CodeGenerator.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>The LLVM Target-Independent Code Generator</title>
<link rel="stylesheet" href="llvm.css" type="text/css">
<style type="text/css">
.unknown { background-color: #C0C0C0; text-align: center; }
.unknown:before { content: "?" }
.no { background-color: #C11B17 }
.no:before { content: "N" }
.partial { background-color: #F88017 }
.yes { background-color: #0F0; }
.yes:before { content: "Y" }
</style>
</head>
<body>
<h1>
The LLVM Target-Independent Code Generator
</h1>
<ol>
<li><a href="#introduction">Introduction</a>
<ul>
<li><a href="#required">Required components in the code generator</a></li>
<li><a href="#high-level-design">The high-level design of the code
generator</a></li>
<li><a href="#tablegen">Using TableGen for target description</a></li>
</ul>
</li>
<li><a href="#targetdesc">Target description classes</a>
<ul>
<li><a href="#targetmachine">The <tt>TargetMachine</tt> class</a></li>
<li><a href="#targetdata">The <tt>TargetData</tt> class</a></li>
<li><a href="#targetlowering">The <tt>TargetLowering</tt> class</a></li>
<li><a href="#targetregisterinfo">The <tt>TargetRegisterInfo</tt> class</a></li>
<li><a href="#targetinstrinfo">The <tt>TargetInstrInfo</tt> class</a></li>
<li><a href="#targetframeinfo">The <tt>TargetFrameInfo</tt> class</a></li>
<li><a href="#targetsubtarget">The <tt>TargetSubtarget</tt> class</a></li>
<li><a href="#targetjitinfo">The <tt>TargetJITInfo</tt> class</a></li>
</ul>
</li>
<li><a href="#codegendesc">The "Machine" Code Generator classes</a>
<ul>
<li><a href="#machineinstr">The <tt>MachineInstr</tt> class</a></li>
<li><a href="#machinebasicblock">The <tt>MachineBasicBlock</tt>
class</a></li>
<li><a href="#machinefunction">The <tt>MachineFunction</tt> class</a></li>
<li><a href="#machineinstrbundle"><tt>MachineInstr Bundles</tt></a></li>
</ul>
</li>
<li><a href="#mc">The "MC" Layer</a>
<ul>
<li><a href="#mcstreamer">The <tt>MCStreamer</tt> API</a></li>
<li><a href="#mccontext">The <tt>MCContext</tt> class</a>
<li><a href="#mcsymbol">The <tt>MCSymbol</tt> class</a></li>
<li><a href="#mcsection">The <tt>MCSection</tt> class</a></li>
<li><a href="#mcinst">The <tt>MCInst</tt> class</a></li>
</ul>
</li>
<li><a href="#codegenalgs">Target-independent code generation algorithms</a>
<ul>
<li><a href="#instselect">Instruction Selection</a>
<ul>
<li><a href="#selectiondag_intro">Introduction to SelectionDAGs</a></li>
<li><a href="#selectiondag_process">SelectionDAG Code Generation
Process</a></li>
<li><a href="#selectiondag_build">Initial SelectionDAG
Construction</a></li>
<li><a href="#selectiondag_legalize_types">SelectionDAG LegalizeTypes Phase</a></li>
<li><a href="#selectiondag_legalize">SelectionDAG Legalize Phase</a></li>
<li><a href="#selectiondag_optimize">SelectionDAG Optimization
Phase: the DAG Combiner</a></li>
<li><a href="#selectiondag_select">SelectionDAG Select Phase</a></li>
<li><a href="#selectiondag_sched">SelectionDAG Scheduling and Formation
Phase</a></li>
<li><a href="#selectiondag_future">Future directions for the
SelectionDAG</a></li>
</ul></li>
<li><a href="#liveintervals">Live Intervals</a>
<ul>
<li><a href="#livevariable_analysis">Live Variable Analysis</a></li>
<li><a href="#liveintervals_analysis">Live Intervals Analysis</a></li>
</ul></li>
<li><a href="#regalloc">Register Allocation</a>
<ul>
<li><a href="#regAlloc_represent">How registers are represented in
LLVM</a></li>
<li><a href="#regAlloc_howTo">Mapping virtual registers to physical
registers</a></li>
<li><a href="#regAlloc_twoAddr">Handling two address instructions</a></li>
<li><a href="#regAlloc_ssaDecon">The SSA deconstruction phase</a></li>
<li><a href="#regAlloc_fold">Instruction folding</a></li>
<li><a href="#regAlloc_builtIn">Built in register allocators</a></li>
</ul></li>
<li><a href="#codeemit">Code Emission</a></li>
<li><a href="#vliw_packetizer">VLIW Packetizer</a>
<ul>
<li><a href="#vliw_mapping">Mapping from instructions to functional
units</a></li>
<li><a href="#vliw_repr">How the packetization tables are
generated and used</a></li>
</ul>
</li>
</ul>
</li>
<li><a href="#nativeassembler">Implementing a Native Assembler</a></li>
<li><a href="#targetimpls">Target-specific Implementation Notes</a>
<ul>
<li><a href="#targetfeatures">Target Feature Matrix</a></li>
<li><a href="#tailcallopt">Tail call optimization</a></li>
<li><a href="#sibcallopt">Sibling call optimization</a></li>
<li><a href="#x86">The X86 backend</a></li>
<li><a href="#ppc">The PowerPC backend</a>
<ul>
<li><a href="#ppc_abi">LLVM PowerPC ABI</a></li>
<li><a href="#ppc_frame">Frame Layout</a></li>
<li><a href="#ppc_prolog">Prolog/Epilog</a></li>
<li><a href="#ppc_dynamic">Dynamic Allocation</a></li>
</ul></li>
<li><a href="#ptx">The PTX backend</a></li>
</ul></li>
</ol>
<div class="doc_author">
<p>Written by the LLVM Team.</p>
</div>
<div class="doc_warning">
<p>Warning: This is a work in progress.</p>
</div>
<!-- *********************************************************************** -->
<h2>
<a name="introduction">Introduction</a>
</h2>
<!-- *********************************************************************** -->
<div>
<p>The LLVM target-independent code generator is a framework that provides a
suite of reusable components for translating the LLVM internal representation
to the machine code for a specified target—either in assembly form
(suitable for a static compiler) or in binary machine code format (usable for
a JIT compiler). The LLVM target-independent code generator consists of six
main components:</p>
<ol>
<li><a href="#targetdesc">Abstract target description</a> interfaces which
capture important properties about various aspects of the machine,
independently of how they will be used. These interfaces are defined in
<tt>include/llvm/Target/</tt>.</li>
<li>Classes used to represent the <a href="#codegendesc">code being
generated</a> for a target. These classes are intended to be abstract
enough to represent the machine code for <i>any</i> target machine. These
classes are defined in <tt>include/llvm/CodeGen/</tt>. At this level,
concepts like "constant pool entries" and "jump tables" are explicitly
exposed.</li>
<li>Classes and algorithms used to represent code as the object file level,
the <a href="#mc">MC Layer</a>. These classes represent assembly level
constructs like labels, sections, and instructions. At this level,
concepts like "constant pool entries" and "jump tables" don't exist.</li>
<li><a href="#codegenalgs">Target-independent algorithms</a> used to implement
various phases of native code generation (register allocation, scheduling,
stack frame representation, etc). This code lives
in <tt>lib/CodeGen/</tt>.</li>
<li><a href="#targetimpls">Implementations of the abstract target description
interfaces</a> for particular targets. These machine descriptions make
use of the components provided by LLVM, and can optionally provide custom
target-specific passes, to build complete code generators for a specific
target. Target descriptions live in <tt>lib/Target/</tt>.</li>
<li><a href="#jit">The target-independent JIT components</a>. The LLVM JIT is
completely target independent (it uses the <tt>TargetJITInfo</tt>
structure to interface for target-specific issues. The code for the
target-independent JIT lives in <tt>lib/ExecutionEngine/JIT</tt>.</li>
</ol>
<p>Depending on which part of the code generator you are interested in working
on, different pieces of this will be useful to you. In any case, you should
be familiar with the <a href="#targetdesc">target description</a>
and <a href="#codegendesc">machine code representation</a> classes. If you
want to add a backend for a new target, you will need
to <a href="#targetimpls">implement the target description</a> classes for
your new target and understand the <a href="LangRef.html">LLVM code
representation</a>. If you are interested in implementing a
new <a href="#codegenalgs">code generation algorithm</a>, it should only
depend on the target-description and machine code representation classes,
ensuring that it is portable.</p>
<!-- ======================================================================= -->
<h3>
<a name="required">Required components in the code generator</a>
</h3>
<div>
<p>The two pieces of the LLVM code generator are the high-level interface to the
code generator and the set of reusable components that can be used to build
target-specific backends. The two most important interfaces
(<a href="#targetmachine"><tt>TargetMachine</tt></a>
and <a href="#targetdata"><tt>TargetData</tt></a>) are the only ones that are
required to be defined for a backend to fit into the LLVM system, but the
others must be defined if the reusable code generator components are going to
be used.</p>
<p>This design has two important implications. The first is that LLVM can
support completely non-traditional code generation targets. For example, the
C backend does not require register allocation, instruction selection, or any
of the other standard components provided by the system. As such, it only
implements these two interfaces, and does its own thing. Another example of
a code generator like this is a (purely hypothetical) backend that converts
LLVM to the GCC RTL form and uses GCC to emit machine code for a target.</p>
<p>This design also implies that it is possible to design and implement
radically different code generators in the LLVM system that do not make use
of any of the built-in components. Doing so is not recommended at all, but
could be required for radically different targets that do not fit into the
LLVM machine description model: FPGAs for example.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="high-level-design">The high-level design of the code generator</a>
</h3>
<div>
<p>The LLVM target-independent code generator is designed to support efficient
and quality code generation for standard register-based microprocessors.
Code generation in this model is divided into the following stages:</p>
<ol>
<li><b><a href="#instselect">Instruction Selection</a></b> — This phase
determines an efficient way to express the input LLVM code in the target
instruction set. This stage produces the initial code for the program in
the target instruction set, then makes use of virtual registers in SSA
form and physical registers that represent any required register
assignments due to target constraints or calling conventions. This step
turns the LLVM code into a DAG of target instructions.</li>
<li><b><a href="#selectiondag_sched">Scheduling and Formation</a></b> —
This phase takes the DAG of target instructions produced by the
instruction selection phase, determines an ordering of the instructions,
then emits the instructions
as <tt><a href="#machineinstr">MachineInstr</a></tt>s with that ordering.
Note that we describe this in the <a href="#instselect">instruction
selection section</a> because it operates on
a <a href="#selectiondag_intro">SelectionDAG</a>.</li>
<li><b><a href="#ssamco">SSA-based Machine Code Optimizations</a></b> —
This optional stage consists of a series of machine-code optimizations
that operate on the SSA-form produced by the instruction selector.
