Merge sign-extension-ops proposal into spec

See the sign-extension-ops proposal here:

https://github.com/WebAssembly/sign-extension-ops

This PR is built on top of #1143 (merge nontrapping-float-to-int)

document/core/appendix/index-instructions.rst

@@ -199,6 +199,11 @@ Instruction                                Binary Opcode              Type
 :math:`\I64.\REINTERPRET\K{\_}\F64`        :math:`\hex{BD}`           :math:`[\F64] \to [\I64]`                   :ref:`validation <valid-cvtop>`           :ref:`execution <exec-cvtop>`, :ref:`operator <op-reinterpret>`
 :math:`\F32.\REINTERPRET\K{\_}\I32`        :math:`\hex{BE}`           :math:`[\I32] \to [\F32]`                   :ref:`validation <valid-cvtop>`           :ref:`execution <exec-cvtop>`, :ref:`operator <op-reinterpret>`
 :math:`\F64.\REINTERPRET\K{\_}\I64`        :math:`\hex{BF}`           :math:`[\I64] \to [\F64]`                   :ref:`validation <valid-cvtop>`           :ref:`execution <exec-cvtop>`, :ref:`operator <op-reinterpret>`
+:math:`\I32.\EXTEND\K{8\_s}`               :math:`\hex{C0}`           :math:`[\I32] \to [\I32]`                   :ref:`validation <valid-unop>`            :ref:`execution <exec-unop>`, :ref:`operator <op-iextendn_s>`
+:math:`\I32.\EXTEND\K{16\_s}`              :math:`\hex{C1}`           :math:`[\I32] \to [\I32]`                   :ref:`validation <valid-unop>`            :ref:`execution <exec-unop>`, :ref:`operator <op-iextendn_s>`
+:math:`\I64.\EXTEND\K{8\_s}`               :math:`\hex{C2}`           :math:`[\I64] \to [\I64]`                   :ref:`validation <valid-unop>`            :ref:`execution <exec-unop>`, :ref:`operator <op-iextendn_s>`
+:math:`\I64.\EXTEND\K{16\_s}`              :math:`\hex{C3}`           :math:`[\I64] \to [\I64]`                   :ref:`validation <valid-unop>`            :ref:`execution <exec-unop>`, :ref:`operator <op-iextendn_s>`
+:math:`\I64.\EXTEND\K{32\_s}`              :math:`\hex{C4}`           :math:`[\I64] \to [\I64]`                   :ref:`validation <valid-unop>`            :ref:`execution <exec-unop>`, :ref:`operator <op-iextendn_s>`
 :math:`\I32.\TRUNC\K{\_sat\_}\F32\K{\_s}`  :math:`\hex{FC}~\hex{00}`  :math:`[\F32] \to [\I32]`                   :ref:`validation <valid-cvtop>`           :ref:`execution <exec-cvtop>`, :ref:`operator <op-trunc_s_sat>`
 :math:`\I32.\TRUNC\K{\_sat\_}\F32\K{\_u}`  :math:`\hex{FC}~\hex{01}`  :math:`[\F32] \to [\I32]`                   :ref:`validation <valid-cvtop>`           :ref:`execution <exec-cvtop>`, :ref:`operator <op-trunc_u_sat>`
 :math:`\I32.\TRUNC\K{\_sat\_}\F64\K{\_s}`  :math:`\hex{FC}~\hex{02}`  :math:`[\F64] \to [\I32]`                   :ref:`validation <valid-cvtop>`           :ref:`execution <exec-cvtop>`, :ref:`operator <op-trunc_s_sat>`

document/core/binary/instructions.rst

@@ -381,6 +381,17 @@ The saturating truncation instructions all have a one byte prefix.
    \end{array}
 
