TableGen backends are at the core of TableGen’s functionality. The source files provide the classes and records that are parsed and end up as a collection of record instances, but it’s up to the backend to interpret and print the records in a way that is meaningful to the user (normally a C++ include file or a textual list of warnings, options, and error messages).
TableGen is used by both LLVM, Clang, and MLIR with very different goals. LLVM uses it as a way to automate the generation of massive amounts of information regarding instructions, schedules, cores, and architecture features. Some backends generate output that is consumed by more than one source file, so they need to be created in a way that makes it is easy for preprocessor tricks to be used. Some backends can also print C++ code structures, so that they can be directly included as-is.
Clang, on the other hand, uses it mainly for diagnostic messages (errors, warnings, tips) and attributes, so more on the textual end of the scale.
MLIR uses TableGen to define operations, operation dialects, and operation traits.
See the TableGen Programmer’s Reference for an in-depth description of TableGen, and the TableGen Backend Developer’s Guide for a guide to writing a new backend.
Warning
This portion is incomplete. Each section below needs three subsections: description of its purpose with a list of users, output generated from generic input, and finally why it needed a new backend (in case there’s something similar).
Overall, each backend will take the same TableGen file type and transform into similar output for different targets/uses. There is an implicit contract between the TableGen files, the back-ends and their users.
For instance, a global contract is that each back-end produces macro-guarded sections. Based on whether the file is included by a header or a source file, or even in which context of each file the include is being used, you have todefine a macro just before including it, to get the right output:
#define GET_REGINFO_TARGET_DESC
#include "ARMGenRegisterInfo.inc"
And just part of the generated file would be included. This is useful if you need the same information in multiple formats (instantiation, initialization, getter/setter functions, etc) from the same source TableGen file without having to re-compile the TableGen file multiple times.
Sometimes, multiple macros might be defined before the same include file to output multiple blocks:
#define GET_REGISTER_MATCHER
#define GET_SUBTARGET_FEATURE_NAME
#define GET_MATCHER_IMPLEMENTATION
#include "ARMGenAsmMatcher.inc"
The macros will be undef’d automatically as they’re used, in the include file.
On all LLVM back-ends, the llvm-tblgen binary will be executed on the root TableGen file <Target>.td, which should include all others. This guarantees that all information needed is accessible, and that no duplication is needed in the TableGen files.
Purpose: CodeEmitterGen uses the descriptions of instructions and their fields to construct an automated code emitter: a function that, given a MachineInstr, returns the (currently, 32-bit unsigned) value of the instruction.
Output: C++ code, implementing the target’s CodeEmitter class by overriding the virtual functions as <Target>CodeEmitter::function().
Usage: Used to include directly at the end of <Target>MCCodeEmitter.cpp.
Purpose: This tablegen backend is responsible for emitting a description of a target register file for a code generator. It uses instances of the Register, RegisterAliases, and RegisterClass classes to gather this information.
Output: C++ code with enums and structures representing the register mappings, properties, masks, etc.
Usage: Both on <Target>BaseRegisterInfo and <Target>MCTargetDesc (headers and source files) with macros defining in which they are for declaration vs. initialization issues.
Purpose: This tablegen backend is responsible for emitting a description of the target instruction set for the code generator. (what are the differences from CodeEmitter?)
Output: C++ code with enums and structures representing the instruction mappings, properties, masks, etc.
Usage: Both on <Target>BaseInstrInfo and <Target>MCTargetDesc (headers and source files) with macros defining in which they are for declaration vs. initialization issues.
Purpose: Emits an assembly printer for the current target.
Output: Implementation of <Target>InstPrinter::printInstruction(), among other things.
Usage: Included directly into InstPrinter/<Target>InstPrinter.cpp.
Purpose: Emits a target specifier matcher for converting parsed assembly operands in the MCInst structures. It also emits a matcher for custom operand parsing. Extensive documentation is written on the AsmMatcherEmitter.cpp file.
Output: Assembler parsers’ matcher functions, declarations, etc.
Usage: Used in back-ends’ AsmParser/<Target>AsmParser.cpp for building the AsmParser class.
Purpose: Contains disassembler table emitters for various architectures. Extensive documentation is written on the DisassemblerEmitter.cpp file.
