Backed out 2 changesets (bug 1416723) for failures in dom/serviceworkers/test/test_serviceworker_interfaces.html on a CLOSED TREE

Backed out changeset b2242216d11b (bug 1416723)
Backed out changeset bfaf82051dfd (bug 1416723)

js/public/ProtoKey.h

@@ -48,6 +48,12 @@
 #define IF_BDATA(real,imaginary) imaginary
 #endif
 
+#ifdef ENABLE_SIMD
+# define IF_SIMD(real,imaginary) real
+#else
+# define IF_SIMD(real,imaginary) imaginary
+#endif
+
 #ifdef ENABLE_SHARED_ARRAY_BUFFER
 #define IF_SAB(real,imaginary) real
 #else
@@ -100,6 +106,7 @@ IF_SAB(real,imaginary)(SharedArrayBuffer,       InitViaClassSpec, OCLASP(SharedA
 IF_INTL(real,imaginary) (Intl,                  InitIntlClass,          CLASP(Intl)) \
 IF_BDATA(real,imaginary)(TypedObject,           InitTypedObjectModuleObject,   OCLASP(TypedObjectModule)) \
     real(Reflect,               InitReflect,            nullptr) \
+IF_SIMD(real,imaginary)(SIMD,                   InitSimdClass, OCLASP(Simd)) \
     real(WeakSet,               InitWeakSetClass,       OCLASP(WeakSet)) \
     real(TypedArray,            InitViaClassSpec,       &js::TypedArrayObject::sharedTypedArrayPrototypeClass) \
 IF_SAB(real,imaginary)(Atomics, InitAtomicsClass, OCLASP(Atomics)) \

js/src/builtin/SIMD.h

@@ -0,0 +1,298 @@
+/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*-
+ * vim: set ts=8 sts=4 et sw=4 tw=99:
+ * This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef builtin_SIMD_h
+#define builtin_SIMD_h
+
+#include "jsapi.h"
+#include "NamespaceImports.h"
+
+#include "builtin/SIMDConstants.h"
+#include "jit/IonTypes.h"
+#include "js/Conversions.h"
+
+/*
+ * JS SIMD functions.
+ * Spec matching polyfill:
+ * https://github.com/tc39/ecmascript_simd/blob/master/src/ecmascript_simd.js
+ */
+
+namespace js {
+
+class GlobalObject;
+
+// These classes implement the concept containing the following constraints:
+// - requires typename Elem: this is the scalar lane type, stored in each lane
+// of the SIMD vector.
+// - requires static const unsigned lanes: this is the number of lanes (length)
+// of the SIMD vector.
+// - requires static const SimdType type: this is the SimdType enum value
+// corresponding to the SIMD type.
+// - requires static bool Cast(JSContext*, JS::HandleValue, Elem*): casts a
+// given Value to the current scalar lane type and saves it in the Elem
+// out-param.
+// - requires static Value ToValue(Elem): returns a Value of the right type
+// containing the given value.
+//
+// This concept is used in the templates above to define the functions
+// associated to a given type and in their implementations, to avoid code
+// redundancy.
+
+struct Float32x4 {
+    typedef float Elem;
+    static const unsigned lanes = 4;
+    static const SimdType type = SimdType::Float32x4;
+    static MOZ_MUST_USE bool Cast(JSContext* cx, JS::HandleValue v, Elem* out) {
+        double d;
+        if (!ToNumber(cx, v, &d))
+            return false;
+        *out = float(d);
+        return true;
+    }
+    static Value ToValue(Elem value) {
+        return DoubleValue(JS::CanonicalizeNaN(value));
+    }
+};
+
+struct Float64x2 {
+    typedef double Elem;
+    static const unsigned lanes = 2;
+    static const SimdType type = SimdType::Float64x2;
+    static MOZ_MUST_USE bool Cast(JSContext* cx, JS::HandleValue v, Elem* out) {
+        return ToNumber(cx, v, out);
+    }
+    static Value ToValue(Elem value) {
+        return DoubleValue(JS::CanonicalizeNaN(value));
+    }
+};
+
+struct Int8x16 {
+    typedef int8_t Elem;
+    static const unsigned lanes = 16;
+    static const SimdType type = SimdType::Int8x16;
+    static MOZ_MUST_USE bool Cast(JSContext* cx, JS::HandleValue v, Elem* out) {
+        return ToInt8(cx, v, out);
+    }
+    static Value ToValue(Elem value) {
+        return NumberValue(value);
+    }
+};
+
+struct Int16x8 {
+    typedef int16_t Elem;
+    static const unsigned lanes = 8;
+    static const SimdType type = SimdType::Int16x8;
+    static MOZ_MUST_USE bool Cast(JSContext* cx, JS::HandleValue v, Elem* out) {
+        return ToInt16(cx, v, out);
+    }
+    static Value ToValue(Elem value) {
+        return NumberValue(value);
+    }
+};
+
+struct Int32x4 {
+    typedef int32_t Elem;
+    static const unsigned lanes = 4;
+    static const SimdType type = SimdType::Int32x4;
+    static MOZ_MUST_USE bool Cast(JSContext* cx, JS::HandleValue v, Elem* out) {
+        return ToInt32(cx, v, out);
+    }
+    static Value ToValue(Elem value) {
+        return NumberValue(value);
+    }
+};
+
+struct Uint8x16 {
+    typedef uint8_t Elem;
+    static const unsigned lanes = 16;
+    static const SimdType type = SimdType::Uint8x16;
+    static MOZ_MUST_USE bool Cast(JSContext* cx, JS::HandleValue v, Elem* out) {
+        return ToUint8(cx, v, out);
+    }
+    static Value ToValue(Elem value) {
+        return NumberValue(value);
+    }
+};
+
+struct Uint16x8 {
+    typedef uint16_t Elem;
+    static const unsigned lanes = 8;
+    static const SimdType type = SimdType::Uint16x8;
+    static MOZ_MUST_USE bool Cast(JSContext* cx, JS::HandleValue v, Elem* out) {
+        return ToUint16(cx, v, out);
+    }
+    static Value ToValue(Elem value) {
+        return NumberValue(value);
+    }
+};
+
+struct Uint32x4 {
+    typedef uint32_t Elem;
+    static const unsigned lanes = 4;
+    static const SimdType type = SimdType::Uint32x4;
+    static MOZ_MUST_USE bool Cast(JSContext* cx, JS::HandleValue v, Elem* out) {
+        return ToUint32(cx, v, out);
+    }
+    static Value ToValue(Elem value) {
+        return NumberValue(value);
+    }
+};
+
+struct Bool8x16 {
+    typedef int8_t Elem;
+    static const unsigned lanes = 16;
+    static const SimdType type = SimdType::Bool8x16;
+    static MOZ_MUST_USE bool Cast(JSContext* cx, JS::HandleValue v, Elem* out) {
+        *out = ToBoolean(v) ? -1 : 0;
+        return true;
+    }
+    static Value ToValue(Elem value) {
+        return BooleanValue(value);
+    }
+};
+
+struct Bool16x8 {
+    typedef int16_t Elem;
+    static const unsigned lanes = 8;
+    static const SimdType type = SimdType::Bool16x8;
+    static MOZ_MUST_USE bool Cast(JSContext* cx, JS::HandleValue v, Elem* out) {
+        *out = ToBoolean(v) ? -1 : 0;
+        return true;
+    }
+    static Value ToValue(Elem value) {
+        return BooleanValue(value);
+    }
+};
+
+struct Bool32x4 {
+    typedef int32_t Elem;
+    static const unsigned lanes = 4;
+    static const SimdType type = SimdType::Bool32x4;
+    static MOZ_MUST_USE bool Cast(JSContext* cx, JS::HandleValue v, Elem* out) {
+        *out = ToBoolean(v) ? -1 : 0;
+        return true;
+    }
+    static Value ToValue(Elem value) {
+        return BooleanValue(value);
+    }
+};
+
+struct Bool64x2 {
+    typedef int64_t Elem;
+    static const unsigned lanes = 2;
+    static const SimdType type = SimdType::Bool64x2;
+    static MOZ_MUST_USE bool Cast(JSContext* cx, JS::HandleValue v, Elem* out) {
+        *out = ToBoolean(v) ? -1 : 0;
+        return true;
+    }
+    static Value ToValue(Elem value) {
+        return BooleanValue(value);
+    }
+};
+
+// Get the well known name of the SIMD.* object corresponding to type.
+PropertyName* SimdTypeToName(const JSAtomState& atoms, SimdType type);
+
+// Check if name is the well known name of a SIMD type.
+// Returns true and sets *type iff name is known.
+bool IsSimdTypeName(const JSAtomState& atoms, const PropertyName* name, SimdType* type);
+
+const char* SimdTypeToString(SimdType type);
+
+template<typename V>
+JSObject* CreateSimd(JSContext* cx, const typename V::Elem* data);
+
+template<typename V>
+bool IsVectorObject(HandleValue v);
+
+template<typename V>
+MOZ_MUST_USE bool ToSimdConstant(JSContext* cx, HandleValue v, jit::SimdConstant* out);
+
+JSObject*
+InitSimdClass(JSContext* cx, Handle<GlobalObject*> global);
+
+namespace jit {
+
+extern const JSJitInfo JitInfo_SimdInt32x4_extractLane;
+extern const JSJitInfo JitInfo_SimdFloat32x4_extractLane;
+
+} // namespace jit
+
+#define DECLARE_SIMD_FLOAT32X4_FUNCTION(Name, Func, Operands)   \
+extern MOZ_MUST_USE bool                                        \
+simd_float32x4_##Name(JSContext* cx, unsigned argc, Value* vp);
+FLOAT32X4_FUNCTION_LIST(DECLARE_SIMD_FLOAT32X4_FUNCTION)
+#undef DECLARE_SIMD_FLOAT32X4_FUNCTION
+
+#define DECLARE_SIMD_FLOAT64X2_FUNCTION(Name, Func, Operands)   \
+extern MOZ_MUST_USE bool                                        \
+simd_float64x2_##Name(JSContext* cx, unsigned argc, Value* vp);
+FLOAT64X2_FUNCTION_LIST(DECLARE_SIMD_FLOAT64X2_FUNCTION)
+#undef DECLARE_SIMD_FLOAT64X2_FUNCTION
+
+#define DECLARE_SIMD_INT8X16_FUNCTION(Name, Func, Operands)     \
+extern MOZ_MUST_USE bool                                        \
+simd_int8x16_##Name(JSContext* cx, unsigned argc, Value* vp);
+INT8X16_FUNCTION_LIST(DECLARE_SIMD_INT8X16_FUNCTION)
+#undef DECLARE_SIMD_INT8X16_FUNCTION
+
+#define DECLARE_SIMD_INT16X8_FUNCTION(Name, Func, Operands)     \
+extern MOZ_MUST_USE bool                                        \
+simd_int16x8_##Name(JSContext* cx, unsigned argc, Value* vp);
+INT16X8_FUNCTION_LIST(DECLARE_SIMD_INT16X8_FUNCTION)
+#undef DECLARE_SIMD_INT16X8_FUNCTION
+
+#define DECLARE_SIMD_INT32X4_FUNCTION(Name, Func, Operands)     \
+extern MOZ_MUST_USE bool                                        \
+simd_int32x4_##Name(JSContext* cx, unsigned argc, Value* vp);
+INT32X4_FUNCTION_LIST(DECLARE_SIMD_INT32X4_FUNCTION)
+#undef DECLARE_SIMD_INT32X4_FUNCTION
+
+#define DECLARE_SIMD_UINT8X16_FUNCTION(Name, Func, Operands)    \
+extern MOZ_MUST_USE bool                                        \
+simd_uint8x16_##Name(JSContext* cx, unsigned argc, Value* vp);
+UINT8X16_FUNCTION_LIST(DECLARE_SIMD_UINT8X16_FUNCTION)
+#undef DECLARE_SIMD_UINT8X16_FUNCTION
+
+#define DECLARE_SIMD_UINT16X8_FUNCTION(Name, Func, Operands)    \
+extern MOZ_MUST_USE bool                                        \
+simd_uint16x8_##Name(JSContext* cx, unsigned argc, Value* vp);
+UINT16X8_FUNCTION_LIST(DECLARE_SIMD_UINT16X8_FUNCTION)
+#undef DECLARE_SIMD_UINT16X8_FUNCTION
+
+#define DECLARE_SIMD_UINT32X4_FUNCTION(Name, Func, Operands)    \
+extern MOZ_MUST_USE bool                                        \
+simd_uint32x4_##Name(JSContext* cx, unsigned argc, Value* vp);
+UINT32X4_FUNCTION_LIST(DECLARE_SIMD_UINT32X4_FUNCTION)
+#undef DECLARE_SIMD_UINT32X4_FUNCTION
+
+#define DECLARE_SIMD_BOOL8X16_FUNCTION(Name, Func, Operands)    \
+extern MOZ_MUST_USE bool                                        \
+simd_bool8x16_##Name(JSContext* cx, unsigned argc, Value* vp);
+BOOL8X16_FUNCTION_LIST(DECLARE_SIMD_BOOL8X16_FUNCTION)
+#undef DECLARE_SIMD_BOOL8X16_FUNCTION
+
+#define DECLARE_SIMD_BOOL16X8_FUNCTION(Name, Func, Operands)    \
+extern MOZ_MUST_USE bool                                        \
+simd_bool16x8_##Name(JSContext* cx, unsigned argc, Value* vp);
+BOOL16X8_FUNCTION_LIST(DECLARE_SIMD_BOOL16X8_FUNCTION)
+#undef DECLARE_SIMD_BOOL16X8_FUNCTION
+
+#define DECLARE_SIMD_BOOL32X4_FUNCTION(Name, Func, Operands)    \
+extern MOZ_MUST_USE bool                                        \
+simd_bool32x4_##Name(JSContext* cx, unsigned argc, Value* vp);
+BOOL32X4_FUNCTION_LIST(DECLARE_SIMD_BOOL32X4_FUNCTION)
+#undef DECLARE_SIMD_BOOL32X4_FUNCTION
+
+#define DECLARE_SIMD_BOOL64X2_FUNCTION(Name, Func, Operands)    \
+extern MOZ_MUST_USE bool                                        \
+simd_bool64x2_##Name(JSContext* cx, unsigned argc, Value* vp);
+BOOL64X2_FUNCTION_LIST(DECLARE_SIMD_BOOL64X2_FUNCTION)
+#undef DECLARE_SIMD_BOOL64X2_FUNCTION
+
+}  /* namespace js */
+
+#endif /* builtin_SIMD_h */

js/src/builtin/SIMDConstants.h

@@ -0,0 +1,941 @@
+/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*-
+ * vim: set ts=8 sts=4 et sw=4 tw=99:
+ * This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef builtin_SIMDConstants_h
+#define builtin_SIMDConstants_h
+
+#include "mozilla/Assertions.h"
+
+#include "builtin/TypedObjectConstants.h"
+
+// Bool8x16.
+#define BOOL8X16_UNARY_FUNCTION_LIST(V)                                               \
+  V(not, (UnaryFunc<Bool8x16, LogicalNot, Bool8x16>), 1)                              \
+  V(check, (UnaryFunc<Bool8x16, Identity, Bool8x16>), 1)                              \
+  V(splat, (FuncSplat<Bool8x16>), 1)                                                  \
+  V(allTrue, (AllTrue<Bool8x16>), 1)                                                  \
+  V(anyTrue, (AnyTrue<Bool8x16>), 1)
+
+#define BOOL8X16_BINARY_FUNCTION_LIST(V)                                              \
+  V(extractLane, (ExtractLane<Bool8x16>), 2)                                          \
+  V(and, (BinaryFunc<Bool8x16, And, Bool8x16>), 2)                                    \
+  V(or, (BinaryFunc<Bool8x16, Or, Bool8x16>), 2)                                      \
+  V(xor, (BinaryFunc<Bool8x16, Xor, Bool8x16>), 2)                                    \
+
+#define BOOL8X16_TERNARY_FUNCTION_LIST(V)                                             \
+  V(replaceLane, (ReplaceLane<Bool8x16>), 3)
+
+#define BOOL8X16_FUNCTION_LIST(V)                                                     \
+  BOOL8X16_UNARY_FUNCTION_LIST(V)                                                     \
+  BOOL8X16_BINARY_FUNCTION_LIST(V)                                                    \
+  BOOL8X16_TERNARY_FUNCTION_LIST(V)
+
+// Bool 16x8.
+#define BOOL16X8_UNARY_FUNCTION_LIST(V)                                               \
+  V(not, (UnaryFunc<Bool16x8, LogicalNot, Bool16x8>), 1)                              \
+  V(check, (UnaryFunc<Bool16x8, Identity, Bool16x8>), 1)                              \
+  V(splat, (FuncSplat<Bool16x8>), 1)                                                  \
+  V(allTrue, (AllTrue<Bool16x8>), 1)                                                  \
+  V(anyTrue, (AnyTrue<Bool16x8>), 1)
+
+#define BOOL16X8_BINARY_FUNCTION_LIST(V)                                              \
+  V(extractLane, (ExtractLane<Bool16x8>), 2)                                          \
+  V(and, (BinaryFunc<Bool16x8, And, Bool16x8>), 2)                                    \
+  V(or, (BinaryFunc<Bool16x8, Or, Bool16x8>), 2)                                      \
+  V(xor, (BinaryFunc<Bool16x8, Xor, Bool16x8>), 2)                                    \
+
+#define BOOL16X8_TERNARY_FUNCTION_LIST(V)                                             \
+  V(replaceLane, (ReplaceLane<Bool16x8>), 3)
+
+#define BOOL16X8_FUNCTION_LIST(V)                                                     \
+  BOOL16X8_UNARY_FUNCTION_LIST(V)                                                     \
+  BOOL16X8_BINARY_FUNCTION_LIST(V)                                                    \
+  BOOL16X8_TERNARY_FUNCTION_LIST(V)
+
+// Bool32x4.
+#define BOOL32X4_UNARY_FUNCTION_LIST(V)                                               \
+  V(not, (UnaryFunc<Bool32x4, LogicalNot, Bool32x4>), 1)                              \
+  V(check, (UnaryFunc<Bool32x4, Identity, Bool32x4>), 1)                              \
+  V(splat, (FuncSplat<Bool32x4>), 1)                                                  \
+  V(allTrue, (AllTrue<Bool32x4>), 1)                                                  \
+  V(anyTrue, (AnyTrue<Bool32x4>), 1)
+
+#define BOOL32X4_BINARY_FUNCTION_LIST(V)                                              \
+  V(extractLane, (ExtractLane<Bool32x4>), 2)                                          \
+  V(and, (BinaryFunc<Bool32x4, And, Bool32x4>), 2)                                    \
+  V(or, (BinaryFunc<Bool32x4, Or, Bool32x4>), 2)                                      \
+  V(xor, (BinaryFunc<Bool32x4, Xor, Bool32x4>), 2)                                    \
+
+#define BOOL32X4_TERNARY_FUNCTION_LIST(V)                                             \
+  V(replaceLane, (ReplaceLane<Bool32x4>), 3)
+
+#define BOOL32X4_FUNCTION_LIST(V)                                                     \
+  BOOL32X4_UNARY_FUNCTION_LIST(V)                                                     \
+  BOOL32X4_BINARY_FUNCTION_LIST(V)                                                    \
+  BOOL32X4_TERNARY_FUNCTION_LIST(V)
+
+// Bool64x2.
+#define BOOL64X2_UNARY_FUNCTION_LIST(V)                                               \
+  V(not, (UnaryFunc<Bool64x2, LogicalNot, Bool64x2>), 1)                              \
+  V(check, (UnaryFunc<Bool64x2, Identity, Bool64x2>), 1)                              \
+  V(splat, (FuncSplat<Bool64x2>), 1)                                                  \
+  V(allTrue, (AllTrue<Bool64x2>), 1)                                                  \
+  V(anyTrue, (AnyTrue<Bool64x2>), 1)
+
+#define BOOL64X2_BINARY_FUNCTION_LIST(V)                                              \
+  V(extractLane, (ExtractLane<Bool64x2>), 2)                                          \
+  V(and, (BinaryFunc<Bool64x2, And, Bool64x2>), 2)                                    \
+  V(or, (BinaryFunc<Bool64x2, Or, Bool64x2>), 2)                                      \
+  V(xor, (BinaryFunc<Bool64x2, Xor, Bool64x2>), 2)                                    \
+
+#define BOOL64X2_TERNARY_FUNCTION_LIST(V)                                             \
+  V(replaceLane, (ReplaceLane<Bool64x2>), 3)
+
+#define BOOL64X2_FUNCTION_LIST(V)                                                     \
+  BOOL64X2_UNARY_FUNCTION_LIST(V)                                                     \
+  BOOL64X2_BINARY_FUNCTION_LIST(V)                                                    \
+  BOOL64X2_TERNARY_FUNCTION_LIST(V)
+
+// Float32x4.
+#define FLOAT32X4_UNARY_FUNCTION_LIST(V)                                              \
+  V(abs, (UnaryFunc<Float32x4, Abs, Float32x4>), 1)                                   \
+  V(check, (UnaryFunc<Float32x4, Identity, Float32x4>), 1)                            \
+  V(fromFloat64x2Bits, (FuncConvertBits<Float64x2, Float32x4>), 1)                    \
+  V(fromInt8x16Bits,   (FuncConvertBits<Int8x16,   Float32x4>), 1)                    \
+  V(fromInt16x8Bits,   (FuncConvertBits<Int16x8,   Float32x4>), 1)                    \
+  V(fromInt32x4,       (FuncConvert<Int32x4,       Float32x4>), 1)                    \
+  V(fromInt32x4Bits,   (FuncConvertBits<Int32x4,   Float32x4>), 1)                    \
+  V(fromUint8x16Bits,  (FuncConvertBits<Uint8x16,  Float32x4>), 1)                    \
+  V(fromUint16x8Bits,  (FuncConvertBits<Uint16x8,  Float32x4>), 1)                    \
+  V(fromUint32x4,      (FuncConvert<Uint32x4,      Float32x4>), 1)                    \
+  V(fromUint32x4Bits,  (FuncConvertBits<Uint32x4,  Float32x4>), 1)                    \
+  V(neg, (UnaryFunc<Float32x4, Neg, Float32x4>), 1)                                   \
+  V(reciprocalApproximation, (UnaryFunc<Float32x4, RecApprox, Float32x4>), 1)         \
+  V(reciprocalSqrtApproximation, (UnaryFunc<Float32x4, RecSqrtApprox, Float32x4>), 1) \
+  V(splat, (FuncSplat<Float32x4>), 1)                                                 \
+  V(sqrt, (UnaryFunc<Float32x4, Sqrt, Float32x4>), 1)
+
+#define FLOAT32X4_BINARY_FUNCTION_LIST(V)                                             \
+  V(add, (BinaryFunc<Float32x4, Add, Float32x4>), 2)                                  \
+  V(div, (BinaryFunc<Float32x4, Div, Float32x4>), 2)                                  \
+  V(equal, (CompareFunc<Float32x4, Equal, Bool32x4>), 2)                              \
+  V(extractLane, (ExtractLane<Float32x4>), 2)                                         \
+  V(greaterThan, (CompareFunc<Float32x4, GreaterThan, Bool32x4>), 2)                  \
+  V(greaterThanOrEqual, (CompareFunc<Float32x4, GreaterThanOrEqual, Bool32x4>), 2)    \
+  V(lessThan, (CompareFunc<Float32x4, LessThan, Bool32x4>), 2)                        \
+  V(lessThanOrEqual, (CompareFunc<Float32x4, LessThanOrEqual, Bool32x4>), 2)          \
+  V(load,  (Load<Float32x4, 4>), 2)                                                   \
+  V(load3, (Load<Float32x4, 3>), 2)                                                   \
+  V(load2, (Load<Float32x4, 2>), 2)                                                   \
+  V(load1, (Load<Float32x4, 1>), 2)                                                   \
+  V(max, (BinaryFunc<Float32x4, Maximum, Float32x4>), 2)                              \
+  V(maxNum, (BinaryFunc<Float32x4, MaxNum, Float32x4>), 2)                            \
+  V(min, (BinaryFunc<Float32x4, Minimum, Float32x4>), 2)                              \
+  V(minNum, (BinaryFunc<Float32x4, MinNum, Float32x4>), 2)                            \
+  V(mul, (BinaryFunc<Float32x4, Mul, Float32x4>), 2)                                  \
+  V(notEqual, (CompareFunc<Float32x4, NotEqual, Bool32x4>), 2)                        \
+  V(sub, (BinaryFunc<Float32x4, Sub, Float32x4>), 2)
+
+#define FLOAT32X4_TERNARY_FUNCTION_LIST(V)                                            \
+  V(replaceLane, (ReplaceLane<Float32x4>), 3)                                         \
+  V(select, (Select<Float32x4, Bool32x4>), 3)                                         \
+  V(store,  (Store<Float32x4, 4>), 3)                                                 \
+  V(store3, (Store<Float32x4, 3>), 3)                                                 \
+  V(store2, (Store<Float32x4, 2>), 3)                                                 \
+  V(store1, (Store<Float32x4, 1>), 3)
+
+#define FLOAT32X4_SHUFFLE_FUNCTION_LIST(V)                                            \
+  V(swizzle, Swizzle<Float32x4>, 5)                                                   \
+  V(shuffle, Shuffle<Float32x4>, 6)
+
+#define FLOAT32X4_FUNCTION_LIST(V)                                                    \
+  FLOAT32X4_UNARY_FUNCTION_LIST(V)                                                    \
+  FLOAT32X4_BINARY_FUNCTION_LIST(V)                                                   \
+  FLOAT32X4_TERNARY_FUNCTION_LIST(V)                                                  \
+  FLOAT32X4_SHUFFLE_FUNCTION_LIST(V)
+
+// Float64x2.
+#define FLOAT64X2_UNARY_FUNCTION_LIST(V)                                              \
+  V(abs, (UnaryFunc<Float64x2, Abs, Float64x2>), 1)                                   \
+  V(check, (UnaryFunc<Float64x2, Identity, Float64x2>), 1)                            \
+  V(fromFloat32x4Bits, (FuncConvertBits<Float32x4, Float64x2>), 1)                    \
+  V(fromInt8x16Bits,   (FuncConvertBits<Int8x16,   Float64x2>), 1)                    \
+  V(fromInt16x8Bits,   (FuncConvertBits<Int16x8,   Float64x2>), 1)                    \
+  V(fromInt32x4Bits,   (FuncConvertBits<Int32x4,   Float64x2>), 1)                    \
+  V(fromUint8x16Bits,  (FuncConvertBits<Uint8x16,  Float64x2>), 1)                    \
+  V(fromUint16x8Bits,  (FuncConvertBits<Uint16x8,  Float64x2>), 1)                    \
+  V(fromUint32x4Bits,  (FuncConvertBits<Uint32x4,  Float64x2>), 1)                    \
+  V(neg, (UnaryFunc<Float64x2, Neg, Float64x2>), 1)                                   \
+  V(reciprocalApproximation, (UnaryFunc<Float64x2, RecApprox, Float64x2>), 1)         \
+  V(reciprocalSqrtApproximation, (UnaryFunc<Float64x2, RecSqrtApprox, Float64x2>), 1) \
+  V(splat, (FuncSplat<Float64x2>), 1)                                                 \
+  V(sqrt, (UnaryFunc<Float64x2, Sqrt, Float64x2>), 1)
+
+#define FLOAT64X2_BINARY_FUNCTION_LIST(V)                                             \
+  V(add, (BinaryFunc<Float64x2, Add, Float64x2>), 2)                                  \
+  V(div, (BinaryFunc<Float64x2, Div, Float64x2>), 2)                                  \
+  V(equal, (CompareFunc<Float64x2, Equal, Bool64x2>), 2)                              \
+  V(extractLane, (ExtractLane<Float64x2>), 2)                                         \
+  V(greaterThan, (CompareFunc<Float64x2, GreaterThan, Bool64x2>), 2)                  \
+  V(greaterThanOrEqual, (CompareFunc<Float64x2, GreaterThanOrEqual, Bool64x2>), 2)    \
+  V(lessThan, (CompareFunc<Float64x2, LessThan, Bool64x2>), 2)                        \
+  V(lessThanOrEqual, (CompareFunc<Float64x2, LessThanOrEqual, Bool64x2>), 2)          \
+  V(load,  (Load<Float64x2, 2>), 2)                                                   \
+  V(load1, (Load<Float64x2, 1>), 2)                                                   \
+  V(max, (BinaryFunc<Float64x2, Maximum, Float64x2>), 2)                              \
+  V(maxNum, (BinaryFunc<Float64x2, MaxNum, Float64x2>), 2)                            \
+  V(min, (BinaryFunc<Float64x2, Minimum, Float64x2>), 2)                              \
+  V(minNum, (BinaryFunc<Float64x2, MinNum, Float64x2>), 2)                            \
+  V(mul, (BinaryFunc<Float64x2, Mul, Float64x2>), 2)                                  \
+  V(notEqual, (CompareFunc<Float64x2, NotEqual, Bool64x2>), 2)                        \
+  V(sub, (BinaryFunc<Float64x2, Sub, Float64x2>), 2)
+
+#define FLOAT64X2_TERNARY_FUNCTION_LIST(V)                                            \
+  V(replaceLane, (ReplaceLane<Float64x2>), 3)                                         \
+  V(select, (Select<Float64x2, Bool64x2>), 3)                                         \
+  V(store,  (Store<Float64x2, 2>), 3)                                                 \
+  V(store1, (Store<Float64x2, 1>), 3)
+
+#define FLOAT64X2_SHUFFLE_FUNCTION_LIST(V)                                            \
+  V(swizzle, Swizzle<Float64x2>, 3)                                                   \
+  V(shuffle, Shuffle<Float64x2>, 4)
+
+#define FLOAT64X2_FUNCTION_LIST(V)                                                    \
+  FLOAT64X2_UNARY_FUNCTION_LIST(V)                                                    \
+  FLOAT64X2_BINARY_FUNCTION_LIST(V)                                                   \
+  FLOAT64X2_TERNARY_FUNCTION_LIST(V)                                                  \
+  FLOAT64X2_SHUFFLE_FUNCTION_LIST(V)
+
+// Int8x16.
+#define INT8X16_UNARY_FUNCTION_LIST(V)                                                \
+  V(check, (UnaryFunc<Int8x16, Identity, Int8x16>), 1)                                \
+  V(fromFloat32x4Bits, (FuncConvertBits<Float32x4, Int8x16>), 1)                      \
+  V(fromFloat64x2Bits, (FuncConvertBits<Float64x2, Int8x16>), 1)                      \
+  V(fromInt16x8Bits,   (FuncConvertBits<Int16x8,   Int8x16>), 1)                      \
+  V(fromInt32x4Bits,   (FuncConvertBits<Int32x4,   Int8x16>), 1)                      \
+  V(fromUint8x16Bits,  (FuncConvertBits<Uint8x16,  Int8x16>), 1)                      \
+  V(fromUint16x8Bits,  (FuncConvertBits<Uint16x8,  Int8x16>), 1)                      \
+  V(fromUint32x4Bits,  (FuncConvertBits<Uint32x4,  Int8x16>), 1)                      \
+  V(neg, (UnaryFunc<Int8x16, Neg, Int8x16>), 1)                                       \
+  V(not, (UnaryFunc<Int8x16, Not, Int8x16>), 1)                                       \
+  V(splat, (FuncSplat<Int8x16>), 1)
+
+#define INT8X16_BINARY_FUNCTION_LIST(V)                                               \
+  V(add, (BinaryFunc<Int8x16, Add, Int8x16>), 2)                                      \
+  V(addSaturate, (BinaryFunc<Int8x16, AddSaturate, Int8x16>), 2)                      \
+  V(and, (BinaryFunc<Int8x16, And, Int8x16>), 2)                                      \
+  V(equal, (CompareFunc<Int8x16, Equal, Bool8x16>), 2)                                \
+  V(extractLane, (ExtractLane<Int8x16>), 2)                                           \
+  V(greaterThan, (CompareFunc<Int8x16, GreaterThan, Bool8x16>), 2)                    \
+  V(greaterThanOrEqual, (CompareFunc<Int8x16, GreaterThanOrEqual, Bool8x16>), 2)      \
+  V(lessThan, (CompareFunc<Int8x16, LessThan, Bool8x16>), 2)                          \
+  V(lessThanOrEqual, (CompareFunc<Int8x16, LessThanOrEqual, Bool8x16>), 2)            \
+  V(load, (Load<Int8x16, 16>), 2)                                                     \
+  V(mul, (BinaryFunc<Int8x16, Mul, Int8x16>), 2)                                      \
+  V(notEqual, (CompareFunc<Int8x16, NotEqual, Bool8x16>), 2)                          \
+  V(or, (BinaryFunc<Int8x16, Or, Int8x16>), 2)                                        \
+  V(sub, (BinaryFunc<Int8x16, Sub, Int8x16>), 2)                                      \
+  V(subSaturate, (BinaryFunc<Int8x16, SubSaturate, Int8x16>), 2)                      \
+  V(shiftLeftByScalar, (BinaryScalar<Int8x16, ShiftLeft>), 2)                         \
+  V(shiftRightByScalar, (BinaryScalar<Int8x16, ShiftRightArithmetic>), 2)             \
+  V(xor, (BinaryFunc<Int8x16, Xor, Int8x16>), 2)
+
+#define INT8X16_TERNARY_FUNCTION_LIST(V)                                              \
+  V(replaceLane, (ReplaceLane<Int8x16>), 3)                                           \
+  V(select, (Select<Int8x16, Bool8x16>), 3)                                           \
+  V(store, (Store<Int8x16, 16>), 3)
+
+#define INT8X16_SHUFFLE_FUNCTION_LIST(V)                                              \
+  V(swizzle, Swizzle<Int8x16>, 17)                                                    \
+  V(shuffle, Shuffle<Int8x16>, 18)
+
+#define INT8X16_FUNCTION_LIST(V)                                                      \
+  INT8X16_UNARY_FUNCTION_LIST(V)                                                      \
+  INT8X16_BINARY_FUNCTION_LIST(V)                                                     \
+  INT8X16_TERNARY_FUNCTION_LIST(V)                                                    \
+  INT8X16_SHUFFLE_FUNCTION_LIST(V)
+
+// Uint8x16.
+#define UINT8X16_UNARY_FUNCTION_LIST(V)                                               \
+  V(check, (UnaryFunc<Uint8x16, Identity, Uint8x16>), 1)                              \
+  V(fromFloat32x4Bits, (FuncConvertBits<Float32x4, Uint8x16>), 1)                     \
+  V(fromFloat64x2Bits, (FuncConvertBits<Float64x2, Uint8x16>), 1)                     \
+  V(fromInt8x16Bits,   (FuncConvertBits<Int8x16,   Uint8x16>), 1)                     \
+  V(fromInt16x8Bits,   (FuncConvertBits<Int16x8,   Uint8x16>), 1)                     \
+  V(fromInt32x4Bits,   (FuncConvertBits<Int32x4,   Uint8x16>), 1)                     \
+  V(fromUint16x8Bits,  (FuncConvertBits<Uint16x8,  Uint8x16>), 1)                     \
+  V(fromUint32x4Bits,  (FuncConvertBits<Uint32x4,  Uint8x16>), 1)                     \
+  V(neg, (UnaryFunc<Uint8x16, Neg, Uint8x16>), 1)                                     \
+  V(not, (UnaryFunc<Uint8x16, Not, Uint8x16>), 1)                                     \
+  V(splat, (FuncSplat<Uint8x16>), 1)
+
+#define UINT8X16_BINARY_FUNCTION_LIST(V)                                              \
+  V(add, (BinaryFunc<Uint8x16, Add, Uint8x16>), 2)                                    \
+  V(addSaturate, (BinaryFunc<Uint8x16, AddSaturate, Uint8x16>), 2)                    \
+  V(and, (BinaryFunc<Uint8x16, And, Uint8x16>), 2)                                    \
+  V(equal, (CompareFunc<Uint8x16, Equal, Bool8x16>), 2)                               \
+  V(extractLane, (ExtractLane<Uint8x16>), 2)                                          \
+  V(greaterThan, (CompareFunc<Uint8x16, GreaterThan, Bool8x16>), 2)                   \
+  V(greaterThanOrEqual, (CompareFunc<Uint8x16, GreaterThanOrEqual, Bool8x16>), 2)     \
+  V(lessThan, (CompareFunc<Uint8x16, LessThan, Bool8x16>), 2)                         \
+  V(lessThanOrEqual, (CompareFunc<Uint8x16, LessThanOrEqual, Bool8x16>), 2)           \
+  V(load, (Load<Uint8x16, 16>), 2)                                                    \
+  V(mul, (BinaryFunc<Uint8x16, Mul, Uint8x16>), 2)                                    \
+  V(notEqual, (CompareFunc<Uint8x16, NotEqual, Bool8x16>), 2)                         \
+  V(or, (BinaryFunc<Uint8x16, Or, Uint8x16>), 2)                                      \
+  V(sub, (BinaryFunc<Uint8x16, Sub, Uint8x16>), 2)                                    \
+  V(subSaturate, (BinaryFunc<Uint8x16, SubSaturate, Uint8x16>), 2)                    \
+  V(shiftLeftByScalar, (BinaryScalar<Uint8x16, ShiftLeft>), 2)                        \
+  V(shiftRightByScalar, (BinaryScalar<Uint8x16, ShiftRightLogical>), 2)               \
+  V(xor, (BinaryFunc<Uint8x16, Xor, Uint8x16>), 2)
+
+#define UINT8X16_TERNARY_FUNCTION_LIST(V)                                             \
+  V(replaceLane, (ReplaceLane<Uint8x16>), 3)                                          \
+  V(select, (Select<Uint8x16, Bool8x16>), 3)                                          \
+  V(store, (Store<Uint8x16, 16>), 3)
+
+#define UINT8X16_SHUFFLE_FUNCTION_LIST(V)                                             \
+  V(swizzle, Swizzle<Uint8x16>, 17)                                                   \
+  V(shuffle, Shuffle<Uint8x16>, 18)
+
+#define UINT8X16_FUNCTION_LIST(V)                                                     \
+  UINT8X16_UNARY_FUNCTION_LIST(V)                                                     \
+  UINT8X16_BINARY_FUNCTION_LIST(V)                                                    \
+  UINT8X16_TERNARY_FUNCTION_LIST(V)                                                   \
+  UINT8X16_SHUFFLE_FUNCTION_LIST(V)
+
+// Int16x8.
+#define INT16X8_UNARY_FUNCTION_LIST(V)                                                \
+  V(check, (UnaryFunc<Int16x8, Identity, Int16x8>), 1)                                \
+  V(fromFloat32x4Bits, (FuncConvertBits<Float32x4, Int16x8>), 1)                      \
+  V(fromFloat64x2Bits, (FuncConvertBits<Float64x2, Int16x8>), 1)                      \
+  V(fromInt8x16Bits,   (FuncConvertBits<Int8x16,   Int16x8>), 1)                      \
+  V(fromInt32x4Bits,   (FuncConvertBits<Int32x4,   Int16x8>), 1)                      \
+  V(fromUint8x16Bits,  (FuncConvertBits<Uint8x16,  Int16x8>), 1)                      \
+  V(fromUint16x8Bits,  (FuncConvertBits<Uint16x8,  Int16x8>), 1)                      \
+  V(fromUint32x4Bits,  (FuncConvertBits<Uint32x4,  Int16x8>), 1)                      \
+  V(neg, (UnaryFunc<Int16x8, Neg, Int16x8>), 1)                                       \
+  V(not, (UnaryFunc<Int16x8, Not, Int16x8>), 1)                                       \
+  V(splat, (FuncSplat<Int16x8>), 1)
+
+#define INT16X8_BINARY_FUNCTION_LIST(V)                                               \
+  V(add, (BinaryFunc<Int16x8, Add, Int16x8>), 2)                                      \
+  V(addSaturate, (BinaryFunc<Int16x8, AddSaturate, Int16x8>), 2)                      \
+  V(and, (BinaryFunc<Int16x8, And, Int16x8>), 2)                                      \
+  V(equal, (CompareFunc<Int16x8, Equal, Bool16x8>), 2)                                \
+  V(extractLane, (ExtractLane<Int16x8>), 2)                                           \
+  V(greaterThan, (CompareFunc<Int16x8, GreaterThan, Bool16x8>), 2)                    \
+  V(greaterThanOrEqual, (CompareFunc<Int16x8, GreaterThanOrEqual, Bool16x8>), 2)      \
+  V(lessThan, (CompareFunc<Int16x8, LessThan, Bool16x8>), 2)                          \
+  V(lessThanOrEqual, (CompareFunc<Int16x8, LessThanOrEqual, Bool16x8>), 2)            \
+  V(load, (Load<Int16x8, 8>), 2)                                                      \
+  V(mul, (BinaryFunc<Int16x8, Mul, Int16x8>), 2)                                      \
+  V(notEqual, (CompareFunc<Int16x8, NotEqual, Bool16x8>), 2)                          \
+  V(or, (BinaryFunc<Int16x8, Or, Int16x8>), 2)                                        \
+  V(sub, (BinaryFunc<Int16x8, Sub, Int16x8>), 2)                                      \
+  V(subSaturate, (BinaryFunc<Int16x8, SubSaturate, Int16x8>), 2)                      \
+  V(shiftLeftByScalar, (BinaryScalar<Int16x8, ShiftLeft>), 2)                         \
+  V(shiftRightByScalar, (BinaryScalar<Int16x8, ShiftRightArithmetic>), 2)             \
+  V(xor, (BinaryFunc<Int16x8, Xor, Int16x8>), 2)
+
+#define INT16X8_TERNARY_FUNCTION_LIST(V)                                              \
+  V(replaceLane, (ReplaceLane<Int16x8>), 3)                                           \
+  V(select, (Select<Int16x8, Bool16x8>), 3)                                           \
+  V(store, (Store<Int16x8, 8>), 3)
+
+#define INT16X8_SHUFFLE_FUNCTION_LIST(V)                                              \
+  V(swizzle, Swizzle<Int16x8>, 9)                                                     \
+  V(shuffle, Shuffle<Int16x8>, 10)
+
+#define INT16X8_FUNCTION_LIST(V)                                                      \
+  INT16X8_UNARY_FUNCTION_LIST(V)                                                      \
+  INT16X8_BINARY_FUNCTION_LIST(V)                                                     \
+  INT16X8_TERNARY_FUNCTION_LIST(V)                                                    \
+  INT16X8_SHUFFLE_FUNCTION_LIST(V)
+
+// Uint16x8.
+#define UINT16X8_UNARY_FUNCTION_LIST(V)                                               \
+  V(check, (UnaryFunc<Uint16x8, Identity, Uint16x8>), 1)                              \
+  V(fromFloat32x4Bits, (FuncConvertBits<Float32x4, Uint16x8>), 1)                     \
+  V(fromFloat64x2Bits, (FuncConvertBits<Float64x2, Uint16x8>), 1)                     \
+  V(fromInt8x16Bits,   (FuncConvertBits<Int8x16,   Uint16x8>), 1)                     \
+  V(fromInt16x8Bits,   (FuncConvertBits<Int16x8,   Uint16x8>), 1)                     \
+  V(fromInt32x4Bits,   (FuncConvertBits<Int32x4,   Uint16x8>), 1)                     \
+  V(fromUint8x16Bits,  (FuncConvertBits<Uint8x16,  Uint16x8>), 1)                     \
+  V(fromUint32x4Bits,  (FuncConvertBits<Uint32x4,  Uint16x8>), 1)                     \
+  V(neg, (UnaryFunc<Uint16x8, Neg, Uint16x8>), 1)                                     \
+  V(not, (UnaryFunc<Uint16x8, Not, Uint16x8>), 1)                                     \
+  V(splat, (FuncSplat<Uint16x8>), 1)
+
+#define UINT16X8_BINARY_FUNCTION_LIST(V)                                              \
+  V(add, (BinaryFunc<Uint16x8, Add, Uint16x8>), 2)                                    \
+  V(addSaturate, (BinaryFunc<Uint16x8, AddSaturate, Uint16x8>), 2)                    \
+  V(and, (BinaryFunc<Uint16x8, And, Uint16x8>), 2)                                    \
+  V(equal, (CompareFunc<Uint16x8, Equal, Bool16x8>), 2)                               \
+  V(extractLane, (ExtractLane<Uint16x8>), 2)                                          \
+  V(greaterThan, (CompareFunc<Uint16x8, GreaterThan, Bool16x8>), 2)                   \
+  V(greaterThanOrEqual, (CompareFunc<Uint16x8, GreaterThanOrEqual, Bool16x8>), 2)     \
+  V(lessThan, (CompareFunc<Uint16x8, LessThan, Bool16x8>), 2)                         \
+  V(lessThanOrEqual, (CompareFunc<Uint16x8, LessThanOrEqual, Bool16x8>), 2)           \
+  V(load, (Load<Uint16x8, 8>), 2)                                                     \
+  V(mul, (BinaryFunc<Uint16x8, Mul, Uint16x8>), 2)                                    \
+  V(notEqual, (CompareFunc<Uint16x8, NotEqual, Bool16x8>), 2)                         \
+  V(or, (BinaryFunc<Uint16x8, Or, Uint16x8>), 2)                                      \
+  V(sub, (BinaryFunc<Uint16x8, Sub, Uint16x8>), 2)                                    \
+  V(subSaturate, (BinaryFunc<Uint16x8, SubSaturate, Uint16x8>), 2)                    \
+  V(shiftLeftByScalar, (BinaryScalar<Uint16x8, ShiftLeft>), 2)                        \
+  V(shiftRightByScalar, (BinaryScalar<Uint16x8, ShiftRightLogical>), 2)               \
+  V(xor, (BinaryFunc<Uint16x8, Xor, Uint16x8>), 2)
+
+#define UINT16X8_TERNARY_FUNCTION_LIST(V)                                             \
+  V(replaceLane, (ReplaceLane<Uint16x8>), 3)                                          \
+  V(select, (Select<Uint16x8, Bool16x8>), 3)                                          \
+  V(store, (Store<Uint16x8, 8>), 3)
+
+#define UINT16X8_SHUFFLE_FUNCTION_LIST(V)                                             \
+  V(swizzle, Swizzle<Uint16x8>, 9)                                                    \
+  V(shuffle, Shuffle<Uint16x8>, 10)
+
+#define UINT16X8_FUNCTION_LIST(V)                                                     \
+  UINT16X8_UNARY_FUNCTION_LIST(V)                                                     \
+  UINT16X8_BINARY_FUNCTION_LIST(V)                                                    \
+  UINT16X8_TERNARY_FUNCTION_LIST(V)                                                   \
+  UINT16X8_SHUFFLE_FUNCTION_LIST(V)
+
+// Int32x4.
+#define INT32X4_UNARY_FUNCTION_LIST(V)                                                \
+  V(check, (UnaryFunc<Int32x4, Identity, Int32x4>), 1)                                \
+  V(fromFloat32x4,     (FuncConvert<Float32x4,     Int32x4>), 1)                      \
+  V(fromFloat32x4Bits, (FuncConvertBits<Float32x4, Int32x4>), 1)                      \
+  V(fromFloat64x2Bits, (FuncConvertBits<Float64x2, Int32x4>), 1)                      \
+  V(fromInt8x16Bits,   (FuncConvertBits<Int8x16,   Int32x4>), 1)                      \
+  V(fromInt16x8Bits,   (FuncConvertBits<Int16x8,   Int32x4>), 1)                      \
+  V(fromUint8x16Bits,  (FuncConvertBits<Uint8x16,  Int32x4>), 1)                      \
+  V(fromUint16x8Bits,  (FuncConvertBits<Uint16x8,  Int32x4>), 1)                      \
+  V(fromUint32x4Bits,  (FuncConvertBits<Uint32x4,  Int32x4>), 1)                      \
+  V(neg, (UnaryFunc<Int32x4, Neg, Int32x4>), 1)                                       \
+  V(not, (UnaryFunc<Int32x4, Not, Int32x4>), 1)                                       \
+  V(splat, (FuncSplat<Int32x4>), 0)
+
+#define INT32X4_BINARY_FUNCTION_LIST(V)                                               \
+  V(add, (BinaryFunc<Int32x4, Add, Int32x4>), 2)                                      \
+  V(and, (BinaryFunc<Int32x4, And, Int32x4>), 2)                                      \
+  V(equal, (CompareFunc<Int32x4, Equal, Bool32x4>), 2)                                \
+  V(extractLane, (ExtractLane<Int32x4>), 2)                                           \
+  V(greaterThan, (CompareFunc<Int32x4, GreaterThan, Bool32x4>), 2)                    \
+  V(greaterThanOrEqual, (CompareFunc<Int32x4, GreaterThanOrEqual, Bool32x4>), 2)      \
+  V(lessThan, (CompareFunc<Int32x4, LessThan, Bool32x4>), 2)                          \
+  V(lessThanOrEqual, (CompareFunc<Int32x4, LessThanOrEqual, Bool32x4>), 2)            \
+  V(load,  (Load<Int32x4, 4>), 2)                                                     \
+  V(load3, (Load<Int32x4, 3>), 2)                                                     \
+  V(load2, (Load<Int32x4, 2>), 2)                                                     \
+  V(load1, (Load<Int32x4, 1>), 2)                                                     \
+  V(mul, (BinaryFunc<Int32x4, Mul, Int32x4>), 2)                                      \
+  V(notEqual, (CompareFunc<Int32x4, NotEqual, Bool32x4>), 2)                          \
+  V(or, (BinaryFunc<Int32x4, Or, Int32x4>), 2)                                        \
+  V(sub, (BinaryFunc<Int32x4, Sub, Int32x4>), 2)                                      \
+  V(shiftLeftByScalar, (BinaryScalar<Int32x4, ShiftLeft>), 2)                         \
+  V(shiftRightByScalar, (BinaryScalar<Int32x4, ShiftRightArithmetic>), 2)             \
+  V(xor, (BinaryFunc<Int32x4, Xor, Int32x4>), 2)
+
+#define INT32X4_TERNARY_FUNCTION_LIST(V)                                              \
+  V(replaceLane, (ReplaceLane<Int32x4>), 3)                                           \
+  V(select, (Select<Int32x4, Bool32x4>), 3)                                           \
+  V(store,  (Store<Int32x4, 4>), 3)                                                   \
+  V(store3, (Store<Int32x4, 3>), 3)                                                   \
+  V(store2, (Store<Int32x4, 2>), 3)                                                   \
+  V(store1, (Store<Int32x4, 1>), 3)
+
+#define INT32X4_SHUFFLE_FUNCTION_LIST(V)                                              \
+  V(swizzle, Swizzle<Int32x4>, 5)                                                     \
+  V(shuffle, Shuffle<Int32x4>, 6)
+
+#define INT32X4_FUNCTION_LIST(V)                                                      \
+  INT32X4_UNARY_FUNCTION_LIST(V)                                                      \
+  INT32X4_BINARY_FUNCTION_LIST(V)                                                     \
+  INT32X4_TERNARY_FUNCTION_LIST(V)                                                    \
+  INT32X4_SHUFFLE_FUNCTION_LIST(V)
+
+// Uint32x4.
+#define UINT32X4_UNARY_FUNCTION_LIST(V)                                               \
+  V(check, (UnaryFunc<Uint32x4, Identity, Uint32x4>), 1)                              \
+  V(fromFloat32x4,     (FuncConvert<Float32x4,     Uint32x4>), 1)                     \
+  V(fromFloat32x4Bits, (FuncConvertBits<Float32x4, Uint32x4>), 1)                     \
+  V(fromFloat64x2Bits, (FuncConvertBits<Float64x2, Uint32x4>), 1)                     \
+  V(fromInt8x16Bits,   (FuncConvertBits<Int8x16,   Uint32x4>), 1)                     \
+  V(fromInt16x8Bits,   (FuncConvertBits<Int16x8,   Uint32x4>), 1)                     \
+  V(fromInt32x4Bits,   (FuncConvertBits<Int32x4,   Uint32x4>), 1)                     \
+  V(fromUint8x16Bits,  (FuncConvertBits<Uint8x16,  Uint32x4>), 1)                     \
+  V(fromUint16x8Bits,  (FuncConvertBits<Uint16x8,  Uint32x4>), 1)                     \
+  V(neg, (UnaryFunc<Uint32x4, Neg, Uint32x4>), 1)                                     \
+  V(not, (UnaryFunc<Uint32x4, Not, Uint32x4>), 1)                                     \
+  V(splat, (FuncSplat<Uint32x4>), 0)
+
+#define UINT32X4_BINARY_FUNCTION_LIST(V)                                              \
+  V(add, (BinaryFunc<Uint32x4, Add, Uint32x4>), 2)                                    \
+  V(and, (BinaryFunc<Uint32x4, And, Uint32x4>), 2)                                    \
+  V(equal, (CompareFunc<Uint32x4, Equal, Bool32x4>), 2)                               \
+  V(extractLane, (ExtractLane<Uint32x4>), 2)                                          \
+  V(greaterThan, (CompareFunc<Uint32x4, GreaterThan, Bool32x4>), 2)                   \
+  V(greaterThanOrEqual, (CompareFunc<Uint32x4, GreaterThanOrEqual, Bool32x4>), 2)     \
+  V(lessThan, (CompareFunc<Uint32x4, LessThan, Bool32x4>), 2)                         \
+  V(lessThanOrEqual, (CompareFunc<Uint32x4, LessThanOrEqual, Bool32x4>), 2)           \
+  V(load,  (Load<Uint32x4, 4>), 2)                                                    \
+  V(load3, (Load<Uint32x4, 3>), 2)                                                    \
+  V(load2, (Load<Uint32x4, 2>), 2)                                                    \
+  V(load1, (Load<Uint32x4, 1>), 2)                                                    \
+  V(mul, (BinaryFunc<Uint32x4, Mul, Uint32x4>), 2)                                    \
+  V(notEqual, (CompareFunc<Uint32x4, NotEqual, Bool32x4>), 2)                         \
+  V(or, (BinaryFunc<Uint32x4, Or, Uint32x4>), 2)                                      \
+  V(sub, (BinaryFunc<Uint32x4, Sub, Uint32x4>), 2)                                    \
+  V(shiftLeftByScalar, (BinaryScalar<Uint32x4, ShiftLeft>), 2)                        \
+  V(shiftRightByScalar, (BinaryScalar<Uint32x4, ShiftRightLogical>), 2)               \
+  V(xor, (BinaryFunc<Uint32x4, Xor, Uint32x4>), 2)
+
+#define UINT32X4_TERNARY_FUNCTION_LIST(V)                                             \
+  V(replaceLane, (ReplaceLane<Uint32x4>), 3)                                          \
+  V(select, (Select<Uint32x4, Bool32x4>), 3)                                          \
+  V(store,  (Store<Uint32x4, 4>), 3)                                                  \
+  V(store3, (Store<Uint32x4, 3>), 3)                                                  \
+  V(store2, (Store<Uint32x4, 2>), 3)                                                  \
+  V(store1, (Store<Uint32x4, 1>), 3)
+
+#define UINT32X4_SHUFFLE_FUNCTION_LIST(V)                                             \
+  V(swizzle, Swizzle<Uint32x4>, 5)                                                    \
+  V(shuffle, Shuffle<Uint32x4>, 6)
+
+#define UINT32X4_FUNCTION_LIST(V)                                                     \
+  UINT32X4_UNARY_FUNCTION_LIST(V)                                                     \
+  UINT32X4_BINARY_FUNCTION_LIST(V)                                                    \
+  UINT32X4_TERNARY_FUNCTION_LIST(V)                                                   \
+  UINT32X4_SHUFFLE_FUNCTION_LIST(V)
+
+/*
+ * The FOREACH macros below partition all of the SIMD operations into disjoint
+ * sets.
+ */
+
+// Operations available on all SIMD types. Mixed arity.
+#define FOREACH_COMMON_SIMD_OP(_)     \
+    _(extractLane)                    \
+    _(replaceLane)                    \
+    _(check)                          \
+    _(splat)
+
+// Lanewise operations available on numeric SIMD types.
+// Include lane-wise select here since it is not arithmetic and defined on
+// numeric types too.
+#define FOREACH_LANE_SIMD_OP(_)       \
+    _(select)                         \
+    _(swizzle)                        \
+    _(shuffle)
+
+// Memory operations available on numeric SIMD types.
+#define FOREACH_MEMORY_SIMD_OP(_)     \
+    _(load)                           \
+    _(store)
+
+// Memory operations available on numeric X4 SIMD types.
+#define FOREACH_MEMORY_X4_SIMD_OP(_)  \
+    _(load1)                          \
+    _(load2)                          \
+    _(load3)                          \
+    _(store1)                         \
+    _(store2)                         \
+    _(store3)
+
+// Unary operations on Bool vectors.
+#define FOREACH_BOOL_SIMD_UNOP(_)     \
+    _(allTrue)                        \
+    _(anyTrue)
+
+// Unary bitwise SIMD operators defined on all integer and boolean SIMD types.
+#define FOREACH_BITWISE_SIMD_UNOP(_)  \
+    _(not)
+
+// Binary bitwise SIMD operators defined on all integer and boolean SIMD types.
+#define FOREACH_BITWISE_SIMD_BINOP(_) \
+    _(and)                            \
+    _(or)                             \
+    _(xor)
+
+// Bitwise shifts defined on integer SIMD types.
+#define FOREACH_SHIFT_SIMD_OP(_)      \
+    _(shiftLeftByScalar)              \
+    _(shiftRightByScalar)
+
+// Unary arithmetic operators defined on numeric SIMD types.
+#define FOREACH_NUMERIC_SIMD_UNOP(_)  \
+    _(neg)
+
+// Binary arithmetic operators defined on numeric SIMD types.
+#define FOREACH_NUMERIC_SIMD_BINOP(_) \
+    _(add)                            \
+    _(sub)                            \
+    _(mul)
+
+// Unary arithmetic operators defined on floating point SIMD types.
+#define FOREACH_FLOAT_SIMD_UNOP(_)    \
+    _(abs)                            \
+    _(sqrt)                           \
+    _(reciprocalApproximation)        \
+    _(reciprocalSqrtApproximation)
+
+// Binary arithmetic operators defined on floating point SIMD types.
+#define FOREACH_FLOAT_SIMD_BINOP(_)   \
+    _(div)                            \
+    _(max)                            \
+    _(min)                            \
+    _(maxNum)                         \
+    _(minNum)
+
+// Binary operations on small integer (< 32 bits) vectors.
+#define FOREACH_SMINT_SIMD_BINOP(_)   \
+    _(addSaturate)                    \
+    _(subSaturate)
+
+// Comparison operators defined on numeric SIMD types.
+#define FOREACH_COMP_SIMD_OP(_)       \
+    _(lessThan)                       \
+    _(lessThanOrEqual)                \
+    _(equal)                          \
+    _(notEqual)                       \
+    _(greaterThan)                    \
+    _(greaterThanOrEqual)
+
+/*
+ * All SIMD operations, excluding casts.
+ */
+#define FORALL_SIMD_NONCAST_OP(_)     \
+    FOREACH_COMMON_SIMD_OP(_)         \
+    FOREACH_LANE_SIMD_OP(_)           \
+    FOREACH_MEMORY_SIMD_OP(_)         \
+    FOREACH_MEMORY_X4_SIMD_OP(_)      \
+    FOREACH_BOOL_SIMD_UNOP(_)         \
+    FOREACH_BITWISE_SIMD_UNOP(_)      \
+    FOREACH_BITWISE_SIMD_BINOP(_)     \
+    FOREACH_SHIFT_SIMD_OP(_)          \
+    FOREACH_NUMERIC_SIMD_UNOP(_)      \
+    FOREACH_NUMERIC_SIMD_BINOP(_)     \
+    FOREACH_FLOAT_SIMD_UNOP(_)        \
+    FOREACH_FLOAT_SIMD_BINOP(_)       \
+    FOREACH_SMINT_SIMD_BINOP(_)       \
+    FOREACH_COMP_SIMD_OP(_)
+
+/*
+ * All operations on integer SIMD types, excluding casts and
+ * FOREACH_MEMORY_X4_OP.
+ */
+#define FORALL_INT_SIMD_OP(_)         \
+    FOREACH_COMMON_SIMD_OP(_)         \
+    FOREACH_LANE_SIMD_OP(_)           \
+    FOREACH_MEMORY_SIMD_OP(_)         \
+    FOREACH_BITWISE_SIMD_UNOP(_)      \
+    FOREACH_BITWISE_SIMD_BINOP(_)     \
+    FOREACH_SHIFT_SIMD_OP(_)          \
+    FOREACH_NUMERIC_SIMD_UNOP(_)      \
+    FOREACH_NUMERIC_SIMD_BINOP(_)     \
+    FOREACH_COMP_SIMD_OP(_)
+
+/*
+ * All operations on floating point SIMD types, excluding casts and
+ * FOREACH_MEMORY_X4_OP.
+ */
+#define FORALL_FLOAT_SIMD_OP(_)       \
+    FOREACH_COMMON_SIMD_OP(_)         \
+    FOREACH_LANE_SIMD_OP(_)           \
+    FOREACH_MEMORY_SIMD_OP(_)         \
+    FOREACH_NUMERIC_SIMD_UNOP(_)      \
+    FOREACH_NUMERIC_SIMD_BINOP(_)     \
+    FOREACH_FLOAT_SIMD_UNOP(_)        \
+    FOREACH_FLOAT_SIMD_BINOP(_)       \
+    FOREACH_COMP_SIMD_OP(_)
+
+/*
+ * All operations on Bool SIMD types.
+ *
+ * These types don't have casts, so no need to specialize.
+ */
+#define FORALL_BOOL_SIMD_OP(_)        \
+    FOREACH_COMMON_SIMD_OP(_)         \
+    FOREACH_BOOL_SIMD_UNOP(_)         \
+    FOREACH_BITWISE_SIMD_UNOP(_)      \
+    FOREACH_BITWISE_SIMD_BINOP(_)
+
+/*
+ * The sets of cast operations are listed per type below.
+ *
+ * These sets are not disjoint.
+ */
+
+#define FOREACH_INT8X16_SIMD_CAST(_)  \
+    _(fromFloat32x4Bits)              \
+    _(fromFloat64x2Bits)              \
+    _(fromInt16x8Bits)                \
+    _(fromInt32x4Bits)
+
+#define FOREACH_INT16X8_SIMD_CAST(_)  \
+    _(fromFloat32x4Bits)              \
+    _(fromFloat64x2Bits)              \
+    _(fromInt8x16Bits)                \
+    _(fromInt32x4Bits)
+
+#define FOREACH_INT32X4_SIMD_CAST(_)  \
+    _(fromFloat32x4)                  \
+    _(fromFloat32x4Bits)              \
+    _(fromFloat64x2Bits)              \
+    _(fromInt8x16Bits)                \
+    _(fromInt16x8Bits)
+
+#define FOREACH_FLOAT32X4_SIMD_CAST(_)\
+    _(fromFloat64x2Bits)              \
+    _(fromInt8x16Bits)                \
+    _(fromInt16x8Bits)                \
+    _(fromInt32x4)                    \
+    _(fromInt32x4Bits)
+
+#define FOREACH_FLOAT64X2_SIMD_CAST(_)\
+    _(fromFloat32x4Bits)              \
+    _(fromInt8x16Bits)                \
+    _(fromInt16x8Bits)                \
+    _(fromInt32x4Bits)
+
+// All operations on Int32x4.
+#define FORALL_INT32X4_SIMD_OP(_)     \
+    FORALL_INT_SIMD_OP(_)             \
+    FOREACH_MEMORY_X4_SIMD_OP(_)      \
+    FOREACH_INT32X4_SIMD_CAST(_)
+
+// All operations on Float32X4
+#define FORALL_FLOAT32X4_SIMD_OP(_)   \
+    FORALL_FLOAT_SIMD_OP(_)           \
+    FOREACH_MEMORY_X4_SIMD_OP(_)      \
+    FOREACH_FLOAT32X4_SIMD_CAST(_)
+
+/*
+ * All SIMD operations assuming only 32x4 types exist.
+ * This is used in the current asm.js impl.
+ */
+#define FORALL_SIMD_ASMJS_OP(_)       \
+    FORALL_SIMD_NONCAST_OP(_)         \
+    _(fromFloat32x4)                  \
+    _(fromFloat32x4Bits)              \
+    _(fromInt8x16Bits)                \
+    _(fromInt16x8Bits)                \
+    _(fromInt32x4)                    \
+    _(fromInt32x4Bits)                \
+    _(fromUint8x16Bits)               \
+    _(fromUint16x8Bits)               \
+    _(fromUint32x4)                   \
+    _(fromUint32x4Bits)
+
+// All operations on Int8x16 or Uint8x16 in the asm.js world.
+// Note: this does not include conversions and casts to/from Uint8x16 because
+// this list is shared between Int8x16 and Uint8x16.
+#define FORALL_INT8X16_ASMJS_OP(_)    \
+    FORALL_INT_SIMD_OP(_)             \
+    FOREACH_SMINT_SIMD_BINOP(_)       \
+    _(fromInt16x8Bits)                \
+    _(fromInt32x4Bits)                \
+    _(fromFloat32x4Bits)
+
+// All operations on Int16x8 or Uint16x8 in the asm.js world.
+// Note: this does not include conversions and casts to/from Uint16x8 because
+// this list is shared between Int16x8 and Uint16x8.
+#define FORALL_INT16X8_ASMJS_OP(_)    \
+    FORALL_INT_SIMD_OP(_)             \
+    FOREACH_SMINT_SIMD_BINOP(_)       \
+    _(fromInt8x16Bits)                \
+    _(fromInt32x4Bits)                \
+    _(fromFloat32x4Bits)
+
+// All operations on Int32x4 or Uint32x4 in the asm.js world.
+// Note: this does not include conversions and casts to/from Uint32x4 because
+// this list is shared between Int32x4 and Uint32x4.
+#define FORALL_INT32X4_ASMJS_OP(_)    \
+    FORALL_INT_SIMD_OP(_)             \
+    FOREACH_MEMORY_X4_SIMD_OP(_)      \
+    _(fromInt8x16Bits)                \
+    _(fromInt16x8Bits)                \
+    _(fromFloat32x4)                  \
+    _(fromFloat32x4Bits)
+
+// All operations on Float32X4 in the asm.js world.
+#define FORALL_FLOAT32X4_ASMJS_OP(_)  \
+    FORALL_FLOAT_SIMD_OP(_)           \
+    FOREACH_MEMORY_X4_SIMD_OP(_)      \
+    _(fromInt8x16Bits)                \
+    _(fromInt16x8Bits)                \
+    _(fromInt32x4Bits)                \
+    _(fromInt32x4)                    \
+    _(fromUint32x4)
+
+namespace js {
+
+// Complete set of SIMD types.
+// It must be kept in sync with the enumeration of values in
+// TypedObjectConstants.h; in particular we need to ensure that Count is
+// appropriately set with respect to the number of actual types.
+enum class SimdType {
+    Int8x16   = JS_SIMDTYPEREPR_INT8X16,
+    Int16x8   = JS_SIMDTYPEREPR_INT16X8,
+    Int32x4   = JS_SIMDTYPEREPR_INT32X4,
+    Uint8x16  = JS_SIMDTYPEREPR_UINT8X16,
+    Uint16x8  = JS_SIMDTYPEREPR_UINT16X8,
+    Uint32x4  = JS_SIMDTYPEREPR_UINT32X4,
+    Float32x4 = JS_SIMDTYPEREPR_FLOAT32X4,
+    Float64x2 = JS_SIMDTYPEREPR_FLOAT64X2,
+    Bool8x16  = JS_SIMDTYPEREPR_BOOL8X16,
+    Bool16x8  = JS_SIMDTYPEREPR_BOOL16X8,
+    Bool32x4  = JS_SIMDTYPEREPR_BOOL32X4,
+    Bool64x2  = JS_SIMDTYPEREPR_BOOL64X2,
+    Count
+};
+
+// The integer SIMD types have a lot of operations that do the exact same thing
+// for signed and unsigned integer types. Sometimes it is simpler to treat
+// signed and unsigned integer SIMD types as the same type, using a SimdSign to
+// distinguish the few cases where there is a difference.
+enum class SimdSign {
+    // Signedness is not applicable to this type. (i.e., Float or Bool).
+    NotApplicable,
+    // Treat as an unsigned integer with a range 0 .. 2^N-1.
+    Unsigned,
+    // Treat as a signed integer in two's complement encoding.
+    Signed,
+};
+
+// Get the signedness of a SIMD type.
+inline SimdSign
+GetSimdSign(SimdType t)
+{
+    switch(t) {
+      case SimdType::Int8x16:
+      case SimdType::Int16x8:
+      case SimdType::Int32x4:
+        return SimdSign::Signed;
+
+      case SimdType::Uint8x16:
+      case SimdType::Uint16x8:
+      case SimdType::Uint32x4:
+        return SimdSign::Unsigned;
+
+      default:
+        return SimdSign::NotApplicable;
+    }
+}
+
+inline bool
+IsSignedIntSimdType(SimdType type)
+{
+    return GetSimdSign(type) == SimdSign::Signed;
+}
+
+// Get the boolean SIMD type with the same shape as t.
+//
+// This is the result type of a comparison operation, and it can also be used to
+// identify the geometry of a SIMD type.
+inline SimdType
+GetBooleanSimdType(SimdType t)
+{
+    switch(t) {
+      case SimdType::Int8x16:
+      case SimdType::Uint8x16:
+      case SimdType::Bool8x16:
+        return SimdType::Bool8x16;
+
+      case SimdType::Int16x8:
+      case SimdType::Uint16x8:
+      case SimdType::Bool16x8:
+        return SimdType::Bool16x8;
+
+      case SimdType::Int32x4:
+      case SimdType::Uint32x4:
+      case SimdType::Float32x4:
+      case SimdType::Bool32x4:
+        return SimdType::Bool32x4;
+
+      case SimdType::Float64x2:
+      case SimdType::Bool64x2:
+        return SimdType::Bool64x2;
+
+      case SimdType::Count:
+        break;
+    }
+    MOZ_MAKE_COMPILER_ASSUME_IS_UNREACHABLE("Bad SIMD type");
+}
+
+// Get the number of lanes in a SIMD type.
+inline unsigned
+GetSimdLanes(SimdType t)
+{
+    switch(t) {
+      case SimdType::Int8x16:
+      case SimdType::Uint8x16:
+      case SimdType::Bool8x16:
+        return 16;
+
+      case SimdType::Int16x8:
+      case SimdType::Uint16x8:
+      case SimdType::Bool16x8:
+        return 8;
+
+      case SimdType::Int32x4:
+      case SimdType::Uint32x4:
+      case SimdType::Float32x4:
+      case SimdType::Bool32x4:
+        return 4;
+
+      case SimdType::Float64x2:
+      case SimdType::Bool64x2:
+        return 2;
+
+      case SimdType::Count:
+        break;
+    }
+    MOZ_MAKE_COMPILER_ASSUME_IS_UNREACHABLE("Bad SIMD type");
+}
+
+// Complete set of SIMD operations.
+//
+// No SIMD types implement all of these operations.
+//
+// C++ defines keywords and/or/xor/not, so prepend Fn_ to all named functions to
+// avoid clashes.
+//
+// Note: because of a gcc < v4.8's compiler bug, uint8_t can't be used as the
+// storage class here. See bug 1243810. See also
+// https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64037 .
+enum class SimdOperation {
+    // The constructor call. No Fn_ prefix here.
+    Constructor,
+
+    // All the operations, except for casts.
+#define DEFOP(x) Fn_##x,
+    FORALL_SIMD_NONCAST_OP(DEFOP)
+#undef DEFOP
+
+    // Int <-> Float conversions.
+    Fn_fromInt32x4,
+    Fn_fromUint32x4,
+    Fn_fromFloat32x4,
+
+    // Bitcasts. One for each type with a memory representation.
+    Fn_fromInt8x16Bits,
+    Fn_fromInt16x8Bits,
+    Fn_fromInt32x4Bits,
+    Fn_fromUint8x16Bits,
+    Fn_fromUint16x8Bits,
+    Fn_fromUint32x4Bits,
+    Fn_fromFloat32x4Bits,
+    Fn_fromFloat64x2Bits,
+
+    Last = Fn_fromFloat64x2Bits
+};
+
+} // namespace js
+
+#endif /* builtin_SIMDConstants_h */