 
+.. math::
+   \begin{array}{llclll}
+   \phantom{\production{instruction}} & \phantom{\Binstr} &\phantom{::=}& \phantom{\dots} && \phantom{thisshouldbeenough} \\[-2ex] &&|&
+     \hex{C0} &\Rightarrow& \I32.\EXTEND\K{8\_s} \\ &&|&
+     \hex{C1} &\Rightarrow& \I32.\EXTEND\K{16\_s} \\ &&|&
+     \hex{C2} &\Rightarrow& \I64.\EXTEND\K{8\_s} \\ &&|&
+     \hex{C3} &\Rightarrow& \I64.\EXTEND\K{16\_s} \\ &&|&
+     \hex{C4} &\Rightarrow& \I64.\EXTEND\K{32\_s} \\
+   \end{array}
+
+
 .. index:: expression
    pair: binary format; expression
    single: expression; constant

document/core/exec/numerics.rst

@@ -612,6 +612,19 @@ The integer result of predicates -- i.e., :ref:`tests <syntax-testop>` and :ref:
    \end{array}
 
 
+.. _op-iextendn_s:
+
+:math:`\iextendns_N(i)`
+.......................
+
+* Return :math:`\extends_{M,N}(i)`.
+
+.. math::
+   \begin{array}{lll@{\qquad}l}
+   \iextendns_{N}(i) &=& \extends_{M,N}(i) \\
+   \end{array}
+
+
 .. index:: floating-point, IEEE 754
 .. _float-ops:
 

document/core/syntax/instructions.rst

@@ -64,8 +64,10 @@ These operations closely match respective operations available in hardware.
      \K{i}\X{nn}\K{.}\itestop \\&&|&
      \K{i}\X{nn}\K{.}\irelop ~|~
      \K{f}\X{nn}\K{.}\frelop \\&&|&
-     \K{i32.}\WRAP\K{\_i64} ~|~
+     \K{i}\X{nn}\K{.}\EXTEND\K{8\_s} ~|~
+     \K{i}\X{nn}\K{.}\EXTEND\K{16\_s} ~|~
      \K{i64.}\EXTEND\K{\_i32}\K{\_}\sx ~|~
+     \K{i32.}\WRAP\K{\_i64} ~|~
      \K{i}\X{nn}\K{.}\TRUNC\K{\_f}\X{mm}\K{\_}\sx \\&&|&
      \K{i}\X{nn}\K{.}\TRUNC\K{\_sat\_f}\X{mm}\K{\_}\sx \\&&|&
      \K{f32.}\DEMOTE\K{\_f64} ~|~
@@ -153,7 +155,10 @@ Occasionally, it is convenient to group operators together according to the foll
 
 .. math::
    \begin{array}{llll}
-   \production{unary operator} & \unop &::=& \iunop ~|~ \funop \\
+   \production{unary operator} & \unop &::=&
+     \iunop ~|~
+     \funop ~|~
+     \EXTEND{N}\K{\_s} ~|~ \\
    \production{binary operator} & \binop &::=& \ibinop ~|~ \fbinop \\
    \production{test operator} & \testop &::=& \itestop \\
    \production{relational operator} & \relop &::=& \irelop ~|~ \frelop \\

document/core/text/instructions.rst

@@ -421,6 +421,16 @@ Numeric Instructions
      \text{f64.reinterpret\_i64} &\Rightarrow& \F64.\REINTERPRET\K{\_}\I64 \\
    \end{array}
 
+.. math::
+   \begin{array}{llclll}
+   \phantom{\production{instruction}} & \phantom{\Tplaininstr_I} &\phantom{::=}& \phantom{thisisenough} && \phantom{thisshouldbeenough} \\[-2ex] &&|&
+     \text{i32.extend8\_s} &\Rightarrow& \I32.\EXTEND\K{8\_s} \\ &&|&
+     \text{i32.extend16\_s} &\Rightarrow& \I32.\EXTEND\K{16\_s} \\ &&|&
+     \text{i64.extend8\_s} &\Rightarrow& \I64.\EXTEND\K{8\_s} \\ &&|&
+     \text{i64.extend16\_s} &\Rightarrow& \I64.\EXTEND\K{16\_s} \\ &&|&
+     \text{i64.extend32\_s} &\Rightarrow& \I64.\EXTEND\K{32\_s} \\
+   \end{array}
+
 
