[InstCombine] Increase test coverage of vector.reverse ready for follow on work.

diff --git a/llvm/test/Transforms/InstCombine/vector-reverse.ll b/llvm/test/Transforms/InstCombine/vector-reverse.ll
index 309c912..11d4c02 100644 (file)
--- a/llvm/test/Transforms/InstCombine/vector-reverse.ll
+++ b/llvm/test/Transforms/InstCombine/vector-reverse.ll
@@ -1,62 +1,707 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt < %s -passes=instcombine -S | FileCheck %s
 
-; Test that the reverse is eliminated if the output and all the inputs
-; of the instruction are calls to reverse.
+; Tests to ensure operand reversals are bubbled to the result when there is no
+; increase in the total number of reversals. Often this resuls in back to back
+; reversals that can be eliminated entirely. This outcome is tested separately.
+
 define <vscale x 4 x i32> @binop_reverse(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
 ; CHECK-LABEL: @binop_reverse(
-; CHECK-NEXT:    [[ADD1:%.*]] = add <vscale x 4 x i32> [[A:%.*]], [[B:%.*]]
-; CHECK-NEXT:    ret <vscale x 4 x i32> [[ADD1]]
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[ADD:%.*]] = add <vscale x 4 x i32> [[A_REV]], [[B_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[ADD]]
+;
+  %a.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %add = add <vscale x 4 x i32> %a.rev, %b.rev
+  ret <vscale x 4 x i32> %add
+}
+
+; %a.rev has multiple uses
+define <vscale x 4 x i32> @binop_reverse_1(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
+; CHECK-LABEL: @binop_reverse_1(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[A_REV]])
+; CHECK-NEXT:    [[ADD:%.*]] = add <vscale x 4 x i32> [[A_REV]], [[B_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[ADD]]
 ;
-  %reva = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
-  %revb = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
-  %add = add <vscale x 4 x i32> %reva, %revb
-  %revadd = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %add)
-  ret <vscale x 4 x i32> %revadd
+  %a.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  call void @use_nxv4i32(<vscale x 4 x i32>  %a.rev)
+  %add = add <vscale x 4 x i32> %a.rev, %b.rev
+  ret <vscale x 4 x i32> %add
+}
+
+; %b.rev has multiple uses
+define <vscale x 4 x i32> @binop_reverse_2(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
+; CHECK-LABEL: @binop_reverse_2(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[B_REV]])
+; CHECK-NEXT:    [[ADD:%.*]] = add <vscale x 4 x i32> [[A_REV]], [[B_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[ADD]]
+;
+  %a.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  call void @use_nxv4i32(<vscale x 4 x i32>  %b.rev)
+  %add = add <vscale x 4 x i32> %a.rev, %b.rev
+  ret <vscale x 4 x i32> %add
+}
+
+; %a.rev and %b.rev have multiple uses
+define <vscale x 4 x i32> @binop_reverse_3(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
+; CHECK-LABEL: @binop_reverse_3(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[A_REV]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[B_REV]])
+; CHECK-NEXT:    [[ADD:%.*]] = add <vscale x 4 x i32> [[A_REV]], [[B_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[ADD]]
+;
+  %a.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  call void @use_nxv4i32(<vscale x 4 x i32> %a.rev)
+  call void @use_nxv4i32(<vscale x 4 x i32> %b.rev)
+  %add = add <vscale x 4 x i32> %a.rev, %b.rev
+  ret <vscale x 4 x i32> %add
+}
+
+; %a.rev used as both operands
+define <vscale x 4 x i32> @binop_reverse_4(<vscale x 4 x i32> %a) {
+; CHECK-LABEL: @binop_reverse_4(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[A:%.*]])
+; CHECK-NEXT:    [[MUL:%.*]] = mul <vscale x 4 x i32> [[A_REV]], [[A_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[MUL]]
+;
+  %a.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
+  %mul = mul <vscale x 4 x i32> %a.rev, %a.rev
+  ret <vscale x 4 x i32> %mul
 }
 
 define <vscale x 4 x i32> @binop_reverse_splat_RHS(<vscale x 4 x i32> %a, i32 %b) {
 ; CHECK-LABEL: @binop_reverse_splat_RHS(
-; CHECK-NEXT:    [[SPLAT_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[B:%.*]], i64 0
-; CHECK-NEXT:    [[SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[SPLAT_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
