|
14 | 14 |
|
15 | 15 | int main() {
|
16 | 16 |
|
17 |
| -sycl::queue q; |
| 17 | + sycl::queue q; |
18 | 18 |
|
19 | 19 | if (!q.get_device().has(sycl::aspect::fp16)) {
|
20 | 20 | std::cout << "skipping fp16 tests: requires fp16 device aspect."
|
21 | 21 | << std::endl;
|
22 | 22 | return 0;
|
23 | 23 | }
|
24 | 24 |
|
25 |
| -const sycl::half tv[16] = {-2.0, -1.5, -1.0, 0.0, 2.0, 1.5, 1.0, 0.0, |
26 |
| - -1.7, 1.7, -1.2, 1.2, -3.0, 3.0, -10.0, 10.0}; |
27 |
| -const sycl::half tl[16] = {-0.97, -0.91, -0.77, -0.1, 0.95, 0.89, 0.75, -0.1, |
28 |
| - -0.94, 0.92, -0.84, 0.82, -1.0, 0.98, -1.10, 0.98}; |
29 |
| -const sycl::half tu[16] = {-0.95, -0.89, -0.75, 0.1, 0.97, 0.91, 0.77, 0.1, |
30 |
| - -0.92, 0.94, -0.82, 0.84, -0.98, 1.00, -0.98, 1.10}; |
| 25 | + const sycl::half tv[16] = {-2.0, -1.5, -1.0, 0.0, 2.0, 1.5, 1.0, 0.0, |
| 26 | + -1.7, 1.7, -1.2, 1.2, -3.0, 3.0, -10.0, 10.0}; |
| 27 | + const sycl::half tl[16] = {-0.97, -0.91, -0.77, -0.1, 0.95, 0.89, |
| 28 | + 0.75, -0.1, -0.94, 0.92, -0.84, 0.82, |
| 29 | + -1.0, 0.98, -1.10, 0.98}; |
| 30 | + const sycl::half tu[16] = {-0.95, -0.89, -0.75, 0.1, 0.97, 0.91, |
| 31 | + 0.77, 0.1, -0.92, 0.94, -0.82, 0.84, |
| 32 | + -0.98, 1.00, -0.98, 1.10}; |
31 | 33 |
|
32 |
| -native_tanh_tester<sycl::half>(q, tv[0], tl[0], tu[0]); |
33 |
| -native_tanh_tester<sycl::half2>(q, {tv[0], tv[1]}, {tl[0], tl[1]}, |
34 |
| - {tu[0], tu[1]}); |
35 |
| -native_tanh_tester<sycl::half3>(q, {tv[0], tv[1], tv[2]}, {tl[0], tl[1], tl[2]}, |
36 |
| - {tu[0], tu[1], tu[2]}); |
37 |
| -native_tanh_tester<sycl::marray<sycl::half, 3>>(q, {tv[0], tv[1], tv[2]}, |
38 |
| - {tl[0], tl[1], tl[2]}, |
39 |
| - {tu[0], tu[1], tu[2]}); |
40 |
| -native_tanh_tester<sycl::half4>(q, {tv[0], tv[1], tv[2], tv[3]}, |
41 |
| - {tl[0], tl[1], tl[2], tl[3]}, |
42 |
| - {tu[0], tu[1], tu[2], tu[3]}); |
43 |
| -native_tanh_tester<sycl::marray<sycl::half, 4>>(q, {tv[0], tv[1], tv[2], tv[3]}, |
44 |
| - {tl[0], tl[1], tl[2], tl[3]}, |
45 |
| - {tu[0], tu[1], tu[2], tu[3]}); |
46 |
| -native_tanh_tester<sycl::half8>( |
47 |
| - q, {tv[0], tv[1], tv[2], tv[3], tv[4], tv[5], tv[6], tv[7]}, |
48 |
| - {tl[0], tl[1], tl[2], tl[3], tl[4], tl[5], tl[6], tl[7]}, |
49 |
| - {tu[0], tu[1], tu[2], tu[3], tu[4], tu[5], tu[6], tu[7]}); |
50 |
| -native_tanh_tester<sycl::half16>( |
51 |
| - q, |
52 |
| - {tv[0], tv[1], tv[2], tv[3], tv[4], tv[5], tv[6], tv[7], tv[8], tv[9], |
53 |
| - tv[10], tv[11], tv[12], tv[13], tv[14], tv[15]}, |
54 |
| - {tl[0], tl[1], tl[2], tl[3], tl[4], tl[5], tl[6], tl[7], tl[8], tl[9], |
55 |
| - tl[10], tl[11], tl[12], tl[13], tl[14], tl[15]}, |
56 |
| - {tu[0], tu[1], tu[2], tu[3], tu[4], tu[5], tu[6], tu[7], tu[8], tu[9], |
57 |
| - tu[10], tu[11], tu[12], tu[13], tu[14], tu[15]}); |
