Cleanup trivial reduction workarounds #2006
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -1597,6 +1597,43 @@ std::vector<IterDomain*> IterDomain::clone( | |
| return cloned_domains; | ||
| } | ||
|
|
||
| IterType inferIterType(IterDomain* i1, IterDomain* i2) { | ||
|
There was a problem hiding this comment. @csarofeen I don't have any concern, but please take a look at here just in case. There was a problem hiding this comment. Only nit I have is changing the name to understand it only handles resolution through serial, broadcast, and trivial reduced domains. In other words it shouldn't be used for like gather/stride IDs. Nice cleanup though. There was a problem hiding this comment. I think for gather and stride, it applies the first rule |
||
| // The itertype inference is a pattern matching of the rules below: | ||
| // | ||
| // X + X = X | ||
| // trivial reduction + X = X | ||
| // X + trivial reduction = X | ||
| // broadcasting + X = X | ||
| // X + broadcasting = X | ||
| // fail | ||
| // | ||
| // The rules are proceeded one by one in order. For each rule, we test if the | ||
| // given (outer, inner) matches the pattern. If it does, then we stop | ||
| // procceeding and get a result. If we have reached the end without finding | ||
| // any matched pattern, then it is a mistake and should be reported. | ||
| // | ||
| // Note that based on the above rule: | ||
| // broadcasting + (non-trivial) reduction = reduction | ||
| // broadcasting + trivial reduction = broadcasting | ||
| if (i1->getIterType() == i2->getIterType()) { | ||
| return i1->getIterType(); | ||
| } | ||
| if (i1->isTrivialReduction()) { | ||
| return i2->getIterType(); | ||
| } | ||
| if (i2->isTrivialReduction()) { | ||
| return i1->getIterType(); | ||
| } | ||
| if (i1->isBroadcast()) { | ||
| return i2->getIterType(); | ||
| } | ||
| if (i2->isBroadcast()) { | ||
| return i1->getIterType(); | ||
| } | ||
| TORCH_CHECK( | ||
| false, "Merging IterDomains requires that their iteration types match."); | ||
| } | ||
|
|
||
| // Merging does not propagate the start and stop values of the input | ||
| // domains to the merged output domain. The actual range of the | ||
| // domains is enforced by predicates. Note that since only root | ||
|
|
@@ -1606,48 +1643,10 @@ IterDomain* IterDomain::merge(IterDomain* outer, IterDomain* inner) { | |
| TORCH_CHECK( | ||
| !outer->extent()->isZeroInt() && !inner->extent()->isZeroInt(), | ||
| "Merging IterDomains with ending values that are 0 is not supported at this time."); | ||
zasdfgbnm marked this conversation as resolved.
Show resolved
Hide resolved
|
||
|
|
||
| Val* merged_id_size = mul(outer->extent(), inner->extent()); | ||
|
|
||
| IterType itype = inferIterType(outer, inner); | ||
|
|
||
| Val* expanded_extent = nullptr; | ||
| if (outer->hasExpandedExtent() || inner->hasExpandedExtent()) { | ||
|
|
||
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -330,13 +330,8 @@ void multiReductionInliner( | |
| } | ||
| } | ||
|
|
||
| // Inline the schedule | ||
| inlineMost(); | ||
| } | ||
|
|
||
| namespace { | ||
|
|
||
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
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.
For now this seems fine to remove, though I'm open minded we want ID based avoidance of inlining certain dimensions, but we probably want a better interface for that.