-; CHECK-NEXT:    [[UDIV1:%.*]] = udiv <vscale x 4 x i32> [[A:%.*]], [[SPLAT]]
-; CHECK-NEXT:    ret <vscale x 4 x i32> [[UDIV1]]
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[A:%.*]])
+; CHECK-NEXT:    [[B_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[B:%.*]], i64 0
+; CHECK-NEXT:    [[B_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[B_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    [[DIV:%.*]] = udiv <vscale x 4 x i32> [[A_REV]], [[B_SPLAT]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[DIV]]
+;
+  %a.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
+  %b.insert = insertelement <vscale x 4 x i32> poison, i32 %b, i32 0
+  %b.splat = shufflevector <vscale x 4 x i32> %b.insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+  %div = udiv <vscale x 4 x i32> %a.rev, %b.splat
+  ret <vscale x 4 x i32> %div
+}
+
+; %a.rev has multiple uses
+define <vscale x 4 x i32> @binop_reverse_splat_RHS_1(<vscale x 4 x i32> %a, i32 %b) {
+; CHECK-LABEL: @binop_reverse_splat_RHS_1(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[A:%.*]])
+; CHECK-NEXT:    [[B_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[B:%.*]], i64 0
+; CHECK-NEXT:    [[B_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[B_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[A_REV]])
+; CHECK-NEXT:    [[DIV:%.*]] = udiv <vscale x 4 x i32> [[A_REV]], [[B_SPLAT]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[DIV]]
 ;
-  %reva = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
-  %splat_insert = insertelement <vscale x 4 x i32> poison, i32 %b, i32 0
-  %splat = shufflevector <vscale x 4 x i32> %splat_insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
-  %udiv = udiv <vscale x 4 x i32> %reva, %splat
-  %revadd = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %udiv)
-  ret <vscale x 4 x i32> %revadd
+  %a.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
+  %b.insert = insertelement <vscale x 4 x i32> poison, i32 %b, i32 0
+  %b.splat = shufflevector <vscale x 4 x i32> %b.insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+  call void @use_nxv4i32(<vscale x 4 x i32> %a.rev)
+  %div = udiv <vscale x 4 x i32> %a.rev, %b.splat
+  ret <vscale x 4 x i32> %div
 }
 
 define <vscale x 4 x i32> @binop_reverse_splat_LHS(<vscale x 4 x i32> %a, i32 %b) {
 ; CHECK-LABEL: @binop_reverse_splat_LHS(
-; CHECK-NEXT:    [[SPLAT_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[B:%.*]], i64 0
-; CHECK-NEXT:    [[SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[SPLAT_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
-; CHECK-NEXT:    [[UDIV1:%.*]] = udiv <vscale x 4 x i32> [[SPLAT]], [[A:%.*]]
-; CHECK-NEXT:    ret <vscale x 4 x i32> [[UDIV1]]
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[A:%.*]])
+; CHECK-NEXT:    [[B_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[B:%.*]], i64 0
+; CHECK-NEXT:    [[B_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[B_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    [[DIV:%.*]] = udiv <vscale x 4 x i32> [[B_SPLAT]], [[A_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[DIV]]
+;
+  %a.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
+  %b.insert = insertelement <vscale x 4 x i32> poison, i32 %b, i32 0
+  %b.splat = shufflevector <vscale x 4 x i32> %b.insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+  %div = udiv <vscale x 4 x i32> %b.splat, %a.rev
+  ret <vscale x 4 x i32> %div
+}
+
+; %a.rev has multiple uses
+define <vscale x 4 x i32> @binop_reverse_splat_LHS_1(<vscale x 4 x i32> %a, i32 %b) {
+; CHECK-LABEL: @binop_reverse_splat_LHS_1(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[A:%.*]])
+; CHECK-NEXT:    [[B_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[B:%.*]], i64 0
+; CHECK-NEXT:    [[B_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[B_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[A_REV]])
+; CHECK-NEXT:    [[DIV:%.*]] = udiv <vscale x 4 x i32> [[B_SPLAT]], [[A_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[DIV]]
 ;
-  %reva = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
-  %splat_insert = insertelement <vscale x 4 x i32> poison, i32 %b, i32 0
