Release 1.0 (Draft, last updated Feb 15, 2018)
WebAssembly (abbreviated Wasm ) is a safe, portable, low-level code format designed for efficient execution and compact representation. Its main goal is to enable high performance applications on the Web, but it does not make any Web-specific assumptions or provide Web-specific features, so it can be employed in other environments as well.
WebAssembly is an open standard developed by a W3C Community Group that includes representatives of all major browser vendors.
This document describes version 1.0 of the core WebAssembly standard. It is intended that it will be superseded by new incremental releases with additional features in the future.
The design goals of WebAssembly are the following:
Fast, safe, and portable semantics:
- Fast: executes with near native code performance, taking advantage of capabilities common to all contemporary hardware.
- Safe: code is validated and executes in a memory-safe , sandboxed environment preventing data corruption or security breaches.
- Well-defined: fully and precisely defines valid programs and their behavior in a way that is easy to reason about informally and formally.
- Hardware-independent: can be compiled on all modern architectures, desktop or mobile devices and embedded systems alike.
- Language-independent: does not privilege any particular language, programming model, or object model.
- Platform-independent: can be embedded in browsers, run as a stand-alone VM, or integrated in other environments.
- Open: programs can interoperate with their environment in a simple and universal manner.
Efficient and portable representation:
- Compact: has a binary format that is fast to transmit by being smaller than typical text or native code formats.
- Modular: programs can be split up in smaller parts that can be transmitted, cached, and consumed separately.
- Efficient: can be decoded, validated, and compiled in a fast single pass, equally with either just-in-time (JIT) or ahead-of-time (AOT) compilation.
- Streamable: allows decoding, validation, and compilation to begin as soon as possible, before all data has been seen.
- Parallelizable: allows decoding, validation, and compilation to be split into many independent parallel tasks.
- Portable: makes no architectural assumptions that are not broadly supported across modern hardware.
WebAssembly code is also intended to be easy to inspect and debug, especially in environments like web browsers, but such features are beyond the scope of this specification.
|||A contraction of “WebAssembly”, not an acronym, hence not using all-caps.|
|||No program can break WebAssembly’s memory model. Of course, it cannot guarantee that an unsafe language compiling to WebAssembly does not corrupt its own memory layout, e.g. inside WebAssembly’s linear memory.|
At its core, WebAssembly is a virtual instruction set architecture (virtual ISA). As such, it has many use cases and can be embedded in many different environments. To encompass their variety and enable maximum reuse, the WebAssembly specification is split and layered into several documents.
This document is concerned with the core ISA layer of WebAssembly. It defines the instruction set, binary encoding, validation, and execution semantics, as well as a textual representation. It does not, however, define how WebAssembly programs can interact with a specific environment they execute in, nor how they are invoked from such an environment.
Instead, this specification is complemented by additional documents defining interfaces to specific embedding environments such as the Web. These will each define a WebAssembly application programming interface (API) suitable for a given environment.
WebAssembly depends on two existing standards:
- IEEE 754-2008, for the representation of floating-point data and the semantics of respective numeric operations.
- Unicode, for the representation of import/export names and the text format.
However, to make this specification self-contained, relevant aspects of the aforementioned standards are defined and formalized as part of this specification, such as the binary representation and rounding of floating-point values, and the value range and UTF-8 encoding of Unicode characters.
The aforementioned standards are the authorative source of all respective definitions. Formalizations given in this specification are intended to match these definitions. Any discrepancy in the syntax or semantics described is to be considered an error.
WebAssembly encodes a low-level, assembly-like programming language. This language is structured around the following concepts.
- WebAssembly provides only four basic value types. These are integers and IEEE 754-2008 numbers, each in 32 and 64 bit width. 32 bit integers also serve as Booleans and as memory addresses. The usual operations on these types are available, including the full matrix of conversions between them. There is no distinction between signed and unsigned integer types. Instead, integers are interpreted by respective operations as either unsigned or signed in two’s complement representation.
- The computational model of WebAssembly is based on a stack machine. Code consists of sequences of instructions that are executed in order. Instructions manipulate values on an implicit operand stack  and fall into two main categories. Simple instructions perform basic operations on data. They pop arguments from the operand stack and push results back to it. Control instructions alter control flow. Control flow is structured, meaning it is expressed with well-nested constructs such as blocks, loops, and conditionals. Branches can only target such constructs.
