed, you should implement a target-specific
|
|
XXXSubtarget class for your architecture. This class should process the
|
|
command-line options <tt>-mcpu=</tt> and <tt>-mattr=</tt>.
|
|
</p>
|
|
|
|
<p>
|
|
TableGen uses definitions in the <tt>Target.td</tt> and <tt>Sparc.td</tt> files
|
|
to generate code in <tt>SparcGenSubtarget.inc</tt>. In <tt>Target.td</tt>, shown
|
|
below, the <tt>SubtargetFeature</tt> interface is defined. The first 4 string
|
|
parameters of the <tt>SubtargetFeature</tt> 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.)
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
class SubtargetFeature<string n, string a, string v, string d,
|
|
list<SubtargetFeature> i = []> {
|
|
string Name = n;
|
|
string Attribute = a;
|
|
string Value = v;
|
|
string Desc = d;
|
|
list<SubtargetFeature> Implies = i;
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
In the <tt>Sparc.td</tt> file, the SubtargetFeature is used to define the
|
|
following features.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
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">;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
Elsewhere in <tt>Sparc.td</tt>, the Proc class is defined and then is used to
|
|
define particular SPARC processor subtypes that may have the previously
|
|
described features.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
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]>;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
From <tt>Target.td</tt> and <tt>Sparc.td</tt> 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 <tt>SparcGenSubtarget.inc</tt> file
|
|
should be included in the <tt>SparcSubtarget.cpp</tt>. The target-specific
|
|
implementation of the XXXSubtarget method should follow this pseudocode:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
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
|
|
// Perform any additional operations
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
</div>
|
|
|
|
<!-- *********************************************************************** -->
|
|
<h2>
|
|
<a name="jitSupport">JIT Support</a>
|
|
</h2>
|
|
<!-- *********************************************************************** -->
|
|
|
|
<div>
|
|
|
|
<p>
|
|
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:
|
|
</p>
|
|
|
|
<ul>
|
|
<li>Write an <tt>XXXCodeEmitter.cpp</tt> file that contains a machine function
|
|
pass that transforms target-machine instructions into relocatable machine
|
|
code.</li>
|
|
|
|
<li>Write an <tt>XXXJITInfo.cpp</tt> file that implements the JIT interfaces for
|
|
target-specific code-generation activities, such as emitting machine code
|
|
and stubs.</li>
|
|
|
|
<li>Modify <tt>XXXTargetMachine</tt> so that it provides a
|
|
<tt>TargetJITInfo</tt> object through its <tt>getJITInfo</tt> method.</li>
|
|
</ul>
|
|
|
|
<p>
|
|
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 <tt>XXXGenCodeEmitter.inc</tt>, which
|
|
contains the binary coding of machine instructions and the
|
|
<tt>getBinaryCodeForInstr</tt> method to access those codes. Other JIT
|
|
implementations do not.
|
|
</p>
|
|
|
|
<p>
|
|
Both <tt>XXXJITInfo.cpp</tt> and <tt>XXXCodeEmitter.cpp</tt> must include the
|
|
<tt>llvm/CodeGen/MachineCodeEmitter.h</tt> header file that defines the
|
|
<tt>MachineCodeEmitter</tt> class containing code for several callback functions
|
|
that write data (in bytes, words, strings, etc.) to the output stream.
|
|
</p>
|
|
|
|
<!-- ======================================================================= -->
|
|
<h3>
|
|
<a name="mce">Machine Code Emitter</a>
|
|
</h3>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
In <tt>XXXCodeEmitter.cpp</tt>, a target-specific of the <tt>Emitter</tt> class
|
|
is implemented as a function pass (subclass
|
|
of <tt>MachineFunctionPass</tt>). The target-specific implementation
|
|
of <tt>runOnMachineFunction</tt> (invoked by
|
|
<tt>runOnFunction</tt> in <tt>MachineFunctionPass</tt>) iterates through the
|
|
<tt>MachineBasicBlock</tt> calls <tt>emitInstruction</tt> to process each
|
|
instruction and emit binary code. <tt>emitInstruction</tt> is largely
|
|
implemented with case statements on the instruction types defined in
|
|
<tt>XXXInstrInfo.h</tt>. For example, in <tt>X86CodeEmitter.cpp</tt>,
|
|
the <tt>emitInstruction</tt> method is built around the following switch/case
|
|
statements:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
switch (Desc->TSFlags & X86::FormMask) {
|
|
case X86II::Pseudo: // for not yet implemented instructions
|
|
... // or pseudo-instructions
|
|
break;
|
|
case X86II::RawFrm: // for instructions with a fixed opcode value
|
|
...
