suwako / VulkanExample

master
ブランチ master
VulkanExample / dep / include / glm / simd / geometric.h
6.6 KB、195行 生 Blame 履歴
1/// @ref simd
2/// @file glm/simd/geometric.h
3
4#pragma once
5
6#include "common.h"
7
8#if GLM_ARCH & GLM_ARCH_SSE2_BIT
9
10GLM_FUNC_DECL glm_vec4 glm_vec4_dot(glm_vec4 v1, glm_vec4 v2);
11GLM_FUNC_DECL glm_vec4 glm_vec1_dot(glm_vec4 v1, glm_vec4 v2);
12
13GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_length(glm_vec4 x)
14{
15 glm_vec4 const dot0 = glm_vec4_dot(x, x);
16 glm_vec4 const sqt0 = _mm_sqrt_ps(dot0);
17 return sqt0;
18}
19
20GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_distance(glm_vec4 p0, glm_vec4 p1)
21{
22 glm_vec4 const sub0 = _mm_sub_ps(p0, p1);
23 glm_vec4 const len0 = glm_vec4_length(sub0);
24 return len0;
25}
26
27GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_dot(glm_vec4 v1, glm_vec4 v2)
28{
29# if GLM_ARCH & GLM_ARCH_SSE41_BIT
30 return _mm_dp_ps(v1, v2, 0xff);
31# elif GLM_ARCH & GLM_ARCH_SSE3_BIT
32 glm_vec4 const mul0 = _mm_mul_ps(v1, v2);
33 glm_vec4 const hadd0 = _mm_hadd_ps(mul0, mul0);
34 glm_vec4 const hadd1 = _mm_hadd_ps(hadd0, hadd0);
35 return hadd1;
36# else
37 glm_vec4 const mul0 = _mm_mul_ps(v1, v2);
38 glm_vec4 const swp0 = _mm_shuffle_ps(mul0, mul0, _MM_SHUFFLE(2, 3, 0, 1));
39 glm_vec4 const add0 = _mm_add_ps(mul0, swp0);
40 glm_vec4 const swp1 = _mm_shuffle_ps(add0, add0, _MM_SHUFFLE(0, 1, 2, 3));
41 glm_vec4 const add1 = _mm_add_ps(add0, swp1);
42 return add1;
43# endif
44}
45
46GLM_FUNC_QUALIFIER glm_vec4 glm_vec3_dot(glm_vec4 v1, glm_vec4 v2)
47{
48# if GLM_ARCH & GLM_ARCH_SSE41_BIT
49 return _mm_dp_ps(v1, v2, 0x77);
50# elif GLM_ARCH & GLM_ARCH_SSE3_BIT
51 glm_vec4 const constant0 = _mm_castsi128_ps(_mm_setr_epi32(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0));
52 glm_vec4 const mul0 = _mm_mul_ps(v1, v2);
53 glm_vec4 const and0 = _mm_and_ps(mul0, constant0);
54 glm_vec4 const hadd0 = _mm_hadd_ps(and0, and0);
55 glm_vec4 const hadd1 = _mm_hadd_ps(hadd0, hadd0);
56 return hadd1;
57# else
58 glm_vec4 const constant0 = _mm_castsi128_ps(_mm_setr_epi32(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0));
59 glm_vec4 const mul0 = _mm_mul_ps(v1, v2);
60 glm_vec4 const and0 = _mm_and_ps(mul0, constant0);
61 glm_vec4 const swp0 = _mm_shuffle_ps(and0, and0, _MM_SHUFFLE(2, 3, 0, 1));
62 glm_vec4 const add0 = _mm_add_ps(and0, swp0);
63 glm_vec4 const swp1 = _mm_shuffle_ps(add0, add0, _MM_SHUFFLE(0, 1, 2, 3));
64 glm_vec4 const add1 = _mm_add_ps(add0, swp1);
65 return add1;
66# endif
67
68}
69
70GLM_FUNC_QUALIFIER glm_vec4 glm_vec1_dot(glm_vec4 v1, glm_vec4 v2)
71{
72# if GLM_ARCH & GLM_ARCH_SSE41_BIT
73 return _mm_dp_ps(v1, v2, 0xff);
74# elif GLM_ARCH & GLM_ARCH_SSE3_BIT
75 glm_vec4 const mul0 = _mm_mul_ps(v1, v2);
76 glm_vec4 const had0 = _mm_hadd_ps(mul0, mul0);
77 glm_vec4 const had1 = _mm_hadd_ps(had0, had0);
78 return had1;
79# else
80 glm_vec4 const mul0 = _mm_mul_ps(v1, v2);
81 glm_vec4 const mov0 = _mm_movehl_ps(mul0, mul0);
82 glm_vec4 const add0 = _mm_add_ps(mov0, mul0);
83 glm_vec4 const swp1 = _mm_shuffle_ps(add0, add0, 1);
84 glm_vec4 const add1 = _mm_add_ss(add0, swp1);
85 return add1;
86# endif
87}
88
89GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_cross(glm_vec4 v1, glm_vec4 v2)
90{
91 glm_vec4 const swp0 = _mm_shuffle_ps(v1, v1, _MM_SHUFFLE(3, 0, 2, 1));
92 glm_vec4 const swp1 = _mm_shuffle_ps(v1, v1, _MM_SHUFFLE(3, 1, 0, 2));
93 glm_vec4 const swp2 = _mm_shuffle_ps(v2, v2, _MM_SHUFFLE(3, 0, 2, 1));
94 glm_vec4 const swp3 = _mm_shuffle_ps(v2, v2, _MM_SHUFFLE(3, 1, 0, 2));
95 glm_vec4 const mul0 = _mm_mul_ps(swp0, swp3);