Optimizations like modulo-scheduling or peephole optimization work
here.</li>
<li><b><a href="#regalloc">Register Allocation</a></b> — The target code
is transformed from an infinite virtual register file in SSA form to the
concrete register file used by the target. This phase introduces spill
code and eliminates all virtual register references from the program.</li>
<li><b><a href="#proepicode">Prolog/Epilog Code Insertion</a></b> — Once
the machine code has been generated for the function and the amount of
stack space required is known (used for LLVM alloca's and spill slots),
the prolog and epilog code for the function can be inserted and "abstract
stack location references" can be eliminated. This stage is responsible
for implementing optimizations like frame-pointer elimination and stack
packing.</li>
<li><b><a href="#latemco">Late Machine Code Optimizations</a></b> —
Optimizations that operate on "final" machine code can go here, such as
spill code scheduling and peephole optimizations.</li>
<li><b><a href="#codeemit">Code Emission</a></b> — The final stage
actually puts out the code for the current function, either in the target
assembler format or in machine code.</li>
</ol>
<p>The code generator is based on the assumption that the instruction selector
will use an optimal pattern matching selector to create high-quality
sequences of native instructions. Alternative code generator designs based
on pattern expansion and aggressive iterative peephole optimization are much
slower. This design permits efficient compilation (important for JIT
environments) and aggressive optimization (used when generating code offline)
by allowing components of varying levels of sophistication to be used for any
step of compilation.</p>
<p>In addition to these stages, target implementations can insert arbitrary
target-specific passes into the flow. For example, the X86 target uses a
special pass to handle the 80x87 floating point stack architecture. Other
targets with unusual requirements can be supported with custom passes as
needed.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="tablegen">Using TableGen for target description</a>
</h3>
<div>
<p>The target description classes require a detailed description of the target
architecture. These target descriptions often have a large amount of common
information (e.g., an <tt>add</tt> instruction is almost identical to a
<tt>sub</tt> instruction). In order to allow the maximum amount of
commonality to be factored out, the LLVM code generator uses
the <a href="TableGenFundamentals.html">TableGen</a> tool to describe big
chunks of the target machine, which allows the use of domain-specific and
target-specific abstractions to reduce the amount of repetition.</p>
<p>As LLVM continues to be developed and refined, we plan to move more and more
of the target description to the <tt>.td</tt> form. Doing so gives us a
number of advantages. The most important is that it makes it easier to port
LLVM because it reduces the amount of C++ code that has to be written, and
the surface area of the code generator that needs to be understood before
someone can get something working. Second, it makes it easier to change
things. In particular, if tables and other things are all emitted
by <tt>tblgen</tt>, we only need a change in one place (<tt>tblgen</tt>) to
update all of the targets to a new interface.</p>
</div>
</div>
<!-- *********************************************************************** -->
<h2>
<a name="targetdesc">Target description classes</a>
</h2>
<!-- *********************************************************************** -->
<div>
<p>The LLVM target description classes (located in the
<tt>include/llvm/Target</tt> directory) provide an abstract description of
the target machine independent of any particular client. These classes are
designed to capture the <i>abstract</i> properties of the target (such as the
instructions and registers it has), and do not incorporate any particular
pieces of code generation algorithms.</p>
<p>All of the target description classes (except the
<tt><a href="#targetdata">TargetData</a></tt> class) are designed to be
subclassed by the concrete target implementation, and have virtual methods
implemented. To get to these implementations, the
<tt><a href="#targetmachine">TargetMachine</a></tt> class provides accessors
that should be implemented by the target.</p>
<!-- ======================================================================= -->
<h3>
<a name="targetmachine">The <tt>TargetMachine</tt> class</a>
</h3>
<div>
<p>The <tt>TargetMachine</tt> class provides virtual methods that are used to
access the target-specific implementations of the various target description
classes via the <tt>get*Info</tt> methods (<tt>getInstrInfo</tt>,
<tt>getRegisterInfo</tt>, <tt>getFrameInfo</tt>, etc.). This class is
designed to be specialized by a concrete target implementation
(e.g., <tt>X86TargetMachine</tt>) which implements the various virtual
methods. The only required target description class is
the <a href="#targetdata"><tt>TargetData</tt></a> class, but if the code
generator components are to be used, the other interfaces should be
implemented as well.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="targetdata">The <tt>TargetData</tt> class</a>
</h3>
<div>
<p>The <tt>TargetData</tt> class is the only required target description class,
and it is the only class that is not extensible (you cannot derived a new
class from it). <tt>TargetData</tt> specifies information about how the
target lays out memory for structures, the alignment requirements for various
data types, the size of pointers in the target, and whether the target is
little-endian or big-endian.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="targetlowering">The <tt>TargetLowering</tt> class</a>
</h3>
<div>
<p>The <tt>TargetLowering</tt> class is used by SelectionDAG based instruction
selectors primarily to describe how LLVM code should be lowered to
SelectionDAG operations. Among other things, this class indicates:</p>
<ul>
<li>an initial register class to use for various <tt>ValueType</tt>s,</li>
<li>which operations are natively supported by the target machine,</li>
<li>the return type of <tt>setcc</tt> operations,</li>
<li>the type to use for shift amounts, and</li>
<li>various high-level characteristics, like whether it is profitable to turn
division by a constant into a multiplication sequence</li>
</ul>
</div>
<!-- ======================================================================= -->
<h3>
<a name="targetregisterinfo">The <tt>TargetRegisterInfo</tt> class</a>
</h3>
<div>
<p>The <tt>TargetRegisterInfo</tt> class is used to describe the register file
of the target and any interactions between the registers.</p>
<p>Registers in the code generator are represented in the code generator by
unsigned integers. Physical registers (those that actually exist in the
target description) are unique small numbers, and virtual registers are
generally large. Note that register #0 is reserved as a flag value.</p>
<p>Each register in the processor description has an associated
<tt>TargetRegisterDesc</tt> entry, which provides a textual name for the
register (used for assembly output and debugging dumps) and a set of aliases
(used to indicate whether one register overlaps with another).</p>
<p>In addition to the per-register description, the <tt>TargetRegisterInfo</tt>
class exposes a set of processor specific register classes (instances of the
<tt>TargetRegisterClass</tt> class). Each register class contains sets of
registers that have the same properties (for example, they are all 32-bit
integer registers). Each SSA virtual register created by the instruction
selector has an associated register class. When the register allocator runs,
it replaces virtual registers with a physical register in the set.</p>
<p>The target-specific implementations of these classes is auto-generated from
a <a href="TableGenFundamentals.html">TableGen</a> description of the
register file.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="targetinstrinfo">The <tt>TargetInstrInfo</tt> class</a>
</h3>
<div>
<p>The <tt>TargetInstrInfo</tt> class is used to describe the machine
instructions supported by the target. It is essentially an array of
<tt>TargetInstrDescriptor</tt> objects, each of which describes one
instruction the target supports. Descriptors define things like the mnemonic
for the opcode, the number of operands, the list of implicit register uses
and defs, whether the instruction has certain target-independent properties
(accesses memory, is commutable, etc), and holds any target-specific
flags.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="targetframeinfo">The <tt>TargetFrameInfo</tt> class</a>
</h3>
<div>
<p>The <tt>TargetFrameInfo</tt> class is used to provide information about the
stack frame layout of the target. It holds the direction of stack growth, the
known stack alignment on entry to each function, and the offset to the local
area. The offset to the local area is the offset from the stack pointer on
function entry to the first location where function data (local variables,
spill locations) can be stored.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="targetsubtarget">The <tt>TargetSubtarget</tt> class</a>
</h3>
<div>
<p>The <tt>TargetSubtarget</tt> class is used to provide information about the
specific chip set being targeted. A sub-target informs code generation of
which instructions are supported, instruction latencies and instruction
execution itinerary; i.e., which processing units are used, in what order,
and for how long.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="targetjitinfo">The <tt>TargetJITInfo</tt> class</a>
</h3>
<div>
<p>The <tt>TargetJITInfo</tt> class exposes an abstract interface used by the
Just-In-Time code generator to perform target-specific activities, such as
emitting stubs. If a <tt>TargetMachine</tt> supports JIT code generation, it
should provide one of these objects through the <tt>getJITInfo</tt>
method.</p>
</div>
</div>
<!-- *********************************************************************** -->
<h2>
<a name="codegendesc">Machine code description classes</a>
</h2>
<!-- *********************************************************************** -->
<div>
<p>At the high-level, LLVM code is translated to a machine specific
representation formed out of
<a href="#machinefunction"><tt>MachineFunction</tt></a>,
<a href="#machinebasicblock"><tt>MachineBasicBlock</tt></a>,
and <a href="#machineinstr"><tt>MachineInstr</tt></a> instances (defined
in <tt>include/llvm/CodeGen</tt>). This representation is completely target
agnostic, representing instructions in their most abstract form: an opcode
and a series of operands. This representation is designed to support both an
SSA representation for machine code, as well as a register allocated, non-SSA
form.</p>
<!-- ======================================================================= -->
<h3>
<a name="machineinstr">The <tt>MachineInstr</tt> class</a>
</h3>
<div>
<p>Target machine instructions are represented as instances of the
<tt>MachineInstr</tt> class. This class is an extremely abstract way of
representing machine instructions. In particular, it only keeps track of an
opcode number and a set of operands.</p>
<p>The opcode number is a simple unsigned integer that only has meaning to a
specific backend. All of the instructions for a target should be defined in
the <tt>*InstrInfo.td</tt> file for the target. The opcode enum values are
auto-generated from this description. The <tt>MachineInstr</tt> class does
not have any information about how to interpret the instruction (i.e., what
the semantics of the instruction are); for that you must refer to the
<tt><a href="#targetinstrinfo">TargetInstrInfo</a></tt> class.</p>
<p>The operands of a machine instruction can be of several different types: a
register reference, a constant integer, a basic block reference, etc. In
addition, a machine operand should be marked as a def or a use of the value
(though only registers are allowed to be defs).</p>
<p>By convention, the LLVM code generator orders instruction operands so that
all register definitions come before the register uses, even on architectures
that are normally printed in other orders. For example, the SPARC add
instruction: "<tt>add %i1, %i2, %i3</tt>" adds the "%i1", and "%i2" registers
and stores the result into the "%i3" register. In the LLVM code generator,
the operands should be stored as "<tt>%i3, %i1, %i2</tt>": with the
destination first.</p>
<p>Keeping destination (definition) operands at the beginning of the operand
list has several advantages. In particular, the debugging printer will print
the instruction like this:</p>
<div class="doc_code">
<pre>
%r3 = add %i1, %i2
</pre>
</div>
<p>Also if the first operand is a def, it is easier to <a href="#buildmi">create
instructions</a> whose only def is the first operand.</p>
<!-- _______________________________________________________________________ -->
<h4>
<a name="buildmi">Using the <tt>MachineInstrBuilder.h</tt> functions</a>
</h4>
<div>
<p>Machine instructions are created by using the <tt>BuildMI</tt> functions,
located in the <tt>include/llvm/CodeGen/MachineInstrBuilder.h</tt> file. The
<tt>BuildMI</tt> functions make it easy to build arbitrary machine
instructions. Usage of the <tt>BuildMI</tt> functions look like this:</p>
<div class="doc_code">
<pre>
// Create a 'DestReg = mov 42' (rendered in X86 assembly as 'mov DestReg, 42')
// instruction. The '1' specifies how many operands will be added.
MachineInstr *MI = BuildMI(X86::MOV32ri, 1, DestReg).addImm(42);
// Create the same instr, but insert it at the end of a basic block.
MachineBasicBlock &MBB = ...
BuildMI(MBB, X86::MOV32ri, 1, DestReg).addImm(42);
// Create the same instr, but insert it before a specified iterator point.
MachineBasicBlock::iterator MBBI = ...
BuildMI(MBB, MBBI, X86::MOV32ri, 1, DestReg).addImm(42);
// Create a 'cmp Reg, 0' instruction, no destination reg.
MI = BuildMI(X86::CMP32ri, 2).addReg(Reg).addImm(0);
// Create an 'sahf' instruction which takes no operands and stores nothing.
MI = BuildMI(X86::SAHF, 0);
// Create a self looping branch instruction.
BuildMI(MBB, X86::JNE, 1).addMBB(&MBB);
</pre>
</div>
<p>The key thing to remember with the <tt>BuildMI</tt> functions is that you
have to specify the number of operands that the machine instruction will
take. This allows for efficient memory allocation. You also need to specify
if operands default to be uses of values, not definitions. If you need to
add a definition operand (other than the optional destination register), you
must explicitly mark it as such:</p>
<div class="doc_code">
<pre>
MI.addReg(Reg, RegState::Define);
</pre>
</div>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="fixedregs">Fixed (preassigned) registers</a>
</h4>
<div>
<p>One important issue that the code generator needs to be aware of is the
presence of fixed registers. In particular, there are often places in the
instruction stream where the register allocator <em>must</em> arrange for a
particular value to be in a particular register. This can occur due to
limitations of the instruction set (e.g., the X86 can only do a 32-bit divide
with the <tt>EAX</tt>/<tt>EDX</tt> registers), or external factors like
calling conventions. In any case, the instruction selector should emit code
that copies a virtual register into or out of a physical register when
needed.</p>
<p>For example, consider this simple LLVM example:</p>
<div class="doc_code">
<pre>
define i32 @test(i32 %X, i32 %Y) {
%Z = udiv i32 %X, %Y
ret i32 %Z
}
</pre>
</div>
<p>The X86 instruction selector produces this machine code for the <tt>div</tt>
and <tt>ret</tt> (use "<tt>llc X.bc -march=x86 -print-machineinstrs</tt>" to
get this):</p>
<div class="doc_code">
<pre>
;; Start of div
%EAX = mov %reg1024 ;; Copy X (in reg1024) into EAX
%reg1027 = sar %reg1024, 31
%EDX = mov %reg1027 ;; Sign extend X into EDX
idiv %reg1025 ;; Divide by Y (in reg1025)
%reg1026 = mov %EAX ;; Read the result (Z) out of EAX
;; Start of ret
%EAX = mov %reg1026 ;; 32-bit return value goes in EAX
ret
</pre>
</div>
<p>By the end of code generation, the register allocator has coalesced the
registers and deleted the resultant identity moves producing the following
code:</p>
<div class="doc_code">
<pre>
;; X is in EAX, Y is in ECX
mov %EAX, %EDX
sar %EDX, 31
idiv %ECX
ret
</pre>
</div>
<p>This approach is extremely general (if it can handle the X86 architecture, it
can handle anything!) and allows all of the target specific knowledge about
the instruction stream to be isolated in the instruction selector. Note that
physical registers should have a short lifetime for good code generation, and
all physical registers are assumed dead on entry to and exit from basic
blocks (before register allocation). Thus, if you need a value to be live
across basic block boundaries, it <em>must</em> live in a virtual
register.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="callclobber">Call-clobbered registers</a>
</h4>
<div>
<p>Some machine instructions, like calls, clobber a large number of physical
registers. Rather than adding <code><def,dead></code> operands for
all of them, it is possible to use an <code>MO_RegisterMask</code> operand
instead. The register mask operand holds a bit mask of preserved registers,
and everything else is considered to be clobbered by the instruction. </p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="ssa">Machine code in SSA form</a>
</h4>
<div>
<p><tt>MachineInstr</tt>'s are initially selected in SSA-form, and are
maintained in SSA-form until register allocation happens. For the most part,
this is trivially simple since LLVM is already in SSA form; LLVM PHI nodes
become machine code PHI nodes, and virtual registers are only allowed to have
a single definition.</p>
<p>After register allocation, machine code is no longer in SSA-form because
there are no virtual registers left in the code.</p>
</div>
</div>
<!-- ======================================================================= -->
<h3>
<a name="machinebasicblock">The <tt>MachineBasicBlock</tt> class</a>
</h3>
<div>
<p>The <tt>MachineBasicBlock</tt> class contains a list of machine instructions
(<tt><a href="#machineinstr">MachineInstr</a></tt> instances). It roughly
corresponds to the LLVM code input to the instruction selector, but there can
be a one-to-many mapping (i.e. one LLVM basic block can map to multiple
machine basic blocks). The <tt>MachineBasicBlock</tt> class has a
"<tt>getBasicBlock</tt>" method, which returns the LLVM basic block that it
comes from.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="machinefunction">The <tt>MachineFunction</tt> class</a>
</h3>
<div>
<p>The <tt>MachineFunction</tt> class contains a list of machine basic blocks
(<tt><a href="#machinebasicblock">MachineBasicBlock</a></tt> instances). It
corresponds one-to-one with the LLVM function input to the instruction
selector. In addition to a list of basic blocks,
the <tt>MachineFunction</tt> contains a a <tt>MachineConstantPool</tt>,
a <tt>MachineFrameInfo</tt>, a <tt>MachineFunctionInfo</tt>, and a
<tt>MachineRegisterInfo</tt>. See
<tt>include/llvm/CodeGen/MachineFunction.h</tt> for more information.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="machineinstrbundle"><tt>MachineInstr Bundles</tt></a>
</h3>
<div>
<p>LLVM code generator can model sequences of instructions as MachineInstr
bundles. A MI bundle can model a VLIW group / pack which contains an
arbitrary number of parallel instructions. It can also be used to model
a sequential list of instructions (potentially with data dependencies) that
cannot be legally separated (e.g. ARM Thumb2 IT blocks).</p>
<p>Conceptually a MI bundle is a MI with a number of other MIs nested within:
</p>
<div class="doc_code">
<pre>
--------------
| Bundle | ---------
-------------- \
| ----------------
| | MI |
| ----------------
| |
| ----------------
| | MI |
| ----------------
| |
| ----------------
| | MI |
| ----------------
|
--------------
| Bundle | --------
-------------- \
| ----------------
| | MI |
| ----------------
| |
| ----------------
| | MI |
| ----------------
| |
| ...
|
--------------
| Bundle | --------
-------------- \
|
...
