|
| 1 | +// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple -fsycl-dead-args-optimization %s -o %t.out |
| 2 | +// RUN: %HOST_RUN_PLACEHOLDER %t.out |
| 3 | +// RUN: %CPU_RUN_PLACEHOLDER %t.out |
| 4 | +// RUN: %GPU_RUN_PLACEHOLDER %t.out |
| 5 | +// RUN: %ACC_RUN_PLACEHOLDER %t.out |
| 6 | +// RUN: %clangxx -DRESTRICT_WRITE_ACCESS_TO_CONSTANT_PTR -fsycl -fsycl-targets=%sycl_triple -fsycl-dead-args-optimization %s -o %t1.out |
| 7 | +// RUN: %HOST_RUN_PLACEHOLDER %t1.out |
| 8 | +// RUN: %CPU_RUN_PLACEHOLDER %t1.out |
| 9 | +// RUN: %GPU_RUN_PLACEHOLDER %t1.out |
| 10 | +// RUN: %ACC_RUN_PLACEHOLDER %t1.out |
| 11 | + |
| 12 | +//==--------------- multi_ptr.cpp - SYCL multi_ptr test --------------------==// |
| 13 | +// |
| 14 | +// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 15 | +// See https://llvm.org/LICENSE.txt for license information. |
| 16 | +// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 17 | +// |
| 18 | +//===----------------------------------------------------------------------===// |
| 19 | + |
| 20 | +#include <CL/sycl.hpp> |
| 21 | +#include <cassert> |
| 22 | +#include <iostream> |
| 23 | +#include <type_traits> |
| 24 | + |
| 25 | +using namespace cl::sycl; |
| 26 | + |
| 27 | +/* This is the class used to name the kernel for the runtime. |
| 28 | + * This must be done when the kernel is expressed as a lambda. */ |
| 29 | +template <typename T> class testMultPtrKernel; |
| 30 | +template <typename T> class testMultPtrArrowOperatorKernel; |
| 31 | + |
| 32 | +template <typename T> struct point { |
| 33 | + point(const point &rhs) : x(rhs.x), y(rhs.y) {} |
| 34 | + point(T x, T y) : x(x), y(y) {} |
| 35 | + point(T v) : x(v), y(v) {} |
| 36 | + point() : x(0), y(0) {} |
| 37 | + bool operator==(const T &rhs) { return rhs == x && rhs == y; } |
| 38 | + bool operator==(const point<T> &rhs) { return rhs.x == x && rhs.y == y; } |
| 39 | + T x; |
| 40 | + T y; |
| 41 | +}; |
| 42 | + |
| 43 | +template <typename T> |
| 44 | +void innerFunc(id<1> wiID, global_ptr<T> ptr_1, global_ptr<T> ptr_2, |
| 45 | + local_ptr<T> local_ptr) { |
| 46 | + T t = ptr_1[wiID.get(0)]; |
| 47 | + local_ptr[wiID.get(0)] = t; |
| 48 | + t = local_ptr[wiID.get(0)]; |
| 49 | + ptr_2[wiID.get(0)] = t; |
| 50 | +} |
| 51 | + |
| 52 | +template <typename T> void testMultPtr() { |
| 53 | + T data_1[10]; |
| 54 | + for (size_t i = 0; i < 10; ++i) { |
| 55 | + data_1[i] = 1; |
| 56 | + } |
| 57 | + T data_2[10]; |
| 58 | + for (size_t i = 0; i < 10; ++i) { |
| 59 | + data_2[i] = 2; |
| 60 | + } |
| 61 | + |
| 62 | + { |
| 63 | + range<1> numOfItems{10}; |
| 64 | + buffer<T, 1> bufferData_1(data_1, numOfItems); |
| 65 | + buffer<T, 1> bufferData_2(data_2, numOfItems); |
| 66 | + queue myQueue; |
| 67 | + myQueue.submit([&](handler &cgh) { |
| 68 | + accessor<T, 1, access::mode::read, access::target::global_buffer, |
| 69 | + access::placeholder::false_t> |
| 70 | + accessorData_1(bufferData_1, cgh); |
| 71 | + accessor<T, 1, access::mode::read_write, access::target::global_buffer, |
| 72 | + access::placeholder::false_t> |
| 73 | + accessorData_2(bufferData_2, cgh); |
| 74 | + accessor<T, 1, access::mode::read_write, access::target::local> |
| 75 | + localAccessor(numOfItems, cgh); |
| 76 | + |
| 77 | + cgh.parallel_for<class testMultPtrKernel<T>>(range<1>{10}, [=](id<1> wiID) { |
| 78 | + auto ptr_1 = make_ptr<T, access::address_space::global_space>( |
| 79 | + accessorData_1.get_pointer()); |
| 80 | + auto ptr_2 = make_ptr<T, access::address_space::global_space>( |
| 81 | + accessorData_2.get_pointer()); |
| 82 | + auto local_ptr = make_ptr<T, access::address_space::local_space>( |
| 83 | + localAccessor.get_pointer()); |
| 84 | + |
| 85 | + // General conversions in multi_ptr class |
| 86 | + T *RawPtr = nullptr; |
| 87 | + global_ptr<T> ptr_4(RawPtr); |
| 88 | + ptr_4 = RawPtr; |
| 89 | + |
| 90 | + global_ptr<T> ptr_5(accessorData_1); |
| 91 | + |
| 92 | + global_ptr<void> ptr_6((void *)RawPtr); |
| 93 | + |
