| // Copyright (c) 2013, the Dart project authors. Please see the AUTHORS file |
| // for details. All rights reserved. Use of this source code is governed by a |
| // BSD-style license that can be found in the LICENSE file. |
| |
| #include "vm/bootstrap_natives.h" |
| |
| #include "vm/exceptions.h" |
| #include "vm/native_entry.h" |
| #include "vm/object.h" |
| #include "vm/symbols.h" |
| |
| namespace dart { |
| |
| static void ThrowMaskRangeException(int64_t m) { |
| if ((m < 0) || (m > 255)) { |
| Exceptions::ThrowRangeError("mask", Integer::Handle(Integer::New(m)), 0, |
| 255); |
| } |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_fromDoubles, 0, 4) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Double, x, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Double, y, arguments->NativeArgAt(1)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Double, z, arguments->NativeArgAt(2)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Double, w, arguments->NativeArgAt(3)); |
| float _x = static_cast<float>(x.value()); |
| float _y = static_cast<float>(y.value()); |
| float _z = static_cast<float>(z.value()); |
| float _w = static_cast<float>(w.value()); |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_splat, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Double, v, arguments->NativeArgAt(0)); |
| float _v = v.value(); |
| return Float32x4::New(_v, _v, _v, _v); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_zero, 0, 1) { |
| ASSERT( |
| TypeArguments::CheckedHandle(zone, arguments->NativeArgAt(0)).IsNull()); |
| return Float32x4::New(0.0f, 0.0f, 0.0f, 0.0f); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_fromInt32x4Bits, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, v, arguments->NativeArgAt(1)); |
| return Float32x4::New(v.value()); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_fromFloat64x2, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, v, arguments->NativeArgAt(1)); |
| float _x = static_cast<float>(v.x()); |
| float _y = static_cast<float>(v.y()); |
| return Float32x4::New(_x, _y, 0.0f, 0.0f); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_add, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, other, arguments->NativeArgAt(1)); |
| float _x = self.x() + other.x(); |
| float _y = self.y() + other.y(); |
| float _z = self.z() + other.z(); |
| float _w = self.w() + other.w(); |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_negate, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| float _x = -self.x(); |
| float _y = -self.y(); |
| float _z = -self.z(); |
| float _w = -self.w(); |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_sub, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, other, arguments->NativeArgAt(1)); |
| float _x = self.x() - other.x(); |
| float _y = self.y() - other.y(); |
| float _z = self.z() - other.z(); |
| float _w = self.w() - other.w(); |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_mul, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, other, arguments->NativeArgAt(1)); |
| float _x = self.x() * other.x(); |
| float _y = self.y() * other.y(); |
| float _z = self.z() * other.z(); |
| float _w = self.w() * other.w(); |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_div, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, other, arguments->NativeArgAt(1)); |
| float _x = self.x() / other.x(); |
| float _y = self.y() / other.y(); |
| float _z = self.z() / other.z(); |
| float _w = self.w() / other.w(); |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_cmplt, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, a, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, b, arguments->NativeArgAt(1)); |
| uint32_t _x = a.x() < b.x() ? 0xFFFFFFFF : 0x0; |
| uint32_t _y = a.y() < b.y() ? 0xFFFFFFFF : 0x0; |
| uint32_t _z = a.z() < b.z() ? 0xFFFFFFFF : 0x0; |
| uint32_t _w = a.w() < b.w() ? 0xFFFFFFFF : 0x0; |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_cmplte, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, a, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, b, arguments->NativeArgAt(1)); |
| uint32_t _x = a.x() <= b.x() ? 0xFFFFFFFF : 0x0; |
| uint32_t _y = a.y() <= b.y() ? 0xFFFFFFFF : 0x0; |
| uint32_t _z = a.z() <= b.z() ? 0xFFFFFFFF : 0x0; |
| uint32_t _w = a.w() <= b.w() ? 0xFFFFFFFF : 0x0; |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_cmpgt, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, a, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, b, arguments->NativeArgAt(1)); |
| uint32_t _x = a.x() > b.x() ? 0xFFFFFFFF : 0x0; |
| uint32_t _y = a.y() > b.y() ? 0xFFFFFFFF : 0x0; |
| uint32_t _z = a.z() > b.z() ? 0xFFFFFFFF : 0x0; |
