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
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
|
/*
* Copyright 2014-2015 The University of Queensland
* http://www.uq.edu.au
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#pragma once
#include <thrust/extrema.h>
#include <cusp/detail/device/arch.h>
#include <cusp/detail/device/common.h>
#include <cusp/detail/device/utils.h>
#include <cusp/detail/device/texture.h>
#include <cusp/detail/device/spmv/cds_symmetric.h>
#include <thrust/device_ptr.h>
namespace cusp
{
namespace detail
{
namespace device
{
////////////////////////////////////////////////////////////////////////
// CDS SpMV kernels
///////////////////////////////////////////////////////////////////////
//
// Diagonal matrices arise in grid-based discretizations using stencils.
// For instance, the standard 5-point discretization of the two-dimensional
// Laplacian operator has the stencil:
// [ 0 -1 0 ]
// [ -1 4 -1 ]
// [ 0 -1 0 ]
// and the resulting CDS format has 5 diagonals.
//
// spmv_cds
// Each thread computes y[i] += A[i,:] * x
// (the dot product of the i-th row of A with the x vector)
//
// spmv_cds_tex
// Same as spmv_cds, except x is accessed via texture cache.
//
template <typename IndexType, typename ValueType, unsigned int BLOCK_SIZE, bool UseCache>
__launch_bounds__(BLOCK_SIZE,1)
__global__ void
spmv_cds_kernel(const IndexType num_rows,
const IndexType num_diagonals,
const IndexType block_size,
const IndexType pitch,
const IndexType* diagonal_offsets,
const ValueType* values,
const ValueType* x,
ValueType* y)
{
__shared__ IndexType offsets[BLOCK_SIZE];
const IndexType thread_id = BLOCK_SIZE * blockIdx.x + threadIdx.x;
const IndexType grid_size = BLOCK_SIZE * gridDim.x;
//const float bsf = (float)block_size;
const IndexType num_cols = num_diagonals * block_size;
// for BLOCK_SIZE=256 this is most likely only executed once
for (IndexType base = 0; base < num_cols; base += BLOCK_SIZE)
{
// read a chunk of the diagonal offsets into shared memory
const IndexType chunk_size = thrust::min(IndexType(BLOCK_SIZE), num_cols - base);
if (threadIdx.x < chunk_size)
offsets[threadIdx.x] = diagonal_offsets[(base + threadIdx.x)/block_size] * block_size + threadIdx.x%block_size;
__syncthreads();
// process chunk
for (IndexType row = thread_id; row < num_rows; row += grid_size)
{
ValueType sum = (base == 0) ? ValueType(0) : y[row];
// index into values array
IndexType idx = row + pitch * base;
// for sm_10 this is faster than integer division
// block_size*(row/block_size)
//const IndexType colbase = block_size*(int)((row/bsf)+0.001f);
const IndexType colbase = block_size*(row/block_size);
for (IndexType n = 0; n < chunk_size; n++)
{
const IndexType col = colbase + offsets[n];
if (col >= 0 && col < num_rows)
{
const ValueType& A_ij = values[idx];
sum += A_ij * fetch_x<UseCache>(col, x);
}
idx += pitch;
}
y[row] = sum;
}
// wait until all threads are done reading offsets
__syncthreads();
}
}
template <bool UseCache,
typename Matrix,
typename ValueType>
void __spmv_cds(const Matrix& A,
const ValueType* x,
ValueType* y)
{
using cusp::detail::device::arch::max_active_blocks;
typedef typename Matrix::index_type IndexType;
const size_t BLOCK_SIZE = 256;
const IndexType num_diagonals = A.values.num_cols / A.block_size;
const IndexType pitch = A.values.pitch;
if (UseCache)
bind_x(x);
const size_t MAX_BLOCKS = max_active_blocks(spmv_cds_kernel<IndexType,
ValueType, BLOCK_SIZE, UseCache>, BLOCK_SIZE,
(size_t) sizeof(IndexType) * BLOCK_SIZE);
const size_t NUM_BLOCKS = std::min<size_t>(MAX_BLOCKS,
DIVIDE_INTO(A.num_rows, BLOCK_SIZE));
spmv_cds_kernel<IndexType, ValueType, BLOCK_SIZE, UseCache>
<<<NUM_BLOCKS, BLOCK_SIZE>>>
(A.num_rows, num_diagonals, A.block_size, pitch,
thrust::raw_pointer_cast(&A.diagonal_offsets[0]),
thrust::raw_pointer_cast(&A.values.values[0]), x, y);
if (UseCache)
unbind_x(x);
}
template <typename Matrix,
typename ValueType>
void spmv_cds(const Matrix& A,
const ValueType* x,
ValueType* y)
{
if (A.symmetric)
__spmv_cds_symmetric<false>(A, x, y);
else
__spmv_cds<false>(A, x, y);
}
template <typename Matrix,
typename ValueType>
void spmv_cds_tex(const Matrix& A,
const ValueType* x,
ValueType* y)
{
if (A.symmetric)
__spmv_cds_symmetric<true>(A, x, y);
else
__spmv_cds<true>(A, x, y);
}
} // end namespace device
} // end namespace detail
} // end namespace cusp
|
