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
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
|
#include <assert.h>
#include <emscripten.h>
#include <math.h>
#include <stdio.h>
#include <wasm_simd128.h>
// SIMD True/False lane values.
#define TRUE 0xFFFF
#define FALSE 0
// Not used below, but helpful for debugging.
void print_f16x8(v128_t v) {
printf("[%g, %g, %g, %g, %g, %g, %g, %g]\n",
wasm_f16x8_extract_lane(v, 0),
wasm_f16x8_extract_lane(v, 1),
wasm_f16x8_extract_lane(v, 2),
wasm_f16x8_extract_lane(v, 3),
wasm_f16x8_extract_lane(v, 4),
wasm_f16x8_extract_lane(v, 5),
wasm_f16x8_extract_lane(v, 6),
wasm_f16x8_extract_lane(v, 7));
}
void print_i16x8(v128_t v) {
printf("[%d, %d, %d, %d, %d, %d, %d, %d]\n",
wasm_i16x8_extract_lane(v, 0),
wasm_i16x8_extract_lane(v, 1),
wasm_i16x8_extract_lane(v, 2),
wasm_i16x8_extract_lane(v, 3),
wasm_i16x8_extract_lane(v, 4),
wasm_i16x8_extract_lane(v, 5),
wasm_i16x8_extract_lane(v, 6),
wasm_i16x8_extract_lane(v, 7));
}
void print_u16x8(v128_t v) {
printf("[%d, %d, %d, %d, %d, %d, %d, %d]\n",
wasm_u16x8_extract_lane(v, 0),
wasm_u16x8_extract_lane(v, 1),
wasm_u16x8_extract_lane(v, 2),
wasm_u16x8_extract_lane(v, 3),
wasm_u16x8_extract_lane(v, 4),
wasm_u16x8_extract_lane(v, 5),
wasm_u16x8_extract_lane(v, 6),
wasm_u16x8_extract_lane(v, 7));
}
void assert_lanes_eq_uint16(v128_t v,
uint16_t l0,
uint16_t l1,
uint16_t l2,
uint16_t l3,
uint16_t l4,
uint16_t l5,
uint16_t l6,
uint16_t l7) {
assert(wasm_u16x8_extract_lane(v, 0) == l0);
assert(wasm_u16x8_extract_lane(v, 1) == l1);
assert(wasm_u16x8_extract_lane(v, 2) == l2);
assert(wasm_u16x8_extract_lane(v, 3) == l3);
assert(wasm_u16x8_extract_lane(v, 4) == l4);
assert(wasm_u16x8_extract_lane(v, 5) == l5);
assert(wasm_u16x8_extract_lane(v, 6) == l6);
assert(wasm_u16x8_extract_lane(v, 7) == l7);
}
void assert_lanes_eq_int16(v128_t v,
int16_t l0,
int16_t l1,
int16_t l2,
int16_t l3,
int16_t l4,
int16_t l5,
int16_t l6,
int16_t l7) {
assert(wasm_i16x8_extract_lane(v, 0) == l0);
assert(wasm_i16x8_extract_lane(v, 1) == l1);
assert(wasm_i16x8_extract_lane(v, 2) == l2);
assert(wasm_i16x8_extract_lane(v, 3) == l3);
assert(wasm_i16x8_extract_lane(v, 4) == l4);
assert(wasm_i16x8_extract_lane(v, 5) == l5);
assert(wasm_i16x8_extract_lane(v, 6) == l6);
assert(wasm_i16x8_extract_lane(v, 7) == l7);
}
void assert_lanes_eq_float(v128_t v,
float l0,
float l1,
float l2,
float l3,
float l4,
float l5,
float l6,
float l7) {
// We can't do a direct comparision since there may be quiet or signaling nan
// values, so if the expected value is nan check that the actual value is nan
// too.
