[VPlan] Summarize recipes used to model inductions.

[Ayal]

Thanks for following up! Adding some thoughts to consider.

llvm/lib/Transforms/Vectorize/VPlan.h

1133

Worth indicating which of these produce a value per iteration, per part, per lane?

1135

Worth adding that this induction controls the vector loop, by comparing it with the vector trip count?
Starting value is specified as zero for main vector loop or where the main vector loop left off for the epilog vector loop (until fixed..).

1137

Worth adding that this is done by introducing an independent vector header phi?

1139

So perhaps a more consistent name would be VPScalarIntOrFpInductionRecipe?
Worth adding that these scalar values are computed based on the canonical IV recipe, rather than forming additional header phi's.

1141

So perhaps a more consistent name would be VPPointerInductionRecipe?

[fhahn]

Update comments, also document derived IV.

[fhahn]

llvm/lib/Transforms/Vectorize/VPlan.h

1133

Thanks, updated!

1135

Added, thanks!

1139

Updated the wording, but kept the name for now to just reflect the current realities in the documentation, thanks!

1141

I am planning to handle the non-widening case by using derived IV + scalar steps in future patches.

[Ayal]

Ship it, thanks!
Add [NFC] to title?
Relate to recent IV recipe patches in commit message?

llvm/lib/Transforms/Vectorize/VPlan.h

1138

nit: worth emphasizing - "a [single] scalar PHI"

1141

nit: "vector phis for each part" >> "a vector PHI per-part", to be more consistent with above and below?

1143

"Produces a single scalar value per iteration"

1147

"vector phis" - better use either "phi" or "PHI" consistently.

Worth adding that vector phis are produced per-part and scalar values per-lane?

[fhahn]

Thanks, comments should be addressed in the committed version!

llvm/lib/Transforms/Vectorize/VPlan.h

#endif
   /// Returns true if a vector phi needs to be created for the induction.
   bool needsVectorIV() const { return NeedsVectorIV; }
 };
 
 /// A pure virtual base class for all recipes modeling header phis, including
 /// phis for first order recurrences, pointer inductions and reductions. The
 /// start value is the first operand of the recipe and the incoming value from
 /// the backedge is the second operand.
+///
+/// Inductions are modeled using the following sub-classes:
+///  * VPCanonicalIVPHIRecipe: Canonical scalar induction of the vector loop,
+///    starting at a specified value (zero for the main vector loop, the resume
+///    value for the epilogue vector loop) and stepping by 1. The induction
+///    controls exiting of the vector loop by comparing against the vector trip
+///    count. Produces a single scalar PHI for the induction value per
+///    iteration.
+///  * VPWidenIntOrFpInductionRecipe: Generates vector values for integer and
+///    floating point inductions with arbitrary start and step values. Produces
+///    a vector PHI per-part.
+///  * VPDerivedIVRecipe: Converts the canonical IV value to the corresponding
+///    value of an IV with different start and step values. Produces a single
+///    scalar value per iteration
+///  * VPScalarIVStepsRecipe: Generates scalar values per-lane based on a
+///    canonical or derived induction.
+///  * VPWidenPointerInductionRecipe: Generate vector and scalar values for a
+///    pointer induction. Produces either a vector PHI per-part or scalar values
+///    per-lane based on the canonical induction.
 class VPHeaderPHIRecipe : public VPRecipeBase, public VPValue {
 protected:
   VPHeaderPHIRecipe(unsigned char VPVID, unsigned char VPDefID, PHINode *Phi,
                     VPValue *Start = nullptr)
       : VPRecipeBase(VPDefID, {}), VPValue(VPVID, Phi, this) {
     if (Start)
       addOperand(Start);
   }