SIMD shuffle API changes

runtime/lib/simd128.cc

 // 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"
@@ skipping 262 matching lines @@
   }
   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_withZWInXY, 2) {
+DEFINE_NATIVE_ENTRY(Float32x4_shuffleMix, 3) {
   GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0));
-  GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, other, arguments->NativeArgAt(1));
-  float _x = other.z();
-  float _y = other.w();
-  float _z = self.z();
-  float _w = self.w();
+  GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, zw, arguments->NativeArgAt(1));

[zra, 2013/10/30 17:39:27]

Just a bit confused about naming, I think. This is called zw, but could the
shuffleMix also take the x or y elements for the final result?

[Cutch, 2013/10/31 01:43:55]

name of "zw" because the resulting ZW lanes will be pulled from it. Renamed it
to "other".

+  GET_NON_NULL_NATIVE_ARGUMENT(Integer, mask, arguments->NativeArgAt(2));
+  int64_t m = mask.AsInt64Value();
+  if (m < 0 || m > 255) {
+    const String& error = String::Handle(
+      String::NewFormatted("mask (%" Pd64 ") must be in the range [0..256)",
+                           m));
+    const Array& args = Array::Handle(Array::New(1));
+    args.SetAt(0, error);
+    Exceptions::ThrowByType(Exceptions::kRange, args);
+  }
+  float data[4] = { self.x(), self.y(), self.z(), self.w() };
+  float zw_data[4] = { zw.x(), zw.y(), zw.z(), zw.w() };
+  float _x = data[m & 0x3];
+  float _y = data[(m >> 2) & 0x3];
+  float _z = zw_data[(m >> 4) & 0x3];
+  float _w = zw_data[(m >> 6) & 0x3];
   return Float32x4::New(_x, _y, _z, _w);
 }
 
-
-DEFINE_NATIVE_ENTRY(Float32x4_interleaveXY, 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();
-  float _y = other.x();
-  float _z = self.y();
-  float _w = other.y();
-  return Float32x4::New(_x, _y, _z, _w);
-}
-
-
-DEFINE_NATIVE_ENTRY(Float32x4_interleaveZW, 2) {
-  GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0));
-  GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, other, arguments->NativeArgAt(1));
-  float _x = self.z();
-  float _y = other.z();
-  float _z = self.w();
-  float _w = other.w();
-  return Float32x4::New(_x, _y, _z, _w);
-}
-
-
-DEFINE_NATIVE_ENTRY(Float32x4_interleaveXYPairs, 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();
-  float _y = self.y();
-  float _z = other.x();
-  float _w = other.y();
-  return Float32x4::New(_x, _y, _z, _w);
-}
-
-
-DEFINE_NATIVE_ENTRY(Float32x4_interleaveZWPairs, 2) {
-  GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, self, arguments->NativeArgAt(0));
-  GET_NON_NULL_NATIVE_ARGUMENT(Float32x4, other, arguments->NativeArgAt(1));
-  float _x = self.z();
-  float _y = self.w();
-  float _z = other.z();
-  float _w = other.w();
-  return Float32x4::New(_x, _y, _z, _w);
-}
-
 
 DEFINE_NATIVE_ENTRY(Float32x4_setX, 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);
 }
@@ skipping 198 matching lines @@
 }
 
 
 DEFINE_NATIVE_ENTRY(Uint32x4_getW, 1) {
   GET_NON_NULL_NATIVE_ARGUMENT(Uint32x4, self, arguments->NativeArgAt(0));
   uint32_t value = self.w();
   return Integer::New(value);
 }
 
 
+DEFINE_NATIVE_ENTRY(Uint32x4_shuffle, 2) {
+  GET_NON_NULL_NATIVE_ARGUMENT(Uint32x4, self, arguments->NativeArgAt(0));
+  GET_NON_NULL_NATIVE_ARGUMENT(Integer, mask, arguments->NativeArgAt(1));
+  int64_t m = mask.AsInt64Value();
+  if (m < 0 || m > 255) {
+    const String& error = String::Handle(
+      String::NewFormatted("mask (%" Pd64 ") must be in the range [0..256)",
+                           m));
+    const Array& args = Array::Handle(Array::New(1));
+    args.SetAt(0, error);
+    Exceptions::ThrowByType(Exceptions::kRange, args);
+  }
+  uint32_t data[4] = { self.x(), self.y(), self.z(), self.w() };
+  uint32_t _x = data[m & 0x3];
+  uint32_t _y = data[(m >> 2) & 0x3];
+  uint32_t _z = data[(m >> 4) & 0x3];
+  uint32_t _w = data[(m >> 6) & 0x3];
+  return Uint32x4::New(_x, _y, _z, _w);
+}
+
+
+DEFINE_NATIVE_ENTRY(Uint32x4_shuffleMix, 3) {
+  GET_NON_NULL_NATIVE_ARGUMENT(Uint32x4, self, arguments->NativeArgAt(0));
+  GET_NON_NULL_NATIVE_ARGUMENT(Uint32x4, zw, arguments->NativeArgAt(1));

[zra, 2013/10/30 17:39:27]

Same question here.

[Cutch, 2013/10/31 01:43:55]

Renamed to other.

+  GET_NON_NULL_NATIVE_ARGUMENT(Integer, mask, arguments->NativeArgAt(2));
+  int64_t m = mask.AsInt64Value();
+  if (m < 0 || m > 255) {
+    const String& error = String::Handle(
+      String::NewFormatted("mask (%" Pd64 ") must be in the range [0..256)",
+                           m));
+    const Array& args = Array::Handle(Array::New(1));
+    args.SetAt(0, error);
+    Exceptions::ThrowByType(Exceptions::kRange, args);
+  }
+  uint32_t data[4] = { self.x(), self.y(), self.z(), self.w() };
+  uint32_t zw_data[4] = { zw.x(), zw.y(), zw.z(), zw.w() };
+  uint32_t _x = data[m & 0x3];
+  uint32_t _y = data[(m >> 2) & 0x3];
+  uint32_t _z = zw_data[(m >> 4) & 0x3];
+  uint32_t _w = zw_data[(m >> 6) & 0x3];
+  return Uint32x4::New(_x, _y, _z, _w);
+}
+
+
 DEFINE_NATIVE_ENTRY(Uint32x4_setX, 2) {
   GET_NON_NULL_NATIVE_ARGUMENT(Uint32x4, self, arguments->NativeArgAt(0));
   GET_NON_NULL_NATIVE_ARGUMENT(Integer, x, arguments->NativeArgAt(1));
   uint32_t _x = static_cast<uint32_t>(x.AsInt64Value() & 0xFFFFFFFF);
   uint32_t _y = self.y();
   uint32_t _z = self.z();
   uint32_t _w = self.w();
   return Uint32x4::New(_x, _y, _z, _w);
 }
 
@@ skipping 141 matching lines @@
   // Perform select.
   float32_uint32 tempX((_maskX & tvx.u) | (~_maskX & fvx.u));
   float32_uint32 tempY((_maskY & tvy.u) | (~_maskY & fvy.u));
   float32_uint32 tempZ((_maskZ & tvz.u) | (~_maskZ & fvz.u));
   float32_uint32 tempW((_maskW & tvw.u) | (~_maskW & fvw.u));
   return Float32x4::New(tempX.f, tempY.f, tempZ.f, tempW.f);
 }
 
 
 }  // namespace dart