Skip to content

Latest commit

 

History

History
402 lines (325 loc) · 19.8 KB

FutureFeatures.md

File metadata and controls

402 lines (325 loc) · 19.8 KB
Raw

Feature to add after the MVP

These are features that make sense in the context of the high-level goals of WebAssembly but are not considered part of the Minimum Viable Product or the essential post-MVP feature set which are expected to be standardized immediately after the MVP. These will be prioritized based on developer feedback, and will be available under feature tests.

Great tooling support

This is covered in the tooling section.

Dynamic linking

This is covered in the dynamic linking section.

Finer-grained control over memory

Provide access to safe OS-provided functionality including:

  • map_file(addr, length, Blob, file-offset): semantically, this operation copies the specified range from Blob into the range [addr, addr+length) (where addr+length <= memory_size) but implementations are encouraged to mmap(addr, length, MAP_FIXED | MAP_PRIVATE, fd)
  • discard(addr, length): semantically, this operation zeroes the given range but the implementation is encouraged to drop the zeroed physical pages from the process's working set (e.g., by calling madvise(MADV_DONTNEED) on POSIX)
  • shmem_create(length): create a memory object that can be simultaneously shared between multiple linear memories
  • map_shmem(addr, length, shmem, shmem-offset): like map_file except MAP_SHARED, which isn't otherwise valid on read-only Blobs
  • mprotect(addr, length, prot-flags): change protection on the range [addr, addr+length) (where addr+length <= memory_size)
  • decommit(addr, length): equivalent to mprotect(addr, length, PROT_NONE) followed by discard(addr, length) and potentially more efficient than performing these operations in sequence.

The addr and length parameters above would be required to be multiples of page_size.

The above list of functionality mostly covers the set of functionality provided by the mmap OS primitive. One significant exception is that mmap can allocate noncontiguous virtual address ranges. See the FAQ for rationale.

More expressive control flow

Some types of control flow (especially irreducible and indirect) cannot be expressed with maximum efficiency in WebAssembly without patterned output by the relooper and jump-threading optimizations in the engine. Target uses for more expressive control flow are:

  • Language interpreters, which often use computed-goto.
  • Functional language support, where guaranteed tail call optimization is expected for correctness and performance.

Options under consideration:

  • No action, while and switch combined with jump-threading are enough.
  • Just add goto (direct and indirect).
  • Add new control-flow primitives that address common patterns.
  • Add signature-restricted Proper Tail Calls.
  • Add proper tail call, expanding upon signature-restricted proper tail calls, and making it easier to support other languages, especially functional programming languages.

GC/DOM Integration

See GC.md.

Linear memory bigger than 4 GiB

The WebAssembly MVP will support the wasm32 mode of WebAssembly, with linear memory sizes up to 4 GiB using 32-bit linear memory indices. To support larger sizes, the wasm64 mode of WebAssembly will be added in the future, supporting much greater linear memory sizes using 64-bit linear memory indices. wasm32 and wasm64 are both just modes of WebAssembly, to be selected by a flag in a module header, and don't imply any semantics differences outside of how linear memory is handled. Platforms will also have APIs for querying which of wasm32 and wasm64 are supported.

Of course, the ability to actually allocate this much memory will always be subject to dynamic resource availability.

It is likely that wasm64 will initially support only 64-bit linear memory indices, and wasm32 will leave 64-bit linear memory indices unsupported, so that implementations don't have to support multiple index sizes in the same instance. However, operators with 32-bit indices and operations with 64-bit indices will be given separate names to leave open the possibility of supporting both in the same instance in the future.

Source maps integration

  • Add a new source maps module section type.
  • Either embed the source maps directly or just a URL from which source maps can be downloaded.
  • Text source maps become intractably large for even moderate-sized compiled codes, so probably need to define new binary format for source maps.
  • Gestate ideas and start discussions at the Source Map RFC repository

Coroutines

Coroutines will eventually be part of C++ and is already popular in other programming languages that WebAssembly will support.

Signature-restricted Proper Tail Calls

See the asm.js RFC for a full description of signature-restricted Proper Tail Calls (PTC).

Useful properties of signature-restricted PTCs:

  • In most cases, can be compiled to a single jump.
  • Can express indirect goto via function-pointer calls.
  • Can be used as a compile target for languages with unrestricted PTCs; the code generator can use a stack in the linear memory to effectively implement a custom call ABI on top of signature-restricted PTCs.
  • An engine that wishes to perform aggressive optimization can fuse a graph of PTCs into a single function.
  • To reduce compile time, a code generator can use PTCs to break up ultra-large functions into smaller functions at low overhead using PTCs.
  • A compiler can exert some amount of control over register allocation via the ordering of arguments in the PTC signature.

General-purpose Proper Tail Calls

General-purpose Proper Tail Calls would have no signature restrictions, and therefore be more broadly usable than Signature-restricted Proper Tail Calls, though there would be some different performance characataristics.