float actual;
actual = wasm_f16x8_extract_lane(v, 0);
assert(isnan(l0) ? isnan(actual) : actual == l0);
actual = wasm_f16x8_extract_lane(v, 1);
assert(isnan(l1) ? isnan(actual) : actual == l1);
actual = wasm_f16x8_extract_lane(v, 2);
assert(isnan(l2) ? isnan(actual) : actual == l2);
actual = wasm_f16x8_extract_lane(v, 3);
assert(isnan(l3) ? isnan(actual) : actual == l3);
actual = wasm_f16x8_extract_lane(v, 4);
assert(isnan(l4) ? isnan(actual) : actual == l4);
actual = wasm_f16x8_extract_lane(v, 5);
assert(isnan(l5) ? isnan(actual) : actual == l5);
actual = wasm_f16x8_extract_lane(v, 6);
assert(isnan(l6) ? isnan(actual) : actual == l6);
actual = wasm_f16x8_extract_lane(v, 7);
assert(isnan(l7) ? isnan(actual) : actual == l7);
}
void assert_all_lanes_eq(v128_t v, float value) {
assert_lanes_eq_float(
v, value, value, value, value, value, value, value, value);
}
v128_t create_f16x8(float l0,
float l1,
float l2,
float l3,
float l4,
float l5,
float l6,
float l7) {
v128_t v;
v = wasm_f16x8_replace_lane(v, 0, l0);
v = wasm_f16x8_replace_lane(v, 1, l1);
v = wasm_f16x8_replace_lane(v, 2, l2);
v = wasm_f16x8_replace_lane(v, 3, l3);
v = wasm_f16x8_replace_lane(v, 4, l4);
v = wasm_f16x8_replace_lane(v, 5, l5);
v = wasm_f16x8_replace_lane(v, 6, l6);
v = wasm_f16x8_replace_lane(v, 7, l7);
return v;
}
v128_t create_i16x8(int16_t l0,
int16_t l1,
int16_t l2,
int16_t l3,
int16_t l4,
int16_t l5,
int16_t l6,
int16_t l7) {
v128_t v;
v = wasm_i16x8_replace_lane(v, 0, l0);
v = wasm_i16x8_replace_lane(v, 1, l1);
v = wasm_i16x8_replace_lane(v, 2, l2);
v = wasm_i16x8_replace_lane(v, 3, l3);
v = wasm_i16x8_replace_lane(v, 4, l4);
v = wasm_i16x8_replace_lane(v, 5, l5);
v = wasm_i16x8_replace_lane(v, 6, l6);
v = wasm_i16x8_replace_lane(v, 7, l7);
return v;
}
v128_t create_u16x8(uint16_t l0,
uint16_t l1,
uint16_t l2,
uint16_t l3,
uint16_t l4,
uint16_t l5,
uint16_t l6,
uint16_t l7) {
v128_t v;
v = wasm_u16x8_replace_lane(v, 0, l0);
v = wasm_u16x8_replace_lane(v, 1, l1);
v = wasm_u16x8_replace_lane(v, 2, l2);
v = wasm_u16x8_replace_lane(v, 3, l3);
v = wasm_u16x8_replace_lane(v, 4, l4);
v = wasm_u16x8_replace_lane(v, 5, l5);
v = wasm_u16x8_replace_lane(v, 6, l6);
v = wasm_u16x8_replace_lane(v, 7, l7);
return v;
}
int main() {
v128_t a, b, c;
a = wasm_f16x8_splat(2.0f);
assert_all_lanes_eq(a, 2.0f);
a = wasm_f16x8_splat(1.0f);
a = wasm_f16x8_replace_lane(a, 0, 99.0f);
a = wasm_f16x8_replace_lane(a, 7, 99.0f);
assert_lanes_eq_float(a, 99.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 99.0f);
// The following tests are not exhaustive, but try to hit some interesting
// floats and how they interact.
a = wasm_f16x8_abs(create_f16x8(-1.0f, 1.0f, 0.0f, -0.0f, INFINITY, -INFINITY, NAN, -NAN));
assert_lanes_eq_float(a, 1.0f, 1.0f, 0.0f, 0.0f, INFINITY, INFINITY, NAN, NAN);
a = wasm_f16x8_neg(create_f16x8(-1.0f, 1.0f, 0.0f, -0.0f, INFINITY, -INFINITY, NAN, -NAN));
assert_lanes_eq_float(a, 1.0f, -1.0f, -0.0f, 0.0f, -INFINITY, INFINITY, NAN, NAN);
a = wasm_f16x8_sqrt(create_f16x8(-1.0f, 1.0f, 0.0f, 4.0f, INFINITY, -INFINITY, NAN, -NAN));
