From ccbac15a287fa2af4a3db882d950db442011a7b1 Mon Sep 17 00:00:00 2001 From: Dan Gohman Date: Wed, 26 Feb 2020 18:30:43 -0800 Subject: Define non-trapping float-to-int conversions. (#1089) * Define non-trapping float-to-int conversions. This also introduces the concept of prefix bytes, and defines the "numeric" prefix byte, used for encodin the new conversion instructions. * Add feature markers. * Add the feature marker in more places. * Rename "numeric" to "misc". See https://github.com/WebAssembly/nontrapping-float-to-int-conversions/issues/5. * Rename opcodes. See https://github.com/WebAssembly/spec/issues/884#issuecomment-426433329. --- BinaryEncoding.md | 28 ++++++++++++++++++++++------ Semantics.md | 23 ++++++++++++++++++++++- 2 files changed, 44 insertions(+), 7 deletions(-) diff --git a/BinaryEncoding.md b/BinaryEncoding.md index f0ec70f..6edacc8 100644 --- a/BinaryEncoding.md +++ b/BinaryEncoding.md @@ -57,12 +57,20 @@ Note: Currently, the only sizes used are `varint7`, `varint32` and `varint64`. ## Instruction Opcodes -In the MVP, the opcodes of [instructions](Semantics.md) are all encoded in a -single byte since there are fewer than 256 opcodes. Future features like -[SIMD][future simd] and [atomics][future threads] -will bring the total count above 256 and so an extension scheme will be -necessary, designating one or more single-byte values as prefixes for multi-byte -opcodes. +The opcodes for many common instructions are encoded in a single byte. + +The opcodes for some families of instructions are encoded as a *prefix byte* +followed by an LEB128 value. Prefix byte values are allocated starting at `0xfe` +and counting downwards. + +### Prefix Bytes + +| Opcode | Name | Description | +| ------ | -------- | ----------- | +| `0xff` | reserved | Reserved for unknown future language evolution :milky_way: | +| `0xfe` | threads | Expected to be used for [atomics :unicorn:][future threads] | +| `0xfd` | simd | Expected to be used for [SIMD :unicorn:][future simd] | +| `0xfc` | misc | Miscellaneous operations :bowling: | ## Language Types @@ -802,6 +810,14 @@ for [future :unicorn:][future multiple tables] use and must be 0 in the MVP. | `f64.convert_s/i64` | `0xb9` | | | | `f64.convert_u/i64` | `0xba` | | | | `f64.promote/f32` | `0xbb` | | | +| `i32.trunc_sat_f32_s` | `0xfc` `0x00` | | :bowling: saturating form of `i32.trunc_f32_s` | +| `i32.trunc_sat_f32_u` | `0xfc` `0x01` | | :bowling: saturating form of `i32.trunc_f32_u` | +| `i32.trunc_sat_f64_s` | `0xfc` `0x02` | | :bowling: saturating form of `i32.trunc_f64_s` | +| `i32.trunc_sat_f64_u` | `0xfc` `0x03` | | :bowling: saturating form of `i32.trunc_f64_u` | +| `i64.trunc_sat_f32_s` | `0xfc` `0x04` | | :bowling: saturating form of `i64.trunc_f32_s` | +| `i64.trunc_sat_f32_u` | `0xfc` `0x05` | | :bowling: saturating form of `i64.trunc_f32_u` | +| `i64.trunc_sat_f64_s` | `0xfc` `0x06` | | :bowling: saturating form of `i64.trunc_f64_s` | +| `i64.trunc_sat_f64_u` | `0xfc` `0x07` | | :bowling: saturating form of `i64.trunc_f64_u` | ## Reinterpretations ([described here](Semantics.md#datatype-conversions-truncations-reinterpretations-promotions-and-demotions)) diff --git a/Semantics.md b/Semantics.md index aa0ed9f..ae6db70 100644 --- a/Semantics.md +++ b/Semantics.md @@ -35,6 +35,14 @@ detailed in this document. [:unicorn:][future general] = Planned [future][future general] feature +## Post-MVP Features + +Some features were added post-MVP. These are indicated with the following symbols: + +| Symbol | Feature | +| ---------- | ------- | +| :bowling: | Saturating float to int conversions | + ## Traps Some operators may *trap* under some conditions, as noted below. In the MVP, @@ -640,6 +648,14 @@ is NaN, and *ordered* otherwise. * `f64.convert_u/i32`: convert an unsigned 32-bit integer to a 64-bit float * `f64.convert_u/i64`: convert an unsigned 64-bit integer to a 64-bit float * `f64.reinterpret/i64`: reinterpret the bits of a 64-bit integer as a 64-bit float + * `i32.trunc_sat_f32_s`: :bowling: truncate a 32-bit float to a signed 32-bit integer with saturation + * `i32.trunc_sat_f64_s`: :bowling: truncate a 64-bit float to a signed 32-bit integer with saturation + * `i32.trunc_sat_f32_u`: :bowling: truncate a 32-bit float to an unsigned 32-bit integer with saturation + * `i32.trunc_sat_f64_u`: :bowling: truncate a 64-bit float to an unsigned 32-bit integer with saturation + * `i64.trunc_sat_f32_s`: :bowling: truncate a 32-bit float to a signed 64-bit integer with saturation + * `i64.trunc_sat_f64_s`: :bowling: truncate a 64-bit float to a signed 64-bit integer with saturation + * `i64.trunc_sat_f32_u`: :bowling: truncate a 32-bit float to an unsigned 64-bit integer with saturation + * `i64.trunc_sat_f64_u`: :bowling: truncate a 64-bit float to an unsigned 64-bit integer with saturation Wrapping and extension of integer values always succeed. Promotion and demotion of floating point values always succeed. @@ -665,7 +681,12 @@ round-to-nearest ties-to-even rounding. Truncation from floating point to integer where IEEE 754-2008 would specify an invalid operator exception (e.g. when the floating point value is NaN or -outside the range which rounds to an integer in range) traps. +outside the range which rounds to an integer in range) is handled as follows: + - For instructions with no exceptional behavior specified, a trap is produced. + - :bowling: For instructions containing `_sat`, no trap is produced, and: + - If the floating-point value is positive, the maximum integer value is returned. + - If the floating-point value is negative, the minimum integer value is returned. + - If the floating-point value is NaN, zero is returned. ## Type-parametric operators -- cgit v1.2.3