[PATCH] D101836: [LoopVectorize] Enable strict reductions when allowReordering() returns false

[Sander de Smalen via Phabricator via llvm-commits]

sdesmalen added inline comments.


================
Comment at: llvm/lib/Transforms/Vectorize/LoopVectorizationLegality.cpp:882
+  // reduction operations in-loop.
+  bool HasExactRdxVar =
+      (any_of(getReductionVars(), [&](auto &Reduction) -> bool {
----------------
Can you rewrite this to:

  // We can now only vectorize if all reductions with Exact FP math also
  // have the isOrdered flag set, which indicates that we can move the
  // reduction operations in-loop.
  return all_of(getReductionVars(), [&](auto &Reduction) -> bool {
    RecurrenceDescriptor RdxDesc = Reduction.second;
    return !RdxDesc.hasExactFPMath() || RdxDesc.isOrdered();
  });


================
Comment at: llvm/test/Transforms/LoopVectorize/AArch64/scalable-strict-fadd.ll:7
 define float @fadd_strict(float* noalias nocapture readonly %a, i64 %n) {
-; CHECK-LABEL: @fadd_strict
-; CHECK: vector.body:
-; CHECK: %[[VEC_PHI:.*]] = phi float [ 0.000000e+00, %vector.ph ], [ %[[RDX:.*]], %vector.body ]
-; CHECK: %[[LOAD:.*]] = load <vscale x 8 x float>, <vscale x 8 x float>*
-; CHECK: %[[RDX]] = call float @llvm.vector.reduce.fadd.nxv8f32(float %[[VEC_PHI]], <vscale x 8 x float> %[[LOAD]])
-; CHECK: for.end
-; CHECK: %[[PHI:.*]] = phi float [ %[[SCALAR:.*]], %for.body ], [ %[[RDX]], %middle.block ]
-; CHECK: ret float %[[PHI]]
+; CHECK-VF8UF1-LABEL: @fadd_strict
+; CHECK-VF8UF1: vector.body:
----------------
Should all test functions have check lines for all VF8UF1, VF8UF4, etc. ? Conversely, is it sufficient to just pass the interleave-count hint (not the vector width) via metadata and have 1 RUN line for VF8UF1, VF8UF4, VF4UF1?

Which also makes me wonder, what is the additional value of having both VF8UF1 and VF4UF1 ?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D101836/new/

https://reviews.llvm.org/D101836