1 =======================
2 Writing an LLVM Backend
3 =======================
16 This document describes techniques for writing compiler backends that convert
17 the LLVM Intermediate Representation (IR) to code for a specified machine or
18 other languages. Code intended for a specific machine can take the form of
19 either assembly code or binary code (usable for a JIT compiler).
21 The backend of LLVM features a target-independent code generator that may
22 create output for several types of target CPUs --- including X86, PowerPC,
23 ARM, and SPARC. The backend may also be used to generate code targeted at SPUs
24 of the Cell processor or GPUs to support the execution of compute kernels.
26 The document focuses on existing examples found in subdirectories of
27 ``llvm/lib/Target`` in a downloaded LLVM release. In particular, this document
28 focuses on the example of creating a static compiler (one that emits text
29 assembly) for a SPARC target, because SPARC has fairly standard
30 characteristics, such as a RISC instruction set and straightforward calling
36 The audience for this document is anyone who needs to write an LLVM backend to
37 generate code for a specific hardware or software target.
42 These essential documents must be read before reading this document:
44 * `LLVM Language Reference Manual <LangRef.html>`_ --- a reference manual for
45 the LLVM assembly language.
47 * :doc:`CodeGenerator` --- a guide to the components (classes and code
48 generation algorithms) for translating the LLVM internal representation into
49 machine code for a specified target. Pay particular attention to the
50 descriptions of code generation stages: Instruction Selection, Scheduling and
51 Formation, SSA-based Optimization, Register Allocation, Prolog/Epilog Code
52 Insertion, Late Machine Code Optimizations, and Code Emission.
54 * :doc:`TableGenFundamentals` --- a document that describes the TableGen
55 (``tblgen``) application that manages domain-specific information to support
56 LLVM code generation. TableGen processes input from a target description
57 file (``.td`` suffix) and generates C++ code that can be used for code
60 * :doc:`WritingAnLLVMPass` --- The assembly printer is a ``FunctionPass``, as
61 are several ``SelectionDAG`` processing steps.
63 To follow the SPARC examples in this document, have a copy of `The SPARC
64 Architecture Manual, Version 8 <http://www.sparc.org/standards/V8.pdf>`_ for
65 reference. For details about the ARM instruction set, refer to the `ARM
66 Architecture Reference Manual <http://infocenter.arm.com/>`_. For more about
67 the GNU Assembler format (``GAS``), see `Using As
68 <http://sourceware.org/binutils/docs/as/index.html>`_, especially for the
69 assembly printer. "Using As" contains a list of target machine dependent
75 To write a compiler backend for LLVM that converts the LLVM IR to code for a
76 specified target (machine or other language), follow these steps:
78 * Create a subclass of the ``TargetMachine`` class that describes
79 characteristics of your target machine. Copy existing examples of specific
80 ``TargetMachine`` class and header files; for example, start with
81 ``SparcTargetMachine.cpp`` and ``SparcTargetMachine.h``, but change the file
82 names for your target. Similarly, change code that references "``Sparc``" to
83 reference your target.
85 * Describe the register set of the target. Use TableGen to generate code for
86 register definition, register aliases, and register classes from a
87 target-specific ``RegisterInfo.td`` input file. You should also write
88 additional code for a subclass of the ``TargetRegisterInfo`` class that
89 represents the class register file data used for register allocation and also
90 describes the interactions between registers.
92 * Describe the instruction set of the target. Use TableGen to generate code
93 for target-specific instructions from target-specific versions of
94 ``TargetInstrFormats.td`` and ``TargetInstrInfo.td``. You should write
95 additional code for a subclass of the ``TargetInstrInfo`` class to represent
96 machine instructions supported by the target machine.
98 * Describe the selection and conversion of the LLVM IR from a Directed Acyclic
99 Graph (DAG) representation of instructions to native target-specific
100 instructions. Use TableGen to generate code that matches patterns and
101 selects instructions based on additional information in a target-specific
102 version of ``TargetInstrInfo.td``. Write code for ``XXXISelDAGToDAG.cpp``,
103 where ``XXX`` identifies the specific target, to perform pattern matching and
104 DAG-to-DAG instruction selection. Also write code in ``XXXISelLowering.cpp``
105 to replace or remove operations and data types that are not supported
106 natively in a SelectionDAG.
108 * Write code for an assembly printer that converts LLVM IR to a GAS format for
109 your target machine. You should add assembly strings to the instructions
110 defined in your target-specific version of ``TargetInstrInfo.td``. You
111 should also write code for a subclass of ``AsmPrinter`` that performs the
112 LLVM-to-assembly conversion and a trivial subclass of ``TargetAsmInfo``.
114 * Optionally, add support for subtargets (i.e., variants with different
115 capabilities). You should also write code for a subclass of the
116 ``TargetSubtarget`` class, which allows you to use the ``-mcpu=`` and
117 ``-mattr=`` command-line options.
119 * Optionally, add JIT support and create a machine code emitter (subclass of
120 ``TargetJITInfo``) that is used to emit binary code directly into memory.
122 In the ``.cpp`` and ``.h``. files, initially stub up these methods and then
123 implement them later. Initially, you may not know which private members that
124 the class will need and which components will need to be subclassed.
129 To actually create your compiler backend, you need to create and modify a few
130 files. The absolute minimum is discussed here. But to actually use the LLVM
131 target-independent code generator, you must perform the steps described in the
132 :doc:`LLVM Target-Independent Code Generator <CodeGenerator>` document.
134 First, you should create a subdirectory under ``lib/Target`` to hold all the
135 files related to your target. If your target is called "Dummy", create the
136 directory ``lib/Target/Dummy``.
138 In this new directory, create a ``Makefile``. It is easiest to copy a
139 ``Makefile`` of another target and modify it. It should at least contain the
140 ``LEVEL``, ``LIBRARYNAME`` and ``TARGET`` variables, and then include
141 ``$(LEVEL)/Makefile.common``. The library can be named ``LLVMDummy`` (for
142 example, see the MIPS target). Alternatively, you can split the library into
143 ``LLVMDummyCodeGen`` and ``LLVMDummyAsmPrinter``, the latter of which should be
144 implemented in a subdirectory below ``lib/Target/Dummy`` (for example, see the
147 Note that these two naming schemes are hardcoded into ``llvm-config``. Using
148 any other naming scheme will confuse ``llvm-config`` and produce a lot of
149 (seemingly unrelated) linker errors when linking ``llc``.
151 To make your target actually do something, you need to implement a subclass of
152 ``TargetMachine``. This implementation should typically be in the file
153 ``lib/Target/DummyTargetMachine.cpp``, but any file in the ``lib/Target``
154 directory will be built and should work. To use LLVM's target independent code
155 generator, you should do what all current machine backends do: create a
156 subclass of ``LLVMTargetMachine``. (To create a target from scratch, create a
157 subclass of ``TargetMachine``.)
159 To get LLVM to actually build and link your target, you need to add it to the
160 ``TARGETS_TO_BUILD`` variable. To do this, you modify the configure script to
161 know about your target when parsing the ``--enable-targets`` option. Search
162 the configure script for ``TARGETS_TO_BUILD``, add your target to the lists
163 there (some creativity required), and then reconfigure. Alternatively, you can
164 change ``autotools/configure.ac`` and regenerate configure by running
165 ``./autoconf/AutoRegen.sh``.
170 ``LLVMTargetMachine`` is designed as a base class for targets implemented with
171 the LLVM target-independent code generator. The ``LLVMTargetMachine`` class
172 should be specialized by a concrete target class that implements the various
173 virtual methods. ``LLVMTargetMachine`` is defined as a subclass of
174 ``TargetMachine`` in ``include/llvm/Target/TargetMachine.h``. The
175 ``TargetMachine`` class implementation (``TargetMachine.cpp``) also processes
176 numerous command-line options.
178 To create a concrete target-specific subclass of ``LLVMTargetMachine``, start
179 by copying an existing ``TargetMachine`` class and header. You should name the
180 files that you create to reflect your specific target. For instance, for the
181 SPARC target, name the files ``SparcTargetMachine.h`` and
182 ``SparcTargetMachine.cpp``.
184 For a target machine ``XXX``, the implementation of ``XXXTargetMachine`` must
185 have access methods to obtain objects that represent target components. These
186 methods are named ``get*Info``, and are intended to obtain the instruction set
187 (``getInstrInfo``), register set (``getRegisterInfo``), stack frame layout
188 (``getFrameInfo``), and similar information. ``XXXTargetMachine`` must also
189 implement the ``getDataLayout`` method to access an object with target-specific
190 data characteristics, such as data type size and alignment requirements.
192 For instance, for the SPARC target, the header file ``SparcTargetMachine.h``
193 declares prototypes for several ``get*Info`` and ``getDataLayout`` methods that
194 simply return a class member.
202 class SparcTargetMachine : public LLVMTargetMachine {
203 const DataLayout DataLayout; // Calculates type size & alignment
204 SparcSubtarget Subtarget;
205 SparcInstrInfo InstrInfo;
206 TargetFrameInfo FrameInfo;
209 virtual const TargetAsmInfo *createTargetAsmInfo() const;
212 SparcTargetMachine(const Module &M, const std::string &FS);
214 virtual const SparcInstrInfo *getInstrInfo() const {return &InstrInfo; }
215 virtual const TargetFrameInfo *getFrameInfo() const {return &FrameInfo; }
216 virtual const TargetSubtarget *getSubtargetImpl() const{return &Subtarget; }
217 virtual const TargetRegisterInfo *getRegisterInfo() const {
218 return &InstrInfo.getRegisterInfo();
220 virtual const DataLayout *getDataLayout() const { return &DataLayout; }
221 static unsigned getModuleMatchQuality(const Module &M);
223 // Pass Pipeline Configuration
224 virtual bool addInstSelector(PassManagerBase &PM, bool Fast);
225 virtual bool addPreEmitPass(PassManagerBase &PM, bool Fast);
228 } // end namespace llvm
231 * ``getRegisterInfo()``
233 * ``getDataLayout()``
234 * ``getSubtargetImpl()``
236 For some targets, you also need to support the following methods:
238 * ``getTargetLowering()``
241 In addition, the ``XXXTargetMachine`` constructor should specify a
242 ``TargetDescription`` string that determines the data layout for the target
243 machine, including characteristics such as pointer size, alignment, and
244 endianness. For example, the constructor for ``SparcTargetMachine`` contains
249 SparcTargetMachine::SparcTargetMachine(const Module &M, const std::string &FS)
250 : DataLayout("E-p:32:32-f128:128:128"),
251 Subtarget(M, FS), InstrInfo(Subtarget),
252 FrameInfo(TargetFrameInfo::StackGrowsDown, 8, 0) {
255 Hyphens separate portions of the ``TargetDescription`` string.
257 * An upper-case "``E``" in the string indicates a big-endian target data model.