- Under some conditions, certain instructions may produce a trap, which immediately aborts execution. Traps cannot be handled by WebAssembly code, but are reported to the outside environment, where they typically can be caught.
- Code is organized into separate functions. Each function takes a sequence of values as parameters and returns a sequence of values as results.  Functions can call each other, including recursively, resulting in an implicit call stack that cannot be accessed directly. Functions may also declare mutable local variables that are usable as virtual registers.
- A table is an array of opaque values of a particular element type. It allows programs to select such values indirectly through a dynamic index operand. Currently, the only available element type is an untyped function reference. Thereby, a program can call functions indirectly through a dynamic index into a table. For example, this allows emulating function pointers by way of table indices.
- Linear Memory
- A linear memory is a contiguous, mutable array of raw bytes. Such a memory is created with an initial size but can be grown dynamically. A program can load and store values from/to a linear memory at any byte address (including unaligned). Integer loads and stores can specify a storage size which is smaller than the size of the respective value type. A trap occurs if an access is not within the bounds of the current memory size.
- A WebAssembly binary takes the form of a module that contains definitions for functions, tables, and linear memories, as well as mutable or immutable global variables. Definitions can also be imported, specifying a module/name pair and a suitable type. Each definition can optionally be exported under one or more names. In addition to definitions, modules can define initialization data for their memories or tables that takes the form of segments copied to given offsets. They can also define a start function that is automatically executed.
- A WebAssembly implementation will typically be embedded into a host environment. This environment defines how loading of modules is initiated, how imports are provided (including host-side definitions), and how exports can be accessed. However, the details of any particular embedding are beyond the scope of this specification, and will instead be provided by complementary, environment-specific API definitions.
|||In practice, implementations need not maintain an actual operand stack. Instead, the stack can be viewed as a set of anonymous registers that are implicitly referenced by instructions. The type system ensures that the stack height, and thus any referenced register, is always known statically.|
|||In the current version of WebAssembly, there may be at most one result value.|
Conceptually, the semantics of WebAssembly is divided into three phases. For each part of the language, the specification specifies each of them.
- WebAssembly modules are distributed in a binary format. Decoding processes that format and converts it into an internal representation of a module. In this specification, this representation is modelled by abstract syntax, but a real implementation could compile directly to machine code instead.
- A decoded module has to be valid. Validation checks a number of well-formedness conditions to guarantee that the module is meaningful and safe. In particular, it performs type checking of functions and the instruction sequences in their bodies, ensuring for example that the operand stack is used consistently.
Finally, a valid module can be executed. Execution can be further divided into two phases:
Instantiation. A module instance is the dynamic representation of a module, complete with its own state and execution stack. Instantiation executes the module body itself, given definitions for all its imports. It initializes globals, memories and tables and invokes the module’s start function if defined. It returns the instances of the module’s exports.
Invocation. Once instantiated, further WebAssembly computations can be initiated by invoking an exported function on a module instance. Given the required arguments, that executes the respective function and returns its results.
Instantiation and invocation are operations within the embedding environment.
WebAssembly is a programming language that has multiple concrete representations (its binary format and the text format). Both map to a common structure. For conciseness, this structure is described in the form of an abstract syntax. All parts of this specification are defined in terms of this abstract syntax.
The following conventions are adopted in defining grammar rules for abstract syntax.
- Terminal symbols (atoms) are written in sans-serif font: .
- Nonterminal symbols are written in italic font: .
- is a sequence of iterations of .
- is a possibly empty sequence of iterations of . (This is a shorthand for used where is not relevant.)
- is a non-empty sequence of iterations of . (This is a shorthand for where .)
- is an optional occurrence of . (This is a shorthand for where .)
- Productions are written .
- Some productions are augmented with side conditions in parentheses, “ ”, that provide a shorthand for a combinatorial expansion of the production into many separate cases.
When dealing with syntactic constructs the following notation is also used:
- denotes the empty sequence.
- denotes the length of a sequence .
- denotes the -th element of a sequence , starting from .
- denotes the sub-sequence of a sequence .
- denotes the same sequence as , except that the -the element is replaced with .