js/src/builtin/TestingFunctions.cpp

@@ -3980,7 +3980,12 @@ static bool
 IsSimdAvailable(JSContext* cx, unsigned argc, Value* vp)
 {
     CallArgs args = CallArgsFromVp(argc, vp);
-    args.rval().set(BooleanValue(cx->jitSupportsSimd()));
+#if defined(JS_CODEGEN_NONE) || !defined(ENABLE_SIMD)
+    bool available = false;
+#else
+    bool available = cx->jitSupportsSimd();
+#endif
+    args.rval().set(BooleanValue(available));
     return true;
 }
 

js/src/builtin/TypedObject.cpp

@@ -11,6 +11,7 @@
 
 #include "jsutil.h"
 
+#include "builtin/SIMDConstants.h"
 #include "gc/Marking.h"
 #include "js/Vector.h"
 #include "util/StringBuffer.h"
@@ -243,6 +244,10 @@ ScalarTypeDescr::typeName(Type type)
         JS_FOR_EACH_SCALAR_TYPE_REPR(NUMERIC_TYPE_TO_STRING)
 #undef NUMERIC_TYPE_TO_STRING
       case Scalar::Int64:
+      case Scalar::Float32x4:
+      case Scalar::Int8x16:
+      case Scalar::Int16x8:
+      case Scalar::Int32x4:
       case Scalar::MaxTypedArrayViewType:
         break;
     }
@@ -280,6 +285,10 @@ ScalarTypeDescr::call(JSContext* cx, unsigned argc, Value* vp)
         JS_FOR_EACH_SCALAR_TYPE_REPR(SCALARTYPE_CALL)
 #undef SCALARTYPE_CALL
       case Scalar::Int64:
+      case Scalar::Float32x4:
+      case Scalar::Int8x16:
+      case Scalar::Int16x8:
+      case Scalar::Int32x4:
       case Scalar::MaxTypedArrayViewType:
         MOZ_CRASH();
     }
@@ -392,6 +401,50 @@ js::ReferenceTypeDescr::call(JSContext* cx, unsigned argc, Value* vp)
     MOZ_CRASH("Unhandled Reference type");
 }
 
+/***************************************************************************
+ * SIMD type objects
+ *
+ * Note: these are partially defined in SIMD.cpp
+ */
+
+SimdType
+SimdTypeDescr::type() const {
+    uint32_t t = uint32_t(getReservedSlot(JS_DESCR_SLOT_TYPE).toInt32());
+    MOZ_ASSERT(t < uint32_t(SimdType::Count));
+    return SimdType(t);
+}
+
+uint32_t
+SimdTypeDescr::size(SimdType t)
+{
+    MOZ_ASSERT(unsigned(t) < unsigned(SimdType::Count));
+    switch (t) {
+      case SimdType::Int8x16:
+      case SimdType::Int16x8:
+      case SimdType::Int32x4:
+      case SimdType::Uint8x16:
+      case SimdType::Uint16x8:
+      case SimdType::Uint32x4:
+      case SimdType::Float32x4:
+      case SimdType::Float64x2:
+      case SimdType::Bool8x16:
+      case SimdType::Bool16x8:
+      case SimdType::Bool32x4:
+      case SimdType::Bool64x2:
+        return 16;
+      case SimdType::Count:
+        break;
+    }
+    MOZ_CRASH("unexpected SIMD type");
+}
+
+uint32_t
+SimdTypeDescr::alignment(SimdType t)
+{
+    MOZ_ASSERT(unsigned(t) < unsigned(SimdType::Count));
+    return size(t);
+}
+
 /***************************************************************************
  * ArrayMetaTypeDescr class
  */
@@ -1613,6 +1666,7 @@ TypeDescr::hasProperty(const JSAtomState& names, jsid id)
     switch (kind()) {
       case type::Scalar:
       case type::Reference:
+      case type::Simd:
         return false;
 
       case type::Array:
@@ -1685,6 +1739,7 @@ TypedObject::obj_hasProperty(JSContext* cx, HandleObject obj, HandleId id, bool*
     switch (typedObj->typeDescr().kind()) {
       case type::Scalar:
       case type::Reference:
+      case type::Simd:
         break;
 
       case type::Array: {
@@ -1736,6 +1791,9 @@ TypedObject::obj_getProperty(JSContext* cx, HandleObject obj, HandleValue receiv
       case type::Reference:
         break;
 
+      case type::Simd:
+        break;
+
       case type::Array:
         if (JSID_IS_ATOM(id, cx->names().length)) {
             if (!typedObj->isAttached()) {
@@ -1782,6 +1840,7 @@ TypedObject::obj_getElement(JSContext* cx, HandleObject obj, HandleValue receive
     switch (descr->kind()) {
       case type::Scalar:
       case type::Reference:
+      case type::Simd:
       case type::Struct:
         break;
 
@@ -1827,6 +1886,9 @@ TypedObject::obj_setProperty(JSContext* cx, HandleObject obj, HandleId id, Handl
       case type::Reference:
         break;
 
+      case type::Simd:
+        break;
+
       case type::Array: {
         if (JSID_IS_ATOM(id, cx->names().length)) {
             if (receiver.isObject() && obj == &receiver.toObject()) {
@@ -1894,6 +1956,7 @@ TypedObject::obj_getOwnPropertyDescriptor(JSContext* cx, HandleObject obj, Handl
     switch (descr->kind()) {
       case type::Scalar:
       case type::Reference:
+      case type::Simd:
         break;
 
       case type::Array:
@@ -1947,6 +2010,7 @@ IsOwnId(JSContext* cx, HandleObject obj, HandleId id)
     switch (typedObj->typeDescr().kind()) {
       case type::Scalar:
       case type::Reference:
+      case type::Simd:
         return false;
 
       case type::Array:
@@ -1985,7 +2049,8 @@ TypedObject::obj_newEnumerate(JSContext* cx, HandleObject obj, AutoIdVector& pro
     RootedId id(cx);
     switch (descr->kind()) {
       case type::Scalar:
-      case type::Reference: {
+      case type::Reference:
+      case type::Simd: {
         // Nothing to enumerate.
         break;
       }
@@ -2470,6 +2535,22 @@ js::GetTypedObjectModule(JSContext* cx, unsigned argc, Value* vp)
     return true;
 }
 
+bool
+js::GetSimdTypeDescr(JSContext* cx, unsigned argc, Value* vp)
+{
+    CallArgs args = CallArgsFromVp(argc, vp);
+    MOZ_ASSERT(args.length() == 1);
+    MOZ_ASSERT(args[0].isInt32());
+    // One of the JS_SIMDTYPEREPR_* constants / a SimdType enum value.
+    // getOrCreateSimdTypeDescr() will do the range check.
+    int32_t simdTypeRepr = args[0].toInt32();
+    Rooted<GlobalObject*> global(cx, cx->global());
+    MOZ_ASSERT(global);
+    auto* obj = GlobalObject::getOrCreateSimdTypeDescr(cx, global, SimdType(simdTypeRepr));
+    args.rval().setObject(*obj);
+    return true;
+}
+
 #define JS_STORE_SCALAR_CLASS_IMPL(_constant, T, _name)                         \
 bool                                                                            \
 js::StoreScalar##T::Func(JSContext* cx, unsigned argc, Value* vp)               \
@@ -2660,6 +2741,7 @@ visitReferences(TypeDescr& descr,
 
     switch (descr.kind()) {
       case type::Scalar:
+      case type::Simd:
         return;
 
       case type::Reference:

js/src/builtin/TypedObject.h

@@ -121,6 +121,7 @@ namespace type {
 enum Kind {
     Scalar = JS_TYPEREPR_SCALAR_KIND,
     Reference = JS_TYPEREPR_REFERENCE_KIND,
+    Simd = JS_TYPEREPR_SIMD_KIND,
     Struct = JS_TYPEREPR_STRUCT_KIND,
     Array = JS_TYPEREPR_ARRAY_KIND
 };
@@ -132,6 +133,7 @@ enum Kind {
 
 class SimpleTypeDescr;
 class ComplexTypeDescr;
+class SimdTypeDescr;
 class StructTypeDescr;
 class TypedProto;
 
@@ -253,6 +255,14 @@ class ScalarTypeDescr : public SimpleTypeDescr
                       "TypedObjectConstants.h must be consistent with Scalar::Type");
         static_assert(Scalar::Uint8Clamped == JS_SCALARTYPEREPR_UINT8_CLAMPED,
                       "TypedObjectConstants.h must be consistent with Scalar::Type");
+        static_assert(Scalar::Float32x4 == JS_SCALARTYPEREPR_FLOAT32X4,
+                      "TypedObjectConstants.h must be consistent with Scalar::Type");
+        static_assert(Scalar::Int8x16 == JS_SCALARTYPEREPR_INT8X16,
+                      "TypedObjectConstants.h must be consistent with Scalar::Type");
+        static_assert(Scalar::Int16x8 == JS_SCALARTYPEREPR_INT16X8,
+                      "TypedObjectConstants.h must be consistent with Scalar::Type");
+        static_assert(Scalar::Int32x4 == JS_SCALARTYPEREPR_INT32X4,
+                      "TypedObjectConstants.h must be consistent with Scalar::Type");
 
         return Type(getReservedSlot(JS_DESCR_SLOT_TYPE).toInt32());
     }
@@ -330,6 +340,25 @@ class ComplexTypeDescr : public TypeDescr
     }
 };
 
+enum class SimdType;
+
+/*
+ * SIMD Type descriptors.
+ */
+class SimdTypeDescr : public ComplexTypeDescr
+{
+  public:
+    static const type::Kind Kind = type::Simd;
+    static const bool Opaque = false;
+    static const Class class_;
+    static uint32_t size(SimdType t);
+    static uint32_t alignment(SimdType t);
+    static MOZ_MUST_USE bool call(JSContext* cx, unsigned argc, Value* vp);
+    static bool is(const Value& v);
+
+    SimdType type() const;
+};
+
 bool IsTypedObjectClass(const Class* clasp); // Defined below
 bool IsTypedObjectArray(JSObject& obj);
 
@@ -765,6 +794,16 @@ class InlineOpaqueTypedObject : public InlineTypedObject
     static const Class class_;
 };
 
+// Class for the global SIMD object.
+class SimdObject : public NativeObject
+{
+  public:
+    static const Class class_;
+    static MOZ_MUST_USE bool toString(JSContext* cx, unsigned int argc, Value* vp);
+    static MOZ_MUST_USE bool resolve(JSContext* cx, JS::HandleObject obj, JS::HandleId,
+                                     bool* resolved);
+};
+
 /*
  * Usage: NewOpaqueTypedObject(typeObj)
  *
@@ -862,6 +901,16 @@ MOZ_MUST_USE bool ClampToUint8(JSContext* cx, unsigned argc, Value* vp);
  */
 MOZ_MUST_USE bool GetTypedObjectModule(JSContext* cx, unsigned argc, Value* vp);
 
+/*
+ * Usage: GetSimdTypeDescr(simdTypeRepr)
+ *
+ * Returns one of the SIMD type objects, identified by `simdTypeRepr` which must
+ * be one of the JS_SIMDTYPEREPR_* constants.
+ *
+ * The SIMD pseudo-module must have been initialized for this to be safe.
+ */
+MOZ_MUST_USE bool GetSimdTypeDescr(JSContext* cx, unsigned argc, Value* vp);
+
 /*
  * Usage: Store_int8(targetDatum, targetOffset, value)
  *        ...
@@ -996,7 +1045,8 @@ inline bool
 IsComplexTypeDescrClass(const Class* clasp)
 {
     return clasp == &StructTypeDescr::class_ ||
-           clasp == &ArrayTypeDescr::class_;
+           clasp == &ArrayTypeDescr::class_ ||
+           clasp == &SimdTypeDescr::class_;
 }
 
 inline bool

js/src/builtin/TypedObject.js

@@ -53,6 +53,9 @@ function TypedObjectGet(descr, typedObj, offset) {
   case JS_TYPEREPR_REFERENCE_KIND:
     return TypedObjectGetReference(descr, typedObj, offset);
 
+  case JS_TYPEREPR_SIMD_KIND:
+    return TypedObjectGetSimd(descr, typedObj, offset);
+
   case JS_TYPEREPR_ARRAY_KIND:
   case JS_TYPEREPR_STRUCT_KIND:
     return TypedObjectGetDerived(descr, typedObj, offset);
@@ -134,6 +137,144 @@ function TypedObjectGetReference(descr, typedObj, offset) {
   return undefined;
 }
 
+function TypedObjectGetSimd(descr, typedObj, offset) {
+  var type = DESCR_TYPE(descr);
+  var simdTypeDescr = GetSimdTypeDescr(type);
+  switch (type) {
+  case JS_SIMDTYPEREPR_FLOAT32X4:
+    var x = Load_float32(typedObj, offset + 0);
+    var y = Load_float32(typedObj, offset + 4);
+    var z = Load_float32(typedObj, offset + 8);
+    var w = Load_float32(typedObj, offset + 12);
+    return simdTypeDescr(x, y, z, w);
+
+  case JS_SIMDTYPEREPR_FLOAT64X2:
+    var x = Load_float64(typedObj, offset + 0);
+    var y = Load_float64(typedObj, offset + 8);
+    return simdTypeDescr(x, y);
+
+  case JS_SIMDTYPEREPR_INT8X16:
+    var s0 = Load_int8(typedObj, offset + 0);
+    var s1 = Load_int8(typedObj, offset + 1);
+    var s2 = Load_int8(typedObj, offset + 2);
+    var s3 = Load_int8(typedObj, offset + 3);
+    var s4 = Load_int8(typedObj, offset + 4);
+    var s5 = Load_int8(typedObj, offset + 5);
+    var s6 = Load_int8(typedObj, offset + 6);
+    var s7 = Load_int8(typedObj, offset + 7);
+    var s8 = Load_int8(typedObj, offset + 8);
+    var s9 = Load_int8(typedObj, offset + 9);
+    var s10 = Load_int8(typedObj, offset + 10);
+    var s11 = Load_int8(typedObj, offset + 11);
+    var s12 = Load_int8(typedObj, offset + 12);
+    var s13 = Load_int8(typedObj, offset + 13);
+    var s14 = Load_int8(typedObj, offset + 14);
+    var s15 = Load_int8(typedObj, offset + 15);
+    return simdTypeDescr(s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, s12, s13, s14, s15);
+
+  case JS_SIMDTYPEREPR_INT16X8:
+    var s0 = Load_int16(typedObj, offset + 0);
+    var s1 = Load_int16(typedObj, offset + 2);
+    var s2 = Load_int16(typedObj, offset + 4);
+    var s3 = Load_int16(typedObj, offset + 6);
+    var s4 = Load_int16(typedObj, offset + 8);
+    var s5 = Load_int16(typedObj, offset + 10);
+    var s6 = Load_int16(typedObj, offset + 12);
+    var s7 = Load_int16(typedObj, offset + 14);
+    return simdTypeDescr(s0, s1, s2, s3, s4, s5, s6, s7);
+
+  case JS_SIMDTYPEREPR_INT32X4:
+    var x = Load_int32(typedObj, offset + 0);
+    var y = Load_int32(typedObj, offset + 4);
+    var z = Load_int32(typedObj, offset + 8);
+    var w = Load_int32(typedObj, offset + 12);
+    return simdTypeDescr(x, y, z, w);
+
+  case JS_SIMDTYPEREPR_UINT8X16:
+    var s0 = Load_uint8(typedObj, offset + 0);
+    var s1 = Load_uint8(typedObj, offset + 1);
+    var s2 = Load_uint8(typedObj, offset + 2);
+    var s3 = Load_uint8(typedObj, offset + 3);
+    var s4 = Load_uint8(typedObj, offset + 4);
+    var s5 = Load_uint8(typedObj, offset + 5);
+    var s6 = Load_uint8(typedObj, offset + 6);
+    var s7 = Load_uint8(typedObj, offset + 7);
+    var s8 = Load_uint8(typedObj, offset + 8);
+    var s9 = Load_uint8(typedObj, offset + 9);
+    var s10 = Load_uint8(typedObj, offset + 10);
+    var s11 = Load_uint8(typedObj, offset + 11);
+    var s12 = Load_uint8(typedObj, offset + 12);
+    var s13 = Load_uint8(typedObj, offset + 13);
+    var s14 = Load_uint8(typedObj, offset + 14);
+    var s15 = Load_uint8(typedObj, offset + 15);
+    return simdTypeDescr(s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, s12, s13, s14, s15);
+
+  case JS_SIMDTYPEREPR_UINT16X8:
+    var s0 = Load_uint16(typedObj, offset + 0);
+    var s1 = Load_uint16(typedObj, offset + 2);
+    var s2 = Load_uint16(typedObj, offset + 4);
+    var s3 = Load_uint16(typedObj, offset + 6);
+    var s4 = Load_uint16(typedObj, offset + 8);
+    var s5 = Load_uint16(typedObj, offset + 10);
+    var s6 = Load_uint16(typedObj, offset + 12);
+    var s7 = Load_uint16(typedObj, offset + 14);
+    return simdTypeDescr(s0, s1, s2, s3, s4, s5, s6, s7);
+
+  case JS_SIMDTYPEREPR_UINT32X4:
+    var x = Load_uint32(typedObj, offset + 0);
+    var y = Load_uint32(typedObj, offset + 4);
+    var z = Load_uint32(typedObj, offset + 8);
+    var w = Load_uint32(typedObj, offset + 12);
+    return simdTypeDescr(x, y, z, w);
+
+  case JS_SIMDTYPEREPR_BOOL8X16:
+    var s0 = Load_int8(typedObj, offset + 0);
+    var s1 = Load_int8(typedObj, offset + 1);
+    var s2 = Load_int8(typedObj, offset + 2);
+    var s3 = Load_int8(typedObj, offset + 3);
+    var s4 = Load_int8(typedObj, offset + 4);
+    var s5 = Load_int8(typedObj, offset + 5);
+    var s6 = Load_int8(typedObj, offset + 6);
+    var s7 = Load_int8(typedObj, offset + 7);
+    var s8 = Load_int8(typedObj, offset + 8);
+    var s9 = Load_int8(typedObj, offset + 9);
+    var s10 = Load_int8(typedObj, offset + 10);
+    var s11 = Load_int8(typedObj, offset + 11);
+    var s12 = Load_int8(typedObj, offset + 12);
+    var s13 = Load_int8(typedObj, offset + 13);
+    var s14 = Load_int8(typedObj, offset + 14);
+    var s15 = Load_int8(typedObj, offset + 15);
+    return simdTypeDescr(s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, s12, s13, s14, s15);
+
+  case JS_SIMDTYPEREPR_BOOL16X8:
+    var s0 = Load_int16(typedObj, offset + 0);
+    var s1 = Load_int16(typedObj, offset + 2);
+    var s2 = Load_int16(typedObj, offset + 4);
+    var s3 = Load_int16(typedObj, offset + 6);
+    var s4 = Load_int16(typedObj, offset + 8);
+    var s5 = Load_int16(typedObj, offset + 10);
+    var s6 = Load_int16(typedObj, offset + 12);
+    var s7 = Load_int16(typedObj, offset + 14);
+    return simdTypeDescr(s0, s1, s2, s3, s4, s5, s6, s7);
+
+  case JS_SIMDTYPEREPR_BOOL32X4:
+    var x = Load_int32(typedObj, offset + 0);
+    var y = Load_int32(typedObj, offset + 4);
+    var z = Load_int32(typedObj, offset + 8);
+    var w = Load_int32(typedObj, offset + 12);
+    return simdTypeDescr(x, y, z, w);
+
+  case JS_SIMDTYPEREPR_BOOL64X2:
+    var x = Load_int32(typedObj, offset + 0);
+    var y = Load_int32(typedObj, offset + 8);
+    return simdTypeDescr(x, y);
+
+  }
+
+  assert(false, "Unhandled SIMD type: " + type);
+  return undefined;
+}
+
 ///////////////////////////////////////////////////////////////////////////
 // Setting values
 //
@@ -155,6 +296,10 @@ function TypedObjectSet(descr, typedObj, offset, name, fromValue) {
     TypedObjectSetReference(descr, typedObj, offset, name, fromValue);
     return;
 
+  case JS_TYPEREPR_SIMD_KIND:
+    TypedObjectSetSimd(descr, typedObj, offset, fromValue);
+    return;
+
   case JS_TYPEREPR_ARRAY_KIND:
     var length = DESCR_ARRAY_LENGTH(descr);
     if (TypedObjectSetArray(descr, length, typedObj, offset, fromValue))
@@ -269,6 +414,106 @@ function TypedObjectSetReference(descr, typedObj, offset, name, fromValue) {
 }
 
 // Sets `fromValue` to `this` assuming that `this` is a scalar type.
+function TypedObjectSetSimd(descr, typedObj, offset, fromValue) {
+  if (!IsObject(fromValue) || !ObjectIsTypedObject(fromValue))
+    ThrowTypeError(JSMSG_CANT_CONVERT_TO,
+                   typeof(fromValue),
+                   DESCR_STRING_REPR(descr));
+
+  if (!DescrsEquiv(descr, TypedObjectTypeDescr(fromValue)))
+    ThrowTypeError(JSMSG_CANT_CONVERT_TO,
+                   typeof(fromValue),
+                   DESCR_STRING_REPR(descr));
+
+  var type = DESCR_TYPE(descr);
+  switch (type) {
+    case JS_SIMDTYPEREPR_FLOAT32X4:
+      Store_float32(typedObj, offset + 0, Load_float32(fromValue, 0));
+      Store_float32(typedObj, offset + 4, Load_float32(fromValue, 4));
+      Store_float32(typedObj, offset + 8, Load_float32(fromValue, 8));
+      Store_float32(typedObj, offset + 12, Load_float32(fromValue, 12));
+      break;
+    case JS_SIMDTYPEREPR_FLOAT64X2:
+      Store_float64(typedObj, offset + 0, Load_float64(fromValue, 0));
+      Store_float64(typedObj, offset + 8, Load_float64(fromValue, 8));
+      break;
+    case JS_SIMDTYPEREPR_INT8X16:
+    case JS_SIMDTYPEREPR_BOOL8X16:
+      Store_int8(typedObj, offset + 0, Load_int8(fromValue, 0));
+      Store_int8(typedObj, offset + 1, Load_int8(fromValue, 1));
+      Store_int8(typedObj, offset + 2, Load_int8(fromValue, 2));
+      Store_int8(typedObj, offset + 3, Load_int8(fromValue, 3));
+      Store_int8(typedObj, offset + 4, Load_int8(fromValue, 4));
+      Store_int8(typedObj, offset + 5, Load_int8(fromValue, 5));
+      Store_int8(typedObj, offset + 6, Load_int8(fromValue, 6));
+      Store_int8(typedObj, offset + 7, Load_int8(fromValue, 7));
+      Store_int8(typedObj, offset + 8, Load_int8(fromValue, 8));
+      Store_int8(typedObj, offset + 9, Load_int8(fromValue, 9));
+      Store_int8(typedObj, offset + 10, Load_int8(fromValue, 10));
+      Store_int8(typedObj, offset + 11, Load_int8(fromValue, 11));
+      Store_int8(typedObj, offset + 12, Load_int8(fromValue, 12));
+      Store_int8(typedObj, offset + 13, Load_int8(fromValue, 13));
+      Store_int8(typedObj, offset + 14, Load_int8(fromValue, 14));
+      Store_int8(typedObj, offset + 15, Load_int8(fromValue, 15));
+      break;
+    case JS_SIMDTYPEREPR_INT16X8:
+    case JS_SIMDTYPEREPR_BOOL16X8:
+      Store_int16(typedObj, offset + 0, Load_int16(fromValue, 0));
+      Store_int16(typedObj, offset + 2, Load_int16(fromValue, 2));
+      Store_int16(typedObj, offset + 4, Load_int16(fromValue, 4));
+      Store_int16(typedObj, offset + 6, Load_int16(fromValue, 6));
+      Store_int16(typedObj, offset + 8, Load_int16(fromValue, 8));
+      Store_int16(typedObj, offset + 10, Load_int16(fromValue, 10));
+      Store_int16(typedObj, offset + 12, Load_int16(fromValue, 12));
+      Store_int16(typedObj, offset + 14, Load_int16(fromValue, 14));
+      break;
+    case JS_SIMDTYPEREPR_INT32X4:
+    case JS_SIMDTYPEREPR_BOOL32X4:
+    case JS_SIMDTYPEREPR_BOOL64X2:
+      Store_int32(typedObj, offset + 0, Load_int32(fromValue, 0));
+      Store_int32(typedObj, offset + 4, Load_int32(fromValue, 4));
+      Store_int32(typedObj, offset + 8, Load_int32(fromValue, 8));
+      Store_int32(typedObj, offset + 12, Load_int32(fromValue, 12));
+      break;
+    case JS_SIMDTYPEREPR_UINT8X16:
+      Store_uint8(typedObj, offset + 0, Load_uint8(fromValue, 0));
+      Store_uint8(typedObj, offset + 1, Load_uint8(fromValue, 1));
+      Store_uint8(typedObj, offset + 2, Load_uint8(fromValue, 2));
+      Store_uint8(typedObj, offset + 3, Load_uint8(fromValue, 3));
+      Store_uint8(typedObj, offset + 4, Load_uint8(fromValue, 4));
+      Store_uint8(typedObj, offset + 5, Load_uint8(fromValue, 5));
+      Store_uint8(typedObj, offset + 6, Load_uint8(fromValue, 6));
+      Store_uint8(typedObj, offset + 7, Load_uint8(fromValue, 7));
+      Store_uint8(typedObj, offset + 8, Load_uint8(fromValue, 8));
+      Store_uint8(typedObj, offset + 9, Load_uint8(fromValue, 9));
+      Store_uint8(typedObj, offset + 10, Load_uint8(fromValue, 10));
+      Store_uint8(typedObj, offset + 11, Load_uint8(fromValue, 11));
+      Store_uint8(typedObj, offset + 12, Load_uint8(fromValue, 12));
+      Store_uint8(typedObj, offset + 13, Load_uint8(fromValue, 13));
+      Store_uint8(typedObj, offset + 14, Load_uint8(fromValue, 14));
+      Store_uint8(typedObj, offset + 15, Load_uint8(fromValue, 15));
+      break;
+    case JS_SIMDTYPEREPR_UINT16X8:
+      Store_uint16(typedObj, offset + 0, Load_uint16(fromValue, 0));
+      Store_uint16(typedObj, offset + 2, Load_uint16(fromValue, 2));
+      Store_uint16(typedObj, offset + 4, Load_uint16(fromValue, 4));
+      Store_uint16(typedObj, offset + 6, Load_uint16(fromValue, 6));
+      Store_uint16(typedObj, offset + 8, Load_uint16(fromValue, 8));
+      Store_uint16(typedObj, offset + 10, Load_uint16(fromValue, 10));
+      Store_uint16(typedObj, offset + 12, Load_uint16(fromValue, 12));
+      Store_uint16(typedObj, offset + 14, Load_uint16(fromValue, 14));
+      break;
+    case JS_SIMDTYPEREPR_UINT32X4:
+      Store_uint32(typedObj, offset + 0, Load_uint32(fromValue, 0));
+      Store_uint32(typedObj, offset + 4, Load_uint32(fromValue, 4));
+      Store_uint32(typedObj, offset + 8, Load_uint32(fromValue, 8));
+      Store_uint32(typedObj, offset + 12, Load_uint32(fromValue, 12));
+      break;
+    default:
+      assert(false, "Unhandled Simd type: " + type);
+  }
+}
+
 ///////////////////////////////////////////////////////////////////////////
 // C++ Wrappers
 //
@@ -384,6 +629,241 @@ function TypedObjectArrayRedimension(newArrayType) {
   return NewDerivedTypedObject(newArrayType, this, 0);
 }
 
+///////////////////////////////////////////////////////////////////////////
+// SIMD
+
+function SimdProtoString(type) {
+  switch (type) {
+  case JS_SIMDTYPEREPR_INT8X16:
+    return "Int8x16";
+  case JS_SIMDTYPEREPR_INT16X8:
+    return "Int16x8";
+  case JS_SIMDTYPEREPR_INT32X4:
+    return "Int32x4";
+  case JS_SIMDTYPEREPR_UINT8X16:
+    return "Uint8x16";
+  case JS_SIMDTYPEREPR_UINT16X8:
+    return "Uint16x8";
+  case JS_SIMDTYPEREPR_UINT32X4:
+    return "Uint32x4";
+  case JS_SIMDTYPEREPR_FLOAT32X4:
+    return "Float32x4";
+  case JS_SIMDTYPEREPR_FLOAT64X2:
+    return "Float64x2";
+  case JS_SIMDTYPEREPR_BOOL8X16:
+    return "Bool8x16";
+  case JS_SIMDTYPEREPR_BOOL16X8:
+    return "Bool16x8";
+  case JS_SIMDTYPEREPR_BOOL32X4:
+    return "Bool32x4";
+  case JS_SIMDTYPEREPR_BOOL64X2:
+    return "Bool64x2";
+  }
+
+  assert(false, "Unhandled type constant");
+  return undefined;
+}
+
+function SimdTypeToLength(type) {
+  switch (type) {
+  case JS_SIMDTYPEREPR_INT8X16:
+  case JS_SIMDTYPEREPR_BOOL8X16:
+    return 16;
+  case JS_SIMDTYPEREPR_INT16X8:
+  case JS_SIMDTYPEREPR_BOOL16X8:
+    return 8;
+  case JS_SIMDTYPEREPR_INT32X4:
+  case JS_SIMDTYPEREPR_FLOAT32X4:
+  case JS_SIMDTYPEREPR_BOOL32X4:
+    return 4;
+  case JS_SIMDTYPEREPR_FLOAT64X2:
+  case JS_SIMDTYPEREPR_BOOL64X2:
+    return 2;
+  }
+
+  assert(false, "Unhandled type constant");
+  return undefined;
+}
+
+// This implements SIMD.*.prototype.valueOf().
+// Once we have proper value semantics for SIMD types, this function should just
+// perform a type check and return this.
+// For now, throw a TypeError unconditionally since valueOf() was probably
+// called from ToNumber() which is supposed to throw when attempting to convert
+// a SIMD value to a number.
+function SimdValueOf() {
+  if (!IsObject(this) || !ObjectIsTypedObject(this))
+    ThrowTypeError(JSMSG_INCOMPATIBLE_PROTO, "SIMD", "valueOf", typeof this);
+
+  var descr = TypedObjectTypeDescr(this);
+
+  if (DESCR_KIND(descr) != JS_TYPEREPR_SIMD_KIND)
+    ThrowTypeError(JSMSG_INCOMPATIBLE_PROTO, "SIMD", "valueOf", typeof this);
+
+  ThrowTypeError(JSMSG_SIMD_TO_NUMBER);
+}
+
+function SimdToSource() {
+  if (!IsObject(this) || !ObjectIsTypedObject(this))
+    ThrowTypeError(JSMSG_INCOMPATIBLE_PROTO, "SIMD.*", "toSource", typeof this);
+
+  var descr = TypedObjectTypeDescr(this);
+
+  if (DESCR_KIND(descr) != JS_TYPEREPR_SIMD_KIND)
+    ThrowTypeError(JSMSG_INCOMPATIBLE_PROTO, "SIMD.*", "toSource", typeof this);
+
+  return SimdFormatString(descr, this);
+}
+
+function SimdToString() {
+  if (!IsObject(this) || !ObjectIsTypedObject(this))
+    ThrowTypeError(JSMSG_INCOMPATIBLE_PROTO, "SIMD.*", "toString", typeof this);
+
+  var descr = TypedObjectTypeDescr(this);
+
+  if (DESCR_KIND(descr) != JS_TYPEREPR_SIMD_KIND)
+    ThrowTypeError(JSMSG_INCOMPATIBLE_PROTO, "SIMD.*", "toString", typeof this);
+
+  return SimdFormatString(descr, this);
+}
+
+function SimdFormatString(descr, typedObj) {
+  var typerepr = DESCR_TYPE(descr);
+  var protoString = SimdProtoString(typerepr);
+  switch (typerepr) {
+      case JS_SIMDTYPEREPR_INT8X16: {
+          var s1 = callFunction(std_SIMD_Int8x16_extractLane, null, typedObj, 0);
+          var s2 = callFunction(std_SIMD_Int8x16_extractLane, null, typedObj, 1);
+          var s3 = callFunction(std_SIMD_Int8x16_extractLane, null, typedObj, 2);
+          var s4 = callFunction(std_SIMD_Int8x16_extractLane, null, typedObj, 3);
+          var s5 = callFunction(std_SIMD_Int8x16_extractLane, null, typedObj, 4);
+          var s6 = callFunction(std_SIMD_Int8x16_extractLane, null, typedObj, 5);
+          var s7 = callFunction(std_SIMD_Int8x16_extractLane, null, typedObj, 6);
+          var s8 = callFunction(std_SIMD_Int8x16_extractLane, null, typedObj, 7);
+          var s9 = callFunction(std_SIMD_Int8x16_extractLane, null, typedObj, 8);
+          var s10 = callFunction(std_SIMD_Int8x16_extractLane, null, typedObj, 9);
+          var s11 = callFunction(std_SIMD_Int8x16_extractLane, null, typedObj, 10);
+          var s12 = callFunction(std_SIMD_Int8x16_extractLane, null, typedObj, 11);
+          var s13 = callFunction(std_SIMD_Int8x16_extractLane, null, typedObj, 12);
+          var s14 = callFunction(std_SIMD_Int8x16_extractLane, null, typedObj, 13);
+          var s15 = callFunction(std_SIMD_Int8x16_extractLane, null, typedObj, 14);
+          var s16 = callFunction(std_SIMD_Int8x16_extractLane, null, typedObj, 15);
+          return `SIMD.${protoString}(${s1}, ${s2}, ${s3}, ${s4}, ${s5}, ${s6}, ${s7}, ${s8}, ${s9}, ${s10}, ${s11}, ${s12}, ${s13}, ${s14}, ${s15}, ${s16})`;
+      }
+      case JS_SIMDTYPEREPR_INT16X8: {
+          var s1 = callFunction(std_SIMD_Int16x8_extractLane, null, typedObj, 0);
+          var s2 = callFunction(std_SIMD_Int16x8_extractLane, null, typedObj, 1);
+          var s3 = callFunction(std_SIMD_Int16x8_extractLane, null, typedObj, 2);
+          var s4 = callFunction(std_SIMD_Int16x8_extractLane, null, typedObj, 3);
+          var s5 = callFunction(std_SIMD_Int16x8_extractLane, null, typedObj, 4);
+          var s6 = callFunction(std_SIMD_Int16x8_extractLane, null, typedObj, 5);
+          var s7 = callFunction(std_SIMD_Int16x8_extractLane, null, typedObj, 6);
+          var s8 = callFunction(std_SIMD_Int16x8_extractLane, null, typedObj, 7);
+          return `SIMD.${protoString}(${s1}, ${s2}, ${s3}, ${s4}, ${s5}, ${s6}, ${s7}, ${s8})`;
+      }
+      case JS_SIMDTYPEREPR_INT32X4: {
+          var x = callFunction(std_SIMD_Int32x4_extractLane, null, typedObj, 0);
+          var y = callFunction(std_SIMD_Int32x4_extractLane, null, typedObj, 1);
+          var z = callFunction(std_SIMD_Int32x4_extractLane, null, typedObj, 2);
+          var w = callFunction(std_SIMD_Int32x4_extractLane, null, typedObj, 3);
+          return `SIMD.${protoString}(${x}, ${y}, ${z}, ${w})`;
+      }
+      case JS_SIMDTYPEREPR_UINT8X16: {
+          var s1 = callFunction(std_SIMD_Uint8x16_extractLane, null, typedObj, 0);
+          var s2 = callFunction(std_SIMD_Uint8x16_extractLane, null, typedObj, 1);
+          var s3 = callFunction(std_SIMD_Uint8x16_extractLane, null, typedObj, 2);
+          var s4 = callFunction(std_SIMD_Uint8x16_extractLane, null, typedObj, 3);
+          var s5 = callFunction(std_SIMD_Uint8x16_extractLane, null, typedObj, 4);
+          var s6 = callFunction(std_SIMD_Uint8x16_extractLane, null, typedObj, 5);
+          var s7 = callFunction(std_SIMD_Uint8x16_extractLane, null, typedObj, 6);
+          var s8 = callFunction(std_SIMD_Uint8x16_extractLane, null, typedObj, 7);
+          var s9 = callFunction(std_SIMD_Uint8x16_extractLane, null, typedObj, 8);
+          var s10 = callFunction(std_SIMD_Uint8x16_extractLane, null, typedObj, 9);
+          var s11 = callFunction(std_SIMD_Uint8x16_extractLane, null, typedObj, 10);
+          var s12 = callFunction(std_SIMD_Uint8x16_extractLane, null, typedObj, 11);
+          var s13 = callFunction(std_SIMD_Uint8x16_extractLane, null, typedObj, 12);
+          var s14 = callFunction(std_SIMD_Uint8x16_extractLane, null, typedObj, 13);
+          var s15 = callFunction(std_SIMD_Uint8x16_extractLane, null, typedObj, 14);
+          var s16 = callFunction(std_SIMD_Uint8x16_extractLane, null, typedObj, 15);
+          return `SIMD.${protoString}(${s1}, ${s2}, ${s3}, ${s4}, ${s5}, ${s6}, ${s7}, ${s8}, ${s9}, ${s10}, ${s11}, ${s12}, ${s13}, ${s14}, ${s15}, ${s16})`;
+      }
+      case JS_SIMDTYPEREPR_UINT16X8: {
+          var s1 = callFunction(std_SIMD_Uint16x8_extractLane, null, typedObj, 0);
+          var s2 = callFunction(std_SIMD_Uint16x8_extractLane, null, typedObj, 1);
+          var s3 = callFunction(std_SIMD_Uint16x8_extractLane, null, typedObj, 2);
+          var s4 = callFunction(std_SIMD_Uint16x8_extractLane, null, typedObj, 3);
+          var s5 = callFunction(std_SIMD_Uint16x8_extractLane, null, typedObj, 4);
+          var s6 = callFunction(std_SIMD_Uint16x8_extractLane, null, typedObj, 5);
+          var s7 = callFunction(std_SIMD_Uint16x8_extractLane, null, typedObj, 6);
+          var s8 = callFunction(std_SIMD_Uint16x8_extractLane, null, typedObj, 7);
+          return `SIMD.${protoString}(${s1}, ${s2}, ${s3}, ${s4}, ${s5}, ${s6}, ${s7}, ${s8})`;
+      }
+      case JS_SIMDTYPEREPR_UINT32X4: {
+          var x = callFunction(std_SIMD_Uint32x4_extractLane, null, typedObj, 0);
+          var y = callFunction(std_SIMD_Uint32x4_extractLane, null, typedObj, 1);
+          var z = callFunction(std_SIMD_Uint32x4_extractLane, null, typedObj, 2);
+          var w = callFunction(std_SIMD_Uint32x4_extractLane, null, typedObj, 3);
+          return `SIMD.${protoString}(${x}, ${y}, ${z}, ${w})`;
+      }
+      case JS_SIMDTYPEREPR_FLOAT32X4: {
+          var x = callFunction(std_SIMD_Float32x4_extractLane, null, typedObj, 0);
+          var y = callFunction(std_SIMD_Float32x4_extractLane, null, typedObj, 1);
+          var z = callFunction(std_SIMD_Float32x4_extractLane, null, typedObj, 2);
+          var w = callFunction(std_SIMD_Float32x4_extractLane, null, typedObj, 3);
+          return `SIMD.${protoString}(${x}, ${y}, ${z}, ${w})`;
+      }
+      case JS_SIMDTYPEREPR_FLOAT64X2: {
+          var x = callFunction(std_SIMD_Float64x2_extractLane, null, typedObj, 0);
+          var y = callFunction(std_SIMD_Float64x2_extractLane, null, typedObj, 1);
+          return `SIMD.${protoString}(${x}, ${y})`;
+      }
+      case JS_SIMDTYPEREPR_BOOL8X16: {
+          var s1 = callFunction(std_SIMD_Bool8x16_extractLane, null, typedObj, 0);
+          var s2 = callFunction(std_SIMD_Bool8x16_extractLane, null, typedObj, 1);
+          var s3 = callFunction(std_SIMD_Bool8x16_extractLane, null, typedObj, 2);
+          var s4 = callFunction(std_SIMD_Bool8x16_extractLane, null, typedObj, 3);
+          var s5 = callFunction(std_SIMD_Bool8x16_extractLane, null, typedObj, 4);
+          var s6 = callFunction(std_SIMD_Bool8x16_extractLane, null, typedObj, 5);
+          var s7 = callFunction(std_SIMD_Bool8x16_extractLane, null, typedObj, 6);
+          var s8 = callFunction(std_SIMD_Bool8x16_extractLane, null, typedObj, 7);
+          var s9 = callFunction(std_SIMD_Bool8x16_extractLane, null, typedObj, 8);
+          var s10 = callFunction(std_SIMD_Bool8x16_extractLane, null, typedObj, 9);
+          var s11 = callFunction(std_SIMD_Bool8x16_extractLane, null, typedObj, 10);
+          var s12 = callFunction(std_SIMD_Bool8x16_extractLane, null, typedObj, 11);
+          var s13 = callFunction(std_SIMD_Bool8x16_extractLane, null, typedObj, 12);
+          var s14 = callFunction(std_SIMD_Bool8x16_extractLane, null, typedObj, 13);
+          var s15 = callFunction(std_SIMD_Bool8x16_extractLane, null, typedObj, 14);
+          var s16 = callFunction(std_SIMD_Bool8x16_extractLane, null, typedObj, 15);
+          return `SIMD.${protoString}(${s1}, ${s2}, ${s3}, ${s4}, ${s5}, ${s6}, ${s7}, ${s8}, ${s9}, ${s10}, ${s11}, ${s12}, ${s13}, ${s14}, ${s15}, ${s16})`;
+      }
+      case JS_SIMDTYPEREPR_BOOL16X8: {
+          var s1 = callFunction(std_SIMD_Bool16x8_extractLane, null, typedObj, 0);
+          var s2 = callFunction(std_SIMD_Bool16x8_extractLane, null, typedObj, 1);
+          var s3 = callFunction(std_SIMD_Bool16x8_extractLane, null, typedObj, 2);
+          var s4 = callFunction(std_SIMD_Bool16x8_extractLane, null, typedObj, 3);
+          var s5 = callFunction(std_SIMD_Bool16x8_extractLane, null, typedObj, 4);
+          var s6 = callFunction(std_SIMD_Bool16x8_extractLane, null, typedObj, 5);
+          var s7 = callFunction(std_SIMD_Bool16x8_extractLane, null, typedObj, 6);
+          var s8 = callFunction(std_SIMD_Bool16x8_extractLane, null, typedObj, 7);
+          return `SIMD.${protoString}(${s1}, ${s2}, ${s3}, ${s4}, ${s5}, ${s6}, ${s7}, ${s8})`;
+      }
+      case JS_SIMDTYPEREPR_BOOL32X4: {
+          var x = callFunction(std_SIMD_Bool32x4_extractLane, null, typedObj, 0);
+          var y = callFunction(std_SIMD_Bool32x4_extractLane, null, typedObj, 1);
+          var z = callFunction(std_SIMD_Bool32x4_extractLane, null, typedObj, 2);
+          var w = callFunction(std_SIMD_Bool32x4_extractLane, null, typedObj, 3);
+          return `SIMD.${protoString}(${x}, ${y}, ${z}, ${w})`;
+      }
+      case JS_SIMDTYPEREPR_BOOL64X2: {
+          var x = callFunction(std_SIMD_Bool64x2_extractLane, null, typedObj, 0);
+          var y = callFunction(std_SIMD_Bool64x2_extractLane, null, typedObj, 1);
+          return `SIMD.${protoString}(${x}, ${y})`;
+      }
+  }
+  assert(false, "unexpected SIMD kind");
+  return "?";
+}
+
 ///////////////////////////////////////////////////////////////////////////
 // Miscellaneous
 

js/src/builtin/TypedObjectConstants.h

@@ -53,7 +53,7 @@
 #define JS_DESCR_SLOT_ARRAYPROTO         6  // Lazily created prototype for arrays
 #define JS_DESCR_SLOT_TRACE_LIST         7  // List of references for use in tracing
 
-// Slots on scalars, references
+// Slots on scalars, references, and SIMD objects
 #define JS_DESCR_SLOT_TYPE               8  // Type code
 
 // Slots on array descriptors
@@ -75,6 +75,7 @@
 #define JS_TYPEREPR_REFERENCE_KIND      2
 #define JS_TYPEREPR_STRUCT_KIND         3
 #define JS_TYPEREPR_ARRAY_KIND          4
+#define JS_TYPEREPR_SIMD_KIND           5
 
 // These constants are for use exclusively in JS code. In C++ code,
 // prefer Scalar::Int8 etc, which allows you to write a switch which will
@@ -88,6 +89,10 @@
 #define JS_SCALARTYPEREPR_FLOAT32       6
 #define JS_SCALARTYPEREPR_FLOAT64       7
 #define JS_SCALARTYPEREPR_UINT8_CLAMPED 8
+#define JS_SCALARTYPEREPR_FLOAT32X4     11
+#define JS_SCALARTYPEREPR_INT8X16       12
+#define JS_SCALARTYPEREPR_INT16X8       13
+#define JS_SCALARTYPEREPR_INT32X4       14
 
 // These constants are for use exclusively in JS code. In C++ code,
 // prefer ReferenceTypeRepresentation::TYPE_ANY etc, which allows
@@ -97,4 +102,20 @@
 #define JS_REFERENCETYPEREPR_OBJECT     1
 #define JS_REFERENCETYPEREPR_STRING     2
 
+// These constants are for use exclusively in JS code. In C++ code, prefer
+// SimdType::Int32x4 etc, since that allows you to write a switch which will
+// receive a warning if you omit a case.
+#define JS_SIMDTYPEREPR_INT8X16         0
+#define JS_SIMDTYPEREPR_INT16X8         1
+#define JS_SIMDTYPEREPR_INT32X4         2
+#define JS_SIMDTYPEREPR_UINT8X16        3
+#define JS_SIMDTYPEREPR_UINT16X8        4
+#define JS_SIMDTYPEREPR_UINT32X4        5
+#define JS_SIMDTYPEREPR_FLOAT32X4       6
+#define JS_SIMDTYPEREPR_FLOAT64X2       7
+#define JS_SIMDTYPEREPR_BOOL8X16        8
+#define JS_SIMDTYPEREPR_BOOL16X8        9
+#define JS_SIMDTYPEREPR_BOOL32X4       10
+#define JS_SIMDTYPEREPR_BOOL64X2       11
+
 #endif

js/src/devtools/automation/cgc-jittest-timeouts.txt

@@ -1,3 +1,4 @@
+SIMD/nursery-overflow.js
 asm.js/testBug1117235.js
 asm.js/testParallelCompile.js
 auto-regress/bug653395.js

js/src/doc/JITOptimizations/Outcomes.md

@@ -152,6 +152,11 @@ Failed to do range check of element access on a typed object.
 
