1 /// @ref simd
2 /// @file glm/simd/common.h
3
4 #pragma once
5
6 #include "platform.h"
7
8 #if GLM_ARCH & GLM_ARCH_SSE2_BIT
9
glm_vec4_add(glm_vec4 a,glm_vec4 b)10 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_add(glm_vec4 a, glm_vec4 b)
11 {
12 return _mm_add_ps(a, b);
13 }
14
glm_vec1_add(glm_vec4 a,glm_vec4 b)15 GLM_FUNC_QUALIFIER glm_vec4 glm_vec1_add(glm_vec4 a, glm_vec4 b)
16 {
17 return _mm_add_ss(a, b);
18 }
19
glm_vec4_sub(glm_vec4 a,glm_vec4 b)20 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_sub(glm_vec4 a, glm_vec4 b)
21 {
22 return _mm_sub_ps(a, b);
23 }
24
glm_vec1_sub(glm_vec4 a,glm_vec4 b)25 GLM_FUNC_QUALIFIER glm_vec4 glm_vec1_sub(glm_vec4 a, glm_vec4 b)
26 {
27 return _mm_sub_ss(a, b);
28 }
29
glm_vec4_mul(glm_vec4 a,glm_vec4 b)30 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_mul(glm_vec4 a, glm_vec4 b)
31 {
32 return _mm_mul_ps(a, b);
33 }
34
glm_vec1_mul(glm_vec4 a,glm_vec4 b)35 GLM_FUNC_QUALIFIER glm_vec4 glm_vec1_mul(glm_vec4 a, glm_vec4 b)
36 {
37 return _mm_mul_ss(a, b);
38 }
39
glm_vec4_div(glm_vec4 a,glm_vec4 b)40 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_div(glm_vec4 a, glm_vec4 b)
41 {
42 return _mm_div_ps(a, b);
43 }
44
glm_vec1_div(glm_vec4 a,glm_vec4 b)45 GLM_FUNC_QUALIFIER glm_vec4 glm_vec1_div(glm_vec4 a, glm_vec4 b)
46 {
47 return _mm_div_ss(a, b);
48 }
49
glm_vec4_div_lowp(glm_vec4 a,glm_vec4 b)50 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_div_lowp(glm_vec4 a, glm_vec4 b)
51 {
52 return glm_vec4_mul(a, _mm_rcp_ps(b));
53 }
54
glm_vec4_swizzle_xyzw(glm_vec4 a)55 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_swizzle_xyzw(glm_vec4 a)
56 {
57 # if GLM_ARCH & GLM_ARCH_AVX2_BIT
58 return _mm_permute_ps(a, _MM_SHUFFLE(3, 2, 1, 0));
59 # else
60 return _mm_shuffle_ps(a, a, _MM_SHUFFLE(3, 2, 1, 0));
61 # endif
62 }
63
glm_vec1_fma(glm_vec4 a,glm_vec4 b,glm_vec4 c)64 GLM_FUNC_QUALIFIER glm_vec4 glm_vec1_fma(glm_vec4 a, glm_vec4 b, glm_vec4 c)
65 {
66 # if GLM_ARCH & GLM_ARCH_AVX2_BIT
67 return _mm_fmadd_ss(a, b, c);
68 # else
69 return _mm_add_ss(_mm_mul_ss(a, b), c);
70 # endif
71 }
72
glm_vec4_fma(glm_vec4 a,glm_vec4 b,glm_vec4 c)73 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_fma(glm_vec4 a, glm_vec4 b, glm_vec4 c)
74 {
75 # if GLM_ARCH & GLM_ARCH_AVX2_BIT
76 return _mm_fmadd_ps(a, b, c);
77 # else
78 return glm_vec4_add(glm_vec4_mul(a, b), c);
79 # endif
80 }
81
glm_vec4_abs(glm_vec4 x)82 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_abs(glm_vec4 x)
83 {
84 return _mm_and_ps(x, _mm_castsi128_ps(_mm_set1_epi32(0x7FFFFFFF)));
85 }
86
glm_ivec4_abs(glm_ivec4 x)87 GLM_FUNC_QUALIFIER glm_ivec4 glm_ivec4_abs(glm_ivec4 x)
88 {
89 # if GLM_ARCH & GLM_ARCH_SSSE3_BIT
90 return _mm_sign_epi32(x, x);
91 # else
92 glm_ivec4 const sgn0 = _mm_srai_epi32(x, 31);
93 glm_ivec4 const inv0 = _mm_xor_si128(x, sgn0);
94 glm_ivec4 const sub0 = _mm_sub_epi32(inv0, sgn0);
95 return sub0;