| uint32_t _w = a.w() > b.w() ? 0xFFFFFFFF : 0x0; |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_cmpgte, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, a, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, b, arguments->NativeArgAt(1)); |
| uint32_t _x = a.x() >= b.x() ? 0xFFFFFFFF : 0x0; |
| uint32_t _y = a.y() >= b.y() ? 0xFFFFFFFF : 0x0; |
| uint32_t _z = a.z() >= b.z() ? 0xFFFFFFFF : 0x0; |
| uint32_t _w = a.w() >= b.w() ? 0xFFFFFFFF : 0x0; |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_cmpequal, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, a, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, b, arguments->NativeArgAt(1)); |
| uint32_t _x = a.x() == b.x() ? 0xFFFFFFFF : 0x0; |
| uint32_t _y = a.y() == b.y() ? 0xFFFFFFFF : 0x0; |
| uint32_t _z = a.z() == b.z() ? 0xFFFFFFFF : 0x0; |
| uint32_t _w = a.w() == b.w() ? 0xFFFFFFFF : 0x0; |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_cmpnequal, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, a, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, b, arguments->NativeArgAt(1)); |
| uint32_t _x = a.x() != b.x() ? 0xFFFFFFFF : 0x0; |
| uint32_t _y = a.y() != b.y() ? 0xFFFFFFFF : 0x0; |
| uint32_t _z = a.z() != b.z() ? 0xFFFFFFFF : 0x0; |
| uint32_t _w = a.w() != b.w() ? 0xFFFFFFFF : 0x0; |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_scale, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Double, scale, arguments->NativeArgAt(1)); |
| float _s = static_cast<float>(scale.value()); |
| float _x = self.x() * _s; |
| float _y = self.y() * _s; |
| float _z = self.z() * _s; |
| float _w = self.w() * _s; |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_abs, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| float _x = fabsf(self.x()); |
| float _y = fabsf(self.y()); |
| float _z = fabsf(self.z()); |
| float _w = fabsf(self.w()); |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_clamp, 0, 3) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, lo, arguments->NativeArgAt(1)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, hi, arguments->NativeArgAt(2)); |
| // The order of the clamping must match the order of the optimized code: |
| // MAX(MIN(self, hi), lo). |
| float _x; |
| float _y; |
| float _z; |
| float _w; |
| // ARM semantics are different from X86/X64 at an instruction level. Ensure |
| // that we match the semantics of the architecture in the C version. |
| #if defined(TARGET_ARCH_IA32) || defined(TARGET_ARCH_X64) |
| _x = self.x() < hi.x() ? self.x() : hi.x(); |
| _y = self.y() < hi.y() ? self.y() : hi.y(); |
| _z = self.z() < hi.z() ? self.z() : hi.z(); |
| _w = self.w() < hi.w() ? self.w() : hi.w(); |
| _x = lo.x() < _x ? _x : lo.x(); |
| _y = lo.y() < _y ? _y : lo.y(); |
| _z = lo.z() < _z ? _z : lo.z(); |
| _w = lo.w() < _w ? _w : lo.w(); |
| #else |
| _x = fminf(self.x(), hi.x()); |
| _y = fminf(self.y(), hi.y()); |
| _z = fminf(self.z(), hi.z()); |
| _w = fminf(self.w(), hi.w()); |
| _x = fmaxf(_x, lo.x()); |
| _y = fmaxf(_y, lo.y()); |
| _z = fmaxf(_z, lo.z()); |
| _w = fmaxf(_w, lo.w()); |
| #endif |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_getX, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| double value = static_cast<double>(self.x()); |
| return Double::New(value); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_getY, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| double value = static_cast<double>(self.y()); |
| return Double::New(value); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_getZ, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| double value = static_cast<double>(self.z()); |
| return Double::New(value); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_getW, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| double value = static_cast<double>(self.w()); |
| return Double::New(value); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_getSignMask, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| uint32_t mx = (bit_cast<uint32_t>(self.x()) & 0x80000000) >> 31; |
| uint32_t my = (bit_cast<uint32_t>(self.y()) & 0x80000000) >> 31; |
| uint32_t mz = (bit_cast<uint32_t>(self.z()) & 0x80000000) >> 31; |
| uint32_t mw = (bit_cast<uint32_t>(self.w()) & 0x80000000) >> 31; |
| uint32_t value = mx | (my << 1) | (mz << 2) | (mw << 3); |
| return Integer::New(value); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_getSignMask, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| uint32_t mx = (self.x() & 0x80000000) >> 31; |
| uint32_t my = (self.y() & 0x80000000) >> 31; |
| uint32_t mz = (self.z() & 0x80000000) >> 31; |
| uint32_t mw = (self.w() & 0x80000000) >> 31; |
| uint32_t value = mx | (my << 1) | (mz << 2) | (mw << 3); |
