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
|
// SPDX-License-Identifier: Apache-2.0
/*
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
/*
Extended example for building on-the-fly kernels with C interface.
Simple examples demonstrating different ways to load source code
and call kernels.
*/
#ifndef GB_REDUCE_JITFACTORY_H
#define GB_REDUCE_JITFACTORY_H
#pragma once
#include "GB_cuda_reduce_factory.hpp"
/**
* This file is responsible for picking all the parameters and what kernel variaiton we will use for a given instance
* - data types
* - semiring types
* - binary ops
* - monoids
*
* Kernel factory says "Here's the actual instance I want you to build with the given parameters"
*/
//bool GB_cuda_reduce(int64_t *index, void *in_data, void *output, unsigned int N, GrB_Monoid op);
//Kernel jitifiers
class reduceFactory ;
template<typename T1, typename T2, typename T3> class spdotFactory ;
//------------------------------------------------------------------------------
// reduceFactory
//------------------------------------------------------------------------------
class reduceFactory
{
std::string base_name = "GB_jit";
std::string kernel_name = "reduceNonZombiesWarp";
int threads_per_block = 320 ;
int work_per_thread = 256;
int number_of_sms = GB_Global_gpu_sm_get (0);
GB_cuda_reduce_factory &reduce_factory_;
public:
reduceFactory(GB_cuda_reduce_factory &myreducefactory) : reduce_factory_(myreducefactory) {}
int get_threads_per_block() {
return threads_per_block;
}
int get_number_of_blocks(unsigned int N) {
return (N + work_per_thread*threads_per_block - 1)/(work_per_thread*threads_per_block);
}
// Note: this does assume the erased types are compatible w/ the monoid's ztype
bool jitGridBlockLaunch(GrB_Matrix A, void* output,
GrB_Monoid op, cudaStream_t stream = 0)
{
GBURBLE ("\n(launch reduce factory) \n") ;
GrB_Scalar temp_scalar;
GrB_Scalar_new(&temp_scalar, op->op->ztype);
cuda::jit::scalar_set_element(temp_scalar, 0);
GrB_Scalar_wait(temp_scalar, GrB_MATERIALIZE);
jit::GBJitCache filecache = jit::GBJitCache::Instance() ;
filecache.getFile (reduce_factory_) ;
auto rcode = std::to_string(reduce_factory_.rcode);
std::string hashable_name = base_name + "_" + kernel_name;
std::stringstream string_to_be_jitted ;
string_to_be_jitted <<
hashable_name << std::endl <<
R"(#include ")" << jit::get_user_home_cache_dir() << "/" << reduce_factory_.filename << R"(")" << std::endl <<
R"(#include ")" << hashable_name << R"(.cuh")" << std::endl;
bool is_sparse = GB_IS_SPARSE(A);
int64_t N = is_sparse ? GB_nnz(A) : GB_NCOLS(A) * GB_NROWS(A);
int blocksz = get_threads_per_block();
int gridsz = get_number_of_blocks(N);
dim3 grid(gridsz);
dim3 block(blocksz);
// FIXME: call GB_stringify_reduce to create GB_ADD and related
// macros, in an include file: GB_reduce_123412341234.h
GBURBLE ("(cuda reduce launch %d threads in %d blocks)", blocksz, gridsz ) ;
jit::launcher(hashable_name + "_" + rcode,
string_to_be_jitted.str(),
header_names,
compiler_flags,
file_callback)
.set_kernel_inst( kernel_name , { A->type->name, op->op->ztype->name, rcode, "true" })
.configure(grid, block, SMEM, stream)
// FIXME: GB_ADD is hardcoded into kernel for now
.launch( A, temp_scalar, N, is_sparse);
// Need to synchronize before copying result to host
CHECK_CUDA( cudaStreamSynchronize(stream) );
memcpy(output, temp_scalar->x, op->op->ztype->size);
rmm_wrap_free(temp_scalar);
return true;
}
};
//------------------------------------------------------------------------------
inline bool GB_cuda_reduce(GB_cuda_reduce_factory &myreducefactory,
GrB_Matrix A, void *output, GrB_Monoid op,
cudaStream_t stream = 0) {
reduceFactory rf(myreducefactory);
GBURBLE ("(starting cuda reduce)" ) ;
bool result = rf.jitGridBlockLaunch(A, output, op, stream);
GBURBLE ("(ending cuda reduce)" ) ;
return (result) ;
}
#endif
|
