| 1 |
// { dg-additional-options -std=c++20 } |
| 2 |
// { dg-output-file target-std__valarray-1.output } |
| 3 |
|
| 4 |
#include <valarray> |
| 5 |
#include <ostream> |
| 6 |
#include <sstream> |
| 7 |
|
| 8 |
|
| 9 |
/*TODO Work around PR118484 "ICE during IPA pass: cp, segfault in determine_versionability ipa-cp.cc:467". |
| 10 |
|
| 11 |
We can't: |
| 12 |
|
| 13 |
#pragma omp declare target(std::basic_streambuf<char, std::char_traits<char>>::basic_streambuf) |
| 14 |
|
| 15 |
... because: |
| 16 |
|
| 17 |
error: overloaded function name ‘std::basic_streambuf<char>::__ct ’ in clause ‘enter’ |
| 18 |
|
| 19 |
Therefore, use dummy classes in '#pragma omp declare target': |
| 20 |
*/ |
| 21 |
|
| 22 |
#pragma omp declare target |
| 23 |
|
| 24 |
// For 'std::basic_streambuf<char, std::char_traits<char> >::basic_streambuf': |
| 25 |
|
| 26 |
class dummy_basic_streambuf__char |
| 27 |
: public std::basic_streambuf<char> |
| 28 |
{ |
| 29 |
public: |
| 30 |
dummy_basic_streambuf__char() {} |
| 31 |
}; |
| 32 |
|
| 33 |
// For 'std::basic_ios<char, std::char_traits<char> >::basic_ios()': |
| 34 |
|
| 35 |
class dummy_basic_ios__char |
| 36 |
: public std::basic_ios<char> |
| 37 |
{ |
| 38 |
public: |
| 39 |
dummy_basic_ios__char() {} |
| 40 |
}; |
| 41 |
|
| 42 |
#pragma omp end declare target |
| 43 |
|
| 44 |
|
| 45 |
int main() |
| 46 |
{ |
| 47 |
// Due to PR120021 "Offloading vs. C++ 'std::initializer_list'", we can't construct these on the device. |
| 48 |
std::initializer_list<int> v1_i = {10, 20, 30, 40, 50}; |
| 49 |
const int *v1_i_data = std::data(v1_i); |
| 50 |
size_t v1_i_size = v1_i.size(); |
| 51 |
std::initializer_list<int> v2_i = {5, 4, 3, 2, 1}; |
| 52 |
const int *v2_i_data = std::data(v2_i); |
| 53 |
size_t v2_i_size = v2_i.size(); |
| 54 |
std::initializer_list<int> shiftData_i = {1, 2, 3, 4, 5}; |
| 55 |
const int *shiftData_i_data = std::data(shiftData_i); |
| 56 |
size_t shiftData_i_size = shiftData_i.size(); |
| 57 |
#pragma omp target \ |
| 58 |
defaultmap(none) \ |
| 59 |
map(to: v1_i_data[:v1_i_size], v1_i_size, \ |
| 60 |
v2_i_data[:v2_i_size], v2_i_size, \ |
| 61 |
shiftData_i_data[:shiftData_i_size], shiftData_i_size) |
| 62 |
{ |
| 63 |
/* Manually set up a buffer we can stream into, similar to 'cout << [...]', and print it at the end of region. */ |
| 64 |
std::stringbuf out_b; |
| 65 |
std::ostream out(&out_b); |
| 66 |
|
| 67 |
std::valarray<int> v1(v1_i_data, v1_i_size); |
| 68 |
out << "\nv1:"; |
| 69 |
for (auto val : v1) |
| 70 |
out << " " << val; |
| 71 |
|
| 72 |
std::valarray<int> v2(v2_i_data, v2_i_size); |
| 73 |
out << "\nv2:"; |
| 74 |
for (auto val : v2) |
| 75 |
out << " " << val; |
| 76 |
|
| 77 |
std::valarray<int> sum = v1 + v2; |
| 78 |
out << "\nv1 + v2:"; |
| 79 |
for (auto val : sum) |
| 80 |
out << " " << val; |
| 81 |
|
| 82 |
std::valarray<int> diff = v1 - v2; |
| 83 |
out << "\nv1 - v2:"; |
| 84 |
for (auto val : diff) |
| 85 |
out << " " << val; |
| 86 |
|
| 87 |
std::valarray<int> product = v1 * v2; |