-  %splat = shufflevector <vscale x 4 x i32> %splat_insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
-  %udiv = udiv <vscale x 4 x i32> %splat, %reva
-  %revadd = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %udiv)
-  ret <vscale x 4 x i32> %revadd
+  %a.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
+  %b.insert = insertelement <vscale x 4 x i32> poison, i32 %b, i32 0
+  %b.splat = shufflevector <vscale x 4 x i32> %b.insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+  call void @use_nxv4i32(<vscale x 4 x i32> %a.rev)
+  %div = udiv <vscale x 4 x i32> %b.splat, %a.rev
+  ret <vscale x 4 x i32> %div
 }
 
 define <vscale x 4 x float> @unop_reverse(<vscale x 4 x float> %a) {
 ; CHECK-LABEL: @unop_reverse(
-; CHECK-NEXT:    [[NEG1:%.*]] = fneg fast <vscale x 4 x float> [[A:%.*]]
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> [[A:%.*]])
+; CHECK-NEXT:    [[NEG:%.*]] = fneg fast <vscale x 4 x float> [[A_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x float> [[NEG]]
+;
+  %a.rev = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %a)
+  %neg = fneg fast <vscale x 4 x float> %a.rev
+  ret <vscale x 4 x float> %neg
+}
+
+; %a.rev has multiple uses
+define <vscale x 4 x float> @unop_reverse_1(<vscale x 4 x float> %a) {
+; CHECK-LABEL: @unop_reverse_1(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> [[A:%.*]])
+; CHECK-NEXT:    call void @use_nxv4f32(<vscale x 4 x float> [[A_REV]])
+; CHECK-NEXT:    [[NEG:%.*]] = fneg fast <vscale x 4 x float> [[A_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x float> [[NEG]]
+;
+  %a.rev = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %a)
+  call void @use_nxv4f32(<vscale x 4 x float> %a.rev)
+  %neg = fneg fast <vscale x 4 x float> %a.rev
+  ret <vscale x 4 x float> %neg
+}
+
+define <vscale x 4 x i1> @icmp_reverse(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
+; CHECK-LABEL: @icmp_reverse(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq <vscale x 4 x i32> [[A_REV]], [[B_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i1> [[CMP]]
+;
+  %a.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %cmp = icmp eq <vscale x 4 x i32> %a.rev, %b.rev
+  ret <vscale x 4 x i1> %cmp
+}
+
+; %a.rev has multiple uses
+define <vscale x 4 x i1> @icmp_reverse_1(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
+; CHECK-LABEL: @icmp_reverse_1(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[A_REV]])
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq <vscale x 4 x i32> [[A_REV]], [[B_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i1> [[CMP]]
+;
+  %a.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  call void @use_nxv4i32(<vscale x 4 x i32> %a.rev)
+  %cmp = icmp eq <vscale x 4 x i32> %a.rev, %b.rev
+  ret <vscale x 4 x i1> %cmp
+}
+
+; %b.rev has multiple uses
+define <vscale x 4 x i1> @icmp_reverse_2(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
+; CHECK-LABEL: @icmp_reverse_2(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[B_REV]])
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq <vscale x 4 x i32> [[A_REV]], [[B_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i1> [[CMP]]
+;
+  %a.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  call void @use_nxv4i32(<vscale x 4 x i32> %b.rev)
+  %cmp = icmp eq <vscale x 4 x i32> %a.rev, %b.rev
+  ret <vscale x 4 x i1> %cmp
+}
+
+; %a.rev and %b.rev have multiple uses
+define <vscale x 4 x i1> @icmp_reverse_3(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
+; CHECK-LABEL: @icmp_reverse_3(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[A_REV]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[B_REV]])
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq <vscale x 4 x i32> [[A_REV]], [[B_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i1> [[CMP]]
+;
+  %a.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  call void @use_nxv4i32(<vscale x 4 x i32> %a.rev)
+  call void @use_nxv4i32(<vscale x 4 x i32> %b.rev)
+  %cmp = icmp eq <vscale x 4 x i32> %a.rev, %b.rev
+  ret <vscale x 4 x i1> %cmp
+}
+
+define <vscale x 4 x i1> @icmp_reverse_splat_RHS(<vscale x 4 x i32> %a, i32 %b) {
+; CHECK-LABEL: @icmp_reverse_splat_RHS(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[A:%.*]])