 ### AccessNotDense
 
+### AccessNotSimdObject
+
+The observed type of the target of the property access doesn't guarantee
+that it is a SIMD object.
+
 ### AccessNotTypedObject
 
 The observed type of the target of the property access doesn't guarantee
@@ -217,6 +222,15 @@ the keys have never been observed to be a String, Symbol, or Int32.
 IonMonkey only generates inline caches for element accesses which are
 either on dense objects (e.g. dense Arrays), or Typed Arrays.
 
+### NoSimdJitSupport
+
+Optimization failed because SIMD JIT support was not enabled.
+
+### SimdTypeNotOptimized
+
+The type observed as being retrieved from this property access did not
+match an optimizable type.
+
 ### HasCommonInliningPath
 
 Inlining was abandoned because the inlining call path was repeated.  A

js/src/jit-test/jit_test.py

@@ -253,6 +253,9 @@ def main(argv):
     # This part is equivalent to:
     # skip-if = coverage
     if os.getenv('GCOV_PREFIX') is not None:
+        # GCOV errors.
+        options.exclude += [os.path.join('asm.js', 'testSIMD.js')]               # Bug 1347245
+
         # JSVM errors.
         options.exclude += [os.path.join('basic', 'functionnames.js')]           # Bug 1369783
         options.exclude += [os.path.join('debug', 'Debugger-findScripts-23.js')]

js/src/jit-test/lib/simd.js

@@ -0,0 +1,109 @@
+if (!this.hasOwnProperty("SIMD"))
+    quit();
+
+function booleanBinaryX4(op, v, w) {
+    var arr = [];
+    var [varr, warr] = [simdToArray(v), simdToArray(w)];
+    for (var i = 0; i < 4; i++)
+        arr[i] = op(varr[i], warr[i]);
+    return arr;
+}
+
+function binaryX(op, v, w) {
+    var arr = [];
+    var [varr, warr] = [simdToArray(v), simdToArray(w)];
+    [varr, warr] = [varr.map(Math.fround), warr.map(Math.fround)];
+    for (var i = 0; i < varr.length; i++)
+        arr[i] = op(varr[i], warr[i]);
+    return arr.map(Math.fround);
+}
+
+function unaryX4(op, v, coerceFunc) {
+    var arr = [];
+    var varr = simdToArray(v).map(coerceFunc);
+    for (var i = 0; i < 4; i++)
+        arr[i] = op(varr[i]);
+    return arr.map(coerceFunc);
+}
+
+function assertNear(a, b) {
+    assertEq((a != a && b != b) || Math.abs(a - b) < 0.001, true);
+}
+
+function GetType(v) {
+    var pt = Object.getPrototypeOf(v);
+    switch (pt) {
+        case SIMD.Int8x16.prototype: return SIMD.Int8x16;
+        case SIMD.Int16x8.prototype: return SIMD.Int16x8;
+        case SIMD.Int32x4.prototype: return SIMD.Int32x4;
+        case SIMD.Uint8x16.prototype: return SIMD.Uint8x16;
+        case SIMD.Uint16x8.prototype: return SIMD.Uint16x8;
+        case SIMD.Uint32x4.prototype: return SIMD.Uint32x4;
+        case SIMD.Float32x4.prototype: return SIMD.Float32x4;
+        case SIMD.Bool8x16.prototype: return SIMD.Bool8x16;
+        case SIMD.Bool16x8.prototype: return SIMD.Bool16x8;
+        case SIMD.Bool32x4.prototype: return SIMD.Bool32x4;
+    }
+    throw "unexpected SIMD type";
+}
+
+function GetLength(t) {
+    switch (t) {
+      case SIMD.Int8x16: return 16;
+      case SIMD.Int16x8: return 8;
+      case SIMD.Int32x4: return 4;
+      case SIMD.Uint8x16: return 16;
+      case SIMD.Uint16x8: return 8;
+      case SIMD.Uint32x4: return 4;
+      case SIMD.Float32x4: return 4;
+      case SIMD.Bool8x16: return 16;
+      case SIMD.Bool16x8: return 8;
+      case SIMD.Bool32x4: return 4;
+    }
+    throw "unexpected SIMD type";
+}
+
+function assertEqVec(v, w) {
+    var typeV = GetType(v);
+    var lengthV = GetLength(typeV);
+    var ext = typeV.extractLane;
+    assertEq(GetType(w), typeV);
+    for (var i = 0; i < lengthV; i++)
+        assertEq(ext(v, i), ext(w, i));
+}
+
+function assertEqVecArr(v, w) {
+    var typeV = GetType(v);
+    var lengthV = GetLength(typeV);
+    var ext = typeV.extractLane;
+    assertEq(w.length, lengthV);
+
+    for (var i = 0; i < lengthV; i++)
+        assertEq(ext(v, i), w[i]);
+}
+
+function assertEqX4(vec, arr, ...opts) {
+
+    var assertFunc;
+    if (opts.length == 1 && typeof opts[0] !== 'undefined') {
+        assertFunc = opts[0];
+    } else {
+        assertFunc = assertEq;
+    }
+
+    var Type = GetType(vec);
+
+    assertFunc(Type.extractLane(vec, 0), arr[0]);
+    assertFunc(Type.extractLane(vec, 1), arr[1]);
+    assertFunc(Type.extractLane(vec, 2), arr[2]);
+    assertFunc(Type.extractLane(vec, 3), arr[3]);
+}
+
+function simdToArray(vec) {
+    var Type = GetType(vec);
+    var Length = GetLength(Type);
+    var a = [];
+    for (var i = 0; i < Length; i++)
+        a.push(Type.extractLane(vec, i));
+    return a;
+}

js/src/jit-test/tests/SIMD/anyall.js

@@ -0,0 +1,38 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 50);
+
+function all(B, n) {
+    var a = B.splat(true);
+    for (var i = 0; i < n; i++) {
+        var b = B.replaceLane(a, i, false);
+        assertEq(B.allTrue(b), false);
+        var c = B.replaceLane(b, i, true);
+        assertEq(B.allTrue(c), true);
+    }
+}
+
+function any(B, n) {
+    var a = B.splat(false);
+    for (var i = 0; i < n; i++) {
+        var b = B.replaceLane(a, i, true);
+        assertEq(B.anyTrue(b), true);
+        var c = B.replaceLane(b, i, false);
+        assertEq(B.anyTrue(c), false);
+    }
+}
+
+function f() {
+    for (var j = 0; j < 200; j++) {
+        all(SIMD.Bool64x2, 2)
+        any(SIMD.Bool64x2, 2)
+        all(SIMD.Bool32x4, 4)
+        any(SIMD.Bool32x4, 4)
+        all(SIMD.Bool16x8, 8)
+        any(SIMD.Bool16x8, 8)
+        all(SIMD.Bool8x16, 16)
+        any(SIMD.Bool8x16, 16)
+    }
+}
+
+f()

js/src/jit-test/tests/SIMD/binary-arith.js

@@ -0,0 +1,30 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 50);
+
+function f() {
+    var i1 = SIMD.Int32x4(1, 2, 3, 4);
+    var i2 = SIMD.Int32x4(4, 3, 2, 1);
+
+    var f1 = SIMD.Float32x4(1, 2, 3, 4);
+    var f2 = SIMD.Float32x4(4, 3, 2, 1);
+
+    var i8_1 = SIMD.Int8x16(1, 2, 3, 4, 20, 30, 40, 50, 100, 115, 120, 125);
+    var i8_2 = SIMD.Int8x16(4, 3, 2, 1,  8,  7,  6,  5,  12,  11,  10,   9);
+
+    for (var i = 0; i < 150; i++) {
+        assertEqX4(SIMD.Float32x4.add(f1, f2), binaryX((x, y) => x + y, f1, f2));
+        assertEqX4(SIMD.Float32x4.sub(f1, f2), binaryX((x, y) => x - y, f1, f2));
+        assertEqX4(SIMD.Float32x4.mul(f1, f2), binaryX((x, y) => x * y, f1, f2));
+
+        assertEqX4(SIMD.Int32x4.add(i1, i2), binaryX((x, y) => x + y, i1, i2));
+        assertEqX4(SIMD.Int32x4.sub(i1, i2), binaryX((x, y) => x - y, i1, i2));
+        assertEqX4(SIMD.Int32x4.mul(i1, i2), binaryX((x, y) => x * y, i1, i2));
+
+        assertEqX4(SIMD.Int8x16.add(i8_1, i8_2), binaryX((x, y) => (x + y) << 24 >> 24, i8_1, i8_2));
+        assertEqX4(SIMD.Int8x16.sub(i8_1, i8_2), binaryX((x, y) => (x - y) << 24 >> 24, i8_1, i8_2));
+        assertEqX4(SIMD.Int8x16.mul(i8_1, i8_2), binaryX((x, y) => (x * y) << 24 >> 24, i8_1, i8_2));
+    }
+}
+
+f();

js/src/jit-test/tests/SIMD/bool32x4-arith.js

@@ -0,0 +1,15 @@
+load(libdir + "simd.js");
+
+setJitCompilerOption("ion.warmup.trigger", 50);
+
+function f() {
+    var b1 = SIMD.Bool32x4(true, false, true, false);
+    var b2 = SIMD.Bool32x4(true, true, true, true);
+    do {
+        assertEqX4(SIMD.Bool32x4.and(b1, b2), booleanBinaryX4((x, y) => x && y, b1, b2));
+        assertEqX4(SIMD.Bool32x4.or(b1, b2),  booleanBinaryX4((x, y) => x || y, b1, b2));
+        assertEqX4(SIMD.Bool32x4.xor(b1, b2), booleanBinaryX4((x, y) => x != y, b1, b2));
+    } while (!inIon());
+}
+
+f();

js/src/jit-test/tests/SIMD/bool32x4-const.js

@@ -0,0 +1,65 @@
+load(libdir + "simd.js");
+
+setJitCompilerOption("ion.warmup.trigger", 50);
+
+// Test constant folding into the Bool32x4 constructor.
+// Verify that we get the truthiness right, c.f. the ECMA ToBoolean() function.
+function f1() {
+    var B = SIMD.Bool32x4;
+    var S = SIMD.Bool32x4.splat;
+    return [
+        B(false, false, false, true),
+        B(true),
+        B(undefined, null, "", "x"),
+        B({}, 0, 1, -0.0),
+        B(NaN, -NaN, Symbol(), createIsHTMLDDA()),
+
+        S(false),
+        S(true),
+        S(undefined),
+        S(null),
+
+        S(""),
+        S("x"),
+        S(0),
+        S(1),
+
+        S({}),
+        S(-0.0),
+        S(NaN),
+        S(Symbol()),
+
+        S(createIsHTMLDDA())
+    ];
+}
+
+function f() {
+    for (var i = 0; i < 100; i++) {
+        var a = f1()
+        assertEqX4(a[0], [false, false, false, true]);
+        assertEqX4(a[1], [true,  false, false, false]);
+        assertEqX4(a[2], [false, false, false, true]);
+        assertEqX4(a[3], [true,  false, true,  false]);
+        assertEqX4(a[4], [false, false, true,  false]);
+
+        // Splats.
+        assertEqX4(a[5], [false, false, false, false]);
+        assertEqX4(a[6], [true, true, true, true]);
+        assertEqX4(a[7], [false, false, false, false]);
+        assertEqX4(a[8], [false, false, false, false]);
+
+        assertEqX4(a[9], [false, false, false, false]);
+        assertEqX4(a[10], [true, true, true, true]);
+        assertEqX4(a[11], [false, false, false, false]);
+        assertEqX4(a[12], [true, true, true, true]);
+
+        assertEqX4(a[13], [true, true, true, true]);
+        assertEqX4(a[14], [false, false, false, false]);
+        assertEqX4(a[15], [false, false, false, false]);
+        assertEqX4(a[16], [true, true, true, true]);
+
+        assertEqX4(a[17], [false, false, false, false]);
+    }
+}
+
+f();

js/src/jit-test/tests/SIMD/bug1109911.js

@@ -0,0 +1,11 @@
+if (typeof TypedObject === "undefined" || typeof SIMD === 'undefined')
+  quit();
+
+var Int32x4 = SIMD.Int32x4;
+var a = Int32x4((4294967295), 200, 300, 400);
+addCase( new Array(Math.pow(2,12)) );
+for ( var arg = "", i = 0; i < Math.pow(2,12); i++ ) {}
+addCase( a );
+function addCase(object) {
+  object.length
+}

js/src/jit-test/tests/SIMD/bug1121299.js

@@ -0,0 +1,31 @@
+if (!this.hasOwnProperty("SIMD"))
+  quit();
+
+setJitCompilerOption("baseline.warmup.trigger", 10);
+setJitCompilerOption("ion.warmup.trigger", 30);
+
+function test_1(i) {
+  if (i >= 40)
+    return;
+  var a = SIMD.Float32x4(1.1, 2.2, 3.3, 4.6);
+  SIMD.Int32x4.fromFloat32x4(a);
+  test_1(i + 1);
+}
+test_1(0);
+
+
+var Float32x4 = SIMD.Float32x4;
+function test_2() {
+    var Array = Float32x4.array(3);
+    var array = new Array([
+        Float32x4(1, 2, 3, 4),
+        Float32x4(5, 6, 7, 8),
+        Float32x4(9, 10, 11, 12)
+    ]);
+    if (typeof reportCompare === "function")
+        reportCompare(true, true);
+}
+test_2();
+evaluate("test_2(); test_2();", {
+    isRunOnce: true,
+});

js/src/jit-test/tests/SIMD/bug1123631.js

@@ -0,0 +1,9 @@
+if (!this.hasOwnProperty("SIMD"))
+  quit();
+
+var Float64x2 = SIMD.Float64x2;
+function test() {
+  var a = Float64x2(1, 2);
+}
+test();
+test();

js/src/jit-test/tests/SIMD/bug1130845.js

@@ -0,0 +1,15 @@
+if (!this.hasOwnProperty("SIMD"))
+  quit();
+
+var Int32x4 = SIMD.Int32x4;
+function test() {
+  var a = Int32x4();
+  var b = Int32x4(10, 20, 30, 40);
+  var c = SIMD.Int32x4.and(a, b);
+  assertEq(Int32x4.extractLane(c, 0), 0);
+  return 0;
+}
+test();
+var u = [], v = [];
+for (var j=0; j<u.length; ++j)
+    v[test()] = t;

js/src/jit-test/tests/SIMD/bug1241872.js

@@ -0,0 +1,10 @@
+if (typeof SIMD !== 'object')
+    quit(0);
+
+function test() {
+    return SIMD.Float32x4().toSource();
+}
+
+var r = '';
+for (var i = 0; i < 10000; i++)
+    r = test();

js/src/jit-test/tests/SIMD/bug1248503.js

@@ -0,0 +1,16 @@
+if (typeof SIMD !== 'object')
+    quit(0);
+
+function assertEqVec(v, w) {
+    [0].forEach(i => v, w);
+    function assertEqX4(...opts) {}
+}
+gczeal(1);
+function f() {
+    SIMD.Float32x4();
+    var i1 = SIMD.Int32x4();
+    for (j = 0; j < 100000; ++j, eval.eval)
+        assertEqVec(SIMD.Int32x4.check(i1), i1);
+}
+f();
+

js/src/jit-test/tests/SIMD/bug1273483.js

@@ -0,0 +1,9 @@
+if (typeof SIMD === 'undefined')
+    quit();
+
+Int8x16 = SIMD.Int8x16;
+var Int32x4 = SIMD.Int32x4;
+function testSwizzleForType(type) { return type(); }
+testSwizzleForType(Int8x16);
+function testSwizzleInt32x4() { return testSwizzleForType(Int32x4); }
+testSwizzleInt32x4();

js/src/jit-test/tests/SIMD/bug1296640-gc-args.js

@@ -0,0 +1,9 @@
+if (typeof gczeal === 'undefined' || typeof SIMD === 'undefined') {
+    quit();
+}
+
+gczeal(9, 2);
+var Int8x16 = SIMD.Int8x16;
+var v = Int8x16();
+var good = { valueOf: () => 21 };
+Int8x16.shiftLeftByScalar(v, good);

js/src/jit-test/tests/SIMD/bug1303780-gc-args.js

@@ -0,0 +1,12 @@
+if (typeof gczeal === 'undefined' || typeof SIMD === 'undefined') {
+    quit();
+}
+
+gczeal(14,2);
+var Float32x4 = SIMD.Float32x4;
+function test() {
+    var v = Float32x4(1,2,3,4);
+    var good = {valueOf: () => 42};
+    Float32x4.replaceLane(v, 0, good);
+}
+test();

js/src/jit-test/tests/SIMD/bug1435317.js

@@ -0,0 +1,25 @@
+load(libdir + 'simd.js');
+
+var ab = new ArrayBuffer(64 * 1024);
+var arr = new Uint8Array(ab);
+
+(function(glob, imp, b) {
+  "use asm";
+  var arr = new glob.Uint8Array(b);
+  return {}
+})(this, null, ab);
+
+function testSimdX4() {
+    for (var i = 10; i --> 0;) {
+        var caught;
+        try {
+            v = SIMD.Int32x4.load(arr, 65534);
+        } catch (e) {
+            caught = e;
+        }
+        assertEq(caught instanceof RangeError, true);
+    }
+}
+
+setJitCompilerOption('ion.warmup.trigger', 0);
+testSimdX4();

js/src/jit-test/tests/SIMD/bug953108.js

@@ -0,0 +1,10 @@
+/*
+ * Any copyright is dedicated to the Public Domain.
+ * http://creativecommons.org/licenses/publicdomain/
+ */
+
+if (!this.hasOwnProperty("TypedObject") || !this.hasOwnProperty("SIMD"))
+  quit();
+
+var Float32x4 = SIMD.Float32x4;
+Float32x4.array(1);

js/src/jit-test/tests/SIMD/check.js

@@ -0,0 +1,25 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 50);
+
+function f() {
+    var f1 = SIMD.Float32x4(1, 2, 3, 4);
+    var i1 = SIMD.Int32x4(1, 2, -3, 4);
+    var b1 = SIMD.Bool32x4(true, true, false, true);
+    var i = 0;
+    try {
+        for (; i < 150; i++) {
+            if (i > 148)
+                i1 = f1;
+            assertEqVec(SIMD.Int32x4.check(i1), i1);
+            assertEqVec(SIMD.Float32x4.check(f1), f1);
+            assertEqVec(SIMD.Bool32x4.check(b1), b1);
+        }
+    } catch (ex) {
+        assertEq(i, 149);
+        assertEq(ex instanceof TypeError, true);
+    }
+}
+
+f();
+

js/src/jit-test/tests/SIMD/compare.js

@@ -0,0 +1,39 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 50);
+
+function f() {
+    var f1 = SIMD.Float32x4(1, 2, 3, 4);
+    var f2 = SIMD.Float32x4(NaN, Infinity, 3.14, -0);
+
+    var i1 = SIMD.Int32x4(1, 2, -3, 4);
+    var i2 = SIMD.Int32x4(1, -2, 3, 0);
+
+    var u1 = SIMD.Uint32x4(1, 2, -3, 4);
+    var u2 = SIMD.Uint32x4(1, -2, 3, 0x80000000);
+
+    for (var i = 0; i < 150; i++) {
+        assertEqX4(SIMD.Int32x4.lessThan(i1, i2),             [false, false, true, false]);
+        assertEqX4(SIMD.Int32x4.lessThanOrEqual(i1, i2),      [true, false, true, false]);
+        assertEqX4(SIMD.Int32x4.equal(i1, i2),                [true, false, false, false]);
+        assertEqX4(SIMD.Int32x4.notEqual(i1, i2),             [false, true, true, true]);
+        assertEqX4(SIMD.Int32x4.greaterThan(i1, i2),          [false, true, false, true]);
+        assertEqX4(SIMD.Int32x4.greaterThanOrEqual(i1, i2),   [true, true, false, true]);
+
+        assertEqX4(SIMD.Uint32x4.lessThan(u1, u2),             [false, true, false, true]);
+        assertEqX4(SIMD.Uint32x4.lessThanOrEqual(u1, u2),      [true,  true, false, true]);
+        assertEqX4(SIMD.Uint32x4.equal(u1, u2),                [true, false, false, false]);
+        assertEqX4(SIMD.Uint32x4.notEqual(u1, u2),             [false, true, true, true]);
+        assertEqX4(SIMD.Uint32x4.greaterThan(u1, u2),          [false, false, true, false]);
+        assertEqX4(SIMD.Uint32x4.greaterThanOrEqual(u1, u2),   [true, false, true, false]);
+
+        assertEqX4(SIMD.Float32x4.lessThan(f1, f2),             [false, true, true, false]);
+        assertEqX4(SIMD.Float32x4.lessThanOrEqual(f1, f2),      [false, true, true, false]);
+        assertEqX4(SIMD.Float32x4.equal(f1, f2),                [false, false, false, false]);
+        assertEqX4(SIMD.Float32x4.notEqual(f1, f2),             [true, true, true, true]);
+        assertEqX4(SIMD.Float32x4.greaterThan(f1, f2),          [false, false, false, true]);
+        assertEqX4(SIMD.Float32x4.greaterThanOrEqual(f1, f2),   [false, false, false, true]);
+    }
+}
+
+f();

js/src/jit-test/tests/SIMD/complex-4.js

@@ -0,0 +1,70 @@
+load(libdir + 'simd.js');
+
+if (typeof SIMD === "undefined")
+    quit();
+
+setJitCompilerOption("baseline.warmup.trigger", 10);
+setJitCompilerOption("ion.warmup.trigger", 90);
+var max = 100; // Make have the warm-up counter high enough to
+               // consider inlining functions.
+
+var f4 = SIMD.Int32x4; // :TODO: Support Float32x4 arith.
+var f4add = f4.add;
+var f4sub = f4.sub;
+var f4mul = f4.mul;
+
+function c4mul(z1, z2) {
+  var { re: re1, im: im1 } = z1;
+  var { re: re2, im: im2 } = z2;
+  var rere = f4mul(re1, re2);
+  var reim = f4mul(re1, im2);
+  var imre = f4mul(im1, re2);
+  var imim = f4mul(im1, im2);
+  return { re: f4sub(rere, imim), im: f4add(reim, imre) };
+}
+
+function c4inv(z) {
+  var { re: re, im: im } = z;
+  var minus = f4(-1, -1, -1, -1);
+  return { re: re, im: f4mul(im, minus) };
+}
+
+function c4inv_inplace(z) {
+  var res = c4inv(z);
+  z.re = res.re;
+  z.im = res.im;
+}
+
+function c4norm(z) {
+  var { re: re, im: im } = c4mul(z, c4inv(z));
+  return re;
+}
+
+function c4scale(z, s) {
+  var { re: re, im: im } = z;
+  var f4s = f4(s, s, s, s);
+  return { re: f4mul(re, f4s), im: f4mul(im, f4s) };
+}
+
+var rotate90 = { re: f4(0, 0, 0, 0), im: f4(1, 1, 1, 1) };
+var cardinals = { re: f4(1, 0, -1, 0), im: f4(0, 1, 0, -1) };
+
+function test(dots) {
+  for (var j = 0; j < 4; j++) {
+    dots = c4mul(rotate90, dots);
+    if (j % 2 == 0) // Magic !
+      c4inv_inplace(dots);
+    dots = c4scale(dots, 2);
+  }
+  return dots;
+}
+
+assertEqX4(c4norm(cardinals), simdToArray(f4.splat(1)));
+var cardinals16 = c4scale(cardinals, 16);
+
+for (var i = 0; i < max; i++) {
+  var res = test(cardinals);
+  assertEqX4(c4norm(res), simdToArray(f4.splat(16 * 16)));
+  assertEqX4(res.re, simdToArray(cardinals16.re));
+  assertEqX4(res.im, simdToArray(cardinals16.im));
+}

js/src/jit-test/tests/SIMD/convert.js

@@ -0,0 +1,68 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 30);
+
+var cast = (function() {
+    var i32 = new Int32Array(1);
+    var f32 = new Float32Array(i32.buffer);
+    return {
+        fromInt32Bits(x) {
+            i32[0] = x;
+            return f32[0];
+        },
+
+        fromFloat32Bits(x) {
+            f32[0] = x;
+            return i32[0];
+        }
+    }
+})();
+
+function f() {
+    // No bailout here.
+    var f4 = SIMD.Float32x4(1, 2, 3, 4);
+    var i4 = SIMD.Int32x4(1, 2, 3, 4);
+    var BitOrZero = (x) => x | 0;
+    for (var i = 0; i < 150; i++) {
+        assertEqX4(SIMD.Float32x4.fromInt32x4(i4), unaryX4(BitOrZero, f4, Math.fround));
+        assertEqX4(SIMD.Float32x4.fromInt32x4Bits(i4), unaryX4(cast.fromInt32Bits, f4, Math.fround));
+        assertEqX4(SIMD.Int32x4.fromFloat32x4(f4), unaryX4(Math.fround, i4, BitOrZero));
+        assertEqX4(SIMD.Int32x4.fromFloat32x4Bits(f4), unaryX4(cast.fromFloat32Bits, i4, BitOrZero));
+    }
+}
+
+function uglyDuckling(val) {
+    // We bail out when i == 149 because the conversion will return
+    // 0x80000000 and the input actually wasn't in bounds.
+    val = Math.fround(val);
+    for (var i = 0; i < 150; i++) {
+        var caught = false;
+        try {
+            var v = SIMD.Float32x4(i < 149 ? 0 : val, 0, 0, 0)
+            SIMD.Int32x4.fromFloat32x4(v);
+        } catch(e) {
+            assertEq(e instanceof RangeError, true);
+            assertEq(i, 149);
+            caught = true;
+        }
+        assertEq(i < 149 || caught, true);
+    }
+}
+
+function dontBail() {
+    // On x86, the conversion will return 0x80000000, which will imply that we
+    // check the input values. However, we shouldn't bail out in this case.
+    for (var i = 0; i < 150; i++) {
+        var v = SIMD.Float32x4(i < 149 ? 0 : -Math.pow(2, 31), 0, 0, 0)
+        SIMD.Int32x4.fromFloat32x4(v);
+    }
+}
+
+f();
+
+dontBail();
+dontBail();
+
+uglyDuckling(Math.pow(2, 31));
+uglyDuckling(NaN);
+uglyDuckling(-Math.pow(2, 32));

js/src/jit-test/tests/SIMD/float32x4-binary-arith.js

@@ -0,0 +1,33 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 50);
+
+function maxNum(x, y) {
+    if (x != x)
+        return y;
+    if (y != y)
+        return x;
+    return Math.max(x, y);
+}
+
+function minNum(x, y) {
+    if (x != x)
+        return y;
+    if (y != y)
+        return x;
+    return Math.min(x, y);
+}
+
+function f() {
+    var f1 = SIMD.Float32x4(1, 2, 3, 4);
+    var f2 = SIMD.Float32x4(4, 3, 2, 1);
+    for (var i = 0; i < 150; i++) {
+        assertEqX4(SIMD.Float32x4.div(f1, f2), binaryX((x, y) => x / y, f1, f2));
+        assertEqX4(SIMD.Float32x4.min(f1, f2), binaryX(Math.min, f1, f2));
+        assertEqX4(SIMD.Float32x4.max(f1, f2), binaryX(Math.max, f1, f2));
+        assertEqX4(SIMD.Float32x4.minNum(f1, f2), binaryX(minNum, f1, f2));
+        assertEqX4(SIMD.Float32x4.maxNum(f1, f2), binaryX(maxNum, f1, f2));
+    }
+}
+
+f();

js/src/jit-test/tests/SIMD/getters.js

@@ -0,0 +1,48 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 50);
+
+function f() {
+    var i4 = SIMD.Int32x4(1, -2, 3, -4);
+    var u4 = SIMD.Uint32x4(1, -2, 3, 0x88000000);
+    var b4 = SIMD.Bool32x4(true, true, false, true);
+
+
+    var bt4 = SIMD.Bool32x4(true, true, true, true);
+    var bf4 = SIMD.Bool32x4(false, false, false, false);
+
+    var v = Math.fround(13.37);
+    var f4 = SIMD.Float32x4(13.37, NaN, Infinity, -0);
+
+    for (var i = 0; i < 150; i++) {
+        assertEq(SIMD.Int32x4.extractLane(i4, 0), 1);
+        assertEq(SIMD.Int32x4.extractLane(i4, 1), -2);
+        assertEq(SIMD.Int32x4.extractLane(i4, 2), 3);
+        assertEq(SIMD.Int32x4.extractLane(i4, 3), -4);
+
+        assertEq(SIMD.Uint32x4.extractLane(u4, 0), 1);
+        assertEq(SIMD.Uint32x4.extractLane(u4, 1), -2 >>> 0);
+        assertEq(SIMD.Uint32x4.extractLane(u4, 2), 3);
+        assertEq(SIMD.Uint32x4.extractLane(u4, 3), 0x88000000);
+
+        assertEq(SIMD.Float32x4.extractLane(f4, 0), v);
+        assertEq(SIMD.Float32x4.extractLane(f4, 1), NaN);
+        assertEq(SIMD.Float32x4.extractLane(f4, 2), Infinity);
+        assertEq(SIMD.Float32x4.extractLane(f4, 3), -0);
+
+        assertEq(SIMD.Bool32x4.extractLane(b4, 0), true);
+        assertEq(SIMD.Bool32x4.extractLane(b4, 1), true);
+        assertEq(SIMD.Bool32x4.extractLane(b4, 2), false);
+        assertEq(SIMD.Bool32x4.extractLane(b4, 3), true);
+
+        assertEq(SIMD.Bool32x4.anyTrue(b4), true);
+        assertEq(SIMD.Bool32x4.allTrue(b4), false);
+
+        assertEq(SIMD.Bool32x4.anyTrue(bt4), true);
+        assertEq(SIMD.Bool32x4.allTrue(bt4), true);
+        assertEq(SIMD.Bool32x4.anyTrue(bf4), false);
+        assertEq(SIMD.Bool32x4.allTrue(bf4), false);
+    }
+}
+
+f();

js/src/jit-test/tests/SIMD/inline-missing-arguments.js

@@ -0,0 +1,81 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 50);
+
+function test(i) {
+    assertEqX4(SIMD.Int32x4(),              [0, 0, 0, 0]);
+    assertEqX4(SIMD.Int32x4(i),             [i, 0, 0, 0]);
+    assertEqX4(SIMD.Int32x4(i, 1),          [i, 1, 0, 0]);
+    assertEqX4(SIMD.Int32x4(i, 1, 2),       [i, 1, 2, 0]);
+    assertEqX4(SIMD.Int32x4(i, 1, 2, 3),    [i, 1, 2, 3]);
+    assertEqX4(SIMD.Int32x4(i, 1, 2, 3, 4), [i, 1, 2, 3]);
+
+    assertEqVecArr(SIMD.Int16x8(),              [0, 0, 0, 0, 0, 0, 0, 0]);
+    assertEqVecArr(SIMD.Int16x8(i),             [i, 0, 0, 0, 0, 0, 0, 0]);
+    assertEqVecArr(SIMD.Int16x8(i, 1),          [i, 1, 0, 0, 0, 0, 0, 0]);
+    assertEqVecArr(SIMD.Int16x8(i, 1, 2),       [i, 1, 2, 0, 0, 0, 0, 0]);
+    assertEqVecArr(SIMD.Int16x8(i, 1, 2, 3),    [i, 1, 2, 3, 0, 0, 0, 0]);
+    assertEqVecArr(SIMD.Int16x8(i, 1, 2, 3, 4), [i, 1, 2, 3, 4, 0, 0, 0]);
+    assertEqVecArr(SIMD.Int16x8(i, 1, 2, 3, 4, 5, 6),
+                               [i, 1, 2, 3, 4, 5, 6, 0]);
+    j = i & 32
+    assertEqVecArr(SIMD.Int8x16(),              [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]);
+    assertEqVecArr(SIMD.Int8x16(j),             [j, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]);
+    assertEqVecArr(SIMD.Int8x16(j, 1),          [j, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]);
+    assertEqVecArr(SIMD.Int8x16(j, 1, 2),       [j, 1, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]);
+    assertEqVecArr(SIMD.Int8x16(j, 1, 2, 3),    [j, 1, 2, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]);
+    assertEqVecArr(SIMD.Int8x16(j, 1, 2, 3, 4), [j, 1, 2, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]);
+    assertEqVecArr(SIMD.Int8x16(j, 1, 2, 3, 4, 5, 6),
+                               [j, 1, 2, 3, 4, 5, 6, 0, 0, 0, 0, 0, 0, 0, 0, 0]);
+    assertEqVecArr(SIMD.Int8x16(j, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12),
+                               [j, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 0, 0, 0]);
+
+    assertEqX4(SIMD.Float32x4(),                [NaN, NaN, NaN, NaN]);
+    assertEqX4(SIMD.Float32x4(i),               [i,   NaN, NaN, NaN]);
+    assertEqX4(SIMD.Float32x4(i, 1),            [i,   1,   NaN, NaN]);
+    assertEqX4(SIMD.Float32x4(i, 1, 2),         [i,   1,   2,   NaN]);
+    assertEqX4(SIMD.Float32x4(i, 1, 2, 3),      [i,   1,   2,   3  ]);
+    assertEqX4(SIMD.Float32x4(i, 1, 2, 3, 4),   [i,   1,   2,   3  ]);
+
+    var b = i % 2 > 0 ;
+    assertEqX4(SIMD.Bool32x4(),                           [false, false, false, false]);
+    assertEqX4(SIMD.Bool32x4(b),                          [b,     false, false, false]);
+    assertEqX4(SIMD.Bool32x4(b, true),                    [b,     true,  false, false]);
+    assertEqX4(SIMD.Bool32x4(b, false, true),             [b,     false, true,  false]);
+    assertEqX4(SIMD.Bool32x4(b, false, true, true),       [b,     false, true,  true ]);
+    assertEqX4(SIMD.Bool32x4(b, false, true, true, true), [b,     false, true,  true ]);
+
+    assertEqVecArr(SIMD.Bool16x8(),
+                                [false, false, false, false, false, false, false, false]);
+    assertEqVecArr(SIMD.Bool16x8(b),
+                                [b,     false, false, false, false, false, false, false]);
+    assertEqVecArr(SIMD.Bool16x8(b,     true),
+                                [b,     true,  false, false, false, false, false, false]);
+    assertEqVecArr(SIMD.Bool16x8(b,     false, true),
+                                [b,     false, true,  false, false, false, false, false]);
+    assertEqVecArr(SIMD.Bool16x8(b,     false, true,  true),
+                                [b,     false, true,  true,  false, false, false, false]);
+    assertEqVecArr(SIMD.Bool16x8(b,     false, true,  true,  true),
+                                [b,     false, true,  true,  true,  false, false, false]);
+    assertEqVecArr(SIMD.Bool16x8(b,     false, true,  true,  true,  true),
+                                [b,     false, true,  true,  true,  true,  false, false]);
+
+    assertEqVecArr(SIMD.Bool8x16(),
+                                [false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false]);
+    assertEqVecArr(SIMD.Bool8x16(b),
+                                [b,     false, false, false, false, false, false, false, false, false, false, false, false, false, false, false]);
+    assertEqVecArr(SIMD.Bool8x16(b,     true),
+                                [b,     true,  false, false, false, false, false, false, false, false, false, false, false, false, false, false]);
+    assertEqVecArr(SIMD.Bool8x16(b,     false, true),
+                                [b,     false, true,  false, false, false, false, false, false, false, false, false, false, false, false, false]);
+    assertEqVecArr(SIMD.Bool8x16(b,     false, true,  true),
+                                [b,     false, true,  true,  false, false, false, false, false, false, false, false, false, false, false, false]);
+    assertEqVecArr(SIMD.Bool8x16(b,     false, true,  true,  true),
+                                [b,     false, true,  true,  true,  false, false, false, false, false, false, false, false, false, false, false]);
+    assertEqVecArr(SIMD.Bool8x16(b,     false, true,  true,  true,  true,  false, true,  true,  true),
+                                [b,     false, true,  true,  true,  true,  false, true,  true,  true,  false, false, false, false, false, false]);
+}
+
+for(var i=0; i<300; i++) {
+    test(i);
+}

js/src/jit-test/tests/SIMD/load.js

@@ -0,0 +1,123 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 40);
+
+function f() {
+    var f32 = new Float32Array(16);
+    for (var i = 0; i < 16; i++)
+        f32[i] = i + 1;
+
+    var f64 = new Float64Array(f32.buffer);
+    var i32 = new Int32Array(f32.buffer);
+    var u32 = new Uint32Array(f32.buffer);
+    var i16 = new Int16Array(f32.buffer);
+    var u16 = new Uint16Array(f32.buffer);
+    var i8  = new Int8Array(f32.buffer);
+    var u8  = new Uint8Array(f32.buffer);
+
+    function testLoad() {
+        assertEqX4(SIMD.Float32x4.load(f64, 0),      [1,2,3,4]);
+        assertEqX4(SIMD.Float32x4.load(f32, 1),      [2,3,4,5]);
+        assertEqX4(SIMD.Float32x4.load(i32, 2),      [3,4,5,6]);
+        assertEqX4(SIMD.Float32x4.load(i16, 3 << 1), [4,5,6,7]);
+        assertEqX4(SIMD.Float32x4.load(u16, 4 << 1), [5,6,7,8]);
+        assertEqX4(SIMD.Float32x4.load(i8 , 5 << 2), [6,7,8,9]);
+        assertEqX4(SIMD.Float32x4.load(u8 , 6 << 2), [7,8,9,10]);
+
+        assertEqX4(SIMD.Float32x4.load(f64, (16 >> 1) - (4 >> 1)), [13,14,15,16]);
+        assertEqX4(SIMD.Float32x4.load(f32, 16 - 4),               [13,14,15,16]);
+        assertEqX4(SIMD.Float32x4.load(i32, 16 - 4),               [13,14,15,16]);
+        assertEqX4(SIMD.Float32x4.load(i16, (16 << 1) - (4 << 1)), [13,14,15,16]);
+        assertEqX4(SIMD.Float32x4.load(u16, (16 << 1) - (4 << 1)), [13,14,15,16]);
+        assertEqX4(SIMD.Float32x4.load(i8,  (16 << 2) - (4 << 2)), [13,14,15,16]);
+        assertEqX4(SIMD.Float32x4.load(u8,  (16 << 2) - (4 << 2)), [13,14,15,16]);
+    }
+
+    function testLoad1() {
+        assertEqX4(SIMD.Float32x4.load1(f64, 0),      [1,0,0,0]);
+        assertEqX4(SIMD.Float32x4.load1(f32, 1),      [2,0,0,0]);
+        assertEqX4(SIMD.Float32x4.load1(i32, 2),      [3,0,0,0]);
+        assertEqX4(SIMD.Float32x4.load1(i16, 3 << 1), [4,0,0,0]);
+        assertEqX4(SIMD.Float32x4.load1(u16, 4 << 1), [5,0,0,0]);
+        assertEqX4(SIMD.Float32x4.load1(i8 , 5 << 2), [6,0,0,0]);
+        assertEqX4(SIMD.Float32x4.load1(u8 , 6 << 2), [7,0,0,0]);
+
+        assertEqX4(SIMD.Float32x4.load1(f64, (16 >> 1) - (4 >> 1)), [13,0,0,0]);
+        assertEqX4(SIMD.Float32x4.load1(f32, 16 - 4),               [13,0,0,0]);
+        assertEqX4(SIMD.Float32x4.load1(i32, 16 - 4),               [13,0,0,0]);
+        assertEqX4(SIMD.Float32x4.load1(i16, (16 << 1) - (4 << 1)), [13,0,0,0]);
+        assertEqX4(SIMD.Float32x4.load1(u16, (16 << 1) - (4 << 1)), [13,0,0,0]);
+        assertEqX4(SIMD.Float32x4.load1(i8,  (16 << 2) - (4 << 2)), [13,0,0,0]);
+        assertEqX4(SIMD.Float32x4.load1(u8,  (16 << 2) - (4 << 2)), [13,0,0,0]);
+    }
+
+    function testLoad2() {
+        assertEqX4(SIMD.Float32x4.load2(f64, 0),      [1,2,0,0]);
+        assertEqX4(SIMD.Float32x4.load2(f32, 1),      [2,3,0,0]);
+        assertEqX4(SIMD.Float32x4.load2(i32, 2),      [3,4,0,0]);
+        assertEqX4(SIMD.Float32x4.load2(i16, 3 << 1), [4,5,0,0]);
+        assertEqX4(SIMD.Float32x4.load2(u16, 4 << 1), [5,6,0,0]);
+        assertEqX4(SIMD.Float32x4.load2(i8 , 5 << 2), [6,7,0,0]);
+        assertEqX4(SIMD.Float32x4.load2(u8 , 6 << 2), [7,8,0,0]);
+
+        assertEqX4(SIMD.Float32x4.load2(f64, (16 >> 1) - (4 >> 1)), [13,14,0,0]);
+        assertEqX4(SIMD.Float32x4.load2(f32, 16 - 4),               [13,14,0,0]);
+        assertEqX4(SIMD.Float32x4.load2(i32, 16 - 4),               [13,14,0,0]);
+        assertEqX4(SIMD.Float32x4.load2(i16, (16 << 1) - (4 << 1)), [13,14,0,0]);
+        assertEqX4(SIMD.Float32x4.load2(u16, (16 << 1) - (4 << 1)), [13,14,0,0]);
+        assertEqX4(SIMD.Float32x4.load2(i8,  (16 << 2) - (4 << 2)), [13,14,0,0]);
+        assertEqX4(SIMD.Float32x4.load2(u8,  (16 << 2) - (4 << 2)), [13,14,0,0]);
+    }
+
+    function testLoad3() {
+        assertEqX4(SIMD.Float32x4.load3(f64, 0),      [1,2,3,0]);
+        assertEqX4(SIMD.Float32x4.load3(f32, 1),      [2,3,4,0]);
+        assertEqX4(SIMD.Float32x4.load3(i32, 2),      [3,4,5,0]);
+        assertEqX4(SIMD.Float32x4.load3(i16, 3 << 1), [4,5,6,0]);
+        assertEqX4(SIMD.Float32x4.load3(u16, 4 << 1), [5,6,7,0]);
+        assertEqX4(SIMD.Float32x4.load3(i8 , 5 << 2), [6,7,8,0]);
+        assertEqX4(SIMD.Float32x4.load3(u8 , 6 << 2), [7,8,9,0]);
+
+        assertEqX4(SIMD.Float32x4.load3(f64, (16 >> 1) - (4 >> 1)), [13,14,15,0]);
+        assertEqX4(SIMD.Float32x4.load3(f32, 16 - 4),               [13,14,15,0]);
+        assertEqX4(SIMD.Float32x4.load3(i32, 16 - 4),               [13,14,15,0]);
+        assertEqX4(SIMD.Float32x4.load3(i16, (16 << 1) - (4 << 1)), [13,14,15,0]);
+        assertEqX4(SIMD.Float32x4.load3(u16, (16 << 1) - (4 << 1)), [13,14,15,0]);
+        assertEqX4(SIMD.Float32x4.load3(i8,  (16 << 2) - (4 << 2)), [13,14,15,0]);
+        assertEqX4(SIMD.Float32x4.load3(u8,  (16 << 2) - (4 << 2)), [13,14,15,0]);
+    }
+
+    for (var i = 0; i < 150; i++) {
+        testLoad();
+        testLoad1();
+        testLoad2();
+        testLoad3();
+    }
+}
+
+f();
+
+function testBailout(uglyDuckling) {
+    var f32 = new Float32Array(16);
+    for (var i = 0; i < 16; i++)
+        f32[i] = i + 1;
+
+    var i8  = new Int8Array(f32.buffer);
+
+    for (var i = 0; i < 150; i++) {
+        var caught = false;
+        try {
+            SIMD.Float32x4.load(i8, (i < 149) ? 0 : uglyDuckling);
+        } catch (e) {
+            print(e);
+            assertEq(e instanceof RangeError, true);
+            caught = true;
+        }
+        assertEq(i < 149 || caught, true);
+    }
+}
+
+print('Testing range checks...');
+testBailout(-1);
+testBailout(-15);
+testBailout(12 * 4 + 1);

js/src/jit-test/tests/SIMD/nursery-overflow.js

@@ -0,0 +1,29 @@
+load(libdir + 'simd.js');
+
+if (typeof SIMD === "undefined")
+    quit();
+
+setJitCompilerOption("baseline.warmup.trigger", 10);
+setJitCompilerOption("ion.warmup.trigger", 30);
+
+var i4 = SIMD.Int32x4;
+var i4sub = SIMD.Int32x4.sub;
+
+function simdbox(i) {
+  return i4(i, i, i, i);
+}
+
+function test() {
+  var arr = [];
+
+  // overflow the nursery with live SIMD objects.
+  for (var i = 0; i < 100000; i++) {
+    arr.push(simdbox(i));
+  }
+
+  return arr;
+}
+
+var arr = test();
+for (var i = 0; i < arr.length; i++)
+  assertEqX4(arr[i], [i, i, i, i]);

js/src/jit-test/tests/SIMD/recover.js

@@ -0,0 +1,70 @@
+load(libdir + 'simd.js');
+
+if (!this.hasOwnProperty("SIMD"))
+  quit();
+
+// This test case ensure that if we are able to optimize SIMD, then we can use
+// recover instructions to get rid of the allocations. So, there is no value
+// (and the test case would fail) if we are not able to inline SIMD
+// constructors.
+if (!isSimdAvailable())
+  quit();
+
+setJitCompilerOption("baseline.warmup.trigger", 10);
+setJitCompilerOption("ion.warmup.trigger", 20);
+
+// This function is used to cause an invalidation after having removed a branch
+// after DCE.  This is made to check if we correctly recover an array
+// allocation.
+var uceFault = function (i) {
+    if (i > 98)
+        uceFault = function (i) { return true; };
+    return false;
+};
+
+// Check that we can correctly recover a boxed value.
+var uceFault_simdBox_i4 = eval(uneval(uceFault).replace('uceFault', 'uceFault_simdBox_i4'));
+function simdBox_i4(i) {
+    var a = SIMD.Int32x4(i, i, i, i);
+    if (uceFault_simdBox_i4(i) || uceFault_simdBox_i4(i))
+        assertEqX4(a, [i, i, i, i]);
+    assertRecoveredOnBailout(a, true);
+    return 0;
+}
+
+var uceFault_simdBox_u4 = eval(uneval(uceFault).replace('uceFault', 'uceFault_simdBox_u4'));
+function simdBox_u4(i) {
+    var a = SIMD.Uint32x4(i, 98 - i, i + 0x7ffffff0, i + 0xffffff00);
+    if (uceFault_simdBox_u4(i) || uceFault_simdBox_u4(i))
+        assertEqX4(a, [i, 98 - i, i + 0x7ffffff0, i + 0xffffff00].map(x => x >>> 0));
+    assertRecoveredOnBailout(a, true);
+    return 0;
+}
+
+var uceFault_simdBox_f4 = eval(uneval(uceFault).replace('uceFault', 'uceFault_simdBox_f4'));
+function simdBox_f4(i) {
+    var a = SIMD.Float32x4(i, i + 0.1, i + 0.2, i + 0.3);
+    if (uceFault_simdBox_f4(i) || uceFault_simdBox_f4(i))
+        assertEqX4(a, [i, i + 0.1, i + 0.2, i + 0.3].map(Math.fround));
+    assertRecoveredOnBailout(a, true);
+    return 0;
+}
+
+var uceFault_simdBox_b4 = eval(uneval(uceFault).replace('uceFault', 'uceFault_simdBox_b4'));
+function simdBox_b4(i) {
+    var val1 = i%2 === 0,
+        val2 = !val1;
+
+    var a = SIMD.Bool32x4(val1, val2, val1, val2);
+    if (uceFault_simdBox_b4(i) || uceFault_simdBox_b4(i))
+        assertEqX4(a, [val1, val2, val1, val2]);
+    assertRecoveredOnBailout(a, true);
+    return 0;
+}
+
+for (var i = 0; i < 100; i++) {
+    simdBox_i4(i);
+    simdBox_u4(i);
+    simdBox_f4(i);
+    simdBox_b4(i);
+}

js/src/jit-test/tests/SIMD/replacelane.js

@@ -0,0 +1,181 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 50);
+
+function f() {
+    var f4 = SIMD.Float32x4(1, 2, 3, 4);
+    var i4 = SIMD.Int32x4(1, 2, 3, 4);
+    var b4 = SIMD.Bool32x4(true, false, true, false);
+
+    for (var i = 0; i < 150; i++) {
+        assertEqX4(SIMD.Int32x4.replaceLane(i4, 0, 42), [42, 2, 3, 4]);
+        assertEqX4(SIMD.Int32x4.replaceLane(i4, 1, 42), [1, 42, 3, 4]);
+        assertEqX4(SIMD.Int32x4.replaceLane(i4, 2, 42), [1, 2, 42, 4]);
+        assertEqX4(SIMD.Int32x4.replaceLane(i4, 3, 42), [1, 2, 3, 42]);
+
+        assertEqX4(SIMD.Float32x4.replaceLane(f4, 0, 42), [42, 2, 3, 4]);
+        assertEqX4(SIMD.Float32x4.replaceLane(f4, 1, 42), [1, 42, 3, 4]);
+        assertEqX4(SIMD.Float32x4.replaceLane(f4, 2, 42), [1, 2, 42, 4]);
+        assertEqX4(SIMD.Float32x4.replaceLane(f4, 3, 42), [1, 2, 3, 42]);
+
+        assertEqX4(SIMD.Bool32x4.replaceLane(b4, 0, false), [false, false, true, false]);
+        assertEqX4(SIMD.Bool32x4.replaceLane(b4, 1, true), [true, true, true, false]);
+        assertEqX4(SIMD.Bool32x4.replaceLane(b4, 2, false), [true, false, false, false]);
+        assertEqX4(SIMD.Bool32x4.replaceLane(b4, 3, true), [true, false, true, true]);
+    }
+}
+
+f();
+
+function e() {
+    var f4 = SIMD.Float32x4(1, 2, 3, 4);
+    var i4 = SIMD.Int32x4(1, 2, 3, 4);
+    var b4 = SIMD.Bool32x4(true, false, true, false);
+
+    for (let i = 0; i < 150; i++) {
+        let caught = false;
+        try {
+            let x = SIMD.Int32x4.replaceLane(i < 149 ? i4 : f4, 0, 42);
+        } catch(e) {
+            assertEq(e instanceof TypeError, true);
+            assertEq(i, 149);
+            caught = true;
+        }
+        assertEq(i < 149 || caught, true);
+    }
+
+    for (let i = 0; i < 150; i++) {
+        let caught = false;
+        try {
+            let x = SIMD.Int32x4.replaceLane(i < 149 ? i4 : b4, 0, 42);
+        } catch(e) {
+            assertEq(e instanceof TypeError, true);
+            assertEq(i, 149);
+            caught = true;
+        }
+        assertEq(i < 149 || caught, true);
+    }
+
+    for (let i = 0; i < 150; i++) {
+        let caught = false;
+        try {
+            let x = SIMD.Int32x4.replaceLane(i4, i < 149 ? 0 : 4, 42);
+        } catch(e) {
+            assertEq(e instanceof RangeError, true);
+            assertEq(i, 149);
+            caught = true;
+        }
+        assertEq(i < 149 || caught, true);
+    }
+
+    for (let i = 0; i < 150; i++) {
+        let caught = false;
+        try {
+            let x = SIMD.Int32x4.replaceLane(i4, i < 149 ? 0 : 1.1, 42);
+        } catch(e) {
+            assertEq(e instanceof RangeError, true);
+            assertEq(i, 149);
+            caught = true;
+        }
+        assertEq(i < 149 || caught, true);
+    }
+
+    for (let i = 0; i < 150; i++) {
+        let caught = false;
+        try {
+            let x = SIMD.Float32x4.replaceLane(i < 149 ? f4 : i4, 0, 42);
+        } catch(e) {
+            assertEq(e instanceof TypeError, true);
+            assertEq(i, 149);
+            caught = true;
+        }
+        assertEq(i < 149 || caught, true);
+    }
+
+    for (let i = 0; i < 150; i++) {
+        let caught = false;
+        try {
+            let x = SIMD.Float32x4.replaceLane(i < 149 ? f4 : b4, 0, 42);
+        } catch(e) {
+            assertEq(e instanceof TypeError, true);
+            assertEq(i, 149);
+            caught = true;
+        }
+        assertEq(i < 149 || caught, true);
+    }
+
+    for (let i = 0; i < 150; i++) {
+        let caught = false;
+        try {
+            let x = SIMD.Float32x4.replaceLane(f4, i < 149 ? 0 : 4, 42);
+        } catch(e) {
+            assertEq(e instanceof RangeError, true);
+            assertEq(i, 149);
+            caught = true;
+        }
+        assertEq(i < 149 || caught, true);
+    }
+
+    for (let i = 0; i < 150; i++) {
+        let caught = false;
+        try {
+            let x = SIMD.Float32x4.replaceLane(f4, i < 149 ? 0 : 1.1, 42);
+        } catch(e) {
+            assertEq(e instanceof RangeError, true);
+            assertEq(i, 149);
+            caught = true;
+        }
+        assertEq(i < 149 || caught, true);
+    }
+
+    for (let i = 0; i < 150; i++) {
+        let caught = false;
+        try {
+            let x = SIMD.Bool32x4.replaceLane(i < 149 ? b4 : i4, 0, true);
+        } catch(e) {
+            assertEq(e instanceof TypeError, true);
+            assertEq(i, 149);
+            caught = true;
+        }
+        assertEq(i < 149 || caught, true);
+    }
+
+    for (let i = 0; i < 150; i++) {
+        let caught = false;
+        try {
+            let x = SIMD.Bool32x4.replaceLane(i < 149 ? b4 : f4, 0, true);
+        } catch(e) {
+            assertEq(e instanceof TypeError, true);
+            assertEq(i, 149);
+            caught = true;
+        }
+        assertEq(i < 149 || caught, true);
+    }
+
+    for (let i = 0; i < 150; i++) {
+        let caught = false;
+        try {
+            let x = SIMD.Bool32x4.replaceLane(b4, i < 149 ? 0 : 4, true);
+        } catch(e) {
+            assertEq(e instanceof RangeError, true);
+            assertEq(i, 149);
+            caught = true;
+        }
+        assertEq(i < 149 || caught, true);
+    }
+
+    for (let i = 0; i < 150; i++) {
+        let caught = false;
+        try {
+            let x = SIMD.Bool32x4.replaceLane(b4, i < 149 ? 0 : 1.1, true);
+        } catch(e) {
+            assertEq(e instanceof RangeError, true);
+            assertEq(i, 149);
+            caught = true;
+        }
+        assertEq(i < 149 || caught, true);
+    }
+
+}
+
+e();

js/src/jit-test/tests/SIMD/saturate.js

@@ -0,0 +1,37 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 50);
+
+const INT8_MIN = -128;
+const INT8_MAX = 127;
+const UINT8_MAX = 255;
+
+function sat8(x) {
+    if (x < INT8_MIN) return INT8_MIN;
+    if (x > INT8_MAX) return INT8_MAX;
+    return x;
+}
+
+function usat8(x) {
+    if (x < 0) return 0;
+    if (x > UINT8_MAX) return UINT8_MAX;
+    return x;
+}
+
+function f() {
+    var i1 = SIMD.Int8x16(1, 100, 3, 4);
+    var i2 = SIMD.Int8x16(4, 30, 2, 1);
+
+    var u1 = SIMD.Uint8x16(1, 2, 3, 4);
+    var u2 = SIMD.Uint8x16(4, 3, 2, 1);
+
+    for (var i = 0; i < 150; i++) {
+        assertEqX4(SIMD.Int8x16.addSaturate(i1, i2), binaryX((x, y) => sat8(x + y), i1, i2));
+        assertEqX4(SIMD.Int8x16.subSaturate(i1, i2), binaryX((x, y) => sat8(x - y), i1, i2));
+
+        assertEqX4(SIMD.Uint8x16.addSaturate(u1, u2), binaryX((x, y) => usat8(x + y), u1, u2));
+        assertEqX4(SIMD.Uint8x16.subSaturate(u1, u2), binaryX((x, y) => usat8(x - y), u1, u2));
+    }
+}
+
+f();

js/src/jit-test/tests/SIMD/select.js

@@ -0,0 +1,35 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 50);
+
+function select(type, mask, ifTrue, ifFalse) {
+    var arr = [];
+    for (var i = 0; i < 4; i++) {
+        var selector = SIMD.Bool32x4.extractLane(mask, i);
+        arr.push(type.extractLane(selector ? ifTrue : ifFalse, i));
+    }
+    return arr;
+}
+
+function f() {
+    var f1 = SIMD.Float32x4(1, 2, 3, 4);
+    var f2 = SIMD.Float32x4(NaN, Infinity, 3.14, -0);
+
+    var i1 = SIMD.Int32x4(2, 3, 5, 8);
+    var i2 = SIMD.Int32x4(13, 37, 24, 42);
+
+    var TTFT = SIMD.Bool32x4(true, true, false, true);
+    var TFTF = SIMD.Bool32x4(true, false, true, false);
+
+    var mask = SIMD.Int32x4(0xdeadbeef, 0xbaadf00d, 0x00ff1ce, 0xdeadc0de);
+
+    for (var i = 0; i < 150; i++) {
+        assertEqX4(SIMD.Float32x4.select(TTFT, f1, f2), select(SIMD.Float32x4, TTFT, f1, f2));
+        assertEqX4(SIMD.Float32x4.select(TFTF, f1, f2), select(SIMD.Float32x4, TFTF, f1, f2));
+
+        assertEqX4(SIMD.Int32x4.select(TFTF, i1, i2), select(SIMD.Int32x4, TFTF, i1, i2));
+        assertEqX4(SIMD.Int32x4.select(TTFT, i1, i2), select(SIMD.Int32x4, TTFT, i1, i2));
+    }
+}
+
+f();

