| Issue |
90418
|
| Summary |
[mlir][Vector] Add support for `i2` vector emulation
|
| Labels |
mlir,
mlir:vector
|
| Assignees |
KoolJBlack,
banach-space
|
| Reporter |
dcaballe
|
[We recently added support](https://github.com/llvm/llvm-project/blob/aa596fa4d974f75ed8d2db3f4880ec0e5be3e176/mlir/lib/Dialect/Vector/Transforms/VectorEmulateNarrowType.cpp#L857) to generate fast vector sequences of instructions to "emulate" conversions from `i4` to byte+ integer and 32-bit floating point types. This issue is to track the same work for `i2` data types. We should add support for signed and unsigned extensions to byte and floating point types. The existing `i4` support should be the foundation of the implementation for `i2` (we may need to loop into that code for the `i2` case, literally). The tests added in the `i4` support can help define the cases that need to be supported for `i2`.
![]()
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