| return Integer::New(value); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_shuffle, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Integer, mask, arguments->NativeArgAt(1)); |
| int64_t m = mask.AsInt64Value(); |
| ThrowMaskRangeException(m); |
| float data[4] = {self.x(), self.y(), self.z(), self.w()}; |
| float _x = data[m & 0x3]; |
| float _y = data[(m >> 2) & 0x3]; |
| float _z = data[(m >> 4) & 0x3]; |
| float _w = data[(m >> 6) & 0x3]; |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_shuffleMix, 0, 3) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, other, arguments->NativeArgAt(1)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Integer, mask, arguments->NativeArgAt(2)); |
| int64_t m = mask.AsInt64Value(); |
| ThrowMaskRangeException(m); |
| float data[4] = {self.x(), self.y(), self.z(), self.w()}; |
| float other_data[4] = {other.x(), other.y(), other.z(), other.w()}; |
| float _x = data[m & 0x3]; |
| float _y = data[(m >> 2) & 0x3]; |
| float _z = other_data[(m >> 4) & 0x3]; |
| float _w = other_data[(m >> 6) & 0x3]; |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_setX, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Double, x, arguments->NativeArgAt(1)); |
| float _x = static_cast<float>(x.value()); |
| float _y = self.y(); |
| float _z = self.z(); |
| float _w = self.w(); |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_setY, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Double, y, arguments->NativeArgAt(1)); |
| float _x = self.x(); |
| float _y = static_cast<float>(y.value()); |
| float _z = self.z(); |
| float _w = self.w(); |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_setZ, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Double, z, arguments->NativeArgAt(1)); |
| float _x = self.x(); |
| float _y = self.y(); |
| float _z = static_cast<float>(z.value()); |
| float _w = self.w(); |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_setW, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Double, w, arguments->NativeArgAt(1)); |
| float _x = self.x(); |
| float _y = self.y(); |
| float _z = self.z(); |
| float _w = static_cast<float>(w.value()); |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_min, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, other, arguments->NativeArgAt(1)); |
| float _x = self.x() < other.x() ? self.x() : other.x(); |
| float _y = self.y() < other.y() ? self.y() : other.y(); |
| float _z = self.z() < other.z() ? self.z() : other.z(); |
| float _w = self.w() < other.w() ? self.w() : other.w(); |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_max, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, other, arguments->NativeArgAt(1)); |
| float _x = self.x() > other.x() ? self.x() : other.x(); |
| float _y = self.y() > other.y() ? self.y() : other.y(); |
| float _z = self.z() > other.z() ? self.z() : other.z(); |
| float _w = self.w() > other.w() ? self.w() : other.w(); |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_sqrt, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| float _x = sqrtf(self.x()); |
| float _y = sqrtf(self.y()); |
| float _z = sqrtf(self.z()); |
| float _w = sqrtf(self.w()); |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_reciprocal, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| float _x = 1.0f / self.x(); |
| float _y = 1.0f / self.y(); |
| float _z = 1.0f / self.z(); |
| float _w = 1.0f / self.w(); |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float32x4_reciprocalSqrt, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0)); |
| float _x = sqrtf(1.0f / self.x()); |
| float _y = sqrtf(1.0f / self.y()); |
| float _z = sqrtf(1.0f / self.z()); |
| float _w = sqrtf(1.0f / self.w()); |
| return Float32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_fromInts, 0, 4) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Integer, x, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Integer, y, arguments->NativeArgAt(1)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Integer, z, arguments->NativeArgAt(2)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Integer, w, arguments->NativeArgAt(3)); |
| int32_t _x = static_cast<int32_t>(x.AsTruncatedUint32Value()); |
| int32_t _y = static_cast<int32_t>(y.AsTruncatedUint32Value()); |
| int32_t _z = static_cast<int32_t>(z.AsTruncatedUint32Value()); |
| int32_t _w = static_cast<int32_t>(w.AsTruncatedUint32Value()); |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_fromBools, 0, 4) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Bool, x, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Bool, y, arguments->NativeArgAt(1)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Bool, z, arguments->NativeArgAt(2)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Bool, w, arguments->NativeArgAt(3)); |
