Fixing bug

Babylon/Math/babylon.math.js

@@ -631,7 +631,7 @@ var BABYLON;
             result.z = rz / rw;
         };
         Vector3.TransformCoordinatesToRefSIMD = function (vector, transformation, result) {
-            var v = SIMD.float32x4.loadXYZ(vector, 0);
+            var v = SIMD.float32x4(vector.x, vector.y, vector.z, 0);
             var m0 = SIMD.float32x4.load(transformation.m, 0);
             var m1 = SIMD.float32x4.load(transformation.m, 4);
             var m2 = SIMD.float32x4.load(transformation.m, 8);
@@ -1393,6 +1393,112 @@ var BABYLON;
             other.m[15] = (((l1 * l36) - (l2 * l38)) + (l3 * l39)) * l27;
             return this;
         };
+        Matrix.prototype.invertToRefSIMD = function (other) {
+            var src = this.m;
+            var dest = other.m;
+            var row0, row1, row2, row3;
+            var tmp1;
+            var minor0, minor1, minor2, minor3;
+            var det;
+            // Load the 4 rows
+            var src0 = SIMD.float32x4.load(src, 0);
+            var src1 = SIMD.float32x4.load(src, 4);
+            var src2 = SIMD.float32x4.load(src, 8);
+            var src3 = SIMD.float32x4.load(src, 12);
+            // Transpose the source matrix.  Sort of.  Not a true transpose operation
+            tmp1 = SIMD.float32x4.shuffle(src0, src1, 0, 1, 4, 5);
+            row1 = SIMD.float32x4.shuffle(src2, src3, 0, 1, 4, 5);
+            row0 = SIMD.float32x4.shuffle(tmp1, row1, 0, 2, 4, 6);
+            row1 = SIMD.float32x4.shuffle(row1, tmp1, 1, 3, 5, 7);
+            tmp1 = SIMD.float32x4.shuffle(src0, src1, 2, 3, 6, 7);
+            row3 = SIMD.float32x4.shuffle(src2, src3, 2, 3, 6, 7);
+            row2 = SIMD.float32x4.shuffle(tmp1, row3, 0, 2, 4, 6);
+            row3 = SIMD.float32x4.shuffle(row3, tmp1, 1, 3, 5, 7);
+            // This is a true transposition, but it will lead to an incorrect result
+            //tmp1 = SIMD.float32x4.shuffle(src0, src1, 0, 1, 4, 5);
+            //tmp2 = SIMD.float32x4.shuffle(src2, src3, 0, 1, 4, 5);
+            //row0  = SIMD.float32x4.shuffle(tmp1, tmp2, 0, 2, 4, 6);
+            //row1  = SIMD.float32x4.shuffle(tmp1, tmp2, 1, 3, 5, 7);
+            //tmp1 = SIMD.float32x4.shuffle(src0, src1, 2, 3, 6, 7);
+            //tmp2 = SIMD.float32x4.shuffle(src2, src3, 2, 3, 6, 7);
+            //row2  = SIMD.float32x4.shuffle(tmp1, tmp2, 0, 2, 4, 6);
+            //row3  = SIMD.float32x4.shuffle(tmp1, tmp2, 1, 3, 5, 7);
+            // ----
+            tmp1 = SIMD.float32x4.mul(row2, row3);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 1, 0, 3, 2); // 0xB1 = 10110001
+            minor0 = SIMD.float32x4.mul(row1, tmp1);
+            minor1 = SIMD.float32x4.mul(row0, tmp1);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 2, 3, 0, 1); // 0x4E = 01001110
+            minor0 = SIMD.float32x4.sub(SIMD.float32x4.mul(row1, tmp1), minor0);
+            minor1 = SIMD.float32x4.sub(SIMD.float32x4.mul(row0, tmp1), minor1);
+            minor1 = SIMD.float32x4.swizzle(minor1, 2, 3, 0, 1); // 0x4E = 01001110
+            // ----
+            tmp1 = SIMD.float32x4.mul(row1, row2);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 1, 0, 3, 2); // 0xB1 = 10110001
+            minor0 = SIMD.float32x4.add(SIMD.float32x4.mul(row3, tmp1), minor0);
+            minor3 = SIMD.float32x4.mul(row0, tmp1);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 2, 3, 0, 1); // 0x4E = 01001110
