1 | // Copyright (C) 2016 Paul Lemire <paul.lemire350@gmail.com> |
2 | // SPDX-License-Identifier: LicenseRef-Qt-Commercial OR LGPL-3.0-only OR GPL-2.0-only OR GPL-3.0-only |
3 | |
4 | #ifndef QT3DCORE_VECTOR3D_SSE_P_H |
5 | #define QT3DCORE_VECTOR3D_SSE_P_H |
6 | |
7 | // |
8 | // W A R N I N G |
9 | // ------------- |
10 | // |
11 | // This file is not part of the Qt3D API. It exists purely as an |
12 | // implementation detail. This header file may change from version to |
13 | // version without notice, or even be removed. |
14 | // |
15 | // We mean it. |
16 | // |
17 | |
18 | #include <Qt3DCore/private/qt3dcore_global_p.h> |
19 | #include <QtCore/private/qsimd_p.h> |
20 | #include <QtCore/QtGlobal> |
21 | #include <QtGui/qvector3d.h> |
22 | #include <QDebug> |
23 | #include <math.h> |
24 | |
25 | #ifdef __SSE2__ |
26 | |
27 | QT_BEGIN_NAMESPACE |
28 | |
29 | namespace Qt3DCore { |
30 | |
31 | class Matrix4x4_SSE; |
32 | class Vector4D_SSE; |
33 | |
34 | class Vector3D_SSE |
35 | { |
36 | public: |
37 | |
38 | Q_ALWAYS_INLINE Vector3D_SSE() |
39 | : m_xyzw(_mm_setzero_ps()) |
40 | { |
41 | } |
42 | |
43 | explicit Q_ALWAYS_INLINE Vector3D_SSE(Qt::Initialization) {} |
44 | |
45 | explicit Q_ALWAYS_INLINE Vector3D_SSE(float x, float y, float z) |
46 | : m_xyzw(_mm_set_ps(z: 0.0f, y: z, x: y, w: x)) |
47 | { |
48 | } |
49 | |
50 | explicit Q_ALWAYS_INLINE Vector3D_SSE(QVector3D v) |
51 | : m_xyzw(_mm_set_ps(z: 0.0f, y: v.z(), x: v.y(), w: v.x())) |
52 | { |
53 | } |
54 | |
55 | explicit Q_3DCORE_PRIVATE_EXPORT Vector3D_SSE(const Vector4D_SSE &v); |
56 | |
57 | Q_ALWAYS_INLINE Vector3D_SSE &operator+=(Vector3D_SSE vector) |
58 | { |
59 | m_xyzw = _mm_add_ps(a: m_xyzw, b: vector.m_xyzw); |
60 | return *this; |
61 | } |
62 | |
63 | Q_ALWAYS_INLINE Vector3D_SSE &operator-=(Vector3D_SSE vector) |
64 | { |
65 | m_xyzw = _mm_sub_ps(a: m_xyzw, b: vector.m_xyzw); |
66 | return *this; |
67 | } |
68 | |
69 | Q_ALWAYS_INLINE Vector3D_SSE &operator*=(Vector3D_SSE vector) |
70 | { |
71 | m_xyzw = _mm_mul_ps(a: m_xyzw, b: vector.m_xyzw); |
72 | return *this; |
73 | } |
74 | |
75 | Q_ALWAYS_INLINE Vector3D_SSE &operator/=(Vector3D_SSE vector) |
76 | { |
77 | m_xyzw = _mm_div_ps(a: m_xyzw, b: vector.m_xyzw); |
78 | return *this; |
79 | } |
80 | |
81 | Q_ALWAYS_INLINE Vector3D_SSE &operator*=(float factor) |
82 | { |
83 | m_xyzw = _mm_mul_ps(a: m_xyzw, b: _mm_set1_ps(w: factor)); |
84 | return *this; |
85 | } |
86 | |
87 | Q_ALWAYS_INLINE Vector3D_SSE &operator/=(float factor) |
88 | { |
89 | m_xyzw = _mm_div_ps(a: m_xyzw, b: _mm_set1_ps(w: factor)); |
90 | return *this; |
91 | } |
92 | |
93 | Q_ALWAYS_INLINE bool operator==(Vector3D_SSE other) const |
94 | { |
95 | // 0b111 == 0x7 |
96 | return ((_mm_movemask_ps(a: _mm_cmpeq_ps(a: m_xyzw, b: other.m_xyzw)) & 0x7) == 0x7); |