| 88 |
out << "\nv1 * v2:"; |
| 89 |
for (auto val : product) |
| 90 |
out << " " << val; |
| 91 |
|
| 92 |
std::valarray<int> quotient = v1 / v2; |
| 93 |
out << "\nv1 / v2:"; |
| 94 |
for (auto val : quotient) |
| 95 |
out << " " << val; |
| 96 |
|
| 97 |
std::valarray<int> squares = pow(v1, 2); |
| 98 |
out << "\npow(v1, 2):"; |
| 99 |
for (auto val : squares) |
| 100 |
out << " " << val; |
| 101 |
|
| 102 |
std::valarray<int> sinhs = sinh(v2); |
| 103 |
out << "\nsinh(v2):"; |
| 104 |
for (auto val : sinhs) |
| 105 |
out << " " << val; |
| 106 |
|
| 107 |
std::valarray<int> logs = log(v1 * v2); |
| 108 |
out << "\nlog(v1 * v2):"; |
| 109 |
for (auto val : logs) |
| 110 |
out << " " << val; |
| 111 |
|
| 112 |
std::valarray<int> data(12); |
| 113 |
for (size_t i = 0; i < data.size(); ++i) |
| 114 |
data[i] = i; |
| 115 |
out << "\nOriginal array:"; |
| 116 |
for (auto val : data) |
| 117 |
out << " " << val; |
| 118 |
|
| 119 |
std::slice slice1(2, 5, 1); |
| 120 |
std::valarray<int> sliced1 = data[slice1]; |
| 121 |
out << "\nSlice(2, 5, 1):"; |
| 122 |
for (auto val : sliced1) |
| 123 |
out << " " << val; |
| 124 |
|
| 125 |
std::slice slice2(1, 4, 3); |
| 126 |
std::valarray<int> sliced2 = data[slice2]; |
| 127 |
out << "\nSlice(1, 4, 3):"; |
| 128 |
for (auto val : sliced2) |
| 129 |
out << " " << val; |
| 130 |
|
| 131 |
data[slice1] = 99; |
| 132 |
out << "\nArray after slice modification:"; |
| 133 |
for (auto val : data) |
| 134 |
out << " " << val; |
| 135 |
|
| 136 |
std::valarray<bool> mask = (v1 > 20); |
| 137 |
out << "\nElements of v1 > 20:"; |
| 138 |
for (size_t i = 0; i < v1.size(); ++i) |
| 139 |
{ |
| 140 |
if (mask[i]) |
| 141 |
out << " " << v1[i]; |
| 142 |
} |
| 143 |
|
| 144 |
std::valarray<int> masked = v1[mask]; |
| 145 |
out << "\nMasked array:"; |
| 146 |
for (auto val : masked) |
| 147 |
out << " " << val; |
| 148 |
|
| 149 |
std::valarray<int> shiftData(shiftData_i_data, shiftData_i_size); |
| 150 |
out << "\nOriginal shiftData:"; |
| 151 |
for (auto val : shiftData) |
| 152 |
out << " " << val; |
| 153 |
|
| 154 |
std::valarray<int> shifted = shiftData.shift(2); |
| 155 |
out << "\nshift(2):"; |
| 156 |
for (auto val : shifted) |
| 157 |
out << " " << val; |
| 158 |
|
| 159 |
std::valarray<int> cshifted = shiftData.cshift(-1); |
| 160 |
out << "\ncshift(-1):"; |
| 161 |
for (auto val : cshifted) |
| 162 |
out << " " << val; |
| 163 |
|
| 164 |
out << "\nSum(v1): " << v1.sum(); |
| 165 |
out << "\nMin(v1): " << v1.min(); |
| 166 |
out << "\nMax(v1): " << v1.max(); |
| 167 |
|
| 168 |
out << "\n"; |
| 169 |
|
| 170 |
/* Terminate with a NUL. Otherwise, we'd have to use: |
| 171 |
__builtin_printf("%.*s", (int) out_b_sv.size(), out_b_sv.data()); |
| 172 |
... which nvptx 'printf', as implemented via PTX 'vprintf', doesn't support (TODO). */ |
| 173 |
out << '\0'; |
| 174 |
std::string_view out_b_sv = out_b.view(); |
| 175 |
__builtin_printf("%s", out_b_sv.data()); |
| 176 |
} |
| 177 |
|
| 178 |
return 0; |
| 179 |
} |