|
|
break;
|
|
case X86II::AddRegFrm: // for instructions that have one register operand
|
|
... // added to their opcode
|
|
break;
|
|
case X86II::MRMDestReg:// for instructions that use the Mod/RM byte
|
|
... // to specify a destination (register)
|
|
break;
|
|
case X86II::MRMDestMem:// for instructions that use the Mod/RM byte
|
|
... // to specify a destination (memory)
|
|
break;
|
|
case X86II::MRMSrcReg: // for instructions that use the Mod/RM byte
|
|
... // to specify a source (register)
|
|
break;
|
|
case X86II::MRMSrcMem: // for instructions that use the Mod/RM byte
|
|
... // to specify a source (memory)
|
|
break;
|
|
case X86II::MRM0r: case X86II::MRM1r: // for instructions that operate on
|
|
case X86II::MRM2r: case X86II::MRM3r: // a REGISTER r/m operand and
|
|
case X86II::MRM4r: case X86II::MRM5r: // use the Mod/RM byte and a field
|
|
case X86II::MRM6r: case X86II::MRM7r: // to hold extended opcode data
|
|
...
|
|
break;
|
|
case X86II::MRM0m: case X86II::MRM1m: // for instructions that operate on
|
|
case X86II::MRM2m: case X86II::MRM3m: // a MEMORY r/m operand and
|
|
case X86II::MRM4m: case X86II::MRM5m: // use the Mod/RM byte and a field
|
|
case X86II::MRM6m: case X86II::MRM7m: // to hold extended opcode data
|
|
...
|
|
break;
|
|
case X86II::MRMInitReg: // for instructions whose source and
|
|
... // destination are the same register
|
|
break;
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
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 <tt>X86CodeEmitter.cpp</tt>,
|
|
for the <tt>X86II::AddRegFrm</tt> case, the first data emitted
|
|
(by <tt>emitByte</tt>) is the opcode added to the register operand. Then an
|
|
object representing the machine operand, <tt>MO1</tt>, is extracted. The helper
|
|
methods such as <tt>isImmediate</tt>,
|
|
<tt>isGlobalAddress</tt>, <tt>isExternalSymbol</tt>, <tt>isConstantPoolIndex</tt>, and
|
|
<tt>isJumpTableIndex</tt> determine the operand
|
|
type. (<tt>X86CodeEmitter.cpp</tt> also has private methods such
|
|
as <tt>emitConstant</tt>, <tt>emitGlobalAddress</tt>,
|
|
<tt>emitExternalSymbolAddress</tt>, <tt>emitConstPoolAddress</tt>,
|
|
and <tt>emitJumpTableAddress</tt> that emit the data into the output stream.)
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
case X86II::AddRegFrm:
|
|
MCE.emitByte(BaseOpcode + getX86RegNum(MI.getOperand(CurOp++).getReg()));
|
|
|
|
if (CurOp != NumOps) {
|
|
const MachineOperand &MO1 = MI.getOperand(CurOp++);
|
|
unsigned Size = X86InstrInfo::sizeOfImm(Desc);
|
|
if (MO1.isImmediate())
|
|
emitConstant(MO1.getImm(), Size);
|
|
else {
|
|
unsigned rt = Is64BitMode ? X86::reloc_pcrel_word
|
|
: (IsPIC ? X86::reloc_picrel_word : X86::reloc_absolute_word);
|
|
if (Opcode == X86::MOV64ri)
|
|
rt = X86::reloc_absolute_dword; // FIXME: add X86II flag?