+; CHECK-NEXT:    [[B_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[B:%.*]], i64 0
+; CHECK-NEXT:    [[B_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[B_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    [[CMP:%.*]] = icmp sgt <vscale x 4 x i32> [[A_REV]], [[B_SPLAT]]
+; CHECK-NEXT:    ret <vscale x 4 x i1> [[CMP]]
+;
+  %a.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
+  %b.insert = insertelement <vscale x 4 x i32> poison, i32 %b, i32 0
+  %b.splat = shufflevector <vscale x 4 x i32> %b.insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+  %cmp = icmp sgt <vscale x 4 x i32> %a.rev, %b.splat
+  ret <vscale x 4 x i1> %cmp
+}
+
+; %a.rev has multiple uses
+define <vscale x 4 x i1> @icmp_reverse_splat_RHS_1(<vscale x 4 x i32> %a, i32 %b) {
+; CHECK-LABEL: @icmp_reverse_splat_RHS_1(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[A:%.*]])
+; CHECK-NEXT:    [[B_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[B:%.*]], i64 0
+; CHECK-NEXT:    [[B_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[B_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[A_REV]])
+; CHECK-NEXT:    [[CMP:%.*]] = icmp sgt <vscale x 4 x i32> [[A_REV]], [[B_SPLAT]]
+; CHECK-NEXT:    ret <vscale x 4 x i1> [[CMP]]
+;
+  %a.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
+  %b.insert = insertelement <vscale x 4 x i32> poison, i32 %b, i32 0
+  %b.splat = shufflevector <vscale x 4 x i32> %b.insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+  call void @use_nxv4i32(<vscale x 4 x i32> %a.rev)
+  %cmp = icmp sgt <vscale x 4 x i32> %a.rev, %b.splat
+  ret <vscale x 4 x i1> %cmp
+}
+
+define <vscale x 4 x i1> @icmp_reverse_splat_LHS(<vscale x 4 x i32> %a, i32 %b) {
+; CHECK-LABEL: @icmp_reverse_splat_LHS(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[A:%.*]])
+; CHECK-NEXT:    [[B_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[B:%.*]], i64 0
+; CHECK-NEXT:    [[B_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[B_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ult <vscale x 4 x i32> [[B_SPLAT]], [[A_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i1> [[CMP]]
+;
+  %a.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
+  %b.insert = insertelement <vscale x 4 x i32> poison, i32 %b, i32 0
+  %b.splat = shufflevector <vscale x 4 x i32> %b.insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+  %cmp = icmp ult <vscale x 4 x i32> %b.splat, %a.rev
+  ret <vscale x 4 x i1> %cmp
+}
+
+; %a.rev has multiple uses
+define <vscale x 4 x i1> @icmp_reverse_splat_LHS_1(<vscale x 4 x i32> %a, i32 %b) {
+; CHECK-LABEL: @icmp_reverse_splat_LHS_1(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[A:%.*]])
+; CHECK-NEXT:    [[B_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[B:%.*]], i64 0
+; CHECK-NEXT:    [[B_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[B_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[A_REV]])
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ult <vscale x 4 x i32> [[B_SPLAT]], [[A_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i1> [[CMP]]
+;
+  %a.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %a)
+  %b.insert = insertelement <vscale x 4 x i32> poison, i32 %b, i32 0
+  %b.splat = shufflevector <vscale x 4 x i32> %b.insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+  call void @use_nxv4i32(<vscale x 4 x i32> %a.rev)
+  %cmp = icmp ult <vscale x 4 x i32> %b.splat, %a.rev
+  ret <vscale x 4 x i1> %cmp
+}
+
+define <vscale x 4 x i32> @select_reverse(<vscale x 4 x i1> %a, <vscale x 4 x i32> %b, <vscale x 4 x i32> %c) {
+; CHECK-LABEL: @select_reverse(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[C_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[C:%.*]])
+; CHECK-NEXT:    [[SELECT:%.*]] = select <vscale x 4 x i1> [[A_REV]], <vscale x 4 x i32> [[B_REV]], <vscale x 4 x i32> [[C_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[SELECT]]
+;
+  %a.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %c.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %c)
+  %select = select <vscale x 4 x i1> %a.rev, <vscale x 4 x i32> %b.rev, <vscale x 4 x i32> %c.rev
+  ret <vscale x 4 x i32> %select
+}
+
+; %a.rev has multiple uses