assert_lanes_eq_float(a, NAN, 1.0f, 0.0f, 2.0f, INFINITY, NAN, NAN, NAN);
a = wasm_f16x8_ceil(create_f16x8(-1.0f, 1.0f, -1.2f, 1.2f, -1.7f, 1.7f, INFINITY, -INFINITY));
assert_lanes_eq_float(a, -1.0f, 1.0f, -1.0f, 2.0f, -1.0f, 2.0f, INFINITY, -INFINITY);
a = wasm_f16x8_floor(create_f16x8(-1.0f, 1.0f, -1.2f, 1.2f, -1.7f, 1.7f, INFINITY, -INFINITY));
assert_lanes_eq_float(a, -1.0f, 1.0f, -2.0f, 1.0f, -2.0f, 1.0f, INFINITY, -INFINITY);
a = wasm_f16x8_trunc(create_f16x8(-1.0f, 1.0f, -1.2f, 1.2f, -1.7f, 1.7f, INFINITY, -INFINITY));
assert_lanes_eq_float(a, -1.0f, 1.0f, -1.0f, 1.0f, -1.0f, 1.0f, INFINITY, -INFINITY);
a = wasm_f16x8_nearest(create_f16x8(-1.0f, 1.0f, -1.2f, 1.2f, -1.7f, 1.7f, INFINITY, -INFINITY));
assert_lanes_eq_float(a, -1.0f, 1.0f, -1.0f, 1.0f, -2.0f, 2.0f, INFINITY, -INFINITY);
a = wasm_f16x8_eq(create_f16x8(-1.0f, 1.0f, 1.0f, -0.0f, INFINITY, -INFINITY, NAN, -NAN),
create_f16x8( 1.0f, -1.0f, 1.0f, -0.0f, INFINITY, -INFINITY, NAN, -NAN));
assert_lanes_eq_uint16(a, FALSE, FALSE, TRUE, TRUE, TRUE, TRUE, FALSE, FALSE);
a = wasm_f16x8_ne(create_f16x8(-1.0f, 1.0f, 1.0f, -0.0f, INFINITY, -INFINITY, NAN, -NAN),
create_f16x8( 1.0f, -1.0f, 1.0f, -0.0f, INFINITY, -INFINITY, NAN, -NAN));
assert_lanes_eq_uint16(a, TRUE, TRUE, FALSE, FALSE, FALSE, FALSE, TRUE, TRUE);
a = wasm_f16x8_lt(create_f16x8(-1.0f, 1.0f, 1.0f, -0.0f, INFINITY, -INFINITY, NAN, -NAN),
create_f16x8( 1.0f, -1.0f, 1.0f, -0.0f, INFINITY, -INFINITY, NAN, -NAN));
assert_lanes_eq_uint16(a, TRUE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE);
a = wasm_f16x8_gt(create_f16x8( -1.0f, 1.0f, 1.0f, -0.0f, INFINITY, -INFINITY, NAN, -NAN),
create_f16x8( 1.0f, -1.0f, 1.0f, -0.0f, INFINITY, -INFINITY, NAN, -NAN));
assert_lanes_eq_uint16(a, FALSE, TRUE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE);
a = wasm_f16x8_le(create_f16x8(-1.0f, 1.0f, 1.0f, -0.0f, INFINITY, -INFINITY, NAN, -NAN),
create_f16x8( 1.0f, -1.0f, 1.0f, -0.0f, INFINITY, -INFINITY, NAN, -NAN));
assert_lanes_eq_uint16(a, TRUE, FALSE, TRUE, TRUE, TRUE, TRUE, FALSE, FALSE);
a = wasm_f16x8_ge(create_f16x8(-1.0f, 1.0f, 1.0f, -0.0f, INFINITY, -INFINITY, NAN, -NAN),
create_f16x8( 1.0f, -1.0f, 1.0f, -0.0f, INFINITY, -INFINITY, NAN, -NAN));
assert_lanes_eq_uint16(a, FALSE, TRUE, TRUE, TRUE, TRUE, TRUE, FALSE, FALSE);
a = wasm_f16x8_add(create_f16x8(-1.0f, 1.0f, 1.0f, 1.0f, INFINITY, INFINITY, NAN, NAN),
create_f16x8( 1.0f, -1.0f, 1.0f, 0.0f, 1.0f, INFINITY, 1.0f, NAN));
assert_lanes_eq_float(a, 0.0f, 0.0f, 2.0f, 1.0f, INFINITY, INFINITY, NAN, NAN);
a = wasm_f16x8_sub(create_f16x8(-1.0f, 1.0f, 1.0f, 1.0f, INFINITY, INFINITY, NAN, NAN),
create_f16x8( 1.0f, -1.0f, 1.0f, 0.0f, 1.0f, INFINITY, 1.0f, NAN));
assert_lanes_eq_float(a, -2.0f, 2.0f, 0.0f, 1.0f, INFINITY, NAN, NAN, NAN);
a = wasm_f16x8_mul(create_f16x8(-1.0f, 1.0f, 1.0f, 1.0f, INFINITY, INFINITY, NAN, NAN),
create_f16x8( 1.0f, -1.0f, 1.0f, 0.0f, 1.0f, INFINITY, 1.0f, NAN));
assert_lanes_eq_float(a, -1.0f, -1.0f, 1.0f, 0.0f, INFINITY, INFINITY, NAN, NAN);
a = wasm_f16x8_div(create_f16x8(-1.0f, 1.0f, 1.0f, 1.0f, INFINITY, INFINITY, NAN, NAN),
