llvm/docs/WritingAnLLVMBackend.html - toolchain/llvm-project - Gitiles

 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
                       "http://www.w3.org/TR/html4/strict.dtd">
 <html>
 <head>
   <title>Writing an LLVM backend</title>
   <link rel="stylesheet" href="llvm.css" type="text/css">
 </head>

 <body>

 <div class="doc_title">
   Writing an LLVM backend
 </div>

 <ol>
   <li><a href="#intro">Introduction</a>
   <li><a href="#backends">Writing a backend</a>
   <ol>
     <li><a href="#machine">Machine backends</a>
     <ol>
       <li><a href="#machineTOC">Outline</a></li>
       <li><a href="#machineDetails">Implementation details</a></li>
     </ol></li>
     <li><a href="#lang">Language backends</a></li>
   </ol></li>
   <li><a href="#related">Related reading material</a>
 </ol>

 <div class="doc_author">
   <p>Written by <a href="http://misha.brukman.net">Misha Brukman</a></p>
 </div>

 <!-- *********************************************************************** -->
 <div class="doc_section">
   <a name="intro">Introduction</a>
 </div>
 <!-- *********************************************************************** -->

 <div class="doc_text">

 <p>This document describes techniques for writing backends for LLVM which
 convert the LLVM representation to machine assembly code or other languages.</p>

 </div>

 <!-- *********************************************************************** -->
 <div class="doc_section">
   <a name="backends">Writing a backend</a>
 </div>
 <!-- *********************************************************************** -->

 <!-- ======================================================================= -->
 <div class="doc_subsection">
   <a name="machine">Machine backends</a>
 </div>

 <!-- _______________________________________________________________________ -->
 <div class="doc_subsubsection">
   <a name="machineTOC">Outline</a>
 </div>

 <div class="doc_text">

 <p>In general, you want to follow the format of SPARC, X86 or PowerPC (in
 <tt>lib/Target</tt>).  SPARC is the simplest backend, and is RISC, so if
 you're working on a RISC target, it is a good one to start with.</p>

 <p>To create a static compiler (one that emits text assembly), you need to
 implement the following:</p>

 <ul>
 <li>Describe the register set.
   <ul>
   <li>Create a <a href="TableGenFundamentals.html">TableGen</a> description of
       the register set and register classes</li>
   <li>Implement a subclass of <tt><a
       href="CodeGenerator.html#targetregisterinfo">TargetRegisterInfo</a></tt></li>
   </ul></li>
 <li>Describe the instruction set.
   <ul>
   <li>Create a <a href="TableGenFundamentals.html">TableGen</a> description of
       the instruction set</li>
   <li>Implement a subclass of <tt><a
       href="CodeGenerator.html#targetinstrinfo">TargetInstrInfo</a></tt></li>
   </ul></li>
 <li>Describe the target machine.
   <ul>
   <li>Create a <a href="TableGenFundamentals.html">TableGen</a> description of
       the target that describes the pointer size and references the instruction
       set</li>
   <li>Implement a subclass of <tt><a
       href="CodeGenerator.html#targetmachine">TargetMachine</a></tt>, which
       configures <tt><a href="CodeGenerator.html#targetdata">TargetData</a></tt>
       correctly</li>
   <li>Register your new target using the <tt>RegisterTarget</tt>
   template:<br><br>
 <div class="doc_code"><pre>
 RegisterTarget&lt;<em>MyTargetMachine</em>&gt; M("short_name", "  Target name");
 </pre></div>
       <br>Here, <em>MyTargetMachine</em> is the name of your implemented
       subclass of <tt><a
       href="CodeGenerator.html#targetmachine">TargetMachine</a></tt>,
       <em>short_name</em> is the option that will be active following
       <tt>-march=</tt> to select a target in llc and lli, and the last string
       is the description of your target to appear in <tt>-help</tt>
       listing.</li>
   </ul></li>
 <li>Implement the assembly printer for the architecture.
   <ul>
   <li>Define all of the assembly strings for your target, adding them to the
       instructions in your *InstrInfo.td file.</li>
   <li>Implement the <tt>llvm::AsmPrinter</tt> interface.</li>
   </ul>
 </li>
 <li>Implement an instruction selector for the architecture.
   <ul>
   <li>The recommended method is the <a href="CodeGenerator.html#instselect">
       pattern-matching DAG-to-DAG instruction selector</a> (for example, see
       the PowerPC backend in PPCISelDAGtoDAG.cpp).  Parts of instruction
       selector creation can be performed by adding patterns to the instructions
       in your <tt>.td</tt> file.</li>
   </ul>
 </li>
 <li>Optionally, add subtarget support.
 <ul>
   <li>If your target has multiple subtargets (e.g. variants with different
       capabilities), implement the <tt>llvm::TargetSubtarget</tt> interface
       for your architecture.  This allows you to add <tt>-mcpu=</tt> and
       <tt>-mattr=</tt> options.</li>
 </ul>
 <li>Optionally, add JIT support.
   <ul>
   <li>Create a subclass of <tt><a
       href="CodeGenerator.html#targetjitinfo">TargetJITInfo</a></tt></li>
   <li>Create a machine code emitter that will be used to emit binary code
       directly into memory, given <tt>MachineInstr</tt>s</li>
   </ul>
 </ul>
 </div>