+define <vscale x 4 x i32> @select_reverse_1(<vscale x 4 x i1> %a, <vscale x 4 x i32> %b, <vscale x 4 x i32> %c) {
+; CHECK-LABEL: @select_reverse_1(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[C_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[C:%.*]])
+; CHECK-NEXT:    call void @use_nxv4i1(<vscale x 4 x i1> [[A_REV]])
+; CHECK-NEXT:    [[SELECT:%.*]] = select <vscale x 4 x i1> [[A_REV]], <vscale x 4 x i32> [[B_REV]], <vscale x 4 x i32> [[C_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[SELECT]]
+;
+  %a.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %c.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %c)
+  call void @use_nxv4i1(<vscale x 4 x i1> %a.rev)
+  %select = select <vscale x 4 x i1> %a.rev, <vscale x 4 x i32> %b.rev, <vscale x 4 x i32> %c.rev
+  ret <vscale x 4 x i32> %select
+}
+
+; %b.rev has multiple uses
+define <vscale x 4 x i32> @select_reverse_2(<vscale x 4 x i1> %a, <vscale x 4 x i32> %b, <vscale x 4 x i32> %c) {
+; CHECK-LABEL: @select_reverse_2(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[C_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[C:%.*]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[B_REV]])
+; CHECK-NEXT:    [[SELECT:%.*]] = select <vscale x 4 x i1> [[A_REV]], <vscale x 4 x i32> [[B_REV]], <vscale x 4 x i32> [[C_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[SELECT]]
+;
+  %a.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %c.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %c)
+  call void @use_nxv4i32(<vscale x 4 x i32> %b.rev)
+  %select = select <vscale x 4 x i1> %a.rev, <vscale x 4 x i32> %b.rev, <vscale x 4 x i32> %c.rev
+  ret <vscale x 4 x i32> %select
+}
+
+; %c.rev has multiple uses
+define <vscale x 4 x i32> @select_reverse_3(<vscale x 4 x i1> %a, <vscale x 4 x i32> %b, <vscale x 4 x i32> %c) {
+; CHECK-LABEL: @select_reverse_3(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[C_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[C:%.*]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[C_REV]])
+; CHECK-NEXT:    [[SELECT:%.*]] = select <vscale x 4 x i1> [[A_REV]], <vscale x 4 x i32> [[B_REV]], <vscale x 4 x i32> [[C_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[SELECT]]
+;
+  %a.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %c.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %c)
+  call void @use_nxv4i32(<vscale x 4 x i32> %c.rev)
+  %select = select <vscale x 4 x i1> %a.rev, <vscale x 4 x i32> %b.rev, <vscale x 4 x i32> %c.rev
+  ret <vscale x 4 x i32> %select
+}
+
+; %a.rev and %b.rev have multiple uses
+define <vscale x 4 x i32> @select_reverse_4(<vscale x 4 x i1> %a, <vscale x 4 x i32> %b, <vscale x 4 x i32> %c) {
+; CHECK-LABEL: @select_reverse_4(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[C_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[C:%.*]])
+; CHECK-NEXT:    call void @use_nxv4i1(<vscale x 4 x i1> [[A_REV]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[B_REV]])
+; CHECK-NEXT:    [[SELECT:%.*]] = select <vscale x 4 x i1> [[A_REV]], <vscale x 4 x i32> [[B_REV]], <vscale x 4 x i32> [[C_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[SELECT]]
+;
+  %a.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %c.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %c)
+  call void @use_nxv4i1(<vscale x 4 x i1> %a.rev)
+  call void @use_nxv4i32(<vscale x 4 x i32> %b.rev)
+  %select = select <vscale x 4 x i1> %a.rev, <vscale x 4 x i32> %b.rev, <vscale x 4 x i32> %c.rev
+  ret <vscale x 4 x i32> %select
+}
+
+; %a.rev and %c.rev have multiple uses
+define <vscale x 4 x i32> @select_reverse_5(<vscale x 4 x i1> %a, <vscale x 4 x i32> %b, <vscale x 4 x i32> %c) {
+; CHECK-LABEL: @select_reverse_5(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[C_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[C:%.*]])
+; CHECK-NEXT:    call void @use_nxv4i1(<vscale x 4 x i1> [[A_REV]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[C_REV]])
+; CHECK-NEXT:    [[SELECT:%.*]] = select <vscale x 4 x i1> [[A_REV]], <vscale x 4 x i32> [[B_REV]], <vscale x 4 x i32> [[C_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[SELECT]]
+;
+  %a.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %c.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %c)
