Warning: This is a work in progress.
The LLVM target-independent code generator is a framework that provides a suite of reusable components for translating the LLVM internal representation to the machine code for a specified target -- either in assembly form (suitable for a static compiler) or in binary machine code format (usable for a JIT compiler). The LLVM target-independent code generator consists of five main components:
Depending on which part of the code generator you are interested in working on, different pieces of this will be useful to you. In any case, you should be familiar with the target description and machine code representation classes. If you want to add a backend for a new target, you will need to implement the target description classes for your new target and understand the LLVM code representation. If you are interested in implementing a new code generation algorithm, it should only depend on the target-description and machine code representation classes, ensuring that it is portable.
The two pieces of the LLVM code generator are the high-level interface to the code generator and the set of reusable components that can be used to build target-specific backends. The two most important interfaces (TargetMachine and TargetData) are the only ones that are required to be defined for a backend to fit into the LLVM system, but the others must be defined if the reusable code generator components are going to be used.
This design has two important implications. The first is that LLVM can support completely non-traditional code generation targets. For example, the C backend does not require register allocation, instruction selection, or any of the other standard components provided by the system. As such, it only implements these two interfaces, and does its own thing. Another example of a code generator like this is a (purely hypothetical) backend that converts LLVM to the GCC RTL form and uses GCC to emit machine code for a target.
This design also implies that it is possible to design and implement radically different code generators in the LLVM system that do not make use of any of the built-in components. Doing so is not recommended at all, but could be required for radically different targets that do not fit into the LLVM machine description model: programmable FPGAs for example.
The LLVM target-independent code generator is designed to support efficient and quality code generation for standard register-based microprocessors. Code generation in this model is divided into the following stages:
The code generator is based on the assumption that the instruction selector will use an optimal pattern matching selector to create high-quality sequences of native instructions. Alternative code generator designs based on pattern expansion and aggressive iterative peephole optimization are much slower. This design permits efficient compilation (important for JIT environments) and aggressive optimization (used when generating code offline) by allowing components of varying levels of sophistication to be used for any step of compilation.
In addition to these stages, target implementations can insert arbitrary target-specific passes into the flow. For example, the X86 target uses a special pass to handle the 80x87 floating point stack architecture. Other targets with unusual requirements can be supported with custom passes as needed.
The target description classes require a detailed description of the target architecture. These target descriptions often have a large amount of common information (e.g., an add instruction is almost identical to a sub instruction). In order to allow the maximum amount of commonality to be factored out, the LLVM code generator uses the TableGen tool to describe big chunks of the target machine, which allows the use of domain-specific and target-specific abstractions to reduce the amount of repetition.
As LLVM continues to be developed and refined, we plan to move more and more of the target description to be in .td form. Doing so gives us a number of advantages. The most important is that it makes it easier to port LLVM, because it reduces the amount of C++ code that has to be written and the surface area of the code generator that needs to be understood before someone can get in an get something working. Second, it is also important to us because it makes it easier to change things: in particular, if tables and other things are all emitted by tblgen, we only need to change one place (tblgen) to update all of the targets to a new interface.
The LLVM target description classes (which are located in the include/llvm/Target directory) provide an abstract description of the target machine; independent of any particular client. These classes are designed to capture the abstract properties of the target (such as the instructions and registers it has), and do not incorporate any particular pieces of code generation algorithms.
All of the target description classes (except the TargetData class) are designed to be subclassed by the concrete target implementation, and have virtual methods implemented. To get to these implementations, the TargetMachine class provides accessors that should be implemented by the target.
The TargetMachine class provides virtual methods that are used to access the target-specific implementations of the various target description classes via the get*Info methods (getInstrInfo, getRegisterInfo, getFrameInfo, etc.). This class is designed to be specialized by a concrete target implementation (e.g., X86TargetMachine) which implements the various virtual methods. The only required target description class is the TargetData class, but if the code generator components are to be used, the other interfaces should be implemented as well.
