1 | /**************************************************************************** |
2 | ** |
3 | ** Copyright (C) 2016 Paul Lemire <paul.lemire350@gmail.com> |
4 | ** Contact: https://www.qt.io/licensing/ |
5 | ** |
6 | ** This file is part of the Qt3D module of the Qt Toolkit. |
7 | ** |
8 | ** $QT_BEGIN_LICENSE:LGPL$ |
9 | ** Commercial License Usage |
10 | ** Licensees holding valid commercial Qt licenses may use this file in |
11 | ** accordance with the commercial license agreement provided with the |
12 | ** Software or, alternatively, in accordance with the terms contained in |
13 | ** a written agreement between you and The Qt Company. For licensing terms |
14 | ** and conditions see https://www.qt.io/terms-conditions. For further |
15 | ** information use the contact form at https://www.qt.io/contact-us. |
16 | ** |
17 | ** GNU Lesser General Public License Usage |
18 | ** Alternatively, this file may be used under the terms of the GNU Lesser |
19 | ** General Public License version 3 as published by the Free Software |
20 | ** Foundation and appearing in the file LICENSE.LGPL3 included in the |
21 | ** packaging of this file. Please review the following information to |
22 | ** ensure the GNU Lesser General Public License version 3 requirements |
23 | ** will be met: https://www.gnu.org/licenses/lgpl-3.0.html. |
24 | ** |
25 | ** GNU General Public License Usage |
26 | ** Alternatively, this file may be used under the terms of the GNU |
27 | ** General Public License version 2.0 or (at your option) the GNU General |
28 | ** Public license version 3 or any later version approved by the KDE Free |
29 | ** Qt Foundation. The licenses are as published by the Free Software |
30 | ** Foundation and appearing in the file LICENSE.GPL2 and LICENSE.GPL3 |
31 | ** included in the packaging of this file. Please review the following |
32 | ** information to ensure the GNU General Public License requirements will |
33 | ** be met: https://www.gnu.org/licenses/gpl-2.0.html and |
34 | ** https://www.gnu.org/licenses/gpl-3.0.html. |
35 | ** |
36 | ** $QT_END_LICENSE$ |
37 | ** |
38 | ****************************************************************************/ |
39 | |
40 | #ifndef QT3DCORE_VECTOR4D_SSE_P_H |
41 | #define QT3DCORE_VECTOR4D_SSE_P_H |
42 | |
43 | // |
44 | // W A R N I N G |
45 | // ------------- |
46 | // |
47 | // This file is not part of the Qt3D API. It exists purely as an |
48 | // implementation detail. This header file may change from version to |
49 | // version without notice, or even be removed. |
50 | // |
51 | // We mean it. |
52 | // |
53 | |
54 | #include <Qt3DCore/private/vector3d_p.h> |
55 | #include <QtGui/qvector4d.h> |
56 | |
57 | #ifdef QT_COMPILER_SUPPORTS_SSE2 |
58 | |
59 | QT_BEGIN_NAMESPACE |
60 | |
61 | namespace Qt3DCore { |
62 | |
63 | class Matrix4x4_SSE; |
64 | class Matrix4x4_AVX2; |
65 | |
66 | class Vector4D_SSE |
67 | { |
68 | public: |
69 | Q_ALWAYS_INLINE Vector4D_SSE() |
70 | : m_xyzw(_mm_setzero_ps()) |
71 | { |
72 | } |
73 | |
74 | explicit Q_ALWAYS_INLINE Vector4D_SSE(Qt::Initialization) {} |
75 | |
76 | explicit Q_ALWAYS_INLINE Vector4D_SSE(float x, float y, float z, float w) |
77 | : m_xyzw(_mm_set_ps(z: w, y: z, x: y, w: x)) |
78 | { |
79 | } |
80 | |
81 | explicit Q_ALWAYS_INLINE Vector4D_SSE(QVector4D v) |
