[PATCH] D30086: Add generic IR vector reductions

[Mehdi AMINI via Phabricator via llvm-commits]

mehdi_amini added inline comments.


================
Comment at: docs/LangRef.rst:11620
+if the vector width is known and a power-of-two then then a shuffle sequence may
+be used instead.
+
----------------
Why this specification about implementation detail here?


================
Comment at: docs/LangRef.rst:11655
+reduction of a vector, returning the result as a scalar. The return type matches
+the element-type of the vector input.
+
----------------
I think it'd be important to clarify the ordering: do we require a "fast" FMF to change it? If not, why?


================
Comment at: docs/LangRef.rst:11767
+
+      declare <scalar-type> @llvm.vector.reduce.umax.<scalar-type>.<vector-type>(<vector-type> %a)
+
----------------
Why the explicit `<scalar-type>` in the signature?


================
Comment at: docs/LangRef.rst:11810
+an ``i32`` constant value which is ``1`` if it can be assumed that there are no
+NaN values in the vector, and ``0`` otherwise. The return type matches the
+element-type of the vector input.
----------------
There is no second operand in the prototype (And we have FMF for this purpose, so remove the sentence).


Repository:
  rL LLVM

https://reviews.llvm.org/D30086