</pre>
</div>
<p> MI bundle support does not change the physical representations of
MachineBasicBlock and MachineInstr. All the MIs (including top level and
nested ones) are stored as sequential list of MIs. The "bundled" MIs are
marked with the 'InsideBundle' flag. A top level MI with the special BUNDLE
opcode is used to represent the start of a bundle. It's legal to mix BUNDLE
MIs with indiviual MIs that are not inside bundles nor represent bundles.
</p>
<p> MachineInstr passes should operate on a MI bundle as a single unit. Member
methods have been taught to correctly handle bundles and MIs inside bundles.
The MachineBasicBlock iterator has been modified to skip over bundled MIs to
enforce the bundle-as-a-single-unit concept. An alternative iterator
instr_iterator has been added to MachineBasicBlock to allow passes to
iterate over all of the MIs in a MachineBasicBlock, including those which
are nested inside bundles. The top level BUNDLE instruction must have the
correct set of register MachineOperand's that represent the cumulative
inputs and outputs of the bundled MIs.</p>
<p> Packing / bundling of MachineInstr's should be done as part of the register
allocation super-pass. More specifically, the pass which determines what
MIs should be bundled together must be done after code generator exits SSA
form (i.e. after two-address pass, PHI elimination, and copy coalescing).
Bundles should only be finalized (i.e. adding BUNDLE MIs and input and
output register MachineOperands) after virtual registers have been
rewritten into physical registers. This requirement eliminates the need to
add virtual register operands to BUNDLE instructions which would effectively
double the virtual register def and use lists.</p>
</div>
</div>
<!-- *********************************************************************** -->
<h2>
<a name="mc">The "MC" Layer</a>
</h2>
<!-- *********************************************************************** -->
<div>
<p>
The MC Layer is used to represent and process code at the raw machine code
level, devoid of "high level" information like "constant pools", "jump tables",
"global variables" or anything like that. At this level, LLVM handles things
like label names, machine instructions, and sections in the object file. The
code in this layer is used for a number of important purposes: the tail end of
the code generator uses it to write a .s or .o file, and it is also used by the
llvm-mc tool to implement standalone machine code assemblers and disassemblers.
</p>
<p>
This section describes some of the important classes. There are also a number
of important subsystems that interact at this layer, they are described later
in this manual.
</p>
<!-- ======================================================================= -->
<h3>
<a name="mcstreamer">The <tt>MCStreamer</tt> API</a>
</h3>
<div>
<p>
MCStreamer is best thought of as an assembler API. It is an abstract API which
is <em>implemented</em> in different ways (e.g. to output a .s file, output an
ELF .o file, etc) but whose API correspond directly to what you see in a .s
file. MCStreamer has one method per directive, such as EmitLabel,
EmitSymbolAttribute, SwitchSection, EmitValue (for .byte, .word), etc, which
directly correspond to assembly level directives. It also has an
EmitInstruction method, which is used to output an MCInst to the streamer.
</p>
<p>
This API is most important for two clients: the llvm-mc stand-alone assembler is
effectively a parser that parses a line, then invokes a method on MCStreamer. In
the code generator, the <a href="#codeemit">Code Emission</a> phase of the code
generator lowers higher level LLVM IR and Machine* constructs down to the MC
layer, emitting directives through MCStreamer.</p>
<p>
On the implementation side of MCStreamer, there are two major implementations:
one for writing out a .s file (MCAsmStreamer), and one for writing out a .o
file (MCObjectStreamer). MCAsmStreamer is a straight-forward implementation
that prints out a directive for each method (e.g. EmitValue -> .byte), but
MCObjectStreamer implements a full assembler.
</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="mccontext">The <tt>MCContext</tt> class</a>
</h3>
<div>
<p>
The MCContext class is the owner of a variety of uniqued data structures at the
MC layer, including symbols, sections, etc. As such, this is the class that you
interact with to create symbols and sections. This class can not be subclassed.
</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="mcsymbol">The <tt>MCSymbol</tt> class</a>
</h3>
<div>
<p>
The MCSymbol class represents a symbol (aka label) in the assembly file. There
are two interesting kinds of symbols: assembler temporary symbols, and normal
symbols. Assembler temporary symbols are used and processed by the assembler
but are discarded when the object file is produced. The distinction is usually
represented by adding a prefix to the label, for example "L" labels are
assembler temporary labels in MachO.
</p>
<p>MCSymbols are created by MCContext and uniqued there. This means that
MCSymbols can be compared for pointer equivalence to find out if they are the
same symbol. Note that pointer inequality does not guarantee the labels will
end up at different addresses though. It's perfectly legal to output something
like this to the .s file:<p>
<pre>
foo:
bar:
.byte 4
</pre>
<p>In this case, both the foo and bar symbols will have the same address.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="mcsection">The <tt>MCSection</tt> class</a>
</h3>
<div>
<p>
The MCSection class represents an object-file specific section. It is subclassed
by object file specific implementations (e.g. <tt>MCSectionMachO</tt>,
<tt>MCSectionCOFF</tt>, <tt>MCSectionELF</tt>) and these are created and uniqued
by MCContext. The MCStreamer has a notion of the current section, which can be
changed with the SwitchToSection method (which corresponds to a ".section"
directive in a .s file).
</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="mcinst">The <tt>MCInst</tt> class</a>
</h3>
<div>
<p>
The MCInst class is a target-independent representation of an instruction. It
is a simple class (much more so than <a href="#machineinstr">MachineInstr</a>)
that holds a target-specific opcode and a vector of MCOperands. MCOperand, in
turn, is a simple discriminated union of three cases: 1) a simple immediate,
2) a target register ID, 3) a symbolic expression (e.g. "Lfoo-Lbar+42") as an
MCExpr.
</p>
<p>MCInst is the common currency used to represent machine instructions at the
MC layer. It is the type used by the instruction encoder, the instruction
printer, and the type generated by the assembly parser and disassembler.
</p>
</div>
</div>
<!-- *********************************************************************** -->
<h2>
<a name="codegenalgs">Target-independent code generation algorithms</a>
</h2>
<!-- *********************************************************************** -->
<div>
<p>This section documents the phases described in the
<a href="#high-level-design">high-level design of the code generator</a>.
It explains how they work and some of the rationale behind their design.</p>
<!-- ======================================================================= -->
<h3>
<a name="instselect">Instruction Selection</a>
</h3>
<div>
<p>Instruction Selection is the process of translating LLVM code presented to
the code generator into target-specific machine instructions. There are
several well-known ways to do this in the literature. LLVM uses a
SelectionDAG based instruction selector.</p>
<p>Portions of the DAG instruction selector are generated from the target
description (<tt>*.td</tt>) files. Our goal is for the entire instruction
selector to be generated from these <tt>.td</tt> files, though currently
there are still things that require custom C++ code.</p>
<!-- _______________________________________________________________________ -->
<h4>
<a name="selectiondag_intro">Introduction to SelectionDAGs</a>
</h4>
<div>
<p>The SelectionDAG provides an abstraction for code representation in a way
that is amenable to instruction selection using automatic techniques
(e.g. dynamic-programming based optimal pattern matching selectors). It is
also well-suited to other phases of code generation; in particular,
instruction scheduling (SelectionDAG's are very close to scheduling DAGs
post-selection). Additionally, the SelectionDAG provides a host
representation where a large variety of very-low-level (but
target-independent) <a href="#selectiondag_optimize">optimizations</a> may be
performed; ones which require extensive information about the instructions
efficiently supported by the target.</p>
<p>The SelectionDAG is a Directed-Acyclic-Graph whose nodes are instances of the
<tt>SDNode</tt> class. The primary payload of the <tt>SDNode</tt> is its
operation code (Opcode) that indicates what operation the node performs and
the operands to the operation. The various operation node types are
described at the top of the <tt>include/llvm/CodeGen/SelectionDAGNodes.h</tt>
file.</p>
<p>Although most operations define a single value, each node in the graph may
define multiple values. For example, a combined div/rem operation will
define both the dividend and the remainder. Many other situations require
multiple values as well. Each node also has some number of operands, which
are edges to the node defining the used value. Because nodes may define
multiple values, edges are represented by instances of the <tt>SDValue</tt>
class, which is a <tt><SDNode, unsigned></tt> pair, indicating the node
and result value being used, respectively. Each value produced by
an <tt>SDNode</tt> has an associated <tt>MVT</tt> (Machine Value Type)
indicating what the type of the value is.</p>
<p>SelectionDAGs contain two different kinds of values: those that represent
data flow and those that represent control flow dependencies. Data values
are simple edges with an integer or floating point value type. Control edges
are represented as "chain" edges which are of type <tt>MVT::Other</tt>.
These edges provide an ordering between nodes that have side effects (such as
loads, stores, calls, returns, etc). All nodes that have side effects should
take a token chain as input and produce a new one as output. By convention,
token chain inputs are always operand #0, and chain results are always the
last value produced by an operation.</p>
<p>A SelectionDAG has designated "Entry" and "Root" nodes. The Entry node is
always a marker node with an Opcode of <tt>ISD::EntryToken</tt>. The Root
node is the final side-effecting node in the token chain. For example, in a
single basic block function it would be the return node.</p>
<p>One important concept for SelectionDAGs is the notion of a "legal" vs.
"illegal" DAG. A legal DAG for a target is one that only uses supported
operations and supported types. On a 32-bit PowerPC, for example, a DAG with
a value of type i1, i8, i16, or i64 would be illegal, as would a DAG that
uses a SREM or UREM operation. The
<a href="#selectinodag_legalize_types">legalize types</a> and
<a href="#selectiondag_legalize">legalize operations</a> phases are
responsible for turning an illegal DAG into a legal DAG.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="selectiondag_process">SelectionDAG Instruction Selection Process</a>
</h4>
<div>
<p>SelectionDAG-based instruction selection consists of the following steps:</p>
<ol>
<li><a href="#selectiondag_build">Build initial DAG</a> — This stage
performs a simple translation from the input LLVM code to an illegal
SelectionDAG.</li>
<li><a href="#selectiondag_optimize">Optimize SelectionDAG</a> — This
stage performs simple optimizations on the SelectionDAG to simplify it,
and recognize meta instructions (like rotates
and <tt>div</tt>/<tt>rem</tt> pairs) for targets that support these meta
operations. This makes the resultant code more efficient and
the <a href="#selectiondag_select">select instructions from DAG</a> phase
(below) simpler.</li>
<li><a href="#selectiondag_legalize_types">Legalize SelectionDAG Types</a>
— This stage transforms SelectionDAG nodes to eliminate any types
that are unsupported on the target.</li>
<li><a href="#selectiondag_optimize">Optimize SelectionDAG</a> — The
SelectionDAG optimizer is run to clean up redundancies exposed by type
legalization.</li>
<li><a href="#selectiondag_legalize">Legalize SelectionDAG Ops</a> —
This stage transforms SelectionDAG nodes to eliminate any operations
that are unsupported on the target.</li>
<li><a href="#selectiondag_optimize">Optimize SelectionDAG</a> — The
SelectionDAG optimizer is run to eliminate inefficiencies introduced by
operation legalization.</li>
<li><a href="#selectiondag_select">Select instructions from DAG</a> —
Finally, the target instruction selector matches the DAG operations to
target instructions. This process translates the target-independent input
DAG into another DAG of target instructions.</li>
<li><a href="#selectiondag_sched">SelectionDAG Scheduling and Formation</a>
— The last phase assigns a linear order to the instructions in the
target-instruction DAG and emits them into the MachineFunction being
compiled. This step uses traditional prepass scheduling techniques.</li>
</ol>
<p>After all of these steps are complete, the SelectionDAG is destroyed and the
rest of the code generation passes are run.</p>
<p>One great way to visualize what is going on here is to take advantage of a
few LLC command line options. The following options pop up a window
displaying the SelectionDAG at specific times (if you only get errors printed
to the console while using this, you probably
<a href="ProgrammersManual.html#ViewGraph">need to configure your system</a>
to add support for it).</p>
<ul>
<li><tt>-view-dag-combine1-dags</tt> displays the DAG after being built,
before the first optimization pass.</li>
<li><tt>-view-legalize-dags</tt> displays the DAG before Legalization.</li>
<li><tt>-view-dag-combine2-dags</tt> displays the DAG before the second
optimization pass.</li>
<li><tt>-view-isel-dags</tt> displays the DAG before the Select phase.</li>
<li><tt>-view-sched-dags</tt> displays the DAG before Scheduling.</li>
</ul>
<p>The <tt>-view-sunit-dags</tt> displays the Scheduler's dependency graph.