82 | : m_xyzw(_mm_set_ps(z: v.w(), y: v.z(), x: v.y(), w: v.x())) |
83 | { |
84 | } |
85 | |
86 | explicit Q_ALWAYS_INLINE Vector4D_SSE(const Vector3D_SSE &vec3, float w = 0.0f) |
87 | : m_xyzw(vec3.m_xyzw) |
88 | { |
89 | setW(w); |
90 | } |
91 | |
92 | explicit Q_ALWAYS_INLINE Vector4D_SSE(QVector3D v, float w = 0.0f) |
93 | : m_xyzw(_mm_set_ps(z: w, y: v.z(), x: v.y(), w: v.x())) |
94 | { |
95 | } |
96 | |
97 | Q_ALWAYS_INLINE Vector4D_SSE &operator+=(Vector4D_SSE vector) |
98 | { |
99 | m_xyzw = _mm_add_ps(a: m_xyzw, b: vector.m_xyzw); |
100 | return *this; |
101 | } |
102 | |
103 | Q_ALWAYS_INLINE Vector4D_SSE &operator-=(Vector4D_SSE vector) |
104 | { |
105 | m_xyzw = _mm_sub_ps(a: m_xyzw, b: vector.m_xyzw); |
106 | return *this; |
107 | } |
108 | |
109 | Q_ALWAYS_INLINE Vector4D_SSE &operator*=(Vector4D_SSE vector) |
110 | { |
111 | m_xyzw = _mm_mul_ps(a: m_xyzw, b: vector.m_xyzw); |
112 | return *this; |
113 | } |
114 | |
115 | Q_ALWAYS_INLINE Vector4D_SSE &operator/=(Vector4D_SSE vector) |
116 | { |
117 | m_xyzw = _mm_div_ps(a: m_xyzw, b: vector.m_xyzw); |
118 | return *this; |
119 | } |
120 | |
121 | Q_ALWAYS_INLINE Vector4D_SSE &operator*=(float factor) |
122 | { |
123 | m_xyzw = _mm_mul_ps(a: m_xyzw, b: _mm_set1_ps(w: factor)); |
124 | return *this; |
125 | } |
126 | |
127 | Q_ALWAYS_INLINE Vector4D_SSE &operator/=(float factor) |
128 | { |
129 | m_xyzw = _mm_div_ps(a: m_xyzw, b: _mm_set1_ps(w: factor)); |
130 | return *this; |
131 | } |
132 | |
133 | Q_ALWAYS_INLINE bool operator==(Vector4D_SSE other) const |
134 | { |
135 | // 0b1111 == 0xf |
136 | return (_mm_movemask_ps(a: _mm_cmpeq_ps(a: m_xyzw, b: other.m_xyzw)) == 0xf); |
137 | } |
138 | |
139 | Q_ALWAYS_INLINE bool operator!=(Vector4D_SSE other) const |
140 | { |
141 | return !(*this == other); |
142 | } |
143 | |
144 | Q_ALWAYS_INLINE QVector4D toQVector4D() const |
145 | { |
146 | return QVector4D(x(), y(), z(), w()); |
147 | } |
148 | |
149 | // TODO: Uncomment when we introduce Vector3D_SSE |
150 | //Q_ALWAYS_INLINE Vector3D_SSE toVector3D() const { return Vector3D_SSE(*this); } |
151 | |
152 | Q_ALWAYS_INLINE float lengthSquared() const |
153 | { |
154 | return dotProduct(a: *this, b: *this); |
155 | } |
156 | |
157 | Q_ALWAYS_INLINE float length() const |
158 | { |
159 | return sqrt(x: dotProduct(a: *this, b: *this)); |
160 | } |
161 | |
162 | Q_ALWAYS_INLINE void normalize() |
163 | { |
164 | const float len = length(); |
165 | m_xyzw = _mm_div_ps(a: m_xyzw, b: _mm_set_ps1(w: len)); |
166 | } |
167 | |
168 | Q_ALWAYS_INLINE Vector4D_SSE normalized() const |
169 | { |
170 | Vector4D_SSE v = *this; |
171 | v.normalize(); |
172 | return v; |
173 | } |
174 | |
175 | Q_ALWAYS_INLINE bool isNull() const |
176 | { |
177 | // 0b1111 == 0xf |
178 | return _mm_movemask_ps(a: _mm_cmpeq_ps(a: m_xyzw, b: _mm_setzero_ps())) == 0xf; |
179 | } |
180 | |
181 | Q_ALWAYS_INLINE float x() const { return _mm_cvtss_f32(a: m_xyzw); } |
182 | |
183 | Q_ALWAYS_INLINE float y() const |
184 | { |
185 | // 0b01010101 = 0x55 |
186 | return _mm_cvtss_f32(_mm_shuffle_ps(m_xyzw, m_xyzw, 0x55)); |
187 | } |
188 | |
189 | Q_ALWAYS_INLINE float z() const |
190 | { |
191 | // 0b10101010 = 0xaa |
192 | return _mm_cvtss_f32(a: _mm_unpackhi_ps(a: m_xyzw, b: m_xyzw)); |
193 | } |
194 | |
195 | Q_ALWAYS_INLINE float w() const |
196 | { |
197 | // 0b11111111 = 0xff |
