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
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
|
#include "common.hpp"
#include <kel/lbm/lbm.hpp>
namespace kel {
namespace lbm {
namespace encode {
template<typename Encode>
struct Sycl {
};
}
/*
namespace impl {
template<typename Schema>
struct struct_has_only_equal_dimension_array{};
}
*/
}
}
namespace saw {
template<typename Sch, uint64_t... Dims, typename Encode>
class data<schema::FixedArray<Sch,Dims...>, kel::lbm::encode::Sycl<Encode>> final {
public:
using Schema = schema::FixedArray<Sch,Dims...>;
private:
acpp::sycl::queue* q_;
data<Sch,Encode>* values_;
//SAW_FORBID_COPY(data);
public:
data(acpp::sycl::queue& q__):
q_{&q__},
values_{nullptr}
{
values_ = acpp::sycl::malloc_device<data<Sch,Encode>>(ct_multiply<uint64_t,Dims...>::value,*q_);
SAW_ASSERT(values_ and q_);
}
~data(){
if(not values_){
return;
}
SAW_ASSERT(q_);
acpp::sycl::free(values_,*q_);
values_ = nullptr;
}
static constexpr data<schema::FixedArray<schema::UInt64, sizeof...(Dims)>> get_dims() {
return saw::data<schema::FixedArray<schema::UInt64, sizeof...(Dims)>>{{Dims...}};
}
constexpr data<Sch,Encode>& at(data<schema::FixedArray<schema::UInt64,sizeof...(Dims)>>& index){
return values_[kel::lbm::flatten_index<schema::UInt64,sizeof...(Dims)>::apply(index,get_dims()).get()];
}
constexpr data<Sch,Encode>& at(data<schema::FixedArray<schema::UInt64,sizeof...(Dims)>>& index) const{
return values_[kel::lbm::flatten_index<schema::UInt64,sizeof...(Dims)>::apply(index,get_dims()).get()];
}
constexpr data<Sch,Encode>* flat_data() {
return values_;
}
};
template<typename Sch, uint64_t Ghost, uint64_t... Sides, typename Encode>
class data<kel::lbm::sch::Chunk<Sch,Ghost,Sides...>,kel::lbm::encode::Sycl<Encode>> final {
public:
using Schema = kel::lbm::sch::Chunk<Sch,Ghost,Sides...>;
private:
using InnerSchema = typename Schema::InnerSchema;
using ValueSchema = typename InnerSchema::ValueType;
data<InnerSchema, kel::lbm::encode::Sycl<Encode>> values_;
public:
data(acpp::sycl::queue& q__):
values_{q__}
{}
data<ValueSchema, kel::lbm::encode::Sycl<Encode>>& ghost_at(const data<schema::FixedArray<schema::UInt64,sizeof...(Sides)>>& index){
return values_.at(index);
}
data<ValueSchema, kel::lbm::encode::Sycl<Encode>>& ghost_at(const data<schema::FixedArray<schema::UInt64,sizeof...(Sides)>>& index) const {
return values_.at(index);
}
static constexpr auto get_ghost_dims() {
return data<InnerSchema,kel::lbm::encode::Sycl<Encode>>::get_dims();
}
data<ValueSchema, Encode>& at(const data<schema::FixedArray<schema::UInt64,sizeof...(Sides)>>& index){
std::decay_t<decltype(index)> ind;
for(uint64_t i = 0u; i < sizeof...(Sides); ++i){
ind.at({i}) = index.at({i}) + Ghost;
}
return values_.at(ind);
}
data<ValueSchema, Encode>& at(const data<schema::FixedArray<schema::UInt64,sizeof...(Sides)>>& index) const {
std::decay_t<decltype(index)> ind;
for(uint64_t i = 0u; i < sizeof...(Sides); ++i){
ind.at({i}) = index.at({i}) + Ghost;
}
return values_.at(ind);
}
static constexpr auto get_dims(){
return data<schema::FixedArray<schema::UInt64, sizeof...(Sides)>,kel::lbm::encode::Sycl<Encode>>{{Sides...}};
}
auto flat_data(){
return values_.flat_data();
}
static constexpr auto flat_size() {
return data<InnerSchema,kel::lbm::encode::Sycl<Encode>>::flat_size();
}
};
template<typename... Members, typename Encode>
class data<schema::Struct<Members...>, kel::lbm::encode::Sycl<Encode> > final {
public:
using StorageT = std::tuple<data<typename Members::ValueType,kel::lbm::encode::Sycl<Encode>>...>;
using Schema = schema::Struct<Members...>;
private:
/**
* @todo Check by static assert that the members all have the same dimensions. Alternatively
* Do it here by specializing.
