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
|
/* $OpenBSD: vecast.c,v 1.1 2022/10/22 17:50:28 gkoehler Exp $ */
/*
* Copyright (c) 2022 George Koehler <gkoehler@openbsd.org>
*
* Permission to use, copy, modify, and distribute this software for any
* purpose with or without fee is hereby granted, provided that the above
* copyright notice and this permission notice appear in all copies.
*
* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
*/
#include <altivec.h>
#include <err.h>
#include <math.h>
#include <stdio.h>
struct double4 {
double d[4];
};
union vu {
vector float vf;
vector int vi;
vector unsigned vu;
float f[4];
int i[4];
unsigned u[4];
};
#define AD(a, b, c, d) (struct double4){a, b, c, d}
#define VF(a, b, c, d) (vector float)(a, b, c, d)
#define VI(a, b, c, d) (vector int)(a, b, c, d)
#define VU(a, b, c, d) (vector unsigned)(a, b, c, d)
#define rsqrt(f) (1.0 / sqrt(f))
int fail;
void
ck_equal(const char *what, vector float out, vector float answer)
{
if (vec_any_ne(out, answer)) {
union vu a, b;
a.vf = out;
b.vf = answer;
warnx("%s: {%a, %a, %a, %a} should be {%a, %a, %a, %a}",
what, a.f[0], a.f[1], a.f[2], a.f[3],
b.f[0], b.f[1], b.f[2], b.f[3]);
fail = 1;
}
}
void
ck_equal_i(const char *what, vector int out, vector int answer)
{
if (vec_any_ne(out, answer)) {
union vu a, b;
a.vi = out;
b.vi = answer;
warnx("%s: {%d, %d, %d, %d} should be {%d, %d, %d, %d}",
what, a.i[0], a.i[1], a.i[2], a.i[3],
b.i[0], b.i[1], b.i[2], b.i[3]);
fail = 1;
}
}
void
ck_equal_u(const char *what, vector unsigned out, vector unsigned answer)
{
if (vec_any_ne(out, answer)) {
union vu a, b;
a.vi = out;
b.vi = answer;
warnx("%s: {%u, %u, %u, %u} should be {%u, %u, %u, %u}",
what, a.u[0], a.u[1], a.u[2], a.u[3],
b.u[0], b.u[1], b.u[2], b.u[3]);
fail = 1;
}
}
enum error_check {REL_1_IN, ABS_1_IN};
/* Checks that error is at most 1 in err_den. */
void
ck_estimate(const char *what, vector float out, struct double4 answer,
enum error_check which, double err_den)
{
union vu u;
int i, warned = 0;
u.vf = out;
for (i = 0; i < 4; i++) {
double estimate = u.f[i];
double target = answer.d[i];
double error;
switch (which) {
case REL_1_IN: /* relative error */
error = fabs(target / (estimate - target));
break;
case ABS_1_IN: /* absolute error */
error = fabs(1 / (estimate - target));
break;
default:
errx(1, "invalid check");
}
if (error < err_den) {
if (!warned) {
warnx("%s: {%a, %a, %a, %a} should be "
"near {%a, %a, %a, %a} (1/%g)",
what, u.f[0], u.f[1], u.f[2], u.f[3],
answer.d[0], answer.d[1], answer.d[2],
answer.d[3], err_den);
warned = 1;
fail = 1;
}
warnx("%a is off %a by 1/%g", estimate, target,
error);
}
}
}
/*
* Tries altivec with denormal or subnormal floats.
