File tree 1 file changed +9
-3
lines changed
1 file changed +9
-3
lines changed Original file line number Diff line number Diff line change @@ -23,8 +23,14 @@ Tensor Mapping APIs
23
23
Motivation
24
24
--------------------------------------------------------------------------------
25
25
26
- Used to target the CUDA entry points cuTensorMapEncodeIm2col and
27
- cuTensorMapEncodeTiled.
26
+ Used to target the CUDA entry points ``cuTensorMapEncodeIm2col `` and
27
+ ``cuTensorMapEncodeTiled ``.
28
+
29
+ For some tensor core operations on ``sm_90+ `` Nvidia devices, a tensor
30
+ descriptor must be built on the host and passed to the kernel. The interfaces
31
+ mentioned above, and mapped to UR in this extension, provide the APIs necessary
32
+ to create these tensor descriptor objects, that can then be passed to the
33
+ kernels.
28
34
29
35
API
30
36
--------------------------------------------------------------------------------
@@ -61,7 +67,7 @@ Changelog
61
67
Support
62
68
--------------------------------------------------------------------------------
63
69
64
- This is only supported in the CUDA adapter .
70
+ This extension is only supported on the `` UR_PLATFORM_BACKEND_CUDA `` backend .
65
71
66
72
Contributors
67
73
--------------------------------------------------------------------------------
You can’t perform that action at this time.
0 commit comments