This graph is based on the final SelectionDAG, with nodes that must be
scheduled together bundled into a single scheduling-unit node, and with
immediate operands and other nodes that aren't relevant for scheduling
omitted.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="selectiondag_build">Initial SelectionDAG Construction</a>
</h4>
<div>
<p>The initial SelectionDAG is naïvely peephole expanded from the LLVM
input by the <tt>SelectionDAGLowering</tt> class in the
<tt>lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp</tt> file. The intent of
this pass is to expose as much low-level, target-specific details to the
SelectionDAG as possible. This pass is mostly hard-coded (e.g. an
LLVM <tt>add</tt> turns into an <tt>SDNode add</tt> while a
<tt>getelementptr</tt> is expanded into the obvious arithmetic). This pass
requires target-specific hooks to lower calls, returns, varargs, etc. For
these features, the <tt><a href="#targetlowering">TargetLowering</a></tt>
interface is used.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="selectiondag_legalize_types">SelectionDAG LegalizeTypes Phase</a>
</h4>
<div>
<p>The Legalize phase is in charge of converting a DAG to only use the types
that are natively supported by the target.</p>
<p>There are two main ways of converting values of unsupported scalar types to
values of supported types: converting small types to larger types
("promoting"), and breaking up large integer types into smaller ones
("expanding"). For example, a target might require that all f32 values are
promoted to f64 and that all i1/i8/i16 values are promoted to i32. The same
target might require that all i64 values be expanded into pairs of i32
values. These changes can insert sign and zero extensions as needed to make
sure that the final code has the same behavior as the input.</p>
<p>There are two main ways of converting values of unsupported vector types to
value of supported types: splitting vector types, multiple times if
necessary, until a legal type is found, and extending vector types by adding
elements to the end to round them out to legal types ("widening"). If a
vector gets split all the way down to single-element parts with no supported
vector type being found, the elements are converted to scalars
("scalarizing").</p>
<p>A target implementation tells the legalizer which types are supported (and
which register class to use for them) by calling the
<tt>addRegisterClass</tt> method in its TargetLowering constructor.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="selectiondag_legalize">SelectionDAG Legalize Phase</a>
</h4>
<div>
<p>The Legalize phase is in charge of converting a DAG to only use the
operations that are natively supported by the target.</p>
<p>Targets often have weird constraints, such as not supporting every operation
on every supported datatype (e.g. X86 does not support byte conditional moves
and PowerPC does not support sign-extending loads from a 16-bit memory
location). Legalize takes care of this by open-coding another sequence of
operations to emulate the operation ("expansion"), by promoting one type to a
larger type that supports the operation ("promotion"), or by using a
target-specific hook to implement the legalization ("custom").</p>
<p>A target implementation tells the legalizer which operations are not
supported (and which of the above three actions to take) by calling the
<tt>setOperationAction</tt> method in its <tt>TargetLowering</tt>
constructor.</p>
<p>Prior to the existence of the Legalize passes, we required that every target
<a href="#selectiondag_optimize">selector</a> supported and handled every
operator and type even if they are not natively supported. The introduction
of the Legalize phases allows all of the canonicalization patterns to be
shared across targets, and makes it very easy to optimize the canonicalized
code because it is still in the form of a DAG.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="selectiondag_optimize">
SelectionDAG Optimization Phase: the DAG Combiner
</a>
</h4>
<div>
<p>The SelectionDAG optimization phase is run multiple times for code
generation, immediately after the DAG is built and once after each
legalization. The first run of the pass allows the initial code to be
cleaned up (e.g. performing optimizations that depend on knowing that the
operators have restricted type inputs). Subsequent runs of the pass clean up
the messy code generated by the Legalize passes, which allows Legalize to be
very simple (it can focus on making code legal instead of focusing on
generating <em>good</em> and legal code).</p>
<p>One important class of optimizations performed is optimizing inserted sign
and zero extension instructions. We currently use ad-hoc techniques, but
could move to more rigorous techniques in the future. Here are some good
papers on the subject:</p>
<p>"<a href="http://www.eecs.harvard.edu/~nr/pubs/widen-abstract.html">Widening
integer arithmetic</a>"<br>
Kevin Redwine and Norman Ramsey<br>
International Conference on Compiler Construction (CC) 2004</p>
<p>"<a href="http://portal.acm.org/citation.cfm?doid=512529.512552">Effective
sign extension elimination</a>"<br>
Motohiro Kawahito, Hideaki Komatsu, and Toshio Nakatani<br>
Proceedings of the ACM SIGPLAN 2002 Conference on Programming Language Design
and Implementation.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="selectiondag_select">SelectionDAG Select Phase</a>
</h4>
<div>
<p>The Select phase is the bulk of the target-specific code for instruction
selection. This phase takes a legal SelectionDAG as input, pattern matches
the instructions supported by the target to this DAG, and produces a new DAG
of target code. For example, consider the following LLVM fragment:</p>
<div class="doc_code">
<pre>
%t1 = fadd float %W, %X
%t2 = fmul float %t1, %Y
%t3 = fadd float %t2, %Z
</pre>
</div>
<p>This LLVM code corresponds to a SelectionDAG that looks basically like
this:</p>
<div class="doc_code">
<pre>
(fadd:f32 (fmul:f32 (fadd:f32 W, X), Y), Z)
</pre>
</div>
<p>If a target supports floating point multiply-and-add (FMA) operations, one of
the adds can be merged with the multiply. On the PowerPC, for example, the
output of the instruction selector might look like this DAG:</p>
<div class="doc_code">
<pre>
(FMADDS (FADDS W, X), Y, Z)
</pre>
</div>
<p>The <tt>FMADDS</tt> instruction is a ternary instruction that multiplies its
first two operands and adds the third (as single-precision floating-point
numbers). The <tt>FADDS</tt> instruction is a simple binary single-precision
add instruction. To perform this pattern match, the PowerPC backend includes
the following instruction definitions:</p>
<div class="doc_code">
<pre>
def FMADDS : AForm_1<59, 29,
(ops F4RC:$FRT, F4RC:$FRA, F4RC:$FRC, F4RC:$FRB),
"fmadds $FRT, $FRA, $FRC, $FRB",
[<b>(set F4RC:$FRT, (fadd (fmul F4RC:$FRA, F4RC:$FRC),
F4RC:$FRB))</b>]>;
def FADDS : AForm_2<59, 21,
(ops F4RC:$FRT, F4RC:$FRA, F4RC:$FRB),
"fadds $FRT, $FRA, $FRB",
[<b>(set F4RC:$FRT, (fadd F4RC:$FRA, F4RC:$FRB))</b>]>;
</pre>
</div>
<p>The portion of the instruction definition in bold indicates the pattern used
to match the instruction. The DAG operators
(like <tt>fmul</tt>/<tt>fadd</tt>) are defined in
the <tt>include/llvm/Target/TargetSelectionDAG.td</tt> file. "
<tt>F4RC</tt>" is the register class of the input and result values.</p>
<p>The TableGen DAG instruction selector generator reads the instruction
patterns in the <tt>.td</tt> file and automatically builds parts of the
pattern matching code for your target. It has the following strengths:</p>
<ul>
<li>At compiler-compiler time, it analyzes your instruction patterns and tells
you if your patterns make sense or not.</li>
<li>It can handle arbitrary constraints on operands for the pattern match. In
particular, it is straight-forward to say things like "match any immediate
that is a 13-bit sign-extended value". For examples, see the
<tt>immSExt16</tt> and related <tt>tblgen</tt> classes in the PowerPC
backend.</li>
<li>It knows several important identities for the patterns defined. For
example, it knows that addition is commutative, so it allows the
<tt>FMADDS</tt> pattern above to match "<tt>(fadd X, (fmul Y, Z))</tt>" as
well as "<tt>(fadd (fmul X, Y), Z)</tt>", without the target author having
to specially handle this case.</li>
<li>It has a full-featured type-inferencing system. In particular, you should
rarely have to explicitly tell the system what type parts of your patterns
are. In the <tt>FMADDS</tt> case above, we didn't have to tell
<tt>tblgen</tt> that all of the nodes in the pattern are of type 'f32'.
It was able to infer and propagate this knowledge from the fact that
<tt>F4RC</tt> has type 'f32'.</li>
<li>Targets can define their own (and rely on built-in) "pattern fragments".
Pattern fragments are chunks of reusable patterns that get inlined into
your patterns during compiler-compiler time. For example, the integer
"<tt>(not x)</tt>" operation is actually defined as a pattern fragment
that expands as "<tt>(xor x, -1)</tt>", since the SelectionDAG does not
have a native '<tt>not</tt>' operation. Targets can define their own
short-hand fragments as they see fit. See the definition of
'<tt>not</tt>' and '<tt>ineg</tt>' for examples.</li>
<li>In addition to instructions, targets can specify arbitrary patterns that
map to one or more instructions using the 'Pat' class. For example, the
PowerPC has no way to load an arbitrary integer immediate into a register
in one instruction. To tell tblgen how to do this, it defines:
<br>
<br>
<div class="doc_code">
<pre>
// Arbitrary immediate support. Implement in terms of LIS/ORI.
def : Pat<(i32 imm:$imm),
(ORI (LIS (HI16 imm:$imm)), (LO16 imm:$imm))>;
</pre>
</div>
<br>
If none of the single-instruction patterns for loading an immediate into a
register match, this will be used. This rule says "match an arbitrary i32
immediate, turning it into an <tt>ORI</tt> ('or a 16-bit immediate') and
an <tt>LIS</tt> ('load 16-bit immediate, where the immediate is shifted to
the left 16 bits') instruction". To make this work, the
<tt>LO16</tt>/<tt>HI16</tt> node transformations are used to manipulate
the input immediate (in this case, take the high or low 16-bits of the
immediate).</li>
<li>While the system does automate a lot, it still allows you to write custom
C++ code to match special cases if there is something that is hard to
express.</li>
</ul>
<p>While it has many strengths, the system currently has some limitations,
primarily because it is a work in progress and is not yet finished:</p>
<ul>
<li>Overall, there is no way to define or match SelectionDAG nodes that define
multiple values (e.g. <tt>SMUL_LOHI</tt>, <tt>LOAD</tt>, <tt>CALL</tt>,
etc). This is the biggest reason that you currently still <em>have
to</em> write custom C++ code for your instruction selector.</li>
<li>There is no great way to support matching complex addressing modes yet.
In the future, we will extend pattern fragments to allow them to define
multiple values (e.g. the four operands of the <a href="#x86_memory">X86
addressing mode</a>, which are currently matched with custom C++ code).
In addition, we'll extend fragments so that a fragment can match multiple
different patterns.</li>
<li>We don't automatically infer flags like isStore/isLoad yet.</li>
<li>We don't automatically generate the set of supported registers and
operations for the <a href="#selectiondag_legalize">Legalizer</a>
yet.</li>
<li>We don't have a way of tying in custom legalized nodes yet.</li>
</ul>
<p>Despite these limitations, the instruction selector generator is still quite
useful for most of the binary and logical operations in typical instruction
sets. If you run into any problems or can't figure out how to do something,
please let Chris know!</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="selectiondag_sched">SelectionDAG Scheduling and Formation Phase</a>
</h4>
<div>
<p>The scheduling phase takes the DAG of target instructions from the selection
phase and assigns an order. The scheduler can pick an order depending on
various constraints of the machines (i.e. order for minimal register pressure
or try to cover instruction latencies). Once an order is established, the
DAG is converted to a list
of <tt><a href="#machineinstr">MachineInstr</a></tt>s and the SelectionDAG is
destroyed.</p>
<p>Note that this phase is logically separate from the instruction selection
phase, but is tied to it closely in the code because it operates on
SelectionDAGs.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="selectiondag_future">Future directions for the SelectionDAG</a>
</h4>
<div>
<ol>
<li>Optional function-at-a-time selection.</li>
<li>Auto-generate entire selector from <tt>.td</tt> file.</li>
</ol>
</div>
</div>
<!-- ======================================================================= -->
<h3>
<a name="ssamco">SSA-based Machine Code Optimizations</a>
</h3>
<div><p>To Be Written</p></div>
<!-- ======================================================================= -->
<h3>
<a name="liveintervals">Live Intervals</a>
</h3>
<div>
<p>Live Intervals are the ranges (intervals) where a variable is <i>live</i>.
They are used by some <a href="#regalloc">register allocator</a> passes to
determine if two or more virtual registers which require the same physical
register are live at the same point in the program (i.e., they conflict).