96 glm_vec4 const mul1 = _mm_mul_ps(swp1, swp2);
97 glm_vec4 const sub0 = _mm_sub_ps(mul0, mul1);
98 return sub0;
99}
100
101GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_normalize_lowp(glm_vec4 v)
102{
103 glm_vec4 const dot0 = glm_vec4_dot(v, v);
104 glm_vec4 const isr0 = _mm_rsqrt_ps(dot0);
105 glm_vec4 const mul0 = _mm_mul_ps(v, isr0);
106 return mul0;
107}
108
109GLM_FUNC_QUALIFIER glm_vec4 glm_vec3_normalize_lowp(glm_vec4 v)
110{
111 glm_vec4 const dot0 = glm_vec3_dot(v, v);
112 glm_vec4 const isr0 = _mm_rsqrt_ps(dot0);
113 glm_vec4 const mul0 = _mm_mul_ps(v, isr0);
114 return mul0;
115}
116
117GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_normalize(glm_vec4 v)
118{
119 glm_vec4 const dot0 = glm_vec4_dot(v, v);
120 glm_vec4 const isr0 = _mm_rsqrt_ps(dot0);
121
122 // One iteration of Newton-Raphson method to improve precision
123 glm_vec4 const mul0 = glm_vec4_mul(dot0, isr0);
124 glm_vec4 const fma0 = glm_vec4_fma(mul0, isr0, _mm_set1_ps(-3.0f));
125 glm_vec4 const mul1 = glm_vec4_mul(isr0, _mm_set1_ps(-0.5f));
126
127 glm_vec4 const mul2 = glm_vec4_mul(mul1, v);
128 glm_vec4 const mul3 = glm_vec4_mul(mul2, fma0);
129
130 return mul3;
131}
132
133GLM_FUNC_QUALIFIER glm_vec4 glm_vec3_normalize(glm_vec4 v)
134{
135 glm_vec4 const dot0 = glm_vec3_dot(v, v);
136 glm_vec4 const isr0 = _mm_rsqrt_ps(dot0);
137
138 // One iteration of Newton-Raphson method to improve precision
139 glm_vec4 const mul0 = glm_vec4_mul(dot0, isr0);
140 glm_vec4 const fma0 = glm_vec4_fma(mul0, isr0, _mm_set1_ps(-3.0f));
141 glm_vec4 const mul1 = glm_vec4_mul(isr0, _mm_set1_ps(-0.5f));
142
143 glm_vec4 const mul2 = glm_vec4_mul(mul1, v);
144 glm_vec4 const mul3 = glm_vec4_mul(mul2, fma0);
145
146 return mul3;
147}
148
149
150GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_faceforward(glm_vec4 N, glm_vec4 I, glm_vec4 Nref)
151{
152 glm_vec4 const dot0 = glm_vec4_dot(Nref, I);
153 glm_vec4 const sgn0 = glm_vec4_sign(dot0);
154 glm_vec4 const mul0 = _mm_mul_ps(sgn0, _mm_set1_ps(-1.0f));
155 glm_vec4 const mul1 = _mm_mul_ps(N, mul0);
156 return mul1;
157}
158
159GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_reflect(glm_vec4 I, glm_vec4 N)
160{
161 glm_vec4 const dot0 = glm_vec4_dot(N, I);
162 glm_vec4 const mul0 = _mm_mul_ps(N, dot0);
163 glm_vec4 const mul1 = _mm_mul_ps(mul0, _mm_set1_ps(2.0f));
164 glm_vec4 const sub0 = _mm_sub_ps(I, mul1);
165 return sub0;
166}
167
168GLM_FUNC_QUALIFIER __m128 glm_vec4_refract(glm_vec4 I, glm_vec4 N, glm_vec4 eta)
169{
170 // k = 1.0 - eta * eta * (1.0 - dot(N, I) * dot(N, I));
171 // if (k < 0.0)
172 // R = genType(0.0); // or genDType(0.0)
173 // else
174 // R = eta * I - (eta * dot(N, I) + sqrt(k)) * N;
175
176 glm_vec4 const dot0 = glm_vec4_dot(N, I); // dot(N, I)
177 glm_vec4 const mul0 = _mm_mul_ps(eta, eta); // eta * eta
178 glm_vec4 const mul1 = _mm_mul_ps(dot0, dot0); // dot(N, I) * dot(N, I)
179 glm_vec4 const sub1 = _mm_sub_ps(_mm_set1_ps(1.0f), mul1); // (1.0 - dot(N, I) * dot(N, I))
180 glm_vec4 const mul2 = _mm_mul_ps(mul0, sub1); // eta * eta * (1.0 - dot(N, I) * dot(N, I))
181 glm_vec4 const sub0 = _mm_sub_ps(_mm_set1_ps(1.0f), mul2); // 1.0 - eta * eta * (1.0 - dot(N, I) * dot(N, I))
182
183 if(_mm_movemask_ps(_mm_cmplt_ss(sub0, _mm_set1_ps(0.0f))) == 0)
184 return _mm_set1_ps(0.0f);
185
186 glm_vec4 const sqt0 = _mm_sqrt_ps(sub0);
187 glm_vec4 const mad0 = glm_vec4_fma(eta, dot0, sqt0);
188 glm_vec4 const mul4 = _mm_mul_ps(mad0, N);
189 glm_vec4 const mul5 = _mm_mul_ps(eta, I);
190 glm_vec4 const sub2 = _mm_sub_ps(mul5, mul4);
191
192 return sub2;
193}
194
195#endif//GLM_ARCH & GLM_ARCH_SSE2_BIT