[llvm] [AArch64] Don't allow mixed partial reductions without i8mm (PR #137602)

[Sander de Smalen via llvm-commits]

================
@@ -1,6 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "^scalar.ph:" --version 4
-; RUN: opt -passes=loop-vectorize -enable-epilogue-vectorization=false -mattr=+i8mm,+dotprod -S < %s | FileCheck %s
-; RUN: opt -passes=loop-vectorize -enable-epilogue-vectorization=false -mattr=+dotprod -S < %s | FileCheck %s --check-prefix=CHECK-NOI8MM
+; RUN: opt -passes=loop-vectorize -enable-epilogue-vectorization=false -mattr=+i8mm,+dotprod --vectorizer-maximize-bandwidth -S < %s | FileCheck %s
+; RUN: opt -passes=loop-vectorize -enable-epilogue-vectorization=false -mattr=+dotprod -S --vectorizer-maximize-bandwidth < %s | FileCheck %s --check-prefix=CHECK-NOI8MM
----------------
sdesmalen-arm wrote:

Sorry for the unfortunate timing, but I've just come to realise that this flag is probably required if you want the SVE sudot/usdot tests to enforce scalable vectors. Otherwise, it would still pick a fixed-length VF for the `+i8mm` case, where you actually want to test this specifically uses a scalable VF. Could you remove the last commit (486aa4ca815823000d9422b447947f2cabf4b50c) ?

https://github.com/llvm/llvm-project/pull/137602