Implement SIMD load splat and load extend

Add more pack types for SIMD load splat and load extends.

Author: ngzhian

interpreter/binary/encode.ml

@@ -561,6 +561,8 @@ let encode m =
       | SimdBitmask V128Op.(I32x4 Bitmask) -> simd_op 0xa4l
       | SimdBitmask (_) -> assert false
 
+      | _ -> assert false
+
     let const c =
       list instr c.it; end_ ()
 

interpreter/exec/simd.ml

@@ -175,6 +175,10 @@ sig
     val widen_low_u : t -> t
     val widen_high_u : t -> t
   end
+  module I64x2_convert : sig
+    val widen_low_s : t -> t
+    val widen_low_u : t -> t
+  end
   module F32x4_convert : sig
     val convert_i32x4_s : t -> t
     val convert_i32x4_u : t -> t
@@ -405,6 +409,16 @@ struct
     let widen_high_u = widen Lib.List.drop 0xffffl
   end
 
+  module I64x2_convert = struct
+    let widen mask x =
+      Rep.of_i64x2
+        (List.map
+           (fun i32 -> Int64.(logand mask (of_int32 i32)))
+           (Lib.List.take 2 (Rep.to_i32x4 x)))
+    let widen_low_s = widen 0xffffffffffffffffL
+    let widen_low_u = widen 0xffffffffL
+  end
+
   module F32x4_convert = struct
     let convert f v = Rep.of_f32x4 (List.map f (Rep.to_i32x4 v))
     let convert_i32x4_s = convert F32_convert.convert_i32_s

interpreter/runtime/memory.ml

@@ -122,13 +122,31 @@ let extend x n = function
   | ZX -> x
   | SX -> let sh = 64 - 8 * n in Int64.(shift_right (shift_left x sh) sh)
 
+let load_simd_packed sz ext x =
+  let b = Bytes.create 16 in
+  Bytes.set_int64_le b 0 x;
+  let v = V128.of_bits (Bytes.to_string b) in
+  match sz, ext with
+  | (Pack8x8, SX) -> V128 (V128.I16x8_convert.widen_low_s v)
+  | (Pack8x8, ZX) -> V128 (V128.I16x8_convert.widen_low_u v)
+  | (Pack16x4, SX) -> V128 (V128.I32x4_convert.widen_low_s v)
+  | (Pack16x4, ZX) -> V128 (V128.I32x4_convert.widen_low_u v)
+  | (Pack32x2, SX) -> V128 (V128.I64x2_convert.widen_low_s v)
+  | (Pack32x2, ZX) -> V128 (V128.I64x2_convert.widen_low_u v)
+  | (Pack8, ZX) -> V128 (V128.I8x16.splat (I8.of_int_s (Int64.to_int x)))
+  | (Pack16, ZX) -> V128 (V128.I16x8.splat (I16.of_int_s (Int64.to_int x)))
+  | (Pack32, ZX) -> V128 (V128.I32x4.splat (I32.of_int_s (Int64.to_int x)))
+  | (Pack64, ZX) -> V128 (V128.I64x2.splat x)
+  | _ -> assert false
+
 let load_packed sz ext mem a o t =
   assert (packed_size sz <= Types.size t);
   let n = packed_size sz in
   let x = extend (loadn mem a o n) n ext in
   match t with
   | I32Type -> I32 (Int64.to_int32 x)
   | I64Type -> I64 x
+  | V128Type -> load_simd_packed sz ext x
   | _ -> raise Type
 
 let store_packed sz mem a o v =

interpreter/syntax/operators.ml

@@ -218,7 +218,28 @@ let memory_grow = MemoryGrow
 
 (* SIMD *)
 let v128_load align offset = Load {ty = V128Type; align; offset; sz = None}
+let i16x8_load8x8_s align offset =
+  Load {ty = V128Type; align; offset; sz = Some (Pack8x8, SX)}
+let i16x8_load8x8_u align offset =
+  Load {ty = V128Type; align; offset; sz = Some (Pack8x8, ZX)}
+let i32x4_load16x4_s align offset =
+  Load {ty = V128Type; align; offset; sz = Some (Pack16x4, SX)}
+let i32x4_load16x4_u align offset =
+  Load {ty = V128Type; align; offset; sz = Some (Pack16x4, ZX)}
+let i64x2_load32x2_s align offset =
+  Load {ty = V128Type; align; offset; sz = Some (Pack32x2, SX)}
+let i64x2_load32x2_u align offset =
+  Load {ty = V128Type; align; offset; sz = Some (Pack32x2, ZX)}
+let v8x16_load_splat align offset =
+  Load {ty= V128Type; align; offset; sz = Some (Pack8, ZX)}
+let v16x8_load_splat align offset =
+  Load {ty= V128Type; align; offset; sz = Some (Pack16, ZX)}
+let v32x4_load_splat align offset =
+  Load {ty= V128Type; align; offset; sz = Some (Pack32, ZX)}
+let v64x2_load_splat align offset =
+  Load {ty= V128Type; align; offset; sz = Some (Pack64, ZX)}
 let v128_store align offset = Store {ty = V128Type; align; offset; sz = None}
+
 let v128_not = Unary (V128 (V128Op.V128 V128Op.Not))
 let v128_and = Binary (V128 (V128Op.V128 V128Op.And))
 let v128_andnot = Binary (V128 (V128Op.V128 V128Op.AndNot))

interpreter/syntax/types.ml

@@ -16,7 +16,7 @@ type extern_type =
   | ExternMemoryType of memory_type
   | ExternGlobalType of global_type
 