258 A lower-case "``e``" indicates little-endian.
260 * "``p:``" is followed by pointer information: size, ABI alignment, and
261 preferred alignment. If only two figures follow "``p:``", then the first
262 value is pointer size, and the second value is both ABI and preferred
265 * Then a letter for numeric type alignment: "``i``", "``f``", "``v``", or
266 "``a``" (corresponding to integer, floating point, vector, or aggregate).
267 "``i``", "``v``", or "``a``" are followed by ABI alignment and preferred
268 alignment. "``f``" is followed by three values: the first indicates the size
269 of a long double, then ABI alignment, and then ABI preferred alignment.
274 You must also register your target with the ``TargetRegistry``, which is what
275 other LLVM tools use to be able to lookup and use your target at runtime. The
276 ``TargetRegistry`` can be used directly, but for most targets there are helper
277 templates which should take care of the work for you.
279 All targets should declare a global ``Target`` object which is used to
280 represent the target during registration. Then, in the target's ``TargetInfo``
281 library, the target should define that object and use the ``RegisterTarget``
282 template to register the target. For example, the Sparc registration code
287 Target llvm::TheSparcTarget;
289 extern "C" void LLVMInitializeSparcTargetInfo() {
290 RegisterTarget<Triple::sparc, /*HasJIT=*/false>
291 X(TheSparcTarget, "sparc", "Sparc");
294 This allows the ``TargetRegistry`` to look up the target by name or by target
295 triple. In addition, most targets will also register additional features which
296 are available in separate libraries. These registration steps are separate,
297 because some clients may wish to only link in some parts of the target --- the
298 JIT code generator does not require the use of the assembler printer, for
299 example. Here is an example of registering the Sparc assembly printer:
303 extern "C" void LLVMInitializeSparcAsmPrinter() {
304 RegisterAsmPrinter<SparcAsmPrinter> X(TheSparcTarget);
307 For more information, see "`llvm/Target/TargetRegistry.h
308 </doxygen/TargetRegistry_8h-source.html>`_".
310 Register Set and Register Classes
311 =================================
313 You should describe a concrete target-specific class that represents the
314 register file of a target machine. This class is called ``XXXRegisterInfo``
315 (where ``XXX`` identifies the target) and represents the class register file
316 data that is used for register allocation. It also describes the interactions
319 You also need to define register classes to categorize related registers. A
320 register class should be added for groups of registers that are all treated the
321 same way for some instruction. Typical examples are register classes for
322 integer, floating-point, or vector registers. A register allocator allows an
323 instruction to use any register in a specified register class to perform the
324 instruction in a similar manner. Register classes allocate virtual registers
325 to instructions from these sets, and register classes let the
326 target-independent register allocator automatically choose the actual
329 Much of the code for registers, including register definition, register
330 aliases, and register classes, is generated by TableGen from
331 ``XXXRegisterInfo.td`` input files and placed in ``XXXGenRegisterInfo.h.inc``
332 and ``XXXGenRegisterInfo.inc`` output files. Some of the code in the
333 implementation of ``XXXRegisterInfo`` requires hand-coding.
338 The ``XXXRegisterInfo.td`` file typically starts with register definitions for
339 a target machine. The ``Register`` class (specified in ``Target.td``) is used
340 to define an object for each register. The specified string ``n`` becomes the
341 ``Name`` of the register. The basic ``Register`` object does not have any
342 subregisters and does not specify any aliases.
346 class Register<string n> {
347 string Namespace = "";
351 int SpillAlignment = 0;
352 list<Register> Aliases = [];
353 list<Register> SubRegs = [];
354 list<int> DwarfNumbers = [];
357 For example, in the ``X86RegisterInfo.td`` file, there are register definitions
358 that utilize the ``Register`` class, such as:
362 def AL : Register<"AL">, DwarfRegNum<[0, 0, 0]>;
364 This defines the register ``AL`` and assigns it values (with ``DwarfRegNum``)
365 that are used by ``gcc``, ``gdb``, or a debug information writer to identify a
366 register. For register ``AL``, ``DwarfRegNum`` takes an array of 3 values
367 representing 3 different modes: the first element is for X86-64, the second for
368 exception handling (EH) on X86-32, and the third is generic. -1 is a special
369 Dwarf number that indicates the gcc number is undefined, and -2 indicates the
370 register number is invalid for this mode.
372 From the previously described line in the ``X86RegisterInfo.td`` file, TableGen
373 generates this code in the ``X86GenRegisterInfo.inc`` file:
377 static const unsigned GR8[] = { X86::AL, ... };
379 const unsigned AL_AliasSet[] = { X86::AX, X86::EAX, X86::RAX, 0 };
381 const TargetRegisterDesc RegisterDescriptors[] = {
383 { "AL", "AL", AL_AliasSet, Empty_SubRegsSet, Empty_SubRegsSet, AL_SuperRegsSet }, ...
385 From the register info file, TableGen generates a ``TargetRegisterDesc`` object
386 for each register. ``TargetRegisterDesc`` is defined in
387 ``include/llvm/Target/TargetRegisterInfo.h`` with the following fields:
391 struct TargetRegisterDesc {
392 const char *AsmName; // Assembly language name for the register
393 const char *Name; // Printable name for the reg (for debugging)
394 const unsigned *AliasSet; // Register Alias Set
395 const unsigned *SubRegs; // Sub-register set
396 const unsigned *ImmSubRegs; // Immediate sub-register set
397 const unsigned *SuperRegs; // Super-register set
400 TableGen uses the entire target description file (``.td``) to determine text
401 names for the register (in the ``AsmName`` and ``Name`` fields of
402 ``TargetRegisterDesc``) and the relationships of other registers to the defined
403 register (in the other ``TargetRegisterDesc`` fields). In this example, other
404 definitions establish the registers "``AX``", "``EAX``", and "``RAX``" as
405 aliases for one another, so TableGen generates a null-terminated array
406 (``AL_AliasSet``) for this register alias set.
408 The ``Register`` class is commonly used as a base class for more complex
409 classes. In ``Target.td``, the ``Register`` class is the base for the
410 ``RegisterWithSubRegs`` class that is used to define registers that need to
411 specify subregisters in the ``SubRegs`` list, as shown here:
415 class RegisterWithSubRegs<string n, list<Register> subregs> : Register<n> {
416 let SubRegs = subregs;
419 In ``SparcRegisterInfo.td``, additional register classes are defined for SPARC:
420 a ``Register`` subclass, ``SparcReg``, and further subclasses: ``Ri``, ``Rf``,
421 and ``Rd``. SPARC registers are identified by 5-bit ID numbers, which is a
422 feature common to these subclasses. Note the use of "``let``" expressions to
423 override values that are initially defined in a superclass (such as ``SubRegs``
424 field in the ``Rd`` class).
428 class SparcReg<string n> : Register<n> {
430 let Namespace = "SP";
432 // Ri - 32-bit integer registers
433 class Ri<bits<5> num, string n> :
437 // Rf - 32-bit floating-point registers
438 class Rf<bits<5> num, string n> :
442 // Rd - Slots in the FP register file for 64-bit floating-point values.
443 class Rd<bits<5> num, string n, list<Register> subregs> : SparcReg<n> {
445 let SubRegs = subregs;
448 In the ``SparcRegisterInfo.td`` file, there are register definitions that
449 utilize these subclasses of ``Register``, such as:
453 def G0 : Ri< 0, "G0">, DwarfRegNum<[0]>;
454 def G1 : Ri< 1, "G1">, DwarfRegNum<[1]>;
456 def F0 : Rf< 0, "F0">, DwarfRegNum<[32]>;
457 def F1 : Rf< 1, "F1">, DwarfRegNum<[33]>;
459 def D0 : Rd< 0, "F0", [F0, F1]>, DwarfRegNum<[32]>;
460 def D1 : Rd< 2, "F2", [F2, F3]>, DwarfRegNum<[34]>;
462 The last two registers shown above (``D0`` and ``D1``) are double-precision
463 floating-point registers that are aliases for pairs of single-precision
464 floating-point sub-registers. In addition to aliases, the sub-register and
465 super-register relationships of the defined register are in fields of a
466 register's ``TargetRegisterDesc``.
468 Defining a Register Class
469 -------------------------
471 The ``RegisterClass`` class (specified in ``Target.td``) is used to define an
472 object that represents a group of related registers and also defines the
473 default allocation order of the registers. A target description file
474 ``XXXRegisterInfo.td`` that uses ``Target.td`` can construct register classes
475 using the following class:
479 class RegisterClass<string namespace,
480 list<ValueType> regTypes, int alignment, dag regList> {
481 string Namespace = namespace;
482 list<ValueType> RegTypes = regTypes;
483 int Size = 0; // spill size, in bits; zero lets tblgen pick the size
484 int Alignment = alignment;
486 // CopyCost is the cost of copying a value between two registers
487 // default value 1 means a single instruction
488 // A negative value means copying is extremely expensive or impossible
490 dag MemberList = regList;
492 // for register classes that are subregisters of this class
493 list<RegisterClass> SubRegClassList = [];
495 code MethodProtos = [{}]; // to insert arbitrary code
496 code MethodBodies = [{}];
499 To define a ``RegisterClass``, use the following 4 arguments:
501 * The first argument of the definition is the name of the namespace.
