tqchen commented on a change in pull request #8338:
URL: https://github.com/apache/tvm/pull/8338#discussion_r663035667
##########
File path: src/tir/transforms/storage_rewrite.cc
##########
@@ -504,8 +503,13 @@ class StoragePlanRewriter : public StmtExprMutator {
return MergeNest(nest, body);
}
// Remap the index
- PrimExpr RemapIndex(DataType dtype, PrimExpr index, StorageEntry* e) {
+ PrimExpr RemapIndex(PrimExpr index, StorageEntry* e) {
Review comment:
In the current setup(our code generation), the Load and Store follows
the dtype specified by the instruction, regardless of the dtype specified in
allocation. So it is better to use dtype in the Node. But you are right that we
should not multiply by lanes when calculating the offset
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
