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
|
//===-- runtime/CUDA/allocatable.cpp --------------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#include "flang/Runtime/CUDA/allocatable.h"
#include "../assign-impl.h"
#include "../stat.h"
#include "../terminator.h"
#include "flang/Runtime/CUDA/common.h"
#include "flang/Runtime/CUDA/descriptor.h"
#include "flang/Runtime/CUDA/memmove-function.h"
#include "flang/Runtime/allocatable.h"
#include "cuda_runtime.h"
namespace Fortran::runtime::cuda {
extern "C" {
RT_EXT_API_GROUP_BEGIN
int RTDEF(CUFAllocatableAllocateSync)(Descriptor &desc, int64_t stream,
bool hasStat, const Descriptor *errMsg, const char *sourceFile,
int sourceLine) {
int stat{RTNAME(CUFAllocatableAllocate)(
desc, stream, hasStat, errMsg, sourceFile, sourceLine)};
#ifndef RT_DEVICE_COMPILATION
// Descriptor synchronization is only done when the allocation is done
// from the host.
if (stat == StatOk) {
void *deviceAddr{
RTNAME(CUFGetDeviceAddress)((void *)&desc, sourceFile, sourceLine)};
RTNAME(CUFDescriptorSync)
((Descriptor *)deviceAddr, &desc, sourceFile, sourceLine);
}
#endif
return stat;
}
int RTDEF(CUFAllocatableAllocate)(Descriptor &desc, int64_t stream,
bool hasStat, const Descriptor *errMsg, const char *sourceFile,
int sourceLine) {
if (desc.HasAddendum()) {
Terminator terminator{sourceFile, sourceLine};
// TODO: This require a bit more work to set the correct type descriptor
// address
terminator.Crash(
"not yet implemented: CUDA descriptor allocation with addendum");
}
// Perform the standard allocation.
int stat{RTNAME(AllocatableAllocate)(
desc, hasStat, errMsg, sourceFile, sourceLine)};
return stat;
}
int RTDEF(CUFAllocatableAllocateSource)(Descriptor &alloc,
const Descriptor &source, int64_t stream, bool hasStat,
const Descriptor *errMsg, const char *sourceFile, int sourceLine) {
int stat{RTNAME(CUFAllocatableAllocate)(
alloc, stream, hasStat, errMsg, sourceFile, sourceLine)};
if (stat == StatOk) {
Terminator terminator{sourceFile, sourceLine};
Fortran::runtime::DoFromSourceAssign(
alloc, source, terminator, &MemmoveHostToDevice);
}
return stat;
}
int RTDEF(CUFAllocatableAllocateSourceSync)(Descriptor &alloc,
const Descriptor &source, int64_t stream, bool hasStat,
const Descriptor *errMsg, const char *sourceFile, int sourceLine) {
int stat{RTNAME(CUFAllocatableAllocateSync)(
alloc, stream, hasStat, errMsg, sourceFile, sourceLine)};
if (stat == StatOk) {
Terminator terminator{sourceFile, sourceLine};
Fortran::runtime::DoFromSourceAssign(
alloc, source, terminator, &MemmoveHostToDevice);
}
return stat;
}
int RTDEF(CUFAllocatableDeallocate)(Descriptor &desc, bool hasStat,
const Descriptor *errMsg, const char *sourceFile, int sourceLine) {
// Perform the standard allocation.
int stat{RTNAME(AllocatableDeallocate)(
desc, hasStat, errMsg, sourceFile, sourceLine)};
#ifndef RT_DEVICE_COMPILATION
// Descriptor synchronization is only done when the deallocation is done
// from the host.
if (stat == StatOk) {
void *deviceAddr{
RTNAME(CUFGetDeviceAddress)((void *)&desc, sourceFile, sourceLine)};
RTNAME(CUFDescriptorSync)
((Descriptor *)deviceAddr, &desc, sourceFile, sourceLine);
}
#endif
return stat;
}
RT_EXT_API_GROUP_END
} // extern "C"
} // namespace Fortran::runtime::cuda
|