+            minor0 = SIMD.float32x4.sub(minor0, SIMD.float32x4.mul(row3, tmp1));
+            minor3 = SIMD.float32x4.sub(SIMD.float32x4.mul(row0, tmp1), minor3);
+            minor3 = SIMD.float32x4.swizzle(minor3, 2, 3, 0, 1); // 0x4E = 01001110
+            // ----
+            tmp1 = SIMD.float32x4.mul(SIMD.float32x4.swizzle(row1, 2, 3, 0, 1), row3); // 0x4E = 01001110
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 1, 0, 3, 2); // 0xB1 = 10110001
+            row2 = SIMD.float32x4.swizzle(row2, 2, 3, 0, 1); // 0x4E = 01001110
+            minor0 = SIMD.float32x4.add(SIMD.float32x4.mul(row2, tmp1), minor0);
+            minor2 = SIMD.float32x4.mul(row0, tmp1);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 2, 3, 0, 1); // 0x4E = 01001110
+            minor0 = SIMD.float32x4.sub(minor0, SIMD.float32x4.mul(row2, tmp1));
+            minor2 = SIMD.float32x4.sub(SIMD.float32x4.mul(row0, tmp1), minor2);
+            minor2 = SIMD.float32x4.swizzle(minor2, 2, 3, 0, 1); // 0x4E = 01001110
+            // ----
+            tmp1 = SIMD.float32x4.mul(row0, row1);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 1, 0, 3, 2); // 0xB1 = 10110001
+            minor2 = SIMD.float32x4.add(SIMD.float32x4.mul(row3, tmp1), minor2);
+            minor3 = SIMD.float32x4.sub(SIMD.float32x4.mul(row2, tmp1), minor3);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 2, 3, 0, 1); // 0x4E = 01001110
+            minor2 = SIMD.float32x4.sub(SIMD.float32x4.mul(row3, tmp1), minor2);
+            minor3 = SIMD.float32x4.sub(minor3, SIMD.float32x4.mul(row2, tmp1));
+            // ----
+            tmp1 = SIMD.float32x4.mul(row0, row3);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 1, 0, 3, 2); // 0xB1 = 10110001
+            minor1 = SIMD.float32x4.sub(minor1, SIMD.float32x4.mul(row2, tmp1));
+            minor2 = SIMD.float32x4.add(SIMD.float32x4.mul(row1, tmp1), minor2);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 2, 3, 0, 1); // 0x4E = 01001110
+            minor1 = SIMD.float32x4.add(SIMD.float32x4.mul(row2, tmp1), minor1);
+            minor2 = SIMD.float32x4.sub(minor2, SIMD.float32x4.mul(row1, tmp1));
+            // ----
+            tmp1 = SIMD.float32x4.mul(row0, row2);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 1, 0, 3, 2); // 0xB1 = 10110001
+            minor1 = SIMD.float32x4.add(SIMD.float32x4.mul(row3, tmp1), minor1);
+            minor3 = SIMD.float32x4.sub(minor3, SIMD.float32x4.mul(row1, tmp1));
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 2, 3, 0, 1); // 0x4E = 01001110
+            minor1 = SIMD.float32x4.sub(minor1, SIMD.float32x4.mul(row3, tmp1));
+            minor3 = SIMD.float32x4.add(SIMD.float32x4.mul(row1, tmp1), minor3);
+            // Compute determinant
+            det = SIMD.float32x4.mul(row0, minor0);
+            det = SIMD.float32x4.add(SIMD.float32x4.swizzle(det, 2, 3, 0, 1), det); // 0x4E = 01001110
+            det = SIMD.float32x4.add(SIMD.float32x4.swizzle(det, 1, 0, 3, 2), det); // 0xB1 = 10110001
+            tmp1 = SIMD.float32x4.reciprocal(det);
+            det = SIMD.float32x4.sub(SIMD.float32x4.add(tmp1, tmp1), SIMD.float32x4.mul(det, SIMD.float32x4.mul(tmp1, tmp1)));
+            det = SIMD.float32x4.swizzle(det, 0, 0, 0, 0);
+            // These shuffles aren't necessary if the faulty transposition is done
+            // up at the top of this function.
