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
|
#include "utest_helper.hpp"
#include <cmath>
#include <algorithm>
static void cpu_compiler_math(float *dst, float *src1, float *src2, float *src3, int i)
{
const float x = src1[i], y = src2[i], z = src3[i];
switch (i%2) {
case 0: dst[i] = x * y + z; break;
case 1: dst[i] = x * y + z; break;
default: dst[i] = 1.f; break;
};
dst[0] = (src1[0]*src2[0]+src3[0]);
}
static void compiler_math_3op_float(void)
{
const size_t n = 32;
float cpu_dst[32], cpu_src1[32], cpu_src2[32], cpu_src3[32];
// Setup kernel and buffers
OCL_CREATE_KERNEL_FROM_FILE("compiler_math_3op",
"compiler_math_3op_float");
OCL_CREATE_BUFFER(buf[0], 0, n * sizeof(float), NULL);
OCL_CREATE_BUFFER(buf[1], 0, n * sizeof(float), NULL);
OCL_CREATE_BUFFER(buf[2], 0, n * sizeof(float), NULL);
OCL_CREATE_BUFFER(buf[3], 0, n * sizeof(float), NULL);
OCL_SET_ARG(0, sizeof(cl_mem), &buf[0]);
OCL_SET_ARG(1, sizeof(cl_mem), &buf[1]);
OCL_SET_ARG(2, sizeof(cl_mem), &buf[2]);
OCL_SET_ARG(3, sizeof(cl_mem), &buf[3]);
globals[0] = 16;
locals[0] = 16;
for (int j = 0; j < 1000; j ++) {
OCL_MAP_BUFFER(1);
OCL_MAP_BUFFER(2);
OCL_MAP_BUFFER(3);
for (uint32_t i = 0; i < 32; ++i) {
cpu_src1[i] = ((float*)buf_data[1])[i] = .001f * (rand() & 15);
cpu_src2[i] = ((float*)buf_data[2])[i] = .002f * (rand() & 15);
cpu_src3[i] = ((float*)buf_data[3])[i] = .003f * (rand() & 15);
}
OCL_UNMAP_BUFFER(1);
OCL_UNMAP_BUFFER(2);
OCL_UNMAP_BUFFER(3);
OCL_NDRANGE(1);
for (int i = 0; i < 16; ++i)
cpu_compiler_math(cpu_dst, cpu_src1, cpu_src2, cpu_src3, i);
OCL_MAP_BUFFER(0);
for (int i = 0; i < 16; ++i) {
const float cpu = cpu_dst[i];
const float gpu = ((float*)buf_data[0])[i];
//printf("cpu:%f, gpu:%f\n", cpu, gpu);
if (std::isinf(cpu))
OCL_ASSERT(std::isinf(gpu));
else if (std::isnan(cpu))
OCL_ASSERT(std::isnan(gpu));
else
OCL_ASSERT(fabs(gpu-cpu) < 1e-3f);
}
OCL_UNMAP_BUFFER(0);
}
}
MAKE_UTEST_FROM_FUNCTION(compiler_math_3op_float)
static void compiler_math_3op_half(void)
{
if (!cl_check_half())
return;
const size_t n = 32;
float cpu_dst[32], cpu_src1[32], cpu_src2[32], cpu_src3[32];
// Setup kernel and buffers
OCL_CALL(cl_kernel_init, "compiler_math_3op.cl",
"compiler_math_3op_half",
SOURCE, "-DHALF");
OCL_CREATE_BUFFER(buf[0], 0, n * sizeof(cl_half), NULL);
OCL_CREATE_BUFFER(buf[1], 0, n * sizeof(cl_half), NULL);
OCL_CREATE_BUFFER(buf[2], 0, n * sizeof(cl_half), NULL);
OCL_CREATE_BUFFER(buf[3], 0, n * sizeof(cl_half), NULL);
OCL_SET_ARG(0, sizeof(cl_mem), &buf[0]);
OCL_SET_ARG(1, sizeof(cl_mem), &buf[1]);
OCL_SET_ARG(2, sizeof(cl_mem), &buf[2]);
OCL_SET_ARG(3, sizeof(cl_mem), &buf[3]);
globals[0] = 16;
locals[0] = 16;
for (int j = 0; j < 1000; j ++) {
OCL_MAP_BUFFER(1);
OCL_MAP_BUFFER(2);
OCL_MAP_BUFFER(3);
for (uint32_t i = 0; i < 32; ++i) {
((cl_half*)buf_data[1])[i] = __float_to_half(as_uint(cpu_src1[i] = 0.1f*(rand() & 63)));
((cl_half*)buf_data[2])[i] = __float_to_half(as_uint(cpu_src2[i] = 0.02f*(rand() & 63)));
((cl_half*)buf_data[3])[i] = __float_to_half(as_uint(cpu_src3[i] = 0.02f*(rand() & 63)));
}
OCL_UNMAP_BUFFER(1);
OCL_UNMAP_BUFFER(2);
OCL_UNMAP_BUFFER(3);
OCL_NDRANGE(1);
for (int i = 0; i < 16; ++i)
cpu_compiler_math(cpu_dst, cpu_src1, cpu_src2, cpu_src3, i);
OCL_MAP_BUFFER(0);
for (int i = 0; i < 16; ++i) {
const float cpu = cpu_dst[i];
bool isInf, infSign;
const float gpu = as_float(__half_to_float(((uint16_t*)buf_data[0])[i], &isInf, &infSign));
//printf("cpu:(%f*%f+%f) = %f, gpu:%f\n", cpu_src1[i], cpu_src2[i], cpu_src3[i],cpu,gpu);
OCL_ASSERT(((fabs(cpu) < 6e-8f) && (gpu < 6e-8f)) || (fabs(cpu - gpu) <= 0.3 * fabs(cpu)) ||
(isInf && ((infSign && cpu > 65504.0f) || (!infSign && cpu < -65504.0f))) ||
(std::isnan(gpu) && std::isnan(cpu)));
}
OCL_UNMAP_BUFFER(0);
}
}
MAKE_UTEST_FROM_FUNCTION(compiler_math_3op_half)
|