97 | } |
98 | |
99 | Q_ALWAYS_INLINE bool operator!=(Vector3D_SSE other) const |
100 | { |
101 | return !(*this == other); |
102 | } |
103 | |
104 | Q_ALWAYS_INLINE QVector3D toQVector3D() const |
105 | { |
106 | return QVector3D(x(), y(), z()); |
107 | } |
108 | |
109 | Q_ALWAYS_INLINE float lengthSquared() const |
110 | { |
111 | return Qt3DCore::Vector3D_SSE::dotProduct(a: *this, b: *this); |
112 | } |
113 | |
114 | Q_ALWAYS_INLINE float length() const |
115 | { |
116 | return sqrt(x: Qt3DCore::Vector3D_SSE::dotProduct(a: *this, b: *this)); |
117 | } |
118 | |
119 | Q_ALWAYS_INLINE float distanceToPoint(const Vector3D_SSE &point) const |
120 | { |
121 | return (*this - point).length(); |
122 | } |
123 | |
124 | Q_ALWAYS_INLINE void normalize() |
125 | { |
126 | const float len = length(); |
127 | m_xyzw = _mm_div_ps(a: m_xyzw, b: _mm_set_ps1(w: len)); |
128 | } |
129 | |
130 | Q_ALWAYS_INLINE Vector3D_SSE normalized() const |
131 | { |
132 | Vector3D_SSE v = *this; |
133 | v.normalize(); |
134 | return v; |
135 | } |
136 | |
137 | Q_ALWAYS_INLINE bool isNull() const |
138 | { |
139 | // Ignore last bit |
140 | // 0b111 = 0x7 |
141 | return ((_mm_movemask_ps(a: _mm_cmpeq_ps(a: m_xyzw, b: _mm_set_ps1(w: 0.0f))) & 0x7) == 0x7); |
142 | } |
143 | |
144 | Q_3DCORE_PRIVATE_EXPORT Vector3D_SSE unproject(const Matrix4x4_SSE &modelView, const Matrix4x4_SSE &projection, const QRect &viewport) const; |
145 | Q_3DCORE_PRIVATE_EXPORT Vector3D_SSE project(const Matrix4x4_SSE &modelView, const Matrix4x4_SSE &projection, const QRect &viewport) const; |
146 | |
147 | Q_ALWAYS_INLINE float x() const { return _mm_cvtss_f32(a: m_xyzw); } |
148 | |
149 | Q_ALWAYS_INLINE float y() const |
150 | { |
151 | // 0b01010101 = 0x55 |
152 | return _mm_cvtss_f32(_mm_shuffle_ps(m_xyzw, m_xyzw, 0x55)); |
153 | } |
154 | |
155 | Q_ALWAYS_INLINE float z() const |
156 | { |
157 | // 0b10101010 = 0xaa |
158 | return _mm_cvtss_f32(a: _mm_unpackhi_ps(a: m_xyzw, b: m_xyzw)); |
159 | } |
160 | |
161 | Q_ALWAYS_INLINE void setX(float x) |
162 | { |
163 | m_xyzw = _mm_move_ss(a: m_xyzw, b: _mm_set_ss(w: x)); |
164 | } |
165 | |
166 | Q_ALWAYS_INLINE void setY(float y) |
167 | { |
168 | // m_xyzw = a, b, c, d |
169 | |
170 | // y, y, y, y |
171 | const __m128 yVec = _mm_set_ps1(w: y); |
172 | |
173 | // y, y, a, a |
174 | // 0b00000000 == 0x0 |
175 | const __m128 yaVec = _mm_shuffle_ps(yVec, m_xyzw, 0x0); |
176 | |
177 | // a, y, c, d |
178 | // 0b11100010 == 0xe2 |
179 | m_xyzw = _mm_shuffle_ps(yaVec, m_xyzw, 0xe2); |
180 | } |
181 | |
182 | Q_ALWAYS_INLINE void setZ(float z) |
183 | { |
184 | // m_xyzw = a, b, c, d |
185 | |
186 | // z, z, z, z |
187 | const __m128 zVec = _mm_set_ps1(w: z); |
188 | |
189 | // z, z, d, d |
190 | // 0b11110000 == 0xf0 |
191 | const __m128 zdVec = _mm_shuffle_ps(zVec, m_xyzw, 0xf0); |
