<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
<title>Writing an LLVM Compiler Backend</title>
- <link rel="stylesheet" href="llvm.css" type="text/css">
+ <link rel="stylesheet" href="_static/llvm.css" type="text/css">
</head>
<body>
<p>
The backend of LLVM features a target-independent code generator that may create
-output for several types of target CPUs — including X86, PowerPC, Alpha,
+output for several types of target CPUs — including X86, PowerPC, ARM,
and SPARC. The backend may also be used to generate code targeted at SPUs of the
Cell processor or GPUs to support the execution of compute kernels.
</p>
<div class="doc_code">
<pre>
class RegisterClass<string namespace,
-list<ValueType> regTypes, int alignment,
- list<Register> regList> {
+list<ValueType> regTypes, int alignment, dag regList> {
string Namespace = namespace;
list<ValueType> RegTypes = regTypes;
int Size = 0; // spill size, in bits; zero lets tblgen pick the size
// default value 1 means a single instruction
// A negative value means copying is extremely expensive or impossible
int CopyCost = 1;
- list<Register> MemberList = regList;
+ dag MemberList = regList;
// for register classes that are subregisters of this class
list<RegisterClass> SubRegClassList = [];
memory.</li>
<li>The final argument, <tt>regList</tt>, specifies which registers are in this
- class. If an <tt>allocation_order_*</tt> method is not specified,
- then <tt>regList</tt> also defines the order of allocation used by the
- register allocator.</li>
+ class. If an alternative allocation order method is not specified, then
+ <tt>regList</tt> also defines the order of allocation used by the register
+ allocator. Besides simply listing registers with <tt>(add R0, R1, ...)</tt>,
+ more advanced set operators are available. See
+ <tt>include/llvm/Target/Target.td</tt> for more information.</li>
</ul>
<p>
'<tt>SP</tt>'. <tt>FPRegs</tt> defines a group of 32 single-precision
floating-point registers (<tt>F0</tt> to <tt>F31</tt>); <tt>DFPRegs</tt> defines
a group of 16 double-precision registers
-(<tt>D0-D15</tt>). For <tt>IntRegs</tt>, the <tt>MethodProtos</tt>
-and <tt>MethodBodies</tt> methods are used by TableGen to insert the specified
-code into generated output.
+(<tt>D0-D15</tt>).
</p>
<div class="doc_code">
<pre>
-def FPRegs : RegisterClass<"SP", [f32], 32,
- [F0, F1, F2, F3, F4, F5, F6, F7, F8, F9, F10, F11, F12, F13, F14, F15,
- F16, F17, F18, F19, F20, F21, F22, F23, F24, F25, F26, F27, F28, F29, F30, F31]>;
+// F0, F1, F2, ..., F31
+def FPRegs : RegisterClass<"SP", [f32], 32, (sequence "F%u", 0, 31)>;
def DFPRegs : RegisterClass<"SP", [f64], 64,
- [D0, D1, D2, D3, D4, D5, D6, D7, D8, D9, D10, D11, D12, D13, D14, D15]>;
+ (add D0, D1, D2, D3, D4, D5, D6, D7, D8,
+ D9, D10, D11, D12, D13, D14, D15)>;
def IntRegs : RegisterClass<"SP", [i32], 32,
- [L0, L1, L2, L3, L4, L5, L6, L7,
- I0, I1, I2, I3, I4, I5,
- O0, O1, O2, O3, O4, O5, O7,
- G1,
- // Non-allocatable regs:
- G2, G3, G4,
- O6, // stack ptr
- I6, // frame ptr
- I7, // return address
- G0, // constant zero
- G5, G6, G7 // reserved for kernel
- ]> {
- let MethodProtos = [{
- iterator allocation_order_end(const MachineFunction &MF) const;
- }];
- let MethodBodies = [{
- IntRegsClass::iterator
- IntRegsClass::allocation_order_end(const MachineFunction &MF) const {
- return end() - 10 // Don't allocate special registers
- -1;
- }
- }];
-}
+ (add L0, L1, L2, L3, L4, L5, L6, L7,
+ I0, I1, I2, I3, I4, I5,
+ O0, O1, O2, O3, O4, O5, O7,
+ G1,
+ // Non-allocatable regs:
+ G2, G3, G4,
+ O6, // stack ptr
+ I6, // frame ptr
+ I7, // return address
+ G0, // constant zero
+ G5, G6, G7 // reserved for kernel
+ )>;
</pre>
</div>
register implementation. The code below shows only the generated integer
registers and associated register classes. The order of registers
in <tt>IntRegs</tt> reflects the order in the definition of <tt>IntRegs</tt> in
-the target description file. Take special note of the use
-of <tt>MethodBodies</tt> in <tt>SparcRegisterInfo.td</tt> to create code in
-<tt>SparcGenRegisterInfo.inc</tt>. <tt>MethodProtos</tt> generates similar code
-in <tt>SparcGenRegisterInfo.h.inc</tt>.
+the target description file.
</p>
<div class="doc_code">
static const TargetRegisterClass* const IntRegsSuperclasses [] = {
NULL
};
-...
- IntRegsClass::iterator
- IntRegsClass::allocation_order_end(const MachineFunction &MF) const {
- return end()-10 // Don't allocate special registers
- -1;
- }
-
+
IntRegsClass::IntRegsClass() : TargetRegisterClass(IntRegsRegClassID,
IntRegsVTs, IntRegsSubclasses, IntRegsSuperclasses, IntRegsSubRegClasses,
IntRegsSuperRegClasses, 4, 4, 1, IntRegs, IntRegs + 32) {}
</pre>
</div>
+<p>
+The register allocators will avoid using reserved registers, and callee saved
+registers are not used until all the volatile registers have been used. That
+is usually good enough, but in some cases it may be necessary to provide custom
+allocation orders.
+</p>
+
</div>
<!-- ======================================================================= -->