- denotes the same sequence as , except that the sub-sequence is replaced with .
- denotes the flat sequence formed by concatenating all sequences in .
Moreover, the following conventions are employed:
- The notation , where is a non-terminal symbol, is treated as a meta variable ranging over respective sequences of (similarly for , , ).
- When given a sequence , then the occurrences of in a sequence written are assumed to be in point-wise correspondence with (similarly for , , ). This implicitly expresses a form of mapping syntactic constructions over a sequence.
Productions of the following form are interpreted as records that map a fixed set of fields to “values” , respectively:
The following notation is adopted for manipulating such records:
denotes the contents of the component of .
denotes the same record as , except that the contents of the component is replaced with .
denotes the composition of two records with the same fields of sequences by appending each sequence point-wise:
denotes the composition of a sequence of records, respectively; if the sequence is empty, then all fields of the resulting record are empty.
The update notation for sequences and records generalizes recursively to nested components accessed by “paths” :
- is short for .
- is short for .
where is shortened to .
WebAssembly programs operate on primitive numeric values. Moreover, in the definition of programs, immutable sequences of values occur to represent more complex data, such as text strings or other vectors.
The simplest form of value are raw uninterpreted bytes. In the abstract syntax they are represented as hexadecimal literals.
Different classes of integers with different value ranges are distinguished by their bit width and by whether they are unsigned or signed.
The latter class defines uninterpreted integers, whose signedness interpretation can vary depending on context. In the abstract syntax, they are represented as unsigned values. However, some operations convert them to signed based on a two’s complement interpretation.
The main integer types occurring in this specification are , , , , , , , . However, other sizes occur as auxiliary constructions, e.g., in the definition of floating-point numbers.
Floating-point data represents 32 or 64 bit values that correspond to the respective binary formats of the IEEE 754-2008 standard (Section 3.3).
Every value has a sign and a magnitude. Magnitudes can either be expressed as normal numbers of the form , where is the exponent and is the significand whose most signifcant bit is , or as a subnormal number where the exponent is fixed to the smallest possible value and is ; among the subnormals are positive and negative zero values. Since the significands are binary values, normals are represented in the form , where is the bit width of ; similarly for subnormals.
Possible magnitudes also include the special values (infinity) and (NaN, not a number). NaN values have a payload that describes the mantissa bits in the underlying binary representation. No distinction is made between signalling and quiet NaNs.
where and with
A canonical NaN is a floating-point value where is a payload whose most significant bit is while all others are :
An arithmetic NaN is a floating-point value with , such that the most significant bit is while all others are arbitrary.
In the abstract syntax, subnormals are distinguished by the leading 0 of the significand. The exponent of subnormals has the same value as the smallest possible exponent of a normal number. Only in the binary representation the exponent of a subnormal is encoded differently than the exponent of any normal number.
Value types classify the individual values that WebAssembly code can compute with and the values that a variable accepts.
The types and classify 32 and 64 bit integers, respectively. Integers are not inherently signed or unsigned, their interpretation is determined by individual operations.
The types and classify 32 and 64 bit floating-point data, respectively. They correspond to the respective binary floating-point representations, also known as single and double precision, as defined by the IEEE 754-2008 standard (Section 3.3).
In the current version of WebAssembly, at most one value is allowed as a result. However, this may be generalized to sequences of values in future versions.
Function types classify the signature of functions, mapping a vector of parameters to a vector of results.
In the current version of WebAssembly, the length of the result type vector of a valid function type may be at most . This restriction may be removed in future versions.
If no maximum is given, the respective storage can grow to any size.
Memory types classify linear memories and their size range.
The limits constrain the minimum and optionally the maximum size of a memory. The limits are given in units of page size.
Table types classify tables over elements of element types within a size range.
Like memories, tables are constrained by limits for their minimum and optionally maximum size. The limits are given in numbers of entries.
The element type is the infinite union of all function types. A table of that type thus contains references to functions of heterogeneous type.
In future versions of WebAssembly, additional element types may be introduced.
Global types classify global variables, which hold a value and can either be mutable or immutable.
WebAssembly code consists of sequences of instructions. Its computational model is based on a stack machine in that instructions manipulate values on an implicit operand stack, consuming (popping) argument values and producing (pushing) result values.