192 | |
193 | // a, b, z, d |
194 | // 0b10000100 == 0x84 |
195 | m_xyzw = _mm_shuffle_ps(m_xyzw, zdVec, 0x84); |
196 | } |
197 | |
198 | Q_ALWAYS_INLINE float operator[](int idx) const |
199 | { |
200 | switch (idx) { |
201 | case 0: |
202 | return x(); |
203 | case 1: |
204 | return y(); |
205 | case 2: |
206 | return z(); |
207 | default: |
208 | Q_UNREACHABLE_RETURN(0.0f); |
209 | } |
210 | } |
211 | |
212 | struct DigitWrapper |
213 | { |
214 | explicit DigitWrapper(int idx, Vector3D_SSE *vec) |
215 | : m_vec(vec) |
216 | , m_idx(idx) |
217 | {} |
218 | |
219 | operator float() const |
220 | { |
221 | switch (m_idx) { |
222 | case 0: |
223 | return m_vec->x(); |
224 | case 1: |
225 | return m_vec->y(); |
226 | case 2: |
227 | return m_vec->z(); |
228 | default: |
229 | Q_UNREACHABLE_RETURN(0.0f); |
230 | } |
231 | } |
232 | |
233 | void operator =(float value) |
234 | { |
235 | switch (m_idx) { |
236 | case 0: |
237 | m_vec->setX(value); |
238 | break; |
239 | case 1: |
240 | m_vec->setY(value); |
241 | break; |
242 | case 2: |
243 | m_vec->setZ(value); |
244 | break; |
245 | default: |
246 | Q_UNREACHABLE(); |
247 | } |
248 | } |
249 | |
250 | private: |
251 | Vector3D_SSE *m_vec; |
252 | const int m_idx; |
253 | }; |
254 | |
255 | Q_ALWAYS_INLINE DigitWrapper operator[](int idx) |
256 | { |
257 | return DigitWrapper(idx, this); |
258 | } |
259 | |
260 | static Q_ALWAYS_INLINE float dotProduct(Vector3D_SSE a, Vector3D_SSE b) |
261 | { |
262 | #if defined(__SSE4_1__) |
263 | // 0b01111111 = 0x7f |
264 | return _mm_cvtss_f32(_mm_dp_ps(a.m_xyzw, b.m_xyzw, 0x7f)); |
265 | #elif defined(__SSE3__) |
266 | const __m128 mult = _mm_mul_ps(a.m_xyzw, b.m_xyzw); |
267 | // a + b, c + d, a + d, c + d |
268 | const __m128 partialSum = _mm_hadd_ps(mult, mult); |
269 | // c + d, ...... |
270 | // 0x00000001 = |
271 | const __m128 partialSumShuffle = _mm_shuffle_ps(partialSum, partialSum, 0x1); |
272 | return _mm_cvtss_f32(_mm_hadd_ps(partialSum, partialSumShuffle)); |
273 | #else |
274 | const __m128 mult = _mm_mul_ps(a: a.m_xyzw, b: b.m_xyzw); |
275 | |
276 | // (multX, 0, 0, 0) + (multY, 0, 0, 0) -> (multX + multY, 0, 0, 0) |
277 | // 0b11111101 == 0xfd |
278 | const __m128 shuffled = _mm_shuffle_ps(mult, mult, 0xfd); |
279 | // (multX + multY, 0, 0, 0) + (multZ, 0, 0, 0); |
280 | // 0b11111110 == 0xfe |
281 | const __m128 shuffled2 = _mm_shuffle_ps(mult, mult, 0xfe); |
282 | const __m128 result = _mm_add_ps(a: _mm_add_ps(a: shuffled, b: mult), b: shuffled2); |
283 | return _mm_cvtss_f32(a: result); |
284 | #endif |
285 | } |
286 | |
287 | static Q_ALWAYS_INLINE Vector3D_SSE crossProduct(Vector3D_SSE a, Vector3D_SSE b) |
288 | { |
289 | // a.y * b.z - a.z * b.y, a.z * b.x - a.x * b.z, a.x * b.y - a.y * b.x |