+  call void @use_nxv4i1(<vscale x 4 x i1> %a.rev)
+  call void @use_nxv4i32(<vscale x 4 x i32> %c.rev)
+  %select = select <vscale x 4 x i1> %a.rev, <vscale x 4 x i32> %b.rev, <vscale x 4 x i32> %c.rev
+  ret <vscale x 4 x i32> %select
+}
+
+; %b.rev and %c.rev have multiple uses
+define <vscale x 4 x i32> @select_reverse_6(<vscale x 4 x i1> %a, <vscale x 4 x i32> %b, <vscale x 4 x i32> %c) {
+; CHECK-LABEL: @select_reverse_6(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[C_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[C:%.*]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[B_REV]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[C_REV]])
+; CHECK-NEXT:    [[SELECT:%.*]] = select <vscale x 4 x i1> [[A_REV]], <vscale x 4 x i32> [[B_REV]], <vscale x 4 x i32> [[C_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[SELECT]]
+;
+  %a.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %c.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %c)
+  call void @use_nxv4i32(<vscale x 4 x i32> %b.rev)
+  call void @use_nxv4i32(<vscale x 4 x i32> %c.rev)
+  %select = select <vscale x 4 x i1> %a.rev, <vscale x 4 x i32> %b.rev, <vscale x 4 x i32> %c.rev
+  ret <vscale x 4 x i32> %select
+}
+
+; %a.rev, %b.rev and %c.rev have multiple uses
+define <vscale x 4 x i32> @select_reverse_7(<vscale x 4 x i1> %a, <vscale x 4 x i32> %b, <vscale x 4 x i32> %c) {
+; CHECK-LABEL: @select_reverse_7(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[C_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[C:%.*]])
+; CHECK-NEXT:    call void @use_nxv4i1(<vscale x 4 x i1> [[A_REV]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[B_REV]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[C_REV]])
+; CHECK-NEXT:    [[SELECT:%.*]] = select <vscale x 4 x i1> [[A_REV]], <vscale x 4 x i32> [[B_REV]], <vscale x 4 x i32> [[C_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[SELECT]]
+;
+  %a.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %c.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %c)
+  call void @use_nxv4i1(<vscale x 4 x i1> %a.rev)
+  call void @use_nxv4i32(<vscale x 4 x i32> %b.rev)
+  call void @use_nxv4i32(<vscale x 4 x i32> %c.rev)
+  %select = select <vscale x 4 x i1> %a.rev, <vscale x 4 x i32> %b.rev, <vscale x 4 x i32> %c.rev
+  ret <vscale x 4 x i32> %select
+}
+
+define <vscale x 4 x i32> @select_reverse_splat_false(<vscale x 4 x i1> %a, <vscale x 4 x i32> %b, i32 %c) {
+; CHECK-LABEL: @select_reverse_splat_false(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[C_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[C:%.*]], i64 0
+; CHECK-NEXT:    [[C_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[C_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    [[SELECT:%.*]] = select <vscale x 4 x i1> [[A_REV]], <vscale x 4 x i32> [[B_REV]], <vscale x 4 x i32> [[C_SPLAT]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[SELECT]]
+;
+  %a.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %c.insert = insertelement <vscale x 4 x i32> poison, i32 %c, i32 0
+  %c.splat = shufflevector <vscale x 4 x i32> %c.insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+  %select = select <vscale x 4 x i1> %a.rev, <vscale x 4 x i32> %b.rev, <vscale x 4 x i32> %c.splat
+  ret <vscale x 4 x i32> %select
+}
+
+; %a.rev has multiple uses
+define <vscale x 4 x i32> @select_reverse_splat_false_1(<vscale x 4 x i1> %a, <vscale x 4 x i32> %b, i32 %c) {
+; CHECK-LABEL: @select_reverse_splat_false_1(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[C_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[C:%.*]], i64 0
+; CHECK-NEXT:    [[C_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[C_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    call void @use_nxv4i1(<vscale x 4 x i1> [[A_REV]])
+; CHECK-NEXT:    [[SELECT:%.*]] = select <vscale x 4 x i1> [[A_REV]], <vscale x 4 x i32> [[B_REV]], <vscale x 4 x i32> [[C_SPLAT]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[SELECT]]
+;
+  %a.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %c.insert = insertelement <vscale x 4 x i32> poison, i32 %c, i32 0
+  %c.splat = shufflevector <vscale x 4 x i32> %c.insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+  call void @use_nxv4i1(<vscale x 4 x i1> %a.rev)