The TargetData class is the only required target description class, and it is the only class that is not extensible (you cannot derived a new class from it). TargetData specifies information about how the target lays out memory for structures, the alignment requirements for various data types, the size of pointers in the target, and whether the target is little-endian or big-endian.
The TargetLowering class is used by SelectionDAG based instruction selectors primarily to describe how LLVM code should be lowered to SelectionDAG operations. Among other things, this class indicates:
The MRegisterInfo class (which will eventually be renamed to TargetRegisterInfo) is used to describe the register file of the target and any interactions between the registers.
Registers in the code generator are represented in the code generator by unsigned numbers. Physical registers (those that actually exist in the target description) are unique small numbers, and virtual registers are generally large. Note that register #0 is reserved as a flag value.
Each register in the processor description has an associated TargetRegisterDesc entry, which provides a textual name for the register (used for assembly output and debugging dumps) and a set of aliases (used to indicate that one register overlaps with another).
In addition to the per-register description, the MRegisterInfo class exposes a set of processor specific register classes (instances of the TargetRegisterClass class). Each register class contains sets of registers that have the same properties (for example, they are all 32-bit integer registers). Each SSA virtual register created by the instruction selector has an associated register class. When the register allocator runs, it replaces virtual registers with a physical register in the set.
The target-specific implementations of these classes is auto-generated from a TableGen description of the register file.
The TargetInstrInfo class is used to describe the machine instructions supported by the target. It is essentially an array of TargetInstrDescriptor objects, each of which describes one instruction the target supports. Descriptors define things like the mnemonic for the opcode, the number of operands, the list of implicit register uses and defs, whether the instruction has certain target-independent properties (accesses memory, is commutable, etc), and holds any target-specific flags.
The TargetFrameInfo class is used to provide information about the stack frame layout of the target. It holds the direction of stack growth, the known stack alignment on entry to each function, and the offset to the locals area. The offset to the local area is the offset from the stack pointer on function entry to the first location where function data (local variables, spill locations) can be stored.
The TargetSubtarget class is used to provide information about the specific chip set being targeted. A sub-target informs code generation of which instructions are supported, instruction latencies and instruction execution itinerary; i.e., which processing units are used, in what order, and for how long.
At the high-level, LLVM code is translated to a machine specific representation formed out of MachineFunction, MachineBasicBlock, and MachineInstr instances (defined in include/llvm/CodeGen). This representation is completely target agnostic, representing instructions in their most abstract form: an opcode and a series of operands. This representation is designed to support both SSA representation for machine code, as well as a register allocated, non-SSA form.
Target machine instructions are represented as instances of the MachineInstr class. This class is an extremely abstract way of representing machine instructions. In particular, it only keeps track of an opcode number and a set of operands.
The opcode number is a simple unsigned number that only has meaning to a specific backend. All of the instructions for a target should be defined in the *InstrInfo.td file for the target. The opcode enum values are auto-generated from this description. The MachineInstr class does not have any information about how to interpret the instruction (i.e., what the semantics of the instruction are): for that you must refer to the TargetInstrInfo class.
The operands of a machine instruction can be of several different types: they can be a register reference, constant integer, basic block reference, etc. In addition, a machine operand should be marked as a def or a use of the value (though only registers are allowed to be defs).
By convention, the LLVM code generator orders instruction operands so that all register definitions come before the register uses, even on architectures that are normally printed in other orders. For example, the SPARC add instruction: "add %i1, %i2, %i3" adds the "%i1", and "%i2" registers and stores the result into the "%i3" register. In the LLVM code generator, the operands should be stored as "%i3, %i1, %i2": with the destination first.
Keeping destination (definition) operands at the beginning of the operand list has several advantages. In particular, the debugging printer will print the instruction like this:
%r3 = add %i1, %i2
If the first operand is a def, and it is also easier to create instructions whose only def is the first operand.