| 34 | + native_tanh_tester<sycl::half>(q, tv[0], tl[0], tu[0]); |
| 35 | + native_tanh_tester<sycl::half2>(q, {tv[0], tv[1]}, {tl[0], tl[1]}, |
| 36 | + {tu[0], tu[1]}); |
| 37 | + native_tanh_tester<sycl::half3>(q, {tv[0], tv[1], tv[2]}, |
| 38 | + {tl[0], tl[1], tl[2]}, {tu[0], tu[1], tu[2]}); |
| 39 | + native_tanh_tester<sycl::marray<sycl::half, 3>>( |
| 40 | + q, {tv[0], tv[1], tv[2]}, {tl[0], tl[1], tl[2]}, {tu[0], tu[1], tu[2]}); |
| 41 | + native_tanh_tester<sycl::half4>(q, {tv[0], tv[1], tv[2], tv[3]}, |
| 42 | + {tl[0], tl[1], tl[2], tl[3]}, |
| 43 | + {tu[0], tu[1], tu[2], tu[3]}); |
| 44 | + native_tanh_tester<sycl::marray<sycl::half, 4>>( |
| 45 | + q, {tv[0], tv[1], tv[2], tv[3]}, {tl[0], tl[1], tl[2], tl[3]}, |
| 46 | + {tu[0], tu[1], tu[2], tu[3]}); |
| 47 | + native_tanh_tester<sycl::half8>( |
| 48 | + q, {tv[0], tv[1], tv[2], tv[3], tv[4], tv[5], tv[6], tv[7]}, |
| 49 | + {tl[0], tl[1], tl[2], tl[3], tl[4], tl[5], tl[6], tl[7]}, |
| 50 | + {tu[0], tu[1], tu[2], tu[3], tu[4], tu[5], tu[6], tu[7]}); |
| 51 | + native_tanh_tester<sycl::half16>( |
| 52 | + q, |
| 53 | + {tv[0], tv[1], tv[2], tv[3], tv[4], tv[5], tv[6], tv[7], tv[8], tv[9], |
| 54 | + tv[10], tv[11], tv[12], tv[13], tv[14], tv[15]}, |
| 55 | + {tl[0], tl[1], tl[2], tl[3], tl[4], tl[5], tl[6], tl[7], tl[8], tl[9], |
| 56 | + tl[10], tl[11], tl[12], tl[13], tl[14], tl[15]}, |
| 57 | + {tu[0], tu[1], tu[2], tu[3], tu[4], tu[5], tu[6], tu[7], tu[8], tu[9], |
| 58 | + tu[10], tu[11], tu[12], tu[13], tu[14], tu[15]}); |
58 | 59 |
|
59 |
| -const sycl::half ev[16] = {-2.0, -1.5, -1.0, 0.0, 2.0, 1.5, 1.0, 0.0, |
60 |
| - -2.0, -1.5, -1.0, 0.0, 2.0, 1.5, 1.0, 0.0}; |
61 |
| -const sycl::half el[16] = {0.1, 0.34, 0.4, -0.9, 3.9, 2.7, 1.9, -0.9, |
62 |
| - 0.1, 0.34, 0.4, -0.9, 3.9, 2.7, 1.9, -0.9}; |
63 |
| -const sycl::half eu[16] = {0.3, 0.36, 0.6, 1.1, 4.1, 2.9, 2.1, 1.1, |
64 |
| - 0.3, 0.36, 0.6, 1.1, 4.1, 2.9, 2.1, 1.1}; |
| 60 | + const sycl::half ev[16] = {-2.0, -1.5, -1.0, 0.0, 2.0, 1.5, 1.0, 0.0, |
| 61 | + -2.0, -1.5, -1.0, 0.0, 2.0, 1.5, 1.0, 0.0}; |
| 62 | + const sycl::half el[16] = {0.1, 0.34, 0.4, -0.9, 3.9, 2.7, 1.9, -0.9, |
| 63 | + 0.1, 0.34, 0.4, -0.9, 3.9, 2.7, 1.9, -0.9}; |
| 64 | + const sycl::half eu[16] = {0.3, 0.36, 0.6, 1.1, 4.1, 2.9, 2.1, 1.1, |
| 65 | + 0.3, 0.36, 0.6, 1.1, 4.1, 2.9, 2.1, 1.1}; |
65 | 66 |
|
66 |
| -native_exp2_tester<sycl::half>(q, ev[0], el[0], eu[0]); |
67 |
| -native_exp2_tester<sycl::half2>(q, {ev[0], ev[1]}, {el[0], el[1]}, |
68 |
| - {eu[0], eu[1]}); |
69 |
| -native_exp2_tester<sycl::half3>(q, {ev[0], ev[1], ev[2]}, {el[0], el[1], el[2]}, |
70 |
| - {eu[0], eu[1], eu[2]}); |
71 |
| -native_exp2_tester<sycl::half4>(q, {ev[0], ev[1], ev[2], ev[3]}, |