+            //minor0 = SIMD.float32x4.swizzle(minor0, 2, 1, 0, 3);
+            //minor1 = SIMD.float32x4.swizzle(minor1, 2, 1, 0, 3);
+            //minor2 = SIMD.float32x4.swizzle(minor2, 2, 1, 0, 3);
+            //minor3 = SIMD.float32x4.swizzle(minor3, 2, 1, 0, 3);
+            // Compute final values by multiplying with 1/det
+            minor0 = SIMD.float32x4.mul(det, minor0);
+            minor1 = SIMD.float32x4.mul(det, minor1);
+            minor2 = SIMD.float32x4.mul(det, minor2);
+            minor3 = SIMD.float32x4.mul(det, minor3);
+            SIMD.float32x4.store(dest, 0, minor0);
+            SIMD.float32x4.store(dest, 4, minor1);
+            SIMD.float32x4.store(dest, 8, minor2);
+            SIMD.float32x4.store(dest, 12, minor3);
+            return this;
+        };
         Matrix.prototype.setTranslation = function (vector3) {
             this.m[12] = vector3.x;
             this.m[13] = vector3.y;
@@ -1747,6 +1853,66 @@ var BABYLON;
             var ez = -Vector3.Dot(this._zAxis, eye);
             return Matrix.FromValuesToRef(this._xAxis.x, this._yAxis.x, this._zAxis.x, 0, this._xAxis.y, this._yAxis.y, this._zAxis.y, 0, this._xAxis.z, this._yAxis.z, this._zAxis.z, 0, ex, ey, ez, 1, result);
         };
+        Matrix.LookAtLHToRefSIMD = function (eyeRef, targetRef, upRef, result) {
+            var out = result.m;
+            var center = SIMD.float32x4(targetRef.x, targetRef.y, targetRef.z, 0);
+            var eye = SIMD.float32x4(eyeRef.x, eyeRef.y, eyeRef.z, 0);
+            var up = SIMD.float32x4(upRef.x, upRef.y, upRef.z, 0);
+            // cc.kmVec3Subtract(f, pCenter, pEye);
+            var f = SIMD.float32x4.sub(center, eye);
+            // cc.kmVec3Normalize(f, f);    
+            var tmp = SIMD.float32x4.mul(f, f);
+            tmp = SIMD.float32x4.add(tmp, SIMD.float32x4.add(SIMD.float32x4.swizzle(tmp, 1, 2, 0, 3), SIMD.float32x4.swizzle(tmp, 2, 0, 1, 3)));
+            f = SIMD.float32x4.mul(f, SIMD.float32x4.reciprocalSqrt(tmp));
+            // cc.kmVec3Assign(up, pUp);
+            // cc.kmVec3Normalize(up, up);
+            tmp = SIMD.float32x4.mul(up, up);
+            tmp = SIMD.float32x4.add(tmp, SIMD.float32x4.add(SIMD.float32x4.swizzle(tmp, 1, 2, 0, 3), SIMD.float32x4.swizzle(tmp, 2, 0, 1, 3)));
+            up = SIMD.float32x4.mul(up, SIMD.float32x4.reciprocalSqrt(tmp));
+            // cc.kmVec3Cross(s, f, up);
+            var s = SIMD.float32x4.sub(SIMD.float32x4.mul(SIMD.float32x4.swizzle(f, 1, 2, 0, 3), SIMD.float32x4.swizzle(up, 2, 0, 1, 3)), SIMD.float32x4.mul(SIMD.float32x4.swizzle(f, 2, 0, 1, 3), SIMD.float32x4.swizzle(up, 1, 2, 0, 3)));
+            // cc.kmVec3Normalize(s, s);
+            tmp = SIMD.float32x4.mul(s, s);
+            tmp = SIMD.float32x4.add(tmp, SIMD.float32x4.add(SIMD.float32x4.swizzle(tmp, 1, 2, 0, 3), SIMD.float32x4.swizzle(tmp, 2, 0, 1, 3)));
+            s = SIMD.float32x4.mul(s, SIMD.float32x4.reciprocalSqrt(tmp));
+            // cc.kmVec3Cross(u, s, f);
+            var u = SIMD.float32x4.sub(SIMD.float32x4.mul(SIMD.float32x4.swizzle(s, 1, 2, 0, 3), SIMD.float32x4.swizzle(f, 2, 0, 1, 3)), SIMD.float32x4.mul(SIMD.float32x4.swizzle(s, 2, 0, 1, 3), SIMD.float32x4.swizzle(f, 1, 2, 0, 3)));
+            // cc.kmVec3Normalize(s, s);
+            tmp = SIMD.float32x4.mul(s, s);
+            tmp = SIMD.float32x4.add(tmp, SIMD.float32x4.add(SIMD.float32x4.swizzle(tmp, 1, 2, 0, 3), SIMD.float32x4.swizzle(tmp, 2, 0, 1, 3)));