| int32_t _x = x.value() ? 0xFFFFFFFF : 0x0; |
| int32_t _y = y.value() ? 0xFFFFFFFF : 0x0; |
| int32_t _z = z.value() ? 0xFFFFFFFF : 0x0; |
| int32_t _w = w.value() ? 0xFFFFFFFF : 0x0; |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_fromFloat32x4Bits, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, v, arguments->NativeArgAt(1)); |
| return Int32x4::New(v.value()); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_or, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, other, arguments->NativeArgAt(1)); |
| int32_t _x = self.x() | other.x(); |
| int32_t _y = self.y() | other.y(); |
| int32_t _z = self.z() | other.z(); |
| int32_t _w = self.w() | other.w(); |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_and, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, other, arguments->NativeArgAt(1)); |
| int32_t _x = self.x() & other.x(); |
| int32_t _y = self.y() & other.y(); |
| int32_t _z = self.z() & other.z(); |
| int32_t _w = self.w() & other.w(); |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_xor, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, other, arguments->NativeArgAt(1)); |
| int32_t _x = self.x() ^ other.x(); |
| int32_t _y = self.y() ^ other.y(); |
| int32_t _z = self.z() ^ other.z(); |
| int32_t _w = self.w() ^ other.w(); |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_add, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, other, arguments->NativeArgAt(1)); |
| int32_t _x = self.x() + other.x(); |
| int32_t _y = self.y() + other.y(); |
| int32_t _z = self.z() + other.z(); |
| int32_t _w = self.w() + other.w(); |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_sub, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, other, arguments->NativeArgAt(1)); |
| int32_t _x = self.x() - other.x(); |
| int32_t _y = self.y() - other.y(); |
| int32_t _z = self.z() - other.z(); |
| int32_t _w = self.w() - other.w(); |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_getX, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| int32_t value = self.x(); |
| return Integer::New(value); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_getY, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| int32_t value = self.y(); |
| return Integer::New(value); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_getZ, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| int32_t value = self.z(); |
| return Integer::New(value); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_getW, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| int32_t value = self.w(); |
| return Integer::New(value); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_shuffle, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Integer, mask, arguments->NativeArgAt(1)); |
| int64_t m = mask.AsInt64Value(); |
| ThrowMaskRangeException(m); |
| int32_t data[4] = {self.x(), self.y(), self.z(), self.w()}; |
| int32_t _x = data[m & 0x3]; |
| int32_t _y = data[(m >> 2) & 0x3]; |
| int32_t _z = data[(m >> 4) & 0x3]; |
| int32_t _w = data[(m >> 6) & 0x3]; |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_shuffleMix, 0, 3) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, zw, arguments->NativeArgAt(1)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Integer, mask, arguments->NativeArgAt(2)); |
| int64_t m = mask.AsInt64Value(); |
| ThrowMaskRangeException(m); |
| int32_t data[4] = {self.x(), self.y(), self.z(), self.w()}; |
| int32_t zw_data[4] = {zw.x(), zw.y(), zw.z(), zw.w()}; |
| int32_t _x = data[m & 0x3]; |
| int32_t _y = data[(m >> 2) & 0x3]; |
| int32_t _z = zw_data[(m >> 4) & 0x3]; |
| int32_t _w = zw_data[(m >> 6) & 0x3]; |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_setX, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Integer, x, arguments->NativeArgAt(1)); |
| int32_t _x = static_cast<int32_t>(x.AsInt64Value() & 0xFFFFFFFF); |
| int32_t _y = self.y(); |
| int32_t _z = self.z(); |
| int32_t _w = self.w(); |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_setY, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Integer, y, arguments->NativeArgAt(1)); |
| int32_t _x = self.x(); |
| int32_t _y = static_cast<int32_t>(y.AsInt64Value() & 0xFFFFFFFF); |
| int32_t _z = self.z(); |
| int32_t _w = self.w(); |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_setZ, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Integer, z, arguments->NativeArgAt(1)); |
| int32_t _x = self.x(); |
| int32_t _y = self.y(); |
| int32_t _z = static_cast<int32_t>(z.AsInt64Value() & 0xFFFFFFFF); |
| int32_t _w = self.w(); |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_setW, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Integer, w, arguments->NativeArgAt(1)); |
