|
145
|
1 /* Copyright (C) 2015-2020 Free Software Foundation, Inc.
|
|
111
|
2 Contributed by Alexander Monakov <amonakov@ispras.ru>
|
|
|
3
|
|
|
4 This file is part of the GNU Offloading and Multi Processing Library
|
|
|
5 (libgomp).
|
|
|
6
|
|
|
7 Libgomp is free software; you can redistribute it and/or modify it
|
|
|
8 under the terms of the GNU General Public License as published by
|
|
|
9 the Free Software Foundation; either version 3, or (at your option)
|
|
|
10 any later version.
|
|
|
11
|
|
|
12 Libgomp is distributed in the hope that it will be useful, but WITHOUT ANY
|
|
|
13 WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
|
|
|
14 FOR A PARTICULAR PURPOSE. See the GNU General Public License for
|
|
|
15 more details.
|
|
|
16
|
|
|
17 Under Section 7 of GPL version 3, you are granted additional
|
|
|
18 permissions described in the GCC Runtime Library Exception, version
|
|
|
19 3.1, as published by the Free Software Foundation.
|
|
|
20
|
|
|
21 You should have received a copy of the GNU General Public License and
|
|
|
22 a copy of the GCC Runtime Library Exception along with this program;
|
|
|
23 see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
|
|
|
24 <http://www.gnu.org/licenses/>. */
|
|
|
25
|
|
|
26 /* This is the NVPTX implementation of doacross spinning. */
|
|
|
27
|
|
|
28 #ifndef GOMP_DOACROSS_H
|
|
|
29 #define GOMP_DOACROSS_H 1
|
|
|
30
|
|
|
31 #include "libgomp.h"
|
|
|
32
|
|
|
33 static int zero;
|
|
|
34
|
|
|
35 static inline int
|
|
|
36 cpu_relax (void)
|
|
|
37 {
|
|
|
38 int r;
|
|
|
39 /* Here we need a long-latency operation to make the current warp yield.
|
|
|
40 We could use ld.cv, uncached load from system (host) memory, but that
|
|
|
41 would require allocating locked memory in the plugin. Alternatively,
|
|
|
42 we can use ld.cg, which evicts from L1 and caches in L2. */
|
|
|
43 asm volatile ("ld.cg.s32 %0, [%1];" : "=r" (r) : "i" (&zero) : "memory");
|
|
|
44 return r;
|
|
|
45 }
|
|
|
46
|
|
|
47 static inline void doacross_spin (unsigned long *addr, unsigned long expected,
|
|
|
48 unsigned long cur)
|
|
|
49 {
|
|
|
50 /* Prevent compiler from optimizing based on bounds of containing object. */
|
|
|
51 asm ("" : "+r" (addr));
|
|
|
52 do
|
|
|
53 {
|
|
|
54 int i = cpu_relax ();
|
|
|
55 cur = addr[i];
|
|
|
56 }
|
|
|
57 while (cur <= expected);
|
|
|
58 }
|
|
|
59
|
|
|
60 #endif /* GOMP_DOACROSS_H */
|
