blob: f43ab7ed155a3b528dcd671d2f5ba020917ff3cf (
plain)
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
|
Mixed Buffer Pointer
--------------------
Segmented address space...
--------------------------
The first challenge with OpenCL is its very liberal use of pointers. The memory
is segment into several address spaces:
- private. This is the memory for each work item
- global. These are buffers in memory shared by all work items and work groups
- constant. These are constant buffers in memory shared by all work items and
work groups as well
- local. These is a memory shared by all work items in the *same* work group
... But with no restriction inside each address space
-----------------------------------------------------
The challenge is that there is no restriction in OpenCL inside each address
space i.e. the full C semantic applies in particular regarding pointer
arithmetic.
Therefore the following code is valid:
<code>
\_\_kernel void example(\_\_global int *dst, \_\_global int *src0, \_\_global int *src1)<br/>
{<br/>
\_\_global int *from;<br/>
if (get\_global\_id(0) % 2)<br/>
from = src0;<br/>
else<br/>
from = src1;<br/>
dst[get\_global\_id(0)] = from[get\_global\_id(0)];<br/>
}
</code>
As one may see, the load done in the last line actually mixes pointers from both
source src0 and src1. This typically makes the use of binding table indices
pretty hard. In we use binding table 0 for dst, 1 for src0 and 2 for src1 (for
example), we are not able to express the load in the last line with one send
only. The pointer "from" in the last line is so called a mixed buffer pointer.
(To be updated)
|