| int32_t _x = self.x(); |
| int32_t _y = self.y(); |
| int32_t _z = self.z(); |
| int32_t _w = static_cast<int32_t>(w.AsInt64Value() & 0xFFFFFFFF); |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_getFlagX, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| int32_t value = self.x(); |
| return Bool::Get(value != 0).raw(); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_getFlagY, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| int32_t value = self.y(); |
| return Bool::Get(value != 0).raw(); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_getFlagZ, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| int32_t value = self.z(); |
| return Bool::Get(value != 0).raw(); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_getFlagW, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| int32_t value = self.w(); |
| return Bool::Get(value != 0).raw(); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_setFlagX, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Bool, flagX, arguments->NativeArgAt(1)); |
| int32_t _x = self.x(); |
| int32_t _y = self.y(); |
| int32_t _z = self.z(); |
| int32_t _w = self.w(); |
| _x = flagX.raw() == Bool::True().raw() ? 0xFFFFFFFF : 0x0; |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_setFlagY, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Bool, flagY, arguments->NativeArgAt(1)); |
| int32_t _x = self.x(); |
| int32_t _y = self.y(); |
| int32_t _z = self.z(); |
| int32_t _w = self.w(); |
| _y = flagY.raw() == Bool::True().raw() ? 0xFFFFFFFF : 0x0; |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_setFlagZ, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Bool, flagZ, arguments->NativeArgAt(1)); |
| int32_t _x = self.x(); |
| int32_t _y = self.y(); |
| int32_t _z = self.z(); |
| int32_t _w = self.w(); |
| _z = flagZ.raw() == Bool::True().raw() ? 0xFFFFFFFF : 0x0; |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_setFlagW, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Bool, flagW, arguments->NativeArgAt(1)); |
| int32_t _x = self.x(); |
| int32_t _y = self.y(); |
| int32_t _z = self.z(); |
| int32_t _w = self.w(); |
| _w = flagW.raw() == Bool::True().raw() ? 0xFFFFFFFF : 0x0; |
| return Int32x4::New(_x, _y, _z, _w); |
| } |
| |
| // Used to convert between int32_t and float32 without breaking strict |
| // aliasing rules. |
| union float32_int32 { |
| float f; |
| int32_t u; |
| float32_int32(float v) { f = v; } |
| float32_int32(int32_t v) { u = v; } |
| }; |
| |
| DEFINE_NATIVE_ENTRY(Int32x4_select, 0, 3) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Int32x4, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, tv, arguments->NativeArgAt(1)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, fv, arguments->NativeArgAt(2)); |
| int32_t _maskX = self.x(); |
| int32_t _maskY = self.y(); |
| int32_t _maskZ = self.z(); |
| int32_t _maskW = self.w(); |
| // Extract floats and interpret them as masks. |
| float32_int32 tvx(tv.x()); |
| float32_int32 tvy(tv.y()); |
| float32_int32 tvz(tv.z()); |
| float32_int32 tvw(tv.w()); |
| float32_int32 fvx(fv.x()); |
| float32_int32 fvy(fv.y()); |
| float32_int32 fvz(fv.z()); |
| float32_int32 fvw(fv.w()); |
| // Perform select. |
| float32_int32 tempX((_maskX & tvx.u) | (~_maskX & fvx.u)); |
| float32_int32 tempY((_maskY & tvy.u) | (~_maskY & fvy.u)); |
| float32_int32 tempZ((_maskZ & tvz.u) | (~_maskZ & fvz.u)); |
| float32_int32 tempW((_maskW & tvw.u) | (~_maskW & fvw.u)); |
| return Float32x4::New(tempX.f, tempY.f, tempZ.f, tempW.f); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_fromDoubles, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Double, x, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Double, y, arguments->NativeArgAt(1)); |
| return Float64x2::New(x.value(), y.value()); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_splat, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Double, v, arguments->NativeArgAt(0)); |
| return Float64x2::New(v.value(), v.value()); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_zero, 0, 1) { |
| ASSERT( |
| TypeArguments::CheckedHandle(zone, arguments->NativeArgAt(0)).IsNull()); |
| return Float64x2::New(0.0, 0.0); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_fromFloat32x4, 0, 2) { |
| ASSERT( |
| TypeArguments::CheckedHandle(zone, arguments->NativeArgAt(0)).IsNull()); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, v, arguments->NativeArgAt(1)); |
| double _x = v.x(); |
| double _y = v.y(); |
| return Float64x2::New(_x, _y); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_add, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, other, arguments->NativeArgAt(1)); |
| double _x = self.x() + other.x(); |
