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
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
|
use ocl::{flags, Buffer, Context, Device, Kernel, Platform, Program, Queue};
use std::sync::mpsc::{Receiver, Sender};
pub struct Job {
rows: Vec<u32>,
bitmask: u64,
}
pub struct GpuSolver {
platform: Platform,
device: Device,
context: Context,
program: Program,
queue: Queue,
n: u32,
h: u32,
w: u32,
wg_size: usize,
permutations: Buffer<u64>,
rec_queues: Vec<ReqestBuffer>,
}
struct ReqestBuffer {
mask_buff: Vec<u64>,
row_buff: Vec<Vec<u32>>,
pointer: usize,
receiver: Receiver<Job>,
}
impl ReqestBuffer {
pub fn new(size: usize, receiver: Receiver<Job>) -> ReqestBuffer {
Self {
mask_buff: Vec::with_capacity(size),
row_buff: Vec::with_capacity(size),
pointer: 0,
receiver,
}
}
pub fn read(&mut self) -> Option<&[u64]> {
for job in self.receiver.try_iter() {
self.mask_buff[self.pointer] = job.bitmask;
self.row_buff[self.pointer] = job.rows;
self.pointer += 1;
if self.pointer == self.mask_buff.len() {
self.pointer = 0;
return Some(self.mask_buff.as_ref());
}
}
None
}
pub fn get_rows(&self) -> &[Vec<u32>] {
self.row_buff.as_ref()
}
}
impl GpuSolver {
pub fn new(
permutation_masks: &[u64],
n: u32,
h: u32,
w: u32,
src: &str,
) -> ocl::Result<Vec<Sender<Job>>> {
let platform = ocl::Platform::default();
let device = ocl::Device::first(platform)?;
let context = ocl::Context::builder()
.platform(platform)
.devices(device.clone())
.build()?;
let queue = ocl::Queue::new(&context, device, None)?;
let program = Program::builder()
.devices(device)
.src(src)
.build(&context)?;
let buffer = ocl::Buffer::builder()
.queue(queue.clone())
.flags(flags::MEM_READ_WRITE)
.copy_host_slice(permutation_masks)
.len(permutation_masks.len())
.build()?;
let mut senders = Vec::with_capacity(h as usize);
let mut receivers = Vec::with_capacity(h as usize);
let wg_size = device.max_wg_size()?;
for _ in 0..h {
let (sx, rx) = std::sync::mpsc::channel();
senders.push(sx);
receivers.push(ReqestBuffer::new(wg_size, rx));
}
let solver = Self {
platform,
device,
context,
program,
queue,
n,
h,
w,
wg_size,
permutations: buffer,
rec_queues: receivers,
};
std::thread::spawn(move || {
solver.run();
});
Ok(senders)
}
fn run(mut self) -> ! {
let queues = self.rec_queues.len();
let mut instruction_buffer = Vec::with_capacity((self.n - self.h) as usize);
let mut result_buffer = Vec::with_capacity((self.n - self.h) as usize);
let chunk = self.permutations.len() / self.n as usize;
for i in 0..queues {
let buffer: Buffer<u64> = Buffer::builder()
.queue(self.queue.clone())
.len(self.wg_size)
.flags(flags::MEM_READ_WRITE)
.build()
.unwrap();
instruction_buffer.push(buffer);
let dim = (i + 1) * chunk;
let results: Buffer<u64> = Buffer::builder()
.queue(self.queue.clone())
.len(self.wg_size * dim / 64)
.flags(flags::MEM_READ_WRITE)
.build()
.unwrap();
result_buffer.push(results);
}
for i in (0..self.rec_queues.len()).cycle() {
if let Some(buffer) = self.rec_queues[i].read() {
instruction_buffer[i].write(buffer).enq().unwrap();
println!("hello world");
let dim = (i + 1) * chunk;
let kernel = Kernel::builder()
.program(&self.program)
.name("check")
.queue(self.queue.clone())
.global_work_size(dim)
.arg(&self.permutations)
.arg(&result_buffer[i])
.arg(&instruction_buffer[i])
.arg(self.n)
.arg(self.w)
.arg((self.n as u64 - i as u64 - 1) * chunk as u64)
.build()
.unwrap();
unsafe {
kernel
.cmd()
.queue(&self.queue)
.global_work_offset(kernel.default_global_work_offset())
.global_work_size(dim)
.local_work_size(self.wg_size)
.enq()
.unwrap();
}
// (5) Read results from the device into a vector (`::block` not shown):
let mut result = vec![0u64; dim * self.wg_size / 64];
result_buffer[i]
.cmd()
.queue(&self.queue)
.offset(0)
.read(&mut result)
.enq()
.unwrap();
println!("{:?}", result);
}
}
panic!();
}
}
/*
pub fn check(permutations: &[u64], w: u32, n: u32, mask: u64, offset: usize) -> ocl::Result<()> {
//println!("read src!");
let src = std::fs::read_to_string("src/solvers/check.cl").expect("failed to open kernel file");
//println!("created queue!");
println!("offset: {}", offset);
println!("length: {}", permutations.len() - offset);
let pro_que = ocl::ProQue::builder()
.src(src)
.dims(permutations.len() - offset)
.build()?;
let results = pro_que.create_buffer::<i32>()?;
let kernel = pro_que
.kernel_builder("check")
.arg(get_buffer())
.arg(&results)
.arg(mask)
.arg(n)
.arg(w)
.arg(offset as u64)
//.global_work_offset(offset)
.build()?;
//println!("starting calculation");
unsafe {
kernel.enq()?;
}
let mut vec = vec![0; results.len()];
results.read(&mut vec).enq()?;
if vec.iter().any(|x| *x != 0) {
println!("The resuts are now '{:?}'!", vec);
}
Ok(())
}*/
|