Machine instructions are created by using the BuildMI functions, located in the include/llvm/CodeGen/MachineInstrBuilder.h file. The BuildMI functions make it easy to build arbitrary machine instructions. Usage of the BuildMI functions look like this:
// Create a 'DestReg = mov 42' (rendered in X86 assembly as 'mov DestReg, 42') // instruction. The '1' specifies how many operands will be added. MachineInstr *MI = BuildMI(X86::MOV32ri, 1, DestReg).addImm(42); // Create the same instr, but insert it at the end of a basic block. MachineBasicBlock &MBB = ... BuildMI(MBB, X86::MOV32ri, 1, DestReg).addImm(42); // Create the same instr, but insert it before a specified iterator point. MachineBasicBlock::iterator MBBI = ... BuildMI(MBB, MBBI, X86::MOV32ri, 1, DestReg).addImm(42); // Create a 'cmp Reg, 0' instruction, no destination reg. MI = BuildMI(X86::CMP32ri, 2).addReg(Reg).addImm(0); // Create an 'sahf' instruction which takes no operands and stores nothing. MI = BuildMI(X86::SAHF, 0); // Create a self looping branch instruction. BuildMI(MBB, X86::JNE, 1).addMBB(&MBB);
The key thing to remember with the BuildMI functions is that you have to specify the number of operands that the machine instruction will take. This allows for efficient memory allocation. You also need to specify if operands default to be uses of values, not definitions. If you need to add a definition operand (other than the optional destination register), you must explicitly mark it as such.
One important issue that the code generator needs to be aware of is the presence of fixed registers. In particular, there are often places in the instruction stream where the register allocator must arrange for a particular value to be in a particular register. This can occur due to limitations of the instruction set (e.g., the X86 can only do a 32-bit divide with the EAX/EDX registers), or external factors like calling conventions. In any case, the instruction selector should emit code that copies a virtual register into or out of a physical register when needed.
For example, consider this simple LLVM example:
int %test(int %X, int %Y) {
%Z = div int %X, %Y
ret int %Z
}
The X86 instruction selector produces this machine code for the div and ret (use "llc X.bc -march=x86 -print-machineinstrs" to get this):
;; Start of div
%EAX = mov %reg1024 ;; Copy X (in reg1024) into EAX
%reg1027 = sar %reg1024, 31
%EDX = mov %reg1027 ;; Sign extend X into EDX
idiv %reg1025 ;; Divide by Y (in reg1025)
%reg1026 = mov %EAX ;; Read the result (Z) out of EAX
;; Start of ret
%EAX = mov %reg1026 ;; 32-bit return value goes in EAX
ret
By the end of code generation, the register allocator has coalesced the registers and deleted the resultant identity moves, producing the following code:
;; X is in EAX, Y is in ECX
mov %EAX, %EDX
sar %EDX, 31
idiv %ECX
ret
This approach is extremely general (if it can handle the X86 architecture, it can handle anything!) and allows all of the target specific knowledge about the instruction stream to be isolated in the instruction selector. Note that physical registers should have a short lifetime for good code generation, and all physical registers are assumed dead on entry and exit of basic blocks (before register allocation). Thus if you need a value to be live across basic block boundaries, it must live in a virtual register.
MachineInstr's are initially selected in SSA-form, and are maintained in SSA-form until register allocation happens. For the most part, this is trivially simple since LLVM is already in SSA form: LLVM PHI nodes become machine code PHI nodes, and virtual registers are only allowed to have a single definition.
After register allocation, machine code is no longer in SSA-form, as there are no virtual registers left in the code.
The MachineBasicBlock class contains a list of machine instructions (MachineInstr instances). It roughly corresponds to the LLVM code input to the instruction selector, but there can be a one-to-many mapping (i.e. one LLVM basic block can map to multiple machine basic blocks). The MachineBasicBlock class has a "getBasicBlock" method, which returns the LLVM basic block that it comes from.
