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
|
use crate::common::CudaDriverFns;
use cuda_types::*;
use rand::{Rng, SeedableRng};
use std::{ffi::c_void, mem, ptr};
mod common;
cuda_driver_test!(shuffle_down);
cuda_driver_test!(shuffle_up);
cuda_driver_test!(shuffle_bfly);
cuda_driver_test!(shuffle_idx);
const KERNEL: &'static str = include_str!("shuffle.ptx");
const WARP_WIDTH: usize = 32;
const TEST_ITERATIONS: usize = 1000;
unsafe fn shuffle_down<T: CudaDriverFns>(cuda: T) {
shuffle(cuda, "down", validate_down);
}
unsafe fn shuffle_up<T: CudaDriverFns>(cuda: T) {
shuffle(cuda, "up", validate_up);
}
unsafe fn shuffle_bfly<T: CudaDriverFns>(cuda: T) {
shuffle(cuda, "bfly", validate_bfly);
}
unsafe fn shuffle_idx<T: CudaDriverFns>(cuda: T) {
shuffle(cuda, "idx", validate_idx);
}
unsafe fn shuffle<T: CudaDriverFns>(
cuda: T,
shuffle_type: &'static str,
mut validate: impl FnMut(&[u32; WARP_WIDTH], u32, u32, &[u32; WARP_WIDTH]) -> bool,
) {
assert_eq!(cuda.cuInit(0), CUresult::CUDA_SUCCESS);
let mut ctx = ptr::null_mut();
assert_eq!(
cuda.cuCtxCreate_v2(&mut ctx, 0, CUdevice_v1(0)),
CUresult::CUDA_SUCCESS
);
let mut kernel_text = KERNEL.replace("#SHUFFLE#", shuffle_type);
kernel_text.push('\0');
let mut module = mem::zeroed();
assert_eq!(
cuda.cuModuleLoadData(&mut module, kernel_text.as_ptr() as _),
CUresult::CUDA_SUCCESS
);
let mut kernel = mem::zeroed();
assert_eq!(
cuda.cuModuleGetFunction(&mut kernel, module, b"shuffle\0".as_ptr() as _),
CUresult::CUDA_SUCCESS
);
let mut input_mem = mem::zeroed();
assert_eq!(
cuda.cuMemAlloc_v2(&mut input_mem, WARP_WIDTH * mem::size_of::<u32>()),
CUresult::CUDA_SUCCESS
);
let mut output_mem = mem::zeroed();
assert_eq!(
cuda.cuMemAlloc_v2(&mut output_mem, WARP_WIDTH * mem::size_of::<u32>()),
CUresult::CUDA_SUCCESS
);
let mut rng = rand_chacha::ChaCha8Rng::seed_from_u64(0x7cb9cbc7c2b95f47);
for _ in 00..TEST_ITERATIONS {
let input = rng.gen::<[u32; WARP_WIDTH]>();
assert_eq!(
cuda.cuMemcpyHtoD_v2(
input_mem,
input.as_ptr() as _,
input.len() * mem::size_of::<u32>(),
),
CUresult::CUDA_SUCCESS
);
let mut b = rng.gen::<u32>();
let mut c = rng.gen::<u32>();
let mut args = [
&mut input_mem as *mut _ as *mut c_void,
&mut output_mem as *mut _ as _,
&mut b as *mut _ as _,
&mut c as *mut _ as _,
];
assert_eq!(
cuda.cuLaunchKernel(
kernel,
1,
1,
1,
32,
1,
1,
0,
0 as _,
args.as_mut_ptr() as _,
ptr::null_mut(),
),
CUresult::CUDA_SUCCESS
);
let output = [0u32; WARP_WIDTH];
assert_eq!(
cuda.cuMemcpyDtoH_v2(
output.as_ptr() as _,
output_mem,
output.len() * mem::size_of::<u32>(),
),
CUresult::CUDA_SUCCESS
);
assert_eq!(cuda.cuCtxSynchronize(), CUresult::CUDA_SUCCESS);
assert!(validate(&input, b, c, &output));
}
}
fn validate_down(input: &[u32; WARP_WIDTH], b: u32, c: u32, result: &[u32; WARP_WIDTH]) -> bool {
validate(mode_down, input, b, c, result)
}
fn validate_up(input: &[u32; WARP_WIDTH], b: u32, c: u32, result: &[u32; WARP_WIDTH]) -> bool {
validate(mode_up, input, b, c, result)
}
fn validate_bfly(input: &[u32; WARP_WIDTH], b: u32, c: u32, result: &[u32; WARP_WIDTH]) -> bool {
validate(mode_bfly, input, b, c, result)
}
fn validate_idx(input: &[u32; WARP_WIDTH], b: u32, c: u32, result: &[u32; WARP_WIDTH]) -> bool {
validate(mode_idx, input, b, c, result)
}
fn validate(
mut mode: impl FnMut(u32, i32, u32, u32, u32) -> (i32, bool),
input: &[u32; WARP_WIDTH],
b: u32,
c: u32,
result: &[u32; WARP_WIDTH],
) -> bool {
let bval = (b & 31) as i32;
let cval = c & 31;
let mask = (c >> 8) & 31;
let source = (0u32..WARP_WIDTH as u32)
.into_iter()
.map(|lane| input[(lane & 31) as usize])
.collect::<Vec<_>>();
let max_lane = (0u32..WARP_WIDTH as u32)
.into_iter()
.map(|lane| ((lane & 31) & (mask)) | (cval & !mask))
.collect::<Vec<_>>();
let min_lane = (0u32..WARP_WIDTH as u32)
.into_iter()
.map(|lane| (lane & 31) & (mask))
.collect::<Vec<_>>();
let expected = (0u32..WARP_WIDTH as u32)
.into_iter()
.zip(max_lane.iter().copied())
.zip(min_lane.iter().copied())
.map(|((lane, max_lane), min_lane)| {
let (mut j, pval) = mode(lane, bval, mask, max_lane, min_lane);
if !pval {
j = lane as i32;
}
source[j as usize]
})
.collect::<Vec<_>>();
eprintln!("{:?} {} {} {:?} {:?}", &input, b, c, &result, &expected);
expected == result
}
fn mode_up(lane: u32, bval: i32, _mask: u32, max_lane: u32, _min_lane: u32) -> (i32, bool) {
let j = (lane as i32) - bval;
let pval = j >= max_lane as i32;
(j, pval)
}
fn mode_down(lane: u32, bval: i32, _mask: u32, max_lane: u32, _min_lane: u32) -> (i32, bool) {
let j = (lane as i32) + bval;
let pval = j <= max_lane as i32;
(j, pval)
}
fn mode_bfly(lane: u32, bval: i32, _mask: u32, max_lane: u32, _min_lane: u32) -> (i32, bool) {
let j = (lane as i32) ^ bval;
let pval = j <= max_lane as i32;
(j, pval)
}
fn mode_idx(_lane: u32, bval: i32, mask: u32, max_lane: u32, min_lane: u32) -> (i32, bool) {
let j = (min_lane as i32) | (bval & !(mask as i32));
let pval = j <= max_lane as i32;
(j, pval)
}
|
