[PATCH] D73572: [mlir] Expand shape functions in ShapeInference doc
Stella Laurenzo via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Sat Feb 1 10:00:31 PST 2020
stellaraccident accepted this revision.
stellaraccident added a comment.
This revision is now accepted and ready to land.
Herald added a subscriber: Joonsoo.
Some typos but lgtm. Thanks!
================
Comment at: mlir/docs/ShapeInference.md:59
+runtime and compiler (for constructions of ops/refinement of shapes, reification
+of dynamic allocations for dialect including TF, TFLite, XLA & tensor computer
+dialect under discussion).
----------------
Typo. s/computer/compute/?
================
Comment at: mlir/docs/ShapeInference.md:128
+ a declarative specification could be reused by both while avoiding a
+ need to map to a from a 3rd representation given these two systems
+ have/and will have different types.
----------------
Typo: should be "map to or from"
================
Comment at: mlir/docs/ShapeInference.md:216
+1. Flow sensitive shape functions;
+ * To enable scalable/cheap shape inference, the shapes functions do not
+ intent to provide flow sensitive information. This facility could
----------------
Typo: should be "the shape function" and "do not intend"
================
Comment at: mlir/docs/ShapeInference.md:229
+Shape functions should be lowerable to runtime checks for validity. E.g. verify
+as much as possible statically, but enable generating instruction to compute the
+shape dynamically and or falling back to runtime checks for attributes not
----------------
... generating instructions...
================
Comment at: mlir/docs/ShapeInference.md:247
+`SameOperandAndResultType` or broadcastable) we should generate the shape
+function from those. Where not an explicit shape function should be specified
+(spelling TBD but currently considering using the MLIR textual form as
----------------
Add comma: "Where not, ..."
================
Comment at: mlir/docs/ShapeInference.md:253
+
+This could be done in future! The extract shape function would use the shape
+inference dialect, so we are starting there. Especially for ops described in a
----------------
... The extracted shape...
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D73572/new/
https://reviews.llvm.org/D73572
More information about the llvm-commits
mailing list