+            s = SIMD.float32x4.mul(s, SIMD.float32x4.reciprocalSqrt(tmp));
+            //cc.kmMat4Identity(pOut);
+            //pOut.mat[0] = s.x;
+            //pOut.mat[4] = s.y;
+            //pOut.mat[8] = s.z;
+            //pOut.mat[1] = u.x;
+            //pOut.mat[5] = u.y;
+            //pOut.mat[9] = u.z;
+            //pOut.mat[2] = -f.x;
+            //pOut.mat[6] = -f.y;
+            //pOut.mat[10] = -f.z;
+            var zero = SIMD.float32x4.splat(0.0);
+            s = SIMD.float32x4.neg(s);
+            var tmp01 = SIMD.float32x4.shuffle(s, u, 0, 1, 4, 5);
+            var tmp23 = SIMD.float32x4.shuffle(f, zero, 0, 1, 4, 5);
+            var a0 = SIMD.float32x4.shuffle(tmp01, tmp23, 0, 2, 4, 6);
+            var a1 = SIMD.float32x4.shuffle(tmp01, tmp23, 1, 3, 5, 7);
+            var tmp01 = SIMD.float32x4.shuffle(s, u, 2, 3, 6, 7);
+            var tmp23 = SIMD.float32x4.shuffle(f, zero, 2, 3, 6, 7);
+            var a2 = SIMD.float32x4.shuffle(tmp01, tmp23, 0, 2, 4, 6);
+            var a3 = SIMD.float32x4(0.0, 0.0, 0.0, 1.0);
+            // cc.kmMat4Translation(translate, -pEye.x, -pEye.y, -pEye.z);
+            var b0 = SIMD.float32x4(1.0, 0.0, 0.0, 0.0);
+            var b1 = SIMD.float32x4(0.0, 1.0, 0.0, 0.0);
+            var b2 = SIMD.float32x4(0.0, 0.0, 1.0, 0.0);
+            var b3 = SIMD.float32x4.neg(eye);
+            b3 = SIMD.float32x4.withW(b3, 1.0);
+            // cc.kmMat4Multiply(pOut, pOut, translate);
+            SIMD.float32x4.store(out, 0, SIMD.float32x4.add(SIMD.float32x4.mul(SIMD.float32x4.swizzle(b0, 0, 0, 0, 0), a0), SIMD.float32x4.add(SIMD.float32x4.mul(SIMD.float32x4.swizzle(b0, 1, 1, 1, 1), a1), SIMD.float32x4.add(SIMD.float32x4.mul(SIMD.float32x4.swizzle(b0, 2, 2, 2, 2), a2), SIMD.float32x4.mul(SIMD.float32x4.swizzle(b0, 3, 3, 3, 3), a3)))));
+            SIMD.float32x4.store(out, 4, SIMD.float32x4.add(SIMD.float32x4.mul(SIMD.float32x4.swizzle(b1, 0, 0, 0, 0), a0), SIMD.float32x4.add(SIMD.float32x4.mul(SIMD.float32x4.swizzle(b1, 1, 1, 1, 1), a1), SIMD.float32x4.add(SIMD.float32x4.mul(SIMD.float32x4.swizzle(b1, 2, 2, 2, 2), a2), SIMD.float32x4.mul(SIMD.float32x4.swizzle(b1, 3, 3, 3, 3), a3)))));
+            SIMD.float32x4.store(out, 8, SIMD.float32x4.add(SIMD.float32x4.mul(SIMD.float32x4.swizzle(b2, 0, 0, 0, 0), a0), SIMD.float32x4.add(SIMD.float32x4.mul(SIMD.float32x4.swizzle(b2, 1, 1, 1, 1), a1), SIMD.float32x4.add(SIMD.float32x4.mul(SIMD.float32x4.swizzle(b2, 2, 2, 2, 2), a2), SIMD.float32x4.mul(SIMD.float32x4.swizzle(b2, 3, 3, 3, 3), a3)))));
+            SIMD.float32x4.store(out, 12, SIMD.float32x4.add(SIMD.float32x4.mul(SIMD.float32x4.swizzle(b3, 0, 0, 0, 0), a0), SIMD.float32x4.add(SIMD.float32x4.mul(SIMD.float32x4.swizzle(b3, 1, 1, 1, 1), a1), SIMD.float32x4.add(SIMD.float32x4.mul(SIMD.float32x4.swizzle(b3, 2, 2, 2, 2), a2), SIMD.float32x4.mul(SIMD.float32x4.swizzle(b3, 3, 3, 3, 3), a3)))));
+        };
         Matrix.OrthoLH = function (width, height, znear, zfar) {
             var matrix = Matrix.Zero();
             Matrix.OrthoLHToRef(width, height, znear, zfar, matrix);
@@ -2571,6 +2737,8 @@ var BABYLON;
     if (window.SIMD !== undefined) {
         // Replace functions
         Matrix.prototype.multiplyToArray = Matrix.prototype.multiplyToArraySIMD;
+        Matrix.prototype.invertToRef = Matrix.prototype.invertToRefSIMD;
+        Matrix.LookAtLHToRef = Matrix.LookAtLHToRefSIMD;