+  %select = select <vscale x 4 x i1> %a.rev, <vscale x 4 x i32> %b.rev, <vscale x 4 x i32> %c.splat
+  ret <vscale x 4 x i32> %select
+}
+
+; %b.rev has multiple uses
+define <vscale x 4 x i32> @select_reverse_splat_false_2(<vscale x 4 x i1> %a, <vscale x 4 x i32> %b, i32 %c) {
+; CHECK-LABEL: @select_reverse_splat_false_2(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[C_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[C:%.*]], i64 0
+; CHECK-NEXT:    [[C_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[C_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[B_REV]])
+; CHECK-NEXT:    [[SELECT:%.*]] = select <vscale x 4 x i1> [[A_REV]], <vscale x 4 x i32> [[B_REV]], <vscale x 4 x i32> [[C_SPLAT]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[SELECT]]
+;
+  %a.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %c.insert = insertelement <vscale x 4 x i32> poison, i32 %c, i32 0
+  %c.splat = shufflevector <vscale x 4 x i32> %c.insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+  call void @use_nxv4i32(<vscale x 4 x i32> %b.rev)
+  %select = select <vscale x 4 x i1> %a.rev, <vscale x 4 x i32> %b.rev, <vscale x 4 x i32> %c.splat
+  ret <vscale x 4 x i32> %select
+}
+
+; %a.rev and %b.rev have multiple uses
+define <vscale x 4 x i32> @select_reverse_splat_false_3(<vscale x 4 x i1> %a, <vscale x 4 x i32> %b, i32 %c) {
+; CHECK-LABEL: @select_reverse_splat_false_3(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[C_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[C:%.*]], i64 0
+; CHECK-NEXT:    [[C_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[C_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    call void @use_nxv4i1(<vscale x 4 x i1> [[A_REV]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[B_REV]])
+; CHECK-NEXT:    [[SELECT:%.*]] = select <vscale x 4 x i1> [[A_REV]], <vscale x 4 x i32> [[B_REV]], <vscale x 4 x i32> [[C_SPLAT]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[SELECT]]
+;
+  %a.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %c.insert = insertelement <vscale x 4 x i32> poison, i32 %c, i32 0
+  %c.splat = shufflevector <vscale x 4 x i32> %c.insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+  call void @use_nxv4i1(<vscale x 4 x i1> %a.rev)
+  call void @use_nxv4i32(<vscale x 4 x i32> %b.rev)
+  %select = select <vscale x 4 x i1> %a.rev, <vscale x 4 x i32> %b.rev, <vscale x 4 x i32> %c.splat
+  ret <vscale x 4 x i32> %select
+}
+
+define <vscale x 4 x i32> @select_reverse_splat_true(<vscale x 4 x i1> %a, <vscale x 4 x i32> %b, i32 %c) {
+; CHECK-LABEL: @select_reverse_splat_true(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[C_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[C:%.*]], i64 0
+; CHECK-NEXT:    [[C_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[C_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    [[SELECT:%.*]] = select <vscale x 4 x i1> [[A_REV]], <vscale x 4 x i32> [[C_SPLAT]], <vscale x 4 x i32> [[B_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[SELECT]]
+;
+  %a.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %c.insert = insertelement <vscale x 4 x i32> poison, i32 %c, i32 0
+  %c.splat = shufflevector <vscale x 4 x i32> %c.insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+  %select = select <vscale x 4 x i1> %a.rev, <vscale x 4 x i32> %c.splat, <vscale x 4 x i32> %b.rev
+  ret <vscale x 4 x i32> %select
+}
+
+; %a.rev has multiple uses
+define <vscale x 4 x i32> @select_reverse_splat_true_1(<vscale x 4 x i1> %a, <vscale x 4 x i32> %b, i32 %c) {
+; CHECK-LABEL: @select_reverse_splat_true_1(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[C_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[C:%.*]], i64 0
+; CHECK-NEXT:    [[C_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[C_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    call void @use_nxv4i1(<vscale x 4 x i1> [[A_REV]])
+; CHECK-NEXT:    [[SELECT:%.*]] = select <vscale x 4 x i1> [[A_REV]], <vscale x 4 x i32> [[C_SPLAT]], <vscale x 4 x i32> [[B_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[SELECT]]
+;
+  %a.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %c.insert = insertelement <vscale x 4 x i32> poison, i32 %c, i32 0