 <!-- _______________________________________________________________________ -->
 <div class="doc_subsubsection">
   <a name="machineDetails">Implementation details</a>
 </div>

 <div class="doc_text">

 <ul>

 <li><p><b>TableGen register info description</b> - describe a class which
 will store the register's number in the binary encoding of the instruction
 (e.g., for JIT purposes).</p>

 <p>You also need to define register classes to contain these registers, such as
 the integer register class and floating-point register class, so that you can
 allocate virtual registers to instructions from these sets, and let the
 target-independent register allocator automatically choose the actual
 architected registers.</p>

 <div class="doc_code">
 <pre>
 // class Register is defined in Target.td
 <b>class</b> <em>Target</em>Reg&lt;string name&gt; : Register&lt;name&gt; {
   <b>let</b> Namespace = "<em>Target</em>";
 }

 <b>class</b> IntReg&lt;<b>bits</b>&lt;5&gt; num, string name&gt; : <em>Target</em>Reg&lt;name&gt; {
   <b>field</b> <b>bits</b>&lt;5&gt; Num = num;
 }

 <b>def</b> R0 : IntReg&lt;0, "%R0"&gt;;
 ...

 // class RegisterClass is defined in Target.td
 <b>def</b> IReg : RegisterClass&lt;i64, 64, [R0, ... ]&gt;;
 </pre>
 </div>
 </li>

 <li><p><b>TableGen instruction info description</b> - break up instructions into
 classes, usually that's already done by the manufacturer (see instruction
 manual).  Define a class for each instruction category.  Define each opcode as a
 subclass of the category, with appropriate parameters such as the fixed binary
 encoding of opcodes and extended opcodes, and map the register bits to the bits
 of the instruction which they are encoded in (for the JIT).  Also specify how
 the instruction should be printed so it can use the automatic assembly printer,
 e.g.:</p>

 <div class="doc_code">
 <pre>
 // class Instruction is defined in Target.td
 <b>class</b> Form&lt;<b>bits</b>&lt;6&gt; opcode, <b>dag</b> OL, <b>string</b> asmstr&gt; : Instruction {
   <b>field</b> <b>bits</b>&lt;42&gt; Inst;

   <b>let</b> Namespace = "<em>Target</em>";
   <b>let</b> Inst{0-6} = opcode;
   <b>let</b> OperandList = OL;
   <b>let</b> AsmString = asmstr;
 }

 <b>def</b> ADD : Form&lt;42, (ops IReg:$rD, IReg:$rA, IReg:$rB), "add $rD, $rA, $rB"&gt;;
 </pre>
 </div>
 </li>

 </ul>

 </div>

 <!-- ======================================================================= -->
 <div class="doc_subsection">
   <a name="lang">Language backends</a>
 </div>

 <div class="doc_text">

 <p>For now, just take a look at <tt>lib/Target/CBackend</tt> for an example of
 how the C backend is written.</p>

 </div>
 <!-- ======================================================================= -->
 <div class="doc_subsection">
   <a name="files">Files to create/modify</a>
 </div>

 <div class="doc_text">

 <p>To actually create your backend, you need to create and modify a few files.
 Here, the absolute minimum will be discussed. To actually use LLVM's target
 independent codegenerator, you must <a href="CodeGenerator.html">implement extra
 things</a>.</p>

 <p>First of all, you should create a subdirectory under <tt>lib/Target</tt>,
 which will hold all the files related to your target. Let's assume that our
 target is called, "Dummy", we would create the directory
 <tt>lib/Target/Dummy</tt>.</p>