The MachineFunction class contains a list of machine basic blocks (MachineBasicBlock instances). It corresponds one-to-one with the LLVM function input to the instruction selector. In addition to a list of basic blocks, the MachineFunction contains a the MachineConstantPool, MachineFrameInfo, MachineFunctionInfo, SSARegMap, and a set of live in and live out registers for the function. See MachineFunction.h for more information.
This section documents the phases described in the high-level design of the code generator. It explains how they work and some of the rationale behind their design.
Instruction Selection is the process of translating LLVM code presented to the code generator into target-specific machine instructions. There are several well-known ways to do this in the literature. In LLVM there are two main forms: the SelectionDAG based instruction selector framework and an old-style 'simple' instruction selector (which effectively peephole selects each LLVM instruction into a series of machine instructions). We recommend that all targets use the SelectionDAG infrastructure.
Portions of the DAG instruction selector are generated from the target description files (*.td) files. Eventually, we aim for the entire instruction selector to be generated from these .td files.
The SelectionDAG provides an abstraction for code representation in a way that is amenable to instruction selection using automatic techniques (e.g. dynamic-programming based optimal pattern matching selectors), It is also well suited to other phases of code generation; in particular, instruction scheduling (SelectionDAG's are very close to scheduling DAGs post-selection). Additionally, the SelectionDAG provides a host representation where a large variety of very-low-level (but target-independent) optimizations may be performed: ones which require extensive information about the instructions efficiently supported by the target.
The SelectionDAG is a Directed-Acyclic-Graph whose nodes are instances of the SDNode class. The primary payload of the SDNode is its operation code (Opcode) that indicates what operation the node performs and the operands to the operation. The various operation node types are described at the top of the include/llvm/CodeGen/SelectionDAGNodes.h file.
Although most operations define a single value, each node in the graph may define multiple values. For example, a combined div/rem operation will define both the dividend and the remainder. Many other situations require multiple values as well. Each node also has some number of operands, which are edges to the node defining the used value. Because nodes may define multiple values, edges are represented by instances of the SDOperand class, which is a <SDNode, unsigned> pair, indicating the node and result value being used, respectively. Each value produced by an SDNode has an associated MVT::ValueType, indicating what type the value is.
SelectionDAGs contain two different kinds of values: those that represent data flow and those that represent control flow dependencies. Data values are simple edges with an integer or floating point value type. Control edges are represented as "chain" edges which are of type MVT::Other. These edges provide an ordering between nodes that have side effects (such as loads/stores/calls/return/etc). All nodes that have side effects should take a token chain as input and produce a new one as output. By convention, token chain inputs are always operand #0, and chain results are always the last value produced by an operation.
A SelectionDAG has designated "Entry" and "Root" nodes. The Entry node is always a marker node with an Opcode of ISD::EntryToken. The Root node is the final side-effecting node in the token chain. For example, in a single basic block function, this would be the return node.
One important concept for SelectionDAGs is the notion of a "legal" vs. "illegal" DAG. A legal DAG for a target is one that only uses supported operations and supported types. On a 32-bit PowerPC, for example, a DAG with any values of i1, i8, i16, or i64 type would be illegal, as would a DAG that uses a SREM or UREM operation. The legalize phase is responsible for turning an illegal DAG into a legal DAG.
SelectionDAG-based instruction selection consists of the following steps:
After all of these steps are complete, the SelectionDAG is destroyed and the rest of the code generation passes are run.
One great way to visualize what is going on here is to take advantage of a few LLC command line options. In particular, the -view-isel-dags option pops up a window with the SelectionDAG input to the Select phase for all of the code compiled (if you only get errors printed to the console while using this, you probably need to configure your system to add support for it). The -view-sched-dags option views the SelectionDAG output from the Select phase and input to the Scheduler phase.
