File: GB_cuda_init.c

package info (click to toggle)
suitesparse 1%3A7.11.0%2Bdfsg-2
  • links: PTS, VCS
  • area: main
  • in suites: forky, sid
  • size: 258,172 kB
  • sloc: ansic: 1,153,566; cpp: 48,145; makefile: 4,997; fortran: 2,087; java: 1,826; sh: 1,113; ruby: 725; python: 676; asm: 371; sed: 166; awk: 44
file content (78 lines) | stat: -rw-r--r-- 2,768 bytes parent folder | download
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
//------------------------------------------------------------------------------
// GraphBLAS/CUDA/GB_cuda_init: initialize the GPUs for use by GraphBLAS
//------------------------------------------------------------------------------

// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2025, All Rights Reserved.
// This file: Copyright (c) 2024-2025, NVIDIA CORPORATION. All rights reserved.
// SPDX-License-Identifier: Apache-2.0

//------------------------------------------------------------------------------

// GB_cuda_init queries the system for the # of GPUs available, their memory
// sizes, SM counts, and other capabilities.  Unified Memory support is
// assumed.  Then each GPU is "warmed up" by allocating a small amount of
// memory.

// FIXME: remove printfs

#include "GB.h"

GrB_Info GB_cuda_init (void)
{
    GrB_Info info ;
    // get the GPU properties
    if (!GB_Global_gpu_count_set (true))
    {
        printf ("GB_cuda_init line %d\n", __LINE__) ;
        return (GxB_GPU_ERROR) ;
    }
    int gpu_count = GB_Global_gpu_count_get ( ) ;
    printf ("GB_cuda_init: ngpus: %d\n", gpu_count) ;
    for (int device = 0 ; device < gpu_count ; device++)
    {
        // query the GPU and then warm it up
        if (!GB_Global_gpu_device_properties_get (device))
        {
            printf ("GB_cuda_init line %d\n", __LINE__) ;
            return (GxB_GPU_ERROR) ;
        }
    }

    // initialize RMM if necessary
    if (!rmm_wrap_is_initialized ())
    {
        rmm_wrap_initialize_all_same (rmm_wrap_managed,
            // FIXME ask the GPU(s) for good default values.  This might be
            // found by GB_cuda_init.  Perhaps GB_cuda_init needs to be split
            // into 2 methods: one to query the sizes(s) of the GPU(s) then
            // call rmm_wrap_initialize_all_same, and the other for the rest
            // of the work.  Alternatively, move GB_cuda_init here (if so,
            // ensure that it doesn't depend on any other initializations
            // below).
            256 * 1000000L, 1024 * 100000000L, 1) ; // FIXME: ask the GPU(s)
    }

    // warm up the GPUs
    for (int device = 0 ; device < gpu_count ; device++)
    {
        if (!GB_cuda_warmup (device))
        {
            printf ("GB_cuda_init line %d\n", __LINE__) ;
            return (GxB_GPU_ERROR) ;
        }
    }

    info = GB_cuda_stream_pool_init ( ) ;
    if (info != GrB_SUCCESS)
    {
        printf ("GB_cuda_init line %d\n", __LINE__) ;
        return info ;
    }

    GB_cuda_set_device (0) ;            // make GPU 0 the default device
    GB_Context_gpu_id_set (NULL, 0) ;   // set GxB_CONTEXT_WORLD->gpu_id to 0

    // also check for jit cache, pre-load library of common kernels ...
    return (GrB_SUCCESS) ;
}