 <p>In this new directory, you should put a <tt>Makefile</tt>. You can probably
 copy one from another target and modify it. It should at least contain the
 <tt>LEVEL</tt>, <tt>LIBRARYNAME</tt> and <tt>TARGET</tt> variables, and then
 include <tt>$(LEVEL)/Makefile.common</tt>. Be careful to give the library the
 correct name, it must be named <tt>LLVMDummy</tt> (see the MIPS target, for
 example). Alternatively, you can split the library into
 <tt>LLVMDummyCodeGen</tt> and <tt>LLVMDummyAsmPrinter</tt>, the latter of which
 should be implemented in a subdirectory below <tt>lib/Target/Dummy</tt> (see the
 PowerPC target, for example).</p>

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

 <p>To make your target actually do something, you need to implement a subclass
 of <tt>TargetMachine</tt>. This implementation should typically be in the file
 <tt>lib/Target/DummyTargetMachine.cpp</tt>, but any file in the
 <tt>lib/Target</tt> directory will be built and should work. To use LLVM's <a
 href="CodeGenerator.html">target independent code generator</a>, you should
 create a subclass of <tt>LLVMTargetMachine</tt>. This is what all current
 machine backends do. To create a target from scratch, create a subclass of
 <tt>TargetMachine</tt>. This is what the current language backends do.</p>

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

 </div>

 <!-- *********************************************************************** -->
 <div class="doc_section">
   <a name="related">Related reading material</a>
 </div>
 <!-- *********************************************************************** -->

 <div class="doc_text">

 <ul>
 <li><a href="CodeGenerator.html">Code generator</a> -
     describes some of the classes in code generation at a high level, but
     it is not (yet) complete</li>
 <li><a href="TableGenFundamentals.html">TableGen fundamentals</a> -
     describes how to use TableGen to describe your target information
     succinctly</li>
 <li><a href="HowToSubmitABug.html#codegen">Debugging code generation with
     bugpoint</a> - shows bugpoint usage scenarios to simplify backend
     development</li>
 </ul>

 </div>

 <!-- *********************************************************************** -->

 <hr>
 <address>
   <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
   src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
   <a href="http://validator.w3.org/check/referer"><img
   src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>

   <a href="http://misha.brukman.net">Misha Brukman</a><br>
   <a href="http://llvm.org">The LLVM Compiler Infrastructure</a>
   <br>
   Last modified: $Date$
 </address>

 </body>
 </html>
	<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
	"http://www.w3.org/TR/html4/strict.dtd">
	<html>
	<head>
	<title>Writing an LLVM backend</title>
	<link rel="stylesheet" href="llvm.css" type="text/css">
	</head>

	<body>

	<div class="doc_title">
	Writing an LLVM backend
	</div>

	<ol>
	<li><a href="#intro">Introduction</a>
	<li><a href="#backends">Writing a backend</a>
	<ol>
	<li><a href="#machine">Machine backends</a>
	<ol>
	<li><a href="#machineTOC">Outline</a></li>
	<li><a href="#machineDetails">Implementation details</a></li>
	</ol></li>
	<li><a href="#lang">Language backends</a></li>
	</ol></li>
	<li><a href="#related">Related reading material</a>
	</ol>

	<div class="doc_author">
	<p>Written by <a href="http://misha.brukman.net">Misha Brukman</a></p>
	</div>

	<!-- *********************************************************************** -->
	<div class="doc_section">
	<a name="intro">Introduction</a>
	</div>
	<!-- *********************************************************************** -->

	<div class="doc_text">

	<p>This document describes techniques for writing backends for LLVM which
	convert the LLVM representation to machine assembly code or other languages.</p>

	</div>

	<!-- *********************************************************************** -->
	<div class="doc_section">
	<a name="backends">Writing a backend</a>
	</div>
	<!-- *********************************************************************** -->

	<!-- ======================================================================= -->
	<div class="doc_subsection">
	<a name="machine">Machine backends</a>
	</div>

	<!-- _______________________________________________________________________ -->
	<div class="doc_subsubsection">
	<a name="machineTOC">Outline</a>
	</div>

	<div class="doc_text">

	<p>In general, you want to follow the format of SPARC, X86 or PowerPC (in
	<tt>lib/Target</tt>). SPARC is the simplest backend, and is RISC, so if
	you're working on a RISC target, it is a good one to start with.</p>

	<p>To create a static compiler (one that emits text assembly), you need to
	implement the following:</p>

