Automerge: [LLVM][InstCombine][SVE] Improve isAllActivePredicate by looking through from.svbool. (#164446)

When a predicate is of the form "%a = sve.from.vsbool(%b)" we know all
bits in %a come from %b and thus if %b is all true then %a must also be
all true.

Author: paulwalker-arm

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

@@ -1577,18 +1577,26 @@ static SVEIntrinsicInfo constructSVEIntrinsicInfo(IntrinsicInst &II) {
 }
 
 static bool isAllActivePredicate(Value *Pred) {
-  // Look through convert.from.svbool(convert.to.svbool(...) chain.
   Value *UncastedPred;
+
+  // Look through predicate casts that only remove lanes.
   if (match(Pred, m_Intrinsic<Intrinsic::aarch64_sve_convert_from_svbool>(
-                      m_Intrinsic<Intrinsic::aarch64_sve_convert_to_svbool>(
-                          m_Value(UncastedPred)))))
-    // If the predicate has the same or less lanes than the uncasted
-    // predicate then we know the casting has no effect.
-    if (cast<ScalableVectorType>(Pred->getType())->getMinNumElements() <=
-        cast<ScalableVectorType>(UncastedPred->getType())->getMinNumElements())
-      Pred = UncastedPred;
+                      m_Value(UncastedPred)))) {
+    auto *OrigPredTy = cast<ScalableVectorType>(Pred->getType());
+    Pred = UncastedPred;
+
+    if (match(Pred, m_Intrinsic<Intrinsic::aarch64_sve_convert_to_svbool>(
+                        m_Value(UncastedPred))))
+      // If the predicate has the same or less lanes than the uncasted predicate
+      // then we know the casting has no effect.
+      if (OrigPredTy->getMinNumElements() <=
+          cast<ScalableVectorType>(UncastedPred->getType())
+              ->getMinNumElements())
+        Pred = UncastedPred;
+  }
+
   auto *C = dyn_cast<Constant>(Pred);
-  return (C && C->isAllOnesValue());
+  return C && C->isAllOnesValue();
 }
 
 // Simplify `V` by only considering the operations that affect active lanes.

llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-to-svbool-binops.ll

@@ -124,6 +124,39 @@ define <vscale x 8 x i1> @try_combine_svbool_binop_orr(<vscale x 8 x i1> %a, <vs
   ret <vscale x 8 x i1> %t3
 }
 
+; Verify predicate cast does not hinder "isAllActive" knowledge.
+define <vscale x 8 x half> @try_combine_svbool_binop_fadd(<vscale x 8 x half> %a, <vscale x 8 x half> %b) {
+; CHECK-LABEL: @try_combine_svbool_binop_fadd(
+; CHECK-NEXT:    [[T2:%.*]] = fadd <vscale x 8 x half> [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT:    ret <vscale x 8 x half> [[T2]]
+;
+  %t1 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> splat (i1 true))
+  %t2 = tail call <vscale x 8 x half> @llvm.aarch64.sve.fadd.nxv8f16(<vscale x 8 x i1> %t1, <vscale x 8 x half> %a, <vscale x 8 x half> %b)
+  ret <vscale x 8 x half> %t2
+}
+
+; Verify predicate cast does not hinder "isAllActive" knowledge.
+define <vscale x 4 x float> @try_combine_svbool_binop_fmul(<vscale x 4 x float> %a, <vscale x 4 x float> %b) {
+; CHECK-LABEL: @try_combine_svbool_binop_fmul(
+; CHECK-NEXT:    [[T2:%.*]] = fmul <vscale x 4 x float> [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT:    ret <vscale x 4 x float> [[T2]]
+;
+  %t1 = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> splat (i1 true))
+  %t2 = tail call <vscale x 4 x float> @llvm.aarch64.sve.fmul.nxv4f32(<vscale x 4 x i1> %t1, <vscale x 4 x float> %a, <vscale x 4 x float> %b)
+  ret <vscale x 4 x float> %t2
+}
+
+; Verify predicate cast does not hinder "isAllActive" knowledge.
+define <vscale x 2 x double> @try_combine_svbool_binop_fsub(<vscale x 2 x double> %a, <vscale x 2 x double> %b) {
+; CHECK-LABEL: @try_combine_svbool_binop_fsub(
+; CHECK-NEXT:    [[T2:%.*]] = fsub <vscale x 2 x double> [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT:    ret <vscale x 2 x double> [[T2]]
+;
+  %t1 = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> splat (i1 true))
+  %t2 = tail call <vscale x 2 x double> @llvm.aarch64.sve.fsub.nxv2f64(<vscale x 2 x i1> %t1, <vscale x 2 x double> %a, <vscale x 2 x double> %b)
+  ret <vscale x 2 x double> %t2
+}
+
 declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1>)
 declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1>)
 declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1>)