503 * The second argument is a list of ``ValueType`` register type values that are
504 defined in ``include/llvm/CodeGen/ValueTypes.td``. Defined values include
505 integer types (such as ``i16``, ``i32``, and ``i1`` for Boolean),
506 floating-point types (``f32``, ``f64``), and vector types (for example,
507 ``v8i16`` for an ``8 x i16`` vector). All registers in a ``RegisterClass``
508 must have the same ``ValueType``, but some registers may store vector data in
509 different configurations. For example a register that can process a 128-bit
510 vector may be able to handle 16 8-bit integer elements, 8 16-bit integers, 4
511 32-bit integers, and so on.
513 * The third argument of the ``RegisterClass`` definition specifies the
514 alignment required of the registers when they are stored or loaded to
517 * The final argument, ``regList``, specifies which registers are in this class.
518 If an alternative allocation order method is not specified, then ``regList``
519 also defines the order of allocation used by the register allocator. Besides
520 simply listing registers with ``(add R0, R1, ...)``, more advanced set
521 operators are available. See ``include/llvm/Target/Target.td`` for more
524 In ``SparcRegisterInfo.td``, three ``RegisterClass`` objects are defined:
525 ``FPRegs``, ``DFPRegs``, and ``IntRegs``. For all three register classes, the
526 first argument defines the namespace with the string "``SP``". ``FPRegs``
527 defines a group of 32 single-precision floating-point registers (``F0`` to
528 ``F31``); ``DFPRegs`` defines a group of 16 double-precision registers
533 // F0, F1, F2, ..., F31
534 def FPRegs : RegisterClass<"SP", [f32], 32, (sequence "F%u", 0, 31)>;
536 def DFPRegs : RegisterClass<"SP", [f64], 64,
537 (add D0, D1, D2, D3, D4, D5, D6, D7, D8,
538 D9, D10, D11, D12, D13, D14, D15)>;
540 def IntRegs : RegisterClass<"SP", [i32], 32,
541 (add L0, L1, L2, L3, L4, L5, L6, L7,
542 I0, I1, I2, I3, I4, I5,
543 O0, O1, O2, O3, O4, O5, O7,
545 // Non-allocatable regs:
549 I7, // return address
551 G5, G6, G7 // reserved for kernel
554 Using ``SparcRegisterInfo.td`` with TableGen generates several output files
555 that are intended for inclusion in other source code that you write.
556 ``SparcRegisterInfo.td`` generates ``SparcGenRegisterInfo.h.inc``, which should
557 be included in the header file for the implementation of the SPARC register
558 implementation that you write (``SparcRegisterInfo.h``). In
559 ``SparcGenRegisterInfo.h.inc`` a new structure is defined called
560 ``SparcGenRegisterInfo`` that uses ``TargetRegisterInfo`` as its base. It also
561 specifies types, based upon the defined register classes: ``DFPRegsClass``,
562 ``FPRegsClass``, and ``IntRegsClass``.
564 ``SparcRegisterInfo.td`` also generates ``SparcGenRegisterInfo.inc``, which is
565 included at the bottom of ``SparcRegisterInfo.cpp``, the SPARC register
566 implementation. The code below shows only the generated integer registers and
567 associated register classes. The order of registers in ``IntRegs`` reflects
568 the order in the definition of ``IntRegs`` in the target description file.
572 // IntRegs Register Class...
573 static const unsigned IntRegs[] = {
574 SP::L0, SP::L1, SP::L2, SP::L3, SP::L4, SP::L5,
575 SP::L6, SP::L7, SP::I0, SP::I1, SP::I2, SP::I3,
576 SP::I4, SP::I5, SP::O0, SP::O1, SP::O2, SP::O3,
577 SP::O4, SP::O5, SP::O7, SP::G1, SP::G2, SP::G3,
578 SP::G4, SP::O6, SP::I6, SP::I7, SP::G0, SP::G5,
582 // IntRegsVTs Register Class Value Types...
583 static const MVT::ValueType IntRegsVTs[] = {
587 namespace SP { // Register class instances
588 DFPRegsClass DFPRegsRegClass;
589 FPRegsClass FPRegsRegClass;
590 IntRegsClass IntRegsRegClass;
592 // IntRegs Sub-register Classess...
593 static const TargetRegisterClass* const IntRegsSubRegClasses [] = {
597 // IntRegs Super-register Classess...
598 static const TargetRegisterClass* const IntRegsSuperRegClasses [] = {
602 // IntRegs Register Class sub-classes...
603 static const TargetRegisterClass* const IntRegsSubclasses [] = {
607 // IntRegs Register Class super-classes...
608 static const TargetRegisterClass* const IntRegsSuperclasses [] = {
612 IntRegsClass::IntRegsClass() : TargetRegisterClass(IntRegsRegClassID,
613 IntRegsVTs, IntRegsSubclasses, IntRegsSuperclasses, IntRegsSubRegClasses,
614 IntRegsSuperRegClasses, 4, 4, 1, IntRegs, IntRegs + 32) {}
617 The register allocators will avoid using reserved registers, and callee saved
618 registers are not used until all the volatile registers have been used. That
619 is usually good enough, but in some cases it may be necessary to provide custom
622 Implement a subclass of ``TargetRegisterInfo``
623 ----------------------------------------------
625 The final step is to hand code portions of ``XXXRegisterInfo``, which
626 implements the interface described in ``TargetRegisterInfo.h`` (see
627 :ref:`TargetRegisterInfo`). These functions return ``0``, ``NULL``, or
628 ``false``, unless overridden. Here is a list of functions that are overridden
629 for the SPARC implementation in ``SparcRegisterInfo.cpp``:
631 * ``getCalleeSavedRegs`` --- Returns a list of callee-saved registers in the
632 order of the desired callee-save stack frame offset.
634 * ``getReservedRegs`` --- Returns a bitset indexed by physical register
635 numbers, indicating if a particular register is unavailable.
637 * ``hasFP`` --- Return a Boolean indicating if a function should have a
638 dedicated frame pointer register.
640 * ``eliminateCallFramePseudoInstr`` --- If call frame setup or destroy pseudo
641 instructions are used, this can be called to eliminate them.
643 * ``eliminateFrameIndex`` --- Eliminate abstract frame indices from
644 instructions that may use them.
646 * ``emitPrologue`` --- Insert prologue code into the function.
648 * ``emitEpilogue`` --- Insert epilogue code into the function.
655 During the early stages of code generation, the LLVM IR code is converted to a
656 ``SelectionDAG`` with nodes that are instances of the ``SDNode`` class
657 containing target instructions. An ``SDNode`` has an opcode, operands, type
658 requirements, and operation properties. For example, is an operation
659 commutative, does an operation load from memory. The various operation node
660 types are described in the ``include/llvm/CodeGen/SelectionDAGNodes.h`` file
661 (values of the ``NodeType`` enum in the ``ISD`` namespace).
663 TableGen uses the following target description (``.td``) input files to
664 generate much of the code for instruction definition:
666 * ``Target.td`` --- Where the ``Instruction``, ``Operand``, ``InstrInfo``, and
667 other fundamental classes are defined.
669 * ``TargetSelectionDAG.td`` --- Used by ``SelectionDAG`` instruction selection
670 generators, contains ``SDTC*`` classes (selection DAG type constraint),
671 definitions of ``SelectionDAG`` nodes (such as ``imm``, ``cond``, ``bb``,
672 ``add``, ``fadd``, ``sub``), and pattern support (``Pattern``, ``Pat``,
673 ``PatFrag``, ``PatLeaf``, ``ComplexPattern``.
675 * ``XXXInstrFormats.td`` --- Patterns for definitions of target-specific
678 * ``XXXInstrInfo.td`` --- Target-specific definitions of instruction templates,
679 condition codes, and instructions of an instruction set. For architecture
680 modifications, a different file name may be used. For example, for Pentium
681 with SSE instruction, this file is ``X86InstrSSE.td``, and for Pentium with
682 MMX, this file is ``X86InstrMMX.td``.
684 There is also a target-specific ``XXX.td`` file, where ``XXX`` is the name of
685 the target. The ``XXX.td`` file includes the other ``.td`` input files, but
686 its contents are only directly important for subtargets.
688 You should describe a concrete target-specific class ``XXXInstrInfo`` that
689 represents machine instructions supported by a target machine.
690 ``XXXInstrInfo`` contains an array of ``XXXInstrDescriptor`` objects, each of
691 which describes one instruction. An instruction descriptor defines:
695 * List of implicit register definitions and uses
696 * Target-independent properties (such as memory access, is commutable)
697 * Target-specific flags
699 The Instruction class (defined in ``Target.td``) is mostly used as a base for
700 more complex instruction classes.
705 string Namespace = "";
706 dag OutOperandList; // A dag containing the MI def operand list.
707 dag InOperandList; // A dag containing the MI use operand list.
708 string AsmString = ""; // The .s format to print the instruction with.
709 list<dag> Pattern; // Set to the DAG pattern for this instruction.
710 list<Register> Uses = [];
711 list<Register> Defs = [];
712 list<Predicate> Predicates = []; // predicates turned into isel match code
713 ... remainder not shown for space ...
716 A ``SelectionDAG`` node (``SDNode``) should contain an object representing a
717 target-specific instruction that is defined in ``XXXInstrInfo.td``. The
718 instruction objects should represent instructions from the architecture manual
719 of the target machine (such as the SPARC Architecture Manual for the SPARC
722 A single instruction from the architecture manual is often modeled as multiple
723 target instructions, depending upon its operands. For example, a manual might
724 describe an add instruction that takes a register or an immediate operand. An
725 LLVM target could model this with two instructions named ``ADDri`` and
728 You should define a class for each instruction category and define each opcode
729 as a subclass of the category with appropriate parameters such as the fixed
730 binary encoding of opcodes and extended opcodes. You should map the register
731 bits to the bits of the instruction in which they are encoded (for the JIT).
732 Also you should specify how the instruction should be printed when the
733 automatic assembly printer is used.
735 As is described in the SPARC Architecture Manual, Version 8, there are three
736 major 32-bit formats for instructions. Format 1 is only for the ``CALL``
737 instruction. Format 2 is for branch on condition codes and ``SETHI`` (set high
738 bits of a register) instructions. Format 3 is for other instructions.
