projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 47e7bdb) | patch
Allow IndexType inside tensors.
authorSean Silva <silvasean@google.com>
Tue, 24 Mar 2020 22:13:31 +0000 (15:13 -0700)
committerSean Silva <silvasean@google.com>
Thu, 26 Mar 2020 17:52:48 +0000 (10:52 -0700)
commit3dceb6d2460550ef8c2e53df2181c8a11a7e6730
tree55b4e4b69f14a14e29e2e4f4e0485373b226c578tree | snapshot
parent47e7bdb10732e6f140adce39a1bc34e3ee2a6ea6commit | diff
Allow IndexType inside tensors.

It's common in many dialects to use tensors to themselves hold tensor shapes (for example, the shape is itself the result of some non-trivial calculation). Currently, such dialects have to use `tensor<?xi64>` or worse (like allowing either i32 or i64 tensors to represent shapes). `tensor<?xindex>` is the natural type to represent this, but is currently disallowed. This patch allows it.

Differential Revision: https://reviews.llvm.org/D76726
mlir/docs/Rationale.md diff | blob | history
mlir/include/mlir/IR/StandardTypes.h diff | blob | history
mlir/test/IR/invalid.mlir diff | blob | history
Domain: System / Toolchain;