Output: Decoding tables, static decoding functions, etc.
Usage: Directly included in Disassembler/<Target>Disassembler.cpp to cater for all default decodings, after all hand-made ones.
Purpose: Generate pseudo instruction lowering.
Output: Implements <Target>AsmPrinter::emitPseudoExpansionLowering().
Usage: Included directly into <Target>AsmPrinter.cpp.
Purpose: Responsible for emitting descriptions of the calling conventions supported by this target.
Output: Implement static functions to deal with calling conventions chained by matching styles, returning false on no match.
Usage: Used in ISelLowering and FastIsel as function pointers to implementation returned by a CC selection function.
Purpose: Generate a DAG instruction selector.
Output: Creates huge functions for automating DAG selection.
Usage: Included in <Target>ISelDAGToDAG.cpp inside the target’s implementation of SelectionDAGISel.
Purpose: This class parses the Schedule.td file and produces an API that can be used to reason about whether an instruction can be added to a packet on a VLIW architecture. The class internally generates a deterministic finite automaton (DFA) that models all possible mappings of machine instructions to functional units as instructions are added to a packet.
Output: Scheduling tables for GPU back-ends (Hexagon, AMD).
Usage: Included directly on <Target>InstrInfo.cpp.
Purpose: This tablegen backend emits code for use by the “fast” instruction selection algorithm. See the comments at the top of lib/CodeGen/SelectionDAG/FastISel.cpp for background. This file scans through the target’s tablegen instruction-info files and extracts instructions with obvious-looking patterns, and it emits code to look up these instructions by type and operator.
Output: Generates Predicate and FastEmit methods.
Usage: Implements private methods of the targets’ implementation of FastISel class.
Purpose: Generate subtarget enumerations.
Output: Enums, globals, local tables for sub-target information.
Usage: Populates <Target>Subtarget and MCTargetDesc/<Target>MCTargetDesc files (both headers and source).
Purpose: Print enum values for a class.
Purpose: Generate custom searchable tables.
Output: Enums, global tables, and lookup helper functions.
Usage: This backend allows generating free-form, target-specific tables from TableGen records. The ARM and AArch64 targets use this backend to generate tables of system registers; the AMDGPU target uses it to generate meta-data about complex image and memory buffer instructions.
See SearchableTables Reference for a detailed description.
Purpose: This tablegen backend emits an index of definitions in ctags(1) format. A helper script, utils/TableGen/tdtags, provides an easier-to-use interface; run ‘tdtags -H’ for documentation.
Purpose: This X86 specific tablegen backend emits tables that map EVEX encoded instructions to their VEX encoded identical instruction.
Purpose: Creates Attrs.inc, which contains semantic attribute class declarations for any attribute in Attr.td that has not set ASTNode = 0. This file is included as part of Attr.h.
Purpose: Creates AttrParserStringSwitches.inc, which contains StringSwitch::Case statements for parser-related string switches. Each switch is given its own macro (such as CLANG_ATTR_ARG_CONTEXT_LIST, or CLANG_ATTR_IDENTIFIER_ARG_LIST), which is expected to be defined before including AttrParserStringSwitches.inc, and undefined after.
Purpose: Creates AttrImpl.inc, which contains semantic attribute class definitions for any attribute in Attr.td that has not set ASTNode = 0. This file is included as part of AttrImpl.cpp.
Purpose: Creates AttrList.inc, which is used when a list of semantic attribute identifiers is required. For instance, AttrKinds.h includes this file to generate the list of attr::Kind enumeration values. This list is separated out into multiple categories: attributes, inheritable attributes, and inheritable parameter attributes. This categorization happens automatically based on information in Attr.td and is used to implement the classof functionality required for dyn_cast and similar APIs.
Purpose: Creates AttrPCHRead.inc, which is used to deserialize attributes in the ASTReader::ReadAttributes function.
Purpose: Creates AttrPCHWrite.inc, which is used to serialize attributes in the ASTWriter::WriteAttributes function.
Purpose: Creates AttrSpellings.inc, which is used to implement the __has_attribute feature test macro.