290 | // (a.y, a.z, a.z, a.x, a.x, a.y) (b.z, b.y, b.x, b.z, b.y, b.x) |
291 | // (a.y, a.z, a.x) * (b.z, b.x, b.y) - (a.z, a.x, a.y) (b.y, b.z, b.x) |
292 | |
293 | // 0b11001001 == 0xc9 |
294 | const __m128 a1 = _mm_shuffle_ps(a.m_xyzw, a.m_xyzw, 0xc9); |
295 | const __m128 b2 = _mm_shuffle_ps(b.m_xyzw, b.m_xyzw, 0xc9); |
296 | // 0b11010010 == 0xd2 |
297 | const __m128 a2 = _mm_shuffle_ps(a.m_xyzw, a.m_xyzw, 0xd2); |
298 | const __m128 b1 = _mm_shuffle_ps(b.m_xyzw, b.m_xyzw, 0xd2); |
299 | |
300 | Vector3D_SSE v(Qt::Uninitialized); |
301 | v.m_xyzw = _mm_sub_ps(a: _mm_mul_ps(a: a1, b: b1), b: _mm_mul_ps(a: a2, b: b2)); |
302 | return v; |
303 | } |
304 | |
305 | friend class Vector4D_SSE; |
306 | friend class Matrix4x4_SSE; |
307 | friend Vector3D_SSE operator*(const Vector3D_SSE &vector, const Matrix4x4_SSE &matrix); |
308 | friend Vector3D_SSE operator*(const Matrix4x4_SSE &matrix, const Vector3D_SSE &vector); |
309 | |
310 | friend Q_ALWAYS_INLINE const Vector3D_SSE operator+(Vector3D_SSE v1, Vector3D_SSE v2) { return v1 += v2; } |
311 | friend Q_ALWAYS_INLINE const Vector3D_SSE operator-(Vector3D_SSE v1, Vector3D_SSE v2) { return v1 -= v2; } |
312 | friend Q_ALWAYS_INLINE const Vector3D_SSE operator*(float factor, Vector3D_SSE vector) { return vector *= factor; } |
313 | friend Q_ALWAYS_INLINE const Vector3D_SSE operator*(Vector3D_SSE vector, float factor) { return vector *= factor; } |
314 | friend Q_ALWAYS_INLINE const Vector3D_SSE operator*(Vector3D_SSE v1, Vector3D_SSE v2) { return v1 *= v2; } |
315 | friend Q_ALWAYS_INLINE const Vector3D_SSE operator-(Vector3D_SSE vector) |
316 | { |
317 | Vector3D_SSE c(Qt::Uninitialized); |
318 | |
319 | c.m_xyzw = _mm_xor_ps(a: vector.m_xyzw, b: _mm_set1_ps(w: -0.0f)); |
320 | |
321 | return c; |
322 | } |
323 | |
324 | friend Q_ALWAYS_INLINE const Vector3D_SSE operator/(Vector3D_SSE vector, float divisor) { return vector /= divisor; } |
325 | friend Q_ALWAYS_INLINE const Vector3D_SSE operator/(Vector3D_SSE vector, Vector3D_SSE divisor) { return vector /= divisor; } |
326 | |
327 | friend Q_3DCORE_PRIVATE_EXPORT QDebug operator<<(QDebug dbg, const Vector3D_SSE &v); |
328 | friend Q_ALWAYS_INLINE bool qFuzzyCompare(const Vector3D_SSE& v1, const Vector3D_SSE& v2) |
329 | { |
330 | return ::qFuzzyCompare(p1: v1.x(), p2: v2.x()) && |
331 | ::qFuzzyCompare(p1: v1.y(), p2: v2.y()) && |
332 | ::qFuzzyCompare(p1: v1.z(), p2: v2.z()); |
333 | } |
334 | |
335 | private: |
336 | // Q_DECL_ALIGN(16) float m[4];// for SSE support |
337 | __m128 m_xyzw; |
338 | }; |
339 | |
340 | } // Qt3DCore |
341 | |
342 | Q_DECLARE_TYPEINFO(Qt3DCore::Vector3D_SSE, Q_PRIMITIVE_TYPE); |
343 | |
344 | QT_END_NAMESPACE |
345 | |
346 | Q_DECLARE_METATYPE(Qt3DCore::Vector3D_SSE) |
347 | |
348 | #endif // __SSE2__ |
349 | |
350 | #endif // QT3DCORE_VECTOR3D_SSE_P_H |
351 | |