         Vector3.TransformCoordinatesToRef = Vector3.TransformCoordinatesToRefSIMD;
         Vector3.TransformCoordinatesFromFloatsToRef = Vector3.TransformCoordinatesFromFloatsToRefSIMD;
     }

Babylon/Math/babylon.math.ts

@@ -781,7 +781,7 @@
         }
 
         public static TransformCoordinatesToRefSIMD(vector: Vector3, transformation: Matrix, result: Vector3): void {
-            var v = SIMD.float32x4.loadXYZ(vector, 0);
+            var v = SIMD.float32x4(vector.x, vector.y, vector.z, 0);
             var m0 = SIMD.float32x4.load(transformation.m, 0);
             var m1 = SIMD.float32x4.load(transformation.m, 4);
             var m2 = SIMD.float32x4.load(transformation.m, 8);
@@ -1730,6 +1730,131 @@
             return this;
         }
 
+        public invertToRefSIMD(other: Matrix): Matrix {
+            var src = this.m;
+            var dest = other.m;
+            var row0, row1, row2, row3;
+            var tmp1;
+            var minor0, minor1, minor2, minor3;
+            var det;
+
+            // Load the 4 rows
+            var src0 = SIMD.float32x4.load(src, 0);
+            var src1 = SIMD.float32x4.load(src, 4);
+            var src2 = SIMD.float32x4.load(src, 8);
+            var src3 = SIMD.float32x4.load(src, 12);
+
+            // Transpose the source matrix.  Sort of.  Not a true transpose operation
+
+            tmp1 = SIMD.float32x4.shuffle(src0, src1, 0, 1, 4, 5);
+            row1 = SIMD.float32x4.shuffle(src2, src3, 0, 1, 4, 5);
+            row0 = SIMD.float32x4.shuffle(tmp1, row1, 0, 2, 4, 6);
+            row1 = SIMD.float32x4.shuffle(row1, tmp1, 1, 3, 5, 7);
+
+            tmp1 = SIMD.float32x4.shuffle(src0, src1, 2, 3, 6, 7);
+            row3 = SIMD.float32x4.shuffle(src2, src3, 2, 3, 6, 7);
+            row2 = SIMD.float32x4.shuffle(tmp1, row3, 0, 2, 4, 6);
+            row3 = SIMD.float32x4.shuffle(row3, tmp1, 1, 3, 5, 7);
+
+            // This is a true transposition, but it will lead to an incorrect result
+
+            //tmp1 = SIMD.float32x4.shuffle(src0, src1, 0, 1, 4, 5);
+            //tmp2 = SIMD.float32x4.shuffle(src2, src3, 0, 1, 4, 5);
+            //row0  = SIMD.float32x4.shuffle(tmp1, tmp2, 0, 2, 4, 6);
+            //row1  = SIMD.float32x4.shuffle(tmp1, tmp2, 1, 3, 5, 7);
+
+            //tmp1 = SIMD.float32x4.shuffle(src0, src1, 2, 3, 6, 7);
+            //tmp2 = SIMD.float32x4.shuffle(src2, src3, 2, 3, 6, 7);
+            //row2  = SIMD.float32x4.shuffle(tmp1, tmp2, 0, 2, 4, 6);
+            //row3  = SIMD.float32x4.shuffle(tmp1, tmp2, 1, 3, 5, 7);
+
+            // ----
+            tmp1 = SIMD.float32x4.mul(row2, row3);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 1, 0, 3, 2); // 0xB1 = 10110001
+            minor0 = SIMD.float32x4.mul(row1, tmp1);
+            minor1 = SIMD.float32x4.mul(row0, tmp1);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 2, 3, 0, 1); // 0x4E = 01001110
+            minor0 = SIMD.float32x4.sub(SIMD.float32x4.mul(row1, tmp1), minor0);
+            minor1 = SIMD.float32x4.sub(SIMD.float32x4.mul(row0, tmp1), minor1);
+            minor1 = SIMD.float32x4.swizzle(minor1, 2, 3, 0, 1); // 0x4E = 01001110
+
+            // ----
+            tmp1 = SIMD.float32x4.mul(row1, row2);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 1, 0, 3, 2); // 0xB1 = 10110001
+            minor0 = SIMD.float32x4.add(SIMD.float32x4.mul(row3, tmp1), minor0);
+            minor3 = SIMD.float32x4.mul(row0, tmp1);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 2, 3, 0, 1); // 0x4E = 01001110