Purpose: Creates AttrSpellingListIndex.inc, which is used to map parsed attribute spellings (including which syntax or scope was used) to an attribute spelling list index. These spelling list index values are internal implementation details exposed via AttributeList::getAttributeSpellingListIndex.
Purpose: Creates AttrVisitor.inc, which is used when implementing recursive AST visitors.
Purpose: Creates AttrTemplateInstantiate.inc, which implements the instantiateTemplateAttribute function, used when instantiating a template that requires an attribute to be cloned.
Purpose: Creates AttrParsedAttrList.inc, which is used to generate the AttributeList::Kind parsed attribute enumeration.
Purpose: Creates AttrParsedAttrImpl.inc, which is used by AttributeList.cpp to implement several functions on the AttributeList class. This functionality is implemented via the AttrInfoMap ParsedAttrInfo array, which contains one element per parsed attribute object.
Purpose: Creates AttrParsedAttrKinds.inc, which is used to implement the AttributeList::getKind function, mapping a string (and syntax) to a parsed attribute AttributeList::Kind enumeration.
Purpose: Creates AttrDump.inc, which dumps information about an attribute. It is used to implement ASTDumper::dumpAttr.
Generate Clang diagnostics definitions.
Generate Clang diagnostic groups.
Generate Clang diagnostic name index.
Generate Clang AST comment nodes.
Generate Clang AST declaration nodes.
Generate Clang AST statement nodes.
Generate Clang Static Analyzer checkers.
Generate efficient matchers for HTML tag names that are used in documentation comments.
Generate efficient matchers for HTML tag properties.
Generate function to translate named character references to UTF-8 sequences.
Generate command properties for commands that are used in documentation comments.
Generate list of commands that are used in documentation comments.
Generate ARM NEON sema support for clang.
Generate ARM NEON tests for clang.
The TableGen command option --print-records invokes a simple backend that prints all the classes and records defined in the source files. This is the default backend option. See the TableGen Backend Developer’s Guide for more information.
The TableGen command option --print-detailed-records invokes a backend that prints all the global variables, classes, and records defined in the source files, with more detail than the default record printer. See the TableGen Backend Developer’s Guide for more information.
Purpose: Output all the values in every def, as a JSON data structure that can be easily parsed by a variety of languages. Useful for writing custom backends without having to modify TableGen itself, or for performing auxiliary analysis on the same TableGen data passed to a built-in backend.
Output:
The root of the output file is a JSON object (i.e. dictionary), containing the following fixed keys:
For each def record, the root object also has a key for the record name. The corresponding value is a subsidiary object containing the following fixed keys:
For each variable defined in a record, the def object for that record also has a key for the variable name. The corresponding value is a translation into JSON of the variable’s value, using the conventions described below.
Some TableGen data types are translated directly into the corresponding JSON type:
All other TableGen value types are emitted as a JSON object, containing two standard fields: kind is a discriminator describing which kind of value the object represents, and printable is a string giving the same representation of the value that would appear in -print-records.
A TableGen include file, SearchableTable.td, provides classes for generating C++ searchable tables. These tables are described in the following sections. To generate the C++ code, run llvm-tblgen with the --gen-searchable-tables option, which invokes the backend that generates the tables from the records you provide.
Each of the data structures generated for searchable tables is guarded by an #ifdef. This allows you to include the generated .inc file and select only certain data structures for inclusion. The examples below show the macro names used in these guards.
The GenericEnum class makes it easy to define a C++ enumerated type and the enumerated elements of that type. To define the type, define a record whose parent class is GenericEnum and whose name is the desired enum type. This class provides three fields, which you can set in the record using the let statement.
Here is an example where the values of the elements are specified explicitly, as a template argument to the BEntry class. The resulting C++ code is shown.
def BValues : GenericEnum {
let FilterClass = "BEntry";
let NameField = "Name";
let ValueField = "Encoding";
}
class BEntry<bits<16> enc> {
string Name = NAME;
bits<16> Encoding = enc;
}
def BFoo : BEntry<0xac>;
def BBar : BEntry<0x14>;
def BZoo : BEntry<0x80>;
def BSnork : BEntry<0x4c>;
#ifdef GET_BValues_DECL
enum BValues {
BBar = 20,
BFoo = 172,
BSnork = 76,
BZoo = 128,
};
#endif
In the following example, the values of the elements are assigned automatically. Note that values are assigned from 0, in alphabetical order by element name.