740 Each of these formats has corresponding classes in ``SparcInstrFormat.td``.
741 ``InstSP`` is a base class for other instruction classes. Additional base
742 classes are specified for more precise formats: for example in
743 ``SparcInstrFormat.td``, ``F2_1`` is for ``SETHI``, and ``F2_2`` is for
744 branches. There are three other base classes: ``F3_1`` for register/register
745 operations, ``F3_2`` for register/immediate operations, and ``F3_3`` for
746 floating-point operations. ``SparcInstrInfo.td`` also adds the base class
747 ``Pseudo`` for synthetic SPARC instructions.
749 ``SparcInstrInfo.td`` largely consists of operand and instruction definitions
750 for the SPARC target. In ``SparcInstrInfo.td``, the following target
751 description file entry, ``LDrr``, defines the Load Integer instruction for a
752 Word (the ``LD`` SPARC opcode) from a memory address to a register. The first
753 parameter, the value 3 (``11``\ :sub:`2`), is the operation value for this
754 category of operation. The second parameter (``000000``\ :sub:`2`) is the
755 specific operation value for ``LD``/Load Word. The third parameter is the
756 output destination, which is a register operand and defined in the ``Register``
757 target description file (``IntRegs``).
761 def LDrr : F3_1 <3, 0b000000, (outs IntRegs:$dst), (ins MEMrr:$addr),
763 [(set i32:$dst, (load ADDRrr:$addr))]>;
765 The fourth parameter is the input source, which uses the address operand
766 ``MEMrr`` that is defined earlier in ``SparcInstrInfo.td``:
770 def MEMrr : Operand<i32> {
771 let PrintMethod = "printMemOperand";
772 let MIOperandInfo = (ops IntRegs, IntRegs);
775 The fifth parameter is a string that is used by the assembly printer and can be
776 left as an empty string until the assembly printer interface is implemented.
777 The sixth and final parameter is the pattern used to match the instruction
778 during the SelectionDAG Select Phase described in :doc:`CodeGenerator`.
779 This parameter is detailed in the next section, :ref:`instruction-selector`.
781 Instruction class definitions are not overloaded for different operand types,
782 so separate versions of instructions are needed for register, memory, or
783 immediate value operands. For example, to perform a Load Integer instruction
784 for a Word from an immediate operand to a register, the following instruction
789 def LDri : F3_2 <3, 0b000000, (outs IntRegs:$dst), (ins MEMri:$addr),
791 [(set i32:$dst, (load ADDRri:$addr))]>;
793 Writing these definitions for so many similar instructions can involve a lot of
794 cut and paste. In ``.td`` files, the ``multiclass`` directive enables the
795 creation of templates to define several instruction classes at once (using the
796 ``defm`` directive). For example in ``SparcInstrInfo.td``, the ``multiclass``
797 pattern ``F3_12`` is defined to create 2 instruction classes each time
798 ``F3_12`` is invoked:
802 multiclass F3_12 <string OpcStr, bits<6> Op3Val, SDNode OpNode> {
803 def rr : F3_1 <2, Op3Val,
804 (outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
805 !strconcat(OpcStr, " $b, $c, $dst"),
806 [(set i32:$dst, (OpNode i32:$b, i32:$c))]>;
807 def ri : F3_2 <2, Op3Val,
808 (outs IntRegs:$dst), (ins IntRegs:$b, i32imm:$c),
809 !strconcat(OpcStr, " $b, $c, $dst"),
810 [(set i32:$dst, (OpNode i32:$b, simm13:$c))]>;
813 So when the ``defm`` directive is used for the ``XOR`` and ``ADD``
814 instructions, as seen below, it creates four instruction objects: ``XORrr``,
815 ``XORri``, ``ADDrr``, and ``ADDri``.
819 defm XOR : F3_12<"xor", 0b000011, xor>;
820 defm ADD : F3_12<"add", 0b000000, add>;
822 ``SparcInstrInfo.td`` also includes definitions for condition codes that are
823 referenced by branch instructions. The following definitions in
824 ``SparcInstrInfo.td`` indicate the bit location of the SPARC condition code.
825 For example, the 10\ :sup:`th` bit represents the "greater than" condition for
826 integers, and the 22\ :sup:`nd` bit represents the "greater than" condition for
831 def ICC_NE : ICC_VAL< 9>; // Not Equal
832 def ICC_E : ICC_VAL< 1>; // Equal
833 def ICC_G : ICC_VAL<10>; // Greater
835 def FCC_U : FCC_VAL<23>; // Unordered
836 def FCC_G : FCC_VAL<22>; // Greater
837 def FCC_UG : FCC_VAL<21>; // Unordered or Greater
840 (Note that ``Sparc.h`` also defines enums that correspond to the same SPARC
841 condition codes. Care must be taken to ensure the values in ``Sparc.h``
842 correspond to the values in ``SparcInstrInfo.td``. I.e., ``SPCC::ICC_NE = 9``,
843 ``SPCC::FCC_U = 23`` and so on.)
845 Instruction Operand Mapping
846 ---------------------------
848 The code generator backend maps instruction operands to fields in the
849 instruction. Operands are assigned to unbound fields in the instruction in the
850 order they are defined. Fields are bound when they are assigned a value. For
851 example, the Sparc target defines the ``XNORrr`` instruction as a ``F3_1``
852 format instruction having three operands.
856 def XNORrr : F3_1<2, 0b000111,
857 (outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
859 [(set i32:$dst, (not (xor i32:$b, i32:$c)))]>;
861 The instruction templates in ``SparcInstrFormats.td`` show the base class for
862 ``F3_1`` is ``InstSP``.
866 class InstSP<dag outs, dag ins, string asmstr, list<dag> pattern> : Instruction {
868 let Namespace = "SP";
870 let Inst{31-30} = op;
871 dag OutOperandList = outs;
872 dag InOperandList = ins;
873 let AsmString = asmstr;
874 let Pattern = pattern;
877 ``InstSP`` leaves the ``op`` field unbound.
881 class F3<dag outs, dag ins, string asmstr, list<dag> pattern>
882 : InstSP<outs, ins, asmstr, pattern> {
886 let op{1} = 1; // Op = 2 or 3
887 let Inst{29-25} = rd;
888 let Inst{24-19} = op3;
889 let Inst{18-14} = rs1;
892 ``F3`` binds the ``op`` field and defines the ``rd``, ``op3``, and ``rs1``
893 fields. ``F3`` format instructions will bind the operands ``rd``, ``op3``, and
898 class F3_1<bits<2> opVal, bits<6> op3val, dag outs, dag ins,
899 string asmstr, list<dag> pattern> : F3<outs, ins, asmstr, pattern> {
900 bits<8> asi = 0; // asi not currently used
904 let Inst{13} = 0; // i field = 0
905 let Inst{12-5} = asi; // address space identifier
909 ``F3_1`` binds the ``op3`` field and defines the ``rs2`` fields. ``F3_1``
910 format instructions will bind the operands to the ``rd``, ``rs1``, and ``rs2``
911 fields. This results in the ``XNORrr`` instruction binding ``$dst``, ``$b``,
912 and ``$c`` operands to the ``rd``, ``rs1``, and ``rs2`` fields respectively.
914 Instruction Operand Name Mapping
915 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
917 TableGen will also generate a function called getNamedOperandIdx() which
918 can be used to look up an operand's index in a MachineInstr based on its
919 TableGen name. Setting the UseNamedOperandTable bit in an instruction's
920 TableGen definition will add all of its operands to an enumeration in the
921 llvm::XXX:OpName namespace and also add an entry for it into the OperandMap
922 table, which can be queried using getNamedOperandIdx()
926 int DstIndex = SP::getNamedOperandIdx(SP::XNORrr, SP::OpName::dst); // => 0
927 int BIndex = SP::getNamedOperandIdx(SP::XNORrr, SP::OpName::b); // => 1
928 int CIndex = SP::getNamedOperandIdx(SP::XNORrr, SP::OpName::c); // => 2
929 int DIndex = SP::getNamedOperandIdx(SP::XNORrr, SP::OpName::d); // => -1
933 The entries in the OpName enum are taken verbatim from the TableGen definitions,
934 so operands with lowercase names will have lower case entries in the enum.
936 To include the getNamedOperandIdx() function in your backend, you will need
937 to define a few preprocessor macros in XXXInstrInfo.cpp and XXXInstrInfo.h.
944 #define GET_INSTRINFO_NAMED_OPS // For getNamedOperandIdx() function
945 #include "XXXGenInstrInfo.inc"
951 #define GET_INSTRINFO_OPERAND_ENUM // For OpName enum
952 #include "XXXGenInstrInfo.inc"
955 int16_t getNamedOperandIdx(uint16_t Opcode, uint16_t NamedIndex);
956 } // End namespace XXX
958 Instruction Scheduling
959 ----------------------
961 Instruction itineraries can be queried using MCDesc::getSchedClass(). The
962 value can be named by an enumemation in llvm::XXX::Sched namespace generated
963 by TableGen in XXXGenInstrInfo.inc. The name of the schedule classes are
964 the same as provided in XXXSchedule.td plus a default NoItinerary class.
966 Instruction Relation Mapping
967 ----------------------------
969 This TableGen feature is used to relate instructions with each other. It is
970 particularly useful when you have multiple instruction formats and need to
971 switch between them after instruction selection. This entire feature is driven
972 by relation models which can be defined in ``XXXInstrInfo.td`` files
973 according to the target-specific instruction set. Relation models are defined
974 using ``InstrMapping`` class as a base. TableGen parses all the models
975 and generates instruction relation maps using the specified information.
976 Relation maps are emitted as tables in the ``XXXGenInstrInfo.inc`` file
977 along with the functions to query them. For the detailed information on how to
978 use this feature, please refer to :doc:`HowToUseInstrMappings`.
980 Implement a subclass of ``TargetInstrInfo``
981 -------------------------------------------
983 The final step is to hand code portions of ``XXXInstrInfo``, which implements
984 the interface described in ``TargetInstrInfo.h`` (see :ref:`TargetInstrInfo`).
