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
|
#include "Halide.h"
#include "halide_thread_pool.h"
#include "test_sharding.h"
#include <stdio.h>
using namespace Halide;
template<typename A>
const char *string_of_type();
#define DECL_SOT(name) \
template<> \
const char *string_of_type<name>() { \
return #name; \
}
DECL_SOT(uint8_t);
DECL_SOT(int8_t);
DECL_SOT(uint16_t);
DECL_SOT(int16_t);
DECL_SOT(uint32_t);
DECL_SOT(int32_t);
DECL_SOT(float);
DECL_SOT(double);
template<typename T>
bool is_type_supported(int vec_width, const Target &target) {
DeviceAPI device = DeviceAPI::Default_GPU;
if (target.has_feature(Target::HVX)) {
device = DeviceAPI::Hexagon;
}
if (target.has_feature(Target::Vulkan)) {
if (type_of<T>() == Float(64)) {
if ((target.os == Target::OSX || target.os == Target::IOS)) {
return false; // MoltenVK doesn't support Float64
}
}
}
return target.supports_type(type_of<T>().with_lanes(vec_width), device);
}
template<typename A, typename B>
bool test(int vec_width, const Target &target) {
// Useful for debugging; leave in (commented out)
// printf("Test %s x %d -> %s x %d\n",
// string_of_type<A>(), vec_width,
// string_of_type<B>(), vec_width);
if (!is_type_supported<A>(vec_width, target) || !is_type_supported<B>(vec_width, target)) {
// Type not supported, return pass.
return true;
}
int W = 1024;
int H = 1;
Buffer<A> input(W, H);
for (int y = 0; y < H; y++) {
for (int x = 0; x < W; x++) {
// Casting from an out-of-range float to an int is UB, so
// we have to pick our values a little carefully.
input(x, y) = (A)((rand() & 0xffff) / 512.0);
}
}
Var x, y;
Func f;
f(x, y) = cast<B>(input(x, y));
if (target.has_gpu_feature()) {
Var xo, xi;
f.gpu_tile(x, xo, xi, 64);
} else {
if (target.has_feature(Target::HVX)) {
// TODO: Non-native vector widths hang the compiler here.
// f.hexagon();
}
if (vec_width > 1) {
f.vectorize(x, vec_width);
}
}
Buffer<B> output = f.realize({W, H});
/*
for (int y = 0; y < H; y++) {
for (int x = 0; x < W; x++) {
printf("%d %d -> %d %d\n", x, y, (int)(input(x, y)), (int)(output(x, y)));
}
}
*/
for (int y = 0; y < H; y++) {
for (int x = 0; x < W; x++) {
bool ok = ((B)(input(x, y)) == output(x, y));
if (!ok) {
fprintf(stderr, "%s x %d -> %s x %d failed\n",
string_of_type<A>(), vec_width,
string_of_type<B>(), vec_width);
fprintf(stderr, "At %d %d, %f -> %f instead of %f\n",
x, y,
(double)(input(x, y)),
(double)(output(x, y)),
(double)((B)(input(x, y))));
return false;
}
}
}
return true;
}
struct Task {
std::function<bool()> fn;
};
template<typename A>
void add_all(int vec_width, const Target &target, std::vector<Task> &tasks) {
tasks.push_back({[=]() { return test<A, float>(vec_width, target); }});
tasks.push_back({[=]() { return test<A, float>(vec_width, target); }});
tasks.push_back({[=]() { return test<A, double>(vec_width, target); }});
tasks.push_back({[=]() { return test<A, uint8_t>(vec_width, target); }});
tasks.push_back({[=]() { return test<A, uint16_t>(vec_width, target); }});
tasks.push_back({[=]() { return test<A, uint32_t>(vec_width, target); }});
tasks.push_back({[=]() { return test<A, int8_t>(vec_width, target); }});
tasks.push_back({[=]() { return test<A, int16_t>(vec_width, target); }});
tasks.push_back({[=]() { return test<A, int32_t>(vec_width, target); }});
}
int main(int argc, char **argv) {
// TODO: is this still relevant?
// We don't test this on windows, because float-to-int conversions
// on windows use _ftol2, which has its own unique calling
// convention, and older LLVMs (e.g. pnacl) don't do it right so
// you get clobbered registers.
#ifdef WIN32
printf("[SKIP] float-to-int conversions don't work with older LLVMs on Windows\n");
return 0;
#endif
Target target = get_jit_target_from_environment();
// We only test power-of-two vector widths for now
int vec_width_max = 64;
if (target.arch == Target::WebAssembly) {
// The wasm jit is very slow, so shorten this test here.
vec_width_max = 16;
}
std::vector<Task> tasks;
for (int vec_width = 1; vec_width <= vec_width_max; vec_width *= 2) {
add_all<float>(vec_width, target, tasks);
add_all<double>(vec_width, target, tasks);
add_all<uint8_t>(vec_width, target, tasks);
add_all<uint16_t>(vec_width, target, tasks);
add_all<uint32_t>(vec_width, target, tasks);
add_all<int8_t>(vec_width, target, tasks);
add_all<int16_t>(vec_width, target, tasks);
add_all<int32_t>(vec_width, target, tasks);
}
using Sharder = Halide::Internal::Test::Sharder;
Sharder sharder;
Halide::Tools::ThreadPool<bool> pool;
std::vector<std::future<bool>> futures;
for (size_t t = 0; t < tasks.size(); t++) {
if (!sharder.should_run(t)) continue;
const auto &task = tasks.at(t);
futures.push_back(pool.async(task.fn));
}
for (auto &f : futures) {
if (!f.get()) {
return 1;
}
}
printf("Success!\n");
return 0;
}
|