Asynchronous Signals

TODO

"Long SIMD"

The initial SIMD API will be a "short SIMD" API, centered around fixed-width 128-bit types and explicit SIMD operations. This is quite portable and useful, but it won't be able to deliver the full performance capabilities of some of today's popular hardware. There is a proposal in the SIMD.js repository for a "long SIMD" model which generalizes to wider hardware vector lengths, making more natural use of advanced features like vector lane predication, gather/scatter, and so on. Interesting questions to ask of such an model will include:

  • How will this model map onto popular modern SIMD hardware architectures?
  • What is this model's relationship to other hardware parallelism features, such as GPUs and threads with shared memory?
  • How will this model be used from higher-level programming languages? For example, the C++ committee is considering a wide variety of possible approaches; which of them might be supported by the model?
  • What is the relationship to the "short SIMD" API? "None" may be an acceptable answer, but it's something to think about.
  • What nondeterminism does this model introduce into the overall platform?
  • What happens when code uses long SIMD on a hardware platform which doesn't support it? Reasonable options may include emulating it without the benefit of hardware acceleration, or indicating a lack of support through feature tests.

Platform-independent Just-in-Time (JIT) compilation

WebAssembly is a new virtual ISA, and as such applications won't be able to simply reuse their existing JIT-compiler backends. Applications will instead have to interface with WebAssembly's instructions as if they were a new ISA.

Applications expect a wide variety of JIT-compilation capabilities. WebAssembly should support:

  • Producing a dynamic library and loading it into the current WebAssembly module.
  • Define lighter-weight mechanisms, such as the ability to add a function to an existing module.
  • Support explicitly patchable constructs within functions to allow for very fine-grained JIT-compilation. This includes:
    • Code patching for polymorphic inline caching;
    • Call patching to chain JIT-compiled functions together;
    • Temporary halt-insertion within functions, to trap if a function start executing while a JIT-compiler's runtime is performing operations dangerous to that function.
  • Provide JITs access to profile feedback for their JIT-compiled code.
  • Code unloading capabilities, especially in the context of code garbage collection and defragmentation.

WebAssembly's JIT interface would likely be fairly low-level. However, there are use cases for higher-level functionality and optimization too. One avenue for addressing these use cases is a JIT and Optimization library.

Multiprocess support

  • vfork.
  • Inter-process communication.
  • Inter-process mmap.

Trapping or non-trapping strategies.

Presently, when an instruction traps, the program is immediately terminated. This suits C/C++ code, where trapping conditions indicate Undefined Behavior at the source level, and it's also nice for handwritten code, where trapping conditions typically indicate an instruction being asked to perform outside its supported range. However, the current facilities do not cover some interesting use cases:

  • Not all likely-bug conditions are covered. For example, it would be very nice to have a signed-integer add which traps on overflow. Such a construct would add too much overhead on today's popular hardware architectures to be used in general, however it may still be useful in some contexts.
  • Some higher-level languages define their own semantics for conditions like division by zero and so on. It's possible for compilers to add explicit checks and handle such cases manually, though more direct support from the platform could have advantages:
    • Non-trapping versions of some opcodes, such as an integer division instruction that returns zero instead of trapping on division by zero, could potentially run faster on some platforms.
    • The ability to recover gracefully from traps in some way could make many things possible. Possibly this could involve throwing or possibly by resuming execution at the trapping instruction with the execution state altered, if there can be a reasonable way to specify how that should work.

Additional integer operations

  • The following operations can be built from other operators already present, however in doing so they read at least one non-constant input multiple times, breaking single-use expression tree formation.

    • i32.rotr: sign-agnostic bitwise rotate right
    • i32.rotl: sign-agnostic bitwise rotate left
    • i32.min_s: signed minimum
    • i32.max_s: signed maximum
    • i32.min_u: unsigned minimum
    • i32.max_u: unsigned maximum
    • i32.sext: sign-agnostic sext(x, y) is shr_s(shl(x,y),y)
    • i32.abs_s: signed absolute value (traps on INT32_MIN)
    • i32.bswap: sign-agnostic reverse bytes (endian conversion)
    • i32.bswap16: sign-agnostic, bswap16(x) is ((x>>8)&255)|((x&255)<<8)

  • The following operations are just potentially interesting.

    • i32.clrs: sign-agnostic count leading redundant sign bits (defined for all values, including 0)
    • i32.floor_div_s: signed division (result is floored)

  • The following 64-bit-only operations are potentially interesting as well.

Additional floating point operations

  • f32.minnum: minimum; if exactly one operand is NaN, returns the other operand
  • f32.maxnum: maximum; if exactly one operand is NaN, returns the other operand
  • f32.fma: fused multiply-add (results always conforming to IEEE 754-2008)
  • f64.minnum: minimum; if exactly one operand is NaN, returns the other operand
  • f64.maxnum: maximum; if exactly one operand is NaN, returns the other operand
  • f64.fma: fused multiply-add (results always conforming to IEEE 754-2008)

minnum and maxnum operations would treat -0.0 as being effectively less than 0.0.

Note that some operations, like fma, may not be available or may not perform well on all platforms. These should be guarded by feature tests so that if available, they behave consistently.

