Simd128Float32, Simd128Mask, and Simd128Float32List additions for dart:scalarlist

sdk/lib/scalarlist/simd128.dart

+// 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.
+
+part of dart.scalarlist;
+
+/**
+ * Interface of Dart Simd128Float32x4 and operations.
+ * Simd128Float32x4 stores 4 32-bit floating point values in "lanes".
+ * The lanes are "x", "y", "z", and "w" respectively.
+ */
+abstract class Simd128Float32x4 {
+  external factory Simd128Float32x4(double x, double y, double z, double w);
+  external factory Simd128Float32x4.zero();
+
+  /// Addition operator.
+  Simd128Float32x4 operator+(Simd128Float32x4 other);

[srdjan, 2013/02/21 21:43:25]

Space between operator and its name (see int.dart). Here and below.

[Cutch, 2013/02/22 03:30:55]

Done.

+  /// Negate operator.
+  Simd128Float32x4 operator-();
+  /// Subtraction operator.
+  Simd128Float32x4 operator-(Simd128Float32x4 other);
+  /// Multiplication operator.
+  Simd128Float32x4 operator*(Simd128Float32x4 other);
+  /// Division operator.
+  Simd128Float32x4 operator/(Simd128Float32x4 other);
+
+  /// Relational less than.
+  Simd128Mask lessThan(Simd128Float32x4 other);
+  /// Relational less than or equal.
+  Simd128Mask lessThanOrEqual(Simd128Float32x4 other);
+  /// Relational greater than.
+  Simd128Mask greaterThan(Simd128Float32x4 other);
+  /// Relational greater than or equal.
+  Simd128Mask greaterThanOrEqual(Simd128Float32x4 other);
+  /// Relational equal.
+  Simd128Mask equal(Simd128Float32x4 other);
+  /// Relational not-equal.
+  Simd128Mask notEqual(Simd128Float32x4 other);
+
+  /// Returns a copy of [this] each lane being scaled by [s].
+  Simd128Float32x4 scale(double s);
+  /// Returns the absolute value of this [Simd128Float32].
+  Simd128Float32x4 abs();
+  /// Clamps [this] to be in the range [lowerLimit]-[upperLimit].
+  Simd128Float32x4 clamp(Simd128Float32x4 lowerLimit,
+                         Simd128Float32x4 upperLimit);
+
+  /// Extracted x value.
+  double get x;
+  /// Extracted y value.
+  double get y;
+  /// Extracted z value.
+  double get z;
+  /// Extracted w value.
+  double get w;
+
+  /// Returns a new [Simd128Float32x4] with [this]' x value in all four lanes.
+  Simd128Float32x4 get xxxx;
+  /// Returns a new [Simd128Float32x4] with [this]' y value in all four lanes.
+  Simd128Float32x4 get yyyy;
+  /// Returns a new [Simd128Float32x4] with [this]' z value in all four lanes.
+  Simd128Float32x4 get zzzz;
+  /// Returns a new [Simd128Float32x4] with [this]' w value in all four lanes.
+  Simd128Float32x4 get wwww;
+  // TODO(johnmccutchan): Add all 256 possible combinations.

[srdjan, 2013/02/21 21:43:25]

What would be better: pass an argument, e.g. a String, specifying how to fill
the lanes, or to write out all 256 possible combinations? The argument (String)
is compile time constant and can be recognized and optimized at compile time. 

[Cutch, 2013/02/22 03:30:55]

So long as it's a constant at the call site that would be fine. On SSE the
combination is embedded into the instruction and cannot change.

[kasperl, 2013/02/22 06:38:47]

That seems like asking for trouble though. We still have to implement a parser
for those strings just in case it isn't a constant so that's extra code to test
and write. Unfolding 256 combinations seems a little bit iffy too, but at least
it's dead simple.