198 | return _mm_cvtss_f32(_mm_shuffle_ps(m_xyzw, m_xyzw, 0xff)); |
199 | } |
200 | |
201 | Q_ALWAYS_INLINE void setX(float x) |
202 | { |
203 | m_xyzw = _mm_move_ss(a: m_xyzw, b: _mm_set_ss(w: x)); |
204 | } |
205 | |
206 | Q_ALWAYS_INLINE void setY(float y) |
207 | { |
208 | // m_xyzw = a, b, c, d |
209 | |
210 | // y, y, y, y |
211 | const __m128 yVec = _mm_set_ps1(w: y); |
212 | |
213 | // y, y, a, a |
214 | // 0b00000000 == 0x0 |
215 | const __m128 yaVec = _mm_shuffle_ps(yVec, m_xyzw, 0x0); |
216 | |
217 | // a, y, c, d |
218 | // 0b11100010 == 0xe2 |
219 | m_xyzw = _mm_shuffle_ps(yaVec, m_xyzw, 0xe2); |
220 | } |
221 | |
222 | Q_ALWAYS_INLINE void setZ(float z) |
223 | { |
224 | // m_xyzw = a, b, c, d |
225 | |
226 | // z, z, z, z |
227 | const __m128 zVec = _mm_set_ps1(w: z); |
228 | |
229 | // z, z, d, d |
230 | // 0b11110000 == 0xf0 |
231 | const __m128 zdVec = _mm_shuffle_ps(zVec, m_xyzw, 0xf0); |
232 | |
233 | // a, b, z, d |
234 | // 0b10000100 == 0x84 |
235 | m_xyzw = _mm_shuffle_ps(m_xyzw, zdVec, 0x84); |
236 | } |
237 | |
238 | Q_ALWAYS_INLINE void setW(float w) |
239 | { |
240 | #ifdef __SSE4_1__ |
241 | const __m128 wVec = _mm_set_ss(w); |
242 | // insert element 0 of wVec into position 3 in vec3, don't zero anything |
243 | m_xyzw = _mm_insert_ps(m_xyzw, wVec, 0x30); |
244 | #else |
245 | // m_xyzw = a, b, c, d |
246 | |
247 | // w, w, w, w |
248 | const __m128 wVec = _mm_set_ps1(w: w); |
249 | |
250 | // c, c, w, w |
251 | const __m128 cwVec = _mm_shuffle_ps(m_xyzw, wVec, _MM_SHUFFLE(0, 0, 2, 2)); |
252 | |
253 | // a, b, c, w |
254 | m_xyzw = _mm_shuffle_ps(m_xyzw, cwVec, _MM_SHUFFLE(2, 0, 1, 0)); |
255 | #endif |
256 | } |
257 | |
258 | Q_ALWAYS_INLINE float operator[](int idx) const |
259 | { |
260 | Q_DECL_ALIGN(16) float vec[4]; |
261 | _mm_store_ps(p: vec, a: m_xyzw); |
262 | return vec[idx]; |
263 | } |
264 | |
265 | struct DigitWrapper |
266 | { |
267 | explicit DigitWrapper(int idx, Vector4D_SSE *vec) |
268 | : m_vec(vec) |
269 | , m_idx(idx) |
270 | {} |
271 | |
272 | operator float() const |
273 | { |
274 | switch (m_idx) { |
275 | case 0: |
276 | return m_vec->x(); |
277 | case 1: |
278 | return m_vec->y(); |
279 | case 2: |
280 | return m_vec->z(); |
281 | case 3: |
282 | return m_vec->w(); |
283 | default: |
284 | Q_UNREACHABLE(); |
285 | return 0.0f; |
286 | } |
287 | } |
288 | void operator =(float value) |
289 | { |
290 | switch (m_idx) { |
291 | case 0: |
292 | m_vec->setX(value); |
293 | break; |
294 | case 1: |
295 | m_vec->setY(value); |
296 | break; |
297 | case 2: |
298 | m_vec->setZ(value); |
299 | break; |
300 | case 3: |
301 | m_vec->setW(value); |
302 | break; |
303 | default: |
304 | Q_UNREACHABLE(); |
305 | } |
306 | } |
307 | |
308 | private: |
309 | Vector4D_SSE *m_vec; |
310 | const int m_idx; |
311 | }; |
312 | |
313 | Q_ALWAYS_INLINE DigitWrapper operator[](int idx) |
314 | { |
315 | return DigitWrapper(idx, this); |
316 | } |
317 | |
318 | static Q_ALWAYS_INLINE float dotProduct(Vector4D_SSE a, Vector4D_SSE b) |
319 | { |
320 | #if defined(__SSE4_1__) |
321 | // 0b11111111 = 0xff |
322 | return _mm_cvtss_f32(_mm_dp_ps(a.m_xyzw, b.m_xyzw, 0xff)); |
323 | #elif defined(__SSE3__) |
324 | const __m128 mult = _mm_mul_ps(a.m_xyzw, b.m_xyzw); |
325 | // a + b, c + d, a + d, c + d |
326 | const __m128 partialSum = _mm_hadd_ps(mult, mult); |