96 # endif
97 }
98
glm_vec4_sign(glm_vec4 x)99 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_sign(glm_vec4 x)
100 {
101 glm_vec4 const zro0 = _mm_setzero_ps();
102 glm_vec4 const cmp0 = _mm_cmplt_ps(x, zro0);
103 glm_vec4 const cmp1 = _mm_cmpgt_ps(x, zro0);
104 glm_vec4 const and0 = _mm_and_ps(cmp0, _mm_set1_ps(-1.0f));
105 glm_vec4 const and1 = _mm_and_ps(cmp1, _mm_set1_ps(1.0f));
106 glm_vec4 const or0 = _mm_or_ps(and0, and1);;
107 return or0;
108 }
109
glm_vec4_round(glm_vec4 x)110 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_round(glm_vec4 x)
111 {
112 # if GLM_ARCH & GLM_ARCH_SSE41_BIT
113 return _mm_round_ps(x, _MM_FROUND_TO_NEAREST_INT);
114 # else
115 glm_vec4 const sgn0 = _mm_castsi128_ps(_mm_set1_epi32(0x80000000));
116 glm_vec4 const and0 = _mm_and_ps(sgn0, x);
117 glm_vec4 const or0 = _mm_or_ps(and0, _mm_set_ps1(8388608.0f));
118 glm_vec4 const add0 = glm_vec4_add(x, or0);
119 glm_vec4 const sub0 = glm_vec4_sub(add0, or0);
120 return sub0;
121 # endif
122 }
123
glm_vec4_floor(glm_vec4 x)124 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_floor(glm_vec4 x)
125 {
126 # if GLM_ARCH & GLM_ARCH_SSE41_BIT
127 return _mm_floor_ps(x);
128 # else
129 glm_vec4 const rnd0 = glm_vec4_round(x);
130 glm_vec4 const cmp0 = _mm_cmplt_ps(x, rnd0);
131 glm_vec4 const and0 = _mm_and_ps(cmp0, _mm_set1_ps(1.0f));
132 glm_vec4 const sub0 = glm_vec4_sub(rnd0, and0);
133 return sub0;
134 # endif
135 }
136
137 /* trunc TODO
138 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_trunc(glm_vec4 x)
139 {
140 return glm_vec4();
141 }
142 */
143
144 //roundEven
glm_vec4_roundEven(glm_vec4 x)145 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_roundEven(glm_vec4 x)
146 {
147 glm_vec4 const sgn0 = _mm_castsi128_ps(_mm_set1_epi32(0x80000000));
148 glm_vec4 const and0 = _mm_and_ps(sgn0, x);
149 glm_vec4 const or0 = _mm_or_ps(and0, _mm_set_ps1(8388608.0f));
150 glm_vec4 const add0 = glm_vec4_add(x, or0);
151 glm_vec4 const sub0 = glm_vec4_sub(add0, or0);
152 return sub0;
153 }
154
glm_vec4_ceil(glm_vec4 x)155 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_ceil(glm_vec4 x)
156 {
157 # if GLM_ARCH & GLM_ARCH_SSE41_BIT
158 return _mm_ceil_ps(x);
159 # else
160 glm_vec4 const rnd0 = glm_vec4_round(x);
161 glm_vec4 const cmp0 = _mm_cmpgt_ps(x, rnd0);
162 glm_vec4 const and0 = _mm_and_ps(cmp0, _mm_set1_ps(1.0f));
163 glm_vec4 const add0 = glm_vec4_add(rnd0, and0);
164 return add0;
165 # endif
166 }
167
glm_vec4_fract(glm_vec4 x)168 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_fract(glm_vec4 x)
169 {
170 glm_vec4 const flr0 = glm_vec4_floor(x);
171 glm_vec4 const sub0 = glm_vec4_sub(x, flr0);
172 return sub0;
173 }
174
glm_vec4_mod(glm_vec4 x,glm_vec4 y)175 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_mod(glm_vec4 x, glm_vec4 y)
176 {
177 glm_vec4 const div0 = glm_vec4_div(x, y);
178 glm_vec4 const flr0 = glm_vec4_floor(div0);