When this situation occurs, one virtual register must be <i>spilled</i>.</p>
<!-- _______________________________________________________________________ -->
<h4>
<a name="livevariable_analysis">Live Variable Analysis</a>
</h4>
<div>
<p>The first step in determining the live intervals of variables is to calculate
the set of registers that are immediately dead after the instruction (i.e.,
the instruction calculates the value, but it is never used) and the set of
registers that are used by the instruction, but are never used after the
instruction (i.e., they are killed). Live variable information is computed
for each <i>virtual</i> register and <i>register allocatable</i> physical
register in the function. This is done in a very efficient manner because it
uses SSA to sparsely compute lifetime information for virtual registers
(which are in SSA form) and only has to track physical registers within a
block. Before register allocation, LLVM can assume that physical registers
are only live within a single basic block. This allows it to do a single,
local analysis to resolve physical register lifetimes within each basic
block. If a physical register is not register allocatable (e.g., a stack
pointer or condition codes), it is not tracked.</p>
<p>Physical registers may be live in to or out of a function. Live in values are
typically arguments in registers. Live out values are typically return values
in registers. Live in values are marked as such, and are given a dummy
"defining" instruction during live intervals analysis. If the last basic
block of a function is a <tt>return</tt>, then it's marked as using all live
out values in the function.</p>
<p><tt>PHI</tt> nodes need to be handled specially, because the calculation of
the live variable information from a depth first traversal of the CFG of the
function won't guarantee that a virtual register used by the <tt>PHI</tt>
node is defined before it's used. When a <tt>PHI</tt> node is encountered,
only the definition is handled, because the uses will be handled in other
basic blocks.</p>
<p>For each <tt>PHI</tt> node of the current basic block, we simulate an
assignment at the end of the current basic block and traverse the successor
basic blocks. If a successor basic block has a <tt>PHI</tt> node and one of
the <tt>PHI</tt> node's operands is coming from the current basic block, then
the variable is marked as <i>alive</i> within the current basic block and all
of its predecessor basic blocks, until the basic block with the defining
instruction is encountered.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="liveintervals_analysis">Live Intervals Analysis</a>
</h4>
<div>
<p>We now have the information available to perform the live intervals analysis
and build the live intervals themselves. We start off by numbering the basic
blocks and machine instructions. We then handle the "live-in" values. These
are in physical registers, so the physical register is assumed to be killed
by the end of the basic block. Live intervals for virtual registers are
computed for some ordering of the machine instructions <tt>[1, N]</tt>. A
live interval is an interval <tt>[i, j)</tt>, where <tt>1 <= i <= j
< N</tt>, for which a variable is live.</p>
<p><i><b>More to come...</b></i></p>
</div>
</div>
<!-- ======================================================================= -->
<h3>
<a name="regalloc">Register Allocation</a>
</h3>
<div>
<p>The <i>Register Allocation problem</i> consists in mapping a program
<i>P<sub>v</sub></i>, that can use an unbounded number of virtual registers,
to a program <i>P<sub>p</sub></i> that contains a finite (possibly small)
number of physical registers. Each target architecture has a different number
of physical registers. If the number of physical registers is not enough to
accommodate all the virtual registers, some of them will have to be mapped
into memory. These virtuals are called <i>spilled virtuals</i>.</p>
<!-- _______________________________________________________________________ -->
<h4>
<a name="regAlloc_represent">How registers are represented in LLVM</a>
</h4>
<div>
<p>In LLVM, physical registers are denoted by integer numbers that normally
range from 1 to 1023. To see how this numbering is defined for a particular
architecture, you can read the <tt>GenRegisterNames.inc</tt> file for that
architecture. For instance, by
inspecting <tt>lib/Target/X86/X86GenRegisterInfo.inc</tt> we see that the
32-bit register <tt>EAX</tt> is denoted by 43, and the MMX register
<tt>MM0</tt> is mapped to 65.</p>
<p>Some architectures contain registers that share the same physical location. A
notable example is the X86 platform. For instance, in the X86 architecture,
the registers <tt>EAX</tt>, <tt>AX</tt> and <tt>AL</tt> share the first eight
bits. These physical registers are marked as <i>aliased</i> in LLVM. Given a
particular architecture, you can check which registers are aliased by
inspecting its <tt>RegisterInfo.td</tt> file. Moreover, the method
<tt>MCRegisterInfo::getAliasSet(p_reg)</tt> returns an array containing
all the physical registers aliased to the register <tt>p_reg</tt>.</p>
<p>Physical registers, in LLVM, are grouped in <i>Register Classes</i>.
Elements in the same register class are functionally equivalent, and can be
interchangeably used. Each virtual register can only be mapped to physical
registers of a particular class. For instance, in the X86 architecture, some
virtuals can only be allocated to 8 bit registers. A register class is
described by <tt>TargetRegisterClass</tt> objects. To discover if a virtual
register is compatible with a given physical, this code can be used:</p>
<div class="doc_code">
<pre>
bool RegMapping_Fer::compatible_class(MachineFunction &mf,
unsigned v_reg,
unsigned p_reg) {
assert(TargetRegisterInfo::isPhysicalRegister(p_reg) &&
"Target register must be physical");
const TargetRegisterClass *trc = mf.getRegInfo().getRegClass(v_reg);
return trc->contains(p_reg);
}
</pre>
</div>
<p>Sometimes, mostly for debugging purposes, it is useful to change the number
of physical registers available in the target architecture. This must be done
statically, inside the <tt>TargetRegsterInfo.td</tt> file. Just <tt>grep</tt>
for <tt>RegisterClass</tt>, the last parameter of which is a list of
registers. Just commenting some out is one simple way to avoid them being
used. A more polite way is to explicitly exclude some registers from
the <i>allocation order</i>. See the definition of the <tt>GR8</tt> register
class in <tt>lib/Target/X86/X86RegisterInfo.td</tt> for an example of this.
</p>
<p>Virtual registers are also denoted by integer numbers. Contrary to physical
registers, different virtual registers never share the same number. Whereas
physical registers are statically defined in a <tt>TargetRegisterInfo.td</tt>
file and cannot be created by the application developer, that is not the case
with virtual registers. In order to create new virtual registers, use the
method <tt>MachineRegisterInfo::createVirtualRegister()</tt>. This method
will return a new virtual register. Use an <tt>IndexedMap<Foo,
VirtReg2IndexFunctor></tt> to hold information per virtual register. If you
need to enumerate all virtual registers, use the function
<tt>TargetRegisterInfo::index2VirtReg()</tt> to find the virtual register
numbers:</p>
<div class="doc_code">
<pre>
for (unsigned i = 0, e = MRI->getNumVirtRegs(); i != e; ++i) {
unsigned VirtReg = TargetRegisterInfo::index2VirtReg(i);
stuff(VirtReg);
}
</pre>
</div>
<p>Before register allocation, the operands of an instruction are mostly virtual
registers, although physical registers may also be used. In order to check if
a given machine operand is a register, use the boolean
function <tt>MachineOperand::isRegister()</tt>. To obtain the integer code of
a register, use <tt>MachineOperand::getReg()</tt>. An instruction may define
or use a register. For instance, <tt>ADD reg:1026 := reg:1025 reg:1024</tt>
defines the registers 1024, and uses registers 1025 and 1026. Given a
register operand, the method <tt>MachineOperand::isUse()</tt> informs if that
register is being used by the instruction. The
method <tt>MachineOperand::isDef()</tt> informs if that registers is being
defined.</p>
<p>We will call physical registers present in the LLVM bitcode before register
allocation <i>pre-colored registers</i>. Pre-colored registers are used in
many different situations, for instance, to pass parameters of functions
calls, and to store results of particular instructions. There are two types
of pre-colored registers: the ones <i>implicitly</i> defined, and
those <i>explicitly</i> defined. Explicitly defined registers are normal
operands, and can be accessed
with <tt>MachineInstr::getOperand(int)::getReg()</tt>. In order to check
which registers are implicitly defined by an instruction, use
the <tt>TargetInstrInfo::get(opcode)::ImplicitDefs</tt>,
where <tt>opcode</tt> is the opcode of the target instruction. One important
difference between explicit and implicit physical registers is that the
latter are defined statically for each instruction, whereas the former may
vary depending on the program being compiled. For example, an instruction
that represents a function call will always implicitly define or use the same
set of physical registers. To read the registers implicitly used by an
instruction,
use <tt>TargetInstrInfo::get(opcode)::ImplicitUses</tt>. Pre-colored
registers impose constraints on any register allocation algorithm. The
register allocator must make sure that none of them are overwritten by
the values of virtual registers while still alive.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="regAlloc_howTo">Mapping virtual registers to physical registers</a>
</h4>
<div>
<p>There are two ways to map virtual registers to physical registers (or to
memory slots). The first way, that we will call <i>direct mapping</i>, is
based on the use of methods of the classes <tt>TargetRegisterInfo</tt>,
and <tt>MachineOperand</tt>. The second way, that we will call <i>indirect
mapping</i>, relies on the <tt>VirtRegMap</tt> class in order to insert loads
and stores sending and getting values to and from memory.</p>
<p>The direct mapping provides more flexibility to the developer of the register
allocator; however, it is more error prone, and demands more implementation
work. Basically, the programmer will have to specify where load and store
instructions should be inserted in the target function being compiled in
order to get and store values in memory. To assign a physical register to a
virtual register present in a given operand,
use <tt>MachineOperand::setReg(p_reg)</tt>. To insert a store instruction,
use <tt>TargetInstrInfo::storeRegToStackSlot(...)</tt>, and to insert a
load instruction, use <tt>TargetInstrInfo::loadRegFromStackSlot</tt>.</p>
<p>The indirect mapping shields the application developer from the complexities
of inserting load and store instructions. In order to map a virtual register
to a physical one, use <tt>VirtRegMap::assignVirt2Phys(vreg, preg)</tt>. In
order to map a certain virtual register to memory,
use <tt>VirtRegMap::assignVirt2StackSlot(vreg)</tt>. This method will return
the stack slot where <tt>vreg</tt>'s value will be located. If it is
necessary to map another virtual register to the same stack slot,
use <tt>VirtRegMap::assignVirt2StackSlot(vreg, stack_location)</tt>. One
important point to consider when using the indirect mapping, is that even if
a virtual register is mapped to memory, it still needs to be mapped to a
physical register. This physical register is the location where the virtual
register is supposed to be found before being stored or after being
reloaded.</p>
<p>If the indirect strategy is used, after all the virtual registers have been
mapped to physical registers or stack slots, it is necessary to use a spiller
object to place load and store instructions in the code. Every virtual that
has been mapped to a stack slot will be stored to memory after been defined
and will be loaded before being used. The implementation of the spiller tries
to recycle load/store instructions, avoiding unnecessary instructions. For an
example of how to invoke the spiller,
see <tt>RegAllocLinearScan::runOnMachineFunction</tt>
in <tt>lib/CodeGen/RegAllocLinearScan.cpp</tt>.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="regAlloc_twoAddr">Handling two address instructions</a>
</h4>
<div>
<p>With very rare exceptions (e.g., function calls), the LLVM machine code
instructions are three address instructions. That is, each instruction is
expected to define at most one register, and to use at most two registers.
However, some architectures use two address instructions. In this case, the
defined register is also one of the used register. For instance, an
instruction such as <tt>ADD %EAX, %EBX</tt>, in X86 is actually equivalent
to <tt>%EAX = %EAX + %EBX</tt>.</p>
<p>In order to produce correct code, LLVM must convert three address
instructions that represent two address instructions into true two address
instructions. LLVM provides the pass <tt>TwoAddressInstructionPass</tt> for
this specific purpose. It must be run before register allocation takes
place. After its execution, the resulting code may no longer be in SSA
form. This happens, for instance, in situations where an instruction such
as <tt>%a = ADD %b %c</tt> is converted to two instructions such as:</p>
<div class="doc_code">
<pre>
%a = MOVE %b
%a = ADD %a %c
</pre>
</div>
<p>Notice that, internally, the second instruction is represented as
<tt>ADD %a[def/use] %c</tt>. I.e., the register operand <tt>%a</tt> is both
used and defined by the instruction.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="regAlloc_ssaDecon">The SSA deconstruction phase</a>
</h4>
<div>
<p>An important transformation that happens during register allocation is called
the <i>SSA Deconstruction Phase</i>. The SSA form simplifies many analyses
that are performed on the control flow graph of programs. However,
traditional instruction sets do not implement PHI instructions. Thus, in
order to generate executable code, compilers must replace PHI instructions
with other instructions that preserve their semantics.</p>
<p>There are many ways in which PHI instructions can safely be removed from the
target code. The most traditional PHI deconstruction algorithm replaces PHI
instructions with copy instructions. That is the strategy adopted by
LLVM. The SSA deconstruction algorithm is implemented
in <tt>lib/CodeGen/PHIElimination.cpp</tt>. In order to invoke this pass, the
identifier <tt>PHIEliminationID</tt> must be marked as required in the code
of the register allocator.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="regAlloc_fold">Instruction folding</a>
</h4>
<div>
<p><i>Instruction folding</i> is an optimization performed during register
allocation that removes unnecessary copy instructions. For instance, a
sequence of instructions such as:</p>
<div class="doc_code">
<pre>
%EBX = LOAD %mem_address
%EAX = COPY %EBX
</pre>
</div>
<p>can be safely substituted by the single instruction:</p>
<div class="doc_code">
<pre>
%EAX = LOAD %mem_address
</pre>
</div>
<p>Instructions can be folded with
the <tt>TargetRegisterInfo::foldMemoryOperand(...)</tt> method. Care must be
taken when folding instructions; a folded instruction can be quite different
from the original
instruction. See <tt>LiveIntervals::addIntervalsForSpills</tt>
in <tt>lib/CodeGen/LiveIntervalAnalysis.cpp</tt> for an example of its
use.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="regAlloc_builtIn">Built in register allocators</a>
</h4>
<div>
<p>The LLVM infrastructure provides the application developer with three
different register allocators:</p>
<ul>
<li><i>Fast</i> — This register allocator is the default for debug
builds. It allocates registers on a basic block level, attempting to keep
values in registers and reusing registers as appropriate.</li>
<li><i>Basic</i> — This is an incremental approach to register
allocation. Live ranges are assigned to registers one at a time in
an order that is driven by heuristics. Since code can be rewritten
on-the-fly during allocation, this framework allows interesting
allocators to be developed as extensions. It is not itself a
production register allocator but is a potentially useful
stand-alone mode for triaging bugs and as a performance baseline.
<li><i>Greedy</i> — <i>The default allocator</i>. This is a
highly tuned implementation of the <i>Basic</i> allocator that
incorporates global live range splitting. This allocator works hard
to minimize the cost of spill code.
