Neko  0.9.0
A portable framework for high-order spectral element flow simulations
fusedcg_aux.hip
Go to the documentation of this file.
1 /*
2  Copyright (c) 2021-2024, The Neko Authors
3  All rights reserved.
4 
5  Redistribution and use in source and binary forms, with or without
6  modification, are permitted provided that the following conditions
7  are met:
8 
9  * Redistributions of source code must retain the above copyright
10  notice, this list of conditions and the following disclaimer.
11 
12  * Redistributions in binary form must reproduce the above
13  copyright notice, this list of conditions and the following
14  disclaimer in the documentation and/or other materials provided
15  with the distribution.
16 
17  * Neither the name of the authors nor the names of its
18  contributors may be used to endorse or promote products derived
19  from this software without specific prior written permission.
20 
21  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25  COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
29  CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
31  ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32  POSSIBILITY OF SUCH DAMAGE.
33 */
34 
35 #include <hip/hip_runtime.h>
36 #include "fusedcg_kernel.h"
37 #include <device/device_config.h>
38 #include <device/hip/check.h>
39 
43 real *fusedcg_buf = NULL;
46 
47 extern "C" {
48 
51 
52 
53 
54  void hip_fusedcg_update_p(void *p, void *z, void *po, real *beta, int *n) {
55 
56  const dim3 nthrds(1024, 1, 1);
57  const dim3 nblcks(((*n)+1024 - 1)/ 1024, 1, 1);
58  const hipStream_t stream = (hipStream_t) glb_cmd_queue;
59 
60  hipLaunchKernelGGL(HIP_KERNEL_NAME(fusedcg_update_p_kernel<real>),
61  nblcks, nthrds, 0, stream,
62  (real *) p, (real *) z, (real *) po, *beta, *n);
63  HIP_CHECK(hipGetLastError());
64 
65  }
66 
67  void hip_fusedcg_update_x(void *x, void *p, void *alpha, int *p_cur, int *n) {
68 
69  const dim3 nthrds(1024, 1, 1);
70  const dim3 nblcks(((*n)+1024 - 1)/ 1024, 1, 1);
71  const hipStream_t stream = (hipStream_t) glb_cmd_queue;
72 
73  hipLaunchKernelGGL(HIP_KERNEL_NAME(fusedcg_update_x_kernel<real>),
74  nblcks, nthrds, 0, stream,
75  (real *) x, (const real **) p,
76  (const real *) alpha, *p_cur, *n);
77  HIP_CHECK(hipGetLastError());
78  }
79 
80 
81  real hip_fusedcg_part2(void *a, void *b, void *c,
82  void *alpha_d , real *alpha, int *p_cur, int * n) {
83 
84  const dim3 nthrds(1024, 1, 1);
85  const dim3 nblcks(((*n)+1024 - 1)/ 1024, 1, 1);
86  const int nb = ((*n) + 1024 - 1)/ 1024;
87  const hipStream_t stream = (hipStream_t) glb_cmd_queue;
88 
89  if (fusedcg_buf != NULL && fusedcg_buf_len < nb) {
90  HIP_CHECK(hipHostFree(fusedcg_buf));
91  HIP_CHECK(hipFree(fusedcg_buf_d));
92  fusedcg_buf = NULL;
93  }
94 
95  if (fusedcg_buf == NULL){
96  HIP_CHECK(hipHostMalloc(&fusedcg_buf, 2*sizeof(real)));
97  HIP_CHECK(hipMalloc(&fusedcg_buf_d, nb*sizeof(real)));
98  fusedcg_buf_len = nb;
99  }
100 
101  /* Store alpha(p_cur) in pinned memory */
102  fusedcg_buf[1] = (*alpha);
103 
104  /* Update alpha_d(p_cur) = alpha(p_cur) */
105  real *alpha_d_p_cur = ((real *) alpha_d) + ((*p_cur - 1));
106  HIP_CHECK(hipMemcpyAsync(alpha_d_p_cur, &fusedcg_buf[1], sizeof(real),
107  hipMemcpyHostToDevice, stream));
108 
109  hipLaunchKernelGGL(HIP_KERNEL_NAME(fusedcg_part2_kernel<real>),
110  nblcks, nthrds, 0, stream,
111  (real *) a, (real *) b, (real *) c,
112  *alpha, fusedcg_buf_d, * n);
113  HIP_CHECK(hipGetLastError());
114 
115  hipLaunchKernelGGL(HIP_KERNEL_NAME(reduce_kernel<real>),
116  1, 1024, 0, stream, fusedcg_buf_d, nb);
117  HIP_CHECK(hipGetLastError());
118 
119 #ifdef HAVE_DEVICE_MPI
120  HIP_CHECK(hipStreamSynchronize(stream));
122  sizeof(real), DEVICE_MPI_SUM);
123 #else
124  HIP_CHECK(hipMemcpyAsync(fusedcg_buf, fusedcg_buf_d, sizeof(real),
125  hipMemcpyDeviceToHost, stream));
126  HIP_CHECK(hipStreamSynchronize(stream));
127 #endif
128 
129  return fusedcg_buf[0];
130  }
131 }
132 
__global__ void const T *__restrict__ x
Definition: cdtp_kernel.h:106
double real
Definition: device_config.h:12
void * glb_cmd_queue
#define DEVICE_MPI_SUM
Definition: device_mpi_op.h:9
void device_mpi_allreduce(void *buf_d, void *buf, int count, int nbytes, int op)
void hip_fusedcg_update_x(void *x, void *p, void *alpha, int *p_cur, int *n)
Definition: fusedcg_aux.hip:67
real hip_fusedcg_part2(void *a, void *b, void *c, void *alpha_d, real *alpha, int *p_cur, int *n)
Definition: fusedcg_aux.hip:81
int fusedcg_buf_len
Definition: fusedcg_aux.hip:45
real * fusedcg_buf
Definition: fusedcg_aux.hip:43
void hip_fusedcg_update_p(void *p, void *z, void *po, real *beta, int *n)
Definition: fusedcg_aux.hip:54
real * fusedcg_buf_d
Definition: fusedcg_aux.hip:44
#define HIP_CHECK(err)
Definition: check.h:8