js/src/jit-test/tests/SIMD/shift.js

@@ -0,0 +1,75 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 50);
+
+function curry(f, arg) { return f.bind(null, arg); }
+
+function binaryLsh(count, v) { count &= 31; return (v << count) | 0; }
+function lsh(count) { return curry(binaryLsh, count); }
+
+function binaryRsh(count, v) { count &= 31; return (v >> count) | 0; }
+function rsh(count) { return curry(binaryRsh, count); }
+
+function binaryUlsh(count, v) { count &= 31; return (v << count) >>> 0; }
+function ulsh(count) { return curry(binaryUlsh, count); }
+
+function binaryUrsh(count, v) { count &= 31; return v >>> count; }
+function ursh(count) { return curry(binaryUrsh, count); }
+
+function f() {
+    var v = SIMD.Int32x4(1, 2, -3, 4);
+    var u = SIMD.Uint32x4(1, 0x55005500, -3, 0xaa00aa00);
+    var a = [1, 2, -3, 4];
+    var b = [1, 0x55005500, -3, 0xaa00aa00];
+
+    var shifts = [-2, -1, 0, 1, 31, 32, 33];
+
+    var r;
+    for (var i = 0; i < 150; i++) {
+        // Constant shift counts
+        assertEqX4(SIMD.Int32x4.shiftLeftByScalar(v, -1), a.map(lsh(-1)));
+        assertEqX4(SIMD.Int32x4.shiftLeftByScalar(v, 0),  a.map(lsh(0)));
+        assertEqX4(SIMD.Int32x4.shiftLeftByScalar(v, 1),  a.map(lsh(1)));
+        assertEqX4(SIMD.Int32x4.shiftLeftByScalar(v, 2),  a.map(lsh(2)));
+        assertEqX4(SIMD.Int32x4.shiftLeftByScalar(v, 31), a.map(lsh(31)));
+        assertEqX4(SIMD.Int32x4.shiftLeftByScalar(v, 32), a.map(lsh(32)));
+        assertEqX4(SIMD.Int32x4.shiftLeftByScalar(v, 33), a.map(lsh(33)));
+
+        assertEqX4(SIMD.Int32x4.shiftRightByScalar(v, -1), a.map(rsh(31)));
+        assertEqX4(SIMD.Int32x4.shiftRightByScalar(v, 0),  a.map(rsh(0)));
+        assertEqX4(SIMD.Int32x4.shiftRightByScalar(v, 1),  a.map(rsh(1)));
+        assertEqX4(SIMD.Int32x4.shiftRightByScalar(v, 2),  a.map(rsh(2)));
+        assertEqX4(SIMD.Int32x4.shiftRightByScalar(v, 31), a.map(rsh(31)));
+        assertEqX4(SIMD.Int32x4.shiftRightByScalar(v, 32), a.map(rsh(32)));
+        assertEqX4(SIMD.Int32x4.shiftRightByScalar(v, 33), a.map(rsh(33)));
+
+        assertEqX4(SIMD.Uint32x4.shiftLeftByScalar(u, -1), b.map(ulsh(-1)));
+        assertEqX4(SIMD.Uint32x4.shiftLeftByScalar(u, 0),  b.map(ulsh(0)));
+        assertEqX4(SIMD.Uint32x4.shiftLeftByScalar(u, 1),  b.map(ulsh(1)));
+        assertEqX4(SIMD.Uint32x4.shiftLeftByScalar(u, 2),  b.map(ulsh(2)));
+        assertEqX4(SIMD.Uint32x4.shiftLeftByScalar(u, 31), b.map(ulsh(31)));
+        assertEqX4(SIMD.Uint32x4.shiftLeftByScalar(u, 32), b.map(ulsh(32)));
+        assertEqX4(SIMD.Uint32x4.shiftLeftByScalar(u, 33), b.map(ulsh(33)));
+
+        assertEqX4(SIMD.Uint32x4.shiftRightByScalar(u, -1), b.map(ursh(-1)));
+        assertEqX4(SIMD.Uint32x4.shiftRightByScalar(u, 0),  b.map(ursh(0)));
+        assertEqX4(SIMD.Uint32x4.shiftRightByScalar(u, 1),  b.map(ursh(1)));
+        assertEqX4(SIMD.Uint32x4.shiftRightByScalar(u, 2),  b.map(ursh(2)));
+        assertEqX4(SIMD.Uint32x4.shiftRightByScalar(u, 31), b.map(ursh(31)));
+        assertEqX4(SIMD.Uint32x4.shiftRightByScalar(u, 32), b.map(ursh(32)));
+        assertEqX4(SIMD.Uint32x4.shiftRightByScalar(u, 33), b.map(ursh(33)));
+
+        // Non constant shift counts
+        var c = shifts[i % shifts.length];
+
+        assertEqX4(SIMD.Int32x4.shiftLeftByScalar(v, c), a.map(lsh(c)));
+        assertEqX4(SIMD.Int32x4.shiftRightByScalar(v, c), a.map(rsh(c)));
+
+        assertEqX4(SIMD.Uint32x4.shiftLeftByScalar(u, c), b.map(ulsh(c)));
+        assertEqX4(SIMD.Uint32x4.shiftRightByScalar(u, c), b.map(ursh(c)));
+    }
+    return r;
+}
+
+f();
+

js/src/jit-test/tests/SIMD/shuffle.js

@@ -0,0 +1,86 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 50);
+
+function f() {
+    var i1 = SIMD.Int32x4(1, 2, 3, 4);
+    var i2 = SIMD.Int32x4(5, 6, 7, 8);
+
+    var leet = Math.fround(13.37);
+    var f1 = SIMD.Float32x4(-.5, -0, Infinity, leet);
+    var f2 = SIMD.Float32x4(42, .5, 23, -10);
+
+    // computes all rotations of a given array
+    function *gen(arr) {
+        var previous = arr.slice().splice(0, 4);
+        var i = 4;
+        for (var j = 0; j < 8; j++) {
+            yield previous.slice();
+            previous = previous.splice(1, previous.length - 1);
+            previous.push(arr[i]);
+            i = (i + 1) % arr.length;
+        }
+    }
+
+    var compI = [];
+    var baseI = [];
+    for (var i = 0; i < 8; i++)
+        baseI.push(SIMD.Int32x4.extractLane(i < 4 ? i1 : i2, i % 4));
+    for (var k of gen(baseI))
+        compI.push(k);
+
+    var compF = [];
+    var baseF = [];
+    for (var i = 0; i < 8; i++)
+        baseF.push(SIMD.Float32x4.extractLane(i < 4 ? f1 : f2, i % 4));
+    for (var k of gen(baseF))
+        compF.push(k);
+
+    for (var i = 0; i < 150; i++) {
+        // Variable lanes
+        var r = SIMD.Float32x4.shuffle(f1, f2, i % 8, (i + 1) % 8, (i + 2) % 8, (i + 3) % 8);
+        assertEqX4(r, compF[i % 8]);
+
+        // Constant lanes
+        assertEqX4(SIMD.Float32x4.shuffle(f1, f2, 3, 2, 4, 5), [leet, Infinity, 42, .5]);
+
+        // Variable lanes
+        var r = SIMD.Int32x4.shuffle(i1, i2, i % 8, (i + 1) % 8, (i + 2) % 8, (i + 3) % 8);
+        assertEqX4(r, compI[i % 8]);
+
+        // Constant lanes
+        assertEqX4(SIMD.Int32x4.shuffle(i1, i2, 3, 2, 4, 5), [4, 3, 5, 6]);
+    }
+}
+
+function testBailouts(expectException, uglyDuckling) {
+    var i1 = SIMD.Int32x4(1, 2, 3, 4);
+    var i2 = SIMD.Int32x4(5, 6, 7, 8);
+
+    for (var i = 0; i < 150; i++) {
+        // Test bailouts
+        var value = i == 149 ? uglyDuckling : 0;
+        var caught = false;
+        try {
+            assertEqX4(SIMD.Int32x4.shuffle(i1, i2, value, 2, 4, 5), [1, 3, 5, 6]);
+        } catch(e) {
+            print(e);
+            caught = true;
+            assertEq(i, 149);
+            assertEq(e instanceof TypeError || e instanceof RangeError, true);
+        }
+        if (i == 149)
+            assertEq(caught, expectException);
+    }
+}
+
+f();
+testBailouts(true, -1);
+testBailouts(true, 8);
+testBailouts(true, 2.5);
+testBailouts(true, undefined);
+testBailouts(true, {});
+testBailouts(true, 'one');
+testBailouts(false, false);
+testBailouts(false, null);
+testBailouts(false, " 0.0 ");

js/src/jit-test/tests/SIMD/splat.js

@@ -0,0 +1,15 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 50);
+
+function f() {
+    for (var i = 0; i < 150; i++) {
+        assertEqX4(SIMD.Int32x4.splat(42),   [42, 42, 42, 42]);
+        assertEqX4(SIMD.Float32x4.splat(42), [42, 42, 42, 42]);
+        assertEqX4(SIMD.Bool32x4.splat(true), [true, true, true, true]);
+        assertEqX4(SIMD.Bool32x4.splat(false), [false, false, false, false]);
+    }
+}
+
+f();
+

js/src/jit-test/tests/SIMD/store.js

@@ -0,0 +1,143 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 40);
+
+function f() {
+    var f32 = new Float32Array(16);
+    for (var i = 0; i < 16; i++)
+        f32[i] = i + 1;
+
+    var f64 = new Float64Array(f32.buffer);
+    var i32 = new Int32Array(f32.buffer);
+    var u32 = new Uint32Array(f32.buffer);
+    var i16 = new Int16Array(f32.buffer);
+    var u16 = new Uint16Array(f32.buffer);
+    var i8  = new Int8Array(f32.buffer);
+    var u8  = new Uint8Array(f32.buffer);
+
+    var f4 = SIMD.Float32x4(42, 43, 44, 45);
+
+    function check(n) {
+        assertEq(f32[0], 42);
+        assertEq(f32[1], n > 1 ? 43 : 2);
+        assertEq(f32[2], n > 2 ? 44 : 3);
+        assertEq(f32[3], n > 3 ? 45 : 4);
+
+        f32[0] = 1;
+        f32[1] = 2;
+        f32[2] = 3;
+        f32[3] = 4;
+    }
+
+    function testStore() {
+        SIMD.Float32x4.store(f64, 0, f4);
+        check(4);
+        SIMD.Float32x4.store(f32, 0, f4);
+        check(4);
+        SIMD.Float32x4.store(i32, 0, f4);
+        check(4);
+        SIMD.Float32x4.store(u32, 0, f4);
+        check(4);
+        SIMD.Float32x4.store(i16, 0, f4);
+        check(4);
+        SIMD.Float32x4.store(u16, 0, f4);
+        check(4);
+        SIMD.Float32x4.store(i8, 0, f4);
+        check(4);
+        SIMD.Float32x4.store(u8, 0, f4);
+        check(4);
+    }
+
+    function testStore1() {
+        SIMD.Float32x4.store1(f64, 0, f4);
+        check(1);
+        SIMD.Float32x4.store1(f32, 0, f4);
+        check(1);
+        SIMD.Float32x4.store1(i32, 0, f4);
+        check(1);
+        SIMD.Float32x4.store1(u32, 0, f4);
+        check(1);
+        SIMD.Float32x4.store1(i16, 0, f4);
+        check(1);
+        SIMD.Float32x4.store1(u16, 0, f4);
+        check(1);
+        SIMD.Float32x4.store1(i8, 0, f4);
+        check(1);
+        SIMD.Float32x4.store1(u8, 0, f4);
+        check(1);
+    }
+
+    function testStore2() {
+        SIMD.Float32x4.store2(f64, 0, f4);
+        check(2);
+        SIMD.Float32x4.store2(f32, 0, f4);
+        check(2);
+        SIMD.Float32x4.store2(i32, 0, f4);
+        check(2);
+        SIMD.Float32x4.store2(u32, 0, f4);
+        check(2);
+        SIMD.Float32x4.store2(i16, 0, f4);
+        check(2);
+        SIMD.Float32x4.store2(u16, 0, f4);
+        check(2);
+        SIMD.Float32x4.store2(i8, 0, f4);
+        check(2);
+        SIMD.Float32x4.store2(u8, 0, f4);
+        check(2);
+    }
+
+    function testStore3() {
+        SIMD.Float32x4.store3(f64, 0, f4);
+        check(3);
+        SIMD.Float32x4.store3(f32, 0, f4);
+        check(3);
+        SIMD.Float32x4.store3(i32, 0, f4);
+        check(3);
+        SIMD.Float32x4.store3(u32, 0, f4);
+        check(3);
+        SIMD.Float32x4.store3(i16, 0, f4);
+        check(3);
+        SIMD.Float32x4.store3(u16, 0, f4);
+        check(3);
+        SIMD.Float32x4.store3(i8, 0, f4);
+        check(3);
+        SIMD.Float32x4.store3(u8, 0, f4);
+        check(3);
+    }
+
+    for (var i = 0; i < 150; i++) {
+        testStore();
+        testStore1();
+        testStore2();
+        testStore3();
+    }
+}
+
+f();
+
+function testBailout(uglyDuckling) {
+    var f32 = new Float32Array(16);
+    for (var i = 0; i < 16; i++)
+        f32[i] = i + 1;
+
+    var i8  = new Int8Array(f32.buffer);
+
+    var f4 = SIMD.Float32x4(42, 43, 44, 45);
+
+    for (var i = 0; i < 150; i++) {
+        var caught = false;
+        try {
+            SIMD.Float32x4.store(i8, (i < 149) ? 0 : (16 << 2) - (4 << 2) + 1, f4);
+        } catch (e) {
+            print(e);
+            assertEq(e instanceof RangeError, true);
+            caught = true;
+        }
+        assertEq(i < 149 || caught, true);
+    }
+}
+
+print('Testing range checks...');
+testBailout(-1);
+testBailout(-15);
+testBailout(12 * 4 + 1);

js/src/jit-test/tests/SIMD/swizzle.js

@@ -0,0 +1,104 @@
+if (!this.hasOwnProperty("SIMD"))
+  quit();
+
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 50);
+
+function f() {
+    var i4 = SIMD.Int32x4(1, 2, 3, 4);
+
+    var leet = Math.fround(13.37);
+    var f4 = SIMD.Float32x4(-.5, -0, Infinity, leet);
+
+    var compI = [
+        [1,2,3,4],
+        [2,3,4,1],
+        [3,4,1,2],
+        [4,1,2,3]
+    ];
+
+    var compF = [
+        [-.5, -0, Infinity, leet],
+        [-0, Infinity, leet, -.5],
+        [Infinity, leet, -.5, -0],
+        [leet, -.5, -0, Infinity]
+    ];
+
+    for (var i = 0; i < 150; i++) {
+        // Variable lanes
+        var r = SIMD.Float32x4.swizzle(f4, i % 4, (i + 1) % 4, (i + 2) % 4, (i + 3) % 4);
+        assertEqX4(r, compF[i % 4]);
+
+        // Constant lanes
+        assertEqX4(SIMD.Float32x4.swizzle(f4, 3, 2, 1, 0), [leet, Infinity, -0, -.5]);
+
+        // Variable lanes
+        var r = SIMD.Int32x4.swizzle(i4, i % 4, (i + 1) % 4, (i + 2) % 4, (i + 3) % 4);
+        assertEqX4(r, compI[i % 4]);
+
+        // Constant lanes
+        assertEqX4(SIMD.Int32x4.swizzle(i4, 3, 2, 1, 0), [4, 3, 2, 1]);
+    }
+}
+
+function testBailouts(expectException, uglyDuckling) {
+    var i4 = SIMD.Int32x4(1, 2, 3, 4);
+    for (var i = 0; i < 150; i++) {
+        // Test bailouts
+        var value = i == 149 ? uglyDuckling : 0;
+        var caught = false;
+        try {
+            assertEqX4(SIMD.Int32x4.swizzle(i4, value, 3, 2, 0), [1, 4, 3, 1]);
+        } catch(e) {
+            print(e);
+            caught = true;
+            assertEq(i, 149);
+            assertEq(e instanceof TypeError || e instanceof RangeError, true);
+        }
+        if (i == 149)
+            assertEq(caught, expectException);
+    }
+}
+
+function testInt32x4SwizzleBailout() {
+    // Test out-of-bounds non-constant indices. This is expected to throw.
+    var i4 = SIMD.Int32x4(1, 2, 3, 4);
+    for (var i = 0; i < 150; i++) {
+        assertEqX4(SIMD.Int32x4.swizzle(i4, i, 3, 2, 0), [i + 1, 4, 3, 1]);
+    }
+}
+
+f();
+testBailouts(true, -1);
+testBailouts(true, 4);
+testBailouts(true, 2.5);
+testBailouts(true, undefined);
+testBailouts(true, {});
+testBailouts(true, 'one');
+testBailouts(false, false);
+testBailouts(false, null);
+testBailouts(false, " 0.0 ");
+
+try {
+    testInt32x4SwizzleBailout();
+    throw 'not caught';
+} catch(e) {
+    assertEq(e instanceof RangeError, true);
+}
+
+(function() {
+    var zappa = 0;
+
+    function testBailouts() {
+        var i4 = SIMD.Int32x4(1, 2, 3, 4);
+        for (var i = 0; i < 300; i++) {
+            var value = i == 299 ? 2.5 : 1;
+            SIMD.Int32x4.swizzle(i4, value, 3, 2, 0);
+            zappa = i;
+        }
+    }
+
+    try { testBailouts(); } catch (e) {}
+    assertEq(zappa, 298);
+})();

js/src/jit-test/tests/SIMD/uconvert.js

@@ -0,0 +1,86 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 30);
+
+// Testing Uint32 <-> Float32 conversions.
+// These conversions deserve special attention because SSE doesn't provide
+// simple conversion instructions.
+
+// Convert an Uint32Array to a Float32Array using scalar conversions.
+function cvt_utof_scalar(u32s, f32s) {
+    assertEq(u32s.length, f32s.length);
+    for (var i = 0; i < u32s.length; i++) {
+        f32s[i] = u32s[i];
+    }
+}
+
+// Convert an Uint32Array to a Float32Array using simd conversions.
+function cvt_utof_simd(u32s, f32s) {
+    assertEq(u32s.length, f32s.length);
+    for (var i = 0; i < u32s.length; i += 4) {
+        SIMD.Float32x4.store(f32s, i, SIMD.Float32x4.fromUint32x4(SIMD.Uint32x4.load(u32s, i)));
+    }
+}
+
+// Convert a Float32Array to an Uint32Array using scalar conversions.
+function cvt_ftou_scalar(f32s, u32s) {
+    assertEq(f32s.length, u32s.length);
+    for (var i = 0; i < f32s.length; i++) {
+        u32s[i] = f32s[i];
+    }
+}
+
+// Convert a Float32Array to an Uint32Array using simd conversions.
+function cvt_ftou_simd(f32s, u32s) {
+    assertEq(f32s.length, u32s.length);
+    for (var i = 0; i < f32s.length; i += 4) {
+        SIMD.Uint32x4.store(u32s, i, SIMD.Uint32x4.fromFloat32x4(SIMD.Float32x4.load(f32s, i)));
+    }
+}
+
+function check(a, b) {
+    assertEq(a.length, b.length);
+    for (var i = 0; i < a.length; i++) {
+        assertEq(a[i], b[i]);
+    }
+}
+
+// Uint32x4 --> Float32x4 tests.
+var src = new Uint32Array(8000);
+var dst1 = new Float32Array(8000);
+var dst2 = new Float32Array(8000);
+
+for (var i = 0; i < 2000; i++) {
+    src[i] = i;
+    src[i + 2000] = 0x7fffffff - i;
+    src[i + 4000] = 0x80000000 + i;
+    src[i + 6000] = 0xffffffff - i;
+}
+
+for (var n = 0; n < 10; n++) {
+    cvt_utof_scalar(src, dst1);
+    cvt_utof_simd(src, dst2);
+    check(dst1, dst2);
+}
+
+// Float32x4 --> Uint32x4 tests.
+var fsrc = dst1;
+var fdst1 = new Uint32Array(8000);
+var fdst2 = new Uint32Array(8000);
+
+// The 0xffffffff entries in fsrc round to 0x1.0p32f which throws.
+// Go as high as 0x0.ffffffp32f.
+for (var i = 0; i < 2000; i++) {
+    fsrc[i + 6000] = 0xffffff7f - i;
+}
+
+// Truncation towards 0.
+fsrc[1990] = -0.9
+fsrc[1991] = 0.9
+fsrc[1992] = 1.9
+
+for (var n = 0; n < 10; n++) {
+    cvt_ftou_scalar(fsrc, fdst1);
+    cvt_ftou_simd(fsrc, fdst2);
+    check(fdst1, fdst2);
+}

js/src/jit-test/tests/SIMD/unary.js

@@ -0,0 +1,35 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("ion.warmup.trigger", 50);
+
+var notf = (function() {
+    var i32 = new Int32Array(1);
+    var f32 = new Float32Array(i32.buffer);
+    return function(x) {
+        f32[0] = x;
+        i32[0] = ~i32[0];
+        return f32[0];
+    }
+})();
+
+function f() {
+    var f4 = SIMD.Float32x4(1, 2, 3, 4);
+    var i4 = SIMD.Int32x4(1, 2, 3, 4);
+    var b4 = SIMD.Bool32x4(true, false, true, false);
+    var BitOrZero = (x) => x | 0;
+    for (var i = 0; i < 150; i++) {
+        assertEqX4(SIMD.Float32x4.neg(f4), unaryX4((x) => -x, f4, Math.fround));
+        assertEqX4(SIMD.Float32x4.abs(f4), unaryX4(Math.abs, f4, Math.fround));
+        assertEqX4(SIMD.Float32x4.sqrt(f4), unaryX4(Math.sqrt, f4, Math.fround));
+
+        assertEqX4(SIMD.Float32x4.reciprocalApproximation(f4), unaryX4((x) => 1 / x, f4, Math.fround), assertNear);
+        assertEqX4(SIMD.Float32x4.reciprocalSqrtApproximation(f4), unaryX4((x) => 1 / Math.sqrt(x), f4, Math.fround), assertNear);
+
+        assertEqX4(SIMD.Int32x4.not(i4), unaryX4((x) => ~x, i4, BitOrZero));
+        assertEqX4(SIMD.Int32x4.neg(i4), unaryX4((x) => -x, i4, BitOrZero));
+
+        assertEqX4(SIMD.Bool32x4.not(b4), unaryX4((x) => !x, b4, (x) => x ));
+    }
+}
+
+f();

js/src/jit-test/tests/SIMD/unbox.js

@@ -0,0 +1,144 @@
+load(libdir + 'simd.js');
+
+setJitCompilerOption("baseline.warmup.trigger", 10);
+setJitCompilerOption("ion.warmup.trigger", 30);
+
+var max = 40, pivot = 35;
+
+var i32x4 = SIMD.Int32x4;
+var f32x4 = SIMD.Float32x4;
+var i32x4Add = SIMD.Int32x4.add;
+
+var FakeSIMDType = function (o) { this.x = o.x; this.y = o.y; this.z = o.z; this.w = o.w; };
+if (this.hasOwnProperty("TypedObject")) {
+  var TO = TypedObject;
+  FakeSIMDType = new TO.StructType({ x: TO.int32, y: TO.int32, z: TO.int32, w: TO.int32 });
+}
+
+function simdunbox_bail_undef(i, lhs, rhs) {
+  return i32x4Add(lhs, rhs);
+}
+
+function simdunbox_bail_object(i, lhs, rhs) {
+  return i32x4Add(lhs, rhs);
+}
+
+function simdunbox_bail_typeobj(i, lhs, rhs) {
+  return i32x4Add(lhs, rhs);
+}
+
+function simdunbox_bail_badsimd(i, lhs, rhs) {
+  return i32x4Add(lhs, rhs);
+}
+
+var arr_undef = [ i32x4(0, 1, 1, 2), i32x4(1, 1, 2, 3) ];
+var fail_undef = 0;
+var arr_object = [ i32x4(0, 1, 1, 2), i32x4(1, 1, 2, 3) ];
+var fail_object = 0;
+var arr_typeobj = [ i32x4(0, 1, 1, 2), i32x4(1, 1, 2, 3) ];
+var fail_typeobj = 0;
+var arr_badsimd = [ i32x4(0, 1, 1, 2), i32x4(1, 1, 2, 3) ];
+var fail_badsimd = 0;
+for (var i = 0; i < max; i++) {
+  try {
+    arr_undef[i + 2] = simdunbox_bail_undef(i, arr_undef[i], arr_undef[i + 1]);
+  } catch (x) {
+    arr_undef[i + 2] = arr_undef[i - 1];
+    fail_undef++;
+  }
+
+  try {
+    arr_object[i + 2] = simdunbox_bail_object(i, arr_object[i], arr_object[i + 1]);
+  } catch (x) {
+    arr_object[i + 2] = arr_object[i - 1];
+    fail_object++;
+  }
+
+  try {
+    arr_typeobj[i + 2] = simdunbox_bail_typeobj(i, arr_typeobj[i], arr_typeobj[i + 1]);
+  } catch (x) {
+    arr_typeobj[i + 2] = arr_typeobj[i - 1];
+    fail_typeobj++;
+  }
+
+  try {
+    arr_badsimd[i + 2] = simdunbox_bail_badsimd(i, arr_badsimd[i], arr_badsimd[i + 1]);
+  } catch (x) {
+    arr_badsimd[i + 2] = arr_badsimd[i - 1];
+    fail_badsimd++;
+  }
+
+  if (i + 2 == pivot) {
+    arr_undef[pivot] = undefined;
+    arr_object[pivot] = { x: 0, y: 1, z: 2, w: 3 };
+    arr_typeobj[pivot] = new FakeSIMDType({ x: 0, y: 1, z: 2, w: 3 });
+    arr_badsimd[pivot] = f32x4(0, 1, 2, 3);
+  }
+}
+
+assertEq(fail_undef, 2);
+assertEq(fail_object, 2);
+assertEq(fail_typeobj, 2);
+assertEq(fail_badsimd, 2);
+
+// Assert that all SIMD values are correct.
+function assertEqX4(real, expected, assertFunc) {
+    if (typeof assertFunc === 'undefined')
+        assertFunc = assertEq;
+
+    assertFunc(real.x, expected[0]);
+    assertFunc(real.y, expected[1]);
+    assertFunc(real.z, expected[2]);
+    assertFunc(real.w, expected[3]);
+}
+
+var fib = [0, 1];
+for (i = 0; i < max + 5; i++)
+  fib[i+2] = (fib[i] + fib[i+1]) | 0;
+
+for (i = 0; i < max; i++) {
+  if (i == pivot)
+    continue;
+  var ref = fib.slice(i < pivot ? i : i - 3);
+  assertEqX4(arr_undef[i], ref);
+  assertEqX4(arr_object[i], ref);
+  assertEqX4(arr_typeobj[i], ref);
+  assertEqX4(arr_badsimd[i], ref);
+}
+
+// Check that unbox operations aren't removed
+(function() {
+
+    function add(i, v, w) {
+        if (i % 2 == 0) {
+            SIMD.Int32x4.add(v, w);
+        } else {
+            SIMD.Float32x4.add(v, w);
+        }
+    }
+
+    var i = 0;
+    var caught = false;
+    var f4 = SIMD.Float32x4(1,2,3,4);
+    var i4 = SIMD.Int32x4(1,2,3,4);
+    try {
+        for (; i < 200; i++) {
+            if (i % 2 == 0) {
+                add(i, i4, i4);
+            } else if (i == 199) {
+                add(i, i4, f4);
+            } else {
+                add(i, f4, f4);
+            }
+        }
+    } catch(e) {
+        print(e);
+        assertEq(e instanceof TypeError, true);
+        assertEq(i, 199);
+        caught = true;
+    }
+
+    assertEq(i < 199 || caught, true);
+
+})();
+

js/src/jit-test/tests/asm.js/bug1126251.js

@@ -13,3 +13,41 @@ var v = asmLink(asmCompile('global', `
 `), this)();
 
 assertEq(v, NaN);
+
+if (!isSimdAvailable() || typeof SIMD === 'undefined') {
+    quit(0);
+}
+
+var v = asmLink(asmCompile('global', `
+    "use asm";
+    var frd = global.Math.fround;
+    var Float32x4 = global.SIMD.Float32x4;
+    var splat = Float32x4.splat;
+    var ext = Float32x4.extractLane;
+    function e() {
+        var v = Float32x4(0,0,0,0);
+        var x = frd(0.);
+        v = splat(.1e+71);
+        x = ext(v,0);
+        x = frd(x / x);
+        return +x;
+    }
+    return e;
+`), this)();
+
+assertEq(v, NaN);
+
+// Bug 1130618: without GVN
+setJitCompilerOption("ion.gvn.enable", 0);
+var v = asmLink(asmCompile('global', `
+    "use asm";
+    var Float32x4 = global.SIMD.Float32x4;
+    var splat = Float32x4.splat;
+    var ext = Float32x4.extractLane;
+    function e() {
+        return +ext(splat(.1e+71),0);
+    }
+    return e;
+`), this)();
+
+assertEq(v, Infinity);

js/src/jit-test/tests/asm.js/bug1201124-simd-proxy.js

@@ -0,0 +1,28 @@
+load(libdir + "asm.js");
+load(libdir + "asserts.js");
+
+if (typeof newGlobal !== 'function' ||
+    !isSimdAvailable() ||
+    typeof SIMD === 'undefined')
+{
+    quit();
+}
+
+var stdlib = new (newGlobal().Proxy)(this, new Proxy({
+    simdGet: 0,
+    getOwnPropertyDescriptor(t, pk) {
+        if (pk === "SIMD" && this.simdGet++ === 1) {
+            return {};
+        }
+        return Reflect.getOwnPropertyDescriptor(t, pk);
+    }
+}, {
+    get(t, pk, r) {
+        print("trap", pk);
+        return Reflect.get(t, pk, r);
+    }
+}));
+
+var m = asmCompile('stdlib', '"use asm"; var i4=stdlib.SIMD.Int32x4; var i4add=i4.add; return {}');
+
+assertAsmLinkFail(m, stdlib);

js/src/jit-test/tests/asm.js/simd-fbirds.js

@@ -0,0 +1,198 @@
+/* -*- Mode: javascript; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 ; js-indent-level : 2 ; js-curly-indent-offset: 0 -*- */
+/* vim: set ts=4 et sw=4 tw=80: */
+
+// Author: Peter Jensen
+
+load(libdir + "asm.js");
+if (!isSimdAvailable() || typeof SIMD === 'undefined') {
+    print("won't run tests as simd extensions aren't activated yet");
+    quit(0);
+}
+
+const NUM_BIRDS = 30;
+const NUM_UPDATES = 20;
+const ACCEL_DATA_STEPS = 30;
+
+var buffer = new ArrayBuffer(0x200000);
+var bufferF32 = new Float32Array(buffer);
+
+var actualBirds = 0;
+
+function init() {
+    actualBirds = 0;
+    // Make it a power of two, for quick modulo wrapping.
+    var accelDataValues = [10.0, 9.5, 9.0, 8.0, 7.0, 6.0, 5.5, 5.0, 5.0, 5.0, 5.5, 6.0, 7.0, 8.0, 9.0, 10.0];
+    accelDataValues = accelDataValues.map(function(v) { return 50*v; });
+    var accelDataValuesLength = accelDataValues.length;
+    assertEq(accelDataValuesLength, 16); // Hard coded in the asm.js module
+    for (i = 0; i < accelDataValuesLength; i++)
+        bufferF32[i + NUM_BIRDS * 2] = accelDataValues[i];
+}
+
+function addBird(pos, vel) {
+    bufferF32[actualBirds] = pos;
+    bufferF32[actualBirds + NUM_BIRDS] = vel;
+    actualBirds++;
+    return actualBirds - 1;
+}
+
+function getActualBirds() {
+    return actualBirds;
+}
+
+var code = `
+    "use asm";
+    var toF = global.Math.fround;
+    var u8 = new global.Uint8Array(buffer);
+    var f32 = new global.Float32Array(buffer);
+    const maxBirds = 100000;
+    const maxBirdsx4 = 400000;
+    const maxBirdsx8 = 800000;
+    const accelMask = 0x3c;
+    const mk4 = 0x000ffff0;
+
+    const getMaxPos = 1000.0;
+    const getAccelDataSteps = imp.accelDataSteps | 0;
+    var getActualBirds = imp.getActualBirds;
+
+    var i4 = global.SIMD.Int32x4;
+    var f4 = global.SIMD.Float32x4;
+    var b4 = global.SIMD.Bool32x4;
+    var i4add = i4.add;
+    var i4and = i4.and;
+    var f4select = f4.select;
+    var f4add = f4.add;
+    var f4sub = f4.sub;
+    var f4mul = f4.mul;
+    var f4greaterThan = f4.greaterThan;
+    var f4splat = f4.splat;
+    var f4load = f4.load;
+    var f4store = f4.store;
+    var b4any = b4.anyTrue;
+
+    const zerox4 = f4(0.0,0.0,0.0,0.0);
+
+    function declareHeapSize() {
+        f32[0x0007ffff] = toF(0.0);
+    }
+
+    function update(timeDelta) {
+        timeDelta = toF(timeDelta);
+        //      var steps               = Math.ceil(timeDelta/accelData.interval);
+        var steps = 0;
+        var subTimeDelta = toF(0.0);
+        var actualBirds = 0;
+        var maxPos = toF(0.0);
+        var maxPosx4 = f4(0.0,0.0,0.0,0.0);
+        var subTimeDeltax4  = f4(0.0,0.0,0.0,0.0);
+        var subTimeDeltaSquaredx4 = f4(0.0,0.0,0.0,0.0);
+        var point5x4 = f4(0.5, 0.5, 0.5, 0.5);
+        var i = 0;
+        var len = 0;
+        var accelIndex = 0;
+        var newPosx4 = f4(0.0,0.0,0.0,0.0);
+        var newVelx4 = f4(0.0,0.0,0.0,0.0);
+        var accel = toF(0.0);
+        var accelx4 = f4(0.0,0.0,0.0,0.0);
+        var a = 0;
+        var posDeltax4 = f4(0.0,0.0,0.0,0.0);
+        var cmpx4 = b4(0,0,0,0);
+        var newVelTruex4 = f4(0.0,0.0,0.0,0.0);
+
+        steps = getAccelDataSteps | 0;
+        subTimeDelta = toF(toF(timeDelta / toF(steps | 0)) / toF(1000.0));
+        actualBirds = getActualBirds() | 0;
+        maxPos = toF(+getMaxPos);
+        maxPosx4 = f4splat(maxPos);
+        subTimeDeltax4 = f4splat(subTimeDelta);
+        subTimeDeltaSquaredx4 = f4mul(subTimeDeltax4, subTimeDeltax4);
+
+        len = ((actualBirds + 3) >> 2) << 4;
+
+        for (i = 0; (i | 0) < (len | 0); i = (i + 16) | 0) {
+            accelIndex = 0;
+            newPosx4 = f4load(u8, i & mk4);
+            newVelx4 = f4load(u8, (i & mk4) + maxBirdsx4);
+            for (a = 0; (a | 0) < (steps | 0); a = (a + 1) | 0) {
+                accel = toF(f32[(accelIndex & accelMask) + maxBirdsx8 >> 2]);
+                accelx4 = f4splat(accel);
+                accelIndex = (accelIndex + 4) | 0;
+                posDeltax4 = f4mul(point5x4, f4mul(accelx4, subTimeDeltaSquaredx4));
+                posDeltax4 = f4add(posDeltax4, f4mul(newVelx4, subTimeDeltax4));
+                newPosx4 = f4add(newPosx4, posDeltax4);
+                newVelx4 = f4add(newVelx4, f4mul(accelx4, subTimeDeltax4));
+                cmpx4 = f4greaterThan(newPosx4, maxPosx4);
+
+                if (b4any(cmpx4)) {
+                    // Work around unimplemented 'neg' operation, using 0 - x.
+                    newVelTruex4 = f4sub(zerox4, newVelx4);
+                    newVelx4 = f4select(cmpx4, newVelTruex4, newVelx4);
+                }
+            }
+            f4store(u8, i & mk4, newPosx4);
+            f4store(u8, (i & mk4) + maxBirdsx4, newVelx4);
+        }
+    }
+
+    return update;
+`
+
+var ffi = {
+    getActualBirds,
+    accelDataSteps: ACCEL_DATA_STEPS
+};
+
+var fbirds = asmLink(asmCompile('global', 'imp', 'buffer', code), this, ffi, buffer);
+
+init();
+for (var i = 0; i < NUM_BIRDS; i++) {
+    addBird(i / 10, Math.exp(2, NUM_BIRDS - i));
+}
+
+var b = dateNow();
+for (var j = 0; j < NUM_UPDATES; j++) {
+    fbirds(16);
+}
+print(dateNow() - b);
+
+assertEq(bufferF32[0], 0);
+assertEq(bufferF32[1], 0.10000000149011612);
+assertEq(bufferF32[2], 0.20000000298023224);
+assertEq(bufferF32[3], 0.30000001192092896);
+assertEq(bufferF32[4], 0.4000000059604645);
+assertEq(bufferF32[5], 0.5);
+assertEq(bufferF32[6], 0.6000000238418579);
+assertEq(bufferF32[7], 0.699999988079071);
+assertEq(bufferF32[8], 0.800000011920929);
+assertEq(bufferF32[9], 0.8999999761581421);
+assertEq(bufferF32[10], 1);
+assertEq(bufferF32[11], 1.100000023841858);
+assertEq(bufferF32[12], 1.2000000476837158);
+assertEq(bufferF32[13], 1.2999999523162842);
+assertEq(bufferF32[14], 1.399999976158142);
+assertEq(bufferF32[15], 1.5);
+assertEq(bufferF32[16], 1.600000023841858);
+assertEq(bufferF32[17], 1.7000000476837158);
+assertEq(bufferF32[18], 1.7999999523162842);
+assertEq(bufferF32[19], 1.899999976158142);
+assertEq(bufferF32[20], 2);
+assertEq(bufferF32[21], 2.0999999046325684);
+assertEq(bufferF32[22], 2.200000047683716);
+assertEq(bufferF32[23], 2.299999952316284);
+assertEq(bufferF32[24], 2.4000000953674316);
+assertEq(bufferF32[25], 2.5);
+assertEq(bufferF32[26], 2.5999999046325684);
+assertEq(bufferF32[27], 2.700000047683716);
+assertEq(bufferF32[28], 2.799999952316284);
+assertEq(bufferF32[29], 2.9000000953674316);
+
+
+// Code used to generate the assertEq list above.
+function generateAssertList() {
+    var buf = '';
+    for (var k = 0; k < NUM_BIRDS; k++) {
+        buf += 'assertEq(bufferF32['+ k + '], ' + bufferF32[k] + ');\n';
+    }
+    print(buf);
+}
+//generateAssertList();

js/src/jit-test/tests/asm.js/testAsmJSWasmMixing.js

@@ -13,6 +13,17 @@ asmLink(asmJS, this, null, asmJSBuf);
 var wasmMem = wasmEvalText('(module (memory 1 1) (export "mem" memory))').exports.mem;
 assertAsmLinkFail(asmJS, this, null, wasmMem.buffer);
 
+if (!getBuildConfiguration().x64 && isSimdAvailable() && this["SIMD"]) {
+    var simdJS = asmCompile('stdlib', 'ffis', 'buf', USE_ASM + 'var i32 = new stdlib.Int32Array(buf); var i32x4 = stdlib.SIMD.Int32x4; return {}');
+    assertAsmLinkFail(simdJS, this, null, asmJSBuf);
+    assertAsmLinkFail(simdJS, this, null, wasmMem.buffer);
+
+    var simdJSBuf = new ArrayBuffer(BUF_MIN);
+    asmLink(simdJS, this, null, simdJSBuf);
+    asmLink(simdJS, this, null, simdJSBuf);  // multiple SIMD.js instantiations succeed
+    assertAsmLinkFail(asmJS, this, null, simdJSBuf);  // but not asm.js
+}
+
 setJitCompilerOption('asmjs.atomics.enable', 1);
 
 var sharedAsmJS = asmCompile('stdlib', 'ffis', 'buf',

js/src/jit-test/tests/asm.js/testBug1099216.js

@@ -0,0 +1,61 @@
+if (typeof SIMD === 'undefined' || !isSimdAvailable()) {
+    print("won't run tests as simd extensions aren't activated yet");
+    quit(0);
+}
+
+(function(global) {
+    "use asm";
+    var frd = global.Math.fround;
+    var fx4 = global.SIMD.Float32x4;
+    var fc4 = fx4.check;
+    var fsp = fx4.splat;
+    function s(){}
+    function d(x){x=fc4(x);}
+    function e() {
+        var x = frd(0);
+        x = frd(x / x);
+        s();
+        d(fsp(x));
+    }
+    return e;
+})(this)();
+
+(function(m) {
+    "use asm"
+    var k = m.SIMD.Bool32x4
+    var g = m.SIMD.Int32x4
+    var gc = g.check;
+    var h = g.select
+    function f() {
+        var x = k(0, 0, 0, 0)
+        var y = g(1, 2, 3, 4)
+        return gc(h(x, y, y))
+    }
+    return f;
+})(this)();
+
+t = (function(global) {
+    "use asm"
+    var toF = global.Math.fround
+    var f4 = global.SIMD.Float32x4
+    var f4c = f4.check
+    function p(x, y, width, value, max_iterations) {
+        x = x | 0
+        y = y | 0
+        width = width | 0
+        value = value | 0
+        max_iterations = max_iterations | 0
+    }
+    function m(xf, yf, yd, max_iterations) {
+        xf = toF(xf)
+        yf = toF(yf)
+        yd = toF(yd)
+        max_iterations = max_iterations | 0
+        var _ = f4(0, 0, 0, 0), c_im4 = f4(0, 0, 0, 0)
+        c_im4 = f4(yf, yd, yd, yf)
+        return f4c(c_im4);
+    }
+    return {p:p,m:m};
+})(this)
+t.p();
+t.m();

js/src/jit-test/tests/asm.js/testJumpRange.js

@@ -18,6 +18,26 @@ for (let threshold of [0, 50, 100, 5000, -1]) {
             return h
         `)()(), 45);
 
+    if (isSimdAvailable() && this.SIMD) {
+        var buf = new ArrayBuffer(BUF_MIN);
+        new Int32Array(buf)[0] = 10;
+        new Float32Array(buf)[1] = 42;
+        assertEq(asmCompile('stdlib', 'ffis', 'buf',
+            USE_ASM + `
+                var H = new stdlib.Uint8Array(buf);
+                var i4 = stdlib.SIMD.Int32x4;
+                var f4 = stdlib.SIMD.Float32x4;
+                var i4load = i4.load;
+                var f4load = f4.load;
+                var toi4 = i4.fromFloat32x4;
+                var i4ext = i4.extractLane;
+                function f(i) { i=i|0; return i4ext(i4load(H, i), 0)|0 }
+                function g(i) { i=i|0; return (i4ext(toi4(f4load(H, i)),1) + (f(i)|0))|0 }
+                function h(i) { i=i|0; return g(i)|0 }
+                return h
+            `)(this, null, buf)(0), 52);
+    }
+
     enableGeckoProfiling();
     asmLink(asmCompile(USE_ASM + 'function f() {} function g() { f() } function h() { g() } return h'))();
     disableGeckoProfiling();

js/src/jit-test/tests/asm.js/testProfiling.js

@@ -210,6 +210,20 @@ if (jitOptions['baseline.enable']) {
     assertStackContainsSeq(stacks, ">,f1,>,<,f1,>,>,<,f1,>,f2,>,<,f1,>,<,f2,>,<,f1,>,f2,>,<,f1,>,>,<,f1,>,<,f1,>,f1,>,>");
 }
 
+
+if (isSimdAvailable() && typeof SIMD !== 'undefined') {
+    // SIMD out-of-bounds exit
+    var buf = new ArrayBuffer(0x10000);
+    var f = asmLink(asmCompile('g','ffi','buf', USE_ASM + 'var f4=g.SIMD.float32x4; var f4l=f4.load; var u8=new g.Uint8Array(buf); function f(i) { i=i|0; return f4l(u8, 0xFFFF + i | 0); } return f'), this, {}, buf);
+    enableSingleStepProfiling();
+    assertThrowsInstanceOf(() => f(4), RangeError);
+    var stacks = disableSingleStepProfiling();
+    // TODO check that expected is actually the correctly expected string, when
+    // SIMD is implemented on ARM.
+    assertStackContainsSeq(stacks, ">,f,>,inline stub,f,>");
+}
+
+
 // Thunks
 setJitCompilerOption("jump-threshold", 0);
 var h = asmLink(asmCompile(USE_ASM + 'function f() {} function g() { f() } function h() { g() } return h'));