|
|
if (MO1.isGlobalAddress()) {
|
|
bool NeedStub = isa<Function>(MO1.getGlobal());
|
|
bool isLazy = gvNeedsLazyPtr(MO1.getGlobal());
|
|
emitGlobalAddress(MO1.getGlobal(), rt, MO1.getOffset(), 0,
|
|
NeedStub, isLazy);
|
|
} else if (MO1.isExternalSymbol())
|
|
emitExternalSymbolAddress(MO1.getSymbolName(), rt);
|
|
else if (MO1.isConstantPoolIndex())
|
|
emitConstPoolAddress(MO1.getIndex(), rt);
|
|
else if (MO1.isJumpTableIndex())
|
|
emitJumpTableAddress(MO1.getIndex(), rt);
|
|
}
|
|
}
|
|
break;
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
In the previous example, <tt>XXXCodeEmitter.cpp</tt> uses the
|
|
variable <tt>rt</tt>, which is a RelocationType enum that may be used to
|
|
relocate addresses (for example, a global address with a PIC base offset). The
|
|
<tt>RelocationType</tt> enum for that target is defined in the short
|
|
target-specific <tt>XXXRelocations.h</tt> file. The <tt>RelocationType</tt> is used by
|
|
the <tt>relocate</tt> method defined in <tt>XXXJITInfo.cpp</tt> to rewrite
|
|
addresses for referenced global symbols.
|
|
</p>
|
|
|
|
<p>
|
|
For example, <tt>X86Relocations.h</tt> 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 <tt>reloc_pcrel_word</tt>
|
|
and <tt>reloc_picrel_word</tt>, there is an additional initial adjustment.
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
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
|
|
};
|
|
</pre>
|
|
</div>
|
|
|
|
</div>
|
|
|
|
<!-- ======================================================================= -->
|
|
<h3>
|
|
<a name="targetJITInfo">Target JIT Info</a>
|
|
</h3>
|
|
|
|
<div>
|
|
|
|
<p>
|
|
<tt>XXXJITInfo.cpp</tt> implements the JIT interfaces for target-specific
|
|
code-generation activities, such as emitting machine code and stubs. At minimum,
|
|
a target-specific version of <tt>XXXJITInfo</tt> implements the following:
|
|
</p>
|
|
|
|
<ul>
|
|
<li><tt>getLazyResolverFunction</tt> — Initializes the JIT, gives the
|
|
target a function that is used for compilation.</li>
|
|
|
|
<li><tt>emitFunctionStub</tt> — Returns a native function with a specified
|
|
address for a callback function.</li>
|
|
|
|
<li><tt>relocate</tt> — Changes the addresses of referenced globals, based
|
|
on relocation types.</li>
|
|
|
|
<li>Callback function that are wrappers to a function stub that is used when the
|
|
real target is not initially known.</li>
|
|
</ul>
|
|
|
|
<p>
|
|
<tt>getLazyResolverFunction</tt> is generally trivial to implement. It makes the
|
|
incoming parameter as the global <tt>JITCompilerFunction</tt> and returns the
|
|
callback function that will be used a function wrapper. For the Alpha target
|
|
(in <tt>AlphaJITInfo.cpp</tt>), the <tt>getLazyResolverFunction</tt>
|
|
implementation is simply:
|
|
</p>
|
|
|
|
<div class="doc_code">
|
|
<pre>
|
|
TargetJITInfo::LazyResolverFn AlphaJITInfo::getLazyResolverFunction(
|
|
JITCompilerFn F) {
|
|
JITCompilerFunction = F;
|
|
return AlphaCompilationCallback;
|
|
}
|
|
</pre>
|
|
</div>
|
|
|
|
<p>
|
|
For the X86 target, the <tt>getLazyResolverFunction</tt> implementation is a
|
|
little more complication, because it returns a different callback function for
|
|
processors with SSE instructions and XMM registers.
|
|
</p>
|
|
|
|
<p>
|
|
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.
|
|
</p>
|
|
|
|
</div>
|
|
|
|
</div>
|
|
|
|
<!-- *********************************************************************** -->
|
|
|
|
<hr>
|
|
<address>
|
|
<a href="http://jigsaw.w3.org/css-validator/check/referer"><img
|
|
src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
|
|
<a href="http://validator.w3.org/check/referer"><img
|
|
src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
|
|
|
|
<a href="http://www.woo.com">Mason Woo</a> and <a href="http://misha.brukman.net">Misha Brukman</a><br>
|
|
<a href="http://llvm.org/">The LLVM Compiler Infrastructure</a>
|
|
<br>
|
|
Last modified: $Date$
|
|
</address>
|
|
|
|
</body>
|
|
</html>
|