In the current version of WebAssembly, at most one result value can be pushed by a single instruction. This restriction may be lifted in future versions.
In addition to dynamic operands from the stack, some instructions also have static immediate arguments, typically indices or type annotations, which are part of the instruction itself.
Some instructions are structured in that they bracket nested sequences of instructions.
The following sections group instructions into a number of different categories.
Numeric instructions are divided by value type. For each type, several subcategories can be distinguished:
- Constants: return a static constant.
- Unary Operators: consume one operand and produce one result of the respective type.
- Binary Operators: consume two operands and produce one result of the respective type.
- Tests: consume one operand of the respective type and produce a Boolean integer result.
- Comparisons: consume two operands of the respective type and produce a Boolean integer result.
- Conversions: consume a value of one type and produce a result of another (the source type of the conversion is the one after the “ ”).
Some integer instructions come in two flavours, where a signedness annotation distinguishes whether the operands are to be interpreted as unsigned or signed integers. For the other integer instructions, the use of two’s complement for the signed interpretation means that they behave the same regardless of signedness.
Instructions in this group can operate on operands of any value type.
The operator simply throws away a single operand.
The operator selects one of its first two operands based on whether its third operand is zero or not.
These instructions get or set the values of variables, respectively. The instruction is like but also returns its argument.
Instructions in this group are concerned with linear memory.
Memory is accessed with and instructions for the different value types. They all take a memory immediate that contains an address offset and an alignment hint (in base 2 logarithmic representation). Integer loads and stores can optionally specify a storage size that is smaller than the bit width of the respective value type. In the case of loads, a sign extension mode is then required to select appropriate behavior.
The static address offset is added to the dynamic address operand, yielding a 33 bit effective address that is the zero-based index at which the memory is accessed. All values are read and written in little endian byte order. A trap results if any of the accessed memory bytes lies outside the address range implied by the memory’s current size.
Future version of WebAssembly might provide memory instructions with 64 bit address ranges.
The instruction returns the current size of a memory. The instruction grows memory by a given delta and returns the previous size, or if enough memory cannot be allocated. Both instructions operate in units of page size.
Instructions in this group affect the flow of control.
The instruction does nothing.
The instruction causes an unconditional trap.
The , and instructions are structured instructions. They bracket nested sequences of instructions, called blocks, terminated with, or separated by, or pseudo-instructions. As the grammar prescribes, they must be well-nested. A structured instruction can produce a value as described by the annotated result type.
Each structured control instruction introduces an implicit label. Labels are targets for branch instructions that reference them with label indices. Unlike with other index spaces, indexing of labels is relative by nesting depth, that is, label refers to the innermost structured control instruction enclosing the referring branch instruction, while increasing indices refer to those farther out. Consequently, labels can only be referenced from within the associated structured control instruction. This also implies that branches can only be directed outwards, “breaking” from the block of the control construct they target. The exact effect depends on that control construct. In case of or it is a forward jump, resuming execution after the matching . In case of it is a backward jump to the beginning of the loop.
This enforces structured control flow. Intuitively, a branch targeting a or behaves like a statement, while a branch targeting a behaves like a statement.
Branch instructions come in several flavors: performs an unconditional branch, performs a conditional branch, and performs an indirect branch through an operand indexing into the label vector that is an immediate to the instruction, or to a default target if the operand is out of bounds. The instruction is a shortcut for an unconditional branch to the outermost block, which implicitly is the body of the current function. Taking a branch unwinds the operand stack up to the height where the targeted structured control instruction was entered. However, forward branches that target a control instruction with a non-empty result type consume matching operands first and push them back on the operand stack after unwinding, as a result for the terminated structured instruction.
The instruction invokes another function, consuming the necessary arguments from the stack and returning the result values of the call. The instruction calls a function indirectly through an operand indexing into a table. Since tables may contain function elements of heterogeneous type , the callee is dynamically checked against the function type indexed by the instruction’s immediate, and the call aborted with a trap if it does not match.
In some places, validation restricts expressions to be constant, which limits the set of allowable instructions.
WebAssembly programs are organized into modules, which are the unit of deployment, loading, and compilation. A module collects definitions for types, functions, tables, memories, and globals. In addition, it can declare imports and exports and provide initialization logic in the form of data and element segments or a start function.