The initial SelectionDAG is naively peephole expanded from the LLVM input by the SelectionDAGLowering class in the SelectionDAGISel.cpp file. The intent of this pass is to expose as much low-level, target-specific details to the SelectionDAG as possible. This pass is mostly hard-coded (e.g. an LLVM add turns into an SDNode add while a geteelementptr is expanded into the obvious arithmetic). This pass requires target-specific hooks to lower calls and returns, varargs, etc. For these features, the TargetLowering interface is used.
The Legalize phase is in charge of converting a DAG to only use the types and operations that are natively supported by the target. This involves two major tasks:
Convert values of unsupported types to values of supported types.
There are two main ways of doing this: converting small types to larger types ("promoting"), and breaking up large integer types into smaller ones ("expanding"). For example, a target might require that all f32 values are promoted to f64 and that all i1/i8/i16 values are promoted to i32. The same target might require that all i64 values be expanded into i32 values. These changes can insert sign and zero extensions as needed to make sure that the final code has the same behavior as the input.
A target implementation tells the legalizer which types are supported (and which register class to use for them) by calling the "addRegisterClass" method in its TargetLowering constructor.
Eliminate operations that are not supported by the target.
Targets often have weird constraints, such as not supporting every operation on every supported datatype (e.g. X86 does not support byte conditional moves and PowerPC does not support sign-extending loads from a 16-bit memory location). Legalize takes care by open-coding another sequence of operations to emulate the operation ("expansion"), by promoting to a larger type that supports the operation (promotion), or using a target-specific hook to implement the legalization (custom).
A target implementation tells the legalizer which operations are not supported (and which of the above three actions to take) by calling the "setOperationAction" method in its TargetLowering constructor.
Prior to the existance of the Legalize pass, we required that every target selector supported and handled every operator and type even if they are not natively supported. The introduction of the Legalize phase allows all of the cannonicalization patterns to be shared across targets, and makes it very easy to optimize the cannonicalized code because it is still in the form of a DAG.
The SelectionDAG optimization phase is run twice for code generation: once immediately after the DAG is built and once after legalization. The first run of the pass allows the initial code to be cleaned up (e.g. performing optimizations that depend on knowing that the operators have restricted type inputs). The second run of the pass cleans up the messy code generated by the Legalize pass, which allows Legalize to be very simple (it can focus on making code legal instead of focusing on generating good and legal code).
One important class of optimizations performed is optimizing inserted sign and zero extension instructions. We currently use ad-hoc techniques, but could move to more rigorous techniques in the future. Here are some good papers on the subject:
"Widening
integer arithmetic"
Kevin Redwine and Norman Ramsey
International Conference on Compiler Construction (CC) 2004
"Effective
sign extension elimination"
Motohiro Kawahito, Hideaki Komatsu, and Toshio Nakatani
Proceedings of the ACM SIGPLAN 2002 Conference on Programming Language Design
and Implementation.
The Select phase is the bulk of the target-specific code for instruction selection. This phase takes a legal SelectionDAG as input, pattern matches the instructions supported by the target to this DAG, and produces a new DAG of target code. For example, consider the following LLVM fragment:
%t1 = add float %W, %X %t2 = mul float %t1, %Y %t3 = add float %t2, %Z
This LLVM code corresponds to a SelectionDAG that looks basically like this:
(fadd:f32 (fmul:f32 (fadd:f32 W, X), Y), Z)
If a target supports floating point multiply-and-add (FMA) operations, one of the adds can be merged with the multiply. On the PowerPC, for example, the output of the instruction selector might look like this DAG:
(FMADDS (FADDS W, X), Y, Z)
The FMADDS instruction is a ternary instruction that multiplies its first two operands and adds the third (as single-precision floating-point numbers). The FADDS instruction is a simple binary single-precision add instruction. To perform this pattern match, the PowerPC backend includes the following instruction definitions:
def FMADDS : AForm_1<59, 29,
(ops F4RC:$FRT, F4RC:$FRA, F4RC:$FRC, F4RC:$FRB),
"fmadds $FRT, $FRA, $FRC, $FRB",
[(set F4RC:$FRT, (fadd (fmul F4RC:$FRA, F4RC:$FRC),
F4RC:$FRB))]>;
def FADDS : AForm_2<59, 21,
(ops F4RC:$FRT, F4RC:$FRA, F4RC:$FRB),
"fadds $FRT, $FRA, $FRB",
[(set F4RC:$FRT, (fadd F4RC:$FRA, F4RC:$FRB))]>;
The portion of the instruction definition in bold indicates the pattern used to match the instruction. The DAG operators (like fmul/fadd) are defined in the lib/Target/TargetSelectionDAG.td file. "F4RC" is the register class of the input and result values.
