1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
|
/*
* Copyright (c), Recep Aslantas.
*
* MIT License (MIT), http://opensource.org/licenses/MIT
* Full license can be found in the LICENSE file
*/
#ifndef cglm_simd_wasm_h
#define cglm_simd_wasm_h
#include "intrin.h"
#ifdef CGLM_SIMD_WASM
#include <wasm_simd128.h>
#define glmm_load(p) wasm_v128_load(p)
#define glmm_store(p, a) wasm_v128_store(p, (a))
#define glmm_set1(x) wasm_f32x4_splat(x)
#define glmm_set1_ptr(x) wasm_f32x4_splat(*x)
#define glmm_set1_rval(x) wasm_f32x4_splat(x)
#define glmm_128 v128_t
#define glmm_shuff1(xmm, z, y, x, w) wasm_i32x4_shuffle(xmm, xmm, w, x, y, z)
#define glmm_splat(x, lane) glmm_shuff1(x, lane, lane, lane, lane)
#define glmm_splat_x(x) glmm_splat(x, 0)
#define glmm_splat_y(x) glmm_splat(x, 1)
#define glmm_splat_z(x) glmm_splat(x, 2)
#define glmm_splat_w(x) glmm_splat(x, 3)
#define GLMM_NEGZEROf 0x80000000 /* 0x80000000 ---> -0.0f */
/* _mm_set_ps(X, Y, Z, W); */
#define GLMM__SIGNMASKf(X, Y, Z, W) wasm_i32x4_const(X, Y, Z, W)
#define glmm_float32x4_SIGNMASK_PNPN GLMM__SIGNMASKf(0, GLMM_NEGZEROf, 0, GLMM_NEGZEROf)
#define glmm_float32x4_SIGNMASK_NPNP GLMM__SIGNMASKf(GLMM_NEGZEROf, 0, GLMM_NEGZEROf, 0)
#define glmm_float32x4_SIGNMASK_NPPN GLMM__SIGNMASKf(GLMM_NEGZEROf, 0, 0, GLMM_NEGZEROf)
#define glmm_float32x4_SIGNMASK_NEG wasm_i32x4_const_splat(GLMM_NEGZEROf)
static inline glmm_128 glmm_abs(glmm_128 x) { return wasm_f32x4_abs(x); }
static inline glmm_128 glmm_min(glmm_128 a, glmm_128 b) { return wasm_f32x4_pmin(b, a); }
static inline glmm_128 glmm_max(glmm_128 a, glmm_128 b) { return wasm_f32x4_pmax(b, a); }
static inline
glmm_128
glmm_vhadd(glmm_128 v) {
glmm_128 x0;
x0 = wasm_f32x4_add(v, glmm_shuff1(v, 0, 1, 2, 3));
x0 = wasm_f32x4_add(x0, glmm_shuff1(x0, 1, 0, 0, 1));
return x0;
}
static inline
glmm_128
glmm_vhadds(glmm_128 v) {
glmm_128 shuf, sums;
shuf = glmm_shuff1(v, 2, 3, 0, 1);
sums = wasm_f32x4_add(v, shuf);
/* shuf = _mm_movehl_ps(shuf, sums); */
shuf = wasm_i32x4_shuffle(shuf, sums, 6, 7, 2, 3);
sums = wasm_i32x4_shuffle(sums, wasm_f32x4_add(sums, shuf), 4, 1, 2, 3);
return sums;
}
static inline
float
glmm_hadd(glmm_128 v) {
return wasm_f32x4_extract_lane(glmm_vhadds(v), 0);
}
static inline
glmm_128
glmm_vhmin(glmm_128 v) {
glmm_128 x0, x1, x2;
x0 = glmm_shuff1(v, 2, 3, 2, 3); /* [2, 3, 2, 3] */
x1 = wasm_f32x4_pmin(x0, v); /* [0|2, 1|3, 2|2, 3|3] */
x2 = glmm_splat(x1, 1); /* [1|3, 1|3, 1|3, 1|3] */