def CEnum : GenericEnum {
let FilterClass = "CEnum";
}
class CEnum;
def CFoo : CEnum;
def CBar : CEnum;
def CBaz : CEnum;
#ifdef GET_CEnum_DECL
enum CEnum {
CBar = 0,
CBaz = 1,
CFoo = 2,
};
#endif
The GenericTable class is used to define a searchable generic table. TableGen produces C++ code to define the table entries and also produces the declaration and definition of a function to search the table based on a primary key. To define the table, define a record whose parent class is GenericTable and whose name is the name of the global table of entries. This class provides six fields.
TableGen attempts to deduce the type of each of the table fields so that it can format the C++ initializers in the emitted table. It can deduce bit, bits<n>, string, Intrinsic, and Instruction. These can be used in the primary key. Any other field types must be specified explicitly; this is done as shown in the second example below. Such fields cannot be used in the primary key.
One special case of the field type has to do with code. Arbitrary code is represented by a string, but has to be emitted as a C++ initializer without quotes. If the code field was defined using a code literal ([{...}]), then TableGen will know to emit it without quotes. However, if it was defined using a string literal or complex string expression, then TableGen will not know. In this case, you can force TableGen to treat the field as code by including the following line in the GenericTable record, where xxx is the code field name.
string TypeOf_xxx = "code";
Here is an example where TableGen can deduce the field types. Note that the table entry records are anonymous; the names of entry records are irrelevant.
def ATable : GenericTable {
let FilterClass = "AEntry";
let Fields = ["Str", "Val1", "Val2"];
let PrimaryKey = ["Val1", "Val2"];
let PrimaryKeyName = "lookupATableByValues";
}
class AEntry<string str, int val1, int val2> {
string Str = str;
bits<8> Val1 = val1;
bits<10> Val2 = val2;
}
def : AEntry<"Bob", 5, 3>;
def : AEntry<"Carol", 2, 6>;
def : AEntry<"Ted", 4, 4>;
def : AEntry<"Alice", 4, 5>;
def : AEntry<"Costa", 2, 1>;
Here is the generated C++ code. The declaration of lookupATableByValues is guarded by GET_ATable_DECL, while the definitions are guarded by GET_ATable_IMPL.
#ifdef GET_ATable_DECL
const AEntry *lookupATableByValues(uint8_t Val1, uint16_t Val2);
#endif
#ifdef GET_ATable_IMPL
constexpr AEntry ATable[] = {
{ "Costa", 0x2, 0x1 }, // 0
{ "Carol", 0x2, 0x6 }, // 1
{ "Ted", 0x4, 0x4 }, // 2
{ "Alice", 0x4, 0x5 }, // 3
{ "Bob", 0x5, 0x3 }, // 4
};
const AEntry *lookupATableByValues(uint8_t Val1, uint16_t Val2) {
struct KeyType {
uint8_t Val1;
uint16_t Val2;
};
KeyType Key = { Val1, Val2 };
auto Table = makeArrayRef(ATable);
auto Idx = std::lower_bound(Table.begin(), Table.end(), Key,
[](const AEntry &LHS, const KeyType &RHS) {
if (LHS.Val1 < RHS.Val1)
return true;
if (LHS.Val1 > RHS.Val1)
return false;
if (LHS.Val2 < RHS.Val2)
return true;
if (LHS.Val2 > RHS.Val2)
return false;
return false;
});
if (Idx == Table.end() ||
Key.Val1 != Idx->Val1 ||
Key.Val2 != Idx->Val2)
return nullptr;
return &*Idx;
}
#endif
The table entries in ATable are sorted in order by Val1, and within each of those values, by Val2. This allows a binary search of the table, which is performed in the lookup function by std::lower_bound. The lookup function returns a reference to the found table entry, or the null pointer if no entry is found.