985 These functions return ``0`` or a Boolean or they assert, unless overridden.
986 Here's a list of functions that are overridden for the SPARC implementation in
987 ``SparcInstrInfo.cpp``:
989 * ``isLoadFromStackSlot`` --- If the specified machine instruction is a direct
990 load from a stack slot, return the register number of the destination and the
991 ``FrameIndex`` of the stack slot.
993 * ``isStoreToStackSlot`` --- If the specified machine instruction is a direct
994 store to a stack slot, return the register number of the destination and the
995 ``FrameIndex`` of the stack slot.
997 * ``copyPhysReg`` --- Copy values between a pair of physical registers.
999 * ``storeRegToStackSlot`` --- Store a register value to a stack slot.
1001 * ``loadRegFromStackSlot`` --- Load a register value from a stack slot.
1003 * ``storeRegToAddr`` --- Store a register value to memory.
1005 * ``loadRegFromAddr`` --- Load a register value from memory.
1007 * ``foldMemoryOperand`` --- Attempt to combine instructions of any load or
1008 store instruction for the specified operand(s).
1010 Branch Folding and If Conversion
1011 --------------------------------
1013 Performance can be improved by combining instructions or by eliminating
1014 instructions that are never reached. The ``AnalyzeBranch`` method in
1015 ``XXXInstrInfo`` may be implemented to examine conditional instructions and
1016 remove unnecessary instructions. ``AnalyzeBranch`` looks at the end of a
1017 machine basic block (MBB) for opportunities for improvement, such as branch
1018 folding and if conversion. The ``BranchFolder`` and ``IfConverter`` machine
1019 function passes (see the source files ``BranchFolding.cpp`` and
1020 ``IfConversion.cpp`` in the ``lib/CodeGen`` directory) call ``AnalyzeBranch``
1021 to improve the control flow graph that represents the instructions.
1023 Several implementations of ``AnalyzeBranch`` (for ARM, Alpha, and X86) can be
1024 examined as models for your own ``AnalyzeBranch`` implementation. Since SPARC
1025 does not implement a useful ``AnalyzeBranch``, the ARM target implementation is
1028 ``AnalyzeBranch`` returns a Boolean value and takes four parameters:
1030 * ``MachineBasicBlock &MBB`` --- The incoming block to be examined.
1032 * ``MachineBasicBlock *&TBB`` --- A destination block that is returned. For a
1033 conditional branch that evaluates to true, ``TBB`` is the destination.
1035 * ``MachineBasicBlock *&FBB`` --- For a conditional branch that evaluates to
1036 false, ``FBB`` is returned as the destination.
1038 * ``std::vector<MachineOperand> &Cond`` --- List of operands to evaluate a
1039 condition for a conditional branch.
1041 In the simplest case, if a block ends without a branch, then it falls through
1042 to the successor block. No destination blocks are specified for either ``TBB``
1043 or ``FBB``, so both parameters return ``NULL``. The start of the
1044 ``AnalyzeBranch`` (see code below for the ARM target) shows the function
1045 parameters and the code for the simplest case.
1049 bool ARMInstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,
1050 MachineBasicBlock *&TBB,
1051 MachineBasicBlock *&FBB,
1052 std::vector<MachineOperand> &Cond) const
1054 MachineBasicBlock::iterator I = MBB.end();
1055 if (I == MBB.begin() || !isUnpredicatedTerminator(--I))
1058 If a block ends with a single unconditional branch instruction, then
1059 ``AnalyzeBranch`` (shown below) should return the destination of that branch in
1060 the ``TBB`` parameter.
1064 if (LastOpc == ARM::B || LastOpc == ARM::tB) {
1065 TBB = LastInst->getOperand(0).getMBB();
1069 If a block ends with two unconditional branches, then the second branch is
1070 never reached. In that situation, as shown below, remove the last branch
1071 instruction and return the penultimate branch in the ``TBB`` parameter.
1075 if ((SecondLastOpc == ARM::B || SecondLastOpc == ARM::tB) &&
1076 (LastOpc == ARM::B || LastOpc == ARM::tB)) {
1077 TBB = SecondLastInst->getOperand(0).getMBB();
1079 I->eraseFromParent();
1083 A block may end with a single conditional branch instruction that falls through
1084 to successor block if the condition evaluates to false. In that case,
1085 ``AnalyzeBranch`` (shown below) should return the destination of that
1086 conditional branch in the ``TBB`` parameter and a list of operands in the
1087 ``Cond`` parameter to evaluate the condition.
1091 if (LastOpc == ARM::Bcc || LastOpc == ARM::tBcc) {
1092 // Block ends with fall-through condbranch.
1093 TBB = LastInst->getOperand(0).getMBB();
1094 Cond.push_back(LastInst->getOperand(1));
1095 Cond.push_back(LastInst->getOperand(2));
1099 If a block ends with both a conditional branch and an ensuing unconditional
1100 branch, then ``AnalyzeBranch`` (shown below) should return the conditional
1101 branch destination (assuming it corresponds to a conditional evaluation of
1102 "``true``") in the ``TBB`` parameter and the unconditional branch destination
1103 in the ``FBB`` (corresponding to a conditional evaluation of "``false``"). A
1104 list of operands to evaluate the condition should be returned in the ``Cond``
1109 unsigned SecondLastOpc = SecondLastInst->getOpcode();
1111 if ((SecondLastOpc == ARM::Bcc && LastOpc == ARM::B) ||
1112 (SecondLastOpc == ARM::tBcc && LastOpc == ARM::tB)) {
1113 TBB = SecondLastInst->getOperand(0).getMBB();
1114 Cond.push_back(SecondLastInst->getOperand(1));
1115 Cond.push_back(SecondLastInst->getOperand(2));
1116 FBB = LastInst->getOperand(0).getMBB();
1120 For the last two cases (ending with a single conditional branch or ending with
1121 one conditional and one unconditional branch), the operands returned in the
1122 ``Cond`` parameter can be passed to methods of other instructions to create new
1123 branches or perform other operations. An implementation of ``AnalyzeBranch``
1124 requires the helper methods ``RemoveBranch`` and ``InsertBranch`` to manage
1125 subsequent operations.
1127 ``AnalyzeBranch`` should return false indicating success in most circumstances.
1128 ``AnalyzeBranch`` should only return true when the method is stumped about what
1129 to do, for example, if a block has three terminating branches.
1130 ``AnalyzeBranch`` may return true if it encounters a terminator it cannot
1131 handle, such as an indirect branch.
1133 .. _instruction-selector:
1135 Instruction Selector
1136 ====================
1138 LLVM uses a ``SelectionDAG`` to represent LLVM IR instructions, and nodes of
1139 the ``SelectionDAG`` ideally represent native target instructions. During code
1140 generation, instruction selection passes are performed to convert non-native
1141 DAG instructions into native target-specific instructions. The pass described
1142 in ``XXXISelDAGToDAG.cpp`` is used to match patterns and perform DAG-to-DAG
1143 instruction selection. Optionally, a pass may be defined (in
1144 ``XXXBranchSelector.cpp``) to perform similar DAG-to-DAG operations for branch
1145 instructions. Later, the code in ``XXXISelLowering.cpp`` replaces or removes
1146 operations and data types not supported natively (legalizes) in a
1149 TableGen generates code for instruction selection using the following target
1150 description input files:
1152 * ``XXXInstrInfo.td`` --- Contains definitions of instructions in a
1153 target-specific instruction set, generates ``XXXGenDAGISel.inc``, which is
1154 included in ``XXXISelDAGToDAG.cpp``.
1156 * ``XXXCallingConv.td`` --- Contains the calling and return value conventions
1157 for the target architecture, and it generates ``XXXGenCallingConv.inc``,
1158 which is included in ``XXXISelLowering.cpp``.
1160 The implementation of an instruction selection pass must include a header that
1161 declares the ``FunctionPass`` class or a subclass of ``FunctionPass``. In
1162 ``XXXTargetMachine.cpp``, a Pass Manager (PM) should add each instruction
1163 selection pass into the queue of passes to run.
1165 The LLVM static compiler (``llc``) is an excellent tool for visualizing the
1166 contents of DAGs. To display the ``SelectionDAG`` before or after specific
1167 processing phases, use the command line options for ``llc``, described at
1168 :ref:`SelectionDAG-Process`.
1170 To describe instruction selector behavior, you should add patterns for lowering
1171 LLVM code into a ``SelectionDAG`` as the last parameter of the instruction
1172 definitions in ``XXXInstrInfo.td``. For example, in ``SparcInstrInfo.td``,
1173 this entry defines a register store operation, and the last parameter describes
1174 a pattern with the store DAG operator.
1176 .. code-block:: llvm
1178 def STrr : F3_1< 3, 0b000100, (outs), (ins MEMrr:$addr, IntRegs:$src),
1179 "st $src, [$addr]", [(store i32:$src, ADDRrr:$addr)]>;
1181 ``ADDRrr`` is a memory mode that is also defined in ``SparcInstrInfo.td``:
1183 .. code-block:: llvm
1185 def ADDRrr : ComplexPattern<i32, 2, "SelectADDRrr", [], []>;
1187 The definition of ``ADDRrr`` refers to ``SelectADDRrr``, which is a function
1188 defined in an implementation of the Instructor Selector (such as
1189 ``SparcISelDAGToDAG.cpp``).
1191 In ``lib/Target/TargetSelectionDAG.td``, the DAG operator for store is defined
1194 .. code-block:: llvm
1196 def store : PatFrag<(ops node:$val, node:$ptr),
1197 (st node:$val, node:$ptr), [{
1198 if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N))
1199 return !ST->isTruncatingStore() &&
1200 ST->getAddressingMode() == ISD::UNINDEXED;
1204 ``XXXInstrInfo.td`` also generates (in ``XXXGenDAGISel.inc``) the
1205 ``SelectCode`` method that is used to call the appropriate processing method
1206 for an instruction. In this example, ``SelectCode`` calls ``Select_ISD_STORE``
1207 for the ``ISD::STORE`` opcode.