return wasm_f32x4_pmin(x1, x2);
}
static inline
float
glmm_hmin(glmm_128 v) {
return wasm_f32x4_extract_lane(glmm_vhmin(v), 0);
}
static inline
glmm_128
glmm_vhmax(glmm_128 v) {
glmm_128 x0, x1, x2;
x0 = glmm_shuff1(v, 2, 3, 2, 3); /* [2, 3, 2, 3] */
x1 = wasm_f32x4_pmax(x0, v); /* [0|2, 1|3, 2|2, 3|3] */
x2 = glmm_splat(x1, 1); /* [1|3, 1|3, 1|3, 1|3] */
/* _mm_max_ss */
return wasm_i32x4_shuffle(x1, wasm_f32x4_pmax(x1, x2), 4, 1, 2, 3);
}
static inline
float
glmm_hmax(glmm_128 v) {
return wasm_f32x4_extract_lane(glmm_vhmax(v), 0);
}
static inline
glmm_128
glmm_vdots(glmm_128 a, glmm_128 b) {
return glmm_vhadds(wasm_f32x4_mul(a, b));
}
static inline
glmm_128
glmm_vdot(glmm_128 a, glmm_128 b) {
glmm_128 x0;
x0 = wasm_f32x4_mul(a, b);
x0 = wasm_f32x4_add(x0, glmm_shuff1(x0, 1, 0, 3, 2));
return wasm_f32x4_add(x0, glmm_shuff1(x0, 0, 1, 0, 1));
}
static inline
float
glmm_dot(glmm_128 a, glmm_128 b) {
return wasm_f32x4_extract_lane(glmm_vdots(a, b), 0);
}
static inline
float
glmm_norm(glmm_128 a) {
glmm_128 x0;
x0 = glmm_vhadds(wasm_f32x4_mul(a, a));
return wasm_f32x4_extract_lane(
wasm_i32x4_shuffle(x0, wasm_f32x4_sqrt(x0),4, 1, 2, 3), 0);
}
static inline
float
glmm_norm2(glmm_128 a) {
return wasm_f32x4_extract_lane(glmm_vhadds(wasm_f32x4_mul(a, a)), 0);
}
static inline
float
glmm_norm_one(glmm_128 a) {
return wasm_f32x4_extract_lane(glmm_vhadds(glmm_abs(a)), 0);
}
static inline
float
glmm_norm_inf(glmm_128 a) {
return wasm_f32x4_extract_lane(glmm_vhmax(glmm_abs(a)), 0);
}
static inline
glmm_128
glmm_load3(float v[3]) {
glmm_128 xy = wasm_v128_load64_zero(v);
return wasm_f32x4_replace_lane(xy, 2, v[2]);
}
static inline
void
glmm_store3(float v[3], glmm_128 vx) {
wasm_v128_store64_lane(v, vx, 0);
wasm_v128_store32_lane(&v[2], vx, 2);
}
static inline
glmm_128
glmm_div(glmm_128 a, glmm_128 b) {
return wasm_f32x4_div(a, b);
}
static inline
glmm_128
glmm_fmadd(glmm_128 a, glmm_128 b, glmm_128 c) {
return wasm_f32x4_add(c, wasm_f32x4_mul(a, b));
}
static inline
glmm_128
glmm_fnmadd(glmm_128 a, glmm_128 b, glmm_128 c) {
return wasm_f32x4_sub(c, wasm_f32x4_mul(a, b));
}
static inline
glmm_128
glmm_fmsub(glmm_128 a, glmm_128 b, glmm_128 c) {
return wasm_f32x4_sub(wasm_f32x4_mul(a, b), c);
}
static inline
glmm_128
glmm_fnmsub(glmm_128 a, glmm_128 b, glmm_128 c) {
return wasm_f32x4_neg(wasm_f32x4_add(wasm_f32x4_mul(a, b), c));
}
#endif
#endif /* cglm_simd_wasm_h */
|
