-
Notifications
You must be signed in to change notification settings - Fork 761
Use CUB version of merge sort #1511
Use CUB version of merge sort #1511
Conversation
|
run tests |
f00fabe to
a212bb6
Compare
|
run tests |
a212bb6 to
6cc34dc
Compare
|
run tests |
|
DVS: 30345552 |
6cc34dc to
83eaecc
Compare
|
run tests |
|
@senior-zero Let's try to get this in for 1.15 if DVS is passing and it's ready to merge, but this is super low priority if it's going to take actual effort. |
alliepiper
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
thrust/system/cuda/detail/sort.h
Outdated
| } | ||
|
|
||
| for (int pass = 0; pass < num_passes; ++pass, ping = !ping) | ||
| if (STABLE::value) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
These may need to be THRUST_IF_CONSTEXPR if tests complain about the constexpr arg.
|
run tests |
13707fb to
5a17e55
Compare
|
run tests |
5a17e55 to
f7e39a8
Compare
|
run tests |
|
DVS CL: 30537652 |
b16eae9 to
ce4c107
Compare
|
run tests |
|
Hoping DVS behaves this time 😅 DVS CL: 30549730 |
|
The DVS builds are all timing out with this update, and gpuCI seems to be taking about an hour longer than before. We should track down why and fix this before merging. Moving to 1.16 since this isn't urgent for 1.15. |
alliepiper
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Need to track down compile time increases between old and new merge sort implementations.
ce4c107 to
9dd2f22
Compare
|
run tests |
9dd2f22 to
6192d9a
Compare
|
run tests |
6192d9a to
91d5e6c
Compare
This PR removes the Thrust version of merge sort and uses CUB one instead.