+            minor0 = SIMD.float32x4.sub(minor0, SIMD.float32x4.mul(row3, tmp1));
+            minor3 = SIMD.float32x4.sub(SIMD.float32x4.mul(row0, tmp1), minor3);
+            minor3 = SIMD.float32x4.swizzle(minor3, 2, 3, 0, 1); // 0x4E = 01001110
+
+            // ----
+            tmp1 = SIMD.float32x4.mul(SIMD.float32x4.swizzle(row1, 2, 3, 0, 1), row3); // 0x4E = 01001110
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 1, 0, 3, 2); // 0xB1 = 10110001
+            row2 = SIMD.float32x4.swizzle(row2, 2, 3, 0, 1);  // 0x4E = 01001110
+            minor0 = SIMD.float32x4.add(SIMD.float32x4.mul(row2, tmp1), minor0);
+            minor2 = SIMD.float32x4.mul(row0, tmp1);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 2, 3, 0, 1); // 0x4E = 01001110
+            minor0 = SIMD.float32x4.sub(minor0, SIMD.float32x4.mul(row2, tmp1));
+            minor2 = SIMD.float32x4.sub(SIMD.float32x4.mul(row0, tmp1), minor2);
+            minor2 = SIMD.float32x4.swizzle(minor2, 2, 3, 0, 1); // 0x4E = 01001110
+
+            // ----
+            tmp1 = SIMD.float32x4.mul(row0, row1);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 1, 0, 3, 2); // 0xB1 = 10110001
+            minor2 = SIMD.float32x4.add(SIMD.float32x4.mul(row3, tmp1), minor2);
+            minor3 = SIMD.float32x4.sub(SIMD.float32x4.mul(row2, tmp1), minor3);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 2, 3, 0, 1); // 0x4E = 01001110
+            minor2 = SIMD.float32x4.sub(SIMD.float32x4.mul(row3, tmp1), minor2);
+            minor3 = SIMD.float32x4.sub(minor3, SIMD.float32x4.mul(row2, tmp1));
+
+            // ----
+            tmp1 = SIMD.float32x4.mul(row0, row3);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 1, 0, 3, 2); // 0xB1 = 10110001
+            minor1 = SIMD.float32x4.sub(minor1, SIMD.float32x4.mul(row2, tmp1));
+            minor2 = SIMD.float32x4.add(SIMD.float32x4.mul(row1, tmp1), minor2);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 2, 3, 0, 1); // 0x4E = 01001110
+            minor1 = SIMD.float32x4.add(SIMD.float32x4.mul(row2, tmp1), minor1);
+            minor2 = SIMD.float32x4.sub(minor2, SIMD.float32x4.mul(row1, tmp1));
+
+            // ----
+            tmp1 = SIMD.float32x4.mul(row0, row2);
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 1, 0, 3, 2); // 0xB1 = 10110001
+            minor1 = SIMD.float32x4.add(SIMD.float32x4.mul(row3, tmp1), minor1);
+            minor3 = SIMD.float32x4.sub(minor3, SIMD.float32x4.mul(row1, tmp1));
+            tmp1 = SIMD.float32x4.swizzle(tmp1, 2, 3, 0, 1); // 0x4E = 01001110
+            minor1 = SIMD.float32x4.sub(minor1, SIMD.float32x4.mul(row3, tmp1));
+            minor3 = SIMD.float32x4.add(SIMD.float32x4.mul(row1, tmp1), minor3);
+
+            // Compute determinant
+            det = SIMD.float32x4.mul(row0, minor0);
+            det = SIMD.float32x4.add(SIMD.float32x4.swizzle(det, 2, 3, 0, 1), det); // 0x4E = 01001110
+            det = SIMD.float32x4.add(SIMD.float32x4.swizzle(det, 1, 0, 3, 2), det); // 0xB1 = 10110001
+            tmp1 = SIMD.float32x4.reciprocal(det);
+            det = SIMD.float32x4.sub(SIMD.float32x4.add(tmp1, tmp1), SIMD.float32x4.mul(det, SIMD.float32x4.mul(tmp1, tmp1)));
+            det = SIMD.float32x4.swizzle(det, 0, 0, 0, 0);
+
+            // These shuffles aren't necessary if the faulty transposition is done
+            // up at the top of this function.