Each of the vectors – and thus the entire module – may be empty.
Definitions are referenced with zero-based indices. Each class of definition has its own index space, as distinguished by the following classes.
The index space for functions, tables, memories and globals includes respective imports declared in the same module. The indices of these imports precede the indices of other definitions in the same index space.
Label indices reference structured control instructions inside an instruction sequence.
The component of a module defines a vector of function types.
All function types used in a module must be defined in this component. They are referenced by type indices.
Future versions of WebAssembly may add additional forms of type definitions.
The component of a module defines a vector of functions with the following structure:
The declare a vector of mutable local variables and their types. These variables are referenced through local indices in the function’s body. The index of the first local is the smallest index not referencing a parameter.
The component of a module defines a vector of tables described by their table type:
A table is a vector of opaque values of a particular table element type. The size in the limits of the table type specifies the initial size of that table, while its , if present, restricts the size to which it can grow later.
Tables can be initialized through element segments.
In the current version of WebAssembly, at most one table may be defined or imported in a single module, and all constructs implicitly reference this table . This restriction may be lifted in future versions.
The component of a module defines a vector of linear memories (or memories for short) as described by their memory type:
A memory is a vector of raw uninterpreted bytes. The size in the limits of the memory type specifies the initial size of that memory, while its , if present, restricts the size to which it can grow later. Both are in units of page size.
Memories can be initialized through data segments.
In the current version of WebAssembly, at most one memory may be defined or imported in a single module, and all constructs implicitly reference this memory . This restriction may be lifted in future versions.
The component of a module defines a vector of global variables (or globals for short):
Each global stores a single value of the given global type. Its also specifies whether a global is immutable or mutable. Moreover, each global is initialized with an value given by a constant initializer expression.
The initial contents of a table is uninitialized. The component of a module defines a vector of element segments that initialize a subrange of a table at a given offset from a static vector of elements.
In the current version of WebAssembly, at most one table is allowed in a module. Consequently, the only valid is .
In the current version of WebAssembly, at most one memory is allowed in a module. Consequently, the only valid is .
The component of a module defines a set of exports that become accessible to the host environment once the module has been instantiated.
In the current version of WebAssembly, only immutable globals may be exported.
The component of a module defines a set of imports that are required for instantiation.
Each import is identified by a two-level name space, consisting of a name and a unique for an entity within that module. Importable definitions are functions, tables, memories, and globals. Each import is specified by a descriptor with a respective type that a definition provided during instantiation is required to match.
Every import defines an index in the respective index space. In each index space, the indices of imports go before the first index of any definition contained in the module itself.
In the current version of WebAssembly, only immutable globals may be imported.
Validation checks that a WebAssembly module is well-formed. Only valid modules can be instantiated.
Validity is defined by a type system over the abstract syntax of a module and its contents. For each piece of abstract syntax, there is a typing rule that specifies the constraints that apply to it. All rules are given in two equivalent forms:
- In prose, describing the meaning in intuitive form.
- In formal notation, describing the rule in mathematical form. 
The prose and formal rules are equivalent, so that understanding of the formal notation is not required to read this specification. The formalism offers a more concise description in notation that is used widely in programming languages semantics and is readily amenable to mathematical proof.
In both cases, the rules are formulated in a declarative manner. That is, they only formulate the constraints, they do not define an algorithm. The skeleton of a sound and complete algorithm for type-checking instruction sequences according to this specification is provided in the appendix.
Validity of an individual definition is specified relative to a context, which collects relevant information about the surrounding module and the definitions in scope:
- Types: the list of types defined in the current module.
- Functions: the list of functions declared in the current module, represented by their function type.
- Tables: the list of tables declared in the current module, represented by their table type.
- Memories: the list of memories declared in the current module, represented by their memory type.
- Globals: the list of globals declared in the current module, represented by their global type.
- Locals: the list of locals declared in the current function (including parameters), represented by their value type.
- Labels: the stack of labels accessible from the current position, represented by their result type.
- Return: the return type of the current function, represented as a result type.