	<ul>
	<li>Describe the register set.
	<ul>
	<li>Create a <a href="TableGenFundamentals.html">TableGen</a> description of
	the register set and register classes</li>
	<li>Implement a subclass of <tt><a
	href="CodeGenerator.html#targetregisterinfo">TargetRegisterInfo</a></tt></li>
	</ul></li>
	<li>Describe the instruction set.
	<ul>
	<li>Create a <a href="TableGenFundamentals.html">TableGen</a> description of
	the instruction set</li>
	<li>Implement a subclass of <tt><a
	href="CodeGenerator.html#targetinstrinfo">TargetInstrInfo</a></tt></li>
	</ul></li>
	<li>Describe the target machine.
	<ul>
	<li>Create a <a href="TableGenFundamentals.html">TableGen</a> description of
	the target that describes the pointer size and references the instruction
	set</li>
	<li>Implement a subclass of <tt><a
	href="CodeGenerator.html#targetmachine">TargetMachine</a></tt>, which
	configures <tt><a href="CodeGenerator.html#targetdata">TargetData</a></tt>
	correctly</li>
	<li>Register your new target using the <tt>RegisterTarget</tt>
	template:<br><br>
	<div class="doc_code"><pre>
	RegisterTarget<<em>MyTargetMachine</em>> M("short_name", " Target name");
	</pre></div>
	<br>Here, <em>MyTargetMachine</em> is the name of your implemented
	subclass of <tt><a
	href="CodeGenerator.html#targetmachine">TargetMachine</a></tt>,
	<em>short_name</em> is the option that will be active following
	<tt>-march=</tt> to select a target in llc and lli, and the last string
	is the description of your target to appear in <tt>-help</tt>
	listing.</li>
	</ul></li>
	<li>Implement the assembly printer for the architecture.
	<ul>
	<li>Define all of the assembly strings for your target, adding them to the
	instructions in your *InstrInfo.td file.</li>
	<li>Implement the <tt>llvm::AsmPrinter</tt> interface.</li>
	</ul>
	</li>
	<li>Implement an instruction selector for the architecture.
	<ul>
	<li>The recommended method is the <a href="CodeGenerator.html#instselect">
	pattern-matching DAG-to-DAG instruction selector</a> (for example, see
	the PowerPC backend in PPCISelDAGtoDAG.cpp). Parts of instruction
	selector creation can be performed by adding patterns to the instructions
	in your <tt>.td</tt> file.</li>
	</ul>
	</li>
	<li>Optionally, add subtarget support.
	<ul>
	<li>If your target has multiple subtargets (e.g. variants with different
	capabilities), implement the <tt>llvm::TargetSubtarget</tt> interface
	for your architecture. This allows you to add <tt>-mcpu=</tt> and
	<tt>-mattr=</tt> options.</li>
	</ul>
	<li>Optionally, add JIT support.
	<ul>
	<li>Create a subclass of <tt><a
	href="CodeGenerator.html#targetjitinfo">TargetJITInfo</a></tt></li>
	<li>Create a machine code emitter that will be used to emit binary code
	directly into memory, given <tt>MachineInstr</tt>s</li>
	</ul>
	</ul>
	</div>

	<!-- _______________________________________________________________________ -->
	<div class="doc_subsubsection">
	<a name="machineDetails">Implementation details</a>
	</div>

	<div class="doc_text">

	<ul>

	<li><p><b>TableGen register info description</b> - describe a class which
	will store the register's number in the binary encoding of the instruction
	(e.g., for JIT purposes).</p>

	<p>You also need to define register classes to contain these registers, such as
	the integer register class and floating-point register class, so that you can
	allocate virtual registers to instructions from these sets, and let the
	target-independent register allocator automatically choose the actual
	architected registers.</p>

	<div class="doc_code">
	<pre>
	// class Register is defined in Target.td
	<b>class</b> <em>Target</em>Reg<string name> : Register<name> {
	<b>let</b> Namespace = "<em>Target</em>";
	}

	<b>class</b> IntReg<<b>bits</b><5> num, string name> : <em>Target</em>Reg<name> {
	<b>field</b> <b>bits</b><5> Num = num;
	}

	<b>def</b> R0 : IntReg<0, "%R0">;
	...