+            //minor0 = SIMD.float32x4.swizzle(minor0, 2, 1, 0, 3);
+            //minor1 = SIMD.float32x4.swizzle(minor1, 2, 1, 0, 3);
+            //minor2 = SIMD.float32x4.swizzle(minor2, 2, 1, 0, 3);
+            //minor3 = SIMD.float32x4.swizzle(minor3, 2, 1, 0, 3);
+
+            // Compute final values by multiplying with 1/det
+            minor0 = SIMD.float32x4.mul(det, minor0);
+            minor1 = SIMD.float32x4.mul(det, minor1);
+            minor2 = SIMD.float32x4.mul(det, minor2);
+            minor3 = SIMD.float32x4.mul(det, minor3);
+
+            SIMD.float32x4.store(dest, 0, minor0);
+            SIMD.float32x4.store(dest, 4, minor1);
+            SIMD.float32x4.store(dest, 8, minor2);
+            SIMD.float32x4.store(dest, 12, minor3);
+
+            return this;
+        }
+
         public setTranslation(vector3: Vector3): Matrix {
             this.m[12] = vector3.x;
             this.m[13] = vector3.y;
@@ -2244,6 +2369,106 @@
                 ex, ey, ez, 1, result);
         }
 
+        public static LookAtLHToRefSIMD(eyeRef: Vector3, targetRef: Vector3, upRef: Vector3, result: Matrix): void {
+            var out = result.m;
+            var center = SIMD.float32x4(targetRef.x, targetRef.y, targetRef.z, 0);
+            var eye = SIMD.float32x4(eyeRef.x, eyeRef.y, eyeRef.z, 0);
+            var up = SIMD.float32x4(upRef.x, upRef.y, upRef.z, 0);
+
+            // cc.kmVec3Subtract(f, pCenter, pEye);
+            var f = SIMD.float32x4.sub(center, eye);
+            // cc.kmVec3Normalize(f, f);    
+            var tmp = SIMD.float32x4.mul(f, f);
+            tmp = SIMD.float32x4.add(tmp, SIMD.float32x4.add(SIMD.float32x4.swizzle(tmp, 1, 2, 0, 3), SIMD.float32x4.swizzle(tmp, 2, 0, 1, 3)));
+            f = SIMD.float32x4.mul(f, SIMD.float32x4.reciprocalSqrt(tmp));
+
+            // cc.kmVec3Assign(up, pUp);
+            // cc.kmVec3Normalize(up, up);
+            tmp = SIMD.float32x4.mul(up, up);
+            tmp = SIMD.float32x4.add(tmp, SIMD.float32x4.add(SIMD.float32x4.swizzle(tmp, 1, 2, 0, 3), SIMD.float32x4.swizzle(tmp, 2, 0, 1, 3)));
+            up = SIMD.float32x4.mul(up, SIMD.float32x4.reciprocalSqrt(tmp));
+
+            // cc.kmVec3Cross(s, f, up);
+            var s = SIMD.float32x4.sub(SIMD.float32x4.mul(SIMD.float32x4.swizzle(f, 1, 2, 0, 3), SIMD.float32x4.swizzle(up, 2, 0, 1, 3)),
+                SIMD.float32x4.mul(SIMD.float32x4.swizzle(f, 2, 0, 1, 3), SIMD.float32x4.swizzle(up, 1, 2, 0, 3)));
+            // cc.kmVec3Normalize(s, s);
+            tmp = SIMD.float32x4.mul(s, s);
+            tmp = SIMD.float32x4.add(tmp, SIMD.float32x4.add(SIMD.float32x4.swizzle(tmp, 1, 2, 0, 3), SIMD.float32x4.swizzle(tmp, 2, 0, 1, 3)));
+            s = SIMD.float32x4.mul(s, SIMD.float32x4.reciprocalSqrt(tmp));
+
+            // cc.kmVec3Cross(u, s, f);
+            var u = SIMD.float32x4.sub(SIMD.float32x4.mul(SIMD.float32x4.swizzle(s, 1, 2, 0, 3), SIMD.float32x4.swizzle(f, 2, 0, 1, 3)),
+                SIMD.float32x4.mul(SIMD.float32x4.swizzle(s, 2, 0, 1, 3), SIMD.float32x4.swizzle(f, 1, 2, 0, 3)));
+            // cc.kmVec3Normalize(s, s);
+            tmp = SIMD.float32x4.mul(s, s);
+            tmp = SIMD.float32x4.add(tmp, SIMD.float32x4.add(SIMD.float32x4.swizzle(tmp, 1, 2, 0, 3), SIMD.float32x4.swizzle(tmp, 2, 0, 1, 3)));
+            s = SIMD.float32x4.mul(s, SIMD.float32x4.reciprocalSqrt(tmp));
+
+            //cc.kmMat4Identity(pOut);
+            //pOut.mat[0] = s.x;
+            //pOut.mat[4] = s.y;
+            //pOut.mat[8] = s.z;
+            //pOut.mat[1] = u.x;
+            //pOut.mat[5] = u.y;
+            //pOut.mat[9] = u.z;
+            //pOut.mat[2] = -f.x;
+            //pOut.mat[6] = -f.y;
+            //pOut.mat[10] = -f.z;
+            var zero = SIMD.float32x4.splat(0.0);
+            s = SIMD.float32x4.neg(s);
+            var tmp01 = SIMD.float32x4.shuffle(s, u, 0, 1, 4, 5);
+            var tmp23 = SIMD.float32x4.shuffle(f, zero, 0, 1, 4, 5);
+            var a0 = SIMD.float32x4.shuffle(tmp01, tmp23, 0, 2, 4, 6);