1211 SDNode *SelectCode(SDValue N) {
1213 MVT::ValueType NVT = N.getNode()->getValueType(0);
1214 switch (N.getOpcode()) {
1218 return Select_ISD_STORE(N);
1225 The pattern for ``STrr`` is matched, so elsewhere in ``XXXGenDAGISel.inc``,
1226 code for ``STrr`` is created for ``Select_ISD_STORE``. The ``Emit_22`` method
1227 is also generated in ``XXXGenDAGISel.inc`` to complete the processing of this
1232 SDNode *Select_ISD_STORE(const SDValue &N) {
1233 SDValue Chain = N.getOperand(0);
1234 if (Predicate_store(N.getNode())) {
1235 SDValue N1 = N.getOperand(1);
1236 SDValue N2 = N.getOperand(2);
1240 // Pattern: (st:void i32:i32:$src,
1241 // ADDRrr:i32:$addr)<<P:Predicate_store>>
1242 // Emits: (STrr:void ADDRrr:i32:$addr, IntRegs:i32:$src)
1243 // Pattern complexity = 13 cost = 1 size = 0
1244 if (SelectADDRrr(N, N2, CPTmp0, CPTmp1) &&
1245 N1.getNode()->getValueType(0) == MVT::i32 &&
1246 N2.getNode()->getValueType(0) == MVT::i32) {
1247 return Emit_22(N, SP::STrr, CPTmp0, CPTmp1);
1251 The SelectionDAG Legalize Phase
1252 -------------------------------
1254 The Legalize phase converts a DAG to use types and operations that are natively
1255 supported by the target. For natively unsupported types and operations, you
1256 need to add code to the target-specific ``XXXTargetLowering`` implementation to
1257 convert unsupported types and operations to supported ones.
1259 In the constructor for the ``XXXTargetLowering`` class, first use the
1260 ``addRegisterClass`` method to specify which types are supported and which
1261 register classes are associated with them. The code for the register classes
1262 are generated by TableGen from ``XXXRegisterInfo.td`` and placed in
1263 ``XXXGenRegisterInfo.h.inc``. For example, the implementation of the
1264 constructor for the SparcTargetLowering class (in ``SparcISelLowering.cpp``)
1265 starts with the following code:
1269 addRegisterClass(MVT::i32, SP::IntRegsRegisterClass);
1270 addRegisterClass(MVT::f32, SP::FPRegsRegisterClass);
1271 addRegisterClass(MVT::f64, SP::DFPRegsRegisterClass);
1273 You should examine the node types in the ``ISD`` namespace
1274 (``include/llvm/CodeGen/SelectionDAGNodes.h``) and determine which operations
1275 the target natively supports. For operations that do **not** have native
1276 support, add a callback to the constructor for the ``XXXTargetLowering`` class,
1277 so the instruction selection process knows what to do. The ``TargetLowering``
1278 class callback methods (declared in ``llvm/Target/TargetLowering.h``) are:
1280 * ``setOperationAction`` --- General operation.
1281 * ``setLoadExtAction`` --- Load with extension.
1282 * ``setTruncStoreAction`` --- Truncating store.
1283 * ``setIndexedLoadAction`` --- Indexed load.
1284 * ``setIndexedStoreAction`` --- Indexed store.
1285 * ``setConvertAction`` --- Type conversion.
1286 * ``setCondCodeAction`` --- Support for a given condition code.
1288 Note: on older releases, ``setLoadXAction`` is used instead of
1289 ``setLoadExtAction``. Also, on older releases, ``setCondCodeAction`` may not
1290 be supported. Examine your release to see what methods are specifically
1293 These callbacks are used to determine that an operation does or does not work
1294 with a specified type (or types). And in all cases, the third parameter is a
1295 ``LegalAction`` type enum value: ``Promote``, ``Expand``, ``Custom``, or
1296 ``Legal``. ``SparcISelLowering.cpp`` contains examples of all four
1297 ``LegalAction`` values.
1302 For an operation without native support for a given type, the specified type
1303 may be promoted to a larger type that is supported. For example, SPARC does
1304 not support a sign-extending load for Boolean values (``i1`` type), so in
1305 ``SparcISelLowering.cpp`` the third parameter below, ``Promote``, changes
1306 ``i1`` type values to a large type before loading.
1310 setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
1315 For a type without native support, a value may need to be broken down further,
1316 rather than promoted. For an operation without native support, a combination
1317 of other operations may be used to similar effect. In SPARC, the
1318 floating-point sine and cosine trig operations are supported by expansion to
1319 other operations, as indicated by the third parameter, ``Expand``, to
1320 ``setOperationAction``:
1324 setOperationAction(ISD::FSIN, MVT::f32, Expand);
1325 setOperationAction(ISD::FCOS, MVT::f32, Expand);
1330 For some operations, simple type promotion or operation expansion may be
1331 insufficient. In some cases, a special intrinsic function must be implemented.
1333 For example, a constant value may require special treatment, or an operation
1334 may require spilling and restoring registers in the stack and working with
1335 register allocators.
1337 As seen in ``SparcISelLowering.cpp`` code below, to perform a type conversion
1338 from a floating point value to a signed integer, first the
1339 ``setOperationAction`` should be called with ``Custom`` as the third parameter:
1343 setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
1345 In the ``LowerOperation`` method, for each ``Custom`` operation, a case
1346 statement should be added to indicate what function to call. In the following
1347 code, an ``FP_TO_SINT`` opcode will call the ``LowerFP_TO_SINT`` method:
1351 SDValue SparcTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) {
1352 switch (Op.getOpcode()) {
1353 case ISD::FP_TO_SINT: return LowerFP_TO_SINT(Op, DAG);
1358 Finally, the ``LowerFP_TO_SINT`` method is implemented, using an FP register to
1359 convert the floating-point value to an integer.
1363 static SDValue LowerFP_TO_SINT(SDValue Op, SelectionDAG &DAG) {
1364 assert(Op.getValueType() == MVT::i32);
1365 Op = DAG.getNode(SPISD::FTOI, MVT::f32, Op.getOperand(0));
1366 return DAG.getNode(ISD::BITCAST, MVT::i32, Op);
1372 The ``Legal`` ``LegalizeAction`` enum value simply indicates that an operation
1373 **is** natively supported. ``Legal`` represents the default condition, so it
1374 is rarely used. In ``SparcISelLowering.cpp``, the action for ``CTPOP`` (an
1375 operation to count the bits set in an integer) is natively supported only for
1376 SPARC v9. The following code enables the ``Expand`` conversion technique for
1377 non-v9 SPARC implementations.
1381 setOperationAction(ISD::CTPOP, MVT::i32, Expand);
1383 if (TM.getSubtarget<SparcSubtarget>().isV9())
1384 setOperationAction(ISD::CTPOP, MVT::i32, Legal);
1389 To support target-specific calling conventions, ``XXXGenCallingConv.td`` uses
1390 interfaces (such as ``CCIfType`` and ``CCAssignToReg``) that are defined in
1391 ``lib/Target/TargetCallingConv.td``. TableGen can take the target descriptor
1392 file ``XXXGenCallingConv.td`` and generate the header file
1393 ``XXXGenCallingConv.inc``, which is typically included in
1394 ``XXXISelLowering.cpp``. You can use the interfaces in
1395 ``TargetCallingConv.td`` to specify:
1397 * The order of parameter allocation.
1399 * Where parameters and return values are placed (that is, on the stack or in
1402 * Which registers may be used.
1404 * Whether the caller or callee unwinds the stack.
1406 The following example demonstrates the use of the ``CCIfType`` and
1407 ``CCAssignToReg`` interfaces. If the ``CCIfType`` predicate is true (that is,
1408 if the current argument is of type ``f32`` or ``f64``), then the action is
1409 performed. In this case, the ``CCAssignToReg`` action assigns the argument
1410 value to the first available register: either ``R0`` or ``R1``.
1412 .. code-block:: llvm
1414 CCIfType<[f32,f64], CCAssignToReg<[R0, R1]>>
1416 ``SparcCallingConv.td`` contains definitions for a target-specific return-value
1417 calling convention (``RetCC_Sparc32``) and a basic 32-bit C calling convention
1418 (``CC_Sparc32``). The definition of ``RetCC_Sparc32`` (shown below) indicates
1419 which registers are used for specified scalar return types. A single-precision
1420 float is returned to register ``F0``, and a double-precision float goes to
1421 register ``D0``. A 32-bit integer is returned in register ``I0`` or ``I1``.
1423 .. code-block:: llvm
1425 def RetCC_Sparc32 : CallingConv<[
1426 CCIfType<[i32], CCAssignToReg<[I0, I1]>>,
1427 CCIfType<[f32], CCAssignToReg<[F0]>>,
1428 CCIfType<[f64], CCAssignToReg<[D0]>>
1431 The definition of ``CC_Sparc32`` in ``SparcCallingConv.td`` introduces
1432 ``CCAssignToStack``, which assigns the value to a stack slot with the specified
1433 size and alignment. In the example below, the first parameter, 4, indicates
1434 the size of the slot, and the second parameter, also 4, indicates the stack
1435 alignment along 4-byte units. (Special cases: if size is zero, then the ABI
1436 size is used; if alignment is zero, then the ABI alignment is used.)
1438 .. code-block:: llvm
1440 def CC_Sparc32 : CallingConv<[
1441 // All arguments get passed in integer registers if there is space.