js/src/jit-test/tests/asm.js/testSIMD-16x8.js

@@ -0,0 +1,525 @@
+load(libdir + "asm.js");
+load(libdir + "simd.js");
+load(libdir + "asserts.js");
+
+// Set to true to see more JS debugging spew.
+const DEBUG = false;
+
+if (!isSimdAvailable()) {
+    DEBUG && print("won't run tests as simd extensions aren't activated yet");
+    quit(0);
+}
+
+// Tests for 16x8 SIMD types: Int16x8, Uint16x8, Bool16x8.
+
+const I16x8 = 'var i16x8 = glob.SIMD.Int16x8;'
+const I16x8CHK = 'var i16x8chk = i16x8.check;'
+const I16x8EXT = 'var i16x8ext = i16x8.extractLane;'
+const I16x8REP = 'var i16x8rep = i16x8.replaceLane;'
+const I16x8U16x8 = 'var i16x8u16x8 = i16x8.fromUint16x8Bits;'
+
+const U16x8 = 'var u16x8 = glob.SIMD.Uint16x8;'
+const U16x8CHK = 'var u16x8chk = u16x8.check;'
+const U16x8EXT = 'var u16x8ext = u16x8.extractLane;'
+const U16x8REP = 'var u16x8rep = u16x8.replaceLane;'
+const U16x8I16x8 = 'var u16x8i16x8 = u16x8.fromInt16x8Bits;'
+
+const B16x8 = 'var b16x8 = glob.SIMD.Bool16x8;'
+const B16x8CHK = 'var b16x8chk = b16x8.check;'
+const B16x8EXT = 'var b16x8ext = b16x8.extractLane;'
+const B16x8REP = 'var b16x8rep = b16x8.replaceLane;'
+
+const INT16_MAX = 0x7fff
+const INT16_MIN = -0x10000
+const UINT16_MAX = 0xffff
+
+// Linking
+assertEq(asmLink(asmCompile('glob', USE_ASM + I16x8 + "function f() {} return f"), {SIMD:{Int16x8: SIMD.Int16x8}})(), undefined);
+assertEq(asmLink(asmCompile('glob', USE_ASM + U16x8 + "function f() {} return f"), {SIMD:{Uint16x8: SIMD.Uint16x8}})(), undefined);
+assertEq(asmLink(asmCompile('glob', USE_ASM + B16x8 + "function f() {} return f"), {SIMD:{Bool16x8: SIMD.Bool16x8}})(), undefined);
+
+// Local variable of Int16x8 type.
+assertAsmTypeFail('glob', USE_ASM + "function f() {var x=Int16x8(1,2,3,4,5,6,7,8);} return f");
+assertAsmTypeFail('glob', USE_ASM + I16x8 + "function f() {var x=i16x8;} return f");
+assertAsmTypeFail('glob', USE_ASM + I16x8 + "function f() {var x=i16x8();} return f");
+assertAsmTypeFail('glob', USE_ASM + I16x8 + "function f() {var x=i16x8(1);} return f");
+assertAsmTypeFail('glob', USE_ASM + I16x8 + "function f() {var x=i16x8(1,2,3,4);} return f");
+assertAsmTypeFail('glob', USE_ASM + I16x8 + "function f() {var x=i16x8(1,2,3,4,5,6,7,8.0);} return f");
+assertAsmTypeFail('glob', USE_ASM + I16x8 + "function f() {var x=i16x8(1,2,3,4,5,6,7,8,9);} return f");
+assertAsmTypeFail('glob', USE_ASM + I16x8 + "function f() {var x=i16x8(1,2,3,4,5,6,7,8|0);} return f");
+assertEq(asmLink(asmCompile('glob', USE_ASM + I16x8 + "function f() {var x=i16x8(1,2,3,4,5,6,7,8);} return f"), this)(), undefined);
+assertEq(asmLink(asmCompile('glob', USE_ASM + I16x8 + "function f() {var x=i16x8(1,2,3,4,5,6,7," + (INT16_MAX + 1) + ");} return f"), this)(), undefined);
+
+// Local variable of Uint16x8 type.
+assertAsmTypeFail('glob', USE_ASM + "function f() {var x=Uint16x8(1,2,3,4,5,6,7,8);} return f");
+assertAsmTypeFail('glob', USE_ASM + U16x8 + "function f() {var x=u16x8;} return f");
+assertAsmTypeFail('glob', USE_ASM + U16x8 + "function f() {var x=u16x8();} return f");
+assertAsmTypeFail('glob', USE_ASM + U16x8 + "function f() {var x=u16x8(1);} return f");
+assertAsmTypeFail('glob', USE_ASM + U16x8 + "function f() {var x=u16x8(1,2,3,4);} return f");
+assertAsmTypeFail('glob', USE_ASM + U16x8 + "function f() {var x=u16x8(1,2,3,4,5,6,7,8.0);} return f");
+assertAsmTypeFail('glob', USE_ASM + U16x8 + "function f() {var x=u16x8(1,2,3,4,5,6,7,8,9);} return f");
+assertAsmTypeFail('glob', USE_ASM + U16x8 + "function f() {var x=u16x8(1,2,3,4,5,6,7,8|0);} return f");
+assertEq(asmLink(asmCompile('glob', USE_ASM + U16x8 + "function f() {var x=u16x8(1,2,3,4,5,6,7,8);} return f"), this)(), undefined);
+assertEq(asmLink(asmCompile('glob', USE_ASM + U16x8 + "function f() {var x=u16x8(1,2,3,4,5,6,7," + (UINT16_MAX + 1) + ");} return f"), this)(), undefined);
+
+// Local variable of Bool16x8 type.
+assertAsmTypeFail('glob', USE_ASM + "function f() {var x=Bool16x8(1,0,0,0, 0,0,0,0);} return f");
+assertAsmTypeFail('glob', USE_ASM + B16x8 + "function f() {var x=b16x8;} return f");
+assertAsmTypeFail('glob', USE_ASM + B16x8 + "function f() {var x=b16x8();} return f");
+assertAsmTypeFail('glob', USE_ASM + B16x8 + "function f() {var x=b16x8(1);} return f");
+assertAsmTypeFail('glob', USE_ASM + B16x8 + "function f() {var x=b16x8(1,0,0,0);} return f");
+assertAsmTypeFail('glob', USE_ASM + B16x8 + "function f() {var x=b16x8(1,0,0,0, 0,0,0,1.0);} return f");
+assertAsmTypeFail('glob', USE_ASM + B16x8 + "function f() {var x=b16x8(1,0,0,0, 0,0,0,0|0);} return f");
+assertAsmTypeFail('glob', USE_ASM + B16x8 + "function f() {var x=b16x8(1,0,0,0, 0,0,0,0, 1);} return f");
+assertEq(asmLink(asmCompile('glob', USE_ASM + B16x8 + "function f() {var x=b16x8(1,0,0,0, 0,-1,-2,0);} return f"), this)(), undefined);
+
+// Global variable of Int16x8 type.
+assertEqVecArr(asmLink(asmCompile('glob', 'ffi', USE_ASM + I16x8 + I16x8CHK + "var g=i16x8chk(ffi.g); function f() { return i16x8chk(g); } return f"), this,
+                       {g: SIMD.Int16x8(1,2,3,4,5,6,7,8)})(), [1,2,3,4,5,6,7,8]);
+assertEqVecArr(asmLink(asmCompile('glob', 'ffi', USE_ASM + I16x8 + I16x8CHK + "var g=i16x8chk(ffi.g); function f() { g=i16x8(5,6,7,8,9,10,11,12); return i16x8chk(g); } return f"), this,
+                       {g: SIMD.Int16x8(1,2,3,4,5,6,7,8)})(), [5,6,7,8,9,10,11,12]);
+
+// Global variable of Bool16x8 type.
+assertEqVecArr(asmLink(asmCompile('glob', 'ffi', USE_ASM + B16x8 + B16x8CHK + "var g=b16x8chk(ffi.g); function f() { return b16x8chk(g); } return f"), this,
+                       {g: SIMD.Bool16x8(1,1,0,1,0,0,1,0)})(), [true,true,false,true,false,false,true,false]);
+assertEqVecArr(asmLink(asmCompile('glob', 'ffi', USE_ASM + B16x8 + B16x8CHK + "var g=b16x8chk(ffi.g); function f() { g=b16x8(1,1,0,1,0,1,1,1); return b16x8chk(g); } return f"), this,
+                       {g: SIMD.Bool16x8(1,1,0,1,0,0,1,0)})(), [true,true,false,true,false,true,true,true]);
+
+// Unsigned SIMD globals are not allowed.
+assertAsmTypeFail('glob', 'ffi', USE_ASM + U16x8 + U16x8CHK + "var g=u16x8chk(ffi.g); function f() { } return f");
+
+// Only signed Int16x8 allowed as return value.
+assertEqVecArr(asmLink(asmCompile('glob', USE_ASM + I16x8 + "function f() {return i16x8(1,2,3,4,5,6,7,8);} return f"), this)(),
+           [1, 2, 3, 4, 5, 6, 7, 8]);
+assertEqVecArr(asmLink(asmCompile('glob', USE_ASM + I16x8 + I16x8CHK + "function f() {return i16x8chk(i16x8(1,2,3,32771,5,6,7,8));} return f"), this)(),
+           [1, 2, 3, -32765, 5, 6, 7, 8]);
+assertAsmTypeFail('glob', USE_ASM + U16x8 + "function f() {return u16x8(1,2,3,4,5,6,7,8);} return f");
+assertAsmTypeFail('glob', USE_ASM + U16x8 + U16x8CHK + "function f() {return u16x8chk(u16x8(1,2,3,4,5,6,7,8));} return f");
+
+// Test splat.
+function splat(x) {
+    let r = []
+    for (let i = 0; i < 8; i++)
+        r.push(x);
+    return r
+}
+
+splatB = asmLink(asmCompile('glob', USE_ASM + B16x8 +
+                            'var splat = b16x8.splat;' +
+                            'function f(x) { x = x|0; return splat(x); } return f'), this);
+assertEqVecArr(splatB(true), splat(true));
+assertEqVecArr(splatB(false), splat(false));
+
+
+splatB0 = asmLink(asmCompile('glob', USE_ASM + B16x8 +
+                             'var splat = b16x8.splat;' +
+                             'function f() { var x = 0; return splat(x); } return f'), this);
+assertEqVecArr(splatB0(), splat(false));
+splatB1 = asmLink(asmCompile('glob', USE_ASM + B16x8 +
+                             'var splat = b16x8.splat;' +
+                             'function f() { var x = 1; return splat(x); } return f'), this);
+assertEqVecArr(splatB1(), splat(true));
+
+splatI = asmLink(asmCompile('glob', USE_ASM + I16x8 +
+                            'var splat = i16x8.splat;' +
+                            'function f(x) { x = x|0; return splat(x); } return f'), this);
+for (let x of [0, 1, -1, 0x12345, 0x1234, -1000, -1000000]) {
+    assertEqVecArr(splatI(x), splat(x << 16 >> 16));
+}
+
+splatIc = asmLink(asmCompile('glob', USE_ASM + I16x8 +
+                             'var splat = i16x8.splat;' +
+                             'function f() { var x = 100; return splat(x); } return f'), this);
+assertEqVecArr(splatIc(), splat(100))
+
+splatU = asmLink(asmCompile('glob', USE_ASM + U16x8 + I16x8 + I16x8U16x8 +
+                            'var splat = u16x8.splat;' +
+                            'function f(x) { x = x|0; return i16x8u16x8(splat(x)); } return f'), this);
+for (let x of [0, 1, -1, 0x12345, 0x1234, -1000, -1000000]) {
+    assertEqVecArr(SIMD.Uint16x8.fromInt16x8Bits(splatI(x)), splat(x << 16 >>> 16));
+}
+
+splatUc = asmLink(asmCompile('glob', USE_ASM + U16x8 + I16x8 + I16x8U16x8 +
+                             'var splat = u16x8.splat;' +
+                             'function f() { var x = 200; return i16x8u16x8(splat(x)); } return f'), this);
+assertEqVecArr(SIMD.Uint16x8.fromInt16x8Bits(splatUc()), splat(200))
+
+
+// Test extractLane.
+//
+// The lane index must be a literal int, and we generate different code for
+// different lanes.
+function extractI(a, i) {
+  return asmLink(asmCompile('glob', USE_ASM + I16x8 + I16x8EXT +
+                            `function f() {var x=i16x8(${a.join(',')}); return i16x8ext(x, ${i})|0; } return f`), this)();
+}
+a = [-1,2,-3,4,-5,6,-7,-8];
+for (var i = 0; i < 8; i++)
+  assertEq(extractI(a, i), a[i]);
+a = a.map(x => -x);
+for (var i = 0; i < 8; i++)
+  assertEq(extractI(a, i), a[i]);
+
+function extractU(a, i) {
+  return asmLink(asmCompile('glob', USE_ASM + U16x8 + U16x8EXT +
+                            `function f() {var x=u16x8(${a.join(',')}); return u16x8ext(x, ${i})|0; } return f`), this)();
+}
+a = [1,255,12,13,14,150,200,3];
+for (var i = 0; i < 8; i++)
+  assertEq(extractU(a, i), a[i]);
+a = a.map(x => UINT16_MAX-x);
+for (var i = 0; i < 8; i++)
+  assertEq(extractU(a, i), a[i]);
+
+function extractB(a, i) {
+  return asmLink(asmCompile('glob', USE_ASM + B16x8 + B16x8EXT +
+                            `function f() {var x=b16x8(${a.join(',')}); return b16x8ext(x, ${i})|0; } return f`), this)();
+}
+a = [1,1,0,1, 1,0,0,0];
+for (var i = 0; i < 8; i++)
+  assertEq(extractB(a, i), a[i]);
+a = a.map(x => 1-x);
+for (var i = 0; i < 8; i++)
+  assertEq(extractB(a, i), a[i]);
+
+// Test replaceLane.
+function replaceI(a, i) {
+  return asmLink(asmCompile('glob', USE_ASM + I16x8 + I16x8REP +
+                            `function f(v) {v=v|0; var x=i16x8(${a.join(',')}); return i16x8rep(x,${i},v); } return f`), this);
+}
+a = [-1,2,-3,4,-5,6,-7,-9];
+for (var i = 0; i < 8; i++) {
+    var f = replaceI(a, i);
+    var b = a.slice(0);
+    b[i] = -20;
+    assertEqVecArr(f(-20), b);
+}
+
+function replaceU(a, i) {
+  return asmLink(asmCompile('glob', USE_ASM + U16x8 + U16x8REP + I16x8 + I16x8U16x8 +
+                            `function f(v) {v=v|0; var x=u16x8(${a.join(',')}); return i16x8u16x8(u16x8rep(x,${i},v)); } return f`), this);
+}
+a = [65000-1,2,65000-3,4,65000-5,6,65000-7,65000-9];
+for (var i = 0; i < 8; i++) {
+    var rawf = replaceU(a, i);
+    var f = x => SIMD.Uint16x8.fromInt16x8Bits(rawf(x))
+    var b = a.slice(0);
+    b[i] = 1000;
+    assertEqVecArr(f(1000), b);
+}
+
+function replaceB(a, i) {
+  return asmLink(asmCompile('glob', USE_ASM + B16x8 + B16x8REP +
+                            `function f(v) {v=v|0; var x=b16x8(${a.join(',')}); return b16x8rep(x,${i},v); } return f`), this);
+}
+a = [1,1,0,1,1,0,0,0];
+for (var i = 0; i < 8; i++) {
+    var f = replaceB(a, i);
+    var b = a.slice(0);
+    let v = 1 - a[i];
+    b[i] = v;
+    assertEqVecArr(f(v), b.map(x => !!x));
+}
+
+
+// Test select.
+selectI = asmLink(asmCompile('glob', USE_ASM + I16x8 + B16x8 + B16x8CHK +
+                             'var select = i16x8.select;' +
+                             'var a = i16x8(-1,2,-3,4,-5, 6,-7, 8);' +
+                             'var b = i16x8( 5,6, 7,8, 9,10,11,12);' +
+                             'function f(x) { x = b16x8chk(x); return select(x, a, b); } return f'), this);
+assertEqVecArr(selectI(SIMD.Bool16x8( 0,0, 1,0, 1,1, 1, 0)),
+                                    [ 5,6,-3,8,-5,6,-7,12]);
+
+selectU = asmLink(asmCompile('glob', USE_ASM + I16x8 + B16x8 + B16x8CHK + U16x8 + I16x8U16x8 + U16x8I16x8 +
+                             'var select = u16x8.select;' +
+                             'var a = i16x8(-1,2,-3,4,-5, 6,-7, 8);' +
+                             'var b = i16x8( 5,6, 7,8, 9,10,11,12);' +
+                             'function f(x) { x = b16x8chk(x); return i16x8u16x8(select(x, u16x8i16x8(a), u16x8i16x8(b))); } return f'), this);
+assertEqVecArr(selectU(SIMD.Bool16x8( 0,0, 1,0, 1,1, 1, 0)),
+                                    [ 5,6,-3,8,-5,6,-7,12]);
+
+// Test swizzle.
+function swizzle(vec, lanes) {
+    let r = [];
+    for (let i = 0; i < 8; i++)
+        r.push(vec[lanes[i]]);
+    return r;
+}
+
+function swizzleI(lanes) {
+    let asm = asmLink(asmCompile('glob', USE_ASM + I16x8 + I16x8CHK +
+                                 'var swz = i16x8.swizzle;' +
+                                 `function f(a) { a = i16x8chk(a); return swz(a, ${lanes.join()}); } return f`), this);
+    let a1 = [    -1,2,    -3,0x8000,0x7f,6,-7, 8].map(x => x << 16 >> 16);
+    let a2 = [0x8000,2,0x8000,0x7fff,   0,0, 8,-9].map(x => x << 16 >> 16);
+    let v1 = SIMD.Int16x8(...a1);
+    let v2 = SIMD.Int16x8(...a2);
+    assertEqVecArr(asm(v1), swizzle(a1, lanes));
+    assertEqVecArr(asm(v2), swizzle(a2, lanes));
+}
+
+swizzleI([3, 4, 7, 1, 4, 3, 1, 2]);
+swizzleI([0, 0, 0, 0, 0, 0, 0, 0]);
+swizzleI([7, 7, 7, 7, 7, 7, 7, 7]);
+
+function swizzleU(lanes) {
+    let asm = asmLink(asmCompile('glob', USE_ASM + I16x8 + I16x8CHK + U16x8 + U16x8I16x8 + I16x8U16x8 +
+                                 'var swz = u16x8.swizzle;' +
+                                 `function f(a) { a = i16x8chk(a); return i16x8u16x8(swz(u16x8i16x8(a), ${lanes.join()})); } return f`), this);
+    let a1 = [    -1,2,    -3,0x8000,0x7f,6,-7, 8].map(x => x << 16 >> 16);
+    let a2 = [0x8000,2,0x8000,0x7fff,   0,0, 8,-9].map(x => x << 16 >> 16);
+    let v1 = SIMD.Int16x8(...a1);
+    let v2 = SIMD.Int16x8(...a2);
+    assertEqVecArr(asm(v1), swizzle(a1, lanes));
+    assertEqVecArr(asm(v2), swizzle(a2, lanes));
+}
+
+swizzleU([3, 4, 7, 1, 4, 3, 1, 2]);
+swizzleU([0, 0, 0, 0, 0, 0, 0, 0]);
+swizzleU([7, 7, 7, 7, 7, 7, 7, 7]);
+
+// Out-of-range lane indexes.
+assertAsmTypeFail('glob', USE_ASM + I16x8 + 'var swz = i16x8.swizzle; ' +
+                  'function f() { var x=i16x8(0,0,0,0,0,0,0,0); swz(x,1,2,3,4,5,6,7,8); } return f');
+assertAsmTypeFail('glob', USE_ASM + U16x8 + 'var swz = u16x8.swizzle; ' +
+                  'function f() { var x=u16x8(0,0,0,0,0,0,0,0); swz(x,1,2,3,4,5,6,7,8); } return f');
+// Missing lane indexes.
+assertAsmTypeFail('glob', USE_ASM + I16x8 + 'var swz = i16x8.swizzle; ' +
+                  'function f() { var x=i16x8(0,0,0,0,0,0,0,0); swz(x,1,2,3,4,5,6,7); } return f');
+assertAsmTypeFail('glob', USE_ASM + U16x8 + 'var swz = u16x8.swizzle; ' +
+                  'function f() { var x=u16x8(0,0,0,0,0,0,0,0); swz(x,1,2,3,4,5,6,7); } return f');
+
+
+// Test shuffle.
+function shuffle(vec1, vec2, lanes) {
+    let r = [];
+    let vec = vec1.concat(vec2)
+    for (let i = 0; i < 8; i++)
+        r.push(vec[lanes[i]]);
+    return r;
+}
+
+function shuffleI(lanes) {
+    let asm = asmLink(asmCompile('glob', USE_ASM + I16x8 + I16x8CHK +
+                                 'var shuf = i16x8.shuffle;' +
+                                 `function f(a1, a2) { a1 = i16x8chk(a1); a2 = i16x8chk(a2); return shuf(a1, a2, ${lanes.join()}); } return f`), this);
+    let a1 = [    -1,2,    -3,0x8000,0x7f,6,-7, 8].map(x => x << 16 >> 16);
+    let a2 = [0x8000,2,0x8000,0x7fff,   0,0, 8,-9].map(x => x << 16 >> 16);
+    let v1 = SIMD.Int16x8(...a1);
+    let v2 = SIMD.Int16x8(...a2);
+    assertEqVecArr(asm(v1, v2), shuffle(a1, a2, lanes));
+}
+
+function shuffleU(lanes) {
+    let asm = asmLink(asmCompile('glob', USE_ASM + I16x8 + I16x8CHK + U16x8 + U16x8I16x8 + I16x8U16x8 +
+                                 'var shuf = u16x8.shuffle;' +
+                                 'function f(a1, a2) { a1 = i16x8chk(a1); a2 = i16x8chk(a2); ' +
+                                 `return i16x8u16x8(shuf(u16x8i16x8(a1), u16x8i16x8(a2), ${lanes.join()})); } return f`), this);
+    let a1 = [    -1,2,    -3,0x8000,0x7f,6,-7, 8].map(x => x << 16 >> 16);
+    let a2 = [0x8000,2,0x8000,0x7fff,   0,0, 8,-9].map(x => x << 16 >> 16);
+    let v1 = SIMD.Int16x8(...a1);
+    let v2 = SIMD.Int16x8(...a2);
+    assertEqVecArr(asm(v1, v2), shuffle(a1, a2, lanes));
+}
+
+shuffleI([0, 0, 0, 0, 0, 0, 0, 0])
+shuffleI([15, 15, 15, 15, 15, 15, 15, 15])
+shuffleI([6, 2, 0, 14, 6, 10, 11, 1])
+
+shuffleU([7, 7, 7, 7, 7, 7, 7, 7])
+shuffleU([8, 15, 15, 15, 15, 15, 15, 15])
+shuffleU([6, 2, 0, 14, 6, 10, 11, 1])
+
+// Test unary operators.
+function unaryI(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + I16x8 + I16x8CHK +
+                                      `var fut = i16x8.${opname};` +
+                                      'function f(v) { v = i16x8chk(v); return fut(v); } return f'), this);
+    let a = [65000-1,2,65000-3,4,65000-5,6,65000-7,65000-9];
+    let v = SIMD.Int16x8(...a);
+    assertEqVecArr(simdfunc(v), a.map(lanefunc));
+}
+
+function unaryU(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + U16x8 + I16x8 + I16x8CHK + U16x8I16x8 + I16x8U16x8 +
+                                      `var fut = u16x8.${opname};` +
+                                      'function f(v) { v = i16x8chk(v); return i16x8u16x8(fut(u16x8i16x8(v))); } return f'), this);
+    let a = [65000-1,2,65000-3,4,65000-5,6,65000-7,65000-9];
+    let v = SIMD.Int16x8(...a);
+    assertEqVecArr(SIMD.Uint16x8.fromInt16x8Bits(simdfunc(v)), a.map(lanefunc));
+}
+
+function unaryB(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + B16x8 + B16x8CHK +
+                                      `var fut = b16x8.${opname};` +
+                                      'function f(v) { v = b16x8chk(v); return fut(v); } return f'), this);
+    let a = [1,1,0,1,1,0,0,0];
+    let v = SIMD.Bool16x8(...a);
+    assertEqVecArr(simdfunc(v), a.map(lanefunc));
+}
+
+unaryI('not', x => ~x << 16 >> 16);
+unaryU('not', x => ~x << 16 >>> 16);
+unaryB('not', x => !x);
+unaryI('neg', x => -x << 16 >> 16);
+unaryU('neg', x => -x << 16 >>> 16);
+
+
+// Test binary operators.
+function zipmap(a1, a2, f) {
+    assertEq(a1.length, a2.length);
+    let r = [];
+    for (var i = 0; i < a1.length; i++)
+        r.push(f(a1[i], a2[i]));
+    return r
+}
+
+function binaryI(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + I16x8 + I16x8CHK +
+                                      `var fut = i16x8.${opname};` +
+                                      'function f(v1, v2) { v1 = i16x8chk(v1); v2 = i16x8chk(v2); return fut(v1, v2); } return f'), this);
+    let a1 = [    -1,2,    -3,0x8000,0x7f,6,-7, 8].map(x => x << 16 >> 16);
+    let a2 = [0x8000,2,0x8000,0x7fff,   0,0, 8,-9].map(x => x << 16 >> 16);
+    let ref = zipmap(a1, a2, lanefunc);
+    let v1 = SIMD.Int16x8(...a1);
+    let v2 = SIMD.Int16x8(...a2);
+    assertEqVecArr(simdfunc(v1, v2), ref);
+}
+
+function binaryU(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + U16x8 + I16x8 + I16x8CHK + U16x8I16x8 + I16x8U16x8 +
+                                      `var fut = u16x8.${opname};` +
+                                      'function f(v1, v2) { v1 = i16x8chk(v1); v2 = i16x8chk(v2); return i16x8u16x8(fut(u16x8i16x8(v1), u16x8i16x8(v2))); } return f'), this);
+    let a1 = [    -1,2,    -3,0x8000,0x7f,6,-7, 8].map(x => x << 16 >>> 16);
+    let a2 = [0x8000,2,0x8000,0x7fff,   0,0, 8,-9].map(x => x << 16 >>> 16);
+    let ref = zipmap(a1, a2, lanefunc);
+    let v1 = SIMD.Int16x8(...a1);
+    let v2 = SIMD.Int16x8(...a2);
+    let res = SIMD.Uint16x8.fromInt16x8Bits(simdfunc(v1, v2));
+    assertEqVecArr(res, ref);
+}
+
+function binaryB(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + B16x8 + B16x8CHK +
+                                      `var fut = b16x8.${opname};` +
+                                      'function f(v1, v2) { v1 = b16x8chk(v1); v2 = b16x8chk(v2); return fut(v1, v2); } return f'), this);
+    let a = [1,1,0,1,1,0,0,0];
+    let v = SIMD.Bool16x8(...a);
+    assertEqVecArr(simdfunc(v), a.map(lanefunc));
+}
+
+binaryI('add', (x, y) => (x + y) << 16 >> 16);
+binaryI('sub', (x, y) => (x - y) << 16 >> 16);
+binaryI('mul', (x, y) => (x * y) << 16 >> 16);
+binaryU('add', (x, y) => (x + y) << 16 >>> 16);
+binaryU('sub', (x, y) => (x - y) << 16 >>> 16);
+binaryU('mul', (x, y) => (x * y) << 16 >>> 16);
+
+binaryI('and', (x, y) => (x & y) << 16 >> 16);
+binaryI('or',  (x, y) => (x | y) << 16 >> 16);
+binaryI('xor', (x, y) => (x ^ y) << 16 >> 16);
+binaryU('and', (x, y) => (x & y) << 16 >>> 16);
+binaryU('or',  (x, y) => (x | y) << 16 >>> 16);
+binaryU('xor', (x, y) => (x ^ y) << 16 >>> 16);
+
+function sat(x, lo, hi) {
+    if (x < lo) return lo;
+    if (x > hi) return hi;
+    return x
+}
+function isat(x) { return sat(x, -32768, 32767); }
+function usat(x) { return sat(x, 0, 0xffff); }
+
+binaryI('addSaturate', (x, y) => isat(x + y))
+binaryI('subSaturate', (x, y) => isat(x - y))
+binaryU('addSaturate', (x, y) => usat(x + y))
+binaryU('subSaturate', (x, y) => usat(x - y))
+
+
+// Test shift operators.
+function zip1map(a, s, f) {
+    return a.map(x => f(x, s));
+}
+
+function shiftI(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + I16x8 + I16x8CHK +
+                                      `var fut = i16x8.${opname};` +
+                                      'function f(v, s) { v = i16x8chk(v); s = s|0; return fut(v, s); } return f'), this);
+    let a = [-1,2,-3,0x80,0x7f,6,0x8000,0x7fff];
+    let v = SIMD.Int16x8(...a);
+    for (let s of [0, 1, 2, 6, 7, 8, 9, 10, 16, 255, -1, -8, -7, -1000]) {
+        let ref = zip1map(a, s, lanefunc);
+        // 1. Test dynamic shift amount.
+        assertEqVecArr(simdfunc(v, s), ref);
+
+        // 2. Test constant shift amount.
+        let cstf = asmLink(asmCompile('glob', USE_ASM + I16x8 + I16x8CHK +
+                                      `var fut = i16x8.${opname};` +
+                                      `function f(v) { v = i16x8chk(v); return fut(v, ${s}); } return f`), this);
+        assertEqVecArr(cstf(v, s), ref);
+    }
+}
+
+function shiftU(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + U16x8 + I16x8 + I16x8CHK + U16x8I16x8 + I16x8U16x8 +
+                                      `var fut = u16x8.${opname};` +
+                                      'function f(v, s) { v = i16x8chk(v); s = s|0; return i16x8u16x8(fut(u16x8i16x8(v), s)); } return f'), this);
+    let a = [-1,2,-3,0x80,0x7f,6,0x8000,0x7fff];
+    let v = SIMD.Int16x8(...a);
+    for (let s of [0, 1, 2, 6, 7, 8, 9, 10, 16, 255, -1, -8, -7, -1000]) {
+        let ref = zip1map(a, s, lanefunc);
+        // 1. Test dynamic shift amount.
+        assertEqVecArr(SIMD.Uint16x8.fromInt16x8Bits(simdfunc(v, s)), ref);
+
+        // 2. Test constant shift amount.
+        let cstf = asmLink(asmCompile('glob', USE_ASM + U16x8 + I16x8 + I16x8CHK + U16x8I16x8 + I16x8U16x8 +
+                                      `var fut = u16x8.${opname};` +
+                                      `function f(v) { v = i16x8chk(v); return i16x8u16x8(fut(u16x8i16x8(v), ${s})); } return f`), this);
+        assertEqVecArr(SIMD.Uint16x8.fromInt16x8Bits(cstf(v, s)), ref);
+    }
+}
+
+shiftI('shiftLeftByScalar', (x,s) => (x << (s & 15)) << 16 >> 16);
+shiftU('shiftLeftByScalar', (x,s) => (x << (s & 15)) << 16 >>> 16);
+shiftI('shiftRightByScalar', (x,s) => ((x << 16 >> 16) >> (s & 15)) << 16 >> 16);
+shiftU('shiftRightByScalar', (x,s) => ((x << 16 >>> 16) >>> (s & 15)) << 16 >>> 16);
+
+
+// Comparisons.
+function compareI(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + I16x8 + I16x8CHK +
+                                      `var fut = i16x8.${opname};` +
+                                      'function f(v1, v2) { v1 = i16x8chk(v1); v2 = i16x8chk(v2); return fut(v1, v2); } return f'), this);
+    let a1 = [    -1,2,    -3,0x8000,0x7f,6,-7, 8].map(x => x << 16 >> 16);
+    let a2 = [0x8000,2,0x8000,0x7fff,   0,0, 8,-9].map(x => x << 16 >> 16);
+    let ref = zipmap(a1, a2, lanefunc);
+    let v1 = SIMD.Int16x8(...a1);
+    let v2 = SIMD.Int16x8(...a2);
+    assertEqVecArr(simdfunc(v1, v2), ref);
+}
+
+function compareU(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + I16x8 + I16x8CHK + U16x8 + U16x8I16x8 +
+                                      `var fut = u16x8.${opname};` +
+                                      'function f(v1, v2) { v1 = i16x8chk(v1); v2 = i16x8chk(v2); return fut(u16x8i16x8(v1), u16x8i16x8(v2)); } return f'), this);
+    let a1 = [    -1,2,    -3,0x8000,0x7f,6,-7, 8].map(x => x << 16 >>> 16);
+    let a2 = [0x8000,2,0x8000,0x7fff,   0,0, 8,-9].map(x => x << 16 >>> 16);
+    let ref = zipmap(a1, a2, lanefunc);
+    let v1 = SIMD.Int16x8(...a1);
+    let v2 = SIMD.Int16x8(...a2);
+    assertEqVecArr(simdfunc(v1, v2), ref);
+}
+
+compareI("equal", (x,y) => x == y);
+compareU("equal", (x,y) => x == y);
+compareI("notEqual", (x,y) => x != y);
+compareU("notEqual", (x,y) => x != y);
+compareI("lessThan", (x,y) => x < y);
+compareU("lessThan", (x,y) => x < y);
+compareI("lessThanOrEqual", (x,y) => x <= y);
+compareU("lessThanOrEqual", (x,y) => x <= y);
+compareI("greaterThan", (x,y) => x > y);
+compareU("greaterThan", (x,y) => x > y);
+compareI("greaterThanOrEqual", (x,y) => x >= y);
+compareU("greaterThanOrEqual", (x,y) => x >= y);

js/src/jit-test/tests/asm.js/testSIMD-8x16.js

@@ -0,0 +1,539 @@
+load(libdir + "asm.js");
+load(libdir + "simd.js");
+load(libdir + "asserts.js");
+
+// Set to true to see more JS debugging spew.
+const DEBUG = false;
+
+if (!isSimdAvailable()) {
+    DEBUG && print("won't run tests as simd extensions aren't activated yet");
+    quit(0);
+}
+
+// Tests for 8x16 SIMD types: Int8x16, Uint8x16, Bool8x16.
+
+const I8x16 = 'var i8x16 = glob.SIMD.Int8x16;'
+const I8x16CHK = 'var i8x16chk = i8x16.check;'
+const I8x16EXT = 'var i8x16ext = i8x16.extractLane;'
+const I8x16REP = 'var i8x16rep = i8x16.replaceLane;'
+const I8x16U8x16 = 'var i8x16u8x16 = i8x16.fromUint8x16Bits;'
+
+const U8x16 = 'var u8x16 = glob.SIMD.Uint8x16;'
+const U8x16CHK = 'var u8x16chk = u8x16.check;'
+const U8x16EXT = 'var u8x16ext = u8x16.extractLane;'
+const U8x16REP = 'var u8x16rep = u8x16.replaceLane;'
+const U8x16I8x16 = 'var u8x16i8x16 = u8x16.fromInt8x16Bits;'
+
+const B8x16 = 'var b8x16 = glob.SIMD.Bool8x16;'
+const B8x16CHK = 'var b8x16chk = b8x16.check;'
+const B8x16EXT = 'var b8x16ext = b8x16.extractLane;'
+const B8x16REP = 'var b8x16rep = b8x16.replaceLane;'
+
+const INT8_MAX = 127
+const INT8_MIN = -128
+const UINT8_MAX = 255
+
+// Linking
+assertEq(asmLink(asmCompile('glob', USE_ASM + I8x16 + "function f() {} return f"), {SIMD:{Int8x16: SIMD.Int8x16}})(), undefined);
+assertEq(asmLink(asmCompile('glob', USE_ASM + U8x16 + "function f() {} return f"), {SIMD:{Uint8x16: SIMD.Uint8x16}})(), undefined);
+assertEq(asmLink(asmCompile('glob', USE_ASM + B8x16 + "function f() {} return f"), {SIMD:{Bool8x16: SIMD.Bool8x16}})(), undefined);
+
+// Local variable of Int8x16 type.
+assertAsmTypeFail('glob', USE_ASM + "function f() {var x=Int8x16(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16);} return f");
+assertAsmTypeFail('glob', USE_ASM + I8x16 + "function f() {var x=i8x16;} return f");
+assertAsmTypeFail('glob', USE_ASM + I8x16 + "function f() {var x=i8x16();} return f");
+assertAsmTypeFail('glob', USE_ASM + I8x16 + "function f() {var x=i8x16(1);} return f");
+assertAsmTypeFail('glob', USE_ASM + I8x16 + "function f() {var x=i8x16(1,2,3,4);} return f");
+assertAsmTypeFail('glob', USE_ASM + I8x16 + "function f() {var x=i8x16(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16.0);} return f");
+assertAsmTypeFail('glob', USE_ASM + I8x16 + "function f() {var x=i8x16(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17);} return f");
+assertAsmTypeFail('glob', USE_ASM + I8x16 + "function f() {var x=i8x16(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16|0);} return f");
+assertEq(asmLink(asmCompile('glob', USE_ASM + I8x16 + "function f() {var x=i8x16(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16);} return f"), this)(), undefined);
+assertEq(asmLink(asmCompile('glob', USE_ASM + I8x16 + "function f() {var x=i8x16(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15," + (INT8_MAX + 1) + ");} return f"), this)(), undefined);
+
+// Local variable of Uint8x16 type.
+assertAsmTypeFail('glob', USE_ASM + "function f() {var x=Uint8x16(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16);} return f");
+assertAsmTypeFail('glob', USE_ASM + U8x16 + "function f() {var x=u8x16;} return f");
+assertAsmTypeFail('glob', USE_ASM + U8x16 + "function f() {var x=u8x16();} return f");
+assertAsmTypeFail('glob', USE_ASM + U8x16 + "function f() {var x=u8x16(1);} return f");
+assertAsmTypeFail('glob', USE_ASM + U8x16 + "function f() {var x=u8x16(1,2,3,4);} return f");
+assertAsmTypeFail('glob', USE_ASM + U8x16 + "function f() {var x=u8x16(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16.0);} return f");
+assertAsmTypeFail('glob', USE_ASM + U8x16 + "function f() {var x=u8x16(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17);} return f");
+assertAsmTypeFail('glob', USE_ASM + U8x16 + "function f() {var x=u8x16(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16|0);} return f");
+assertEq(asmLink(asmCompile('glob', USE_ASM + U8x16 + "function f() {var x=u8x16(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16);} return f"), this)(), undefined);
+assertEq(asmLink(asmCompile('glob', USE_ASM + U8x16 + "function f() {var x=u8x16(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15," + (UINT8_MAX + 1) + ");} return f"), this)(), undefined);
+
+// Local variable of Bool8x16 type.
+assertAsmTypeFail('glob', USE_ASM + "function f() {var x=Bool8x16(1,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1);} return f");
+assertAsmTypeFail('glob', USE_ASM + B8x16 + "function f() {var x=b8x16;} return f");
+assertAsmTypeFail('glob', USE_ASM + B8x16 + "function f() {var x=b8x16();} return f");
+assertAsmTypeFail('glob', USE_ASM + B8x16 + "function f() {var x=b8x16(1);} return f");
+assertAsmTypeFail('glob', USE_ASM + B8x16 + "function f() {var x=b8x16(1,0,0,0);} return f");
+assertAsmTypeFail('glob', USE_ASM + B8x16 + "function f() {var x=b8x16(1,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1.0);} return f");
+assertAsmTypeFail('glob', USE_ASM + B8x16 + "function f() {var x=b8x16(1,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1|0);} return f");
+assertAsmTypeFail('glob', USE_ASM + B8x16 + "function f() {var x=b8x16(1,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1);} return f");
+assertEq(asmLink(asmCompile('glob', USE_ASM + B8x16 + "function f() {var x=b8x16(1,0,0,0,0,0,0,0,0,1,-1,2,-2,1,1,1);} return f"), this)(), undefined);
+
+// Global variable of Int8x16 type.
+assertEqVecArr(asmLink(asmCompile('glob', 'ffi', USE_ASM + I8x16 + I8x16CHK + "var g=i8x16chk(ffi.g); function f() { return i8x16chk(g); } return f"), this,
+                       {g: SIMD.Int8x16(1,2,3,4,5,6,7,8,10,11,12,13,14,15,16,17)})(), [1,2,3,4,5,6,7,8,10,11,12,13,14,15,16,17]);
+assertEqVecArr(asmLink(asmCompile('glob', 'ffi', USE_ASM + I8x16 + I8x16CHK + "var g=i8x16chk(ffi.g); function f() { g=i8x16(5,6,7,8,9,10,11,12,1,2,3,4,5,6,7,8); return i8x16chk(g); } return f"), this,
+                       {g: SIMD.Int8x16(1,2,3,4,5,6,7,8,10,11,12,13,14,15,16,17)})(), [5,6,7,8,9,10,11,12,1,2,3,4,5,6,7,8]);
+
+// Global variable of Bool8x16 type.
+assertEqVecArr(asmLink(asmCompile('glob', 'ffi', USE_ASM + B8x16 + B8x16CHK + "var g=b8x16chk(ffi.g); function f() { return b8x16chk(g); } return f"), this,
+                       {g: SIMD.Bool8x16(1,1,0,1,0,0,1,0,0,1,0,1,0,0,1,0)})(), [true,true,false,true,false,false,true,false,false,true,false,true,false,false,true,false]);
+assertEqVecArr(asmLink(asmCompile('glob', 'ffi', USE_ASM + B8x16 + B8x16CHK + "var g=b8x16chk(ffi.g); function f() { g=b8x16(1,1,0,1,0,1,1,1,0,1,0,1,1,1,0,0); return b8x16chk(g); } return f"), this,
+                       {g: SIMD.Bool8x16(1,1,0,1,0,0,1,0)})(), [true,true,false,true,false,true,true,true,false,true,false,true,true,true,false,false]);
+
+// Unsigned SIMD globals are not allowed.
+assertAsmTypeFail('glob', 'ffi', USE_ASM + U8x16 + U8x16CHK + "var g=u8x16chk(ffi.g); function f() { } return f");
+
+// Only signed Int8x16 allowed as return value.
+assertEqVecArr(asmLink(asmCompile('glob', USE_ASM + I8x16 + "function f() {return i8x16(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16);} return f"), this)(),
+           [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16]);
+assertEqVecArr(asmLink(asmCompile('glob', USE_ASM + I8x16 + I8x16CHK + "function f() {return i8x16chk(i8x16(1,2,3,132,5,6,7,8,9,10,11,12,13,14,15,16));} return f"), this)(),
+           [1, 2, 3, -124, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16]);
+assertAsmTypeFail('glob', USE_ASM + U8x16 + "function f() {return u8x16(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16);} return f");
+assertAsmTypeFail('glob', USE_ASM + U8x16 + U8x16CHK + "function f() {return u8x16chk(u8x16(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16));} return f");
+
+// Test splat.
+function splat(x) {
+    let r = []
+    for (let i = 0; i < 16; i++)
+        r.push(x);
+    return r
+}
+
+splatB = asmLink(asmCompile('glob', USE_ASM + B8x16 +
+                            'var splat = b8x16.splat;' +
+                            'function f(x) { x = x|0; return splat(x); } return f'), this);
+assertEqVecArr(splatB(true), splat(true));
+assertEqVecArr(splatB(false), splat(false));
+
+
+splatB0 = asmLink(asmCompile('glob', USE_ASM + B8x16 +
+                             'var splat = b8x16.splat;' +
+                             'function f() { var x = 0; return splat(x); } return f'), this);
+assertEqVecArr(splatB0(), splat(false));
+splatB1 = asmLink(asmCompile('glob', USE_ASM + B8x16 +
+                             'var splat = b8x16.splat;' +
+                             'function f() { var x = 1; return splat(x); } return f'), this);
+assertEqVecArr(splatB1(), splat(true));
+
+splatI = asmLink(asmCompile('glob', USE_ASM + I8x16 +
+                            'var splat = i8x16.splat;' +
+                            'function f(x) { x = x|0; return splat(x); } return f'), this);
+for (let x of [0, 1, -1, 0x1234, 0x12, 1000, -1000000]) {
+    assertEqVecArr(splatI(x), splat(x << 24 >> 24));
+}
+
+splatIc = asmLink(asmCompile('glob', USE_ASM + I8x16 +
+                             'var splat = i8x16.splat;' +
+                             'function f() { var x = 100; return splat(x); } return f'), this);
+assertEqVecArr(splatIc(), splat(100))
+
+splatU = asmLink(asmCompile('glob', USE_ASM + U8x16 + I8x16 + I8x16U8x16 +
+                            'var splat = u8x16.splat;' +
+                            'function f(x) { x = x|0; return i8x16u8x16(splat(x)); } return f'), this);
+for (let x of [0, 1, -1, 0x1234, 0x12, 1000, -1000000]) {
+    assertEqVecArr(SIMD.Uint8x16.fromInt8x16Bits(splatI(x)), splat(x << 24 >>> 24));
+}
+
+splatUc = asmLink(asmCompile('glob', USE_ASM + U8x16 + I8x16 + I8x16U8x16 +
+                            'var splat = u8x16.splat;' +
+                            'function f() { var x = 200; return i8x16u8x16(splat(x)); } return f'), this);
+assertEqVecArr(SIMD.Uint8x16.fromInt8x16Bits(splatUc()), splat(200))
+
+
+// Test extractLane.
+//
+// The lane index must be a literal int, and we generate different code for
+// different lanes.
+function extractI(a, i) {
+  return asmLink(asmCompile('glob', USE_ASM + I8x16 + I8x16EXT +
+                            `function f() {var x=i8x16(${a.join(',')}); return i8x16ext(x, ${i})|0; } return f`), this)();
+}
+a = [-1,2,-3,4,-5,6,-7,8,-9,10,-11,12,-13,-14,-15,-16];
+for (var i = 0; i < 16; i++)
+  assertEq(extractI(a, i), a[i]);
+a = a.map(x => -x);
+for (var i = 0; i < 16; i++)
+  assertEq(extractI(a, i), a[i]);
+
+function extractU(a, i) {
+  return asmLink(asmCompile('glob', USE_ASM + U8x16 + U8x16EXT +
+                            `function f() {var x=u8x16(${a.join(',')}); return u8x16ext(x, ${i})|0; } return f`), this)();
+}
+a = [1,255,12,13,14,150,200,3,4,5,6,7,8,9,10,16];
+for (var i = 0; i < 16; i++)
+  assertEq(extractU(a, i), a[i]);
+a = a.map(x => 255-x);
+for (var i = 0; i < 16; i++)
+  assertEq(extractU(a, i), a[i]);
+
+function extractB(a, i) {
+  return asmLink(asmCompile('glob', USE_ASM + B8x16 + B8x16EXT +
+                            `function f() {var x=b8x16(${a.join(',')}); return b8x16ext(x, ${i})|0; } return f`), this)();
+}
+a = [1,1,0,1,1,0,0,0,1,1,1,1,0,0,0,1];
+for (var i = 0; i < 16; i++)
+  assertEq(extractB(a, i), a[i]);
+a = a.map(x => 1-x);
+for (var i = 0; i < 16; i++)
+  assertEq(extractB(a, i), a[i]);
+
+// Test replaceLane.
+function replaceI(a, i) {
+  return asmLink(asmCompile('glob', USE_ASM + I8x16 + I8x16REP +
+                            `function f(v) {v=v|0; var x=i8x16(${a.join(',')}); return i8x16rep(x,${i},v); } return f`), this);
+}
+a = [-1,2,-3,4,-5,6,-7,8,-9,10,-11,12,-13,-14,-15,-16];
+for (var i = 0; i < 16; i++) {
+    var f = replaceI(a, i);
+    var b = a.slice(0);
+    b[i] = -20;
+    assertEqVecArr(f(-20), b);
+}
+
+function replaceU(a, i) {
+  return asmLink(asmCompile('glob', USE_ASM + U8x16 + U8x16REP + I8x16 + I8x16U8x16 +
+                            `function f(v) {v=v|0; var x=u8x16(${a.join(',')}); x=u8x16rep(x,${i},v); return i8x16u8x16(x); } return f`), this);
+}
+a = [256-1,2,256-3,4,256-5,6,256-7,8,256-9,10,256-11,12,256-13,256-14,256-15,256-16];
+for (var i = 0; i < 16; i++) {
+    // Result returned as Int8x16, convert back.
+    var rawf = replaceU(a, i);
+    var f = x => SIMD.Uint8x16.fromInt8x16Bits(rawf(x));
+    var b = a.slice(0);
+    b[i] = 100;
+    assertEqVecArr(f(100), b);
+}
+
+function replaceB(a, i) {
+  return asmLink(asmCompile('glob', USE_ASM + B8x16 + B8x16REP +
+                            `function f(v) {v=v|0; var x=b8x16(${a.join(',')}); return b8x16rep(x,${i},v); } return f`), this);
+}
+a = [1,1,0,1,1,0,0,0,1,1,1,1,0,0,0,1];
+for (var i = 0; i < 16; i++) {
+    var f = replaceB(a, i);
+    var b = a.slice(0);
+    v = 1 - a[i];
+    b[i] = v;
+    assertEqVecArr(f(v), b.map(x => !!x));
+}
+
+
+// Test select.
+selectI = asmLink(asmCompile('glob', USE_ASM + I8x16 + B8x16 + B8x16CHK +
+                             'var select = i8x16.select;' +
+                             'var a = i8x16(-1,2,-3,4,-5, 6,-7, 8,-9,10,-11,12,-13,-14,-15,-16);' +
+                             'var b = i8x16( 5,6, 7,8, 9,10,11,12,13,14, 15,16,-77, 45, 32,  0);' +
+                             'function f(x) { x = b8x16chk(x); return select(x, a, b); } return f'), this);
+assertEqVecArr(selectI(SIMD.Bool8x16( 0,0, 1,0, 1,1, 1, 0, 1, 1, 0, 0,  1,  1, 0,  1)),
+                                    [ 5,6,-3,8,-5,6,-7,12,-9,10,15,16,-13,-14,32,-16]);
+
+selectU = asmLink(asmCompile('glob', USE_ASM + I8x16 + B8x16 + B8x16CHK + U8x16 + I8x16U8x16 + U8x16I8x16 +
+                             'var select = u8x16.select;' +
+                             'var a = i8x16(-1,2,-3,4,-5, 6,-7, 8,-9,10,-11,12,-13,-14,-15,-16);' +
+                             'var b = i8x16( 5,6, 7,8, 9,10,11,12,13,14, 15,16,-77, 45, 32,  0);' +
+                             'function f(x) { x = b8x16chk(x); return i8x16u8x16(select(x, u8x16i8x16(a), u8x16i8x16(b))); } return f'), this);
+assertEqVecArr(selectU(SIMD.Bool8x16( 0,0, 1,0, 1,1, 1, 0, 1, 1, 0, 0,  1,  1, 0,  1)),
+                                    [ 5,6,-3,8,-5,6,-7,12,-9,10,15,16,-13,-14,32,-16]);
+
+
+// Test swizzle.
+function swizzle(vec, lanes) {
+    let r = [];
+    for (let i = 0; i < 16; i++)
+        r.push(vec[lanes[i]]);
+    return r;
+}
+
+function swizzleI(lanes) {
+    let asm = asmLink(asmCompile('glob', USE_ASM + I8x16 + I8x16CHK +
+                                 'var swz = i8x16.swizzle;' +
+                                 `function f(a) { a = i8x16chk(a); return swz(a, ${lanes.join()}); } return f`), this);
+    let a1 = [  -1,2,  -3,-128,0x7f,6,-7, 8,-9, 10,-11, 12,-13,-14,-15, -16];
+    let a2 = [-128,2,-128,0x7f,   0,0, 8,-9,10,-11, 12,-13,-14,-15,-16,  -1];
+    let v1 = SIMD.Int8x16(...a1);
+    let v2 = SIMD.Int8x16(...a2);
+    assertEqVecArr(asm(v1), swizzle(a1, lanes));
+    assertEqVecArr(asm(v2), swizzle(a2, lanes));
+}
+
+swizzleI([10, 1, 7, 5, 1, 2, 6, 8, 5, 13, 0, 6, 2, 8, 0, 9]);
+swizzleI([ 0, 0, 0, 0, 0, 0, 0, 0, 0,  0, 0, 0, 0, 0, 0, 0]);
+swizzleI([15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15]);
+
+function swizzleU(lanes) {
+    let asm = asmLink(asmCompile('glob', USE_ASM + I8x16 + I8x16CHK + U8x16 + U8x16I8x16 + I8x16U8x16 +
+                                 'var swz = u8x16.swizzle;' +
+                                 `function f(a) { a = i8x16chk(a); return i8x16u8x16(swz(u8x16i8x16(a), ${lanes.join()})); } return f`), this);
+    let a1 = [  -1,2,  -3,-128,0x7f,6,-7, 8,-9, 10,-11, 12,-13,-14,-15, -16];
+    let a2 = [-128,2,-128,0x7f,   0,0, 8,-9,10,-11, 12,-13,-14,-15,-16,  -1];
+    let v1 = SIMD.Int8x16(...a1);
+    let v2 = SIMD.Int8x16(...a2);
+    assertEqVecArr(asm(v1), swizzle(a1, lanes));
+    assertEqVecArr(asm(v2), swizzle(a2, lanes));
+}
+
+swizzleU([10, 1, 7, 5, 1, 2, 6, 8, 5, 13, 0, 6, 2, 8, 0, 9]);
+swizzleU([ 0, 0, 0, 0, 0, 0, 0, 0, 0,  0, 0, 0, 0, 0, 0, 0]);
+swizzleU([15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15]);
+
+// Out-of-range lane indexes.
+assertAsmTypeFail('glob', USE_ASM + I8x16 + 'var swz = i8x16.swizzle; ' +
+                  'function f() { var x=i8x16(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0); swz(x,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16); } return f');
+assertAsmTypeFail('glob', USE_ASM + U8x16 + 'var swz = u8x16.swizzle; ' +
+                  'function f() { var x=u8x16(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0); swz(x,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16); } return f');
+// Missing lane indexes.
+assertAsmTypeFail('glob', USE_ASM + I8x16 + 'var swz = i8x16.swizzle; ' +
+                  'function f() { var x=i8x16(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0); swz(x,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15); } return f');
+assertAsmTypeFail('glob', USE_ASM + U8x16 + 'var swz = u8x16.swizzle; ' +
+                  'function f() { var x=u8x16(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0); swz(x,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15); } return f');
+
+
+// Test shuffle.
+function shuffle(vec1, vec2, lanes) {
+    let r = [];
+    let vec = vec1.concat(vec2);
+    for (let i = 0; i < 16; i++)
+        r.push(vec[lanes[i]]);
+    return r;
+}
+
+function shuffleI(lanes) {
+    let asm = asmLink(asmCompile('glob', USE_ASM + I8x16 + I8x16CHK +
+                                 'var shuf = i8x16.shuffle;' +
+                                 `function f(a1, a2) { a1 = i8x16chk(a1); a2 = i8x16chk(a2); return shuf(a1, a2, ${lanes.join()}); } return f`), this);
+    let a1 = [  -1,2,  -3,-128,0x7f,6,-7, 8,-9, 10,-11, 12,-13,-14,-15, -16];
+    let a2 = [-128,2,-128,0x7f,   0,0, 8,-9,10,-11, 12,-13,-14,-15,-16,  -1];
+    let v1 = SIMD.Int8x16(...a1);
+    let v2 = SIMD.Int8x16(...a2);
+    assertEqVecArr(asm(v1, v2), shuffle(a1, a2, lanes));
+}
+
+shuffleI([31, 9, 5, 4, 29, 12, 19, 10, 16, 22, 10, 9, 6, 18, 9, 8]);
+shuffleI([ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]);
+shuffleI([31, 31, 31, 31, 31, 31, 31, 31, 31, 31, 31, 31, 31, 31, 31, 31]);
+
+function shuffleU(lanes) {
+    let asm = asmLink(asmCompile('glob', USE_ASM + I8x16 + I8x16CHK + U8x16 + U8x16I8x16 + I8x16U8x16 +
+                                 'var shuf = u8x16.shuffle;' +
+                                 'function f(a1, a2) { a1 = i8x16chk(a1); a2 = i8x16chk(a2); ' +
+                                 `return i8x16u8x16(shuf(u8x16i8x16(a1), u8x16i8x16(a2), ${lanes.join()})); } return f`), this);
+    let a1 = [  -1,2,  -3,-128,0x7f,6,-7, 8,-9, 10,-11, 12,-13,-14,-15, -16];
+    let a2 = [-128,2,-128,0x7f,   0,0, 8,-9,10,-11, 12,-13,-14,-15,-16,  -1];
+    let v1 = SIMD.Int8x16(...a1);
+    let v2 = SIMD.Int8x16(...a2);
+    assertEqVecArr(asm(v1, v2), shuffle(a1, a2, lanes));
+}
+
+shuffleU([31, 9, 5, 4, 29, 12, 19, 10, 16, 22, 10, 9, 6, 18, 9, 8]);
+shuffleU([ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]);
+shuffleU([31, 31, 31, 31, 31, 31, 31, 31, 31, 31, 31, 31, 31, 31, 31, 31]);
+
+
+// Out-of-range lane indexes.
+assertAsmTypeFail('glob', USE_ASM + I8x16 + 'var shuf = i8x16.shuffle; ' +
+                  'function f() { var x=i8x16(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0); shuf(x,x,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,32); } return f');
+assertAsmTypeFail('glob', USE_ASM + U8x16 + 'var shuf = u8x16.shuffle; ' +
+                  'function f() { var x=u8x16(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0); shuf(x,x,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,32); } return f');
+// Missing lane indexes.
+assertAsmTypeFail('glob', USE_ASM + I8x16 + 'var shuf = i8x16.shuffle; ' +
+                  'function f() { var x=i8x16(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0); shuf(x,x,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15); } return f');
+assertAsmTypeFail('glob', USE_ASM + U8x16 + 'var shuf = u8x16.shuffle; ' +
+                  'function f() { var x=u8x16(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0); shuf(x,x,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15); } return f');
+
+
+// Test unary operators.
+function unaryI(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + I8x16 + I8x16CHK +
+                                      `var fut = i8x16.${opname};` +
+                                      'function f(v) { v = i8x16chk(v); return fut(v); } return f'), this);
+    let a = [-1,2,-3,4,-5,6,-7,8,-9,10,-11,12,-13,-14,-15,-16];
+    let v = SIMD.Int8x16(...a);
+    assertEqVecArr(simdfunc(v), a.map(lanefunc));
+}
+
+function unaryU(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + U8x16 + I8x16 + I8x16CHK + U8x16I8x16 + I8x16U8x16 +
+                                      `var fut = u8x16.${opname};` +
+                                      'function f(v) { v = i8x16chk(v); return i8x16u8x16(fut(u8x16i8x16(v))); } return f'), this);
+    let a = [256-1,2,256-3,4,256-5,6,256-7,8,256-9,10,256-11,12,256-13,256-14,256-15,256-16];
+    let v = SIMD.Int8x16(...a);
+    assertEqVecArr(SIMD.Uint8x16.fromInt8x16Bits(simdfunc(v)), a.map(lanefunc));
+}
+
+function unaryB(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + B8x16 + B8x16CHK +
+                                      `var fut = b8x16.${opname};` +
+                                      'function f(v) { v = b8x16chk(v); return fut(v); } return f'), this);
+    let a = [1,1,0,1,1,0,0,0,1,1,1,1,0,0,0,1];
+    let v = SIMD.Bool8x16(...a);
+    assertEqVecArr(simdfunc(v), a.map(lanefunc));
+}
+
+unaryI('not', x => ~x << 24 >> 24);
+unaryU('not', x => ~x << 24 >>> 24);
+unaryB('not', x => !x);
+unaryI('neg', x => -x << 24 >> 24);
+unaryU('neg', x => -x << 24 >>> 24);
+
+
+// Test binary operators.
+function zipmap(a1, a2, f) {
+    assertEq(a1.length, a2.length);
+    let r = [];
+    for (var i = 0; i < a1.length; i++)
+        r.push(f(a1[i], a2[i]));
+    return r
+}
+
+function binaryI(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + I8x16 + I8x16CHK +
+                                      `var fut = i8x16.${opname};` +
+                                      'function f(v1, v2) { v1 = i8x16chk(v1); v2 = i8x16chk(v2); return fut(v1, v2); } return f'), this);
+    let a1 = [  -1,2,  -3,-128,0x7f,6,-7, 8,-9, 10,-11, 12,-13,-14,-15, -16];
+    let a2 = [-128,2,-128,0x7f,   0,0, 8,-9,10,-11, 12,-13,-14,-15,-16,  -1];
+    let ref = zipmap(a1, a2, lanefunc);
+    let v1 = SIMD.Int8x16(...a1);
+    let v2 = SIMD.Int8x16(...a2);
+    assertEqVecArr(simdfunc(v1, v2), ref);
+}
+
+function binaryU(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + U8x16 + I8x16 + I8x16CHK + U8x16I8x16 + I8x16U8x16 +
+                                      `var fut = u8x16.${opname};` +
+                                      'function f(v1, v2) { v1 = i8x16chk(v1); v2 = i8x16chk(v2); return i8x16u8x16(fut(u8x16i8x16(v1), u8x16i8x16(v2))); } return f'), this);
+    let a1 = [  -1,2,  -3,0x80,0x7f,6,-7, 8,-9, 10,-11, 12,-13,-14,-15, -16].map(x => x & 0xff);
+    let a2 = [0x80,2,0x80,0x7f,   0,0, 8,-9,10,-11, 12,-13,-14,-15,-16,0xff].map(x => x & 0xff);
+    let ref = zipmap(a1, a2, lanefunc);
+    let v1 = SIMD.Int8x16(...a1);
+    let v2 = SIMD.Int8x16(...a2);
+    let res = SIMD.Uint8x16.fromInt8x16Bits(simdfunc(v1, v2));
+    assertEqVecArr(res, ref);
+}
+
+function binaryB(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + B8x16 + B8x16CHK +
+                                      `var fut = b8x16.${opname};` +
+                                      'function f(v1, v2) { v1 = b8x16chk(v1); v2 = b8x16chk(v2); return fut(v1, v2); } return f'), this);
+    let a = [1,1,0,1,1,0,0,0,1,1,1,1,0,0,0,1];
+    let v = SIMD.Bool8x16(...a);
+    assertEqVecArr(simdfunc(v), a.map(lanefunc));
+}
+
+binaryI('add', (x, y) => (x + y) << 24 >> 24);
+binaryI('sub', (x, y) => (x - y) << 24 >> 24);
+binaryI('mul', (x, y) => (x * y) << 24 >> 24);
+binaryU('add', (x, y) => (x + y) << 24 >>> 24);
+binaryU('sub', (x, y) => (x - y) << 24 >>> 24);
+binaryU('mul', (x, y) => (x * y) << 24 >>> 24);
+
+binaryI('and', (x, y) => (x & y) << 24 >> 24);
+binaryI('or',  (x, y) => (x | y) << 24 >> 24);
+binaryI('xor', (x, y) => (x ^ y) << 24 >> 24);
+binaryU('and', (x, y) => (x & y) << 24 >>> 24);
+binaryU('or',  (x, y) => (x | y) << 24 >>> 24);
+binaryU('xor', (x, y) => (x ^ y) << 24 >>> 24);
+
+function sat(x, lo, hi) {
+    if (x < lo) return lo;
+    if (x > hi) return hi;
+    return x
+}
+function isat(x) { return sat(x, -128, 127); }
+function usat(x) { return sat(x, 0, 255); }
+
+binaryI('addSaturate', (x, y) => isat(x + y))
+binaryI('subSaturate', (x, y) => isat(x - y))
+binaryU('addSaturate', (x, y) => usat(x + y))
+binaryU('subSaturate', (x, y) => usat(x - y))
+
+// Test shift operators.
+function zip1map(a, s, f) {
+    return a.map(x => f(x, s));
+}
+
+function shiftI(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + I8x16 + I8x16CHK +
+                                      `var fut = i8x16.${opname};` +
+                                      'function f(v, s) { v = i8x16chk(v); s = s|0; return fut(v, s); } return f'), this);
+    let a = [0x80,2,0x80,0x7f,   0,0, 8,-9,10,-11, 12,-13,-14,-15,-16,0xff];
+    let v = SIMD.Int8x16(...a);
+    for (let s of [0, 1, 2, 6, 7, 8, 9, 10, 16, 255, -1, -8, -7, -1000]) {
+        let ref = zip1map(a, s, lanefunc);
+        // 1. Test dynamic shift amount.
+        assertEqVecArr(simdfunc(v, s), ref);
+
+        // 2. Test constant shift amount.
+        let cstf = asmLink(asmCompile('glob', USE_ASM + I8x16 + I8x16CHK +
+                                      `var fut = i8x16.${opname};` +
+                                      `function f(v) { v = i8x16chk(v); return fut(v, ${s}); } return f`), this);
+        assertEqVecArr(cstf(v, s), ref);
+    }
+}
+
+function shiftU(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + U8x16 + I8x16 + I8x16CHK + U8x16I8x16 + I8x16U8x16 +
+                                      `var fut = u8x16.${opname};` +
+                                      'function f(v, s) { v = i8x16chk(v); s = s|0; return i8x16u8x16(fut(u8x16i8x16(v), s)); } return f'), this);
+    let a = [0x80,2,0x80,0x7f,   0,0, 8,-9,10,-11, 12,-13,-14,-15,-16,0xff];
+    let v = SIMD.Int8x16(...a);
+    for (let s of [0, 1, 2, 6, 7, 8, 9, 10, 16, 255, -1, -8, -7, -1000]) {
+        let ref = zip1map(a, s, lanefunc);
+        // 1. Test dynamic shift amount.
+        assertEqVecArr(SIMD.Uint8x16.fromInt8x16Bits(simdfunc(v, s)), ref);
+
+        // 2. Test constant shift amount.
+        let cstf = asmLink(asmCompile('glob', USE_ASM + U8x16 + I8x16 + I8x16CHK + U8x16I8x16 + I8x16U8x16 +
+                                      `var fut = u8x16.${opname};` +
+                                      `function f(v) { v = i8x16chk(v); return i8x16u8x16(fut(u8x16i8x16(v), ${s})); } return f`), this);
+        assertEqVecArr(SIMD.Uint8x16.fromInt8x16Bits(cstf(v, s)), ref);
+    }
+}
+
+shiftI('shiftLeftByScalar', (x,s) => (x << (s & 7)) << 24 >> 24);
+shiftU('shiftLeftByScalar', (x,s) => (x << (s & 7)) << 24 >>> 24);
+shiftI('shiftRightByScalar', (x,s) => ((x << 24 >> 24) >> (s & 7)) << 24 >> 24);
+shiftU('shiftRightByScalar', (x,s) => ((x << 24 >>> 24) >>> (s & 7)) << 24 >>> 24);
+
+
+// Comparisons.
+function compareI(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + I8x16 + I8x16CHK +
+                                      `var fut = i8x16.${opname};` +
+                                      'function f(v1, v2) { v1 = i8x16chk(v1); v2 = i8x16chk(v2); return fut(v1, v2); } return f'), this);
+    let a1 = [  -1,2,  -3,-128,0x7f,6,-7, 8,-9, 10,-11, 12,-13,-14,-15, -16];
+    let a2 = [-128,2,-128,0x7f,   0,0, 8,-9,10,-11, 12,-13,-14,-15,-16,  -1];
+    let ref = zipmap(a1, a2, lanefunc);
+    let v1 = SIMD.Int8x16(...a1);
+    let v2 = SIMD.Int8x16(...a2);
+    assertEqVecArr(simdfunc(v1, v2), ref);
+}
+
+function compareU(opname, lanefunc) {
+    let simdfunc = asmLink(asmCompile('glob', USE_ASM + I8x16 + I8x16CHK + U8x16 + U8x16I8x16 +
+                                      `var fut = u8x16.${opname};` +
+                                      'function f(v1, v2) { v1 = i8x16chk(v1); v2 = i8x16chk(v2); return fut(u8x16i8x16(v1), u8x16i8x16(v2)); } return f'), this);
+    let a1 = [  -1,2,  -3,-128,0x7f,6,-7, 8,-9, 10,-11, 12,-13,-14,-15, -16].map(x => x << 24 >>> 24);
+    let a2 = [-128,2,-128,0x7f,   0,0, 8,-9,10,-11, 12,-13,-14,-15,-16,  -1].map(x => x << 24 >>> 24);
+    let ref = zipmap(a1, a2, lanefunc);
+    let v1 = SIMD.Int8x16(...a1);
+    let v2 = SIMD.Int8x16(...a2);
+    assertEqVecArr(simdfunc(v1, v2), ref);
+}
+
+compareI("equal", (x,y) => x == y);
+compareU("equal", (x,y) => x == y);
+compareI("notEqual", (x,y) => x != y);
+compareU("notEqual", (x,y) => x != y);
+compareI("lessThan", (x,y) => x < y);
+compareU("lessThan", (x,y) => x < y);
+compareI("lessThanOrEqual", (x,y) => x <= y);
+compareU("lessThanOrEqual", (x,y) => x <= y);
+compareI("greaterThan", (x,y) => x > y);
+compareU("greaterThan", (x,y) => x > y);
+compareI("greaterThanOrEqual", (x,y) => x >= y);
+compareU("greaterThanOrEqual", (x,y) => x >= y);

js/src/jit-test/tests/asm.js/testSIMD-bitcasts.js

@@ -0,0 +1,84 @@
+load(libdir + "asm.js");
+load(libdir + "simd.js");
+load(libdir + "asserts.js");
+
+// Set to true to see more JS debugging spew.
+const DEBUG = false;
+
+if (!isSimdAvailable()) {
+    DEBUG && print("won't run tests as simd extensions aren't activated yet");
+    quit(0);
+}
+
+// Test all bit-casts and normal loads and stores.
+var heap = new ArrayBuffer(BUF_MIN);
+var asU8 = new Uint8Array(heap);
+var allTypes = [
+    "Int8x16",
+    "Int16x8",
+    "Int32x4",
+    "Uint8x16",
+    "Uint16x8",
+    "Uint32x4",
+    "Float32x4"
+];
+
+// Generate a load bit-cast store test function that performs:
+//
+// function f(a, b) {
+//     vec = src.load(H, a);
+//     cast = dst.from«src»Bits(vec);
+//     store(H, b, cast);
+// }
+//
+// Here, `H` is the heap provided by `heap`.
+function test_func(src, dst) {
+    text = `
+        "use asm";
+        var src = glob.SIMD.${src};
+        var dst = glob.SIMD.${dst};
+        var ld = src.load;
+        var st = dst.store;
+        var bc = dst.from${src}Bits;
+
+        var H = new glob.Uint8Array(heap);
+
+        function f(a, b) {
+            a = a|0;
+            b = b|0;
+
+            st(H, b, bc(ld(H, a)));
+        }
+
+        return f;
+    `;
+    return asmLink(asmCompile('glob', 'ffi', 'heap', text), this, null, heap);
+}
+
+function assertBuf16(a, b) {
+    for (let i=0; i < 16; i++) {
+        assertEq(asU8[a+i], asU8[b+i]);
+    }
+}
+
+for (let src of allTypes) {
+    for (let dst of allTypes) {
+        // Skip identity conversions.
+        if (src == dst) continue;
+
+        print(src, dst);
+        let f = test_func(src, dst);
+        // Initialize with pseudo-random data.
+        for (let i = 0; i < 64; i++) {
+            asU8[i] = (i + 17) * 97;
+        }
+
+        // Aligned load/store.
+        f(0, 16);
+        assertBuf16(0, 16);
+
+        // Unaligned access.
+        f(1, 27);
+        assertBuf16(1, 27);
+    }
+}