In other words, a context contains a sequence of suitable types for each index space, describing each defined entry in that space. Locals, labels and return type are only used for validating instructions in function bodies, and are left empty elsewhere. The label stack is the only part of the context that changes as validation of an instruction sequence proceeds.
It is convenient to define contexts as records with abstract syntax:
The fields of a context are not defined as vectors, since their lengths are not bounded by the maximum vector size.
In addition to field access the following notation is adopted for manipulating contexts:
- When spelling out a context, empty fields are omitted.
- denotes the same context as but with the elements prepended to its component sequence.
This notation is defined to prepend not append. It is only used in situations where the original is either empty or is . In the latter case adding to the front is desired because the label index space is indexed relatively, that is, in reverse order of addition.
Validation is specified by stylised rules for each relevant part of the abstract syntax. The rules not only state constraints defining when a phrase is valid, they also classify it with a type. The following conventions are adopted in stating these rules.
A phrase is said to be “valid with type ” if and only if all constraints expressed by the respective rules are met. The form of depends on what is.
The rules implicitly assume a given context .
In some places, this context is locally extended to a context with additional entries. The formulation “Under context , … statement …” is adopted to express that the following statement must apply under the assumptions embodied in the extended context.
This section gives a brief explanation of the notation for specifying typing rules formally. For the interested reader, a more thorough introduction can be found in respective text books. 
The proposition that a phrase has a respective type is written . In general, however, typing is dependent on a context . To express this explicitly, the complete form is a judgement , which says that holds under the assumptions encoded in .
The formal typing rules use a standard approach for specifying type systems, rendering them into deduction rules. Every rule has the following general form:
Such a rule is read as a big implication: if all premises hold, then the conclusion holds. Some rules have no premises; they are axioms whose conclusion holds unconditionally. The conclusion always is a judgment , and there is one respective rule for each relevant construct of the abstract syntax.
For example, the typing rule for the instruction can be given as an axiom:
The instruction is always valid with type ] (saying that it consumes two values and produces one), independent of any side conditions.
An instruction like can be typed as follows:
Here, the premise enforces that the immediate local index exists in the context. The instruction produces a value of its respective type (and does not consume any values). If does not exist then the premise does not hold, and the instruction is ill-typed.
Finally, a structured instruction requires a recursive rule, where the premise is itself a typing judgement:
A instruction is only valid when the instruction sequence in its body is. Moreover, the result type must match the block’s annotation . If so, then the instruction has the same type as the body. Inside the body an additional label of the same type is available, which is expressed by extending the context with the additional label information for the premise.
|||The semantics is derived from the following article: Andreas Haas, Andreas Rossberg, Derek Schuff, Ben Titzer, Dan Gohman, Luke Wagner, Alon Zakai, JF Bastien, Michael Holman. Bringing the Web up to Speed with WebAssembly. Proceedings of the 38th ACM SIGPLAN Conference on Programming Language Design and Implementation (PLDI 2017). ACM 2017.|
|||For example: Benjamin Pierce. Types and Programming Languages. The MIT Press 2002|
Limits must have meaningful bounds.
Function types may not specify more than one result.
- The limits must be valid.
- Then the table type is valid.
- The limits must be valid.
- Then the memory type is valid.
Instructions are classified by function types that describe how they manipulate the operand stack. The types describe the required input stack with argument values of types that an instruction pops off and the provided output stack with result values of types that it pushes back.
For example, the instruction has type , consuming two values and producing one.
Typing extends to instruction sequences . Such a sequence has a function types if the accumulative effect of executing the instructions is consuming values of types off the operand stack and pushing new values of types .
For some instructions, the typing rules do not fully constrain the type, and therefore allow for multiple types. Such instructions are called polymorphic. Two degrees of polymorphism can be distinguished:
- value-polymorphic: the value type of one or several individual operands is unconstrained. That is the case for all parametric instructions like and .
- stack-polymorphic: the entire (or most of the) function type of the instruction is unconstrained. That is the case for all control instructions that perform an unconditional control transfer, such as , , , and .
In both cases, the unconstrained types or type sequences can be chosen arbitrarily, as long as they meet the constraints imposed for the surrounding parts of the program.
For example, the instruction is valid with type , for any possible value type . Consequently, both instruction sequences
are valid, with in the typing of being instantiated to or , respectively.
The instruction is valid with type