 .. index:: ! folded instruction, S-expression
 .. _text-foldedinstr:

document/core/util/macros.def

@@ -920,6 +920,7 @@
 .. |iles| mathdef:: \xref{exec/numerics}{op-ile_s}{\F{ile\_s}}
 .. |igeu| mathdef:: \xref{exec/numerics}{op-ige_u}{\F{ige\_u}}
 .. |iges| mathdef:: \xref{exec/numerics}{op-ige_s}{\F{ige\_s}}
+.. |iextendns| mathdef:: \xref{exec/numerics}{op-iextendn_s}{\F{iextend}M\F{\_s}}
 
 .. |fadd| mathdef:: \xref{exec/numerics}{op-fadd}{\F{fadd}}
 .. |fsub| mathdef:: \xref{exec/numerics}{op-fsub}{\F{fsub}}

interpreter/binary/decode.ml

@@ -445,6 +445,12 @@ let rec instr s =
   | 0xbe -> f32_reinterpret_i32
   | 0xbf -> f64_reinterpret_i64
 
+  | 0xc0 -> i32_extend8_s
+  | 0xc1 -> i32_extend16_s
+  | 0xc2 -> i64_extend8_s
+  | 0xc3 -> i64_extend16_s
+  | 0xc4 -> i64_extend32_s
+
   | 0xfc -> math_prefix s
 
   | b -> illegal s pos b

interpreter/binary/encode.ml

@@ -130,7 +130,6 @@ let encode m =
     open Source
     open Ast
     open Values
-    open Memory
 
     let op n = u8 n
     let end_ () = op 0x0b
@@ -260,10 +259,16 @@ let encode m =
       | Unary (I32 I32Op.Clz) -> op 0x67
       | Unary (I32 I32Op.Ctz) -> op 0x68
       | Unary (I32 I32Op.Popcnt) -> op 0x69
+      | Unary (I32 (I32Op.ExtendS Pack8)) -> op 0xc0
+      | Unary (I32 (I32Op.ExtendS Pack16)) -> op 0xc1
+      | Unary (I32 (I32Op.ExtendS Pack32)) -> assert false
 
       | Unary (I64 I64Op.Clz) -> op 0x79
       | Unary (I64 I64Op.Ctz) -> op 0x7a
       | Unary (I64 I64Op.Popcnt) -> op 0x7b
+      | Unary (I64 (I64Op.ExtendS Pack8)) -> op 0xc2
+      | Unary (I64 (I64Op.ExtendS Pack16)) -> op 0xc3
+      | Unary (I64 (I64Op.ExtendS Pack32)) -> op 0xc4
 
       | Unary (F32 F32Op.Abs) -> op 0x8b
       | Unary (F32 F32Op.Neg) -> op 0x8c

interpreter/exec/eval_numeric.ml

@@ -24,6 +24,7 @@ struct
       | Clz -> IXX.clz
       | Ctz -> IXX.ctz
       | Popcnt -> IXX.popcnt
+      | ExtendS sz -> IXX.extend_s (8 * packed_size sz)
     in fun v -> to_value (f (of_value 1 v))
 
   let binop op =

interpreter/exec/int.ml

@@ -58,6 +58,7 @@ sig
   val clz : t -> t
   val ctz : t -> t
   val popcnt : t -> t
+  val extend_s : int -> t -> t
   val eqz : t -> bool
   val eq : t -> t -> bool
   val ne : t -> t -> bool
@@ -201,6 +202,10 @@ struct
         loop acc' (i - 1) (Rep.shift_right_logical n 1)
     in Rep.of_int (loop 0 Rep.bitwidth x)
 
+  let extend_s n x =
+    let shift = Rep.bitwidth - n in
+    Rep.shift_right (Rep.shift_left x shift) shift
+
   let eqz x = x = Rep.zero
 
   let eq x y = x = y

interpreter/runtime/memory.ml

@@ -7,9 +7,6 @@ type size = int32  (* number of pages *)
 type address = int64
 type offset = int32
 