create_f16x8( 1.0f, -1.0f, 1.0f, 0.0f, 1.0f, INFINITY, 1.0f, NAN));
assert_lanes_eq_float(a, -1.0f, -1.0f, 1.0f, INFINITY, INFINITY, NAN, NAN, NAN);
a = wasm_f16x8_min(create_f16x8(-1.0f, 1.0f, 1.0f, 1.0f, INFINITY, INFINITY, NAN, NAN),
create_f16x8( 1.0f, -1.0f, 1.0f, 0.0f, 1.0f, INFINITY, 1.0f, NAN));
assert_lanes_eq_float(a, -1.0f, -1.0f, 1.0f, 0.0f, 1.0f, INFINITY, NAN, NAN);
a = wasm_f16x8_max(create_f16x8(-1.0f, 1.0f, 1.0f, 1.0f, INFINITY, INFINITY, NAN, NAN),
create_f16x8( 1.0f, -1.0f, 1.0f, 0.0f, 1.0f, INFINITY, 1.0f, NAN));
assert_lanes_eq_float(a, 1.0f, 1.0f, 1.0f, 1.0f, INFINITY, INFINITY, NAN, NAN);
a = wasm_f16x8_pmin(create_f16x8(-1.0f, 1.0f, 1.0f, 1.0f, INFINITY, INFINITY, NAN, NAN),
create_f16x8( 1.0f, -1.0f, 1.0f, 0.0f, 1.0f, INFINITY, 1.0f, NAN));
assert_lanes_eq_float(a, -1.0f, -1.0f, 1.0f, 0.0f, 1.0f, INFINITY, NAN, NAN);
a = wasm_f16x8_pmax(create_f16x8(-1.0f, 1.0f, 1.0f, 1.0f, INFINITY, INFINITY, NAN, NAN),
create_f16x8( 1.0f, -1.0f, 1.0f, 0.0f, 1.0f, INFINITY, 1.0f, NAN));
assert_lanes_eq_float(a, 1.0f, 1.0f, 1.0f, 1.0f, INFINITY, INFINITY, NAN, NAN);
// TODO needed in binaryen
// wasm_i16x8_trunc_sat_f16x8;
// wasm_u16x8_trunc_sat_f16x8
// wasm_f16x8_convert_i16x8
// wasm_f16x8_convert_u16x8
// Lane 0 illustrates the difference between fused/unfused. e.g.
// fused: (positive overflow) + -inf = -inf
// unfused: (inf) + -inf = NaN
a = wasm_f16x8_relaxed_madd(create_f16x8( 1e4, INFINITY, -1.0f, 0.0f, 1.0f, 1.5f, -2.0f, 1.0f),
create_f16x8( 1e4, INFINITY, -1.0f, 0.0f, 1.0f, 1.5f, 4.0f, 1.0f),
create_f16x8(-INFINITY, INFINITY, -1.0f, 0.0f, 1.0f, 2.0f, 1.0f, -1.0f));
assert_lanes_eq_float(a, -INFINITY, INFINITY, 0.0f, 0.0f, 2.0f, 4.25f, -7.0f, 0.0f);
// Lane 0 illustrates the difference between fused/unfused. e.g.
// fused: -(positive overflow) + inf = inf
// unfused: (-inf) + -inf = NaN
a = wasm_f16x8_relaxed_nmadd(create_f16x8( 1e4, -INFINITY, -1.0f, 0.0f, 1.0f, 1.5f, -2.0f, 1.0f),
create_f16x8( 1e4, INFINITY, -1.0f, 0.0f, 1.0f, 1.5f, 4.0f, 1.0f),
create_f16x8(INFINITY, INFINITY, -1.0f, 0.0f, 1.0f, 2.0f, 1.0f, -1.0f));
assert_lanes_eq_float(a, INFINITY, INFINITY, -2.0f, 0.0f, 0.0f, -0.25f, 9.0f, -2.0f);
a = wasm_i16x8_trunc_sat_f16x8(create_f16x8(42.0f, -42.0f, NAN, INFINITY, -INFINITY, 65504.0f, -65504.0f, 0));
assert_lanes_eq_int16(a, 42, -42, 0, 32767, -32768, 32767, -32768, 0);
a = wasm_u16x8_trunc_sat_f16x8(create_f16x8(42.0f, -42.0f, NAN, INFINITY, -INFINITY, 65504.0f, -65504.0f, 0));
assert_lanes_eq_uint16(a, 42, 0, 0, 65535, 0, 65504, 0, 0);
// One confusing one below is 32767 which cannot be represented in FP16, so it becomes 32768.
a = wasm_f16x8_convert_i16x8(create_i16x8(0, 1, -1, 32767, -32768, 0, 0, 0));
assert_lanes_eq_float(a, 0, 1.0f, -1.0f, 32768.0f, -32768, 0, 0, 0);
// Another tricky one below is 65535, FP16 can't represent the max of uint16 so it becomes INFINITY.
a = wasm_f16x8_convert_u16x8(create_u16x8(0, 1, 65535, 65504U, 0, 0, 0, 0));
assert_lanes_eq_float(a, 0, 1.0f, INFINITY, 65504.0f, 0, 0, 0, 0);
}
|