This example includes a field whose type TableGen cannot deduce. The Kind field uses the enumerated type CEnum defined above. To inform TableGen of the type, the record derived from GenericTable must include a string field named TypeOf_field, where field is the name of the field whose type is required.
def CTable : GenericTable {
let FilterClass = "CEntry";
let Fields = ["Name", "Kind", "Encoding"];
string TypeOf_Kind = "CEnum";
let PrimaryKey = ["Encoding"];
let PrimaryKeyName = "lookupCEntryByEncoding";
}
class CEntry<string name, CEnum kind, int enc> {
string Name = name;
CEnum Kind = kind;
bits<16> Encoding = enc;
}
def : CEntry<"Apple", CFoo, 10>;
def : CEntry<"Pear", CBaz, 15>;
def : CEntry<"Apple", CBar, 13>;
Here is the generated C++ code.
#ifdef GET_CTable_DECL
const CEntry *lookupCEntryByEncoding(uint16_t Encoding);
#endif
#ifdef GET_CTable_IMPL
constexpr CEntry CTable[] = {
{ "Apple", CFoo, 0xA }, // 0
{ "Apple", CBar, 0xD }, // 1
{ "Pear", CBaz, 0xF }, // 2
};
const CEntry *lookupCEntryByEncoding(uint16_t Encoding) {
struct KeyType {
uint16_t Encoding;
};
KeyType Key = { Encoding };
auto Table = makeArrayRef(CTable);
auto Idx = std::lower_bound(Table.begin(), Table.end(), Key,
[](const CEntry &LHS, const KeyType &RHS) {
if (LHS.Encoding < RHS.Encoding)
return true;
if (LHS.Encoding > RHS.Encoding)
return false;
return false;
});
if (Idx == Table.end() ||
Key.Encoding != Idx->Encoding)
return nullptr;
return &*Idx;
}
The PrimaryKeyEarlyOut field, when set to 1, modifies the lookup function so that it tests the first field of the primary key to determine whether it is within the range of the collected records’ primary keys. If not, the function returns the null pointer without performing the binary search. This is useful for tables that provide data for only some of the elements of a larger enum-based space. The first field of the primary key must be an integral type; it cannot be a string.
Adding let PrimaryKeyEarlyOut = 1 to the ATable above:
def ATable : GenericTable {
let FilterClass = "AEntry";
let Fields = ["Str", "Val1", "Val2"];
let PrimaryKey = ["Val1", "Val2"];
let PrimaryKeyName = "lookupATableByValues";
let PrimaryKeyEarlyOut = 1;
}
causes the lookup function to change as follows:
const AEntry *lookupATableByValues(uint8_t Val1, uint16_t Val2) {
if ((Val1 < 0x2) ||
(Val1 > 0x5))
return nullptr;
struct KeyType {
...
The SearchIndex class is used to define additional lookup functions for generic tables. To define an additional function, define a record whose parent class is SearchIndex and whose name is the name of the desired lookup function. This class provides three fields.
Here is an example of a secondary key added to the CTable above. The generated function looks up entries based on the Name and Kind fields.
def lookupCEntry : SearchIndex {
let Table = CTable;
let Key = ["Name", "Kind"];
}
This use of SearchIndex generates the following additional C++ code.
const CEntry *lookupCEntry(StringRef Name, unsigned Kind);
...
const CEntry *lookupCEntryByName(StringRef Name, unsigned Kind) {
struct IndexType {
const char * Name;
unsigned Kind;
unsigned _index;
};
static const struct IndexType Index[] = {
{ "APPLE", CBar, 1 },
{ "APPLE", CFoo, 0 },
{ "PEAR", CBaz, 2 },
};
struct KeyType {
std::string Name;
unsigned Kind;
};
KeyType Key = { Name.upper(), Kind };
auto Table = makeArrayRef(Index);
auto Idx = std::lower_bound(Table.begin(), Table.end(), Key,
[](const IndexType &LHS, const KeyType &RHS) {
int CmpName = StringRef(LHS.Name).compare(RHS.Name);
if (CmpName < 0) return true;
if (CmpName > 0) return false;
if ((unsigned)LHS.Kind < (unsigned)RHS.Kind)
return true;
if ((unsigned)LHS.Kind > (unsigned)RHS.Kind)
return false;
return false;
});
if (Idx == Table.end() ||
Key.Name != Idx->Name ||
Key.Kind != Idx->Kind)
return nullptr;
return &CTable[Idx->_index];
}