72 |
| - {el[0], el[1], el[2], el[3]}, |
73 |
| - {eu[0], eu[1], eu[2], eu[3]}); |
74 |
| -native_exp2_tester<sycl::marray<sycl::half, 3>>(q, {ev[0], ev[1], ev[2]}, |
75 |
| - {el[0], el[1], el[2]}, |
76 |
| - {eu[0], eu[1], eu[2]}); |
77 |
| -native_exp2_tester<sycl::marray<sycl::half, 4>>(q, {ev[0], ev[1], ev[2], ev[3]}, |
78 |
| - {el[0], el[1], el[2], el[3]}, |
79 |
| - {eu[0], eu[1], eu[2], eu[3]}); |
80 |
| -native_exp2_tester<sycl::half8>( |
81 |
| - q, {ev[0], ev[1], ev[2], ev[3], ev[4], ev[5], ev[6], ev[7]}, |
82 |
| - {el[0], el[1], el[2], el[3], el[4], el[5], el[6], el[7]}, |
83 |
| - {eu[0], eu[1], eu[2], eu[3], eu[4], eu[5], eu[6], eu[7]}); |
84 |
| -native_exp2_tester<sycl::half16>( |
85 |
| - q, |
86 |
| - {ev[0], ev[1], ev[2], ev[3], ev[4], ev[5], ev[6], ev[7], ev[8], ev[9], |
87 |
| - ev[10], ev[11], ev[12], ev[13], ev[14], ev[15]}, |
88 |
| - {el[0], el[1], el[2], el[3], el[4], el[5], el[6], el[7], el[8], el[9], |
89 |
| - el[10], el[11], el[12], el[13], el[14], el[15]}, |
90 |
| - {eu[0], eu[1], eu[2], eu[3], eu[4], eu[5], eu[6], eu[7], eu[8], eu[9], |
91 |
| - eu[10], eu[11], eu[12], eu[13], eu[14], eu[15]}); |
| 67 | + native_exp2_tester<sycl::half>(q, ev[0], el[0], eu[0]); |
| 68 | + native_exp2_tester<sycl::half2>(q, {ev[0], ev[1]}, {el[0], el[1]}, |
| 69 | + {eu[0], eu[1]}); |
| 70 | + native_exp2_tester<sycl::half3>(q, {ev[0], ev[1], ev[2]}, |
| 71 | + {el[0], el[1], el[2]}, {eu[0], eu[1], eu[2]}); |
| 72 | + native_exp2_tester<sycl::half4>(q, {ev[0], ev[1], ev[2], ev[3]}, |
| 73 | + {el[0], el[1], el[2], el[3]}, |
| 74 | + {eu[0], eu[1], eu[2], eu[3]}); |
| 75 | + native_exp2_tester<sycl::marray<sycl::half, 3>>( |
| 76 | + q, {ev[0], ev[1], ev[2]}, {el[0], el[1], el[2]}, {eu[0], eu[1], eu[2]}); |
| 77 | + native_exp2_tester<sycl::marray<sycl::half, 4>>( |
| 78 | + q, {ev[0], ev[1], ev[2], ev[3]}, {el[0], el[1], el[2], el[3]}, |
| 79 | + {eu[0], eu[1], eu[2], eu[3]}); |
| 80 | + native_exp2_tester<sycl::half8>( |
| 81 | + q, {ev[0], ev[1], ev[2], ev[3], ev[4], ev[5], ev[6], ev[7]}, |
| 82 | + {el[0], el[1], el[2], el[3], el[4], el[5], el[6], el[7]}, |
| 83 | + {eu[0], eu[1], eu[2], eu[3], eu[4], eu[5], eu[6], eu[7]}); |
| 84 | + native_exp2_tester<sycl::half16>( |
| 85 | + q, |
| 86 | + {ev[0], ev[1], ev[2], ev[3], ev[4], ev[5], ev[6], ev[7], ev[8], ev[9], |
| 87 | + ev[10], ev[11], ev[12], ev[13], ev[14], ev[15]}, |
| 88 | + {el[0], el[1], el[2], el[3], el[4], el[5], el[6], el[7], el[8], el[9], |
| 89 | + el[10], el[11], el[12], el[13], el[14], el[15]}, |
| 90 | + {eu[0], eu[1], eu[2], eu[3], eu[4], eu[5], eu[6], eu[7], eu[8], eu[9], |
| 91 | + eu[10], eu[11], eu[12], eu[13], eu[14], eu[15]}); |
92 | 92 |
|
93 | 93 | return 0;
|
94 | 94 | }
|
0 commit comments