-type pack_size = Pack8 | Pack16 | Pack32
-type extension = SX | ZX
-
 type memory' = (int, int8_unsigned_elt, c_layout) Array1.t
 type memory = {mutable content : memory'; max : size option}
 type t = memory
@@ -22,11 +19,6 @@ exception OutOfMemory
 
 let page_size = 0x10000L (* 64 KiB *)
 
-let packed_size = function
-  | Pack8 -> 1
-  | Pack16 -> 2
-  | Pack32 -> 4
-
 let within_limits n = function
   | None -> true
   | Some max -> I32.le_u n max

interpreter/runtime/memory.mli

@@ -8,17 +8,13 @@ type size = int32  (* number of pages *)
 type address = int64
 type offset = int32
 
-type pack_size = Pack8 | Pack16 | Pack32
-type extension = SX | ZX
-
 exception Type
 exception Bounds
 exception SizeOverflow
 exception SizeLimit
 exception OutOfMemory
 
 val page_size : int64
-val packed_size : pack_size -> int
 
 val alloc : memory_type -> memory (* raises SizeOverflow, OutOfMemory *)
 val type_of : memory -> memory_type

interpreter/syntax/ast.ml

@@ -23,7 +23,7 @@ open Types
 
 module IntOp =
 struct
-  type unop = Clz | Ctz | Popcnt
+  type unop = Clz | Ctz | Popcnt | ExtendS of pack_size
   type binop = Add | Sub | Mul | DivS | DivU | RemS | RemU
              | And | Or | Xor | Shl | ShrS | ShrU | Rotl | Rotr
   type testop = Eqz
@@ -58,8 +58,8 @@ type cvtop = (I32Op.cvtop, I64Op.cvtop, F32Op.cvtop, F64Op.cvtop) Values.op
 
 type 'a memop =
   {ty : value_type; align : int; offset : Memory.offset; sz : 'a option}
-type loadop = (Memory.pack_size * Memory.extension) memop
-type storeop = Memory.pack_size memop
+type loadop = (pack_size * extension) memop
+type storeop = pack_size memop
 
 
 (* Expressions *)

interpreter/syntax/operators.ml

@@ -1,7 +1,6 @@
 open Source
 open Types
 open Values
-open Memory
 open Ast
 
 
@@ -207,6 +206,12 @@ let i64_reinterpret_f64 = Convert (I64 I64Op.ReinterpretFloat)
 let f32_reinterpret_i32 = Convert (F32 F32Op.ReinterpretInt)
 let f64_reinterpret_i64 = Convert (F64 F64Op.ReinterpretInt)
 
+let i32_extend8_s = Unary (I32 (I32Op.ExtendS Pack8))
+let i32_extend16_s = Unary (I32 (I32Op.ExtendS Pack16))
+let i64_extend8_s = Unary (I64 (I64Op.ExtendS Pack8))
+let i64_extend16_s = Unary (I64 (I64Op.ExtendS Pack16))
+let i64_extend32_s = Unary (I64 (I64Op.ExtendS Pack32))
+
 let memory_size = MemorySize
 let memory_grow = MemoryGrow
 

interpreter/syntax/types.ml

@@ -16,13 +16,21 @@ type extern_type =
   | ExternMemoryType of memory_type
   | ExternGlobalType of global_type
 
+type pack_size = Pack8 | Pack16 | Pack32
+type extension = SX | ZX
+
 
 (* Attributes *)
 
 let size = function
   | I32Type | F32Type -> 4
   | I64Type | F64Type -> 8
 
+let packed_size = function
+  | Pack8 -> 1
+  | Pack16 -> 2
+  | Pack32 -> 4
+
 
 (* Subtyping *)
 

interpreter/text/arrange.ml

@@ -75,6 +75,15 @@ let global_type = function
   | GlobalType (t, Immutable) -> atom string_of_value_type t
   | GlobalType (t, Mutable) -> Node ("mut", [atom string_of_value_type t])
 