<li><i>PBQP</i> — A Partitioned Boolean Quadratic Programming (PBQP)
based register allocator. This allocator works by constructing a PBQP
problem representing the register allocation problem under consideration,
solving this using a PBQP solver, and mapping the solution back to a
register assignment.</li>
</ul>
<p>The type of register allocator used in <tt>llc</tt> can be chosen with the
command line option <tt>-regalloc=...</tt>:</p>
<div class="doc_code">
<pre>
$ llc -regalloc=linearscan file.bc -o ln.s;
$ llc -regalloc=fast file.bc -o fa.s;
$ llc -regalloc=pbqp file.bc -o pbqp.s;
</pre>
</div>
</div>
</div>
<!-- ======================================================================= -->
<h3>
<a name="proepicode">Prolog/Epilog Code Insertion</a>
</h3>
<div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="compact_unwind">Compact Unwind</a>
</h4>
<div>
<p>Throwing an exception requires <em>unwinding</em> out of a function. The
information on how to unwind a given function is traditionally expressed in
DWARF unwind (a.k.a. frame) info. But that format was originally developed
for debuggers to backtrace, and each Frame Description Entry (FDE) requires
~20-30 bytes per function. There is also the cost of mapping from an address
in a function to the corresponding FDE at runtime. An alternative unwind
encoding is called <em>compact unwind</em> and requires just 4-bytes per
function.</p>
<p>The compact unwind encoding is a 32-bit value, which is encoded in an
architecture-specific way. It specifies which registers to restore and from
where, and how to unwind out of the function. When the linker creates a final
linked image, it will create a <code>__TEXT,__unwind_info</code>
section. This section is a small and fast way for the runtime to access
unwind info for any given function. If we emit compact unwind info for the
function, that compact unwind info will be encoded in
the <code>__TEXT,__unwind_info</code> section. If we emit DWARF unwind info,
the <code>__TEXT,__unwind_info</code> section will contain the offset of the
FDE in the <code>__TEXT,__eh_frame</code> section in the final linked
image.</p>
<p>For X86, there are three modes for the compact unwind encoding:</p>
<dl>
<dt><i>Function with a Frame Pointer (<code>EBP</code> or <code>RBP</code>)</i></dt>
<dd><p><code>EBP/RBP</code>-based frame, where <code>EBP/RBP</code> is pushed
onto the stack immediately after the return address,
then <code>ESP/RSP</code> is moved to <code>EBP/RBP</code>. Thus to
unwind, <code>ESP/RSP</code> is restored with the
current <code>EBP/RBP</code> value, then <code>EBP/RBP</code> is restored
by popping the stack, and the return is done by popping the stack once
more into the PC. All non-volatile registers that need to be restored must
have been saved in a small range on the stack that
starts <code>EBP-4</code> to <code>EBP-1020</code> (<code>RBP-8</code>
to <code>RBP-1020</code>). The offset (divided by 4 in 32-bit mode and 8
in 64-bit mode) is encoded in bits 16-23 (mask: <code>0x00FF0000</code>).
The registers saved are encoded in bits 0-14
(mask: <code>0x00007FFF</code>) as five 3-bit entries from the following
table:</p>
<table border="1" cellspacing="0">
<tr>
<th>Compact Number</th>
<th>i386 Register</th>
<th>x86-64 Regiser</th>
</tr>
<tr>
<td>1</td>
<td><code>EBX</code></td>
<td><code>RBX</code></td>
</tr>
<tr>
<td>2</td>
<td><code>ECX</code></td>
<td><code>R12</code></td>
</tr>
<tr>
<td>3</td>
<td><code>EDX</code></td>
<td><code>R13</code></td>
</tr>
<tr>
<td>4</td>
<td><code>EDI</code></td>
<td><code>R14</code></td>
</tr>
<tr>
<td>5</td>
<td><code>ESI</code></td>
<td><code>R15</code></td>
</tr>
<tr>
<td>6</td>
<td><code>EBP</code></td>
<td><code>RBP</code></td>
</tr>
</table>
</dd>
<dt><i>Frameless with a Small Constant Stack Size (<code>EBP</code>
or <code>RBP</code> is not used as a frame pointer)</i></dt>
<dd><p>To return, a constant (encoded in the compact unwind encoding) is added
to the <code>ESP/RSP</code>. Then the return is done by popping the stack
into the PC. All non-volatile registers that need to be restored must have
been saved on the stack immediately after the return address. The stack
size (divided by 4 in 32-bit mode and 8 in 64-bit mode) is encoded in bits
16-23 (mask: <code>0x00FF0000</code>). There is a maximum stack size of
1024 bytes in 32-bit mode and 2048 in 64-bit mode. The number of registers
saved is encoded in bits 9-12 (mask: <code>0x00001C00</code>). Bits 0-9
(mask: <code>0x000003FF</code>) contain which registers were saved and
their order. (See
the <code>encodeCompactUnwindRegistersWithoutFrame()</code> function
in <code>lib/Target/X86FrameLowering.cpp</code> for the encoding
algorithm.)</p></dd>
<dt><i>Frameless with a Large Constant Stack Size (<code>EBP</code>
or <code>RBP</code> is not used as a frame pointer)</i></dt>
<dd><p>This case is like the "Frameless with a Small Constant Stack Size"
case, but the stack size is too large to encode in the compact unwind
encoding. Instead it requires that the function contains "<code>subl
$nnnnnn, %esp</code>" in its prolog. The compact encoding contains the
offset to the <code>$nnnnnn</code> value in the function in bits 9-12
(mask: <code>0x00001C00</code>).</p></dd>
</dl>
</div>
</div>
<!-- ======================================================================= -->
<h3>
<a name="latemco">Late Machine Code Optimizations</a>
</h3>
<div><p>To Be Written</p></div>
<!-- ======================================================================= -->
<h3>
<a name="codeemit">Code Emission</a>
</h3>
<div>
<p>The code emission step of code generation is responsible for lowering from
the code generator abstractions (like <a
href="#machinefunction">MachineFunction</a>, <a
href="#machineinstr">MachineInstr</a>, etc) down
to the abstractions used by the MC layer (<a href="#mcinst">MCInst</a>,
<a href="#mcstreamer">MCStreamer</a>, etc). This is
done with a combination of several different classes: the (misnamed)
target-independent AsmPrinter class, target-specific subclasses of AsmPrinter
(such as SparcAsmPrinter), and the TargetLoweringObjectFile class.</p>
<p>Since the MC layer works at the level of abstraction of object files, it
doesn't have a notion of functions, global variables etc. Instead, it thinks
about labels, directives, and instructions. A key class used at this time is
the MCStreamer class. This is an abstract API that is implemented in different
ways (e.g. to output a .s file, output an ELF .o file, etc) that is effectively
an "assembler API". MCStreamer has one method per directive, such as EmitLabel,
EmitSymbolAttribute, SwitchSection, etc, which directly correspond to assembly
level directives.
</p>
<p>If you are interested in implementing a code generator for a target, there
are three important things that you have to implement for your target:</p>
<ol>
<li>First, you need a subclass of AsmPrinter for your target. This class
implements the general lowering process converting MachineFunction's into MC
label constructs. The AsmPrinter base class provides a number of useful methods
and routines, and also allows you to override the lowering process in some
important ways. You should get much of the lowering for free if you are
implementing an ELF, COFF, or MachO target, because the TargetLoweringObjectFile
class implements much of the common logic.</li>
<li>Second, you need to implement an instruction printer for your target. The
instruction printer takes an <a href="#mcinst">MCInst</a> and renders it to a
raw_ostream as text. Most of this is automatically generated from the .td file
(when you specify something like "<tt>add $dst, $src1, $src2</tt>" in the
instructions), but you need to implement routines to print operands.</li>
<li>Third, you need to implement code that lowers a <a
href="#machineinstr">MachineInstr</a> to an MCInst, usually implemented in
"<target>MCInstLower.cpp". This lowering process is often target
specific, and is responsible for turning jump table entries, constant pool
indices, global variable addresses, etc into MCLabels as appropriate. This
translation layer is also responsible for expanding pseudo ops used by the code
generator into the actual machine instructions they correspond to. The MCInsts
that are generated by this are fed into the instruction printer or the encoder.
</li>
</ol>
<p>Finally, at your choosing, you can also implement an subclass of
MCCodeEmitter which lowers MCInst's into machine code bytes and relocations.
This is important if you want to support direct .o file emission, or would like
to implement an assembler for your target.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="vliw_packetizer">VLIW Packetizer</a>
</h3>
<div>
<p>In a Very Long Instruction Word (VLIW) architecture, the compiler is
responsible for mapping instructions to functional-units available on
the architecture. To that end, the compiler creates groups of instructions
called <i>packets</i> or <i>bundles</i>. The VLIW packetizer in LLVM is
a target-independent mechanism to enable the packetization of machine
instructions.</p>
<!-- _______________________________________________________________________ -->
<h4>
<a name="vliw_mapping">Mapping from instructions to functional units</a>
</h4>
<div>
<p>Instructions in a VLIW target can typically be mapped to multiple functional
units. During the process of packetizing, the compiler must be able to reason
about whether an instruction can be added to a packet. This decision can be
complex since the compiler has to examine all possible mappings of instructions
to functional units. Therefore to alleviate compilation-time complexity, the
VLIW packetizer parses the instruction classes of a target and generates tables
at compiler build time. These tables can then be queried by the provided
machine-independent API to determine if an instruction can be accommodated in a
packet.</p>
</div>
<!-- ======================================================================= -->
<h4>
<a name="vliw_repr">
How the packetization tables are generated and used
</a>
</h4>
<div>
<p>The packetizer reads instruction classes from a target's itineraries and
creates a deterministic finite automaton (DFA) to represent the state of a
packet. A DFA consists of three major elements: inputs, states, and
transitions. The set of inputs for the generated DFA represents the instruction
being added to a packet. The states represent the possible consumption
of functional units by instructions in a packet. In the DFA, transitions from
one state to another occur on the addition of an instruction to an existing
packet. If there is a legal mapping of functional units to instructions, then
the DFA contains a corresponding transition. The absence of a transition
indicates that a legal mapping does not exist and that the instruction cannot
be added to the packet.</p>
<p>To generate tables for a VLIW target, add <i>Target</i>GenDFAPacketizer.inc
as a target to the Makefile in the target directory. The exported API provides
three functions: <tt>DFAPacketizer::clearResources()</tt>,
<tt>DFAPacketizer::reserveResources(MachineInstr *MI)</tt>, and
<tt>DFAPacketizer::canReserveResources(MachineInstr *MI)</tt>. These functions
allow a target packetizer to add an instruction to an existing packet and to
check whether an instruction can be added to a packet. See
<tt>llvm/CodeGen/DFAPacketizer.h</tt> for more information.</p>
</div>
</div>
</div>
<!-- *********************************************************************** -->
<h2>
<a name="nativeassembler">Implementing a Native Assembler</a>
</h2>
<!-- *********************************************************************** -->
<div>
<p>Though you're probably reading this because you want to write or maintain a
compiler backend, LLVM also fully supports building a native assemblers too.
We've tried hard to automate the generation of the assembler from the .td files
(in particular the instruction syntax and encodings), which means that a large
part of the manual and repetitive data entry can be factored and shared with the
compiler.</p>
<!-- ======================================================================= -->
<h3 id="na_instparsing">Instruction Parsing</h3>
<div><p>To Be Written</p></div>
<!-- ======================================================================= -->
<h3 id="na_instaliases">
Instruction Alias Processing
</h3>
<div>
<p>Once the instruction is parsed, it enters the MatchInstructionImpl function.
The MatchInstructionImpl function performs alias processing and then does
actual matching.</p>
<p>Alias processing is the phase that canonicalizes different lexical forms of
the same instructions down to one representation. There are several different
kinds of alias that are possible to implement and they are listed below in the
order that they are processed (which is in order from simplest/weakest to most
complex/powerful). Generally you want to use the first alias mechanism that
meets the needs of your instruction, because it will allow a more concise
description.</p>
<!-- _______________________________________________________________________ -->
<h4>Mnemonic Aliases</h4>
<div>
<p>The first phase of alias processing is simple instruction mnemonic
remapping for classes of instructions which are allowed with two different
mnemonics. This phase is a simple and unconditionally remapping from one input
mnemonic to one output mnemonic. It isn't possible for this form of alias to
look at the operands at all, so the remapping must apply for all forms of a
given mnemonic. Mnemonic aliases are defined simply, for example X86 has:
</p>
<div class="doc_code">
<pre>
def : MnemonicAlias<"cbw", "cbtw">;
def : MnemonicAlias<"smovq", "movsq">;
def : MnemonicAlias<"fldcww", "fldcw">;
def : MnemonicAlias<"fucompi", "fucomip">;
def : MnemonicAlias<"ud2a", "ud2">;
</pre>
</div>
<p>... and many others. With a MnemonicAlias definition, the mnemonic is
remapped simply and directly. Though MnemonicAlias's can't look at any aspect
of the instruction (such as the operands) they can depend on global modes (the
same ones supported by the matcher), through a Requires clause:</p>
<div class="doc_code">
<pre>
def : MnemonicAlias<"pushf", "pushfq">, Requires<[In64BitMode]>;
def : MnemonicAlias<"pushf", "pushfl">, Requires<[In32BitMode]>;
</pre>
</div>
<p>In this example, the mnemonic gets mapped into different a new one depending
on the current instruction set.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>Instruction Aliases</h4>
<div>
<p>The most general phase of alias processing occurs while matching is
happening: it provides new forms for the matcher to match along with a specific
instruction to generate. An instruction alias has two parts: the string to
match and the instruction to generate. For example:
</p>
<div class="doc_code">
<pre>
def : InstAlias<"movsx $src, $dst", (MOVSX16rr8W GR16:$dst, GR8 :$src)>;
def : InstAlias<"movsx $src, $dst", (MOVSX16rm8W GR16:$dst, i8mem:$src)>;
def : InstAlias<"movsx $src, $dst", (MOVSX32rr8 GR32:$dst, GR8 :$src)>;
def : InstAlias<"movsx $src, $dst", (MOVSX32rr16 GR32:$dst, GR16 :$src)>;
def : InstAlias<"movsx $src, $dst", (MOVSX64rr8 GR64:$dst, GR8 :$src)>;
def : InstAlias<"movsx $src, $dst", (MOVSX64rr16 GR64:$dst, GR16 :$src)>;
def : InstAlias<"movsx $src, $dst", (MOVSX64rr32 GR64:$dst, GR32 :$src)>;
</pre>
</div>
<p>This shows a powerful example of the instruction aliases, matching the
same mnemonic in multiple different ways depending on what operands are present
in the assembly. The result of instruction aliases can include operands in a
different order than the destination instruction, and can use an input
multiple times, for example:</p>
<div class="doc_code">
<pre>
def : InstAlias<"clrb $reg", (XOR8rr GR8 :$reg, GR8 :$reg)>;
def : InstAlias<"clrw $reg", (XOR16rr GR16:$reg, GR16:$reg)>;
def : InstAlias<"clrl $reg", (XOR32rr GR32:$reg, GR32:$reg)>;
def : InstAlias<"clrq $reg", (XOR64rr GR64:$reg, GR64:$reg)>;
</pre>
</div>
<p>This example also shows that tied operands are only listed once. In the X86
backend, XOR8rr has two input GR8's and one output GR8 (where an input is tied
to the output). InstAliases take a flattened operand list without duplicates
for tied operands. The result of an instruction alias can also use immediates
and fixed physical registers which are added as simple immediate operands in the
result, for example:</p>
<div class="doc_code">
<pre>
// Fixed Immediate operand.