327 | // c + d, ...... |
328 | // 0x00000001 = |
329 | const __m128 partialSumShuffle = _mm_shuffle_ps(partialSum, partialSum, 0x1); |
330 | return _mm_cvtss_f32(_mm_hadd_ps(partialSum, partialSumShuffle)); |
331 | #else |
332 | const __m128 mult = _mm_mul_ps(a: a.m_xyzw, b: b.m_xyzw); |
333 | // (multX, multY, 0, 0) + (multZ, multW, 0, 0) -> (multX + multZ, multY + multW, 0, 0) |
334 | // 0b00001110 == 0xe |
335 | const __m128 shuffled = _mm_shuffle_ps(mult, mult, 0xe); |
336 | __m128 result = _mm_add_ps(a: shuffled, b: mult); |
337 | // (multX + multZ, 0, 0, 0) + (multY + multW, 0, 0, 0); |
338 | // 0b00000001 == 0x1 |
339 | const __m128 shuffled2 = _mm_shuffle_ps(result, result, 0x1); |
340 | result = _mm_add_ps(a: result, b: shuffled2); |
341 | return _mm_cvtss_f32(a: result); |
342 | #endif |
343 | } |
344 | |
345 | friend class Matrix4x4_SSE; |
346 | |
347 | #ifdef __AVX2__ |
348 | friend class Matrix4x4_AVX2; |
349 | friend Vector4D_SSE operator*(const Vector4D_SSE &vector, const Matrix4x4_AVX2 &matrix); |
350 | friend Vector4D_SSE operator*(const Matrix4x4_AVX2 &matrix, const Vector4D_SSE &vector); |
351 | #endif |
352 | |
353 | friend class Vector3D_SSE; |
354 | friend Vector4D_SSE operator*(const Vector4D_SSE &vector, const Matrix4x4_SSE &matrix); |
355 | friend Vector4D_SSE operator*(const Matrix4x4_SSE &matrix, const Vector4D_SSE &vector); |
356 | |
357 | friend Q_ALWAYS_INLINE const Vector4D_SSE operator+(Vector4D_SSE v1, Vector4D_SSE v2) { return v1 += v2; } |
358 | friend Q_ALWAYS_INLINE const Vector4D_SSE operator-(Vector4D_SSE v1, Vector4D_SSE v2) { return v1 -= v2; } |
359 | friend Q_ALWAYS_INLINE const Vector4D_SSE operator*(float factor, Vector4D_SSE vector) { return vector *= factor; } |
360 | friend Q_ALWAYS_INLINE const Vector4D_SSE operator*(Vector4D_SSE vector, float factor) { return vector *= factor; } |
361 | friend Q_ALWAYS_INLINE const Vector4D_SSE operator*(Vector4D_SSE v1, Vector4D_SSE v2) { return v1 *= v2; } |
362 | friend Q_ALWAYS_INLINE const Vector4D_SSE operator-(Vector4D_SSE vector) |
363 | { |
364 | Vector4D_SSE c(Qt::Uninitialized); |
365 | |
366 | c.m_xyzw = _mm_xor_ps(a: vector.m_xyzw, b: _mm_set1_ps(w: -0.0f)); |
367 | |
368 | return c; |
369 | } |
370 | |
371 | friend Q_ALWAYS_INLINE const Vector4D_SSE operator/(Vector4D_SSE vector, float divisor) { return vector /= divisor; } |
372 | friend Q_ALWAYS_INLINE const Vector4D_SSE operator/(Vector4D_SSE vector, Vector4D_SSE divisor) { return vector /= divisor; } |
373 | |
374 | friend Q_3DCORE_PRIVATE_EXPORT QDebug operator<<(QDebug dbg, const Vector4D_SSE &v); |
375 | friend Q_ALWAYS_INLINE bool qFuzzyCompare(const Vector4D_SSE& v1, const Vector4D_SSE& v2) |
376 | { |
377 | return ::qFuzzyCompare(p1: v1.x(), p2: v2.x()) && |
378 | ::qFuzzyCompare(p1: v1.y(), p2: v2.y()) && |
379 | ::qFuzzyCompare(p1: v1.z(), p2: v2.z()) && |
380 | ::qFuzzyCompare(p1: v1.w(), p2: v2.w()); |
381 | } |
382 | |
383 | private: |
384 | // Q_DECL_ALIGN(16) float m[4];// for SSE support |
385 | __m128 m_xyzw; |
386 | }; |
387 | |
388 | } // Qt3DCore |
389 | |
390 | Q_DECLARE_TYPEINFO(Qt3DCore::Vector4D_SSE, Q_PRIMITIVE_TYPE); |
391 | |
392 | QT_END_NAMESPACE |
393 | |
394 | Q_DECLARE_METATYPE(Qt3DCore::Vector4D_SSE) |
395 | |
396 | #endif // QT_COMPILER_SUPPORTS_SSE2 |
397 | |
398 | #endif // QT3DCORE_VECTOR4D_SSE_P_H |
399 | |