1442 CCIfType<[i32, f32, f64], CCAssignToReg<[I0, I1, I2, I3, I4, I5]>>,
1443 CCAssignToStack<4, 4>
1446 ``CCDelegateTo`` is another commonly used interface, which tries to find a
1447 specified sub-calling convention, and, if a match is found, it is invoked. In
1448 the following example (in ``X86CallingConv.td``), the definition of
1449 ``RetCC_X86_32_C`` ends with ``CCDelegateTo``. After the current value is
1450 assigned to the register ``ST0`` or ``ST1``, the ``RetCC_X86Common`` is
1453 .. code-block:: llvm
1455 def RetCC_X86_32_C : CallingConv<[
1456 CCIfType<[f32], CCAssignToReg<[ST0, ST1]>>,
1457 CCIfType<[f64], CCAssignToReg<[ST0, ST1]>>,
1458 CCDelegateTo<RetCC_X86Common>
1461 ``CCIfCC`` is an interface that attempts to match the given name to the current
1462 calling convention. If the name identifies the current calling convention,
1463 then a specified action is invoked. In the following example (in
1464 ``X86CallingConv.td``), if the ``Fast`` calling convention is in use, then
1465 ``RetCC_X86_32_Fast`` is invoked. If the ``SSECall`` calling convention is in
1466 use, then ``RetCC_X86_32_SSE`` is invoked.
1468 .. code-block:: llvm
1470 def RetCC_X86_32 : CallingConv<[
1471 CCIfCC<"CallingConv::Fast", CCDelegateTo<RetCC_X86_32_Fast>>,
1472 CCIfCC<"CallingConv::X86_SSECall", CCDelegateTo<RetCC_X86_32_SSE>>,
1473 CCDelegateTo<RetCC_X86_32_C>
1476 Other calling convention interfaces include:
1478 * ``CCIf <predicate, action>`` --- If the predicate matches, apply the action.
1480 * ``CCIfInReg <action>`` --- If the argument is marked with the "``inreg``"
1481 attribute, then apply the action.
1483 * ``CCIfNest <action>`` --- If the argument is marked with the "``nest``"
1484 attribute, then apply the action.
1486 * ``CCIfNotVarArg <action>`` --- If the current function does not take a
1487 variable number of arguments, apply the action.
1489 * ``CCAssignToRegWithShadow <registerList, shadowList>`` --- similar to
1490 ``CCAssignToReg``, but with a shadow list of registers.
1492 * ``CCPassByVal <size, align>`` --- Assign value to a stack slot with the
1493 minimum specified size and alignment.
1495 * ``CCPromoteToType <type>`` --- Promote the current value to the specified
1498 * ``CallingConv <[actions]>`` --- Define each calling convention that is
1504 During the code emission stage, the code generator may utilize an LLVM pass to
1505 produce assembly output. To do this, you want to implement the code for a
1506 printer that converts LLVM IR to a GAS-format assembly language for your target
1507 machine, using the following steps:
1509 * Define all the assembly strings for your target, adding them to the
1510 instructions defined in the ``XXXInstrInfo.td`` file. (See
1511 :ref:`instruction-set`.) TableGen will produce an output file
1512 (``XXXGenAsmWriter.inc``) with an implementation of the ``printInstruction``
1513 method for the ``XXXAsmPrinter`` class.
1515 * Write ``XXXTargetAsmInfo.h``, which contains the bare-bones declaration of
1516 the ``XXXTargetAsmInfo`` class (a subclass of ``TargetAsmInfo``).
1518 * Write ``XXXTargetAsmInfo.cpp``, which contains target-specific values for
1519 ``TargetAsmInfo`` properties and sometimes new implementations for methods.
1521 * Write ``XXXAsmPrinter.cpp``, which implements the ``AsmPrinter`` class that
1522 performs the LLVM-to-assembly conversion.
1524 The code in ``XXXTargetAsmInfo.h`` is usually a trivial declaration of the
1525 ``XXXTargetAsmInfo`` class for use in ``XXXTargetAsmInfo.cpp``. Similarly,
1526 ``XXXTargetAsmInfo.cpp`` usually has a few declarations of ``XXXTargetAsmInfo``
1527 replacement values that override the default values in ``TargetAsmInfo.cpp``.
1528 For example in ``SparcTargetAsmInfo.cpp``:
1532 SparcTargetAsmInfo::SparcTargetAsmInfo(const SparcTargetMachine &TM) {
1533 Data16bitsDirective = "\t.half\t";
1534 Data32bitsDirective = "\t.word\t";
1535 Data64bitsDirective = 0; // .xword is only supported by V9.
1536 ZeroDirective = "\t.skip\t";
1537 CommentString = "!";
1538 ConstantPoolSection = "\t.section \".rodata\",#alloc\n";
1541 The X86 assembly printer implementation (``X86TargetAsmInfo``) is an example
1542 where the target specific ``TargetAsmInfo`` class uses an overridden methods:
1543 ``ExpandInlineAsm``.
1545 A target-specific implementation of ``AsmPrinter`` is written in
1546 ``XXXAsmPrinter.cpp``, which implements the ``AsmPrinter`` class that converts
1547 the LLVM to printable assembly. The implementation must include the following
1548 headers that have declarations for the ``AsmPrinter`` and
1549 ``MachineFunctionPass`` classes. The ``MachineFunctionPass`` is a subclass of
1554 #include "llvm/CodeGen/AsmPrinter.h"
1555 #include "llvm/CodeGen/MachineFunctionPass.h"
1557 As a ``FunctionPass``, ``AsmPrinter`` first calls ``doInitialization`` to set
1558 up the ``AsmPrinter``. In ``SparcAsmPrinter``, a ``Mangler`` object is
1559 instantiated to process variable names.
1561 In ``XXXAsmPrinter.cpp``, the ``runOnMachineFunction`` method (declared in
1562 ``MachineFunctionPass``) must be implemented for ``XXXAsmPrinter``. In
1563 ``MachineFunctionPass``, the ``runOnFunction`` method invokes
1564 ``runOnMachineFunction``. Target-specific implementations of
1565 ``runOnMachineFunction`` differ, but generally do the following to process each
1568 * Call ``SetupMachineFunction`` to perform initialization.
1570 * Call ``EmitConstantPool`` to print out (to the output stream) constants which
1571 have been spilled to memory.
1573 * Call ``EmitJumpTableInfo`` to print out jump tables used by the current
1576 * Print out the label for the current function.
1578 * Print out the code for the function, including basic block labels and the
1579 assembly for the instruction (using ``printInstruction``)
1581 The ``XXXAsmPrinter`` implementation must also include the code generated by
1582 TableGen that is output in the ``XXXGenAsmWriter.inc`` file. The code in
1583 ``XXXGenAsmWriter.inc`` contains an implementation of the ``printInstruction``
1584 method that may call these methods:
1587 * ``printMemOperand``
1588 * ``printCCOperand`` (for conditional statements)
1589 * ``printDataDirective``
1591 * ``printImplicitDef``
1592 * ``printInlineAsm``
1594 The implementations of ``printDeclare``, ``printImplicitDef``,
1595 ``printInlineAsm``, and ``printLabel`` in ``AsmPrinter.cpp`` are generally
1596 adequate for printing assembly and do not need to be overridden.
1598 The ``printOperand`` method is implemented with a long ``switch``/``case``
1599 statement for the type of operand: register, immediate, basic block, external
1600 symbol, global address, constant pool index, or jump table index. For an
1601 instruction with a memory address operand, the ``printMemOperand`` method
1602 should be implemented to generate the proper output. Similarly,
1603 ``printCCOperand`` should be used to print a conditional operand.
1605 ``doFinalization`` should be overridden in ``XXXAsmPrinter``, and it should be
1606 called to shut down the assembly printer. During ``doFinalization``, global
1607 variables and constants are printed to output.
1612 Subtarget support is used to inform the code generation process of instruction
1613 set variations for a given chip set. For example, the LLVM SPARC
1614 implementation provided covers three major versions of the SPARC microprocessor
1615 architecture: Version 8 (V8, which is a 32-bit architecture), Version 9 (V9, a
1616 64-bit architecture), and the UltraSPARC architecture. V8 has 16
1617 double-precision floating-point registers that are also usable as either 32
1618 single-precision or 8 quad-precision registers. V8 is also purely big-endian.
1619 V9 has 32 double-precision floating-point registers that are also usable as 16
1620 quad-precision registers, but cannot be used as single-precision registers.
1621 The UltraSPARC architecture combines V9 with UltraSPARC Visual Instruction Set
1624 If subtarget support is needed, you should implement a target-specific
1625 ``XXXSubtarget`` class for your architecture. This class should process the
1626 command-line options ``-mcpu=`` and ``-mattr=``.
1628 TableGen uses definitions in the ``Target.td`` and ``Sparc.td`` files to
1629 generate code in ``SparcGenSubtarget.inc``. In ``Target.td``, shown below, the
1630 ``SubtargetFeature`` interface is defined. The first 4 string parameters of
1631 the ``SubtargetFeature`` interface are a feature name, an attribute set by the
1632 feature, the value of the attribute, and a description of the feature. (The
1633 fifth parameter is a list of features whose presence is implied, and its
1634 default value is an empty array.)