def : InstAlias<"aad", (AAD8i8 10)>;
// Fixed register operand.
def : InstAlias<"fcomi", (COM_FIr ST1)>;
// Simple alias.
def : InstAlias<"fcomi $reg", (COM_FIr RST:$reg)>;
</pre>
</div>
<p>Instruction aliases can also have a Requires clause to make them
subtarget specific.</p>
<p>If the back-end supports it, the instruction printer can automatically emit
the alias rather than what's being aliased. It typically leads to better,
more readable code. If it's better to print out what's being aliased, then
pass a '0' as the third parameter to the InstAlias definition.</p>
</div>
</div>
<!-- ======================================================================= -->
<h3 id="na_matching">Instruction Matching</h3>
<div><p>To Be Written</p></div>
</div>
<!-- *********************************************************************** -->
<h2>
<a name="targetimpls">Target-specific Implementation Notes</a>
</h2>
<!-- *********************************************************************** -->
<div>
<p>This section of the document explains features or design decisions that are
specific to the code generator for a particular target. First we start
with a table that summarizes what features are supported by each target.</p>
<!-- ======================================================================= -->
<h3>
<a name="targetfeatures">Target Feature Matrix</a>
</h3>
<div>
<p>Note that this table does not include the C backend or Cpp backends, since
they do not use the target independent code generator infrastructure. It also
doesn't list features that are not supported fully by any target yet. It
considers a feature to be supported if at least one subtarget supports it. A
feature being supported means that it is useful and works for most cases, it
does not indicate that there are zero known bugs in the implementation. Here
is the key:</p>
<table border="1" cellspacing="0">
<tr>
<th>Unknown</th>
<th>No support</th>
<th>Partial Support</th>
<th>Complete Support</th>
</tr>
<tr>
<td class="unknown"></td>
<td class="no"></td>
<td class="partial"></td>
<td class="yes"></td>
</tr>
</table>
<p>Here is the table:</p>
<table width="689" border="1" cellspacing="0">
<tr><td></td>
<td colspan="13" align="center" style="background-color:#ffc">Target</td>
</tr>
<tr>
<th>Feature</th>
<th>ARM</th>
<th>CellSPU</th>
<th>Hexagon</th>
<th>MBlaze</th>
<th>MSP430</th>
<th>Mips</th>
<th>PTX</th>
<th>PowerPC</th>
<th>Sparc</th>
<th>X86</th>
<th>XCore</th>
</tr>
<tr>
<td><a href="#feat_reliable">is generally reliable</a></td>
<td class="yes"></td> <!-- ARM -->
<td class="no"></td> <!-- CellSPU -->
<td class="yes"></td> <!-- Hexagon -->
<td class="no"></td> <!-- MBlaze -->
<td class="unknown"></td> <!-- MSP430 -->
<td class="yes"></td> <!-- Mips -->
<td class="no"></td> <!-- PTX -->
<td class="yes"></td> <!-- PowerPC -->
<td class="yes"></td> <!-- Sparc -->
<td class="yes"></td> <!-- X86 -->
<td class="unknown"></td> <!-- XCore -->
</tr>
<tr>
<td><a href="#feat_asmparser">assembly parser</a></td>
<td class="no"></td> <!-- ARM -->
<td class="no"></td> <!-- CellSPU -->
<td class="no"></td> <!-- Hexagon -->
<td class="yes"></td> <!-- MBlaze -->
<td class="no"></td> <!-- MSP430 -->
<td class="no"></td> <!-- Mips -->
<td class="no"></td> <!-- PTX -->
<td class="no"></td> <!-- PowerPC -->
<td class="no"></td> <!-- Sparc -->
<td class="yes"></td> <!-- X86 -->
<td class="no"></td> <!-- XCore -->
</tr>
<tr>
<td><a href="#feat_disassembler">disassembler</a></td>
<td class="yes"></td> <!-- ARM -->
<td class="no"></td> <!-- CellSPU -->
<td class="no"></td> <!-- Hexagon -->
<td class="yes"></td> <!-- MBlaze -->
<td class="no"></td> <!-- MSP430 -->
<td class="no"></td> <!-- Mips -->
<td class="no"></td> <!-- PTX -->
<td class="no"></td> <!-- PowerPC -->
<td class="no"></td> <!-- Sparc -->
<td class="yes"></td> <!-- X86 -->
<td class="no"></td> <!-- XCore -->
</tr>
<tr>
<td><a href="#feat_inlineasm">inline asm</a></td>
<td class="yes"></td> <!-- ARM -->
<td class="no"></td> <!-- CellSPU -->
<td class="yes"></td> <!-- Hexagon -->
<td class="yes"></td> <!-- MBlaze -->
<td class="unknown"></td> <!-- MSP430 -->
<td class="no"></td> <!-- Mips -->
<td class="unknown"></td> <!-- PTX -->
<td class="yes"></td> <!-- PowerPC -->
<td class="unknown"></td> <!-- Sparc -->
<td class="yes"></td> <!-- X86 -->
<td class="unknown"></td> <!-- XCore -->
</tr>
<tr>
<td><a href="#feat_jit">jit</a></td>
<td class="partial"><a href="#feat_jit_arm">*</a></td> <!-- ARM -->
<td class="no"></td> <!-- CellSPU -->
<td class="no"></td> <!-- Hexagon -->
<td class="no"></td> <!-- MBlaze -->
<td class="unknown"></td> <!-- MSP430 -->
<td class="yes"></td> <!-- Mips -->
<td class="unknown"></td> <!-- PTX -->
<td class="yes"></td> <!-- PowerPC -->
<td class="unknown"></td> <!-- Sparc -->
<td class="yes"></td> <!-- X86 -->
<td class="unknown"></td> <!-- XCore -->
</tr>
<tr>
<td><a href="#feat_objectwrite">.o file writing</a></td>
<td class="no"></td> <!-- ARM -->
<td class="no"></td> <!-- CellSPU -->
<td class="no"></td> <!-- Hexagon -->
<td class="yes"></td> <!-- MBlaze -->
<td class="no"></td> <!-- MSP430 -->
<td class="no"></td> <!-- Mips -->
<td class="no"></td> <!-- PTX -->
<td class="no"></td> <!-- PowerPC -->
<td class="no"></td> <!-- Sparc -->
<td class="yes"></td> <!-- X86 -->
<td class="no"></td> <!-- XCore -->
</tr>
<tr>
<td><a href="#feat_tailcall">tail calls</a></td>
<td class="yes"></td> <!-- ARM -->
<td class="no"></td> <!-- CellSPU -->
<td class="yes"></td> <!-- Hexagon -->
<td class="no"></td> <!-- MBlaze -->
<td class="unknown"></td> <!-- MSP430 -->
<td class="no"></td> <!-- Mips -->
<td class="unknown"></td> <!-- PTX -->
<td class="yes"></td> <!-- PowerPC -->
<td class="unknown"></td> <!-- Sparc -->
<td class="yes"></td> <!-- X86 -->
<td class="unknown"></td> <!-- XCore -->
</tr>
<tr>
<td><a href="#feat_segstacks">segmented stacks</a></td>
<td class="no"></td> <!-- ARM -->
<td class="no"></td> <!-- CellSPU -->
<td class="no"></td> <!-- Hexagon -->
<td class="no"></td> <!-- MBlaze -->
<td class="no"></td> <!-- MSP430 -->
<td class="no"></td> <!-- Mips -->
<td class="no"></td> <!-- PTX -->
<td class="no"></td> <!-- PowerPC -->
<td class="no"></td> <!-- Sparc -->
<td class="partial"><a href="#feat_segstacks_x86">*</a></td> <!-- X86 -->
<td class="no"></td> <!-- XCore -->
</tr>
</table>
<!-- _______________________________________________________________________ -->
<h4 id="feat_reliable">Is Generally Reliable</h4>
<div>
<p>This box indicates whether the target is considered to be production quality.
This indicates that the target has been used as a static compiler to
compile large amounts of code by a variety of different people and is in
continuous use.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4 id="feat_asmparser">Assembly Parser</h4>
<div>
<p>This box indicates whether the target supports parsing target specific .s
files by implementing the MCAsmParser interface. This is required for llvm-mc
to be able to act as a native assembler and is required for inline assembly
support in the native .o file writer.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4 id="feat_disassembler">Disassembler</h4>
<div>
<p>This box indicates whether the target supports the MCDisassembler API for
disassembling machine opcode bytes into MCInst's.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4 id="feat_inlineasm">Inline Asm</h4>
<div>
<p>This box indicates whether the target supports most popular inline assembly
constraints and modifiers.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4 id="feat_jit">JIT Support</h4>
<div>
<p>This box indicates whether the target supports the JIT compiler through
the ExecutionEngine interface.</p>
<p id="feat_jit_arm">The ARM backend has basic support for integer code
in ARM codegen mode, but lacks NEON and full Thumb support.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4 id="feat_objectwrite">.o File Writing</h4>
<div>
<p>This box indicates whether the target supports writing .o files (e.g. MachO,
ELF, and/or COFF) files directly from the target. Note that the target also
must include an assembly parser and general inline assembly support for full
inline assembly support in the .o writer.</p>
<p>Targets that don't support this feature can obviously still write out .o
files, they just rely on having an external assembler to translate from a .s
file to a .o file (as is the case for many C compilers).</p>
</div>
<!-- _______________________________________________________________________ -->
<h4 id="feat_tailcall">Tail Calls</h4>
<div>
<p>This box indicates whether the target supports guaranteed tail calls. These
are calls marked "<a href="LangRef.html#i_call">tail</a>" and use the fastcc
calling convention. Please see the <a href="#tailcallopt">tail call section
more more details</a>.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4 id="feat_segstacks">Segmented Stacks</h4>
<div>
<p>This box indicates whether the target supports segmented stacks. This
replaces the traditional large C stack with many linked segments. It
is compatible with the <a href="http://gcc.gnu.org/wiki/SplitStacks">gcc
implementation</a> used by the Go front end.</p>
<p id="feat_segstacks_x86">Basic support exists on the X86 backend. Currently
vararg doesn't work and the object files are not marked the way the gold
linker expects, but simple Go programs can be built by dragonegg.</p>
</div>
</div>
<!-- ======================================================================= -->
<h3>
<a name="tailcallopt">Tail call optimization</a>
</h3>
<div>
<p>Tail call optimization, callee reusing the stack of the caller, is currently
supported on x86/x86-64 and PowerPC. It is performed if:</p>
<ul>
<li>Caller and callee have the calling convention <tt>fastcc</tt> or
<tt>cc 10</tt> (GHC call convention).</li>
<li>The call is a tail call - in tail position (ret immediately follows call
and ret uses value of call or is void).</li>
<li>Option <tt>-tailcallopt</tt> is enabled.</li>
<li>Platform specific constraints are met.</li>
</ul>
<p>x86/x86-64 constraints:</p>
<ul>
<li>No variable argument lists are used.</li>
<li>On x86-64 when generating GOT/PIC code only module-local calls (visibility
= hidden or protected) are supported.</li>
</ul>
<p>PowerPC constraints:</p>
<ul>
<li>No variable argument lists are used.</li>
<li>No byval parameters are used.</li>
<li>On ppc32/64 GOT/PIC only module-local calls (visibility = hidden or protected) are supported.</li>
</ul>
<p>Example:</p>
<p>Call as <tt>llc -tailcallopt test.ll</tt>.</p>
<div class="doc_code">
<pre>
declare fastcc i32 @tailcallee(i32 inreg %a1, i32 inreg %a2, i32 %a3, i32 %a4)
define fastcc i32 @tailcaller(i32 %in1, i32 %in2) {
%l1 = add i32 %in1, %in2
%tmp = tail call fastcc i32 @tailcallee(i32 %in1 inreg, i32 %in2 inreg, i32 %in1, i32 %l1)
ret i32 %tmp
}
</pre>
</div>
<p>Implications of <tt>-tailcallopt</tt>:</p>
<p>To support tail call optimization in situations where the callee has more
arguments than the caller a 'callee pops arguments' convention is used. This
currently causes each <tt>fastcc</tt> call that is not tail call optimized
(because one or more of above constraints are not met) to be followed by a
readjustment of the stack. So performance might be worse in such cases.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="sibcallopt">Sibling call optimization</a>
</h3>
<div>
<p>Sibling call optimization is a restricted form of tail call optimization.
Unlike tail call optimization described in the previous section, it can be
performed automatically on any tail calls when <tt>-tailcallopt</tt> option
is not specified.</p>
<p>Sibling call optimization is currently performed on x86/x86-64 when the
following constraints are met:</p>
<ul>
<li>Caller and callee have the same calling convention. It can be either
<tt>c</tt> or <tt>fastcc</tt>.
<li>The call is a tail call - in tail position (ret immediately follows call
and ret uses value of call or is void).</li>
<li>Caller and callee have matching return type or the callee result is not
used.
<li>If any of the callee arguments are being passed in stack, they must be
available in caller's own incoming argument stack and the frame offsets
must be the same.