* These are single-precision floats f, where
* 0 < |f| < 2**-126 = 0x1p-126 = 0x10p-130 = 1.17549435E-38F
*/
int
main(void)
{
struct double4 dan;
volatile vector float in1, in2, in3;
vector float ans;
vector int ian;
vector unsigned uan;
/* in1 + in2 */
in1 = VF(10, 0x10p-140, 0x20p-130, -0x2000p-134);
in2 = VF( 4, 0x5p-140, -0x1p-130, 0x1fffp-134);
ans = VF(14, 0x15p-140, 0x1fp-130, -0x1p-134);
ck_equal("vec_add", vec_add(in1, in2), ans);
/* in1 - in2 */
in1 = VF(0x4000p-134, 10, 0x10p-140, 0x3p-130);
in2 = VF(0x3ffep-134, 4, 0x5p-140, 0x40p-130);
ans = VF( 0x2p-134, 6, 0xbp-140, -0x3dp-130);
ck_equal("vec_sub", vec_sub(in1, in2), ans);
/* in1 * in2 + in3 */
in1 = VF( 0x6p-70, 0x6p-140, 6, 0x6p-100);
in2 = VF( 0x7p-70, 0x7p50, 7, 0x7p-30);
in3 = VF( 0, 0, 1, -0x20p-130);
ans = VF(0x2ap-140, 0x2ap-90, 43, 0xap-130);
ck_equal("vec_madd", vec_madd(in1, in2, in3), ans);
/* in3 - in1 * in2 */
in1 = VF( 0xbp-30, 0xbp-70, 0xbp44, 11);
in2 = VF( 0x3p-100, 0x3p-70, -0x3p-138, 3);
in3 = VF(0x25p-130, 0, 0, 35);
ans = VF( 0x4p-130, -0x21p-140, 0x21p-94, 2);
ck_equal("vec_nmsub", vec_nmsub(in1, in2, in3), ans);
/* 1 / in1 */
in1 = VF( 3, 0x3p126, 0x3p-126, 0x1p127);
dan = AD(1.0 / 3, 1.0 / 0x3p126, 1.0 / 0x3p-126, 0x1p-127);
ck_estimate("vec_re", vec_re(in1), dan, REL_1_IN, 4096);
/* 1 / sqrt(in1) */
in1 = VF(1, 2, 0x1p-128, 0x5p-135);
dan = AD(1, rsqrt(2), rsqrt(0x1p-128), rsqrt(0x5p-135));
ck_estimate("vec_rsqrt", vec_rsqrte(in1), dan, REL_1_IN, 4096);
/* log2(in1) */
in1 = VF(0x1p-130, 0x1p-149, 32, 0x1p-10);
dan = AD( -130, -149, 5, -10);
ck_estimate("vec_loge", vec_loge(in1), dan, ABS_1_IN, 32);
in1 = VF( 0x123p-139, 0xabcp-145, 1, 1);
dan = AD(log2(0x123p-139), log2(0xabcp-145), 0, 0);
ck_estimate("vec_loge", vec_loge(in1), dan, ABS_1_IN, 32);
/* 2**in1 */
in1 = VF( -149, -138, -127, 10);
ans = VF(0x1p-149, 0x1p-138, 0x1p-127, 1024);
ck_equal("vec_expte", vec_expte(in1), ans);
in1 = VF( -10, -145.3, -136.9, -127.1);
dan = AD(0x1p-10, exp2(-145.3), exp2(-136.9), exp2(-127.1));
ck_estimate("vec_expte", vec_expte(in1), dan, REL_1_IN, 16);
/* (int)(in1 * 2**exponent) */
in1 = VF(0x1p-127, 2.34, -0xfedp-140, -19.8);
ian = VI( 0, 2, 0, -19);
ck_equal_i("vec_cts", vec_cts(in1, 0), ian);
in1 = VF(0x1p-113, -1, -0xabcp-143, 0x1fp-10);
ian = VI( 0, -1024, 0, 0x1f);
ck_equal_i("vec_cts", vec_cts(in1, 10), ian);
/* (unsigned)(in1 * 2**exponent) */
in1 = VF(0x1.ap-130, 0x1.ep-140, 24000012, 0);
uan = VU( 0, 0, 3072001536, 0);
ck_equal_u("vec_ctu", vec_ctu(in1, 7), uan);
return fail;
}
|
