compiler: Introduce symbolic fencing #2244
Conversation
Codecov Report
@@ Coverage Diff @@
## master #2244 +/- ##
==========================================
- Coverage 86.91% 86.89% -0.03%
==========================================
Files 229 229
Lines 41573 41687 +114
Branches 7665 7695 +30
==========================================
+ Hits 36133 36223 +90
- Misses 4818 4835 +17
- Partials 622 629 +7
|
|
What is the overarching target of this PR? |
| all(a.is_regular for a in self.scope.accesses)) | ||
|
|
||
| @cached_property | ||
| def is_sparse(self): | ||
| """ | ||
| A Cluster is sparse if it represents a sparse operation, i.e iff | ||
| There's at least one irregular access. | ||
| True if it represents a sparse operation, i.e iff there's at least |
There was a problem hiding this comment.
if and only if?
| # Synchronization operations prevent lifting | ||
| if c.syncs.get(dim): | ||
| # Synchronization prevents lifting | ||
| if c.syncs.get(dim) or \ |
There was a problem hiding this comment.
Does this get too long as one line?
There was a problem hiding this comment.
I do the splitting on purpose for clarity when I feel there's (relatively speaking) little correlation between the two operands of the relation
FabioLuporini
force-pushed
the
patch-lazy-threading
branch
from
f3b05b9 to
a000c49
Compare
Necessary tweaks for stuff in PRO |
|
basically it provides a series of mechanisms to impose that given subsequences of Clusters don't float around upon topological sorting, even if deemed possible by data dependence analysis Clusters sitckiness helps with:
I'm not sure how to explain this any better without showing the actual low-level details |
Thanks, these things could easily get into description to help review. |
No description provided.