js/src/jit-test/tests/asm.js/testSIMD-load-store.js

@@ -0,0 +1,457 @@
+// |jit-test|
+load(libdir + "asm.js");
+load(libdir + "simd.js");
+load(libdir + "asserts.js");
+
+// Avoid pathological --ion-eager compile times due to bails in loops
+setJitCompilerOption('ion.warmup.trigger', 1000000);
+
+// Set to true to see more JS debugging spew
+const DEBUG = false;
+
+if (!isSimdAvailable() || typeof SIMD === 'undefined' || !isAsmJSCompilationAvailable()) {
+    DEBUG && print("won't run tests as simd extensions aren't activated yet");
+    quit(0);
+}
+
+const RuntimeError = WebAssembly.RuntimeError;
+
+const INT32_MAX = Math.pow(2, 31) - 1;
+const INT32_MIN = INT32_MAX + 1 | 0;
+
+try {
+
+// Load / Store
+var IMPORTS = USE_ASM + 'var H=new glob.Uint8Array(heap); var i4=glob.SIMD.Int32x4; var ci4=i4.check; var load=i4.load; var store=i4.store;';
+
+//      Bad number of args
+assertAsmTypeFail('glob', 'ffi', 'heap', IMPORTS + "function f(){load();} return f");
+assertAsmTypeFail('glob', 'ffi', 'heap', IMPORTS + "function f(){load(3);} return f");
+assertAsmTypeFail('glob', 'ffi', 'heap', IMPORTS + "function f(){load(3, 4, 5);} return f");
+
+//      Bad type of args
+assertAsmTypeFail('glob', 'ffi', 'heap', IMPORTS + "function f(){load(3, 5);} return f");
+assertAsmTypeFail('glob', 'ffi', 'heap', IMPORTS + "function f(){load(H, 5.0);} return f");
+assertAsmTypeFail('glob', 'ffi', 'heap', IMPORTS + "function f(){var i=0.;load(H, i);} return f");
+assertAsmTypeFail('glob', 'ffi', 'heap', IMPORTS + "var H2=new glob.Int32Array(heap); function f(){var i=0;load(H2, i)} return f");
+assertAsmTypeFail('glob', 'ffi', 'heap', IMPORTS + "var H2=42; function f(){var i=0;load(H2, i)} return f");
+assertAsmTypeFail('glob', 'ffi', 'heap', IMPORTS + "function f(){var i=0;load(H2, i)} return f");
+assertAsmTypeFail('glob', 'ffi', 'heap', IMPORTS + "var f4=glob.SIMD.Float32x4; function f(){var i=0;var vec=f4(1,2,3,4); store(H, i, vec)} return f");
+
+//      Bad coercions of returned values
+assertAsmTypeFail('glob', 'ffi', 'heap', IMPORTS + "function f(){var i=0;return load(H, i)|0;} return f");
+assertAsmTypeFail('glob', 'ffi', 'heap', IMPORTS + "function f(){var i=0;return +load(H, i);} return f");
+
+//      Literal index constants
+var buf = new ArrayBuffer(BUF_MIN);
+var SIZE_TA = BUF_MIN >> 2
+var asI32 = new Int32Array(buf);
+asI32[SIZE_TA - 4] = 4;
+asI32[SIZE_TA - 3] = 3;
+asI32[SIZE_TA - 2] = 2;
+asI32[SIZE_TA - 1] = 1;
+
+assertAsmTypeFail('glob', 'ffi', 'heap', IMPORTS + "function f(){load(H, -1);} return f");
+assertAsmTypeFail('glob', 'ffi', 'heap', IMPORTS + "function f(){load(H, " + (INT32_MAX + 1) + ");} return f");
+assertAsmTypeFail('glob', 'ffi', 'heap', IMPORTS + "function f(){load(H, " + (INT32_MAX + 1 - 15) + ");} return f");
+asmCompile('glob', 'ffi', 'heap', IMPORTS + "function f(){load(H, " + (INT32_MAX + 1 - 16) + ");} return f");
+
+assertAsmLinkFail(asmCompile('glob', 'ffi', 'heap', IMPORTS + "function f() {return ci4(load(H, " + (BUF_MIN - 15) + "));} return f"), this, {}, buf);
+assertEqX4(asmLink(asmCompile('glob', 'ffi', 'heap', IMPORTS + "function f() {return ci4(load(H, " + (BUF_MIN - 16) + "));} return f"), this, {}, buf)(), [4, 3, 2, 1]);
+assertEqX4(asmLink(asmCompile('glob', 'ffi', 'heap', IMPORTS + "function f() {return ci4(load(H, " + BUF_MIN + " - 16 | 0));} return f"), this, {}, buf)(), [4, 3, 2, 1]);
+
+var CONSTANT_INDEX = 42;
+var CONSTANT_BYTE_INDEX = CONSTANT_INDEX << 2;
+
+var loadStoreCode = `
+    "use asm";
+
+    var H = new glob.Uint8Array(heap);
+
+    var i4 = glob.SIMD.Int32x4;
+    var i4load = i4.load;
+    var i4store = i4.store;
+    var ci4 = i4.check;
+
+    var f4 = glob.SIMD.Float32x4;
+    var f4load = f4.load;
+    var f4store = f4.store;
+    var cf4 = f4.check;
+
+    function f32l(i) { i=i|0; return cf4(f4load(H, i|0)); }
+    function f32lcst() { return cf4(f4load(H, ${CONSTANT_BYTE_INDEX})); }
+    function f32s(i, vec) { i=i|0; vec=cf4(vec); f4store(H, i|0, vec); }
+    function f32scst(vec) { vec=cf4(vec); f4store(H, ${CONSTANT_BYTE_INDEX}, vec); }
+
+    function i32l(i) { i=i|0; return ci4(i4load(H, i|0)); }
+    function i32lcst() { return ci4(i4load(H, ${CONSTANT_BYTE_INDEX})); }
+    function i32s(i, vec) { i=i|0; vec=ci4(vec); i4store(H, i|0, vec); }
+    function i32scst(vec) { vec=ci4(vec); i4store(H, ${CONSTANT_BYTE_INDEX}, vec); }
+
+    function f32lbndcheck(i) {
+        i=i|0;
+        if ((i|0) > ${CONSTANT_BYTE_INDEX}) i=${CONSTANT_BYTE_INDEX};
+        if ((i|0) < 0) i = 0;
+        return cf4(f4load(H, i|0));
+    }
+    function f32sbndcheck(i, vec) {
+        i=i|0;
+        vec=cf4(vec);
+        if ((i|0) > ${CONSTANT_BYTE_INDEX}) i=${CONSTANT_BYTE_INDEX};
+        if ((i|0) < 0) i = 0;
+        return cf4(f4store(H, i|0, vec));
+    }
+
+    return {
+        f32l: f32l,
+        f32lcst: f32lcst,
+        f32s: f32s,
+        f32scst: f32scst,
+        f32lbndcheck: f32lbndcheck,
+        f32sbndcheck: f32sbndcheck,
+        i32l: i32l,
+        i32lcst: i32lcst,
+        i32s: i32s,
+        i32scst: i32scst
+    }
+`;
+
+const SIZE = 0x8000;
+
+var F32 = new Float32Array(SIZE);
+var reset = function() {
+    for (var i = 0; i < SIZE; i++)
+        F32[i] = i + 1;
+};
+reset();
+
+var buf = F32.buffer;
+var m = asmLink(asmCompile('glob', 'ffi', 'heap', loadStoreCode), this, null, buf);
+
+function slice(TA, i, n) { return Array.prototype.slice.call(TA, i, i + n); }
+
+// Float32x4.load
+function f32l(n) { return m.f32l((n|0) << 2 | 0); };
+
+//      Correct accesses
+assertEqX4(f32l(0), slice(F32, 0, 4));
+assertEqX4(f32l(1), slice(F32, 1, 4));
+assertEqX4(f32l(SIZE - 4), slice(F32, SIZE - 4, 4));
+
+assertEqX4(m.f32lcst(), slice(F32, CONSTANT_INDEX, 4));
+assertEqX4(m.f32lbndcheck(CONSTANT_BYTE_INDEX), slice(F32, CONSTANT_INDEX, 4));
+
+//      OOB
+assertThrowsInstanceOf(() => f32l(-1), RuntimeError);
+assertThrowsInstanceOf(() => f32l(SIZE), RuntimeError);
+assertThrowsInstanceOf(() => f32l(SIZE - 1), RuntimeError);
+assertThrowsInstanceOf(() => f32l(SIZE - 2), RuntimeError);
+assertThrowsInstanceOf(() => f32l(SIZE - 3), RuntimeError);
+
+var code = `
+    "use asm";
+    var f4 = glob.SIMD.Float32x4;
+    var f4l = f4.load;
+    var u8 = new glob.Uint8Array(heap);
+
+    function g(x) {
+        x = x|0;
+        // set a constraint on the size of the heap
+        var ptr = 0;
+        ptr = u8[0xFFFF] | 0;
+        // give a precise range to x
+        x = (x>>0) > 5 ? 5 : x;
+        x = (x>>0) < 0 ? 0 : x;
+        // ptr value gets a precise range but the bounds check shouldn't get
+        // eliminated.
+        return f4l(u8, 0xFFFA + x | 0);
+    }
+
+    return g;
+`;
+assertThrowsInstanceOf(() => asmLink(asmCompile('glob', 'ffi', 'heap', code), this, {}, new ArrayBuffer(0x10000))(0), RuntimeError);
+
+// Float32x4.store
+function f32s(n, v) { return m.f32s((n|0) << 2 | 0, v); };
+
+var vec  = SIMD.Float32x4(5,6,7,8);
+var vec2 = SIMD.Float32x4(0,1,2,3);
+var vecWithNaN = SIMD.Float32x4(NaN, 2, NaN, 4);
+
+reset();
+f32s(0, vec);
+assertEqX4(vec, slice(F32, 0, 4));
+
+reset();
+f32s(0, vec2);
+assertEqX4(vec2, slice(F32, 0, 4));
+
+reset();
+f32s(4, vec);
+assertEqX4(vec, slice(F32, 4, 4));
+
+reset();
+f32s(4, vecWithNaN);
+assertEqX4(vecWithNaN, slice(F32, 4, 4));
+
+reset();
+m.f32scst(vec2);
+assertEqX4(vec2, slice(F32, CONSTANT_INDEX, 4));
+
+reset();
+m.f32sbndcheck(CONSTANT_BYTE_INDEX, vec);
+assertEqX4(vec, slice(F32, CONSTANT_INDEX, 4));
+
+//      OOB
+reset();
+assertThrowsInstanceOf(() => f32s(SIZE - 3, vec), RuntimeError);
+assertThrowsInstanceOf(() => f32s(SIZE - 2, vec), RuntimeError);
+assertThrowsInstanceOf(() => f32s(SIZE - 1, vec), RuntimeError);
+assertThrowsInstanceOf(() => f32s(SIZE, vec), RuntimeError);
+for (var i = 0; i < SIZE; i++)
+    assertEq(F32[i], i + 1);
+
+// Int32x4.load
+var I32 = new Int32Array(buf);
+reset = function () {
+    for (var i = 0; i < SIZE; i++)
+        I32[i] = i + 1;
+};
+reset();
+
+function i32(n) { return m.i32l((n|0) << 2 | 0); };
+
+//      Correct accesses
+assertEqX4(i32(0), slice(I32, 0, 4));
+assertEqX4(i32(1), slice(I32, 1, 4));
+assertEqX4(i32(SIZE - 4), slice(I32, SIZE - 4, 4));
+
+assertEqX4(m.i32lcst(), slice(I32, CONSTANT_INDEX, 4));
+
+//      OOB
+assertThrowsInstanceOf(() => i32(-1), RuntimeError);
+assertThrowsInstanceOf(() => i32(SIZE), RuntimeError);
+assertThrowsInstanceOf(() => i32(SIZE - 1), RuntimeError);
+assertThrowsInstanceOf(() => i32(SIZE - 2), RuntimeError);
+assertThrowsInstanceOf(() => i32(SIZE - 3), RuntimeError);
+
+// Int32x4.store
+function i32s(n, v) { return m.i32s((n|0) << 2 | 0, v); };
+
+var vec  = SIMD.Int32x4(5,6,7,8);
+var vec2 = SIMD.Int32x4(0,1,2,3);
+
+reset();
+i32s(0, vec);
+assertEqX4(vec, slice(I32, 0, 4));
+
+reset();
+i32s(0, vec2);
+assertEqX4(vec2, slice(I32, 0, 4));
+
+reset();
+i32s(4, vec);
+assertEqX4(vec, slice(I32, 4, 4));
+
+reset();
+m.i32scst(vec2);
+assertEqX4(vec2, slice(I32, CONSTANT_INDEX, 4));
+
+//      OOB
+reset();
+assertThrowsInstanceOf(() => i32s(SIZE - 3, vec), RuntimeError);
+assertThrowsInstanceOf(() => i32s(SIZE - 2, vec), RuntimeError);
+assertThrowsInstanceOf(() => i32s(SIZE - 1, vec), RuntimeError);
+assertThrowsInstanceOf(() => i32s(SIZE - 0, vec), RuntimeError);
+for (var i = 0; i < SIZE; i++)
+    assertEq(I32[i], i + 1);
+
+// Partial loads and stores
+(function() {
+
+//      Variable indexes
+function MakeCodeFor(typeName) {
+    return `
+    "use asm";
+    var type = glob.SIMD.${typeName};
+    var c = type.check;
+
+    var l1 = type.load1;
+    var l2 = type.load2;
+
+    var s1 = type.store1;
+    var s2 = type.store2;
+
+    var u8 = new glob.Uint8Array(heap);
+
+    function load1(i) { i=i|0; return l1(u8, i); }
+    function load2(i) { i=i|0; return l2(u8, i); }
+
+    function loadCst1() { return l1(u8, 41 << 2); }
+    function loadCst2() { return l2(u8, 41 << 2); }
+
+    function store1(i, x) { i=i|0; x=c(x); return s1(u8, i, x); }
+    function store2(i, x) { i=i|0; x=c(x); return s2(u8, i, x); }
+
+    function storeCst1(x) { x=c(x); return s1(u8, 41 << 2, x); }
+    function storeCst2(x) { x=c(x); return s2(u8, 41 << 2, x); }
+
+    return {
+        load1: load1,
+        load2: load2,
+        loadCst1: loadCst1,
+        loadCst2: loadCst2,
+        store1: store1,
+        store2: store2,
+        storeCst1: storeCst1,
+        storeCst2: storeCst2,
+    }
+`;
+}
+
+var SIZE = 0x10000;
+
+function TestPartialLoads(m, typedArray, x, y, z, w) {
+    // Fill array with predictable values
+    for (var i = 0; i < SIZE; i += 4) {
+        typedArray[i] =     x(i);
+        typedArray[i + 1] = y(i);
+        typedArray[i + 2] = z(i);
+        typedArray[i + 3] = w(i);
+    }
+
+    // Test correct loads
+    var i = 0, j = 0; // i in elems, j in bytes
+    assertEqX4(m.load1(j), [x(i), 0, 0, 0]);
+    assertEqX4(m.load2(j), [x(i), y(i), 0, 0]);
+
+    j += 4;
+    assertEqX4(m.load1(j), [y(i), 0, 0, 0]);
+    assertEqX4(m.load2(j), [y(i), z(i), 0, 0]);
+
+    j += 4;
+    assertEqX4(m.load1(j), [z(i), 0, 0, 0]);
+    assertEqX4(m.load2(j), [z(i), w(i), 0, 0]);
+
+    j += 4;
+    assertEqX4(m.load1(j), [w(i), 0, 0, 0]);
+    assertEqX4(m.load2(j), [w(i), x(i+4), 0, 0]);
+
+    j += 4;
+    i += 4;
+    assertEqX4(m.load1(j), [x(i), 0, 0, 0]);
+    assertEqX4(m.load2(j), [x(i), y(i), 0, 0]);
+
+    // Test loads with constant indexes (41)
+    assertEqX4(m.loadCst1(), [y(40), 0, 0, 0]);
+    assertEqX4(m.loadCst2(), [y(40), z(40), 0, 0]);
+
+    // Test limit and OOB accesses
+    assertEqX4(m.load1((SIZE - 1) << 2), [w(SIZE - 4), 0, 0, 0]);
+    assertThrowsInstanceOf(() => m.load1(((SIZE - 1) << 2) + 1), RuntimeError);
+
+    assertEqX4(m.load2((SIZE - 2) << 2), [z(SIZE - 4), w(SIZE - 4), 0, 0]);
+    assertThrowsInstanceOf(() => m.load2(((SIZE - 2) << 2) + 1), RuntimeError);
+}
+
+// Partial stores
+function TestPartialStores(m, typedArray, typeName, x, y, z, w) {
+    var val = SIMD[typeName](x, y, z, w);
+
+    function Reset() {
+        for (var i = 0; i < SIZE; i++)
+            typedArray[i] = i + 1;
+    }
+    function CheckNotModified(low, high) {
+        for (var i = low; i < high; i++)
+            assertEq(typedArray[i], i + 1);
+    }
+
+    function TestStore1(i) {
+        m.store1(i, val);
+        CheckNotModified(0, i >> 2);
+        assertEq(typedArray[i >> 2], x);
+        CheckNotModified((i >> 2) + 1, SIZE);
+        typedArray[i >> 2] = (i >> 2) + 1;
+    }
+
+    function TestStore2(i) {
+        m.store2(i, val);
+        CheckNotModified(0, i >> 2);
+        assertEq(typedArray[i >> 2], x);
+        assertEq(typedArray[(i >> 2) + 1], y);
+        CheckNotModified((i >> 2) + 2, SIZE);
+        typedArray[i >> 2] = (i >> 2) + 1;
+        typedArray[(i >> 2) + 1] = (i >> 2) + 2;
+    }
+
+    function TestOOBStore(f) {
+        assertThrowsInstanceOf(f, RuntimeError);
+        CheckNotModified(0, SIZE);
+    }
+
+    Reset();
+
+    TestStore1(0);
+    TestStore1(1 << 2);
+    TestStore1(2 << 2);
+    TestStore1(3 << 2);
+    TestStore1(1337 << 2);
+
+    var i = (SIZE - 1) << 2;
+    TestStore1(i);
+    TestOOBStore(() => m.store1(i + 1, val));
+    TestOOBStore(() => m.store1(-1, val));
+
+    TestStore2(0);
+    TestStore2(1 << 2);
+    TestStore2(2 << 2);
+    TestStore2(3 << 2);
+    TestStore2(1337 << 2);
+
+    var i = (SIZE - 2) << 2;
+    TestStore2(i);
+    TestOOBStore(() => m.store2(i + 1, val));
+    TestOOBStore(() => m.store2(-1, val));
+
+    // Constant indexes (41)
+    m.storeCst1(val);
+    CheckNotModified(0, 41);
+    assertEq(typedArray[41], x);
+    CheckNotModified(42, SIZE);
+    typedArray[41] = 42;
+
+    m.storeCst2(val);
+    CheckNotModified(0, 41);
+    assertEq(typedArray[41], x);
+    assertEq(typedArray[42], y);
+    CheckNotModified(43, SIZE);
+    typedArray[41] = 42;
+    typedArray[42] = 43;
+}
+
+var f32 = new Float32Array(SIZE);
+var mFloat32x4 = asmLink(asmCompile('glob', 'ffi', 'heap', MakeCodeFor('Float32x4')), this, null, f32.buffer);
+
+TestPartialLoads(mFloat32x4, f32,
+            (i) => i + 1,
+            (i) => Math.fround(13.37),
+            (i) => Math.fround(1/i),
+            (i) => Math.fround(Math.sqrt(0x2000 - i)));
+
+TestPartialStores(mFloat32x4, f32, 'Float32x4', 42, -0, NaN, 0.1337);
+
+var i32 = new Int32Array(f32.buffer);
+var mInt32x4 = asmLink(asmCompile('glob', 'ffi', 'heap', MakeCodeFor('Int32x4')), this, null, i32.buffer);
+
+TestPartialLoads(mInt32x4, i32,
+            (i) => i + 1 | 0,
+            (i) => -i | 0,
+            (i) => i * 2 | 0,
+            (i) => 42);
+
+TestPartialStores(mInt32x4, i32, 'Int32x4', 42, -3, 13, 37);
+
+})();
+
+} catch (e) { print('stack: ', e.stack); throw e }

js/src/jit-test/tests/asm.js/testZOOB.js

@@ -100,6 +100,106 @@ function assertEqX4(observed, expected) {
     assertEq(observed.w, expected.w);
 }
 
+function testSimdX4(ctor, shift, scale, disp, simdName, simdCtor) {
+    var arr = new ctor(ab);
+
+    var c = asmCompile('glob', 'imp', 'b',
+                       USE_ASM +
+                       'var arr=new glob.' + ctor.name + '(b); ' +
+                       'var SIMD_' + simdName + ' = glob.SIMD.' + simdName + '; ' +
+                       'var SIMD_' + simdName + '_check = SIMD_' + simdName + '.check; ' +
+                       'var SIMD_' + simdName + '_load = SIMD_' + simdName + '.load; ' +
+                       'var SIMD_' + simdName + '_load2 = SIMD_' + simdName + '.load2; ' +
+                       'var SIMD_' + simdName + '_load1 = SIMD_' + simdName + '.load1; ' +
+                       'var SIMD_' + simdName + '_store = SIMD_' + simdName + '.store; ' +
+                       'var SIMD_' + simdName + '_store2 = SIMD_' + simdName + '.store2; ' +
+                       'var SIMD_' + simdName + '_store1 = SIMD_' + simdName + '.store1; ' +
+                       'function load(i) {i=i|0; return SIMD_' + simdName + '_check(SIMD_' + simdName + '_load(arr, ((i<<' + scale + ')+' + disp + ')>>' + shift + ')) } ' +
+                       'function load2(i) {i=i|0; return SIMD_' + simdName + '_check(SIMD_' + simdName + '_load2(arr, ((i<<' + scale + ')+' + disp + ')>>' + shift + ')) } ' +
+                       'function load1(i) {i=i|0; return SIMD_' + simdName + '_check(SIMD_' + simdName + '_load1(arr, ((i<<' + scale + ')+' + disp + ')>>' + shift + ')) } ' +
+                       'function store(i,j) {i=i|0;j=SIMD_' + simdName + '_check(j); SIMD_' + simdName + '_store(arr, ((i<<' + scale + ')+' + disp + ')>>' + shift + ', j) } ' +
+                       'function store2(i,j) {i=i|0;j=SIMD_' + simdName + '_check(j); SIMD_' + simdName + '_store2(arr, ((i<<' + scale + ')+' + disp + ')>>' + shift + ', j) } ' +
+                       'function store1(i,j) {i=i|0;j=SIMD_' + simdName + '_check(j); SIMD_' + simdName + '_store1(arr, ((i<<' + scale + ')+' + disp + ')>>' + shift + ', j) } ' +
+                       'return { load: load, load2: load2, load1: load1, store: store, store2 : store2, store1 : store1 }');
+    var f = asmLink(c, this, null, ab);
+
+    const RuntimeError = WebAssembly.RuntimeError;
+
+    for (var i of indices) {
+        var index = ((i<<scale)+disp)>>shift;
+
+        var v, v2, v1;
+        var t = false, t2 = false, t1 = false;
+        try { v = simdCtor.load(arr, index); }
+        catch (e) {
+            assertEq(e instanceof RangeError, true);
+            t = true;
+        }
+        try { v2 = simdCtor.load2(arr, index); }
+        catch (e) {
+            assertEq(e instanceof RangeError, true);
+            t2 = true;
+        }
+        try { v1 = simdCtor.load1(arr, index); }
+        catch (e) {
+            assertEq(e instanceof RangeError, true);
+            t1 = true;
+        }
+
+        // Loads
+        var l, l2, l1;
+        var r = false, r2 = false, r1 = false;
+        try { l = f.load(i); }
+        catch (e) {
+            assertEq(e instanceof RuntimeError, true);
+            r = true;
+        }
+        try { l2 = f.load2(i); }
+        catch (e) {
+            assertEq(e instanceof RuntimeError, true);
+            r2 = true;
+        }
+        try { l1 = f.load1(i); }
+        catch (e) {
+            assertEq(e instanceof RuntimeError, true);
+            r1 = true;
+        }
+        assertEq(t, r);
+        assertEq(t2, r2);
+        assertEq(t1, r1);
+        if (!t) assertEqX4(v, l);
+        if (!t2) assertEqX4(v2, l2);
+        if (!t1) assertEqX4(v1, l1);
+
+        // Stores
+        if (!t) {
+            simdCtor.store(arr, index, simdCtor.neg(v));
+            f.store(i, v);
+            assertEqX4(simdCtor.load(arr, index), v);
+        } else
+            assertThrowsInstanceOf(() => f.store(i, simdCtor()), RuntimeError);
+        if (!t2) {
+            simdCtor.store2(arr, index, simdCtor.neg(v2));
+            f.store2(i, v2);
+            assertEqX4(simdCtor.load2(arr, index), v2);
+        } else
+            assertThrowsInstanceOf(() => f.store2(i, simdCtor()), RuntimeError);
+        if (!t1) {
+            simdCtor.store1(arr, index, simdCtor.neg(v1));
+            f.store1(i, v1);
+            assertEqX4(simdCtor.load1(arr, index), v1);
+        } else
+            assertThrowsInstanceOf(() => f.store1(i, simdCtor()), RuntimeError);
+    }
+}
+
+function testFloat32x4(ctor, shift, scale, disp) {
+    testSimdX4(ctor, shift, scale, disp, 'Float32x4', SIMD.Float32x4);
+}
+function testInt32x4(ctor, shift, scale, disp) {
+    testSimdX4(ctor, shift, scale, disp, 'Int32x4', SIMD.Int32x4);
+}
+
 function test(tester, ctor, shift) {
     var arr = new ctor(ab);
     for (var i = 0; i < arr.length; i++)
@@ -123,3 +223,15 @@ test(testInt, Int32Array, 2);
 test(testInt, Uint32Array, 2);
 test(testFloat32, Float32Array, 2);
 test(testFloat64, Float64Array, 3);
+if (typeof SIMD !== 'undefined' && isSimdAvailable()) {
+    // Avoid pathological --ion-eager compile times due to bails in loops
+    setJitCompilerOption('ion.warmup.trigger', 1000000);
+
+    // Use a fresh ArrayBuffer so prepareForAsmJS can allocated a guard page
+    // which SIMD.js needs. Since the original ArrayBuffer was prepared for
+    // asm.js that didn't use SIMD.js, it has no guard page (on 32-bit).
+    ab = new ArrayBuffer(BUF_MIN);
+
+    test(testInt32x4, Uint8Array, 0);
+    test(testFloat32x4, Uint8Array, 0);
+}

js/src/jit/BaselineBailouts.cpp

@@ -2012,6 +2012,7 @@ jit::FinishBailoutToBaseline(BaselineBailoutInfo* bailoutInfo)
       case Bailout_NonObjectInput:
       case Bailout_NonStringInput:
       case Bailout_NonSymbolInput:
+      case Bailout_UnexpectedSimdInput:
       case Bailout_NonSharedTypedArrayInput:
       case Bailout_Debugger:
       case Bailout_UninitializedThis:

js/src/jit/BaselineIC.cpp

@@ -14,6 +14,7 @@
 #include "jstypes.h"
 
 #include "builtin/Eval.h"
+#include "builtin/SIMDConstants.h"
 #include "gc/Policy.h"
 #include "jit/BaselineCacheIRCompiler.h"
 #include "jit/BaselineDebugModeOSR.h"
@@ -2012,6 +2013,71 @@ TryAttachFunCallStub(JSContext* cx, ICCall_Fallback* stub, HandleScript script,
     return true;
 }
 
+// Check if target is a native SIMD operation which returns a SIMD type.
+// If so, set res to a template object matching the SIMD type produced and return true.
+static bool
+GetTemplateObjectForSimd(JSContext* cx, JSFunction* target, MutableHandleObject res)
+{
+    if (!target->hasJitInfo())
+        return false;
+
+    const JSJitInfo* jitInfo = target->jitInfo();
+    if (jitInfo->type() != JSJitInfo::InlinableNative)
+        return false;
+
+    // Check if this is a native inlinable SIMD operation.
+    SimdType ctrlType;
+    switch (jitInfo->inlinableNative) {
+      case InlinableNative::SimdInt8x16:   ctrlType = SimdType::Int8x16;   break;
+      case InlinableNative::SimdUint8x16:  ctrlType = SimdType::Uint8x16;  break;
+      case InlinableNative::SimdInt16x8:   ctrlType = SimdType::Int16x8;   break;
+      case InlinableNative::SimdUint16x8:  ctrlType = SimdType::Uint16x8;  break;
+      case InlinableNative::SimdInt32x4:   ctrlType = SimdType::Int32x4;   break;
+      case InlinableNative::SimdUint32x4:  ctrlType = SimdType::Uint32x4;  break;
+      case InlinableNative::SimdFloat32x4: ctrlType = SimdType::Float32x4; break;
+      case InlinableNative::SimdBool8x16:  ctrlType = SimdType::Bool8x16;  break;
+      case InlinableNative::SimdBool16x8:  ctrlType = SimdType::Bool16x8;  break;
+      case InlinableNative::SimdBool32x4:  ctrlType = SimdType::Bool32x4;  break;
+      // This is not an inlinable SIMD operation.
+      default: return false;
+    }
+
+    // The controlling type is not necessarily the return type.
+    // Check the actual operation.
+    SimdOperation simdOp = SimdOperation(jitInfo->nativeOp);
+    SimdType retType;
+
+    switch(simdOp) {
+      case SimdOperation::Fn_allTrue:
+      case SimdOperation::Fn_anyTrue:
+      case SimdOperation::Fn_extractLane:
+        // These operations return a scalar. No template object needed.
+        return false;
+
+      case SimdOperation::Fn_lessThan:
+      case SimdOperation::Fn_lessThanOrEqual:
+      case SimdOperation::Fn_equal:
+      case SimdOperation::Fn_notEqual:
+      case SimdOperation::Fn_greaterThan:
+      case SimdOperation::Fn_greaterThanOrEqual:
+        // These operations return a boolean vector with the same shape as the
+        // controlling type.
+        retType = GetBooleanSimdType(ctrlType);
+        break;
+
+      default:
+        // All other operations return the controlling type.
+        retType = ctrlType;
+        break;
+    }
+
+    // Create a template object based on retType.
+    RootedGlobalObject global(cx, cx->global());
+    Rooted<SimdTypeDescr*> descr(cx, GlobalObject::getOrCreateSimdTypeDescr(cx, global, retType));
+    res.set(cx->realm()->jitRealm()->getSimdTemplateObjectFor(cx, descr));
+    return true;
+}
+
 static bool
 GetTemplateObjectForNative(JSContext* cx, HandleFunction target, const CallArgs& args,
                            MutableHandleObject res, bool* skipAttach)
@@ -2095,6 +2161,9 @@ GetTemplateObjectForNative(JSContext* cx, HandleFunction target, const CallArgs&
         return !!res;
     }
 
+    if (JitSupportsSimd() && GetTemplateObjectForSimd(cx, target, res))
+       return !!res;
+
     return true;
 }
 
@@ -2111,6 +2180,12 @@ GetTemplateObjectForClassHook(JSContext* cx, JSNative hook, CallArgs& args,
         return !!templateObject;
     }
 
+    if (hook == SimdTypeDescr::call && JitSupportsSimd()) {
+        Rooted<SimdTypeDescr*> descr(cx, &args.callee().as<SimdTypeDescr>());
+        templateObject.set(cx->realm()->jitRealm()->getSimdTemplateObjectFor(cx, descr));
+        return !!templateObject;
+    }
+
     return true;
 }
 

js/src/jit/BaselineInspector.cpp

@@ -663,6 +663,25 @@ BaselineInspector::getTemplateObjectForClassHook(jsbytecode* pc, const Class* cl
     return nullptr;
 }
 
+JSObject*
+BaselineInspector::getTemplateObjectForSimdCtor(jsbytecode* pc, SimdType simdType)
+{
+    if (!hasBaselineScript())
+        return nullptr;
+
+    const ICEntry& entry = icEntryFromPC(pc);
+    for (ICStub* stub = entry.firstStub(); stub; stub = stub->next()) {
+        if (stub->isCall_ClassHook() && stub->toCall_ClassHook()->clasp() == &SimdTypeDescr::class_) {
+            JSObject* templateObj = stub->toCall_ClassHook()->templateObject();
+            InlineTypedObject& typedObj = templateObj->as<InlineTypedObject>();
+            if (typedObj.typeDescr().as<SimdTypeDescr>().type() == simdType)
+                return templateObj;
+        }
+    }
+
+    return nullptr;
+}
+
 LexicalEnvironmentObject*
 BaselineInspector::templateNamedLambdaObject()
 {

js/src/jit/BaselineInspector.h

@@ -131,6 +131,7 @@ class BaselineInspector
     JSObject* getTemplateObject(jsbytecode* pc);
     JSObject* getTemplateObjectForNative(jsbytecode* pc, Native native);
     JSObject* getTemplateObjectForClassHook(jsbytecode* pc, const Class* clasp);
+    JSObject* getTemplateObjectForSimdCtor(jsbytecode* pc, SimdType simdType);
 
     // Sometimes the group a template object will have is known, even if the
     // object itself isn't.

js/src/jit/CodeGenerator.cpp

@@ -422,6 +422,7 @@ CodeGenerator::CodeGenerator(MIRGenerator* gen, LIRGraph* graph, MacroAssembler*
   : CodeGeneratorSpecific(gen, graph, masm)
   , ionScriptLabels_(gen->alloc())
   , scriptCounts_(nullptr)
+  , simdTemplatesToReadBarrier_(0)
   , realmStubsToReadBarrier_(0)
 {
 }
@@ -6426,6 +6427,82 @@ CodeGenerator::visitNewTypedObject(LNewTypedObject* lir)
     masm.bind(ool->rejoin());
 }
 
+void
+CodeGenerator::visitSimdBox(LSimdBox* lir)
+{
+    FloatRegister in = ToFloatRegister(lir->input());
+    Register object = ToRegister(lir->output());
+    Register temp = ToRegister(lir->temp());
+    InlineTypedObject* templateObject = lir->mir()->templateObject();
+    gc::InitialHeap initialHeap = lir->mir()->initialHeap();
+    MIRType type = lir->mir()->input()->type();
+
+    addSimdTemplateToReadBarrier(lir->mir()->simdType());
+
+    MOZ_ASSERT(lir->safepoint()->liveRegs().has(in), "Save the input register across oolCallVM");
+    OutOfLineCode* ool = oolCallVM(NewTypedObjectInfo, lir,
+                                   ArgList(ImmGCPtr(templateObject), Imm32(initialHeap)),
+                                   StoreRegisterTo(object));
+
+    TemplateObject templateObj(templateObject);
+    masm.createGCObject(object, temp, templateObj, initialHeap, ool->entry());
+    masm.bind(ool->rejoin());
+
+    Address objectData(object, InlineTypedObject::offsetOfDataStart());
+    switch (type) {
+      case MIRType::Int8x16:
+      case MIRType::Int16x8:
+      case MIRType::Int32x4:
+      case MIRType::Bool8x16:
+      case MIRType::Bool16x8:
+      case MIRType::Bool32x4:
+        masm.storeUnalignedSimd128Int(in, objectData);
+        break;
+      case MIRType::Float32x4:
+        masm.storeUnalignedSimd128Float(in, objectData);
+        break;
+      default:
+        MOZ_CRASH("Unknown SIMD kind when generating code for SimdBox.");
+    }
+}
+
+void
+CodeGenerator::addSimdTemplateToReadBarrier(SimdType simdType)
+{
+    simdTemplatesToReadBarrier_ |= 1 << uint32_t(simdType);
+}
+
+void
+CodeGenerator::visitSimdUnbox(LSimdUnbox* lir)
+{
+    Register object = ToRegister(lir->input());
+    FloatRegister simd = ToFloatRegister(lir->output());
+    Register temp = ToRegister(lir->temp());
+    Label bail;
+
+    masm.branchIfNotSimdObject(object, temp, lir->mir()->simdType(), &bail);
+
+    // Load the value from the data of the InlineTypedObject.
+    Address objectData(object, InlineTypedObject::offsetOfDataStart());
+    switch (lir->mir()->type()) {
+      case MIRType::Int8x16:
+      case MIRType::Int16x8:
+      case MIRType::Int32x4:
+      case MIRType::Bool8x16:
+      case MIRType::Bool16x8:
+      case MIRType::Bool32x4:
+        masm.loadUnalignedSimd128Int(objectData, simd);
+        break;
+      case MIRType::Float32x4:
+        masm.loadUnalignedSimd128Float(objectData, simd);
+        break;
+      default:
+        MOZ_CRASH("The impossible happened!");
+    }
+
+    bailoutFrom(&bail, lir->snapshot());
+}
+
 typedef js::NamedLambdaObject* (*NewNamedLambdaObjectFn)(JSContext*, HandleFunction, gc::InitialHeap);
 static const VMFunction NewNamedLambdaObjectInfo =
     FunctionInfo<NewNamedLambdaObjectFn>(NamedLambdaObject::createTemplateObject,
@@ -7156,7 +7233,7 @@ CodeGenerator::visitWasmLoadGlobalVar(LWasmLoadGlobalVar* ins)
     MWasmLoadGlobalVar* mir = ins->mir();
 
     MIRType type = mir->type();
-    MOZ_ASSERT(IsNumberType(type));
+    MOZ_ASSERT(IsNumberType(type) || IsSimdType(type));
 
     Register tls = ToRegister(ins->tlsPtr());
     Address addr(tls, offsetof(wasm::TlsData, globalArea) + mir->globalDataOffset());
@@ -7183,7 +7260,11 @@ CodeGenerator::visitWasmLoadGlobalVar(LWasmLoadGlobalVar* ins)
       case MIRType::Bool8x16:
       case MIRType::Bool16x8:
       case MIRType::Bool32x4:
+        masm.loadInt32x4(addr, ToFloatRegister(ins->output()));
+        break;
       case MIRType::Float32x4:
+        masm.loadFloat32x4(addr, ToFloatRegister(ins->output()));
+        break;
       default:
         MOZ_CRASH("unexpected type in visitWasmLoadGlobalVar");
     }
@@ -7195,7 +7276,7 @@ CodeGenerator::visitWasmStoreGlobalVar(LWasmStoreGlobalVar* ins)
     MWasmStoreGlobalVar* mir = ins->mir();
 
     MIRType type = mir->value()->type();
-    MOZ_ASSERT(IsNumberType(type));
+    MOZ_ASSERT(IsNumberType(type) || IsSimdType(type));
 
     Register tls = ToRegister(ins->tlsPtr());
     Address addr(tls, offsetof(wasm::TlsData, globalArea) + mir->globalDataOffset());
@@ -7222,7 +7303,11 @@ CodeGenerator::visitWasmStoreGlobalVar(LWasmStoreGlobalVar* ins)
       case MIRType::Bool8x16:
       case MIRType::Bool16x8:
       case MIRType::Bool32x4:
+        masm.storeInt32x4(ToFloatRegister(ins->value()), addr);
+        break;
       case MIRType::Float32x4:
+        masm.storeFloat32x4(ToFloatRegister(ins->value()), addr);
+        break;
       default:
         MOZ_CRASH("unexpected type in visitWasmStoreGlobalVar");
     }
@@ -10277,6 +10362,7 @@ CodeGenerator::link(JSContext* cx, CompilerConstraintList* constraints)
     // script, which may have happened off-thread.
     const JitRealm* jr = gen->realm->jitRealm();
     jr->performStubReadBarriers(realmStubsToReadBarrier_);
+    jr->performSIMDTemplateReadBarriers(simdTemplatesToReadBarrier_);
 
     // We finished the new IonScript. Invalidate the current active IonScript,
     // so we can replace it with this new (probably higher optimized) version.
@@ -11500,17 +11586,19 @@ CodeGenerator::visitLoadUnboxedScalar(LLoadUnboxedScalar* lir)
     const MLoadUnboxedScalar* mir = lir->mir();
 
     Scalar::Type readType = mir->readType();
+    unsigned numElems = mir->numElems();
+
     int width = Scalar::byteSize(mir->storageType());
     bool canonicalizeDouble = mir->canonicalizeDoubles();
 
     Label fail;
     if (lir->index()->isConstant()) {
         Address source(elements, ToInt32(lir->index()) * width + mir->offsetAdjustment());
-        masm.loadFromTypedArray(readType, source, out, temp, &fail, canonicalizeDouble);
+        masm.loadFromTypedArray(readType, source, out, temp, &fail, canonicalizeDouble, numElems);
     } else {
         BaseIndex source(elements, ToRegister(lir->index()), ScaleFromElemWidth(width),
                          mir->offsetAdjustment());
-        masm.loadFromTypedArray(readType, source, out, temp, &fail, canonicalizeDouble);
+        masm.loadFromTypedArray(readType, source, out, temp, &fail, canonicalizeDouble, numElems);
     }
 
     if (fail.used())
@@ -11789,10 +11877,13 @@ CodeGenerator::visitLoadElementFromStateV(LLoadElementFromStateV* lir)
 template <typename T>
 static inline void
 StoreToTypedArray(MacroAssembler& masm, Scalar::Type writeType, const LAllocation* value,
-                  const T& dest)
+                  const T& dest, unsigned numElems = 0)
 {
-    if (writeType == Scalar::Float32 || writeType == Scalar::Float64) {
+    if (Scalar::isSimdType(writeType) ||
-        masm.storeToTypedFloatArray(writeType, ToFloatRegister(value), dest);
+        writeType == Scalar::Float32 ||
+        writeType == Scalar::Float64)
+    {
+        masm.storeToTypedFloatArray(writeType, ToFloatRegister(value), dest, numElems);
     } else {
         if (value->isConstant())
             masm.storeToTypedIntArray(writeType, Imm32(ToInt32(value)), dest);
@@ -11810,16 +11901,17 @@ CodeGenerator::visitStoreUnboxedScalar(LStoreUnboxedScalar* lir)
     const MStoreUnboxedScalar* mir = lir->mir();
 
     Scalar::Type writeType = mir->writeType();
+    unsigned numElems = mir->numElems();
 
     int width = Scalar::byteSize(mir->storageType());
 
     if (lir->index()->isConstant()) {
         Address dest(elements, ToInt32(lir->index()) * width + mir->offsetAdjustment());
-        StoreToTypedArray(masm, writeType, value, dest);
+        StoreToTypedArray(masm, writeType, value, dest, numElems);
     } else {
         BaseIndex dest(elements, ToRegister(lir->index()), ScaleFromElemWidth(width),
                        mir->offsetAdjustment());
-        StoreToTypedArray(masm, writeType, value, dest);
+        StoreToTypedArray(masm, writeType, value, dest, numElems);
     }
 }
 

js/src/jit/CodeGenerator.h

@@ -317,9 +317,24 @@ class CodeGenerator final : public CodeGeneratorSpecific
     PerfSpewer perfSpewer_;
 #endif
 
+    // This integer is a bit mask of all SimdTypeDescr::Type indexes.  When a
+    // MSimdBox instruction is encoded, it might have either been created by
+    // IonBuilder, or by the Eager Simd Unbox phase.
+    //
+    // As the template objects are weak references, the JitRealm is using
+    // Read Barriers, but such barrier cannot be used during the compilation. To
+    // work around this issue, the barriers are captured during
+    // CodeGenerator::link.
+    //
+    // Instead of saving the pointers, we just save the index of the Read
+    // Barriered objects in a bit mask.
+    uint32_t simdTemplatesToReadBarrier_;
+
     // Bit mask of JitRealm stubs that are to be read-barriered.
     uint32_t realmStubsToReadBarrier_;
 
+    void addSimdTemplateToReadBarrier(SimdType simdType);
+
 #define LIR_OP(op) void visit##op(L##op* ins);
     LIR_OPCODE_LIST(LIR_OP)
 #undef LIR_OP

js/src/jit/EagerSimdUnbox.cpp

@@ -0,0 +1,128 @@
+/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*-
+ * vim: set ts=8 sts=4 et sw=4 tw=99:
+ * This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "jit/EagerSimdUnbox.h"
+
+#include "jit/MIR.h"
+#include "jit/MIRGenerator.h"
+#include "jit/MIRGraph.h"
+
+namespace js {
+namespace jit {
+
+// Do not optimize any Phi instruction which has conflicting Unbox operations,
+// as this might imply some intended polymorphism.
+static bool
+CanUnboxSimdPhi(const JitRealm* jitRealm, MPhi* phi, SimdType unboxType)
+{
+    MOZ_ASSERT(phi->type() == MIRType::Object);
+
+    // If we are unboxing, we are more than likely to have boxed this SIMD type
+    // once in baseline, otherwise, we cannot create a MSimdBox as we have no
+    // template object to use.
+    if (!jitRealm->maybeGetSimdTemplateObjectFor(unboxType))
+        return false;
+
+    MResumePoint* entry = phi->block()->entryResumePoint();
+    MIRType mirType = SimdTypeToMIRType(unboxType);
+    for (MUseIterator i(phi->usesBegin()), e(phi->usesEnd()); i != e; i++) {
+        // If we cannot recover the Simd object at the entry of the basic block,
+        // then we would have to box the content anyways.
+        if ((*i)->consumer() == entry && !entry->isRecoverableOperand(*i))
+            return false;
+
+        if (!(*i)->consumer()->isDefinition())
+            continue;
+
+        MDefinition* def = (*i)->consumer()->toDefinition();
+        if (def->isSimdUnbox() && def->toSimdUnbox()->type() != mirType)
+            return false;
+    }
+
+    return true;
+}
+
+static void
+UnboxSimdPhi(const JitRealm* jitRealm, MIRGraph& graph, MPhi* phi, SimdType unboxType)
+{
+    TempAllocator& alloc = graph.alloc();
+
+    // Unbox and replace all operands.
+    for (size_t i = 0, e = phi->numOperands(); i < e; i++) {
+        MDefinition* op = phi->getOperand(i);
+        MSimdUnbox* unbox = MSimdUnbox::New(alloc, op, unboxType);
+        op->block()->insertAtEnd(unbox);
+        phi->replaceOperand(i, unbox);
+    }
+
+    // Change the MIRType of the Phi.
+    MIRType mirType = SimdTypeToMIRType(unboxType);
+    phi->setResultType(mirType);
+
+    MBasicBlock* phiBlock = phi->block();
+    MInstruction* atRecover = phiBlock->safeInsertTop(nullptr, MBasicBlock::IgnoreRecover);
+    MInstruction* at = phiBlock->safeInsertTop(atRecover);
+
+    // Note, we capture the uses-list now, as new instructions are not visited.
+    MUseIterator i(phi->usesBegin()), e(phi->usesEnd());
+
+    // Add a MSimdBox, and replace all the Phi uses with it.
+    JSObject* templateObject = jitRealm->maybeGetSimdTemplateObjectFor(unboxType);
+    InlineTypedObject* inlineTypedObject = &templateObject->as<InlineTypedObject>();
+    MSimdBox* recoverBox = MSimdBox::New(alloc, nullptr, phi, inlineTypedObject, unboxType, gc::DefaultHeap);
+    recoverBox->setRecoveredOnBailout();
+    phiBlock->insertBefore(atRecover, recoverBox);
+
+    MSimdBox* box = nullptr;
+    while (i != e) {
+        MUse* use = *i++;
+        MNode* ins = use->consumer();
+
+        if ((ins->isDefinition() && ins->toDefinition()->isRecoveredOnBailout()) ||
+            (ins->isResumePoint() && ins->toResumePoint()->isRecoverableOperand(use)))
+        {
+            use->replaceProducer(recoverBox);
+            continue;
+        }
+
+        if (!box) {
+            box = MSimdBox::New(alloc, nullptr, phi, inlineTypedObject, unboxType, gc::DefaultHeap);
+            phiBlock->insertBefore(at, box);
+        }
+
+        use->replaceProducer(box);
+    }
+}
+
+bool
+EagerSimdUnbox(MIRGenerator* mir, MIRGraph& graph)
+{
+    const JitRealm* jitRealm = mir->realm->jitRealm();
+    for (PostorderIterator block = graph.poBegin(); block != graph.poEnd(); block++) {
+        if (mir->shouldCancel("Eager Simd Unbox"))
+            return false;
+
+        for (MInstructionReverseIterator ins = block->rbegin(); ins != block->rend(); ins++) {
+            if (!ins->isSimdUnbox())
+                continue;
+
+            MSimdUnbox* unbox = ins->toSimdUnbox();
+            if (!unbox->input()->isPhi())
+                continue;
+
+            MPhi* phi = unbox->input()->toPhi();
+            if (!CanUnboxSimdPhi(jitRealm, phi, unbox->simdType()))
+                continue;
+
+            UnboxSimdPhi(jitRealm, graph, phi, unbox->simdType());
+        }
+    }
+
+    return true;
+}
+
+} /* namespace jit */
+} /* namespace js */

js/src/jit/EagerSimdUnbox.h

@@ -0,0 +1,25 @@
+/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*-
+ * vim: set ts=8 sts=4 et sw=4 tw=99:
+ * This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+// This file declares eager SIMD unboxing.
+#ifndef jit_EagerSimdUnbox_h
+#define jit_EagerSimdUnbox_h
+
+#include "mozilla/Attributes.h"
+
+namespace js {
+namespace jit {
+
+class MIRGenerator;
+class MIRGraph;
+
+MOZ_MUST_USE bool
+EagerSimdUnbox(MIRGenerator* mir, MIRGraph& graph);
+
+} // namespace jit
+} // namespace js
+
+#endif /* jit_EagerSimdUnbox_h */

js/src/jit/InlinableNatives.h

@@ -97,6 +97,17 @@
     _(ObjectIs)                     \
     _(ObjectToString)               \
                                     \
+    _(SimdInt32x4)                  \
+    _(SimdUint32x4)                 \
+    _(SimdInt16x8)                  \
+    _(SimdUint16x8)                 \
+    _(SimdInt8x16)                  \
+    _(SimdUint8x16)                 \
+    _(SimdFloat32x4)                \
+    _(SimdBool32x4)                 \
+    _(SimdBool16x8)                 \
+    _(SimdBool8x16)                 \
+                                    \
     _(TestBailout)                  \
     _(TestAssertFloat32)            \
     _(TestAssertRecoveredOnBailout) \

js/src/jit/Ion.cpp

@@ -21,6 +21,7 @@
 #include "jit/BaselineJIT.h"
 #include "jit/CacheIRSpewer.h"
 #include "jit/CodeGenerator.h"
+#include "jit/EagerSimdUnbox.h"
 #include "jit/EdgeCaseAnalysis.h"
 #include "jit/EffectiveAddressAnalysis.h"
 #include "jit/FoldLinearArithConstants.h"
@@ -443,6 +444,17 @@ JitRealm::performStubReadBarriers(uint32_t stubsToBarrier) const
     }
 }
 
+void
+JitRealm::performSIMDTemplateReadBarriers(uint32_t simdTemplatesToBarrier) const
+{
+    while (simdTemplatesToBarrier) {
+        auto type = PopNextBitmaskValue<SimdType>(&simdTemplatesToBarrier);
+        const ReadBarrieredObject& tpl = simdTemplateObjects_[type];
+        MOZ_ASSERT(tpl);
+        tpl.get();
+    }
+}
+
 bool
 JitZone::init(JSContext* cx)
 {
@@ -637,6 +649,11 @@ JitRealm::sweep(JS::Realm* realm)
         if (stub && IsAboutToBeFinalized(&stub))
             stub.set(nullptr);
     }
+
+    for (ReadBarrieredObject& obj : simdTemplateObjects_) {
+        if (obj && IsAboutToBeFinalized(&obj))
+            obj.set(nullptr);
+    }
 }
 
 void
@@ -1467,6 +1484,17 @@ OptimizeMIR(MIRGenerator* mir)
             return false;
     }
 
+    if (!JitOptions.disableRecoverIns && mir->optimizationInfo().eagerSimdUnboxEnabled()) {
+        AutoTraceLog log(logger, TraceLogger_EagerSimdUnbox);
+        if (!EagerSimdUnbox(mir, graph))
+            return false;
+        gs.spewPass("Eager Simd Unbox");
+        AssertGraphCoherency(graph);
+
+        if (mir->shouldCancel("Eager Simd Unbox"))
+            return false;
+    }
+
     if (mir->optimizationInfo().amaEnabled()) {
         AutoTraceLog log(logger, TraceLogger_AlignmentMaskAnalysis);
         AlignmentMaskAnalysis ama(graph);

js/src/jit/IonBuilder.cpp

@@ -7902,6 +7902,11 @@ IonBuilder::getElemTryTypedObject(bool* emitted, MDefinition* obj, MDefinition*
         return Ok();
 
     switch (elemPrediction.kind()) {
+      case type::Simd:
+        // FIXME (bug 894105): load into a MIRType::float32x4 etc
+        trackOptimizationOutcome(TrackedOutcome::GenericFailure);
+        return Ok();
+
       case type::Struct:
       case type::Array:
         return getElemTryComplexElemOfTypedObject(emitted,
@@ -8959,6 +8964,11 @@ IonBuilder::setElemTryTypedObject(bool* emitted, MDefinition* obj,
         return Ok();
 
     switch (elemPrediction.kind()) {
+      case type::Simd:
+        // FIXME (bug 894105): store a MIRType::float32x4 etc
+        trackOptimizationOutcome(TrackedOutcome::GenericFailure);
+        return Ok();
+
       case type::Reference:
         return setElemTryReferenceElemOfTypedObject(emitted, obj, index,
                                                     objPrediction, value, elemPrediction);
@@ -10586,6 +10596,10 @@ IonBuilder::getPropTryTypedObject(bool* emitted,
         return Ok();
 
     switch (fieldPrediction.kind()) {
+      case type::Simd:
+        // FIXME (bug 894104): load into a MIRType::float32x4 etc
+        return Ok();
+
       case type::Struct:
       case type::Array:
         return getPropTryComplexPropOfTypedObject(emitted,
@@ -11728,6 +11742,10 @@ IonBuilder::setPropTryTypedObject(bool* emitted, MDefinition* obj,
         return Ok();
 
     switch (fieldPrediction.kind()) {
+      case type::Simd:
+        // FIXME (bug 894104): store into a MIRType::float32x4 etc
+        return Ok();
+
       case type::Reference:
         return setPropTryReferencePropOfTypedObject(emitted, obj, fieldOffset,
                                                     value, fieldPrediction, name);

js/src/jit/IonBuilder.h

@@ -730,6 +730,51 @@ class IonBuilder
     InliningResult inlineSetTypedObjectOffset(CallInfo& callInfo);
     InliningResult inlineConstructTypedObject(CallInfo& callInfo, TypeDescr* target);
 
+    // SIMD intrinsics and natives.
+    InliningResult inlineConstructSimdObject(CallInfo& callInfo, SimdTypeDescr* target);
+
+    // SIMD helpers.
+    bool canInlineSimd(CallInfo& callInfo, JSNative native, unsigned numArgs,
+                       InlineTypedObject** templateObj);
+    MDefinition* unboxSimd(MDefinition* ins, SimdType type);
+    InliningResult boxSimd(CallInfo& callInfo, MDefinition* ins, InlineTypedObject* templateObj);
+    MDefinition* convertToBooleanSimdLane(MDefinition* scalar);
+
+    InliningResult inlineSimd(CallInfo& callInfo, JSFunction* target, SimdType type);
+
+    InliningResult inlineSimdBinaryArith(CallInfo& callInfo, JSNative native,
+                                         MSimdBinaryArith::Operation op, SimdType type);
+    InliningResult inlineSimdBinaryBitwise(CallInfo& callInfo, JSNative native,
+                                           MSimdBinaryBitwise::Operation op, SimdType type);
+    InliningResult inlineSimdBinarySaturating(CallInfo& callInfo, JSNative native,
+                                              MSimdBinarySaturating::Operation op, SimdType type);
+    InliningResult inlineSimdShift(CallInfo& callInfo, JSNative native, MSimdShift::Operation op,
+                                   SimdType type);
+    InliningResult inlineSimdComp(CallInfo& callInfo, JSNative native,
+                                  MSimdBinaryComp::Operation op, SimdType type);
+    InliningResult inlineSimdUnary(CallInfo& callInfo, JSNative native,
+                                   MSimdUnaryArith::Operation op, SimdType type);
+    InliningResult inlineSimdExtractLane(CallInfo& callInfo, JSNative native, SimdType type);
+    InliningResult inlineSimdReplaceLane(CallInfo& callInfo, JSNative native, SimdType type);
+    InliningResult inlineSimdSplat(CallInfo& callInfo, JSNative native, SimdType type);
+    InliningResult inlineSimdShuffle(CallInfo& callInfo, JSNative native, SimdType type,
+                                     unsigned numVectors);
+    InliningResult inlineSimdCheck(CallInfo& callInfo, JSNative native, SimdType type);
+    InliningResult inlineSimdConvert(CallInfo& callInfo, JSNative native, bool isCast,
+                                     SimdType from, SimdType to);
+    InliningResult inlineSimdSelect(CallInfo& callInfo, JSNative native, SimdType type);
+
+    bool prepareForSimdLoadStore(CallInfo& callInfo, Scalar::Type simdType,
+                                 MInstruction** elements, MDefinition** index,
+                                 Scalar::Type* arrayType);
+    InliningResult inlineSimdLoad(CallInfo& callInfo, JSNative native, SimdType type,
+                                  unsigned numElems);
+    InliningResult inlineSimdStore(CallInfo& callInfo, JSNative native, SimdType type,
+                                   unsigned numElems);
+
+    InliningResult inlineSimdAnyAllTrue(CallInfo& callInfo, bool IsAllTrue, JSNative native,
+                                        SimdType type);
+
     // Utility intrinsics.
     InliningResult inlineIsCallable(CallInfo& callInfo);
     InliningResult inlineIsConstructor(CallInfo& callInfo);

js/src/jit/IonOptimizationLevels.cpp

@@ -27,6 +27,7 @@ OptimizationInfo::initNormalOptimizationInfo()
 
     autoTruncate_ = true;
     eaa_ = true;
+    eagerSimdUnbox_ = true;
     edgeCaseAnalysis_ = true;
     eliminateRedundantChecks_ = true;
     inlineInterpreted_ = true;
@@ -68,6 +69,7 @@ OptimizationInfo::initWasmOptimizationInfo()
 
     ama_ = true;
     autoTruncate_ = false;
+    eagerSimdUnbox_ = false;           // wasm has no boxing / unboxing.
     edgeCaseAnalysis_ = false;
     eliminateRedundantChecks_ = false;
     scalarReplacement_ = false;        // wasm has no objects.

js/src/jit/IonOptimizationLevels.h

@@ -65,6 +65,9 @@ class OptimizationInfo
     // Toggles whether native scripts get inlined.
     bool inlineNative_;
 
+    // Toggles whether eager unboxing of SIMD is used.
+    bool eagerSimdUnbox_;
+
     // Toggles whether global value numbering is used.
     bool gvn_;
 
@@ -154,6 +157,7 @@ class OptimizationInfo
         eliminateRedundantChecks_(false),
         inlineInterpreted_(false),
         inlineNative_(false),
+        eagerSimdUnbox_(false),
         gvn_(false),
         licm_(false),
         rangeAnalysis_(false),
@@ -194,6 +198,10 @@ class OptimizationInfo
 
     uint32_t compilerWarmUpThreshold(JSScript* script, jsbytecode* pc = nullptr) const;
 
+    bool eagerSimdUnboxEnabled() const {
+        return eagerSimdUnbox_ && !JitOptions.disableEagerSimdUnbox;
+    }
+
     bool gvnEnabled() const {
         return gvn_ && !JitOptions.disableGvn;
     }

js/src/jit/IonTypes.h

@@ -104,6 +104,9 @@ enum BailoutKind
     Bailout_NonStringInput,
     Bailout_NonSymbolInput,
 
+    // SIMD Unbox expects a given type, bails out if it doesn't match.
+    Bailout_UnexpectedSimdInput,
+
     // Atomic operations require shared memory, bail out if the typed array
     // maps unshared memory.
     Bailout_NonSharedTypedArrayInput,
@@ -205,6 +208,8 @@ BailoutKindString(BailoutKind kind)
         return "Bailout_NonStringInput";
       case Bailout_NonSymbolInput:
         return "Bailout_NonSymbolInput";
+      case Bailout_UnexpectedSimdInput:
+        return "Bailout_UnexpectedSimdInput";
       case Bailout_NonSharedTypedArrayInput:
         return "Bailout_NonSharedTypedArrayInput";
       case Bailout_Debugger:
@@ -247,19 +252,6 @@ static const uint32_t VECTOR_SCALE_BITS = 3;
 static const uint32_t VECTOR_SCALE_SHIFT = ELEMENT_TYPE_BITS + ELEMENT_TYPE_SHIFT;
 static const uint32_t VECTOR_SCALE_MASK = (1 << VECTOR_SCALE_BITS) - 1;
 