*/
StorageT members_;
template<std::size_t... Is>
constexpr data(acpp::sycl::queue& q, std::index_sequence<Is...>):
members_{(static_cast<void>(Is), q)...}
{}
public:
data(acpp::sycl::queue& q__):
data{q__, std::make_index_sequence<sizeof...(Members)>{}}
{
q__.wait();
}
template<saw::string_literal K>
auto& get(){
return std::get<parameter_key_pack_index<K, Members::KeyLiteral...>::value>(members_);
}
template<saw::string_literal K>
const auto& get() const {
return std::get<parameter_key_pack_index<K, Members::KeyLiteral...>::value>(members_);
}
};
}
namespace kel {
namespace lbm {
namespace impl {
template<typename Sch, typename Encode>
struct sycl_copy_helper;
template<typename... Members, typename Encode>
struct sycl_copy_helper<sch::Struct<Members...>, Encode> final {
using Schema = sch::Struct<Members...>;
template<uint64_t i>
static saw::error_or<void> copy_to_device_member(saw::data<Schema,Encode>& host_data, saw::data<Schema,encode::Sycl<Encode>>& sycl_data, sycl::queue& q){
if constexpr (i < sizeof...(Members)){
using M = typename saw::parameter_pack_type<i,Members...>::type;
auto& host_member_data = host_data.template get<M::KeyLiteral>();
auto& sycl_member_data = sycl_data.template get<M::KeyLiteral>();
auto host_ptr = host_member_data.flat_data();
auto sycl_ptr = sycl_member_data.flat_data();
static_assert(sizeof(std::decay_t<decltype(sycl_ptr)>) == sizeof(std::decay_t<decltype(host_ptr)>), "Unequal size");
q.submit([&](acpp::sycl::handler& h){
h.copy(host_ptr,sycl_ptr, host_member_data.flat_size().get());
}).wait();
return copy_to_device_member<i+1u>(host_data,sycl_data,q);
}
return saw::make_void();
}
static saw::error_or<void> copy_to_device(saw::data<Schema,Encode>& host_data, saw::data<Schema, encode::Sycl<Encode>>& sycl_data, sycl::queue& q){
return copy_to_device_member<0u>(host_data, sycl_data, q);
}
template<uint64_t i>
static saw::error_or<void> copy_to_host_member(saw::data<Schema,encode::Sycl<Encode>>& sycl_data, saw::data<Schema,Encode>& host_data, sycl::queue& q){
if constexpr (i < sizeof...(Members)){
using M = typename saw::parameter_pack_type<i,Members...>::type;
auto& host_member_data = host_data.template get<M::KeyLiteral>();
auto& sycl_member_data = sycl_data.template get<M::KeyLiteral>();
auto host_ptr = host_member_data.flat_data();
auto sycl_ptr = sycl_member_data.flat_data();
q.submit([&](acpp::sycl::handler& h){
h.copy(sycl_ptr,host_ptr, host_member_data.flat_size().get());
}).wait();
return copy_to_host_member<i+1u>(sycl_data,host_data,q);
}
return saw::make_void();
}
static saw::error_or<void> copy_to_host(saw::data<Schema,Encode>& sycl_data, saw::data<Schema,encode::Sycl<Encode>>& host_data, sycl::queue& q){
return copy_to_host_member<0u>(sycl_data, host_data, q);
}
};
}
class device final {
private:
sycl::queue q_;
SAW_FORBID_COPY(device);
SAW_FORBID_MOVE(device);
public:
device() = default;
~device() = default;
template<typename Sch, typename Encode>
saw::error_or<void> copy_to_device(saw::data<Sch,Encode>& host_data, saw::data<Sch,encode::Sycl<Encode>>& sycl_data){
return impl::sycl_copy_helper<Sch,Encode>::copy_to_device(host_data, sycl_data, q_);
}
template<typename Sch, typename Encode>
saw::error_or<void> copy_to_host(saw::data<Sch,encode::Sycl<Encode>>& sycl_data, saw::data<Sch,Encode>& host_data){
return impl::sycl_copy_helper<Sch,Encode>::copy_to_host(sycl_data, host_data, q_);
}
auto& get_handle(){
return q_;
}
};
}
}
|