1636 .. code-block:: llvm
1638 class SubtargetFeature<string n, string a, string v, string d,
1639 list<SubtargetFeature> i = []> {
1641 string Attribute = a;
1644 list<SubtargetFeature> Implies = i;
1647 In the ``Sparc.td`` file, the ``SubtargetFeature`` is used to define the
1650 .. code-block:: llvm
1652 def FeatureV9 : SubtargetFeature<"v9", "IsV9", "true",
1653 "Enable SPARC-V9 instructions">;
1654 def FeatureV8Deprecated : SubtargetFeature<"deprecated-v8",
1655 "V8DeprecatedInsts", "true",
1656 "Enable deprecated V8 instructions in V9 mode">;
1657 def FeatureVIS : SubtargetFeature<"vis", "IsVIS", "true",
1658 "Enable UltraSPARC Visual Instruction Set extensions">;
1660 Elsewhere in ``Sparc.td``, the ``Proc`` class is defined and then is used to
1661 define particular SPARC processor subtypes that may have the previously
1664 .. code-block:: llvm
1666 class Proc<string Name, list<SubtargetFeature> Features>
1667 : Processor<Name, NoItineraries, Features>;
1669 def : Proc<"generic", []>;
1670 def : Proc<"v8", []>;
1671 def : Proc<"supersparc", []>;
1672 def : Proc<"sparclite", []>;
1673 def : Proc<"f934", []>;
1674 def : Proc<"hypersparc", []>;
1675 def : Proc<"sparclite86x", []>;
1676 def : Proc<"sparclet", []>;
1677 def : Proc<"tsc701", []>;
1678 def : Proc<"v9", [FeatureV9]>;
1679 def : Proc<"ultrasparc", [FeatureV9, FeatureV8Deprecated]>;
1680 def : Proc<"ultrasparc3", [FeatureV9, FeatureV8Deprecated]>;
1681 def : Proc<"ultrasparc3-vis", [FeatureV9, FeatureV8Deprecated, FeatureVIS]>;
1683 From ``Target.td`` and ``Sparc.td`` files, the resulting
1684 ``SparcGenSubtarget.inc`` specifies enum values to identify the features,
1685 arrays of constants to represent the CPU features and CPU subtypes, and the
1686 ``ParseSubtargetFeatures`` method that parses the features string that sets
1687 specified subtarget options. The generated ``SparcGenSubtarget.inc`` file
1688 should be included in the ``SparcSubtarget.cpp``. The target-specific
1689 implementation of the ``XXXSubtarget`` method should follow this pseudocode:
1693 XXXSubtarget::XXXSubtarget(const Module &M, const std::string &FS) {
1694 // Set the default features
1695 // Determine default and user specified characteristics of the CPU
1696 // Call ParseSubtargetFeatures(FS, CPU) to parse the features string
1697 // Perform any additional operations
1703 The implementation of a target machine optionally includes a Just-In-Time (JIT)
1704 code generator that emits machine code and auxiliary structures as binary
1705 output that can be written directly to memory. To do this, implement JIT code
1706 generation by performing the following steps:
1708 * Write an ``XXXCodeEmitter.cpp`` file that contains a machine function pass
1709 that transforms target-machine instructions into relocatable machine
1712 * Write an ``XXXJITInfo.cpp`` file that implements the JIT interfaces for
1713 target-specific code-generation activities, such as emitting machine code and
1716 * Modify ``XXXTargetMachine`` so that it provides a ``TargetJITInfo`` object
1717 through its ``getJITInfo`` method.
1719 There are several different approaches to writing the JIT support code. For
1720 instance, TableGen and target descriptor files may be used for creating a JIT
1721 code generator, but are not mandatory. For the Alpha and PowerPC target
1722 machines, TableGen is used to generate ``XXXGenCodeEmitter.inc``, which
1723 contains the binary coding of machine instructions and the
1724 ``getBinaryCodeForInstr`` method to access those codes. Other JIT
1725 implementations do not.
1727 Both ``XXXJITInfo.cpp`` and ``XXXCodeEmitter.cpp`` must include the
1728 ``llvm/CodeGen/MachineCodeEmitter.h`` header file that defines the
1729 ``MachineCodeEmitter`` class containing code for several callback functions
1730 that write data (in bytes, words, strings, etc.) to the output stream.
1732 Machine Code Emitter
1733 --------------------
1735 In ``XXXCodeEmitter.cpp``, a target-specific of the ``Emitter`` class is
1736 implemented as a function pass (subclass of ``MachineFunctionPass``). The
1737 target-specific implementation of ``runOnMachineFunction`` (invoked by
1738 ``runOnFunction`` in ``MachineFunctionPass``) iterates through the
1739 ``MachineBasicBlock`` calls ``emitInstruction`` to process each instruction and
1740 emit binary code. ``emitInstruction`` is largely implemented with case
1741 statements on the instruction types defined in ``XXXInstrInfo.h``. For
1742 example, in ``X86CodeEmitter.cpp``, the ``emitInstruction`` method is built
1743 around the following ``switch``/``case`` statements:
1747 switch (Desc->TSFlags & X86::FormMask) {
1748 case X86II::Pseudo: // for not yet implemented instructions
1749 ... // or pseudo-instructions
1751 case X86II::RawFrm: // for instructions with a fixed opcode value
1754 case X86II::AddRegFrm: // for instructions that have one register operand
1755 ... // added to their opcode
1757 case X86II::MRMDestReg:// for instructions that use the Mod/RM byte
1758 ... // to specify a destination (register)
1760 case X86II::MRMDestMem:// for instructions that use the Mod/RM byte
1761 ... // to specify a destination (memory)
1763 case X86II::MRMSrcReg: // for instructions that use the Mod/RM byte
1764 ... // to specify a source (register)
1766 case X86II::MRMSrcMem: // for instructions that use the Mod/RM byte
1767 ... // to specify a source (memory)
1769 case X86II::MRM0r: case X86II::MRM1r: // for instructions that operate on
1770 case X86II::MRM2r: case X86II::MRM3r: // a REGISTER r/m operand and
1771 case X86II::MRM4r: case X86II::MRM5r: // use the Mod/RM byte and a field
1772 case X86II::MRM6r: case X86II::MRM7r: // to hold extended opcode data
1775 case X86II::MRM0m: case X86II::MRM1m: // for instructions that operate on
1776 case X86II::MRM2m: case X86II::MRM3m: // a MEMORY r/m operand and
1777 case X86II::MRM4m: case X86II::MRM5m: // use the Mod/RM byte and a field
1778 case X86II::MRM6m: case X86II::MRM7m: // to hold extended opcode data
1781 case X86II::MRMInitReg: // for instructions whose source and
1782 ... // destination are the same register
1786 The implementations of these case statements often first emit the opcode and
1787 then get the operand(s). Then depending upon the operand, helper methods may
1788 be called to process the operand(s). For example, in ``X86CodeEmitter.cpp``,
1789 for the ``X86II::AddRegFrm`` case, the first data emitted (by ``emitByte``) is
1790 the opcode added to the register operand. Then an object representing the
1791 machine operand, ``MO1``, is extracted. The helper methods such as
1792 ``isImmediate``, ``isGlobalAddress``, ``isExternalSymbol``,
1793 ``isConstantPoolIndex``, and ``isJumpTableIndex`` determine the operand type.
1794 (``X86CodeEmitter.cpp`` also has private methods such as ``emitConstant``,
1795 ``emitGlobalAddress``, ``emitExternalSymbolAddress``, ``emitConstPoolAddress``,
1796 and ``emitJumpTableAddress`` that emit the data into the output stream.)
1800 case X86II::AddRegFrm:
1801 MCE.emitByte(BaseOpcode + getX86RegNum(MI.getOperand(CurOp++).getReg()));
1803 if (CurOp != NumOps) {
1804 const MachineOperand &MO1 = MI.getOperand(CurOp++);
1805 unsigned Size = X86InstrInfo::sizeOfImm(Desc);
1806 if (MO1.isImmediate())
1807 emitConstant(MO1.getImm(), Size);
1809 unsigned rt = Is64BitMode ? X86::reloc_pcrel_word
1810 : (IsPIC ? X86::reloc_picrel_word : X86::reloc_absolute_word);
1811 if (Opcode == X86::MOV64ri)
1812 rt = X86::reloc_absolute_dword; // FIXME: add X86II flag?
1813 if (MO1.isGlobalAddress()) {
1814 bool NeedStub = isa<Function>(MO1.getGlobal());
1815 bool isLazy = gvNeedsLazyPtr(MO1.getGlobal());
1816 emitGlobalAddress(MO1.getGlobal(), rt, MO1.getOffset(), 0,
1818 } else if (MO1.isExternalSymbol())
1819 emitExternalSymbolAddress(MO1.getSymbolName(), rt);
1820 else if (MO1.isConstantPoolIndex())
1821 emitConstPoolAddress(MO1.getIndex(), rt);
1822 else if (MO1.isJumpTableIndex())
1823 emitJumpTableAddress(MO1.getIndex(), rt);
1828 In the previous example, ``XXXCodeEmitter.cpp`` uses the variable ``rt``, which
1829 is a ``RelocationType`` enum that may be used to relocate addresses (for
1830 example, a global address with a PIC base offset). The ``RelocationType`` enum
1831 for that target is defined in the short target-specific ``XXXRelocations.h``
1832 file. The ``RelocationType`` is used by the ``relocate`` method defined in
1833 ``XXXJITInfo.cpp`` to rewrite addresses for referenced global symbols.
1835 For example, ``X86Relocations.h`` specifies the following relocation types for
1836 the X86 addresses. In all four cases, the relocated value is added to the
1837 value already in memory. For ``reloc_pcrel_word`` and ``reloc_picrel_word``,
1838 there is an additional initial adjustment.
1842 enum RelocationType {
1843 reloc_pcrel_word = 0, // add reloc value after adjusting for the PC loc
1844 reloc_picrel_word = 1, // add reloc value after adjusting for the PIC base
1845 reloc_absolute_word = 2, // absolute relocation; no additional adjustment
1846 reloc_absolute_dword = 3 // absolute relocation; no additional adjustment
1852 ``XXXJITInfo.cpp`` implements the JIT interfaces for target-specific
1853 code-generation activities, such as emitting machine code and stubs. At
1854 minimum, a target-specific version of ``XXXJITInfo`` implements the following:
1856 * ``getLazyResolverFunction`` --- Initializes the JIT, gives the target a
1857 function that is used for compilation.
1859 * ``emitFunctionStub`` --- Returns a native function with a specified address
1860 for a callback function.
1862 * ``relocate`` --- Changes the addresses of referenced globals, based on
1865 * Callback function that are wrappers to a function stub that is used when the
1866 real target is not initially known.
1868 ``getLazyResolverFunction`` is generally trivial to implement. It makes the
1869 incoming parameter as the global ``JITCompilerFunction`` and returns the
1870 callback function that will be used a function wrapper. For the Alpha target
1871 (in ``AlphaJITInfo.cpp``), the ``getLazyResolverFunction`` implementation is
1876 TargetJITInfo::LazyResolverFn AlphaJITInfo::getLazyResolverFunction(
1878 JITCompilerFunction = F;
1879 return AlphaCompilationCallback;
1882 For the X86 target, the ``getLazyResolverFunction`` implementation is a little
1883 more complicated, because it returns a different callback function for
1884 processors with SSE instructions and XMM registers.
1886 The callback function initially saves and later restores the callee register
1887 values, incoming arguments, and frame and return address. The callback
1888 function needs low-level access to the registers or stack, so it is typically
1889 implemented with assembler.