| double _y = self.y() + other.y(); |
| return Float64x2::New(_x, _y); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_negate, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, self, arguments->NativeArgAt(0)); |
| double _x = -self.x(); |
| double _y = -self.y(); |
| return Float64x2::New(_x, _y); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_sub, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, other, arguments->NativeArgAt(1)); |
| double _x = self.x() - other.x(); |
| double _y = self.y() - other.y(); |
| return Float64x2::New(_x, _y); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_mul, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, other, arguments->NativeArgAt(1)); |
| double _x = self.x() * other.x(); |
| double _y = self.y() * other.y(); |
| return Float64x2::New(_x, _y); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_div, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, other, arguments->NativeArgAt(1)); |
| double _x = self.x() / other.x(); |
| double _y = self.y() / other.y(); |
| return Float64x2::New(_x, _y); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_scale, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Double, scale, arguments->NativeArgAt(1)); |
| double _s = scale.value(); |
| double _x = self.x() * _s; |
| double _y = self.y() * _s; |
| return Float64x2::New(_x, _y); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_abs, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, self, arguments->NativeArgAt(0)); |
| double _x = fabs(self.x()); |
| double _y = fabs(self.y()); |
| return Float64x2::New(_x, _y); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_clamp, 0, 3) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, lo, arguments->NativeArgAt(1)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, hi, arguments->NativeArgAt(2)); |
| // The order of the clamping must match the order of the optimized code: |
| // MAX(MIN(self, hi), lo). |
| double _x; |
| double _y; |
| |
| // ARM semantics are different from X86/X64 at an instruction level. Ensure |
| // that we match the semantics of the architecture in the C version. |
| #if defined(TARGET_ARCH_IA32) || defined(TARGET_ARCH_X64) |
| _x = self.x() < hi.x() ? self.x() : hi.x(); |
| _y = self.y() < hi.y() ? self.y() : hi.y(); |
| _x = lo.x() < _x ? _x : lo.x(); |
| _y = lo.y() < _y ? _y : lo.y(); |
| #else |
| _x = fmin(self.x(), hi.x()); |
| _y = fmin(self.y(), hi.y()); |
| _x = fmax(_x, lo.x()); |
| _y = fmax(_y, lo.y()); |
| #endif |
| return Float64x2::New(_x, _y); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_getX, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, self, arguments->NativeArgAt(0)); |
| return Double::New(self.x()); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_getY, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, self, arguments->NativeArgAt(0)); |
| return Double::New(self.y()); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_getSignMask, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, self, arguments->NativeArgAt(0)); |
| uint32_t mx = (bit_cast<uint64_t>(self.x()) & 0x8000000000000000LL) >> 63; |
| uint32_t my = (bit_cast<uint64_t>(self.y()) & 0x8000000000000000LL) >> 63; |
| uint32_t value = mx | (my << 1); |
| return Integer::New(value); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_setX, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Double, x, arguments->NativeArgAt(1)); |
| double _x = x.value(); |
| double _y = self.y(); |
| return Float64x2::New(_x, _y); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_setY, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Double, y, arguments->NativeArgAt(1)); |
| double _x = self.x(); |
| double _y = y.value(); |
| return Float64x2::New(_x, _y); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_min, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, other, arguments->NativeArgAt(1)); |
| double _x = self.x() < other.x() ? self.x() : other.x(); |
| double _y = self.y() < other.y() ? self.y() : other.y(); |
| return Float64x2::New(_x, _y); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_max, 0, 2) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, self, arguments->NativeArgAt(0)); |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, other, arguments->NativeArgAt(1)); |
| double _x = self.x() > other.x() ? self.x() : other.x(); |
| double _y = self.y() > other.y() ? self.y() : other.y(); |
| return Float64x2::New(_x, _y); |
| } |
| |
| DEFINE_NATIVE_ENTRY(Float64x2_sqrt, 0, 1) { |
| GET_NON_NULL_NATIVE_ARGUMENT(Float64x2, self, arguments->NativeArgAt(0)); |
| double _x = sqrt(self.x()); |
| double _y = sqrt(self.y()); |
| return Float64x2::New(_x, _y); |
| } |
| |
| } // namespace dart |