+
+  /// Returns a new [Simd128Float32x4] copied from [this] with a new x value.
+  Simd128Float32x4 withX(double x);
+  /// Returns a new [Simd128Float32x4] copied from [this] with a new y value.
+  Simd128Float32x4 withY(double y);
+  /// Returns a new [Simd128Float32x4] copied from [this] with a new z value.
+  Simd128Float32x4 withZ(double z);
+  /// Returns a new [Simd128Float32x4] copied from [this] with a new w value.
+  Simd128Float32x4 withW(double w);
+
+  /// Returns the lane-wise minimum value in [this] or [other].
+  Simd128Float32x4 min(Simd128Float32x4 other);
+
+  /// Returns the lane-wise maximum value in [this] or [other].
+  Simd128Float32x4 max(Simd128Float32x4 other);
+
+  /// Returns the square root of [this].
+  Simd128Float32x4 sqrt();
+
+  /// Returns the reciprocal of [this].
+  Simd128Float32x4 reciprocal();
+
+  /// Returns the square root of the reciprocal of [this].
+  Simd128Float32x4 reciprocalSqrt();
+
+  /// Returns a bit-wise copy of [this] as a [Simd128Mask].
+  Simd128Mask toSimd128Mask();
+}
+
+/**
+ * Interface of Dart Simd128Mask and operations.
+ * Simd128Mask stores 4 32-bit bit-masks in "lanes".
+ * The lanes are "x", "y", "z", and "w" respectively.
+ */
+abstract class Simd128Mask {
+  external factory Simd128Mask(int x, int y, int z, int w);
+  external factory Simd128Mask.bool(bool x, bool y, bool z, bool w);
+
+  /// The bit-wise or operator.
+  Simd128Mask operator|(Simd128Mask other);
+  /// The bit-wise and operator.
+  Simd128Mask operator&(Simd128Mask other);
+  /// The bit-wise xor operator.
+  Simd128Mask operator^(Simd128Mask other);
+
+  /// Extract 32-bit mask from x lane.
+  int get x;
+  /// Extract 32-bit mask from y lane.
+  int get y;
+  /// Extract 32-bit mask from z lane.
+  int get z;
+  /// Extract 32-bit mask from w lane.
+  int get w;
+
+  /// Returns a new [Simd128Mask] copied from [this] with a new x value.
+  Simd128Mask withX(int x);
+  /// Returns a new [Simd128Mask] copied from [this] with a new y value.
+  Simd128Mask withY(int y);
+  /// Returns a new [Simd128Mask] copied from [this] with a new z value.
+  Simd128Mask withZ(int z);
+  /// Returns a new [Simd128Mask] copied from [this] with a new w value.
+  Simd128Mask withW(int w);
+
+  /// Extracted x value. Returns false for 0, true for any other value.
+  bool get flagX;
+  /// Extracted y value. Returns false for 0, true for any other value.
+  bool get flagY;
+  /// Extracted z value. Returns false for 0, true for any other value.
+  bool get flagZ;
+  /// Extracted w value. Returns false for 0, true for any other value.
+  bool get flagW;
+
+  /// Returns a new [Simd128Mask] copied from [this] with a new x value.
+  Simd128Mask withFlagX(bool x);
+  /// Returns a new [Simd128Mask] copied from [this] with a new y value.
+  Simd128Mask withFlagY(bool y);
+  /// Returns a new [Simd128Mask] copied from [this] with a new z value.
+  Simd128Mask withFlagZ(bool z);
+  /// Returns a new [Simd128Mask] copied from [this] with a new w value.
+  Simd128Mask withFlagW(bool w);
+
+  /// Merge [trueValue] and [falseValue] based on [this]' bit mask:
+  /// Select bit from [trueValue] when bit in [this] is on.
+  /// Select bit from [falseValue] when bit in [this] is off.
+  Simd128Float32x4 select(Simd128Float32x4 trueValue,
+                          Simd128Float32x4 falseValue);
+
+  /// Returns a bit-wise copy of [this] as a [Simd128Float32x4].
+  Simd128Float32x4 toSimd128Float32x4();
+}
+
+/**
+ * A fixed-length list of Simd128Float32x4 numbers that is viewable as a
+ * [ByteArray]. For long lists, this implementation will be considerably more
+ * space- and time-efficient than the default [List] implementation.
+ */
+abstract class Simd128Float32x4List implements List<Simd128Float32x4>,
+    ByteArrayViewable {
+  /**
+   * Creates a [Simd128Float32List] of the specified length (in elements),
+   * all of whose elements are initially zero.
+   */
+  external factory Simd128Float32x4List(int length);
+
+  /**
+   * Creates a [Simd128Float32x4List] _view_ of the specified region in the
+   * specified byte [array]. Changes in the [Simd128Float32x4List] will be
+   * visible in the byte array and vice versa. If the [start] index of the
+   * region is not specified, it defaults to zero (the first byte in the byte
+   * array). If the length is not specified, it defaults to null, which
+   * indicates that the view extends to the end of the byte array.
+   *
+   * Throws [ArgumentError] if the length of the specified region is not
+   * divisible by 16 (the size of a "Simd128Float32x4" in bytes), or if the
+   * [start] of the region is not divisible by 16. If, however, [array] is a
+   * view of another byte array, this constructor will throw [ArgumentError]
+   * if the implicit starting position in the "ultimately backing" byte array
+   * is not divisible by 16. In plain terms, this constructor throws
+   * [ArgumentError] if the specified region does not contain an integral
+   * number of "Simd128Float32x4s," or if it is not "Simd128Float32x4-aligned."
+   */
+  external factory Simd128Float32x4List.view(ByteArray array,
+                                             [int start = 0, int length]);
+}