Precise dependencies compatibility #785
Conversation
7222048 to
8b96893
Compare
8b96893 to
0cae494
Compare
|
@inducer This is ready for another look. |
bcb356a to
47d3c01
Compare
47d3c01 to
5c84e95
Compare
inducer
left a comment
There was a problem hiding this comment.
FWIW, I will fix this up in the interest of getting it in soon.
| shared_inames_order_before = [ | ||
| domain_before.get_dim_name(dim_type.out, idim) | ||
| for idim in range(domain_before.dim(dim_type.out)) | ||
| if domain_before.get_dim_name(dim_type.out, idim) | ||
| in shared_inames] | ||
|
|
||
| shared_inames_order_after = [ | ||
| domain_after.get_dim_name(dim_type.out, idim) | ||
| for idim in range(domain_after.dim(dim_type.out)) | ||
| if domain_after.get_dim_name(dim_type.out, idim) | ||
| in shared_inames] | ||
|
|
||
| assert shared_inames_order_after == shared_inames_order_before | ||
| shared_inames_order = shared_inames_order_after |
There was a problem hiding this comment.
This snippet here would imply that the order of variables in a set is significant. This is not the case anywhere else in loopy, and I would prefer to not start having it be the case now. This behavior is also not documented.
There was a problem hiding this comment.
So... I tried, but I ran out of time. 🙂 There's a bit of algorithmic work to be done here, and some important algorithmic pieces are in #690. We can discuss in more detail during our meeting Monday.
Rather than hijack this PR for my big pile of unrelated changes, that's now #864, which this depends on. The relevant pieces from #690 are also in #864, so you'll be able to rely on them.
b475e5a to
507792b
Compare
507792b to
587f381
Compare
587f381 to
1a6548c
Compare
HappensAfterdata structureloopy/kernel/dependency.py)InstructionBaseto accommodate precise dependency semantics (depends_on->happens_after)depends_onis nowhappens_afterNeeds:
UniqueNameGeneratorpytools#255 and a pytools release