Floating point approximation operations

  • f32.reciprocal_approximation: reciprocal approximation
  • f64.reciprocal_approximation: reciprocal approximation
  • f32.reciprocal_sqrt_approximation: reciprocal sqrt approximation
  • f64.reciprocal_sqrt_approximation: reciprocal sqrt approximation

These operations would not required to be fully precise, but the specifics would need clarification.

16-bit and 128-bit floating point support

For 16-bit floating point support, it may make sense to split the feature into two parts: support for just converting between 16-bit and 32-bit or 64-bit formats possibly folded into load and store operations, and full support for actual 16-bit arithmetic.

128-bit is an interesting question because hardware support for it is very rare, so it's usually going to be implemented with software emulation anyway, so there's nothing preventing WebAssembly applications from linking to an appropriate emulation library and getting similarly performant results. Emulation libraries would have more flexibility to offer approximation techniques such as double-double arithmetic. If we standardize 128-bit floating point in WebAssembly, it will probably be standard IEEE 754-2008 quadruple precision.

Full IEEE 754-2008 conformance

WebAssembly floating point conforms IEEE 754-2008 in most respects, but there are a few areas that are not yet covered.

IEEE 754-2008 NaN bit pattern propagation is presently permitted but not required. It would be possible for WebAssembly to require it in the future.

To support exceptions and alternate rounding modes, one option is to define an alternate form for each of add, sub, mul, div, sqrt, and fma. These alternate forms would have extra operands for rounding mode, masked traps, and old flags, and an extra result for a new flags value. These operations would be fairly verbose, but it's expected that their use cases will specialized. This approach has the advantage of exposing no global (even if only per-thread) control and status registers to applications, and to avoid giving the common operations the possibility of having side effects.

Debugging techniques are also important, but they don't necessarily need to be in the spec itself. Implementations are welcome (and encouraged) to support non-standard execution modes, enabled only from developer tools, such as modes with alternate rounding, or evaluation of floating point expressions at greater precision, to support [techniques for detecting numerical instability] (https://www.cs.berkeley.edu/~wkahan/Mindless.pdf).

To help developers find the sources of floating point exceptions, implementations may wish to provide a mode where NaN values are produced with payloads containing identifiers helping programmers locate where the NaNs first appeared. Another option would be to offer another non-standard execution mode, enabled only from developer tools, that would enable traps on selected floating point exceptions, however care should be taken, since not all floating point exceptions indicate bugs.

Integer Overflow Detection

There are two different use cases here, one where the application wishes to handle overflow locally, and one where it doesn't.

When the application is prepared to handle overflow locally, it would be useful to have arithmetic operations which can indicate when overflow occurred. An example of this is the checked arithmetic builtins available in compilers such as clang and GCC. If WebAssembly is made to support nodes with multiple return values, that could be used instead of passing a pointer.

There are also several use cases where an application does not wish to handle overflow locally. One family of examples includes implementing optimized bignum arithmetic, or optimizing JS Numbers to use int32 operations. Another family includes compiling code that doesn't expect overflow to occur, but which wishes to have overflow detected and reported if it does happen. These use cases would ideally like to have overflow trap, and to allow them to handle trap specially. Following the rule that explicitly signed and unsigned operations trap whenever the result value can not be represented in the result type, it would be possible to add explicitly signed and unsigned versions of integer add, sub, and mul, which would trap on overflow. The main reason we haven't added these already is that they're not efficient for general-purpose use on several of today's popular hardware architectures.

Better feature testing support

The MVP will provide a basic feature detection query which can be used to allow an application to conditionally use features added after the MVP. However, the usual feature detection pattern in JS:

if (foo)
    foo();
else
    alternativeToFoo();

won't work in WebAssembly since, when foo isn't supported, the use of foo will fail at decode/validation time.

In the MVP, applications wanting to conditionally use a new feature can employ a few brute-force strategies:

  • compile several different versions of a module, each assuming different feature support and use feature testing to decide which one to load;
  • during layer 1 decoding, which happens in user code anyway, use feature detection to translate unsupported opcodes into something that validates (a call to abort() or a polyfill).

However, with SIMD and other proposed extensions, it would be good to have a strategy that didn't require these brute force techniques (which have developer and load-time cost). The basic challenge is to allow a WebAssembly decoder to decode "through" an AST node that it knows nothing about. There are a number of ways to achieve this and more concrete experience with the realities of polyfilling is necessary to suggest the right design.

Mutable global variables

In the MVP, there are no global variables; C/C++ global variables are stored in linear memory and thus accessed through normal linear memory operations. Dynamic linking will add some form of immutable global variable analogous to "symbols" in native binaries. In some cases, though, it may be useful to have a fully mutable global variable which lives outside linear memory. This would allow more aggressive compiler optimizations (due to better alias information). If globals are additionally allowed array types, significant portions of memory could be moved out of linear memory which could reduce fragmentation issues. Languages like Fortran which limit aliasing would be one use case. C/C++ compilers could also determine that some global variables never have their address taken.