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
|
#include "Halide.h"
using namespace Halide;
template<typename T>
void test() {
{
// Test div_round_to_zero
Func f;
Var x, y;
Expr d = cast<T>(y - 128);
Expr n = cast<T>(x - 128);
d = select(d == 0 || (d == -1 && n == d.type().min()),
cast<T>(1),
d);
f(x, y) = div_round_to_zero(n, d);
f.vectorize(x, 8);
Buffer<T> result = f.realize({256, 256});
for (int d = -128; d < 128; d++) {
if (d == 0) {
continue;
}
for (int n = -128; n < 128; n++) {
if (d == -1 && n == std::numeric_limits<T>::min()) {
continue;
}
int correct = d == 0 ? n : (T)(n / d);
int r = result(n + 128, d + 128);
if (r != correct) {
printf("result(%d, %d) = %d instead of %d\n", n, d, r, correct);
exit(1);
}
}
}
}
{
// Test the fast version
Func f;
Var x, y;
f(x, y) = fast_integer_divide_round_to_zero(cast<T>(x - 128), cast<uint8_t>(y + 1));
f.vectorize(x, 8);
Buffer<T> result_fast = f.realize({256, 255});
for (int d = 1; d < 256; d++) {
for (int n = -128; n < 128; n++) {
int correct = (T)(n / d);
int r = result_fast(n + 128, d - 1);
if (r != correct) {
printf("result_fast(%d, %d) = %d instead of %d\n", n, d, r, correct);
exit(1);
}
}
}
}
{
// Try some constant denominators
for (int d : {-128, -54, -3, -1, 1, 2, 25, 32, 127}) {
if (d == 0) {
continue;
}
Func f;
Var x;
f(x) = div_round_to_zero(cast<T>(x - 128), cast<T>(d));
f.vectorize(x, 8);
Buffer<T> result_const = f.realize({256});
for (int n = -128; n < 128; n++) {
int correct = (T)(n / d);
int r = result_const(n + 128);
if (r != correct) {
printf("result_const(%d, %d) = %d instead of %d\n", n, d, r, correct);
exit(1);
}
}
}
}
}
int main(int argc, char **argv) {
test<int8_t>();
test<int16_t>();
test<int32_t>();
printf("Success!\n");
return 0;
}
|