| 94 | + ptr_6 = (void *)RawPtr; |
| 95 | + |
| 96 | + // Explicit conversions for device_ptr/host_ptr to global_ptr |
| 97 | + device_ptr<void> ptr_7((void *)RawPtr); |
| 98 | + global_ptr<void> ptr_8 = global_ptr<void>(ptr_7); |
| 99 | + host_ptr<void> ptr_9((void *)RawPtr); |
| 100 | + global_ptr<void> ptr_10 = global_ptr<void>(ptr_9); |
| 101 | + // TODO: need propagation of a7b763b26 patch to acl tool before testing |
| 102 | + // these conversions - otherwise the test would fail on accelerator |
| 103 | + // device during reversed translation from SPIR-V to LLVM IR |
| 104 | + // device_ptr<T> ptr_11(accessorData_1); |
| 105 | + // global_ptr<T> ptr_12 = global_ptr<T>(ptr_11); |
| 106 | + |
| 107 | + innerFunc<T>(wiID.get(0), ptr_1, ptr_2, local_ptr); |
| 108 | + }); |
| 109 | + }); |
| 110 | + } |
| 111 | + for (size_t i = 0; i < 10; ++i) { |
| 112 | + assert(data_1[i] == 1 && "Expected data_1[i] == 1"); |
| 113 | + } |
| 114 | + for (size_t i = 0; i < 10; ++i) { |
| 115 | + assert(data_2[i] == 1 && "Expected data_2[i] == 1"); |
| 116 | + } |
| 117 | +} |
| 118 | + |
| 119 | +template <typename T> void testMultPtrArrowOperator() { |
| 120 | + point<T> data_1[1] = {1}; |
| 121 | + point<T> data_2[1] = {2}; |
| 122 | + point<T> data_3[1] = {3}; |
| 123 | + point<T> data_4[1] = {4}; |
| 124 | + |
| 125 | + { |
| 126 | + range<1> numOfItems{1}; |
| 127 | + buffer<point<T>, 1> bufferData_1(data_1, numOfItems); |
| 128 | + buffer<point<T>, 1> bufferData_2(data_2, numOfItems); |
| 129 | + buffer<point<T>, 1> bufferData_3(data_3, numOfItems); |
| 130 | + buffer<point<T>, 1> bufferData_4(data_4, numOfItems); |
| 131 | + queue myQueue; |
| 132 | + myQueue.submit([&](handler &cgh) { |
| 133 | + accessor<point<T>, 1, access::mode::read, access::target::global_buffer, |
| 134 | + access::placeholder::false_t> |
| 135 | + accessorData_1(bufferData_1, cgh); |
| 136 | + accessor<point<T>, 1, access::mode::read, access::target::constant_buffer, |
| 137 | + access::placeholder::false_t> |
| 138 | + accessorData_2(bufferData_2, cgh); |
| 139 | + accessor<point<T>, 1, access::mode::read_write, access::target::local, |
| 140 | + access::placeholder::false_t> |
| 141 | + accessorData_3(1, cgh); |
| 142 | + accessor<point<T>, 1, access::mode::read, access::target::global_buffer, |
| 143 | + access::placeholder::false_t> |
| 144 | + accessorData_4(bufferData_4, cgh); |
| 145 | + |
| 146 | + cgh.single_task<class testMultPtrArrowOperatorKernel<T>>([=]() { |
| 147 | + auto ptr_1 = make_ptr<point<T>, access::address_space::global_space>( |
| 148 | + accessorData_1.get_pointer()); |
| 149 | + auto ptr_2 = make_ptr<point<T>, access::address_space::constant_space>( |
| 150 | + accessorData_2.get_pointer()); |
| 151 | + auto ptr_3 = make_ptr<point<T>, access::address_space::local_space>( |
| 152 | + accessorData_3.get_pointer()); |
| 153 | + auto ptr_4 = |
| 154 | + make_ptr<point<T>, access::address_space::global_device_space>( |
| 155 | + accessorData_4.get_pointer()); |
| 156 | + |
| 157 | + auto x1 = ptr_1 -> x; |
| 158 | + auto x2 = ptr_2 -> x; |
| 159 | + auto x3 = ptr_3 -> x; |
| 160 | + auto x4 = ptr_4 -> x; |
| 161 | + |
| 162 | + static_assert(std::is_same<decltype(x1), T>::value, |
| 163 | + "Expected decltype(ptr_1->x) == T"); |
| 164 | + static_assert(std::is_same<decltype(x2), T>::value, |
| 165 | + "Expected decltype(ptr_2->x) == T"); |
| 166 | + static_assert(std::is_same<decltype(x3), T>::value, |
| 167 | + "Expected decltype(ptr_3->x) == T"); |
| 168 | + static_assert(std::is_same<decltype(x4), T>::value, |
| 169 | + "Expected decltype(ptr_4->x) == T"); |
| 170 | + }); |
| 171 | + }); |
| 172 | + } |
| 173 | +} |
| 174 | + |
| 175 | +int main() { |
| 176 | + testMultPtr<int>(); |
| 177 | + testMultPtr<float>(); |
| 178 | + testMultPtr<point<int>>(); |
| 179 | + testMultPtr<point<float>>(); |
| 180 | + |
| 181 | + testMultPtrArrowOperator<int>(); |
| 182 | + testMultPtrArrowOperator<float>(); |
| 183 | + |
| 184 | + return 0; |
| 185 | +} |
0 commit comments