</ul>
<p>Example:</p>
<div class="doc_code">
<pre>
declare i32 @bar(i32, i32)
define i32 @foo(i32 %a, i32 %b, i32 %c) {
entry:
%0 = tail call i32 @bar(i32 %a, i32 %b)
ret i32 %0
}
</pre>
</div>
</div>
<!-- ======================================================================= -->
<h3>
<a name="x86">The X86 backend</a>
</h3>
<div>
<p>The X86 code generator lives in the <tt>lib/Target/X86</tt> directory. This
code generator is capable of targeting a variety of x86-32 and x86-64
processors, and includes support for ISA extensions such as MMX and SSE.</p>
<!-- _______________________________________________________________________ -->
<h4>
<a name="x86_tt">X86 Target Triples supported</a>
</h4>
<div>
<p>The following are the known target triples that are supported by the X86
backend. This is not an exhaustive list, and it would be useful to add those
that people test.</p>
<ul>
<li><b>i686-pc-linux-gnu</b> — Linux</li>
<li><b>i386-unknown-freebsd5.3</b> — FreeBSD 5.3</li>
<li><b>i686-pc-cygwin</b> — Cygwin on Win32</li>
<li><b>i686-pc-mingw32</b> — MingW on Win32</li>
<li><b>i386-pc-mingw32msvc</b> — MingW crosscompiler on Linux</li>
<li><b>i686-apple-darwin*</b> — Apple Darwin on X86</li>
<li><b>x86_64-unknown-linux-gnu</b> — Linux</li>
</ul>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="x86_cc">X86 Calling Conventions supported</a>
</h4>
<div>
<p>The following target-specific calling conventions are known to backend:</p>
<ul>
<li><b>x86_StdCall</b> — stdcall calling convention seen on Microsoft
Windows platform (CC ID = 64).</li>
<li><b>x86_FastCall</b> — fastcall calling convention seen on Microsoft
Windows platform (CC ID = 65).</li>
<li><b>x86_ThisCall</b> — Similar to X86_StdCall. Passes first argument
in ECX, others via stack. Callee is responsible for stack cleaning. This
convention is used by MSVC by default for methods in its ABI
(CC ID = 70).</li>
</ul>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="x86_memory">Representing X86 addressing modes in MachineInstrs</a>
</h4>
<div>
<p>The x86 has a very flexible way of accessing memory. It is capable of
forming memory addresses of the following expression directly in integer
instructions (which use ModR/M addressing):</p>
<div class="doc_code">
<pre>
SegmentReg: Base + [1,2,4,8] * IndexReg + Disp32
</pre>
</div>
<p>In order to represent this, LLVM tracks no less than 5 operands for each
memory operand of this form. This means that the "load" form of
'<tt>mov</tt>' has the following <tt>MachineOperand</tt>s in this order:</p>
<div class="doc_code">
<pre>
Index: 0 | 1 2 3 4 5
Meaning: DestReg, | BaseReg, Scale, IndexReg, Displacement Segment
OperandTy: VirtReg, | VirtReg, UnsImm, VirtReg, SignExtImm PhysReg
</pre>
</div>
<p>Stores, and all other instructions, treat the four memory operands in the
same way and in the same order. If the segment register is unspecified
(regno = 0), then no segment override is generated. "Lea" operations do not
have a segment register specified, so they only have 4 operands for their
memory reference.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="x86_memory">X86 address spaces supported</a>
</h4>
<div>
<p>x86 has a feature which provides
the ability to perform loads and stores to different address spaces
via the x86 segment registers. A segment override prefix byte on an
instruction causes the instruction's memory access to go to the specified
segment. LLVM address space 0 is the default address space, which includes
the stack, and any unqualified memory accesses in a program. Address spaces
1-255 are currently reserved for user-defined code. The GS-segment is
represented by address space 256, while the FS-segment is represented by
address space 257. Other x86 segments have yet to be allocated address space
numbers.</p>
<p>While these address spaces may seem similar to TLS via the
<tt>thread_local</tt> keyword, and often use the same underlying hardware,
there are some fundamental differences.</p>
<p>The <tt>thread_local</tt> keyword applies to global variables and
specifies that they are to be allocated in thread-local memory. There are
no type qualifiers involved, and these variables can be pointed to with
normal pointers and accessed with normal loads and stores.
The <tt>thread_local</tt> keyword is target-independent at the LLVM IR
level (though LLVM doesn't yet have implementations of it for some
configurations).<p>
<p>Special address spaces, in contrast, apply to static types. Every
load and store has a particular address space in its address operand type,
and this is what determines which address space is accessed.
LLVM ignores these special address space qualifiers on global variables,
and does not provide a way to directly allocate storage in them.
At the LLVM IR level, the behavior of these special address spaces depends
in part on the underlying OS or runtime environment, and they are specific
to x86 (and LLVM doesn't yet handle them correctly in some cases).</p>
<p>Some operating systems and runtime environments use (or may in the future
use) the FS/GS-segment registers for various low-level purposes, so care
should be taken when considering them.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="x86_names">Instruction naming</a>
</h4>
<div>
<p>An instruction name consists of the base name, a default operand size, and a
a character per operand with an optional special size. For example:</p>
<div class="doc_code">
<pre>
ADD8rr -> add, 8-bit register, 8-bit register
IMUL16rmi -> imul, 16-bit register, 16-bit memory, 16-bit immediate
IMUL16rmi8 -> imul, 16-bit register, 16-bit memory, 8-bit immediate
MOVSX32rm16 -> movsx, 32-bit register, 16-bit memory
</pre>
</div>
</div>
</div>
<!-- ======================================================================= -->
<h3>
<a name="ppc">The PowerPC backend</a>
</h3>
<div>
<p>The PowerPC code generator lives in the lib/Target/PowerPC directory. The
code generation is retargetable to several variations or <i>subtargets</i> of
the PowerPC ISA; including ppc32, ppc64 and altivec.</p>
<!-- _______________________________________________________________________ -->
<h4>
<a name="ppc_abi">LLVM PowerPC ABI</a>
</h4>
<div>
<p>LLVM follows the AIX PowerPC ABI, with two deviations. LLVM uses a PC
relative (PIC) or static addressing for accessing global values, so no TOC
(r2) is used. Second, r31 is used as a frame pointer to allow dynamic growth
of a stack frame. LLVM takes advantage of having no TOC to provide space to
save the frame pointer in the PowerPC linkage area of the caller frame.
Other details of PowerPC ABI can be found at <a href=
"http://developer.apple.com/documentation/DeveloperTools/Conceptual/LowLevelABI/Articles/32bitPowerPC.html"
>PowerPC ABI.</a> Note: This link describes the 32 bit ABI. The 64 bit ABI
is similar except space for GPRs are 8 bytes wide (not 4) and r13 is reserved
for system use.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="ppc_frame">Frame Layout</a>
</h4>
<div>
<p>The size of a PowerPC frame is usually fixed for the duration of a
function's invocation. Since the frame is fixed size, all references
into the frame can be accessed via fixed offsets from the stack pointer. The
exception to this is when dynamic alloca or variable sized arrays are
present, then a base pointer (r31) is used as a proxy for the stack pointer
and stack pointer is free to grow or shrink. A base pointer is also used if
llvm-gcc is not passed the -fomit-frame-pointer flag. The stack pointer is
always aligned to 16 bytes, so that space allocated for altivec vectors will
be properly aligned.</p>
<p>An invocation frame is laid out as follows (low memory at top);</p>
<table class="layout">
<tr>
<td>Linkage<br><br></td>
</tr>
<tr>
<td>Parameter area<br><br></td>
</tr>
<tr>
<td>Dynamic area<br><br></td>
</tr>
<tr>
<td>Locals area<br><br></td>
</tr>
<tr>
<td>Saved registers area<br><br></td>
</tr>
<tr style="border-style: none hidden none hidden;">
<td><br></td>
</tr>
<tr>
<td>Previous Frame<br><br></td>
</tr>
</table>
<p>The <i>linkage</i> area is used by a callee to save special registers prior
to allocating its own frame. Only three entries are relevant to LLVM. The
first entry is the previous stack pointer (sp), aka link. This allows
probing tools like gdb or exception handlers to quickly scan the frames in
the stack. A function epilog can also use the link to pop the frame from the
stack. The third entry in the linkage area is used to save the return
address from the lr register. Finally, as mentioned above, the last entry is
used to save the previous frame pointer (r31.) The entries in the linkage
area are the size of a GPR, thus the linkage area is 24 bytes long in 32 bit
mode and 48 bytes in 64 bit mode.</p>
<p>32 bit linkage area</p>
<table class="layout">
<tr>
<td>0</td>
<td>Saved SP (r1)</td>
</tr>
<tr>
<td>4</td>
<td>Saved CR</td>
</tr>
<tr>
<td>8</td>
<td>Saved LR</td>
</tr>
<tr>
<td>12</td>
<td>Reserved</td>
</tr>
<tr>
<td>16</td>
<td>Reserved</td>
</tr>
<tr>
<td>20</td>
<td>Saved FP (r31)</td>
</tr>
</table>
<p>64 bit linkage area</p>
<table class="layout">
<tr>
<td>0</td>
<td>Saved SP (r1)</td>
</tr>
<tr>
<td>8</td>
<td>Saved CR</td>
</tr>
<tr>
<td>16</td>
<td>Saved LR</td>
</tr>
<tr>
<td>24</td>
<td>Reserved</td>
</tr>
<tr>
<td>32</td>
<td>Reserved</td>
</tr>
<tr>
<td>40</td>
<td>Saved FP (r31)</td>
</tr>
</table>
<p>The <i>parameter area</i> is used to store arguments being passed to a callee
function. Following the PowerPC ABI, the first few arguments are actually
passed in registers, with the space in the parameter area unused. However,
if there are not enough registers or the callee is a thunk or vararg
function, these register arguments can be spilled into the parameter area.
Thus, the parameter area must be large enough to store all the parameters for
the largest call sequence made by the caller. The size must also be
minimally large enough to spill registers r3-r10. This allows callees blind
to the call signature, such as thunks and vararg functions, enough space to
cache the argument registers. Therefore, the parameter area is minimally 32
bytes (64 bytes in 64 bit mode.) Also note that since the parameter area is
a fixed offset from the top of the frame, that a callee can access its spilt
arguments using fixed offsets from the stack pointer (or base pointer.)</p>
<p>Combining the information about the linkage, parameter areas and alignment. A
stack frame is minimally 64 bytes in 32 bit mode and 128 bytes in 64 bit
mode.</p>
<p>The <i>dynamic area</i> starts out as size zero. If a function uses dynamic
alloca then space is added to the stack, the linkage and parameter areas are
shifted to top of stack, and the new space is available immediately below the
linkage and parameter areas. The cost of shifting the linkage and parameter
areas is minor since only the link value needs to be copied. The link value
can be easily fetched by adding the original frame size to the base pointer.
Note that allocations in the dynamic space need to observe 16 byte
alignment.</p>
<p>The <i>locals area</i> is where the llvm compiler reserves space for local
variables.</p>
<p>The <i>saved registers area</i> is where the llvm compiler spills callee
saved registers on entry to the callee.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="ppc_prolog">Prolog/Epilog</a>
</h4>
<div>
<p>The llvm prolog and epilog are the same as described in the PowerPC ABI, with
the following exceptions. Callee saved registers are spilled after the frame
is created. This allows the llvm epilog/prolog support to be common with
other targets. The base pointer callee saved register r31 is saved in the
TOC slot of linkage area. This simplifies allocation of space for the base
pointer and makes it convenient to locate programatically and during
debugging.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="ppc_dynamic">Dynamic Allocation</a>
</h4>
<div>
<p><i>TODO - More to come.</i></p>
</div>
</div>
<!-- ======================================================================= -->
<h3>
<a name="ptx">The PTX backend</a>
</h3>
<div>
<p>The PTX code generator lives in the lib/Target/PTX directory. It is
currently a work-in-progress, but already supports most of the code
generation functionality needed to generate correct PTX kernels for
CUDA devices.</p>
<p>The code generator can target PTX 2.0+, and shader model 1.0+. The
PTX ISA Reference Manual is used as the primary source of ISA
information, though an effort is made to make the output of the code
generator match the output of the NVidia nvcc compiler, whenever
possible.</p>
<p>Code Generator Options:</p>
<table border="1" cellspacing="0">
<tr>
<th>Option</th>
<th>Description</th>
</tr>
<tr>
<td><code>double</code></td>
<td align="left">If enabled, the map_f64_to_f32 directive is
disabled in the PTX output, allowing native double-precision
arithmetic</td>
</tr>
<tr>
<td><code>no-fma</code></td>
<td align="left">Disable generation of Fused-Multiply Add
instructions, which may be beneficial for some devices</td>
</tr>
<tr>
<td><code>smxy / computexy</code></td>
<td align="left">Set shader model/compute capability to x.y,
e.g. sm20 or compute13</td>
</tr>
</table>
<p>Working:</p>
<ul>
<li>Arithmetic instruction selection (including combo FMA)</li>
<li>Bitwise instruction selection</li>
<li>Control-flow instruction selection</li>
<li>Function calls (only on SM 2.0+ and no return arguments)</li>
<li>Addresses spaces (0 = global, 1 = constant, 2 = local, 4 =
shared)</li>
<li>Thread synchronization (bar.sync)</li>
<li>Special register reads ([N]TID, [N]CTAID, PMx, CLOCK, etc.)</li>
</ul>
<p>In Progress:</p>
<ul>
<li>Robust call instruction selection</li>
<li>Stack frame allocation</li>
<li>Device-specific instruction scheduling optimizations</li>
</ul>
</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="mailto:sabre@nondot.org">Chris Lattner</a><br>
<a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
Last modified: $Date: 2012-04-15 13:22:36 -0700 (Sun, 15 Apr 2012) $
</address>
</body>
</html>