+let pack_size = function
+  | Pack8 -> "8"
+  | Pack16 -> "16"
+  | Pack32 -> "32"
+
+let extension = function
+  | SX -> "_s"
+  | ZX -> "_u"
+
 
 (* Operators *)
 
@@ -101,6 +110,7 @@ struct
     | Clz -> "clz"
     | Ctz -> "ctz"
     | Popcnt -> "popcnt"
+    | ExtendS sz -> "extend" ^ pack_size sz ^ "_s"
 
   let binop xx = function
     | Add -> "add"
@@ -191,15 +201,6 @@ let testop = oper (IntOp.testop, FloatOp.testop)
 let relop = oper (IntOp.relop, FloatOp.relop)
 let cvtop = oper (IntOp.cvtop, FloatOp.cvtop)
 
-let pack_size = function
-  | Memory.Pack8 -> "8"
-  | Memory.Pack16 -> "16"
-  | Memory.Pack32 -> "32"
-
-let extension = function
-  | Memory.SX -> "_s"
-  | Memory.ZX -> "_u"
-
 let memop name {ty; align; offset; _} sz =
   value_type ty ^ "." ^ name ^
   (if offset = 0l then "" else " offset=" ^ nat32 offset) ^
@@ -209,12 +210,12 @@ let loadop op =
   match op.sz with
   | None -> memop "load" op (size op.ty)
   | Some (sz, ext) ->
-    memop ("load" ^ pack_size sz ^ extension ext) op (Memory.packed_size sz)
+    memop ("load" ^ pack_size sz ^ extension ext) op (packed_size sz)
 
 let storeop op =
   match op.sz with
   | None -> memop "store" op (size op.ty)
-  | Some sz -> memop ("store" ^ pack_size sz) op (Memory.packed_size sz)
+  | Some sz -> memop ("store" ^ pack_size sz) op (packed_size sz)
 
 
 (* Expressions *)

interpreter/text/lexer.mll

@@ -238,6 +238,9 @@ rule token = parse
   | (ixx as t)".clz" { UNARY (intop t i32_clz i64_clz) }
   | (ixx as t)".ctz" { UNARY (intop t i32_ctz i64_ctz) }
   | (ixx as t)".popcnt" { UNARY (intop t i32_popcnt i64_popcnt) }
+  | (ixx as t)".extend8_s" { UNARY (intop t i32_extend8_s i64_extend8_s) }
+  | (ixx as t)".extend16_s" { UNARY (intop t i32_extend16_s i64_extend16_s) }
+  | "i64.extend32_s" { UNARY i64_extend32_s }
   | (fxx as t)".neg" { UNARY (floatop t f32_neg f64_neg) }
   | (fxx as t)".abs" { UNARY (floatop t f32_abs f64_abs) }
   | (fxx as t)".sqrt" { UNARY (floatop t f32_sqrt f64_sqrt) }

interpreter/valid/valid.ml

@@ -139,15 +139,23 @@ let type_cvtop at = function
 
 (* Expressions *)
 
+let check_pack sz t at =
+  require (packed_size sz < size t) at "invalid sign extension"
+
+let check_unop unop at =
+  match unop with
+  | Values.I32 (IntOp.ExtendS sz) | Values.I64 (IntOp.ExtendS sz) ->
+    check_pack sz (Values.type_of unop) at
+  | _ -> ()
+
 let check_memop (c : context) (memop : 'a memop) get_sz at =
   ignore (memory c (0l @@ at));
   let size =
     match get_sz memop.sz with
     | None -> size memop.ty
     | Some sz ->
-      require (memop.ty = I64Type || sz <> Memory.Pack32) at
-        "memory size too big";
-      Memory.packed_size sz
+      check_pack sz memop.ty at;
+      packed_size sz
   in
   require (1 lsl memop.align <= size) at
     "alignment must not be larger than natural"
@@ -279,6 +287,7 @@ let rec check_instr (c : context) (e : instr) (s : infer_stack_type) : op_type =
     [t; t] --> [I32Type]
 
   | Unary unop ->
+    check_unop unop e.at;
     let t = type_unop unop in
     [t] --> [t]