179 glm_vec4 const mul0 = glm_vec4_mul(y, flr0);
180 glm_vec4 const sub0 = glm_vec4_sub(x, mul0);
181 return sub0;
182 }
183
glm_vec4_clamp(glm_vec4 v,glm_vec4 minVal,glm_vec4 maxVal)184 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_clamp(glm_vec4 v, glm_vec4 minVal, glm_vec4 maxVal)
185 {
186 glm_vec4 const min0 = _mm_min_ps(v, maxVal);
187 glm_vec4 const max0 = _mm_max_ps(min0, minVal);
188 return max0;
189 }
190
glm_vec4_mix(glm_vec4 v1,glm_vec4 v2,glm_vec4 a)191 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_mix(glm_vec4 v1, glm_vec4 v2, glm_vec4 a)
192 {
193 glm_vec4 const sub0 = glm_vec4_sub(_mm_set1_ps(1.0f), a);
194 glm_vec4 const mul0 = glm_vec4_mul(v1, sub0);
195 glm_vec4 const mad0 = glm_vec4_fma(v2, a, mul0);
196 return mad0;
197 }
198
glm_vec4_step(glm_vec4 edge,glm_vec4 x)199 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_step(glm_vec4 edge, glm_vec4 x)
200 {
201 glm_vec4 const cmp = _mm_cmple_ps(x, edge);
202 return _mm_movemask_ps(cmp) == 0 ? _mm_set1_ps(1.0f) : _mm_setzero_ps();
203 }
204
glm_vec4_smoothstep(glm_vec4 edge0,glm_vec4 edge1,glm_vec4 x)205 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_smoothstep(glm_vec4 edge0, glm_vec4 edge1, glm_vec4 x)
206 {
207 glm_vec4 const sub0 = glm_vec4_sub(x, edge0);
208 glm_vec4 const sub1 = glm_vec4_sub(edge1, edge0);
209 glm_vec4 const div0 = glm_vec4_sub(sub0, sub1);
210 glm_vec4 const clp0 = glm_vec4_clamp(div0, _mm_setzero_ps(), _mm_set1_ps(1.0f));
211 glm_vec4 const mul0 = glm_vec4_mul(_mm_set1_ps(2.0f), clp0);
212 glm_vec4 const sub2 = glm_vec4_sub(_mm_set1_ps(3.0f), mul0);
213 glm_vec4 const mul1 = glm_vec4_mul(clp0, clp0);
214 glm_vec4 const mul2 = glm_vec4_mul(mul1, sub2);
215 return mul2;
216 }
217
218 // Agner Fog method
glm_vec4_nan(glm_vec4 x)219 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_nan(glm_vec4 x)
220 {
221 glm_ivec4 const t1 = _mm_castps_si128(x); // reinterpret as 32-bit integer
222 glm_ivec4 const t2 = _mm_sll_epi32(t1, _mm_cvtsi32_si128(1)); // shift out sign bit
223 glm_ivec4 const t3 = _mm_set1_epi32(0xFF000000); // exponent mask
224 glm_ivec4 const t4 = _mm_and_si128(t2, t3); // exponent
225 glm_ivec4 const t5 = _mm_andnot_si128(t3, t2); // fraction
226 glm_ivec4 const Equal = _mm_cmpeq_epi32(t3, t4);
227 glm_ivec4 const Nequal = _mm_cmpeq_epi32(t5, _mm_setzero_si128());
228 glm_ivec4 const And = _mm_and_si128(Equal, Nequal);
229 return _mm_castsi128_ps(And); // exponent = all 1s and fraction != 0
230 }
231
232 // Agner Fog method
glm_vec4_inf(glm_vec4 x)233 GLM_FUNC_QUALIFIER glm_vec4 glm_vec4_inf(glm_vec4 x)
234 {
235 glm_ivec4 const t1 = _mm_castps_si128(x); // reinterpret as 32-bit integer
236 glm_ivec4 const t2 = _mm_sll_epi32(t1, _mm_cvtsi32_si128(1)); // shift out sign bit
237 return _mm_castsi128_ps(_mm_cmpeq_epi32(t2, _mm_set1_epi32(0xFF000000))); // exponent is all 1s, fraction is 0
238 }
239
240 #endif//GLM_ARCH & GLM_ARCH_SSE2_BIT
241