-type pack_size = Pack8 | Pack16 | Pack32
+type pack_size = Pack8 | Pack16 | Pack32 | Pack64 | Pack8x8 | Pack16x4 | Pack32x2
 type extension = SX | ZX
 
 
@@ -31,7 +31,7 @@ let packed_size = function
   | Pack8 -> 1
   | Pack16 -> 2
   | Pack32 -> 4
-
+  | Pack64 | Pack8x8 | Pack16x4 | Pack32x2 -> 8
 
 (* Subtyping *)
 

interpreter/text/arrange.ml

@@ -77,6 +77,10 @@ let pack_size = function
   | Pack8 -> "8"
   | Pack16 -> "16"
   | Pack32 -> "32"
+  | Pack64 -> "64"
+  | Pack8x8 -> "8x8"
+  | Pack16x4 -> "16x4"
+  | Pack32x2 -> "32x2"
 
 let extension = function
   | SX -> "_s"

interpreter/text/lexer.mll

@@ -288,6 +288,20 @@ rule token = parse
             (ext s i64_load8_s i64_load8_u (opt a 0))
             (ext s i64_load16_s i64_load16_u (opt a 1))
             (ext s i64_load32_s i64_load32_u (opt a 2)) o)) }
+  | "i16x8.load8x8_"(sign as s)
+  { LOAD (fun a o -> (ext s i16x8_load8x8_s i16x8_load8x8_u (opt a 3)) o) }
+  | "i32x4.load16x4_"(sign as s)
+  { LOAD (fun a o -> (ext s i32x4_load16x4_s i32x4_load16x4_u (opt a 3)) o) }
+  | "i64x2.load32x2_"(sign as s)
+  { LOAD (fun a o -> (ext s i64x2_load32x2_s i64x2_load32x2_u (opt a 3)) o) }
+  | "v8x16.load_splat"
+  { LOAD (fun a o -> (v8x16_load_splat (opt a 0)) o) }
+  | "v16x8.load_splat"
+  { LOAD (fun a o -> (v16x8_load_splat (opt a 1)) o) }
+  | "v32x4.load_splat"
+  { LOAD (fun a o -> (v32x4_load_splat (opt a 2)) o) }
+  | "v64x2.load_splat"
+  { LOAD (fun a o -> (v64x2_load_splat (opt a 3)) o) }
   | (ixx as t)".store"(mem_size as sz)
     { if t = "i32" && sz = "32" then error lexbuf "unknown operator";
       STORE (fun a o ->

test/core/simd/simd_align.wast

@@ -105,7 +105,7 @@
   (module quote
     "(memory 1) (func (drop (v128.load align=-1 (i32.const 0))))"
   )
-  "alignment must be a power of two"
+  "unknown operator"
 )
 (assert_malformed
   (module quote
@@ -123,7 +123,7 @@
   (module quote
     "(memory 1) (func (v128.store align=-1 (i32.const 0) (v128.const i32x4 0 0 0 0)))"
   )
-  "alignment must be a power of two"
+  "unknown operator"
 )
 (assert_malformed
   (module quote
@@ -141,7 +141,7 @@
   (module quote
     "(memory 1) (func (result v128) (i16x8.load8x8_s align=-1 (i32.const 0)))"
   )
-  "alignment must be a power of two"
+  "unknown operator"
 )
 (assert_malformed
   (module quote
@@ -159,7 +159,7 @@
   (module quote
     "(memory 1) (func (result v128) (i16x8.load8x8_u align=-1 (i32.const 0)))"
   )
-  "alignment must be a power of two"
+  "unknown operator"
 )
 (assert_malformed
   (module quote
@@ -177,7 +177,7 @@
   (module quote
     "(memory 1) (func (result v128) (i32x4.load16x4_s align=-1 (i32.const 0)))"
   )
-  "alignment must be a power of two"
+  "unknown operator"
 )
 (assert_malformed
   (module quote
@@ -195,7 +195,7 @@
   (module quote
     "(memory 1) (func (result v128) (i32x4.load16x4_u align=-1 (i32.const 0)))"
   )
-  "alignment must be a power of two"
+  "unknown operator"
 )
 (assert_malformed
   (module quote
@@ -213,7 +213,7 @@
   (module quote
     "(memory 1) (func (result v128) (i64x2.load32x2_s align=-1 (i32.const 0)))"
   )
-  "alignment must be a power of two"
+  "unknown operator"
 )
 (assert_malformed
   (module quote
@@ -231,7 +231,7 @@
   (module quote
     "(memory 1) (func (result v128) (i64x2.load32x2_u align=-1 (i32.const 0)))"
   )
-  "alignment must be a power of two"
+  "unknown operator"
 )
 (assert_malformed
   (module quote
@@ -249,7 +249,7 @@
   (module quote
     "(memory 1) (func (result v128) (v8x16.load_splat align=-1 (i32.const 0)))"
   )
-  "alignment must be a power of two"
+  "unknown operator"
 )
 (assert_malformed
   (module quote
@@ -261,7 +261,7 @@
   (module quote
     "(memory 1) (func (result v128) (v16x8.load_splat align=-1 (i32.const 0)))"
   )
-  "alignment must be a power of two"
+  "unknown operator"
 )
 (assert_malformed
   (module quote
@@ -273,7 +273,7 @@
   (module quote
     "(memory 1) (func (result v128) (v32x4.load_splat align=-1 (i32.const 0)))"
   )
-  "alignment must be a power of two"
+  "unknown operator"
 )
 (assert_malformed
   (module quote
@@ -291,7 +291,7 @@
   (module quote
     "(memory 1) (func (result v128) (v64x2.load_splat align=-1 (i32.const 0)))"
   )
-  "alignment must be a power of two"
+  "unknown operator"
 )
 (assert_malformed
   (module quote