+            var a1 = SIMD.float32x4.shuffle(tmp01, tmp23, 1, 3, 5, 7);
+
+            var tmp01 = SIMD.float32x4.shuffle(s, u, 2, 3, 6, 7);
+            var tmp23 = SIMD.float32x4.shuffle(f, zero, 2, 3, 6, 7);
+            var a2 = SIMD.float32x4.shuffle(tmp01, tmp23, 0, 2, 4, 6);
+            var a3 = SIMD.float32x4(0.0, 0.0, 0.0, 1.0);
+
+            // cc.kmMat4Translation(translate, -pEye.x, -pEye.y, -pEye.z);
+            var b0 = SIMD.float32x4(1.0, 0.0, 0.0, 0.0);
+            var b1 = SIMD.float32x4(0.0, 1.0, 0.0, 0.0);
+            var b2 = SIMD.float32x4(0.0, 0.0, 1.0, 0.0);
+            var b3 = SIMD.float32x4.neg(eye);
+            b3 = SIMD.float32x4.withW(b3, 1.0);
+
+            // cc.kmMat4Multiply(pOut, pOut, translate);
+            SIMD.float32x4.store(out, 0, SIMD.float32x4.add(
+                SIMD.float32x4.mul(
+                    SIMD.float32x4.swizzle(b0, 0, 0, 0, 0), a0),
+                SIMD.float32x4.add(
+                    SIMD.float32x4.mul(SIMD.float32x4.swizzle(b0, 1, 1, 1, 1), a1),
+                    SIMD.float32x4.add(
+                        SIMD.float32x4.mul(SIMD.float32x4.swizzle(b0, 2, 2, 2, 2), a2),
+                        SIMD.float32x4.mul(SIMD.float32x4.swizzle(b0, 3, 3, 3, 3), a3)))));
+            SIMD.float32x4.store(out, 4, SIMD.float32x4.add(
+                SIMD.float32x4.mul(
+                    SIMD.float32x4.swizzle(b1, 0, 0, 0, 0), a0),
+                SIMD.float32x4.add(
+                    SIMD.float32x4.mul(SIMD.float32x4.swizzle(b1, 1, 1, 1, 1), a1),
+                    SIMD.float32x4.add(
+                        SIMD.float32x4.mul(SIMD.float32x4.swizzle(b1, 2, 2, 2, 2), a2),
+                        SIMD.float32x4.mul(SIMD.float32x4.swizzle(b1, 3, 3, 3, 3), a3)))));
+            SIMD.float32x4.store(out, 8, SIMD.float32x4.add(
+                SIMD.float32x4.mul(
+                    SIMD.float32x4.swizzle(b2, 0, 0, 0, 0), a0),
+                SIMD.float32x4.add(
+                    SIMD.float32x4.mul(SIMD.float32x4.swizzle(b2, 1, 1, 1, 1), a1),
+                    SIMD.float32x4.add(
+                        SIMD.float32x4.mul(SIMD.float32x4.swizzle(b2, 2, 2, 2, 2), a2),
+                        SIMD.float32x4.mul(SIMD.float32x4.swizzle(b2, 3, 3, 3, 3), a3)))));
+            SIMD.float32x4.store(out, 12, SIMD.float32x4.add(
+                SIMD.float32x4.mul(
+                    SIMD.float32x4.swizzle(b3, 0, 0, 0, 0), a0),
+                SIMD.float32x4.add(
+                    SIMD.float32x4.mul(SIMD.float32x4.swizzle(b3, 1, 1, 1, 1), a1),
+                    SIMD.float32x4.add(
+                        SIMD.float32x4.mul(SIMD.float32x4.swizzle(b3, 2, 2, 2, 2), a2),
+                        SIMD.float32x4.mul(SIMD.float32x4.swizzle(b3, 3, 3, 3, 3), a3)))));
+        }
+
+
         public static OrthoLH(width: number, height: number, znear: number, zfar: number): Matrix {
             var matrix = Matrix.Zero();
 
@@ -3235,6 +3460,8 @@
     if (window.SIMD !== undefined) {
         // Replace functions
         Matrix.prototype.multiplyToArray = <any>Matrix.prototype.multiplyToArraySIMD;
+        Matrix.prototype.invertToRef = <any>Matrix.prototype.invertToRefSIMD;
+        Matrix.LookAtLHToRef = <any>Matrix.LookAtLHToRefSIMD;
         Vector3.TransformCoordinatesToRef = <any>Vector3.TransformCoordinatesToRefSIMD;
         Vector3.TransformCoordinatesFromFloatsToRef = <any>Vector3.TransformCoordinatesFromFloatsToRefSIMD;
     }

Babylon/Mesh/babylon.subMesh.js

@@ -15,6 +15,7 @@ var BABYLON;
             this._id = mesh.subMeshes.length - 1;
             if (createBoundingBox) {
                 this.refreshBoundingInfo();
+                mesh.computeWorldMatrix(true);
             }
         }
         SubMesh.prototype.getBoundingInfo = function () {

Babylon/Mesh/babylon.subMesh.ts

@@ -24,6 +24,7 @@
 
             if (createBoundingBox) {
                 this.refreshBoundingInfo();
+                mesh.computeWorldMatrix(true);
             }
         }