	// class RegisterClass is defined in Target.td
	<b>def</b> IReg : RegisterClass<i64, 64, [R0, ... ]>;
	</pre>
	</div>
	</li>

	<li><p><b>TableGen instruction info description</b> - break up instructions into
	classes, usually that's already done by the manufacturer (see instruction
	manual). Define a class for each instruction category. Define each opcode as a
	subclass of the category, with appropriate parameters such as the fixed binary
	encoding of opcodes and extended opcodes, and map the register bits to the bits
	of the instruction which they are encoded in (for the JIT). Also specify how
	the instruction should be printed so it can use the automatic assembly printer,
	e.g.:</p>

	<div class="doc_code">
	<pre>
	// class Instruction is defined in Target.td
	<b>class</b> Form<<b>bits</b><6> opcode, <b>dag</b> OL, <b>string</b> asmstr> : Instruction {
	<b>field</b> <b>bits</b><42> Inst;

	<b>let</b> Namespace = "<em>Target</em>";
	<b>let</b> Inst{0-6} = opcode;
	<b>let</b> OperandList = OL;
	<b>let</b> AsmString = asmstr;
	}

	<b>def</b> ADD : Form<42, (ops IReg:$rD, IReg:$rA, IReg:$rB), "add $rD, $rA, $rB">;
	</pre>
	</div>
	</li>

	</ul>

	</div>

	<!-- ======================================================================= -->
	<div class="doc_subsection">
	<a name="lang">Language backends</a>
	</div>

	<div class="doc_text">

	<p>For now, just take a look at <tt>lib/Target/CBackend</tt> for an example of
	how the C backend is written.</p>

	</div>
	<!-- ======================================================================= -->
	<div class="doc_subsection">
	<a name="files">Files to create/modify</a>
	</div>

	<div class="doc_text">

	<p>To actually create your backend, you need to create and modify a few files.
	Here, the absolute minimum will be discussed. To actually use LLVM's target
	independent codegenerator, you must <a href="CodeGenerator.html">implement extra
	things</a>.</p>

	<p>First of all, you should create a subdirectory under <tt>lib/Target</tt>,
	which will hold all the files related to your target. Let's assume that our
	target is called, "Dummy", we would create the directory
	<tt>lib/Target/Dummy</tt>.</p>

	<p>In this new directory, you should put a <tt>Makefile</tt>. You can probably
	copy one from another target and modify it. It should at least contain the
	<tt>LEVEL</tt>, <tt>LIBRARYNAME</tt> and <tt>TARGET</tt> variables, and then
	include <tt>$(LEVEL)/Makefile.common</tt>. Be careful to give the library the
	correct name, it must be named <tt>LLVMDummy</tt> (see the MIPS target, for
	example). Alternatively, you can split the library into
	<tt>LLVMDummyCodeGen</tt> and <tt>LLVMDummyAsmPrinter</tt>, the latter of which
	should be implemented in a subdirectory below <tt>lib/Target/Dummy</tt> (see the
	PowerPC target, for example).</p>

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

	<p>To make your target actually do something, you need to implement a subclass
	of <tt>TargetMachine</tt>. This implementation should typically be in the file
	<tt>lib/Target/DummyTargetMachine.cpp</tt>, but any file in the
	<tt>lib/Target</tt> directory will be built and should work. To use LLVM's <a
	href="CodeGenerator.html">target independent code generator</a>, you should
	create a subclass of <tt>LLVMTargetMachine</tt>. This is what all current
	machine backends do. To create a target from scratch, create a subclass of
	<tt>TargetMachine</tt>. This is what the current language backends do.</p>

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

	</div>

	<!-- *********************************************************************** -->
	<div class="doc_section">
	<a name="related">Related reading material</a>
	</div>
	<!-- *********************************************************************** -->

	<div class="doc_text">

	<ul>
	<li><a href="CodeGenerator.html">Code generator</a> -
	describes some of the classes in code generation at a high level, but
	it is not (yet) complete</li>
	<li><a href="TableGenFundamentals.html">TableGen fundamentals</a> -
	describes how to use TableGen to describe your target information
	succinctly</li>
	<li><a href="HowToSubmitABug.html#codegen">Debugging code generation with
	bugpoint</a> - shows bugpoint usage scenarios to simplify backend
	development</li>
	</ul>

	</div>

	<!-- *********************************************************************** -->

	<hr>
	<address>
	<a href="http://jigsaw.w3.org/css-validator/check/referer"><img
	src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
	<a href="http://validator.w3.org/check/referer"><img
	src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>

	<a href="http://misha.brukman.net">Misha Brukman</a><br>
	<a href="http://llvm.org">The LLVM Compiler Infrastructure</a>
	<br>
	Last modified: $Date$
	</address>

	</body>
	</html>