The TableGen DAG instruction selector generator reads the instruction patterns in the .td and automatically builds parts of the pattern matching code for your target. It has the following strengths:
// Arbitrary immediate support. Implement in terms of LIS/ORI.
def : Pat<(i32 imm:$imm),
(ORI (LIS (HI16 imm:$imm)), (LO16 imm:$imm))>;
If none of the single-instruction patterns for loading an immediate into a
register match, this will be used. This rule says "match an arbitrary i32
immediate, turning it into an ORI ('or a 16-bit immediate') and an LIS
('load 16-bit immediate, where the immediate is shifted to the left 16
bits') instruction". To make this work, the LO16/HI16 node transformations
are used to manipulate the input immediate (in this case, take the high or
low 16-bits of the immediate).
While it has many strengths, the system currently has some limitations, primarily because it is a work in progress and is not yet finished:
Despite these limitations, the instruction selector generator is still quite useful for most of the binary and logical operations in typical instruction sets. If you run into any problems or can't figure out how to do something, please let Chris know!
The scheduling phase takes the DAG of target instructions from the selection phase and assigns an order. The scheduler can pick an order depending on various constraints of the machines (i.e. order for minimal register pressure or try to cover instruction latencies). Once an order is established, the DAG is converted to a list of MachineInstrs and the Selection DAG is destroyed.
Note that this phase is logically separate from the instruction selection phase, but is tied to it closely in the code because it operates on SelectionDAGs.
To Be Written
To Be Written
To Be Written
To Be Written
For the JIT or .o file writer
This section of the document explains features or design decisions that are specific to the code generator for a particular target.
The X86 code generator lives in the lib/Target/X86 directory. This code generator currently targets a generic P6-like processor. As such, it produces a few P6-and-above instructions (like conditional moves), but it does not make use of newer features like MMX or SSE. In the future, the X86 backend will have sub-target support added for specific processor families and implementations.
The following are the known target triples that are supported by the X86 backend. This is not an exhaustive list, but it would be useful to add those that people test.
The x86 has a very flexible way of accessing memory. It is capable of forming memory addresses of the following expression directly in integer instructions (which use ModR/M addressing):
Base+[1,2,4,8]*IndexReg+Disp32
In order to represent this, LLVM tracks no less than 4 operands for each memory operand of this form. This means that the "load" form of 'mov' has the following MachineOperands in this order:
Index: 0 | 1 2 3 4 Meaning: DestReg, | BaseReg, Scale, IndexReg, Displacement OperandTy: VirtReg, | VirtReg, UnsImm, VirtReg, SignExtImm
Stores, and all other instructions, treat the four memory operands in the same way, in the same order.
An instruction name consists of the base name, a default operand size, and a a character per operand with an optional special size. For example:
ADD8rr -> add, 8-bit register, 8-bit register
IMUL16rmi -> imul, 16-bit register, 16-bit memory, 16-bit immediate
IMUL16rmi8 -> imul, 16-bit register, 16-bit memory, 8-bit immediate
MOVSX32rm16 -> movsx, 32-bit register, 16-bit memory