-// The integer SIMD types have a lot of operations that do the exact same thing
-// for signed and unsigned integer types. Sometimes it is simpler to treat
-// signed and unsigned integer SIMD types as the same type, using a SimdSign to
-// distinguish the few cases where there is a difference.
-enum class SimdSign {
-    // Signedness is not applicable to this type. (i.e., Float or Bool).
-    NotApplicable,
-    // Treat as an unsigned integer with a range 0 .. 2^N-1.
-    Unsigned,
-    // Treat as a signed integer in two's complement encoding.
-    Signed,
-};
-
 class SimdConstant {
   public:
     enum Type {
@@ -465,6 +457,39 @@ IsSimdType(MIRType type)
     return ((unsigned(type) >> VECTOR_SCALE_SHIFT) & VECTOR_SCALE_MASK) != 0;
 }
 
+// Returns the number of vector elements (hereby called "length") for a given
+// SIMD kind. It is the Y part of the name "Foo x Y".
+static inline unsigned
+SimdTypeToLength(MIRType type)
+{
+    MOZ_ASSERT(IsSimdType(type));
+    return 1 << ((unsigned(type) >> VECTOR_SCALE_SHIFT) & VECTOR_SCALE_MASK);
+}
+
+// Get the type of the individual lanes in a SIMD type.
+// For example, Int32x4 -> Int32, Float32x4 -> Float32 etc.
+static inline MIRType
+SimdTypeToLaneType(MIRType type)
+{
+    MOZ_ASSERT(IsSimdType(type));
+    static_assert(unsigned(MIRType::Last) <= ELEMENT_TYPE_MASK,
+                  "ELEMENT_TYPE_MASK should be larger than the last MIRType");
+    return MIRType((unsigned(type) >> ELEMENT_TYPE_SHIFT) & ELEMENT_TYPE_MASK);
+}
+
+// Get the type expected when inserting a lane into a SIMD type.
+// This is the argument type expected by the MSimdValue constructors as well as
+// MSimdSplat and MSimdInsertElement.
+static inline MIRType
+SimdTypeToLaneArgumentType(MIRType type)
+{
+    MIRType laneType = SimdTypeToLaneType(type);
+
+    // Boolean lanes should be pre-converted to an Int32 with the values 0 or -1.
+    // All other lane types are inserted directly.
+    return laneType == MIRType::Boolean ? MIRType::Int32 : laneType;
+}
+
 static inline MIRType
 MIRTypeFromValueType(JSValueType type)
 {
@@ -664,6 +689,24 @@ IsNullOrUndefined(MIRType type)
     return type == MIRType::Null || type == MIRType::Undefined;
 }
 
+static inline bool
+IsFloatingPointSimdType(MIRType type)
+{
+    return type == MIRType::Float32x4;
+}
+
+static inline bool
+IsIntegerSimdType(MIRType type)
+{
+    return IsSimdType(type) && SimdTypeToLaneType(type) == MIRType::Int32;
+}
+
+static inline bool
+IsBooleanSimdType(MIRType type)
+{
+    return IsSimdType(type) && SimdTypeToLaneType(type) == MIRType::Boolean;
+}
+
 static inline bool
 IsMagicType(MIRType type)
 {
@@ -692,10 +735,18 @@ ScalarTypeToMIRType(Scalar::Type type)
         return MIRType::Float32;
       case Scalar::Float64:
         return MIRType::Double;
+      case Scalar::Float32x4:
+        return MIRType::Float32x4;
+      case Scalar::Int8x16:
+        return MIRType::Int8x16;
+      case Scalar::Int16x8:
+        return MIRType::Int16x8;
+      case Scalar::Int32x4:
+        return MIRType::Int32x4;
       case Scalar::MaxTypedArrayViewType:
         break;
     }
-    MOZ_CRASH("unexpected kind");
+    MOZ_CRASH("unexpected SIMD kind");
 }
 
 static inline unsigned
@@ -713,10 +764,17 @@ ScalarTypeToLength(Scalar::Type type)
       case Scalar::Float64:
       case Scalar::Uint8Clamped:
         return 1;
+      case Scalar::Float32x4:
+      case Scalar::Int32x4:
+        return 4;
+      case Scalar::Int16x8:
+        return 8;
+      case Scalar::Int8x16:
+        return 16;
       case Scalar::MaxTypedArrayViewType:
         break;
     }
-    MOZ_CRASH("unexpected kind");
+    MOZ_CRASH("unexpected SIMD kind");
 }
 
 static inline const char*

js/src/jit/JSJitFrameIter.h

@@ -407,7 +407,9 @@ struct MaybeReadFallback
     }
 };
 
+
 class RResumePoint;
+class RSimdBox;
 
 // Reads frame information in snapshot-encoding order (that is, outermost frame
 // to innermost frame).
@@ -469,6 +471,7 @@ class SnapshotIterator
     void warnUnreadableAllocation();
 
   private:
+    friend class RSimdBox;
     const FloatRegisters::RegisterContent* floatAllocationPointer(const RValueAllocation& a) const;
 
   public:

js/src/jit/JitOptions.cpp

@@ -86,6 +86,9 @@ DefaultJitOptions::DefaultJitOptions()
     // Toggles whether Effective Address Analysis is globally disabled.
     SET_DEFAULT(disableEaa, false);
 
+    // Toggle whether eager simd unboxing is globally disabled.
+    SET_DEFAULT(disableEagerSimdUnbox, false);
+
     // Toggles whether Edge Case Analysis is gobally disabled.
     SET_DEFAULT(disableEdgeCaseAnalysis, false);
 

js/src/jit/JitOptions.h

@@ -50,6 +50,7 @@ struct DefaultJitOptions
     bool disableInlineBacktracking;
     bool disableAma;
     bool disableEaa;
+    bool disableEagerSimdUnbox;
     bool disableEdgeCaseAnalysis;
     bool disableGvn;
     bool disableInlining;

js/src/jit/JitRealm.h

@@ -507,6 +507,10 @@ class JitRealm
 
     mozilla::EnumeratedArray<StubIndex, StubIndex::Count, ReadBarrieredJitCode> stubs_;
 
+    // The same approach is taken for SIMD template objects.
+
+    mozilla::EnumeratedArray<SimdType, SimdType::Count, ReadBarrieredObject> simdTemplateObjects_;
+
     JitCode* generateStringConcatStub(JSContext* cx);
     JitCode* generateRegExpMatcherStub(JSContext* cx);
     JitCode* generateRegExpSearcherStub(JSContext* cx);
@@ -519,6 +523,23 @@ class JitRealm
     }
 
   public:
+    JSObject* getSimdTemplateObjectFor(JSContext* cx, Handle<SimdTypeDescr*> descr) {
+        ReadBarrieredObject& tpl = simdTemplateObjects_[descr->type()];
+        if (!tpl)
+            tpl.set(TypedObject::createZeroed(cx, descr, gc::TenuredHeap));
+        return tpl.get();
+    }
+
+    JSObject* maybeGetSimdTemplateObjectFor(SimdType type) const {
+        // This function is used by Eager Simd Unbox phase which can run
+        // off-thread, so we cannot use the usual read barrier. For more
+        // information, see the comment above
+        // CodeGenerator::simdRefreshTemplatesDuringLink_.
+
+        MOZ_ASSERT(CurrentThreadIsIonCompiling());
+        return simdTemplateObjects_[type].unbarrieredGet();
+    }
+
     JitCode* getStubCode(uint32_t key) {
         ICStubCodeMap::Ptr p = stubCodes_->lookup(key);
         if (p)
@@ -599,13 +620,15 @@ class JitRealm
         return stubs_[RegExpTester];
     }
 
-    // Perform the necessary read barriers on stubs described by the bitmasks
+    // Perform the necessary read barriers on stubs and SIMD template object
-    // passed in. This function can only be called from the main thread.
+    // described by the bitmasks passed in. This function can only be called
+    // from the main thread.
     //
-    // The stub pointers must still be valid by the time these methods are
+    // The stub and template object pointers must still be valid by the time
-    // called. This is arranged by cancelling off-thread Ion compilation at the
+    // these methods are called. This is arranged by cancelling off-thread Ion
-    // start of GC and at the start of sweeping.
+    // compilation at the start of GC and at the start of sweeping.
     void performStubReadBarriers(uint32_t stubsToBarrier) const;
+    void performSIMDTemplateReadBarriers(uint32_t simdTemplatesToBarrier) const;
 
     size_t sizeOfIncludingThis(mozilla::MallocSizeOf mallocSizeOf) const;
 

js/src/jit/Lowering.cpp

@@ -3689,7 +3689,8 @@ LIRGenerator::visitLoadUnboxedScalar(MLoadUnboxedScalar* ins)
     const LUse elements = useRegister(ins->elements());
     const LAllocation index = useRegisterOrConstant(ins->index());
 
-    MOZ_ASSERT(IsNumberType(ins->type()) || ins->type() == MIRType::Boolean);
+    MOZ_ASSERT(IsNumberType(ins->type()) || IsSimdType(ins->type()) ||
+               ins->type() == MIRType::Boolean);
 
     // We need a temp register for Uint32Array with known double result.
     LDefinition tempDef = LDefinition::BogusTemp();
@@ -3770,7 +3771,12 @@ LIRGenerator::visitStoreUnboxedScalar(MStoreUnboxedScalar* ins)
     MOZ_ASSERT(IsValidElementsType(ins->elements(), ins->offsetAdjustment()));
     MOZ_ASSERT(ins->index()->type() == MIRType::Int32);
 
-    if (ins->isFloatWrite()) {
+    if (ins->isSimdWrite()) {
+        MOZ_ASSERT_IF(ins->writeType() == Scalar::Float32x4, ins->value()->type() == MIRType::Float32x4);
+        MOZ_ASSERT_IF(ins->writeType() == Scalar::Int8x16, ins->value()->type() == MIRType::Int8x16);
+        MOZ_ASSERT_IF(ins->writeType() == Scalar::Int16x8, ins->value()->type() == MIRType::Int16x8);
+        MOZ_ASSERT_IF(ins->writeType() == Scalar::Int32x4, ins->value()->type() == MIRType::Int32x4);
+    } else if (ins->isFloatWrite()) {
         MOZ_ASSERT_IF(ins->writeType() == Scalar::Float32, ins->value()->type() == MIRType::Float32);
         MOZ_ASSERT_IF(ins->writeType() == Scalar::Float64, ins->value()->type() == MIRType::Double);
     } else {
@@ -4704,7 +4710,7 @@ LIRGenerator::visitWasmParameter(MWasmParameter* ins)
 #endif
         );
     } else {
-        MOZ_ASSERT(IsNumberType(ins->type()));
+        MOZ_ASSERT(IsNumberType(ins->type()) || IsSimdType(ins->type()));
         defineFixed(new(alloc()) LWasmParameter, ins, LArgument(abi.offsetFromArgBase()));
     }
 }
@@ -4724,6 +4730,8 @@ LIRGenerator::visitWasmReturn(MWasmReturn* ins)
         lir->setOperand(0, useFixed(rval, ReturnFloat32Reg));
     else if (rval->type() == MIRType::Double)
         lir->setOperand(0, useFixed(rval, ReturnDoubleReg));
+    else if (IsSimdType(rval->type()))
+        lir->setOperand(0, useFixed(rval, ReturnSimd128Reg));
     else if (rval->type() == MIRType::Int32)
         lir->setOperand(0, useFixed(rval, ReturnReg));
     else
@@ -4743,7 +4751,7 @@ LIRGenerator::visitWasmStackArg(MWasmStackArg* ins)
 {
     if (ins->arg()->type() == MIRType::Int64) {
         add(new(alloc()) LWasmStackArgI64(useInt64RegisterOrConstantAtStart(ins->arg())), ins);
-    } else if (IsFloatingPointType(ins->arg()->type())) {
+    } else if (IsFloatingPointType(ins->arg()->type()) || IsSimdType(ins->arg()->type())) {
         MOZ_ASSERT(!ins->arg()->isEmittedAtUses());
         add(new(alloc()) LWasmStackArg(useRegisterAtStart(ins->arg())), ins);
     } else {
@@ -4878,6 +4886,217 @@ LIRGenerator::visitRecompileCheck(MRecompileCheck* ins)
     assignSafepoint(lir, ins);
 }
 
+void
+LIRGenerator::visitSimdBox(MSimdBox* ins)
+{
+    MOZ_ASSERT(IsSimdType(ins->input()->type()));
+    LUse in = useRegister(ins->input());
+    LSimdBox* lir = new(alloc()) LSimdBox(in, temp());
+    define(lir, ins);
+    assignSafepoint(lir, ins);
+}
+
+void
+LIRGenerator::visitSimdUnbox(MSimdUnbox* ins)
+{
+    MOZ_ASSERT(ins->input()->type() == MIRType::Object);
+    MOZ_ASSERT(IsSimdType(ins->type()));
+    LUse in = useRegister(ins->input());
+    LSimdUnbox* lir = new(alloc()) LSimdUnbox(in, temp());
+    assignSnapshot(lir, Bailout_UnexpectedSimdInput);
+    define(lir, ins);
+}
+
+void
+LIRGenerator::visitSimdConstant(MSimdConstant* ins)
+{
+    MOZ_ASSERT(IsSimdType(ins->type()));
+
+    switch (ins->type()) {
+      case MIRType::Int8x16:
+      case MIRType::Int16x8:
+      case MIRType::Int32x4:
+      case MIRType::Bool8x16:
+      case MIRType::Bool16x8:
+      case MIRType::Bool32x4:
+        define(new(alloc()) LSimd128Int(), ins);
+        break;
+      case MIRType::Float32x4:
+        define(new(alloc()) LSimd128Float(), ins);
+        break;
+      default:
+        MOZ_CRASH("Unknown SIMD kind when generating constant");
+    }
+}
+
+void
+LIRGenerator::visitSimdConvert(MSimdConvert* ins)
+{
+    MOZ_ASSERT(IsSimdType(ins->type()));
+    MDefinition* input = ins->input();
+    LUse use = useRegister(input);
+    if (ins->type() == MIRType::Int32x4) {
+        MOZ_ASSERT(input->type() == MIRType::Float32x4);
+        switch (ins->signedness()) {
+          case SimdSign::Signed: {
+              LFloat32x4ToInt32x4* lir = new(alloc()) LFloat32x4ToInt32x4(use, temp());
+              if (!gen->compilingWasm())
+                  assignSnapshot(lir, Bailout_BoundsCheck);
+              define(lir, ins);
+              break;
+          }
+          case SimdSign::Unsigned: {
+              LFloat32x4ToUint32x4* lir =
+                new (alloc()) LFloat32x4ToUint32x4(use, temp(), temp(LDefinition::SIMD128INT));
+              if (!gen->compilingWasm())
+                  assignSnapshot(lir, Bailout_BoundsCheck);
+              define(lir, ins);
+              break;
+          }
+          default:
+            MOZ_CRASH("Unexpected SimdConvert sign");
+        }
+    } else if (ins->type() == MIRType::Float32x4) {
+        MOZ_ASSERT(input->type() == MIRType::Int32x4);
+        MOZ_ASSERT(ins->signedness() == SimdSign::Signed, "Unexpected SimdConvert sign");
+        define(new(alloc()) LInt32x4ToFloat32x4(use), ins);
+    } else {
+        MOZ_CRASH("Unknown SIMD kind when generating constant");
+    }
+}
+
+void
+LIRGenerator::visitSimdReinterpretCast(MSimdReinterpretCast* ins)
+{
+    MOZ_ASSERT(IsSimdType(ins->type()) && IsSimdType(ins->input()->type()));
+    MDefinition* input = ins->input();
+    LUse use = useRegisterAtStart(input);
+    // :TODO: (Bug 1132894) We have to allocate a different register as redefine
+    // and/or defineReuseInput are not yet capable of reusing the same register
+    // with a different register type.
+    define(new(alloc()) LSimdReinterpretCast(use), ins);
+}
+
+void
+LIRGenerator::visitSimdAllTrue(MSimdAllTrue* ins)
+{
+    MDefinition* input = ins->input();
+    MOZ_ASSERT(IsBooleanSimdType(input->type()));
+
+    LUse use = useRegisterAtStart(input);
+    define(new(alloc()) LSimdAllTrue(use), ins);
+}
+
+void
+LIRGenerator::visitSimdAnyTrue(MSimdAnyTrue* ins)
+{
+    MDefinition* input = ins->input();
+    MOZ_ASSERT(IsBooleanSimdType(input->type()));
+
+    LUse use = useRegisterAtStart(input);
+    define(new(alloc()) LSimdAnyTrue(use), ins);
+}
+
+void
+LIRGenerator::visitSimdUnaryArith(MSimdUnaryArith* ins)
+{
+    MOZ_ASSERT(IsSimdType(ins->input()->type()));
+    MOZ_ASSERT(IsSimdType(ins->type()));
+
+    // Cannot be at start, as the ouput is used as a temporary to store values.
+    LUse in = use(ins->input());
+
+    switch (ins->type()) {
+      case MIRType::Int8x16:
+      case MIRType::Bool8x16:
+        define(new (alloc()) LSimdUnaryArithIx16(in), ins);
+        break;
+      case MIRType::Int16x8:
+      case MIRType::Bool16x8:
+        define(new (alloc()) LSimdUnaryArithIx8(in), ins);
+        break;
+      case MIRType::Int32x4:
+      case MIRType::Bool32x4:
+        define(new (alloc()) LSimdUnaryArithIx4(in), ins);
+        break;
+      case MIRType::Float32x4:
+        define(new (alloc()) LSimdUnaryArithFx4(in), ins);
+        break;
+      default:
+        MOZ_CRASH("Unknown SIMD kind for unary operation");
+    }
+}
+
+void
+LIRGenerator::visitSimdBinaryComp(MSimdBinaryComp* ins)
+{
+    MOZ_ASSERT(IsSimdType(ins->lhs()->type()));
+    MOZ_ASSERT(IsSimdType(ins->rhs()->type()));
+    MOZ_ASSERT(IsBooleanSimdType(ins->type()));
+
+    if (ShouldReorderCommutative(ins->lhs(), ins->rhs(), ins))
+        ins->reverse();
+
+    switch (ins->specialization()) {
+      case MIRType::Int8x16: {
+          MOZ_ASSERT(ins->signedness() == SimdSign::Signed);
+          LSimdBinaryCompIx16* add = new (alloc()) LSimdBinaryCompIx16();
+          lowerForFPU(add, ins, ins->lhs(), ins->rhs());
+          return;
+      }
+      case MIRType::Int16x8: {
+          MOZ_ASSERT(ins->signedness() == SimdSign::Signed);
+          LSimdBinaryCompIx8* add = new (alloc()) LSimdBinaryCompIx8();
+          lowerForFPU(add, ins, ins->lhs(), ins->rhs());
+          return;
+      }
+      case MIRType::Int32x4: {
+          MOZ_ASSERT(ins->signedness() == SimdSign::Signed);
+          LSimdBinaryCompIx4* add = new (alloc()) LSimdBinaryCompIx4();
+          lowerForCompIx4(add, ins, ins->lhs(), ins->rhs());
+          return;
+      }
+      case MIRType::Float32x4: {
+          MOZ_ASSERT(ins->signedness() == SimdSign::NotApplicable);
+          LSimdBinaryCompFx4* add = new (alloc()) LSimdBinaryCompFx4();
+          lowerForCompFx4(add, ins, ins->lhs(), ins->rhs());
+          return;
+      }
+      default:
+        MOZ_CRASH("Unknown compare type when comparing values");
+    }
+}
+
+void
+LIRGenerator::visitSimdBinaryBitwise(MSimdBinaryBitwise* ins)
+{
+    MOZ_ASSERT(IsSimdType(ins->lhs()->type()));
+    MOZ_ASSERT(IsSimdType(ins->rhs()->type()));
+    MOZ_ASSERT(IsSimdType(ins->type()));
+
+    MDefinition* lhs = ins->lhs();
+    MDefinition* rhs = ins->rhs();
+    ReorderCommutative(&lhs, &rhs, ins);
+    LSimdBinaryBitwise* lir = new(alloc()) LSimdBinaryBitwise;
+    lowerForFPU(lir, ins, lhs, rhs);
+}
+
+void
+LIRGenerator::visitSimdShift(MSimdShift* ins)
+{
+    MOZ_ASSERT(IsIntegerSimdType(ins->type()));
+    MOZ_ASSERT(ins->lhs()->type() == ins->type());
+    MOZ_ASSERT(ins->rhs()->type() == MIRType::Int32);
+
+    LUse vector = useRegisterAtStart(ins->lhs());
+    LAllocation value = useRegisterOrConstant(ins->rhs());
+    // We need a temp register to mask the shift amount, but not if the shift
+    // amount is a constant.
+    LDefinition tempReg = value.isConstant() ? LDefinition::BogusTemp() : temp();
+    LSimdShift* lir = new(alloc()) LSimdShift(vector, value, tempReg);
+    defineReuseInput(lir, ins, 0);
+}
+
 void
 LIRGenerator::visitLexicalCheck(MLexicalCheck* ins)
 {

js/src/jit/MCallOptimize.cpp

@@ -15,6 +15,7 @@
 #include "builtin/intl/PluralRules.h"
 #include "builtin/intl/RelativeTimeFormat.h"
 #include "builtin/MapObject.h"
+#include "builtin/SIMDConstants.h"
 #include "builtin/String.h"
 #include "builtin/TestingFunctions.h"
 #include "builtin/TypedObject.h"
@@ -262,6 +263,28 @@ IonBuilder::inlineNativeCall(CallInfo& callInfo, JSFunction* target)
       case InlinableNative::ObjectToString:
         return inlineObjectToString(callInfo);
 
+      // SIMD natives.
+      case InlinableNative::SimdInt32x4:
+        return inlineSimd(callInfo, target, SimdType::Int32x4);
+      case InlinableNative::SimdUint32x4:
+        return inlineSimd(callInfo, target, SimdType::Uint32x4);
+      case InlinableNative::SimdInt16x8:
+        return inlineSimd(callInfo, target, SimdType::Int16x8);
+      case InlinableNative::SimdUint16x8:
+        return inlineSimd(callInfo, target, SimdType::Uint16x8);
+      case InlinableNative::SimdInt8x16:
+        return inlineSimd(callInfo, target, SimdType::Int8x16);
+      case InlinableNative::SimdUint8x16:
+        return inlineSimd(callInfo, target, SimdType::Uint8x16);
+      case InlinableNative::SimdFloat32x4:
+        return inlineSimd(callInfo, target, SimdType::Float32x4);
+      case InlinableNative::SimdBool32x4:
+        return inlineSimd(callInfo, target, SimdType::Bool32x4);
+      case InlinableNative::SimdBool16x8:
+        return inlineSimd(callInfo, target, SimdType::Bool16x8);
+      case InlinableNative::SimdBool8x16:
+        return inlineSimd(callInfo, target, SimdType::Bool8x16);
+
       // Testing functions.
       case InlinableNative::TestBailout:
         return inlineBailout(callInfo);
@@ -456,6 +479,9 @@ IonBuilder::inlineNonFunctionCall(CallInfo& callInfo, JSObject* target)
     if (callInfo.constructing() && target->constructHook() == TypedObject::construct)
         return inlineConstructTypedObject(callInfo, &target->as<TypeDescr>());
 
+    if (!callInfo.constructing() && target->callHook() == SimdTypeDescr::call)
+        return inlineConstructSimdObject(callInfo, &target->as<SimdTypeDescr>());
+
     return InliningStatus_NotInlined;
 }
 
@@ -3771,6 +3797,766 @@ IonBuilder::inlineConstructTypedObject(CallInfo& callInfo, TypeDescr* descr)
     return InliningStatus_Inlined;
 }
 
+// Main entry point for SIMD inlining.
+// When the controlling simdType is an integer type, sign indicates whether the lanes should
+// be treated as signed or unsigned integers.
+IonBuilder::InliningResult
+IonBuilder::inlineSimd(CallInfo& callInfo, JSFunction* target, SimdType type)
+{
+    if (!JitSupportsSimd()) {
+        trackOptimizationOutcome(TrackedOutcome::NoSimdJitSupport);
+        return InliningStatus_NotInlined;
+    }
+
+    JSNative native = target->native();
+    SimdOperation simdOp = SimdOperation(target->jitInfo()->nativeOp);
+
+    switch(simdOp) {
+      case SimdOperation::Constructor:
+        // SIMD constructor calls are handled via inlineNonFunctionCall(), so
+        // they won't show up here where target is required to be a JSFunction.
+        // See also inlineConstructSimdObject().
+        MOZ_CRASH("SIMD constructor call not expected.");
+      case SimdOperation::Fn_check:
+        return inlineSimdCheck(callInfo, native, type);
+      case SimdOperation::Fn_splat:
+        return inlineSimdSplat(callInfo, native, type);
+      case SimdOperation::Fn_extractLane:
+        return inlineSimdExtractLane(callInfo, native, type);
+      case SimdOperation::Fn_replaceLane:
+        return inlineSimdReplaceLane(callInfo, native, type);
+      case SimdOperation::Fn_select:
+        return inlineSimdSelect(callInfo, native, type);
+      case SimdOperation::Fn_swizzle:
+        return inlineSimdShuffle(callInfo, native, type, 1);
+      case SimdOperation::Fn_shuffle:
+        return inlineSimdShuffle(callInfo, native, type, 2);
+
+        // Unary arithmetic.
+      case SimdOperation::Fn_abs:
+        return inlineSimdUnary(callInfo, native, MSimdUnaryArith::abs, type);
+      case SimdOperation::Fn_neg:
+        return inlineSimdUnary(callInfo, native, MSimdUnaryArith::neg, type);
+      case SimdOperation::Fn_not:
+        return inlineSimdUnary(callInfo, native, MSimdUnaryArith::not_, type);
+      case SimdOperation::Fn_reciprocalApproximation:
+        return inlineSimdUnary(callInfo, native, MSimdUnaryArith::reciprocalApproximation,
+                               type);
+      case SimdOperation::Fn_reciprocalSqrtApproximation:
+        return inlineSimdUnary(callInfo, native, MSimdUnaryArith::reciprocalSqrtApproximation,
+                               type);
+      case SimdOperation::Fn_sqrt:
+        return inlineSimdUnary(callInfo, native, MSimdUnaryArith::sqrt, type);
+
+        // Binary arithmetic.
+      case SimdOperation::Fn_add:
+        return inlineSimdBinaryArith(callInfo, native, MSimdBinaryArith::Op_add, type);
+      case SimdOperation::Fn_sub:
+        return inlineSimdBinaryArith(callInfo, native, MSimdBinaryArith::Op_sub, type);
+      case SimdOperation::Fn_mul:
+        return inlineSimdBinaryArith(callInfo, native, MSimdBinaryArith::Op_mul, type);
+      case SimdOperation::Fn_div:
+        return inlineSimdBinaryArith(callInfo, native, MSimdBinaryArith::Op_div, type);
+      case SimdOperation::Fn_max:
+        return inlineSimdBinaryArith(callInfo, native, MSimdBinaryArith::Op_max, type);
+      case SimdOperation::Fn_min:
+        return inlineSimdBinaryArith(callInfo, native, MSimdBinaryArith::Op_min, type);
+      case SimdOperation::Fn_maxNum:
+        return inlineSimdBinaryArith(callInfo, native, MSimdBinaryArith::Op_maxNum, type);
+      case SimdOperation::Fn_minNum:
+        return inlineSimdBinaryArith(callInfo, native, MSimdBinaryArith::Op_minNum, type);
+
+        // Binary saturating.
+      case SimdOperation::Fn_addSaturate:
+        return inlineSimdBinarySaturating(callInfo, native, MSimdBinarySaturating::add, type);
+      case SimdOperation::Fn_subSaturate:
+        return inlineSimdBinarySaturating(callInfo, native, MSimdBinarySaturating::sub, type);
+
+        // Binary bitwise.
+      case SimdOperation::Fn_and:
+        return inlineSimdBinaryBitwise(callInfo, native, MSimdBinaryBitwise::and_, type);
+      case SimdOperation::Fn_or:
+        return inlineSimdBinaryBitwise(callInfo, native, MSimdBinaryBitwise::or_, type);
+      case SimdOperation::Fn_xor:
+        return inlineSimdBinaryBitwise(callInfo, native, MSimdBinaryBitwise::xor_, type);
+
+      // Shifts.
+      case SimdOperation::Fn_shiftLeftByScalar:
+        return inlineSimdShift(callInfo, native, MSimdShift::lsh, type);
+      case SimdOperation::Fn_shiftRightByScalar:
+        return inlineSimdShift(callInfo, native, MSimdShift::rshForSign(GetSimdSign(type)), type);
+
+        // Boolean unary.
+      case SimdOperation::Fn_allTrue:
+        return inlineSimdAnyAllTrue(callInfo, /* IsAllTrue= */true, native, type);
+      case SimdOperation::Fn_anyTrue:
+        return inlineSimdAnyAllTrue(callInfo, /* IsAllTrue= */false, native, type);
+
+        // Comparisons.
+      case SimdOperation::Fn_lessThan:
+        return inlineSimdComp(callInfo, native, MSimdBinaryComp::lessThan, type);
+      case SimdOperation::Fn_lessThanOrEqual:
+        return inlineSimdComp(callInfo, native, MSimdBinaryComp::lessThanOrEqual, type);
+      case SimdOperation::Fn_equal:
+        return inlineSimdComp(callInfo, native, MSimdBinaryComp::equal, type);
+      case SimdOperation::Fn_notEqual:
+        return inlineSimdComp(callInfo, native, MSimdBinaryComp::notEqual, type);
+      case SimdOperation::Fn_greaterThan:
+        return inlineSimdComp(callInfo, native, MSimdBinaryComp::greaterThan, type);
+      case SimdOperation::Fn_greaterThanOrEqual:
+        return inlineSimdComp(callInfo, native, MSimdBinaryComp::greaterThanOrEqual, type);
+
+        // Int <-> Float conversions.
+      case SimdOperation::Fn_fromInt32x4:
+        return inlineSimdConvert(callInfo, native, false, SimdType::Int32x4, type);
+      case SimdOperation::Fn_fromUint32x4:
+        return inlineSimdConvert(callInfo, native, false, SimdType::Uint32x4, type);
+      case SimdOperation::Fn_fromFloat32x4:
+        return inlineSimdConvert(callInfo, native, false, SimdType::Float32x4, type);
+
+        // Load/store.
+      case SimdOperation::Fn_load:
+        return inlineSimdLoad(callInfo, native, type, GetSimdLanes(type));
+      case SimdOperation::Fn_load1:
+        return inlineSimdLoad(callInfo, native, type, 1);
+      case SimdOperation::Fn_load2:
+        return inlineSimdLoad(callInfo, native, type, 2);
+      case SimdOperation::Fn_load3:
+        return inlineSimdLoad(callInfo, native, type, 3);
+      case SimdOperation::Fn_store:
+        return inlineSimdStore(callInfo, native, type, GetSimdLanes(type));
+      case SimdOperation::Fn_store1:
+        return inlineSimdStore(callInfo, native, type, 1);
+      case SimdOperation::Fn_store2:
+        return inlineSimdStore(callInfo, native, type, 2);
+      case SimdOperation::Fn_store3:
+        return inlineSimdStore(callInfo, native, type, 3);
+
+        // Bitcasts. One for each type with a memory representation.
+      case SimdOperation::Fn_fromInt32x4Bits:
+        return inlineSimdConvert(callInfo, native, true, SimdType::Int32x4, type);
+      case SimdOperation::Fn_fromUint32x4Bits:
+        return inlineSimdConvert(callInfo, native, true, SimdType::Uint32x4, type);
+      case SimdOperation::Fn_fromInt16x8Bits:
+        return inlineSimdConvert(callInfo, native, true, SimdType::Int16x8, type);
+      case SimdOperation::Fn_fromUint16x8Bits:
+        return inlineSimdConvert(callInfo, native, true, SimdType::Uint16x8, type);
+      case SimdOperation::Fn_fromInt8x16Bits:
+        return inlineSimdConvert(callInfo, native, true, SimdType::Int8x16, type);
+      case SimdOperation::Fn_fromUint8x16Bits:
+        return inlineSimdConvert(callInfo, native, true, SimdType::Uint8x16, type);
+      case SimdOperation::Fn_fromFloat32x4Bits:
+        return inlineSimdConvert(callInfo, native, true, SimdType::Float32x4, type);
+      case SimdOperation::Fn_fromFloat64x2Bits:
+        return InliningStatus_NotInlined;
+    }
+
+    MOZ_CRASH("Unexpected SIMD opcode");
+}
+
+// The representation of boolean SIMD vectors is the same as the corresponding
+// integer SIMD vectors with -1 lanes meaning true and 0 lanes meaning false.
+//
+// Functions that set the value of a boolean vector lane work by applying
+// ToBoolean on the input argument, so they accept any argument type, just like
+// the MNot and MTest instructions.
+//
+// Convert any scalar value into an appropriate SIMD lane value: An Int32 value
+// that is either 0 for false or -1 for true.
+MDefinition*
+IonBuilder::convertToBooleanSimdLane(MDefinition* scalar)
+{
+    MSub* result;
+
+    if (scalar->type() == MIRType::Boolean) {
+        // The input scalar is already a boolean with the int32 values 0 / 1.
+        // Compute result = 0 - scalar.
+        result = MSub::New(alloc(), constant(Int32Value(0)), scalar);
+    } else {
+        // For any other type, let MNot handle the conversion to boolean.
+        // Compute result = !scalar - 1.
+        MNot* inv = MNot::New(alloc(), scalar);
+        current->add(inv);
+        result = MSub::New(alloc(), inv, constant(Int32Value(1)));
+    }
+
+    result->setInt32Specialization();
+    current->add(result);
+    return result;
+}
+
+IonBuilder::InliningResult
+IonBuilder::inlineConstructSimdObject(CallInfo& callInfo, SimdTypeDescr* descr)
+{
+    if (!JitSupportsSimd()) {
+        trackOptimizationOutcome(TrackedOutcome::NoSimdJitSupport);
+        return InliningStatus_NotInlined;
+    }
+
+    // Generic constructor of SIMD valuesX4.
+    MIRType simdType;
+    if (!MaybeSimdTypeToMIRType(descr->type(), &simdType)) {
+        trackOptimizationOutcome(TrackedOutcome::SimdTypeNotOptimized);
+        return InliningStatus_NotInlined;
+    }
+
+    // Take the templateObject out of Baseline ICs, such that we can box
+    // SIMD value type in the same kind of objects.
+    MOZ_ASSERT(InlineTypedObject::canAccommodateType(descr));
+    MOZ_ASSERT(descr->getClass() == &SimdTypeDescr::class_,
+               "getTemplateObjectForSimdCtor needs an update");
+
+    JSObject* templateObject = inspector->getTemplateObjectForSimdCtor(pc, descr->type());
+    if (!templateObject)
+        return InliningStatus_NotInlined;
+
+    // The previous assertion ensures this will never fail if we were able to
+    // allocate a templateObject in Baseline.
+    InlineTypedObject* inlineTypedObject = &templateObject->as<InlineTypedObject>();
+    MOZ_ASSERT(&inlineTypedObject->typeDescr() == descr);
+
+    // When there are missing arguments, provide a default value
+    // containing the coercion of 'undefined' to the right type.
+    MConstant* defVal = nullptr;
+    MIRType laneType = SimdTypeToLaneType(simdType);
+    unsigned lanes = SimdTypeToLength(simdType);
+    if (lanes != 4 || callInfo.argc() < lanes) {
+        if (laneType == MIRType::Int32 || laneType == MIRType::Boolean) {
+            // The default lane for a boolean vector is |false|, but
+            // |MSimdSplat|, |MSimdValueX4|, and |MSimdInsertElement| all
+            // require an Int32 argument with the value 0 or 01 to initialize a
+            // boolean lane. See also convertToBooleanSimdLane() which is
+            // idempotent with a 0 argument after constant folding.
+            defVal = constant(Int32Value(0));
+        } else if (laneType == MIRType::Double) {
+            defVal = constant(DoubleNaNValue());
+        } else {
+            MOZ_ASSERT(laneType == MIRType::Float32);
+            defVal = MConstant::NewFloat32(alloc(), JS::GenericNaN());
+            current->add(defVal);
+        }
+    }
+
+    MInstruction *values = nullptr;
+
+    // Use the MSimdValueX4 constructor for X4 vectors.
+    if (lanes == 4) {
+        MDefinition* lane[4];
+        for (unsigned i = 0; i < 4; i++)
+            lane[i] = callInfo.getArgWithDefault(i, defVal);
+
+        // Convert boolean lanes into Int32 0 / -1.
+        if (laneType == MIRType::Boolean) {
+            for (unsigned i = 0; i < 4; i++)
+                lane[i] = convertToBooleanSimdLane(lane[i]);
+        }
+
+        values = MSimdValueX4::New(alloc(), simdType, lane[0], lane[1], lane[2], lane[3]);
+        current->add(values);
+    } else {
+        // For general constructor calls, start from splat(defVal), insert one
+        // lane at a time.
+        values = MSimdSplat::New(alloc(), defVal, simdType);
+        current->add(values);
+
+        // Stop early if constructor doesn't have enough arguments. These lanes
+        // then get the default value.
+        if (callInfo.argc() < lanes)
+            lanes = callInfo.argc();
+
+        for (unsigned i = 0; i < lanes; i++) {
+            MDefinition* lane = callInfo.getArg(i);
+            if (laneType == MIRType::Boolean)
+                lane = convertToBooleanSimdLane(lane);
+            values = MSimdInsertElement::New(alloc(), values, lane, i);
+            current->add(values);
+        }
+    }
+
+    MSimdBox* obj = MSimdBox::New(alloc(), constraints(), values, inlineTypedObject, descr->type(),
+                                  inlineTypedObject->group()->initialHeap(constraints()));
+    current->add(obj);
+    current->push(obj);
+
+    callInfo.setImplicitlyUsedUnchecked();
+    return InliningStatus_Inlined;
+}
+
+bool
+IonBuilder::canInlineSimd(CallInfo& callInfo, JSNative native, unsigned numArgs,
+                          InlineTypedObject** templateObj)
+{
+    if (callInfo.argc() != numArgs)
+        return false;
+
+    JSObject* templateObject = inspector->getTemplateObjectForNative(pc, native);
+    if (!templateObject)
+        return false;
+
+    *templateObj = &templateObject->as<InlineTypedObject>();
+    return true;
+}
+
+IonBuilder::InliningResult
+IonBuilder::inlineSimdCheck(CallInfo& callInfo, JSNative native, SimdType type)
+{
+    InlineTypedObject* templateObj = nullptr;
+    if (!canInlineSimd(callInfo, native, 1, &templateObj))
+        return InliningStatus_NotInlined;
+
+    // Unboxing checks the SIMD object type and throws a TypeError if it doesn't
+    // match type.
+    MDefinition *arg = unboxSimd(callInfo.getArg(0), type);
+
+    // Create an unbox/box pair, expecting the box to be optimized away if
+    // anyone use the return value from this check() call. This is what you want
+    // for code like this:
+    //
+    // function f(x) {
+    //   x = Int32x4.check(x)
+    //   for(...) {
+    //     y = Int32x4.add(x, ...)
+    //   }
+    //
+    // The unboxing of x happens as early as possible, and only once.
+    return boxSimd(callInfo, arg, templateObj);
+}
+
+// Given a value or object, insert a dynamic check that this is a SIMD object of
+// the required SimdType, and unbox it into the corresponding SIMD MIRType.
+//
+// This represents the standard type checking that all the SIMD operations
+// perform on their arguments.
+MDefinition*
+IonBuilder::unboxSimd(MDefinition* ins, SimdType type)
+{
+    // Trivial optimization: If ins is a MSimdBox of the same SIMD type, there
+    // is no way the unboxing could fail, and we can skip it altogether.
+    // This is the same thing MSimdUnbox::foldsTo() does, but we can save the
+    // memory allocation here.
+    if (ins->isSimdBox()) {
+        MSimdBox* box = ins->toSimdBox();
+        if (box->simdType() == type) {
+            MDefinition* value = box->input();
+            MOZ_ASSERT(value->type() == SimdTypeToMIRType(type));
+            return value;
+        }
+    }
+
+    MSimdUnbox* unbox = MSimdUnbox::New(alloc(), ins, type);
+    current->add(unbox);
+    return unbox;
+}
+
+IonBuilder::InliningResult
+IonBuilder::boxSimd(CallInfo& callInfo, MDefinition* ins, InlineTypedObject* templateObj)
+{
+    SimdType simdType = templateObj->typeDescr().as<SimdTypeDescr>().type();
+    MSimdBox* obj = MSimdBox::New(alloc(), constraints(), ins, templateObj, simdType,
+                                  templateObj->group()->initialHeap(constraints()));
+
+    // In some cases, ins has already been added to current.
+    if (!ins->block() && ins->isInstruction())
+        current->add(ins->toInstruction());
+    current->add(obj);
+    current->push(obj);
+
+    callInfo.setImplicitlyUsedUnchecked();
+    return InliningStatus_Inlined;
+}
+
+IonBuilder::InliningResult
+IonBuilder::inlineSimdBinaryArith(CallInfo& callInfo, JSNative native,
+                                  MSimdBinaryArith::Operation op, SimdType type)
+{
+    InlineTypedObject* templateObj = nullptr;
+    if (!canInlineSimd(callInfo, native, 2, &templateObj))
+        return InliningStatus_NotInlined;
+
+    MDefinition* lhs = unboxSimd(callInfo.getArg(0), type);
+    MDefinition* rhs = unboxSimd(callInfo.getArg(1), type);
+
+    auto* ins = MSimdBinaryArith::AddLegalized(alloc(), current, lhs, rhs, op);
+    return boxSimd(callInfo, ins, templateObj);
+}
+
+IonBuilder::InliningResult
+IonBuilder::inlineSimdBinaryBitwise(CallInfo& callInfo, JSNative native,
+                                    MSimdBinaryBitwise::Operation op, SimdType type)
+{
+    InlineTypedObject* templateObj = nullptr;
+    if (!canInlineSimd(callInfo, native, 2, &templateObj))
+        return InliningStatus_NotInlined;
+
+    MDefinition* lhs = unboxSimd(callInfo.getArg(0), type);
+    MDefinition* rhs = unboxSimd(callInfo.getArg(1), type);
+
+    auto* ins = MSimdBinaryBitwise::New(alloc(), lhs, rhs, op);
+    return boxSimd(callInfo, ins, templateObj);
+}
+
+// Inline a binary SIMD operation where both arguments are SIMD types.
+IonBuilder::InliningResult
+IonBuilder::inlineSimdBinarySaturating(CallInfo& callInfo, JSNative native,
+                                       MSimdBinarySaturating::Operation op, SimdType type)
+{
+    InlineTypedObject* templateObj = nullptr;
+    if (!canInlineSimd(callInfo, native, 2, &templateObj))
+        return InliningStatus_NotInlined;
+
+    MDefinition* lhs = unboxSimd(callInfo.getArg(0), type);
+    MDefinition* rhs = unboxSimd(callInfo.getArg(1), type);
+
+    MSimdBinarySaturating* ins =
+      MSimdBinarySaturating::New(alloc(), lhs, rhs, op, GetSimdSign(type));
+    return boxSimd(callInfo, ins, templateObj);
+}
+
+// Inline a SIMD shiftByScalar operation.
+IonBuilder::InliningResult
+IonBuilder::inlineSimdShift(CallInfo& callInfo, JSNative native, MSimdShift::Operation op,
+                            SimdType type)
+{
+    InlineTypedObject* templateObj = nullptr;
+    if (!canInlineSimd(callInfo, native, 2, &templateObj))
+        return InliningStatus_NotInlined;
+
+    MDefinition* vec = unboxSimd(callInfo.getArg(0), type);
+
+    MInstruction* ins = MSimdShift::AddLegalized(alloc(), current, vec, callInfo.getArg(1), op);
+    return boxSimd(callInfo, ins, templateObj);
+}
+
+IonBuilder::InliningResult
+IonBuilder::inlineSimdComp(CallInfo& callInfo, JSNative native, MSimdBinaryComp::Operation op,
+                           SimdType type)
+{
+    InlineTypedObject* templateObj = nullptr;
+    if (!canInlineSimd(callInfo, native, 2, &templateObj))
+        return InliningStatus_NotInlined;
+
+    MDefinition* lhs = unboxSimd(callInfo.getArg(0), type);
+    MDefinition* rhs = unboxSimd(callInfo.getArg(1), type);
+    MInstruction* ins =
+      MSimdBinaryComp::AddLegalized(alloc(), current, lhs, rhs, op, GetSimdSign(type));
+    return boxSimd(callInfo, ins, templateObj);
+}
+
+IonBuilder::InliningResult
+IonBuilder::inlineSimdUnary(CallInfo& callInfo, JSNative native, MSimdUnaryArith::Operation op,
+                            SimdType type)
+{
+    InlineTypedObject* templateObj = nullptr;
+    if (!canInlineSimd(callInfo, native, 1, &templateObj))
+        return InliningStatus_NotInlined;
+
+    MDefinition* arg = unboxSimd(callInfo.getArg(0), type);
+
+    MSimdUnaryArith* ins = MSimdUnaryArith::New(alloc(), arg, op);
+    return boxSimd(callInfo, ins, templateObj);
+}
+
+IonBuilder::InliningResult
+IonBuilder::inlineSimdSplat(CallInfo& callInfo, JSNative native, SimdType type)
+{
+    InlineTypedObject* templateObj = nullptr;
+    if (!canInlineSimd(callInfo, native, 1, &templateObj))
+        return InliningStatus_NotInlined;
+
+    MIRType mirType = SimdTypeToMIRType(type);
+    MDefinition* arg = callInfo.getArg(0);
+
+    // Convert to 0 / -1 before splatting a boolean lane.
+    if (SimdTypeToLaneType(mirType) == MIRType::Boolean)
+        arg = convertToBooleanSimdLane(arg);
+
+    MSimdSplat* ins = MSimdSplat::New(alloc(), arg, mirType);
+    return boxSimd(callInfo, ins, templateObj);
+}
+
+IonBuilder::InliningResult
+IonBuilder::inlineSimdExtractLane(CallInfo& callInfo, JSNative native, SimdType type)
+{
+    // extractLane() returns a scalar, so don't use canInlineSimd() which looks
+    // for a template object.
+    if (callInfo.argc() != 2 || callInfo.constructing()) {
+        trackOptimizationOutcome(TrackedOutcome::CantInlineNativeBadForm);
+        return InliningStatus_NotInlined;
+    }
+
+    // Lane index.
+    MDefinition* arg = callInfo.getArg(1);
+    if (!arg->isConstant() || arg->type() != MIRType::Int32)
+        return InliningStatus_NotInlined;
+    unsigned lane = arg->toConstant()->toInt32();
+    if (lane >= GetSimdLanes(type))
+        return InliningStatus_NotInlined;
+
+    // Original vector.
+    MDefinition* orig = unboxSimd(callInfo.getArg(0), type);
+    MIRType vecType = orig->type();
+    MIRType laneType = SimdTypeToLaneType(vecType);
+    SimdSign sign = GetSimdSign(type);
+
+    // An Uint32 lane can't be represented in MIRType::Int32. Get it as a double.
+    if (type == SimdType::Uint32x4)
+        laneType = MIRType::Double;
+
+    MSimdExtractElement* ins =
+      MSimdExtractElement::New(alloc(), orig, laneType, lane, sign);
+    current->add(ins);
+    current->push(ins);
+    callInfo.setImplicitlyUsedUnchecked();
+    return InliningStatus_Inlined;
+}
+
+IonBuilder::InliningResult
+IonBuilder::inlineSimdReplaceLane(CallInfo& callInfo, JSNative native, SimdType type)
+{
+    InlineTypedObject* templateObj = nullptr;
+    if (!canInlineSimd(callInfo, native, 3, &templateObj))
+        return InliningStatus_NotInlined;
+
+    // Lane index.
+    MDefinition* arg = callInfo.getArg(1);
+    if (!arg->isConstant() || arg->type() != MIRType::Int32)
+        return InliningStatus_NotInlined;
+
+    unsigned lane = arg->toConstant()->toInt32();
+    if (lane >= GetSimdLanes(type))
+        return InliningStatus_NotInlined;
+
+    // Original vector.
+    MDefinition* orig = unboxSimd(callInfo.getArg(0), type);
+    MIRType vecType = orig->type();
+
+    // Convert to 0 / -1 before inserting a boolean lane.
+    MDefinition* value = callInfo.getArg(2);
+    if (SimdTypeToLaneType(vecType) == MIRType::Boolean)
+        value = convertToBooleanSimdLane(value);
+
+    MSimdInsertElement* ins = MSimdInsertElement::New(alloc(), orig, value, lane);
+    return boxSimd(callInfo, ins, templateObj);
+}
+
+// Inline a SIMD conversion or bitcast. When isCast==false, one of the types
+// must be floating point and the other integer. In this case, sign indicates if
+// the integer lanes should be treated as signed or unsigned integers.
+IonBuilder::InliningResult
+IonBuilder::inlineSimdConvert(CallInfo& callInfo, JSNative native, bool isCast, SimdType fromType,
+                              SimdType toType)
+{
+    InlineTypedObject* templateObj = nullptr;
+    if (!canInlineSimd(callInfo, native, 1, &templateObj))
+        return InliningStatus_NotInlined;
+
+    MDefinition* arg = unboxSimd(callInfo.getArg(0), fromType);
+    MIRType mirType = SimdTypeToMIRType(toType);
+
+    MInstruction* ins;
+    if (isCast) {
+        // Signed/Unsigned doesn't matter for bitcasts.
+        ins = MSimdReinterpretCast::New(alloc(), arg, mirType);
+    } else {
+        // Exactly one of fromType, toType must be an integer type.
+        SimdSign sign = GetSimdSign(fromType);
+        if (sign == SimdSign::NotApplicable)
+            sign = GetSimdSign(toType);
+
+        // Possibly expand into multiple instructions.
+        ins = MSimdConvert::AddLegalized(alloc(), current, arg, mirType, sign);
+    }
+
+    return boxSimd(callInfo, ins, templateObj);
+}
+
+IonBuilder::InliningResult
+IonBuilder::inlineSimdSelect(CallInfo& callInfo, JSNative native, SimdType type)
+{
+    InlineTypedObject* templateObj = nullptr;
+    if (!canInlineSimd(callInfo, native, 3, &templateObj))
+        return InliningStatus_NotInlined;
+
+    MDefinition* mask = unboxSimd(callInfo.getArg(0), GetBooleanSimdType(type));
+    MDefinition* tval = unboxSimd(callInfo.getArg(1), type);
+    MDefinition* fval = unboxSimd(callInfo.getArg(2), type);
+
+    MSimdSelect* ins = MSimdSelect::New(alloc(), mask, tval, fval);
+    return boxSimd(callInfo, ins, templateObj);
+}
+
+IonBuilder::InliningResult
+IonBuilder::inlineSimdShuffle(CallInfo& callInfo, JSNative native, SimdType type,
+                              unsigned numVectors)
+{
+    unsigned numLanes = GetSimdLanes(type);
+    InlineTypedObject* templateObj = nullptr;
+    if (!canInlineSimd(callInfo, native, numVectors + numLanes, &templateObj))
+        return InliningStatus_NotInlined;
+
+    MIRType mirType = SimdTypeToMIRType(type);
+
+    MSimdGeneralShuffle* ins = MSimdGeneralShuffle::New(alloc(), numVectors, numLanes, mirType);
+
+    if (!ins->init(alloc()))
+        return abort(AbortReason::Alloc);
+
+    for (unsigned i = 0; i < numVectors; i++)
+        ins->setVector(i, unboxSimd(callInfo.getArg(i), type));
+    for (size_t i = 0; i < numLanes; i++)
+        ins->setLane(i, callInfo.getArg(numVectors + i));
+
+    return boxSimd(callInfo, ins, templateObj);
+}
+
+IonBuilder::InliningResult
+IonBuilder::inlineSimdAnyAllTrue(CallInfo& callInfo, bool IsAllTrue, JSNative native,
+                                 SimdType type)
+{
+    // anyTrue() / allTrue() return a scalar, so don't use canInlineSimd() which looks
+    // for a template object.
+    if (callInfo.argc() != 1 || callInfo.constructing()) {
+        trackOptimizationOutcome(TrackedOutcome::CantInlineNativeBadForm);
+        return InliningStatus_NotInlined;
+    }
+
+    MDefinition* arg = unboxSimd(callInfo.getArg(0), type);
+
+    MUnaryInstruction* ins;
+    if (IsAllTrue)
+        ins = MSimdAllTrue::New(alloc(), arg, MIRType::Boolean);
+    else
+        ins = MSimdAnyTrue::New(alloc(), arg, MIRType::Boolean);
+
+    current->add(ins);
+    current->push(ins);
+    callInfo.setImplicitlyUsedUnchecked();
+    return InliningStatus_Inlined;
+}
+
+// Get the typed array element type corresponding to the lanes in a SIMD vector type.
+// This only applies to SIMD types that can be loaded and stored to a typed array.
+static Scalar::Type
+SimdTypeToArrayElementType(SimdType type)
+{
+    switch (type) {
+      case SimdType::Float32x4: return Scalar::Float32x4;
+      case SimdType::Int8x16:
+      case SimdType::Uint8x16:  return Scalar::Int8x16;
+      case SimdType::Int16x8:
+      case SimdType::Uint16x8:  return Scalar::Int16x8;
+      case SimdType::Int32x4:
+      case SimdType::Uint32x4:  return Scalar::Int32x4;
+      default:                MOZ_CRASH("unexpected simd type");
+    }
+}
+
+bool
+IonBuilder::prepareForSimdLoadStore(CallInfo& callInfo, Scalar::Type simdType,
+                                    MInstruction** elements, MDefinition** index,
+                                    Scalar::Type* arrayType)
+{
+    MDefinition* array = callInfo.getArg(0);
+    *index = callInfo.getArg(1);
+
+    if (!ElementAccessIsTypedArray(constraints(), array, *index, arrayType))
+        return false;
+
+    MInstruction* indexAsInt32 = MToNumberInt32::New(alloc(), *index);
+    current->add(indexAsInt32);
+    *index = indexAsInt32;
+
+    MDefinition* indexLoadEnd = *index;
+
+    MOZ_ASSERT(Scalar::byteSize(simdType) % Scalar::byteSize(*arrayType) == 0);
+    int32_t byteLoadSize = Scalar::byteSize(simdType) / Scalar::byteSize(*arrayType);
+    if (byteLoadSize > 1) {
+        // Add the number of supplementary needed slots. Overflows are fine
+        // because the bounds check code uses an unsigned comparison.
+        MAdd* addedIndex = MAdd::New(alloc(), *index, constant(Int32Value(byteLoadSize - 1)));
+        addedIndex->setInt32Specialization();
+        current->add(addedIndex);
+        indexLoadEnd = addedIndex;
+    }
+
+    MInstruction* length;
+    addTypedArrayLengthAndData(array, SkipBoundsCheck, index, &length, elements);
+
+    // If the index+size addition overflows, then indexLoadEnd might be
+    // in bounds while the actual index isn't, so we need two bounds checks
+    // here.
+    if (byteLoadSize > 1) {
+        indexLoadEnd = addBoundsCheck(indexLoadEnd, length);
+        auto* sub = MSub::New(alloc(), indexLoadEnd, constant(Int32Value(byteLoadSize - 1)));
+        sub->setInt32Specialization();
+        current->add(sub);
+        *index = sub;
+    }
+
+    *index = addBoundsCheck(*index, length);
+
+    return true;
+}
+
+IonBuilder::InliningResult
+IonBuilder::inlineSimdLoad(CallInfo& callInfo, JSNative native, SimdType type, unsigned numElems)
+{
+    InlineTypedObject* templateObj = nullptr;
+    if (!canInlineSimd(callInfo, native, 2, &templateObj))
+        return InliningStatus_NotInlined;
+
+    Scalar::Type elemType = SimdTypeToArrayElementType(type);
+
+    MDefinition* index = nullptr;
+    MInstruction* elements = nullptr;
+    Scalar::Type arrayType;
+    if (!prepareForSimdLoadStore(callInfo, elemType, &elements, &index, &arrayType))
+        return InliningStatus_NotInlined;
+
+    MLoadUnboxedScalar* load = MLoadUnboxedScalar::New(alloc(), elements, index, arrayType);
+    load->setResultType(SimdTypeToMIRType(type));
+    load->setSimdRead(elemType, numElems);
+
+    return boxSimd(callInfo, load, templateObj);
+}
+
+IonBuilder::InliningResult
+IonBuilder::inlineSimdStore(CallInfo& callInfo, JSNative native, SimdType type, unsigned numElems)
+{
+    InlineTypedObject* templateObj = nullptr;
+    if (!canInlineSimd(callInfo, native, 3, &templateObj))
+        return InliningStatus_NotInlined;
+
+    Scalar::Type elemType = SimdTypeToArrayElementType(type);
+
+    MDefinition* index = nullptr;
+    MInstruction* elements = nullptr;
+    Scalar::Type arrayType;
+    if (!prepareForSimdLoadStore(callInfo, elemType, &elements, &index, &arrayType))
+        return InliningStatus_NotInlined;
+
+    MDefinition* valueToWrite = unboxSimd(callInfo.getArg(2), type);
+    MStoreUnboxedScalar* store = MStoreUnboxedScalar::New(alloc(), elements, index,
+                                                          valueToWrite, arrayType,
+                                                          MStoreUnboxedScalar::TruncateInput);
+    store->setSimdWrite(elemType, numElems);
+
+    current->add(store);
+    // Produce the original boxed value as our return value.
+    // This is unlikely to be used, so don't bother reboxing valueToWrite.
+    current->push(callInfo.getArg(2));
+
+    callInfo.setImplicitlyUsedUnchecked();
+
+    MOZ_TRY(resumeAfter(store));
+    return InliningStatus_Inlined;
+}
+
+// Note that SIMD.cpp provides its own JSJitInfo objects for SIMD.foo.* functions.
+// The Simd* objects defined here represent SIMD.foo() constructor calls.
+// They are encoded with .nativeOp = 0. That is the sub-opcode within the SIMD type.
+static_assert(uint16_t(SimdOperation::Constructor) == 0, "Constructor opcode must be 0");
+
 #define ADD_NATIVE(native) const JSJitInfo JitInfo_##native { \
     { nullptr }, { uint16_t(InlinableNative::native) }, { 0 }, JSJitInfo::InlinableNative };
     INLINABLE_NATIVE_LIST(ADD_NATIVE)

js/src/jit/MIR.cpp

@@ -17,6 +17,7 @@
 #include "jslibmath.h"
 
 #include "builtin/RegExp.h"
+#include "builtin/SIMD.h"
 #include "builtin/String.h"
 #include "jit/AtomicOperations.h"
 #include "jit/BaselineInspector.h"
@@ -1311,7 +1312,531 @@ MWasmFloatConstant::congruentTo(const MDefinition* ins) const
            u.bits_ == ins->toWasmFloatConstant()->u.bits_;
 }
 
