[RISCV] Allow non-loop invariant steps in RISCVGatherScatterLowering (#122244)

The motivation for this is to allow us to match strided accesses that
are emitted from the loop vectorizer with EVL tail folding (see #122232)

In these loops the step isn't loop invariant and is based off of
@llvm.experimental.get.vector.length.

We can relax this as long as we make sure to construct the updates after
the definition inside the loop, instead of the preheader.

I presume the restriction was previously added so that the step would
dominate the insertion point in the preheader. I can't think of why it
wouldn't be safe to calculate it in the loop otherwise.

Author: lukel97

llvm/lib/Target/RISCV/RISCVGatherScatterLowering.cpp

@@ -211,10 +211,6 @@ bool RISCVGatherScatterLowering::matchStridedRecurrence(Value *Index, Loop *L,
     assert(Phi->getIncomingValue(IncrementingBlock) == Inc &&
            "Expected one operand of phi to be Inc");
 
-    // Only proceed if the step is loop invariant.
-    if (!L->isLoopInvariant(Step))
-      return false;
-
     // Step should be a splat.
     Step = getSplatValue(Step);
     if (!Step)
@@ -298,6 +294,7 @@ bool RISCVGatherScatterLowering::matchStridedRecurrence(Value *Index, Loop *L,
       BasePtr->getIncomingBlock(StartBlock)->getTerminator());
   Builder.SetCurrentDebugLocation(DebugLoc());
 
+  // TODO: Share this switch with matchStridedStart?
   switch (BO->getOpcode()) {
   default:
     llvm_unreachable("Unexpected opcode!");
@@ -310,18 +307,32 @@ bool RISCVGatherScatterLowering::matchStridedRecurrence(Value *Index, Loop *L,
   }
   case Instruction::Mul: {
     Start = Builder.CreateMul(Start, SplatOp, "start");
-    Step = Builder.CreateMul(Step, SplatOp, "step");
     Stride = Builder.CreateMul(Stride, SplatOp, "stride");
     break;
   }
   case Instruction::Shl: {
     Start = Builder.CreateShl(Start, SplatOp, "start");
-    Step = Builder.CreateShl(Step, SplatOp, "step");
     Stride = Builder.CreateShl(Stride, SplatOp, "stride");
     break;
   }
   }
 
+  // If the Step was defined inside the loop, adjust it before its definition
+  // instead of in the preheader.
+  if (auto *StepI = dyn_cast<Instruction>(Step); StepI && L->contains(StepI))
+    Builder.SetInsertPoint(*StepI->getInsertionPointAfterDef());
+
+  switch (BO->getOpcode()) {
+  default:
+    break;
+  case Instruction::Mul:
+    Step = Builder.CreateMul(Step, SplatOp, "step");
+    break;
+  case Instruction::Shl:
+    Step = Builder.CreateShl(Step, SplatOp, "step");
+    break;
+  }
+
   Inc->setOperand(StepIndex, Step);
   BasePtr->setIncomingValue(StartBlock, Start);
   return true;

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-strided-load-store.ll

@@ -320,8 +320,8 @@ for.cond.cleanup:                                 ; preds = %vector.body
 define void @gather_unknown_pow2(ptr noalias nocapture %A, ptr noalias nocapture readonly %B, i64 %shift) {
 ; CHECK-LABEL: @gather_unknown_pow2(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[STEP:%.*]] = shl i64 8, [[SHIFT:%.*]]
-; CHECK-NEXT:    [[STRIDE:%.*]] = shl i64 1, [[SHIFT]]
+; CHECK-NEXT:    [[STRIDE:%.*]] = shl i64 1, [[SHIFT:%.*]]
+; CHECK-NEXT:    [[STEP:%.*]] = shl i64 8, [[SHIFT]]
 ; CHECK-NEXT:    [[TMP0:%.*]] = mul i64 [[STRIDE]], 4
 ; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
 ; CHECK:       vector.body: