MAINT: Bump rapids to 25.08 and some other versions - #145
Conversation
This is just a start/try. CI will pick up a legate with an older CCCL so this may or may not go through. Locally, I still have problems around `__half` (as typical, I guess), but without cupynumeric, so I should have one with CCCL/rapids-cmake already bumped... Signed-off-by: Sebastian Berg <sebastianb@nvidia.com>
| # unnecessary in the future. | ||
| target_compile_definitions("LegateDataframe" PUBLIC _LIBCUDACXX_HAS_NVFP16=1) | ||
| endif() | ||
| target_compile_definitions("LegateDataframe" PUBLIC LIBCUDACXX_ENABLE_HOST_NVFP16=1) |
There was a problem hiding this comment.
Should try to remove these before merging, but as a quick check couldn't hurt...
Signed-off-by: Sebastian Berg <sebastianb@nvidia.com>
This comment was marked as outdated.
This comment was marked as outdated.
Signed-off-by: Sebastian Berg <sebastianb@nvidia.com>
Signed-off-by: Sebastian Berg <sebastianb@nvidia.com>
|
Ack, this had worked locally with maybe a slightly older legate... Well, let's see if I can mess with some other definition, heh... |
…n't exist Signed-off-by: Sebastian Berg <sebastianb@nvidia.com>
Signed-off-by: Sebastian Berg <sebastianb@nvidia.com>
Signed-off-by: Sebastian Berg <sebastianb@nvidia.com>
Signed-off-by: Sebastian Berg <sebastianb@nvidia.com>
|
Ah, runners and cuda-version were all unrelated, but seems just as well (presuming it works). Somehow The actual problem for CPU is that |
|
Ah, I guess it's plausible that the import succeeds but actually doing anything doesn't... let me add a mini operation instead... |
…ated round for now Signed-off-by: Sebastian Berg <sebastianb@nvidia.com>
Signed-off-by: Sebastian Berg <sebastianb@nvidia.com>
Signed-off-by: Sebastian Berg <sebastianb@nvidia.com>
seberg
changed the title
|
OK, some of the updates here aren't strictly necessary, but I think they don't hurt either and things are passing. I have some bad PTX build locally (not sure why), but I don't think that should be a problem (although it means that the new legate isn't tested with multi-gpu setup yet). |
2 participants
This is just a start/try. CI will pick up a legate with an older CCCL so this may or may not go through.
Locally, I still have problems around
__half(as typical, I guess), but without cupynumeric, so I should have one with CCCL/rapids-cmake already bumped...