+MDefinition*
+MSimdValueX4::foldsTo(TempAllocator& alloc)
+{
+#ifdef DEBUG
+    MIRType laneType = SimdTypeToLaneArgumentType(type());
+#endif
+    bool allConstants = true;
+    bool allSame = true;
+
+    for (size_t i = 0; i < 4; ++i) {
+        MDefinition* op = getOperand(i);
+        MOZ_ASSERT(op->type() == laneType);
+        if (!op->isConstant())
+            allConstants = false;
+        if (i > 0 && op != getOperand(i - 1))
+            allSame = false;
+    }
+
+    if (!allConstants && !allSame)
+        return this;
+
+    if (allConstants) {
+        SimdConstant cst;
+        switch (type()) {
+          case MIRType::Bool32x4: {
+            int32_t a[4];
+            for (size_t i = 0; i < 4; ++i)
+                a[i] = getOperand(i)->toConstant()->valueToBooleanInfallible() ? -1 : 0;
+            cst = SimdConstant::CreateX4(a);
+            break;
+          }
+          case MIRType::Int32x4: {
+            int32_t a[4];
+            for (size_t i = 0; i < 4; ++i)
+                a[i] = getOperand(i)->toConstant()->toInt32();
+            cst = SimdConstant::CreateX4(a);
+            break;
+          }
+          case MIRType::Float32x4: {
+            float a[4];
+            for (size_t i = 0; i < 4; ++i)
+                a[i] = getOperand(i)->toConstant()->numberToDouble();
+            cst = SimdConstant::CreateX4(a);
+            break;
+          }
+          default: MOZ_CRASH("unexpected type in MSimdValueX4::foldsTo");
+        }
+
+        return MSimdConstant::New(alloc, cst, type());
+    }
+
+    MOZ_ASSERT(allSame);
+    return MSimdSplat::New(alloc, getOperand(0), type());
+}
+
+MDefinition*
+MSimdSplat::foldsTo(TempAllocator& alloc)
+{
+#ifdef DEBUG
+    MIRType laneType = SimdTypeToLaneArgumentType(type());
+#endif
+    MDefinition* op = getOperand(0);
+    if (!op->isConstant())
+        return this;
+    MOZ_ASSERT(op->type() == laneType);
+
+    SimdConstant cst;
+    switch (type()) {
+      case MIRType::Bool8x16: {
+        int8_t v = op->toConstant()->valueToBooleanInfallible() ? -1 : 0;
+        cst = SimdConstant::SplatX16(v);
+        break;
+      }
+      case MIRType::Bool16x8: {
+        int16_t v = op->toConstant()->valueToBooleanInfallible() ? -1 : 0;
+        cst = SimdConstant::SplatX8(v);
+        break;
+      }
+      case MIRType::Bool32x4: {
+        int32_t v = op->toConstant()->valueToBooleanInfallible() ? -1 : 0;
+        cst = SimdConstant::SplatX4(v);
+        break;
+      }
+      case MIRType::Int8x16: {
+        int32_t v = op->toConstant()->toInt32();
+        cst = SimdConstant::SplatX16(v);
+        break;
+      }
+      case MIRType::Int16x8: {
+        int32_t v = op->toConstant()->toInt32();
+        cst = SimdConstant::SplatX8(v);
+        break;
+      }
+      case MIRType::Int32x4: {
+        int32_t v = op->toConstant()->toInt32();
+        cst = SimdConstant::SplatX4(v);
+        break;
+      }
+      case MIRType::Float32x4: {
+        float v = op->toConstant()->numberToDouble();
+        cst = SimdConstant::SplatX4(v);
+        break;
+      }
+      default: MOZ_CRASH("unexpected type in MSimdSplat::foldsTo");
+    }
+
+    return MSimdConstant::New(alloc, cst, type());
+}
+
+MDefinition*
+MSimdUnbox::foldsTo(TempAllocator& alloc)
+{
+    MDefinition* in = input();
+
+    if (in->isSimdBox()) {
+        MSimdBox* box = in->toSimdBox();
+        // If the operand is a MSimdBox, then we just reuse the operand of the
+        // MSimdBox as long as the type corresponds to what we are supposed to
+        // unbox.
+        in = box->input();
+        if (box->simdType() != simdType())
+            return this;
+        MOZ_ASSERT(in->type() == type());
+        return in;
+    }
+
+    return this;
+}
+
+MDefinition*
+MSimdSwizzle::foldsTo(TempAllocator& alloc)
+{
+    if (lanesMatch(0, 1, 2, 3))
+        return input();
+    return this;
+}
+
+MDefinition*
+MSimdGeneralShuffle::foldsTo(TempAllocator& alloc)
+{
+    FixedList<uint8_t> lanes;
+    if (!lanes.init(alloc, numLanes()))
+        return this;
+
+    for (size_t i = 0; i < numLanes(); i++) {
+        if (!lane(i)->isConstant() || lane(i)->type() != MIRType::Int32)
+            return this;
+        int32_t temp = lane(i)->toConstant()->toInt32();
+        if (temp < 0 || unsigned(temp) >= numLanes() * numVectors())
+            return this;
+        lanes[i] = uint8_t(temp);
+    }
+
+    if (numVectors() == 1)
+        return MSimdSwizzle::New(alloc, vector(0), lanes.data());
+
+    MOZ_ASSERT(numVectors() == 2);
+    return MSimdShuffle::New(alloc, vector(0), vector(1), lanes.data());
+}
+
+MInstruction*
+MSimdConvert::AddLegalized(TempAllocator& alloc, MBasicBlock* addTo, MDefinition* obj,
+                           MIRType toType, SimdSign sign, wasm::BytecodeOffset bytecodeOffset)
+{
+    MIRType fromType = obj->type();
+
+    if (SupportsUint32x4FloatConversions || sign != SimdSign::Unsigned) {
+        MInstruction* ins = New(alloc, obj, toType, sign, bytecodeOffset);
+        addTo->add(ins);
+        return ins;
+    }
+
+    // This architecture can't do Uint32x4 <-> Float32x4 conversions (Hi SSE!)
+    MOZ_ASSERT(sign == SimdSign::Unsigned);
+    if (fromType == MIRType::Int32x4 && toType == MIRType::Float32x4) {
+        // Converting Uint32x4 -> Float32x4. This algorithm is from LLVM.
+        //
+        // Split the input number into high and low parts:
+        //
+        // uint32_t hi = x >> 16;
+        // uint32_t lo = x & 0xffff;
+        //
+        // Insert these parts as the low mantissa bits in a float32 number with
+        // the corresponding exponent:
+        //
+        // float fhi = (bits-as-float)(hi | 0x53000000); // 0x1.0p39f + hi*2^16
+        // float flo = (bits-as-float)(lo | 0x4b000000); // 0x1.0p23f + lo
+        //
+        // Subtract the bias from the hi part:
+        //
+        // fhi -= (0x1.0p39 + 0x1.0p23) // hi*2^16 - 0x1.0p23
+        //
+        // And finally combine:
+        //
+        // result = flo + fhi // lo + hi*2^16.
+
+        // Compute hi = obj >> 16 (lane-wise unsigned shift).
+        MInstruction* c16 = MConstant::New(alloc, Int32Value(16));
+        addTo->add(c16);
+        MInstruction* hi = MSimdShift::AddLegalized(alloc, addTo, obj, c16, MSimdShift::ursh);
+
+        // Compute lo = obj & 0xffff (lane-wise).
+        MInstruction* m16 =
+          MSimdConstant::New(alloc, SimdConstant::SplatX4(0xffff), MIRType::Int32x4);
+        addTo->add(m16);
+        MInstruction* lo = MSimdBinaryBitwise::New(alloc, obj, m16, MSimdBinaryBitwise::and_);
+        addTo->add(lo);
+
+        // Mix in the exponents.
+        MInstruction* exphi =
+          MSimdConstant::New(alloc, SimdConstant::SplatX4(0x53000000), MIRType::Int32x4);
+        addTo->add(exphi);
+        MInstruction* mhi = MSimdBinaryBitwise::New(alloc, hi, exphi, MSimdBinaryBitwise::or_);
+        addTo->add(mhi);
+        MInstruction* explo =
+          MSimdConstant::New(alloc, SimdConstant::SplatX4(0x4b000000), MIRType::Int32x4);
+        addTo->add(explo);
+        MInstruction* mlo = MSimdBinaryBitwise::New(alloc, lo, explo, MSimdBinaryBitwise::or_);
+        addTo->add(mlo);
+
+        // Bit-cast both to Float32x4.
+        MInstruction* fhi = MSimdReinterpretCast::New(alloc, mhi, MIRType::Float32x4);
+        addTo->add(fhi);
+        MInstruction* flo = MSimdReinterpretCast::New(alloc, mlo, MIRType::Float32x4);
+        addTo->add(flo);
+
+        // Subtract out the bias: 0x1.0p39f + 0x1.0p23f.
+        // MSVC doesn't support the hexadecimal float syntax.
+        const float BiasValue = 549755813888.f + 8388608.f;
+        MInstruction* bias =
+          MSimdConstant::New(alloc, SimdConstant::SplatX4(BiasValue), MIRType::Float32x4);
+        addTo->add(bias);
+        MInstruction* fhi_debiased =
+          MSimdBinaryArith::AddLegalized(alloc, addTo, fhi, bias, MSimdBinaryArith::Op_sub);
+
+        // Compute the final result.
+        return MSimdBinaryArith::AddLegalized(alloc, addTo, fhi_debiased, flo,
+                                              MSimdBinaryArith::Op_add);
+    }
+
+    if (fromType == MIRType::Float32x4 && toType == MIRType::Int32x4) {
+        // The Float32x4 -> Uint32x4 conversion can throw if the input is out of
+        // range. This is handled by the LFloat32x4ToUint32x4 expansion.
+        MInstruction* ins = New(alloc, obj, toType, sign, bytecodeOffset);
+        addTo->add(ins);
+        return ins;
+    }
+
+    MOZ_CRASH("Unhandled SIMD type conversion");
+}
+
+MInstruction*
+MSimdBinaryComp::AddLegalized(TempAllocator& alloc, MBasicBlock* addTo, MDefinition* left,
+                              MDefinition* right, Operation op, SimdSign sign)
+{
+    MOZ_ASSERT(left->type() == right->type());
+    MIRType opType = left->type();
+    MOZ_ASSERT(IsSimdType(opType));
+    bool IsEquality = op == equal || op == notEqual;
+
+    // Check if this is an unsupported unsigned compare that needs to be biased.
+    // If so, put the bias vector in `bias`.
+    if (sign == SimdSign::Unsigned && !IsEquality) {
+        MInstruction* bias = nullptr;
+
+        // This is an order comparison of Uint32x4 vectors which are not supported on this target.
+        // Simply offset |left| and |right| by INT_MIN, then do a signed comparison.
+        if (!SupportsUint32x4Compares && opType == MIRType::Int32x4)
+            bias = MSimdConstant::New(alloc, SimdConstant::SplatX4(int32_t(0x80000000)), opType);
+        else if (!SupportsUint16x8Compares && opType == MIRType::Int16x8)
+            bias = MSimdConstant::New(alloc, SimdConstant::SplatX8(int16_t(0x8000)), opType);
+        if (!SupportsUint8x16Compares && opType == MIRType::Int8x16)
+            bias = MSimdConstant::New(alloc, SimdConstant::SplatX16(int8_t(0x80)), opType);
+
+        if (bias) {
+            addTo->add(bias);
+
+            // Add the bias.
+            MInstruction* bleft =
+              MSimdBinaryArith::AddLegalized(alloc, addTo, left, bias, MSimdBinaryArith::Op_add);
+            MInstruction* bright =
+              MSimdBinaryArith::AddLegalized(alloc, addTo, right, bias, MSimdBinaryArith::Op_add);
+
+            // Do the equivalent signed comparison.
+            MInstruction* result =
+              MSimdBinaryComp::New(alloc, bleft, bright, op, SimdSign::Signed);
+            addTo->add(result);
+
+            return result;
+        }
+    }
+
+    if (sign == SimdSign::Unsigned &&
+        ((!SupportsUint32x4Compares && opType == MIRType::Int32x4) ||
+         (!SupportsUint16x8Compares && opType == MIRType::Int16x8) ||
+         (!SupportsUint8x16Compares && opType == MIRType::Int8x16))) {
+        // The sign doesn't matter for equality tests. Flip it to make the
+        // backend assertions happy.
+        MOZ_ASSERT(IsEquality);
+        sign = SimdSign::Signed;
+    }
+
+    // This is a legal operation already. Just create the instruction requested.
+    MInstruction* result = MSimdBinaryComp::New(alloc, left, right, op, sign);
+    addTo->add(result);
+    return result;
+}
+
+MInstruction*
+MSimdBinaryArith::AddLegalized(TempAllocator& alloc, MBasicBlock* addTo, MDefinition* left,
+                               MDefinition* right, Operation op)
+{
+    MOZ_ASSERT(left->type() == right->type());
+    MIRType opType = left->type();
+    MOZ_ASSERT(IsSimdType(opType));
+
+    // SSE does not have 8x16 multiply instructions.
+    if (opType == MIRType::Int8x16 && op == Op_mul) {
+        // Express the multiply in terms of Int16x8 multiplies by handling the
+        // even and odd lanes separately.
+
+        MInstruction* wideL = MSimdReinterpretCast::New(alloc, left, MIRType::Int16x8);
+        addTo->add(wideL);
+        MInstruction* wideR = MSimdReinterpretCast::New(alloc, right, MIRType::Int16x8);
+        addTo->add(wideR);
+
+        // wideL = yyxx yyxx yyxx yyxx yyxx yyxx yyxx yyxx
+        // wideR = bbaa bbaa bbaa bbaa bbaa bbaa bbaa bbaa
+
+        // Shift the odd lanes down to the low bits of the 16x8 vectors.
+        MInstruction* eight = MConstant::New(alloc, Int32Value(8));
+        addTo->add(eight);
+        MInstruction* evenL = wideL;
+        MInstruction* evenR = wideR;
+        MInstruction* oddL =
+          MSimdShift::AddLegalized(alloc, addTo, wideL, eight, MSimdShift::ursh);
+        MInstruction* oddR =
+          MSimdShift::AddLegalized(alloc, addTo, wideR, eight, MSimdShift::ursh);
+
+        // evenL = yyxx yyxx yyxx yyxx yyxx yyxx yyxx yyxx
+        // evenR = bbaa bbaa bbaa bbaa bbaa bbaa bbaa bbaa
+        // oddL  = 00yy 00yy 00yy 00yy 00yy 00yy 00yy 00yy
+        // oddR  = 00bb 00bb 00bb 00bb 00bb 00bb 00bb 00bb
+
+        // Now do two 16x8 multiplications. We can use the low bits of each.
+        MInstruction* even = MSimdBinaryArith::AddLegalized(alloc, addTo, evenL, evenR, Op_mul);
+        MInstruction* odd = MSimdBinaryArith::AddLegalized(alloc, addTo, oddL, oddR, Op_mul);
+
+        // even = ~~PP ~~PP ~~PP ~~PP ~~PP ~~PP ~~PP ~~PP
+        // odd  = ~~QQ ~~QQ ~~QQ ~~QQ ~~QQ ~~QQ ~~QQ ~~QQ
+
+        MInstruction* mask =
+          MSimdConstant::New(alloc, SimdConstant::SplatX8(int16_t(0x00ff)), MIRType::Int16x8);
+        addTo->add(mask);
+        even = MSimdBinaryBitwise::New(alloc, even, mask, MSimdBinaryBitwise::and_);
+        addTo->add(even);
+        odd = MSimdShift::AddLegalized(alloc, addTo, odd, eight, MSimdShift::lsh);
+
+        // even = 00PP 00PP 00PP 00PP 00PP 00PP 00PP 00PP
+        // odd  = QQ00 QQ00 QQ00 QQ00 QQ00 QQ00 QQ00 QQ00
+
+        // Combine:
+        MInstruction* result = MSimdBinaryBitwise::New(alloc, even, odd, MSimdBinaryBitwise::or_);
+        addTo->add(result);
+        result = MSimdReinterpretCast::New(alloc, result, opType);
+        addTo->add(result);
+        return result;
+    }
+
+    // This is a legal operation already. Just create the instruction requested.
+    MInstruction* result = MSimdBinaryArith::New(alloc, left, right, op);
+    addTo->add(result);
+    return result;
+}
+
+MInstruction*
+MSimdShift::AddLegalized(TempAllocator& alloc, MBasicBlock* addTo, MDefinition* left,
+                         MDefinition* right, Operation op)
+{
+    MIRType opType = left->type();
+    MOZ_ASSERT(IsIntegerSimdType(opType));
+
+    // SSE does not provide 8x16 shift instructions.
+    if (opType == MIRType::Int8x16) {
+        // Express the shift in terms of Int16x8 shifts by splitting into even
+        // and odd lanes, place 8-bit lanes into the high bits of Int16x8
+        // vectors `even` and `odd`. Shift, mask, combine.
+        //
+        //   wide = Int16x8.fromInt8x16Bits(left);
+        //   shiftBy = right & 7
+        //   mask = Int16x8.splat(0xff00);
+        //
+        MInstruction* wide = MSimdReinterpretCast::New(alloc, left, MIRType::Int16x8);
+        addTo->add(wide);
+
+        // wide = yyxx yyxx yyxx yyxx yyxx yyxx yyxx yyxx
+
+        MInstruction* shiftMask = MConstant::New(alloc, Int32Value(7));
+        addTo->add(shiftMask);
+        MBinaryBitwiseInstruction* shiftBy = MBitAnd::New(alloc, right, shiftMask);
+        shiftBy->setInt32Specialization();
+        addTo->add(shiftBy);
+
+        // Move the even 8x16 lanes into the high bits of the 16x8 lanes.
+        MInstruction* eight = MConstant::New(alloc, Int32Value(8));
+        addTo->add(eight);
+        MInstruction* even = MSimdShift::AddLegalized(alloc, addTo, wide, eight, lsh);
+
+        // Leave the odd lanes in place.
+        MInstruction* odd = wide;
+
+        // even = xx00 xx00 xx00 xx00 xx00 xx00 xx00 xx00
+        // odd  = yyxx yyxx yyxx yyxx yyxx yyxx yyxx yyxx
+
+        MInstruction* mask =
+          MSimdConstant::New(alloc, SimdConstant::SplatX8(int16_t(0xff00)), MIRType::Int16x8);
+        addTo->add(mask);
+
+        // Left-shift: Clear the low bits in `odd` before shifting.
+        if (op == lsh) {
+            odd = MSimdBinaryBitwise::New(alloc, odd, mask, MSimdBinaryBitwise::and_);
+            addTo->add(odd);
+            // odd  = yy00 yy00 yy00 yy00 yy00 yy00 yy00 yy00
+        }
+
+        // Do the real shift twice: once for the even lanes, once for the odd
+        // lanes. This is a recursive call, but with a different type.
+        even = MSimdShift::AddLegalized(alloc, addTo, even, shiftBy, op);
+        odd = MSimdShift::AddLegalized(alloc, addTo, odd, shiftBy, op);
+
+        // even = XX~~ XX~~ XX~~ XX~~ XX~~ XX~~ XX~~ XX~~
+        // odd  = YY~~ YY~~ YY~~ YY~~ YY~~ YY~~ YY~~ YY~~
+
+        // Right-shift: Clear the low bits in `odd` after shifting.
+        if (op != lsh) {
+            odd = MSimdBinaryBitwise::New(alloc, odd, mask, MSimdBinaryBitwise::and_);
+            addTo->add(odd);
+            // odd  = YY00 YY00 YY00 YY00 YY00 YY00 YY00 YY00
+        }
+
+        // Move the even lanes back to their original place.
+        even = MSimdShift::AddLegalized(alloc, addTo, even, eight, ursh);
+
+        // Now, `odd` contains the odd lanes properly shifted, and `even`
+        // contains the even lanes properly shifted:
+        //
+        // even = 00XX 00XX 00XX 00XX 00XX 00XX 00XX 00XX
+        // odd  = YY00 YY00 YY00 YY00 YY00 YY00 YY00 YY00
+        //
+        // Combine:
+        MInstruction* result = MSimdBinaryBitwise::New(alloc, even, odd, MSimdBinaryBitwise::or_);
+        addTo->add(result);
+        result = MSimdReinterpretCast::New(alloc, result, opType);
+        addTo->add(result);
+        return result;
+    }
+
+    // This is a legal operation already. Just create the instruction requested.
+    MInstruction* result = MSimdShift::New(alloc, left, right, op);
+    addTo->add(result);
+    return result;
+}
+
+template <typename T>
+static void
+PrintOpcodeOperation(T* mir, GenericPrinter& out)
+{
+    mir->MDefinition::printOpcode(out);
+    out.printf(" (%s)", T::OperationName(mir->operation()));
+}
+
 #ifdef JS_JITSPEW
+void
+MSimdBinaryArith::printOpcode(GenericPrinter& out) const
+{
+    PrintOpcodeOperation(this, out);
+}
+void
+MSimdBinarySaturating::printOpcode(GenericPrinter& out) const
+{
+    PrintOpcodeOperation(this, out);
+}
+void
+MSimdBinaryBitwise::printOpcode(GenericPrinter& out) const
+{
+    PrintOpcodeOperation(this, out);
+}
+void
+MSimdUnaryArith::printOpcode(GenericPrinter& out) const
+{
+    PrintOpcodeOperation(this, out);
+}
+void
+MSimdBinaryComp::printOpcode(GenericPrinter& out) const
+{
+    PrintOpcodeOperation(this, out);
+}
+void
+MSimdShift::printOpcode(GenericPrinter& out) const
+{
+    PrintOpcodeOperation(this, out);
+}
+
+void
+MSimdInsertElement::printOpcode(GenericPrinter& out) const
+{
+    MDefinition::printOpcode(out);
+    out.printf(" (lane %u)", lane());
+}
+
+void
+MSimdBox::printOpcode(GenericPrinter& out) const
+{
+    MDefinition::printOpcode(out);
+    out.printf(" (%s%s)", SimdTypeToString(simdType()),
+               initialHeap() == gc::TenuredHeap ? ", tenured" : "");
+}
+
+void
+MSimdUnbox::printOpcode(GenericPrinter& out) const
+{
+    MDefinition::printOpcode(out);
+    out.printf(" (%s)", SimdTypeToString(simdType()));
+}
+
 void
 MControlInstruction::printOpcode(GenericPrinter& out) const
 {

js/src/jit/MIRGenerator.h

@@ -158,6 +158,10 @@ class MIRGenerator
         return needsOverrecursedCheck_;
     }
 
+    // Traverses the graph to find if there's any SIMD instruction. Costful but
+    // the value is cached, so don't worry about calling it several times.
+    bool usesSimd();
+
     bool modifiesFrameArguments() const {
         return modifiesFrameArguments_;
     }
@@ -187,6 +191,8 @@ class MIRGenerator
     uint32_t wasmMaxStackArgBytes_;
     bool needsOverrecursedCheck_;
     bool needsStaticStackAlignment_;
+    bool usesSimd_;
+    bool cachedUsesSimd_;
 
     // Keep track of whether frame arguments are modified during execution.
     // RegAlloc needs to know this as spilling values back to their register

js/src/jit/MIRGraph.cpp

@@ -32,6 +32,8 @@ MIRGenerator::MIRGenerator(CompileRealm* realm, const JitCompileOptions& options
     wasmMaxStackArgBytes_(0),
     needsOverrecursedCheck_(false),
     needsStaticStackAlignment_(false),
+    usesSimd_(false),
+    cachedUsesSimd_(false),
     modifiesFrameArguments_(false),
     instrumentedProfiling_(false),
     instrumentedProfilingIsCached_(false),
@@ -42,6 +44,37 @@ MIRGenerator::MIRGenerator(CompileRealm* realm, const JitCompileOptions& options
     gs_(alloc)
 { }
 
+bool
+MIRGenerator::usesSimd()
+{
+    if (cachedUsesSimd_)
+        return usesSimd_;
+
+    cachedUsesSimd_ = true;
+    for (ReversePostorderIterator block = graph_->rpoBegin(),
+                                  end   = graph_->rpoEnd();
+         block != end;
+         block++)
+    {
+        // It's fine to use MInstructionIterator here because we don't have to
+        // worry about Phis, since any reachable phi (or phi cycle) will have at
+        // least one instruction as an input.
+        for (MInstructionIterator inst = block->begin(); inst != block->end(); inst++) {
+            // Instructions that have SIMD inputs but not a SIMD type are fine
+            // to ignore, as their inputs are also reached at some point. By
+            // induction, at least one instruction with a SIMD type is reached
+            // at some point.
+            if (IsSimdType(inst->type())) {
+                MOZ_ASSERT(SupportsSimd);
+                usesSimd_ = true;
+                return true;
+            }
+        }
+    }
+    usesSimd_ = false;
+    return false;
+}
+
 mozilla::GenericErrorResult<AbortReason>
 MIRGenerator::abort(AbortReason r)
 {

js/src/jit/MacroAssembler.cpp

@@ -340,7 +340,8 @@ template void MacroAssembler::guardTypeSet(const TypedOrValueRegister& value, co
 
 template<typename S, typename T>
 static void
-StoreToTypedFloatArray(MacroAssembler& masm, int arrayType, const S& value, const T& dest)
+StoreToTypedFloatArray(MacroAssembler& masm, int arrayType, const S& value, const T& dest,
+                       unsigned numElems)
 {
     switch (arrayType) {
       case Scalar::Float32:
@@ -349,6 +350,48 @@ StoreToTypedFloatArray(MacroAssembler& masm, int arrayType, const S& value, cons
       case Scalar::Float64:
         masm.storeDouble(value, dest);
         break;
+      case Scalar::Float32x4:
+        switch (numElems) {
+          case 1:
+            masm.storeFloat32(value, dest);
+            break;
+          case 2:
+            masm.storeDouble(value, dest);
+            break;
+          case 3:
+            masm.storeFloat32x3(value, dest);
+            break;
+          case 4:
+            masm.storeUnalignedSimd128Float(value, dest);
+            break;
+          default: MOZ_CRASH("unexpected number of elements in simd write");
+        }
+        break;
+      case Scalar::Int32x4:
+        switch (numElems) {
+          case 1:
+            masm.storeInt32x1(value, dest);
+            break;
+          case 2:
+            masm.storeInt32x2(value, dest);
+            break;
+          case 3:
+            masm.storeInt32x3(value, dest);
+            break;
+          case 4:
+            masm.storeUnalignedSimd128Int(value, dest);
+            break;
+          default: MOZ_CRASH("unexpected number of elements in simd write");
+        }
+        break;
+      case Scalar::Int8x16:
+        MOZ_ASSERT(numElems == 16, "unexpected partial store");
+        masm.storeUnalignedSimd128Int(value, dest);
+        break;
+      case Scalar::Int16x8:
+        MOZ_ASSERT(numElems == 8, "unexpected partial store");
+        masm.storeUnalignedSimd128Int(value, dest);
+        break;
       default:
         MOZ_CRASH("Invalid typed array type");
     }
@@ -356,21 +399,21 @@ StoreToTypedFloatArray(MacroAssembler& masm, int arrayType, const S& value, cons
 
 void
 MacroAssembler::storeToTypedFloatArray(Scalar::Type arrayType, FloatRegister value,
-                                       const BaseIndex& dest)
+                                       const BaseIndex& dest, unsigned numElems)
 {
-    StoreToTypedFloatArray(*this, arrayType, value, dest);
+    StoreToTypedFloatArray(*this, arrayType, value, dest, numElems);
 }
 void
 MacroAssembler::storeToTypedFloatArray(Scalar::Type arrayType, FloatRegister value,
-                                       const Address& dest)
+                                       const Address& dest, unsigned numElems)
 {
-    StoreToTypedFloatArray(*this, arrayType, value, dest);
+    StoreToTypedFloatArray(*this, arrayType, value, dest, numElems);
 }
 
 template<typename T>
 void
 MacroAssembler::loadFromTypedArray(Scalar::Type arrayType, const T& src, AnyRegister dest, Register temp,
-                                   Label* fail, bool canonicalizeDoubles)
+                                   Label* fail, bool canonicalizeDoubles, unsigned numElems)
 {
     switch (arrayType) {
       case Scalar::Int8:
@@ -411,17 +454,59 @@ MacroAssembler::loadFromTypedArray(Scalar::Type arrayType, const T& src, AnyRegi
         if (canonicalizeDoubles)
             canonicalizeDouble(dest.fpu());
         break;
+      case Scalar::Int32x4:
+        switch (numElems) {
+          case 1:
+            loadInt32x1(src, dest.fpu());
+            break;
+          case 2:
+            loadInt32x2(src, dest.fpu());
+            break;
+          case 3:
+            loadInt32x3(src, dest.fpu());
+            break;
+          case 4:
+            loadUnalignedSimd128Int(src, dest.fpu());
+            break;
+          default: MOZ_CRASH("unexpected number of elements in SIMD load");
+        }
+        break;
+      case Scalar::Float32x4:
+        switch (numElems) {
+          case 1:
+            loadFloat32(src, dest.fpu());
+            break;
+          case 2:
+            loadDouble(src, dest.fpu());
+            break;
+          case 3:
+            loadFloat32x3(src, dest.fpu());
+            break;
+          case 4:
+            loadUnalignedSimd128Float(src, dest.fpu());
+            break;
+          default: MOZ_CRASH("unexpected number of elements in SIMD load");
+        }
+        break;
+      case Scalar::Int8x16:
+        MOZ_ASSERT(numElems == 16, "unexpected partial load");
+        loadUnalignedSimd128Int(src, dest.fpu());
+        break;
+      case Scalar::Int16x8:
+        MOZ_ASSERT(numElems == 8, "unexpected partial load");
+        loadUnalignedSimd128Int(src, dest.fpu());
+        break;
       default:
         MOZ_CRASH("Invalid typed array type");
     }
 }
 
-template void MacroAssembler::loadFromTypedArray(Scalar::Type arrayType, const Address& src,
+template void MacroAssembler::loadFromTypedArray(Scalar::Type arrayType, const Address& src, AnyRegister dest,
-                                                 AnyRegister dest, Register temp, Label* fail,
+                                                 Register temp, Label* fail, bool canonicalizeDoubles,
-                                                 bool canonicalizeDoubles);
+                                                 unsigned numElems);
-template void MacroAssembler::loadFromTypedArray(Scalar::Type arrayType, const BaseIndex& src,
+template void MacroAssembler::loadFromTypedArray(Scalar::Type arrayType, const BaseIndex& src, AnyRegister dest,
-                                                 AnyRegister dest, Register temp, Label* fail,
+                                                 Register temp, Label* fail, bool canonicalizeDoubles,
-                                                 bool canonicalizeDoubles);
+                                                 unsigned numElems);
 
 template<typename T>
 void
@@ -3310,6 +3395,41 @@ MacroAssembler::branchIfInlineTypedObject(Register obj, Register scratch, Label*
     branchPtr(Assembler::Equal, scratch, ImmPtr(&InlineTransparentTypedObject::class_), label);
 }
 
+void
+MacroAssembler::branchIfNotSimdObject(Register obj, Register scratch, SimdType simdType,
+                                      Label* label)
+{
+    loadPtr(Address(obj, JSObject::offsetOfGroup()), scratch);
+
+    // Guard that the object has the same representation as the one produced for
+    // SIMD value-type.
+    Address clasp(scratch, ObjectGroup::offsetOfClasp());
+    static_assert(!SimdTypeDescr::Opaque, "SIMD objects are transparent");
+    branchPtr(Assembler::NotEqual, clasp, ImmPtr(&InlineTransparentTypedObject::class_),
+              label);
+
+    // obj->type()->typeDescr()
+    // The previous class pointer comparison implies that the addendumKind is
+    // Addendum_TypeDescr.
+    loadPtr(Address(scratch, ObjectGroup::offsetOfAddendum()), scratch);
+
+    // Check for the /Kind/ reserved slot of the TypeDescr.  This is an Int32
+    // Value which is equivalent to the object class check.
+    static_assert(JS_DESCR_SLOT_KIND < NativeObject::MAX_FIXED_SLOTS, "Load from fixed slots");
+    Address typeDescrKind(scratch, NativeObject::getFixedSlotOffset(JS_DESCR_SLOT_KIND));
+    assertTestInt32(Assembler::Equal, typeDescrKind,
+      "MOZ_ASSERT(obj->type()->typeDescr()->getReservedSlot(JS_DESCR_SLOT_KIND).isInt32())");
+    branch32(Assembler::NotEqual, ToPayload(typeDescrKind), Imm32(js::type::Simd), label);
+
+    // Check if the SimdTypeDescr /Type/ matches the specialization of this
+    // MSimdUnbox instruction.
+    static_assert(JS_DESCR_SLOT_TYPE < NativeObject::MAX_FIXED_SLOTS, "Load from fixed slots");
+    Address typeDescrType(scratch, NativeObject::getFixedSlotOffset(JS_DESCR_SLOT_TYPE));
+    assertTestInt32(Assembler::Equal, typeDescrType,
+      "MOZ_ASSERT(obj->type()->typeDescr()->getReservedSlot(JS_DESCR_SLOT_TYPE).isInt32())");
+    branch32(Assembler::NotEqual, ToPayload(typeDescrType), Imm32(int32_t(simdType)), label);
+}
+
 void
 MacroAssembler::copyObjGroupNoPreBarrier(Register sourceObj, Register destObj, Register scratch)
 {

js/src/jit/MacroAssembler.h

@@ -1158,6 +1158,8 @@ class MacroAssembler : public MacroAssemblerSpecific
 
     void branchIfInlineTypedObject(Register obj, Register scratch, Label* label);
 
+    void branchIfNotSimdObject(Register obj, Register scratch, SimdType simdType, Label* label);
+
     inline void branchTestClassIsProxy(bool proxy, Register clasp, Label* label);
 
     inline void branchTestObjectIsProxy(bool proxy, Register object, Register scratch, Label* label);
@@ -1374,6 +1376,9 @@ class MacroAssembler : public MacroAssemblerSpecific
     inline void canonicalizeFloat(FloatRegister reg);
     inline void canonicalizeFloatIfDeterministic(FloatRegister reg);
 
+    inline void canonicalizeFloat32x4(FloatRegister reg, FloatRegister scratch)
+        DEFINED_ON(x86_shared);
+
   public:
     // ========================================================================
     // Memory access primitives.
@@ -2148,7 +2153,7 @@ class MacroAssembler : public MacroAssemblerSpecific
 
     template<typename T>
     void loadFromTypedArray(Scalar::Type arrayType, const T& src, AnyRegister dest, Register temp, Label* fail,
-                            bool canonicalizeDoubles = true);
+                            bool canonicalizeDoubles = true, unsigned numElems = 0);
 
     template<typename T>
     void loadFromTypedArray(Scalar::Type arrayType, const T& src, const ValueOperand& dest, bool allowDouble,
@@ -2175,8 +2180,10 @@ class MacroAssembler : public MacroAssemblerSpecific
         }
     }
 
-    void storeToTypedFloatArray(Scalar::Type arrayType, FloatRegister value, const BaseIndex& dest);
+    void storeToTypedFloatArray(Scalar::Type arrayType, FloatRegister value, const BaseIndex& dest,
-    void storeToTypedFloatArray(Scalar::Type arrayType, FloatRegister value, const Address& dest);
+                                unsigned numElems = 0);
+    void storeToTypedFloatArray(Scalar::Type arrayType, FloatRegister value, const Address& dest,
+                                unsigned numElems = 0);
 
     void memoryBarrierBefore(const Synchronization& sync);
     void memoryBarrierAfter(const Synchronization& sync);

js/src/jit/RangeAnalysis.cpp

@@ -1785,6 +1785,10 @@ GetTypedArrayRange(TempAllocator& alloc, Scalar::Type type)
       case Scalar::Int64:
       case Scalar::Float32:
       case Scalar::Float64:
+      case Scalar::Float32x4:
+      case Scalar::Int8x16:
+      case Scalar::Int16x8:
+      case Scalar::Int32x4:
       case Scalar::MaxTypedArrayViewType:
         break;
     }

js/src/jit/Recover.cpp

@@ -10,6 +10,7 @@
 #include "jsmath.h"
 
 #include "builtin/RegExp.h"
+#include "builtin/SIMD.h"
 #include "builtin/String.h"
 #include "builtin/TypedObject.h"
 #include "gc/Heap.h"
@@ -1661,6 +1662,79 @@ RNewCallObject::recover(JSContext* cx, SnapshotIterator& iter) const
     return true;
 }
 
+bool
+MSimdBox::writeRecoverData(CompactBufferWriter& writer) const
+{
+    MOZ_ASSERT(canRecoverOnBailout());
+    writer.writeUnsigned(uint32_t(RInstruction::Recover_SimdBox));
+    static_assert(unsigned(SimdType::Count) < 0x100, "assuming SimdType fits in 8 bits");
+    writer.writeByte(uint8_t(simdType()));
+    return true;
+}
+
+RSimdBox::RSimdBox(CompactBufferReader& reader)
+{
+    type_ = reader.readByte();
+}
+
+bool
+RSimdBox::recover(JSContext* cx, SnapshotIterator& iter) const
+{
+    JSObject* resultObject = nullptr;
+    RValueAllocation a = iter.readAllocation();
+    MOZ_ASSERT(iter.allocationReadable(a));
+    MOZ_ASSERT_IF(a.mode() == RValueAllocation::ANY_FLOAT_REG, a.fpuReg().isSimd128());
+    const FloatRegisters::RegisterContent* raw = iter.floatAllocationPointer(a);
+    switch (SimdType(type_)) {
+      case SimdType::Bool8x16:
+        resultObject = js::CreateSimd<Bool8x16>(cx, (const Bool8x16::Elem*) raw);
+        break;
+      case SimdType::Int8x16:
+        resultObject = js::CreateSimd<Int8x16>(cx, (const Int8x16::Elem*) raw);
+        break;
+      case SimdType::Uint8x16:
+        resultObject = js::CreateSimd<Uint8x16>(cx, (const Uint8x16::Elem*) raw);
+        break;
+      case SimdType::Bool16x8:
+        resultObject = js::CreateSimd<Bool16x8>(cx, (const Bool16x8::Elem*) raw);
+        break;
+      case SimdType::Int16x8:
+        resultObject = js::CreateSimd<Int16x8>(cx, (const Int16x8::Elem*) raw);
+        break;
+      case SimdType::Uint16x8:
+        resultObject = js::CreateSimd<Uint16x8>(cx, (const Uint16x8::Elem*) raw);
+        break;
+      case SimdType::Bool32x4:
+        resultObject = js::CreateSimd<Bool32x4>(cx, (const Bool32x4::Elem*) raw);
+        break;
+      case SimdType::Int32x4:
+        resultObject = js::CreateSimd<Int32x4>(cx, (const Int32x4::Elem*) raw);
+        break;
+      case SimdType::Uint32x4:
+        resultObject = js::CreateSimd<Uint32x4>(cx, (const Uint32x4::Elem*) raw);
+        break;
+      case SimdType::Float32x4:
+        resultObject = js::CreateSimd<Float32x4>(cx, (const Float32x4::Elem*) raw);
+        break;
+      case SimdType::Float64x2:
+        MOZ_CRASH("NYI, RSimdBox of Float64x2");
+        break;
+      case SimdType::Bool64x2:
+        MOZ_CRASH("NYI, RSimdBox of Bool64x2");
+        break;
+      case SimdType::Count:
+        MOZ_CRASH("RSimdBox of Count is unreachable");
+    }
+
+    if (!resultObject)
+        return false;
+
+    RootedValue result(cx);
+    result.setObject(*resultObject);
+    iter.storeInstructionResult(result);
+    return true;
+}
+
 bool
 MObjectState::writeRecoverData(CompactBufferWriter& writer) const
 {

js/src/jit/Recover.h

@@ -112,6 +112,7 @@ namespace jit {
     _(CreateThisWithTemplate)                   \
     _(Lambda)                                   \
     _(LambdaArrow)                              \
+    _(SimdBox)                                  \
     _(ObjectState)                              \
     _(ArrayState)                               \
     _(SetArrayLength)                           \
@@ -689,6 +690,17 @@ class RNewCallObject final : public RInstruction
     MOZ_MUST_USE bool recover(JSContext* cx, SnapshotIterator& iter) const override;
 };
 
+class RSimdBox final : public RInstruction
+{
+  private:
+    uint8_t type_;
+
+  public:
+    RINSTRUCTION_HEADER_NUM_OP_(SimdBox, 1)
+
+    MOZ_MUST_USE bool recover(JSContext* cx, SnapshotIterator& iter) const override;
+};
+
 class RObjectState final : public RInstruction
 {
   private:

js/src/jit/TypePolicy.cpp

@@ -637,6 +637,45 @@ template bool NoFloatPolicyAfter<0>::adjustInputs(TempAllocator& alloc, MInstruc
 template bool NoFloatPolicyAfter<1>::adjustInputs(TempAllocator& alloc, MInstruction* def) const;
 template bool NoFloatPolicyAfter<2>::adjustInputs(TempAllocator& alloc, MInstruction* def) const;
 
+template <unsigned Op>
+bool
+SimdScalarPolicy<Op>::staticAdjustInputs(TempAllocator& alloc, MInstruction* ins)
+{
+    MOZ_ASSERT(IsSimdType(ins->type()));
+    MIRType laneType = SimdTypeToLaneType(ins->type());
+
+    MDefinition* in = ins->getOperand(Op);
+
+    // A vector with boolean lanes requires Int32 inputs that have already been
+    // converted to 0/-1.
+    // We can't insert a MIRType::Boolean lane directly - it requires conversion.
+    if (laneType == MIRType::Boolean) {
+        MOZ_ASSERT(in->type() == MIRType::Int32, "Boolean SIMD vector requires Int32 lanes.");
+        return true;
+    }
+
+    if (in->type() == laneType)
+        return true;
+
+    MInstruction* replace;
+    if (laneType == MIRType::Int32) {
+        replace = MTruncateToInt32::New(alloc, in);
+    } else {
+        MOZ_ASSERT(laneType == MIRType::Float32);
+        replace = MToFloat32::New(alloc, in);
+    }
+
+    ins->block()->insertBefore(ins, replace);
+    ins->replaceOperand(Op, replace);
+
+    return replace->typePolicy()->adjustInputs(alloc, replace);
+}
+
+template bool SimdScalarPolicy<0>::staticAdjustInputs(TempAllocator& alloc, MInstruction* def);
+template bool SimdScalarPolicy<1>::staticAdjustInputs(TempAllocator& alloc, MInstruction* def);
+template bool SimdScalarPolicy<2>::staticAdjustInputs(TempAllocator& alloc, MInstruction* def);
+template bool SimdScalarPolicy<3>::staticAdjustInputs(TempAllocator& alloc, MInstruction* def);
+
 template <unsigned Op>
 bool
 BoxPolicy<Op>::staticAdjustInputs(TempAllocator& alloc, MInstruction* ins)
@@ -818,6 +857,75 @@ template bool ObjectPolicy<1>::staticAdjustInputs(TempAllocator& alloc, MInstruc
 template bool ObjectPolicy<2>::staticAdjustInputs(TempAllocator& alloc, MInstruction* ins);
 template bool ObjectPolicy<3>::staticAdjustInputs(TempAllocator& alloc, MInstruction* ins);
 
+template <unsigned Op>
+bool
+SimdSameAsReturnedTypePolicy<Op>::staticAdjustInputs(TempAllocator& alloc, MInstruction* ins)
+{
+    MOZ_ASSERT(ins->type() == ins->getOperand(Op)->type());
+    return true;
+}
+
+template bool
+SimdSameAsReturnedTypePolicy<0>::staticAdjustInputs(TempAllocator& alloc, MInstruction* ins);
+template bool
+SimdSameAsReturnedTypePolicy<1>::staticAdjustInputs(TempAllocator& alloc, MInstruction* ins);
+
+bool
+SimdAllPolicy::adjustInputs(TempAllocator& alloc, MInstruction* ins) const
+{
+    for (unsigned i = 0, e = ins->numOperands(); i < e; i++)
+        MOZ_ASSERT(ins->getOperand(i)->type() == ins->typePolicySpecialization());
+    return true;
+}
+
+template <unsigned Op>
+bool
+SimdPolicy<Op>::adjustInputs(TempAllocator& alloc, MInstruction* ins) const
+{
+    MOZ_ASSERT(ins->typePolicySpecialization() == ins->getOperand(Op)->type());
+    return true;
+}
+
+template bool
+SimdPolicy<0>::adjustInputs(TempAllocator& alloc, MInstruction* ins) const;
+
+bool
+SimdShufflePolicy::adjustInputs(TempAllocator& alloc, MInstruction* ins) const
+{
+    MSimdGeneralShuffle* s = ins->toSimdGeneralShuffle();
+
+    for (unsigned i = 0; i < s->numVectors(); i++)
+        MOZ_ASSERT(ins->getOperand(i)->type() == ins->typePolicySpecialization());
+
+    // Next inputs are the lanes, which need to be int32
+    for (unsigned i = 0; i < s->numLanes(); i++) {
+        MDefinition* in = ins->getOperand(s->numVectors() + i);
+        if (in->type() == MIRType::Int32)
+            continue;
+
+        auto* replace = MToNumberInt32::New(alloc, in, IntConversionInputKind::NumbersOnly);
+        ins->block()->insertBefore(ins, replace);
+        ins->replaceOperand(s->numVectors() + i, replace);
+        if (!replace->typePolicy()->adjustInputs(alloc, replace))
+            return false;
+    }
+
+    return true;
+}
+
+bool
+SimdSelectPolicy::adjustInputs(TempAllocator& alloc, MInstruction* ins) const
+{
+    // First input is the mask, which has to be a boolean.
+    MOZ_ASSERT(IsBooleanSimdType(ins->getOperand(0)->type()));
+
+    // Next inputs are the two vectors of a particular type.
+    for (unsigned i = 1; i < 3; i++)
+        MOZ_ASSERT(ins->getOperand(i)->type() == ins->typePolicySpecialization());
+
+    return true;
+}
+
 bool
 CallPolicy::adjustInputs(TempAllocator& alloc, MInstruction* ins) const
 {
@@ -875,6 +983,13 @@ StoreUnboxedScalarPolicy::adjustValueInput(TempAllocator& alloc, MInstruction* i
                                            Scalar::Type writeType, MDefinition* value,
                                            int valueOperand)
 {
+    // Storing a SIMD value requires a valueOperand that has already been
+    // SimdUnboxed. See IonBuilder::inlineSimdStore(()
+    if (Scalar::isSimdType(writeType)) {
+        MOZ_ASSERT(IsSimdType(value->type()));
+        return true;
+    }
+
     MDefinition* curValue = value;
     // First, ensure the value is int32, boolean, double or Value.
     // The conversion is based on TypedArrayObjectTemplate::setElementTail.
@@ -1155,6 +1270,9 @@ FilterTypeSetPolicy::adjustInputs(TempAllocator& alloc, MInstruction* ins) const
     _(PowPolicy)                                \
     _(SameValuePolicy)                          \
     _(SignPolicy)                               \
+    _(SimdAllPolicy)                            \
+    _(SimdSelectPolicy)                         \
+    _(SimdShufflePolicy)                        \
     _(StoreTypedArrayHolePolicy)                \
     _(StoreUnboxedScalarPolicy)                 \
     _(StoreUnboxedObjectOrNullPolicy)           \
@@ -1192,6 +1310,7 @@ FilterTypeSetPolicy::adjustInputs(TempAllocator& alloc, MInstruction* ins) const
     _(MixPolicy<ObjectPolicy<0>, UnboxedInt32Policy<1>, UnboxedInt32Policy<2>, UnboxedInt32Policy<3>>)        \
     _(MixPolicy<ObjectPolicy<0>, UnboxedInt32Policy<1>, TruncateToInt32Policy<2>, TruncateToInt32Policy<3> >) \
     _(MixPolicy<ObjectPolicy<0>, CacheIdPolicy<1>, NoFloatPolicy<2>>)                                         \
+    _(MixPolicy<SimdScalarPolicy<0>, SimdScalarPolicy<1>, SimdScalarPolicy<2>, SimdScalarPolicy<3> >)         \
     _(MixPolicy<ObjectPolicy<0>, BoxExceptPolicy<1, MIRType::Object>, CacheIdPolicy<2>>)                      \
     _(MixPolicy<BoxPolicy<0>, ObjectPolicy<1> >)                                                              \
     _(MixPolicy<ConvertToStringPolicy<0>, ConvertToStringPolicy<1> >)                                         \
@@ -1211,6 +1330,8 @@ FilterTypeSetPolicy::adjustInputs(TempAllocator& alloc, MInstruction* ins) const
     _(MixPolicy<ObjectPolicy<0>, StringPolicy<1> >)                                                           \
     _(MixPolicy<ObjectPolicy<0>, ConvertToStringPolicy<2> >)                                                  \
     _(MixPolicy<ObjectPolicy<1>, ConvertToStringPolicy<0> >)                                                  \
+    _(MixPolicy<SimdSameAsReturnedTypePolicy<0>, SimdSameAsReturnedTypePolicy<1> >)                           \
+    _(MixPolicy<SimdSameAsReturnedTypePolicy<0>, SimdScalarPolicy<1> >)                                       \
     _(MixPolicy<StringPolicy<0>, UnboxedInt32Policy<1> >)                                                     \
     _(MixPolicy<StringPolicy<0>, StringPolicy<1> >)                                                           \
     _(MixPolicy<BoxPolicy<0>, BoxPolicy<1> >)                                                                 \
@@ -1221,6 +1342,9 @@ FilterTypeSetPolicy::adjustInputs(TempAllocator& alloc, MInstruction* ins) const
     _(ObjectPolicy<0>)                                                                                        \
     _(ObjectPolicy<1>)                                                                                        \
     _(ObjectPolicy<3>)                                                                                        \
+    _(SimdPolicy<0>)                                                                                          \
+    _(SimdSameAsReturnedTypePolicy<0>)                                                                        \
+    _(SimdScalarPolicy<0>)                                                                                    \
     _(StringPolicy<0>)
 
 

js/src/jit/TypePolicy.h

@@ -365,6 +365,67 @@ class ObjectPolicy final : public TypePolicy
 // a primitive, we use ValueToNonNullObject.
 typedef ObjectPolicy<0> SingleObjectPolicy;
 
+// Convert an operand to have a type identical to the scalar type of the
+// returned type of the instruction.
+template <unsigned Op>
+class SimdScalarPolicy final : public TypePolicy
+{
+  public:
+    constexpr SimdScalarPolicy() { }
+    EMPTY_DATA_;
+    static MOZ_MUST_USE bool staticAdjustInputs(TempAllocator& alloc, MInstruction* def);
+    MOZ_MUST_USE bool adjustInputs(TempAllocator& alloc, MInstruction* def) const override {
+        return staticAdjustInputs(alloc, def);
+    }
+};
+
+class SimdAllPolicy final : public TypePolicy
+{
+  public:
+    constexpr SimdAllPolicy () { }
+    SPECIALIZATION_DATA_;
+    MOZ_MUST_USE bool adjustInputs(TempAllocator& alloc, MInstruction* ins) const override;
+};
+
+template <unsigned Op>
+class SimdPolicy final : public TypePolicy
+{
+  public:
+    constexpr SimdPolicy() { }
+    SPECIALIZATION_DATA_;
+    MOZ_MUST_USE bool adjustInputs(TempAllocator& alloc, MInstruction* ins) const override;
+};
+
+class SimdSelectPolicy final : public TypePolicy
+{
+  public:
+    constexpr SimdSelectPolicy() { }
+    SPECIALIZATION_DATA_;
+    MOZ_MUST_USE bool adjustInputs(TempAllocator& alloc, MInstruction* ins) const override;
+};
+
+class SimdShufflePolicy final : public TypePolicy
+{
+  public:
+    constexpr SimdShufflePolicy() { }
+    SPECIALIZATION_DATA_;
+    MOZ_MUST_USE bool adjustInputs(TempAllocator& alloc, MInstruction* ins) const override;
+};
+
+// SIMD value-type policy, use the returned type of the instruction to determine
+// how to unbox its operand.
+template <unsigned Op>
+class SimdSameAsReturnedTypePolicy final : public TypePolicy
+{
+  public:
+    constexpr SimdSameAsReturnedTypePolicy() { }
+    EMPTY_DATA_;
+    static MOZ_MUST_USE bool staticAdjustInputs(TempAllocator& alloc, MInstruction* ins);
+    MOZ_MUST_USE bool adjustInputs(TempAllocator& alloc, MInstruction* ins) const override {
+        return staticAdjustInputs(alloc, ins);
+    }
+};
+
 template <unsigned Op>
 class BoxPolicy final : public TypePolicy
 {

js/src/jit/TypedObjectPrediction.cpp

@@ -119,6 +119,7 @@ TypedObjectPrediction::ofArrayKind() const
     switch (kind()) {
       case type::Scalar:
       case type::Reference:
+      case type::Simd:
       case type::Struct:
         return false;
 
@@ -206,6 +207,12 @@ TypedObjectPrediction::referenceType() const
     return extractType<ReferenceTypeDescr>();
 }
 
+SimdType
+TypedObjectPrediction::simdType() const
+{
+    return descr().as<SimdTypeDescr>().type();
+}
+
 bool
 TypedObjectPrediction::hasKnownArrayLength(int32_t* length) const
 {

js/src/jit/TypedObjectPrediction.h

@@ -165,10 +165,11 @@ class TypedObjectPrediction {
     //////////////////////////////////////////////////////////////////////
     // Simple operations
     //
-    // Only valid when |kind()| is Scalar or Reference.
+    // Only valid when |kind()| is Scalar, Reference, or Simd (as appropriate).
 
     Scalar::Type scalarType() const;
     ReferenceType referenceType() const;
+    SimdType simdType() const;
 
     ///////////////////////////////////////////////////////////////////////////
     // Queries valid only for arrays.

js/src/jit/arm/CodeGenerator-arm.cpp

@@ -3156,14 +3156,248 @@ CodeGenerator::visitWasmAtomicExchangeI64(LWasmAtomicExchangeI64* lir)
     masm.atomicExchange64(Synchronization::Full(), addr, value, out);
 }
 
+void
+CodeGenerator::visitSimdSplatX4(LSimdSplatX4* lir)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimd128Int(LSimd128Int* ins)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimd128Float(LSimd128Float* ins)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdExtractElementI(LSimdExtractElementI* ins)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdExtractElementF(LSimdExtractElementF* ins)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdBinaryCompIx4(LSimdBinaryCompIx4* lir)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdBinaryCompFx4(LSimdBinaryCompFx4* lir)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdBinaryArithIx4(LSimdBinaryArithIx4* lir)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdBinaryArithFx4(LSimdBinaryArithFx4* lir)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdBinaryBitwise(LSimdBinaryBitwise* lir)
+{
+    MOZ_CRASH("NYI");
+}
+
 void
 CodeGenerator::visitNearbyInt(LNearbyInt*)
 {
     MOZ_CRASH("NYI");
 }
 
+void
+CodeGenerator::visitSimdShift(LSimdShift*)
+{
+    MOZ_CRASH("NYI");
+}
+
 void
 CodeGenerator::visitNearbyIntF(LNearbyIntF*)
 {
     MOZ_CRASH("NYI");
 }
+
+void
+CodeGenerator::visitSimdSelect(LSimdSelect*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdAllTrue(LSimdAllTrue*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdAnyTrue(LSimdAnyTrue*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdShuffle(LSimdShuffle*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdSplatX8(LSimdSplatX8*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdSplatX16(LSimdSplatX16*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdSwizzleF(LSimdSwizzleF*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdSwizzleI(LSimdSwizzleI*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdShuffleX4(LSimdShuffleX4*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdBinaryCompIx8(LSimdBinaryCompIx8*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdUnaryArithFx4(LSimdUnaryArithFx4*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdUnaryArithIx4(LSimdUnaryArithIx4*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdUnaryArithIx8(LSimdUnaryArithIx8*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitFloat32x4ToInt32x4(LFloat32x4ToInt32x4*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitInt32x4ToFloat32x4(LInt32x4ToFloat32x4*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdBinaryArithIx8(LSimdBinaryArithIx8*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdBinaryCompIx16(LSimdBinaryCompIx16*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdInsertElementF(LSimdInsertElementF*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdInsertElementI(LSimdInsertElementI*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdUnaryArithIx16(LSimdUnaryArithIx16*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitFloat32x4ToUint32x4(LFloat32x4ToUint32x4*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdBinaryArithIx16(LSimdBinaryArithIx16*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdExtractElementB(LSimdExtractElementB*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdGeneralShuffleF(LSimdGeneralShuffleF*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdGeneralShuffleI(LSimdGeneralShuffleI*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdReinterpretCast(LSimdReinterpretCast*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdBinarySaturating(LSimdBinarySaturating*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+CodeGenerator::visitSimdExtractElementU2D(LSimdExtractElementU2D*)
+{
+    MOZ_CRASH("NYI");
+}

js/src/jit/arm/Lowering-arm.cpp

@@ -1058,3 +1058,63 @@ LIRGenerator::visitSignExtendInt64(MSignExtendInt64* ins)
 {
     defineInt64(new(alloc()) LSignExtendInt64(useInt64RegisterAtStart(ins->input())), ins);
 }
+
+void
+LIRGenerator::visitSimdInsertElement(MSimdInsertElement*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+LIRGenerator::visitSimdExtractElement(MSimdExtractElement*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+LIRGenerator::visitSimdBinaryArith(MSimdBinaryArith*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+LIRGenerator::visitSimdSelect(MSimdSelect*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+LIRGenerator::visitSimdSplat(MSimdSplat*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+LIRGenerator::visitSimdValueX4(MSimdValueX4*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+LIRGenerator::visitSimdBinarySaturating(MSimdBinarySaturating*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+LIRGenerator::visitSimdSwizzle(MSimdSwizzle*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+LIRGenerator::visitSimdShuffle(MSimdShuffle*)
+{
+    MOZ_CRASH("NYI");
+}
+
+void
+LIRGenerator::visitSimdGeneralShuffle(MSimdGeneralShuffle*)
+{
+    MOZ_CRASH("NYI");
+}

js/src/jit/arm/Lowering-arm.h

@@ -63,6 +63,17 @@ class LIRGeneratorARM : public LIRGeneratorShared
     void lowerForFPU(LInstructionHelper<1, 2, Temps>* ins, MDefinition* mir,
                      MDefinition* lhs, MDefinition* rhs);
 
+    void lowerForCompIx4(LSimdBinaryCompIx4* ins, MSimdBinaryComp* mir,
+                         MDefinition* lhs, MDefinition* rhs)
+    {
+        return lowerForFPU(ins, mir, lhs, rhs);
+    }
+    void lowerForCompFx4(LSimdBinaryCompFx4* ins, MSimdBinaryComp* mir,
+                         MDefinition* lhs, MDefinition* rhs)
+    {
+        return lowerForFPU(ins, mir, lhs, rhs);
+    }
+
     void lowerForBitAndAndBranch(LBitAndAndBranch* baab, MInstruction* mir,
                                  MDefinition* lhs, MDefinition* rhs);
     void lowerTruncateDToInt32(MTruncateToInt32* ins);