+  %c.splat = shufflevector <vscale x 4 x i32> %c.insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+  call void @use_nxv4i1(<vscale x 4 x i1> %a.rev)
+  %select = select <vscale x 4 x i1> %a.rev, <vscale x 4 x i32> %c.splat, <vscale x 4 x i32> %b.rev
+  ret <vscale x 4 x i32> %select
+}
+
+; %b.rev has multiple uses
+define <vscale x 4 x i32> @select_reverse_splat_true_2(<vscale x 4 x i1> %a, <vscale x 4 x i32> %b, i32 %c) {
+; CHECK-LABEL: @select_reverse_splat_true_2(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[C_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[C:%.*]], i64 0
+; CHECK-NEXT:    [[C_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[C_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[B_REV]])
+; CHECK-NEXT:    [[SELECT:%.*]] = select <vscale x 4 x i1> [[A_REV]], <vscale x 4 x i32> [[C_SPLAT]], <vscale x 4 x i32> [[B_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[SELECT]]
+;
+  %a.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %c.insert = insertelement <vscale x 4 x i32> poison, i32 %c, i32 0
+  %c.splat = shufflevector <vscale x 4 x i32> %c.insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+  call void @use_nxv4i32(<vscale x 4 x i32> %b.rev)
+  %select = select <vscale x 4 x i1> %a.rev, <vscale x 4 x i32> %c.splat, <vscale x 4 x i32> %b.rev
+  ret <vscale x 4 x i32> %select
+}
+
+; %a.rev and %b.rev have multiple uses
+define <vscale x 4 x i32> @select_reverse_splat_true_3(<vscale x 4 x i1> %a, <vscale x 4 x i32> %b, i32 %c) {
+; CHECK-LABEL: @select_reverse_splat_true_3(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[C_INSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[C:%.*]], i64 0
+; CHECK-NEXT:    [[C_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[C_INSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    call void @use_nxv4i1(<vscale x 4 x i1> [[A_REV]])
+; CHECK-NEXT:    call void @use_nxv4i32(<vscale x 4 x i32> [[B_REV]])
+; CHECK-NEXT:    [[SELECT:%.*]] = select <vscale x 4 x i1> [[A_REV]], <vscale x 4 x i32> [[C_SPLAT]], <vscale x 4 x i32> [[B_REV]]
+; CHECK-NEXT:    ret <vscale x 4 x i32> [[SELECT]]
+;
+  %a.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %a)
+  %b.rev = tail call <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32> %b)
+  %c.insert = insertelement <vscale x 4 x i32> poison, i32 %c, i32 0
+  %c.splat = shufflevector <vscale x 4 x i32> %c.insert, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+  call void @use_nxv4i1(<vscale x 4 x i1> %a.rev)
+  call void @use_nxv4i32(<vscale x 4 x i32> %b.rev)
+  %select = select <vscale x 4 x i1> %a.rev, <vscale x 4 x i32> %c.splat, <vscale x 4 x i32> %b.rev
+  ret <vscale x 4 x i32> %select
+}
+
+; Tests to ensure no reversals exist when all operands are reversed and the
+; result is also reversed.
+
+define <vscale x 4 x float> @reverse_binop_reverse(<vscale x 4 x float> %a, <vscale x 4 x float> %b) {
+; CHECK-LABEL: @reverse_binop_reverse(
+; CHECK-NEXT:    [[ADD1:%.*]] = fadd <vscale x 4 x float> [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT:    ret <vscale x 4 x float> [[ADD1]]
+;
+  %a.rev = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %a)
+  %b.rev = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %b)
+  %add = fadd <vscale x 4 x float> %a.rev, %b.rev
+  %add.rev = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %add)
+  ret <vscale x 4 x float> %add.rev
+}
+
+define <vscale x 4 x float> @reverse_binop_reverse_splat_RHS(<vscale x 4 x float> %a, float %b) {
+; CHECK-LABEL: @reverse_binop_reverse_splat_RHS(
+; CHECK-NEXT:    [[B_INSERT:%.*]] = insertelement <vscale x 4 x float> poison, float [[B:%.*]], i64 0
+; CHECK-NEXT:    [[B_SPLAT:%.*]] = shufflevector <vscale x 4 x float> [[B_INSERT]], <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    [[DIV1:%.*]] = fdiv <vscale x 4 x float> [[A:%.*]], [[B_SPLAT]]
+; CHECK-NEXT:    ret <vscale x 4 x float> [[DIV1]]
+;
+  %a.rev = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %a)
+  %b.insert = insertelement <vscale x 4 x float> poison, float %b, i32 0
+  %b.splat = shufflevector <vscale x 4 x float> %b.insert, <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer
+  %div = fdiv <vscale x 4 x float> %a.rev, %b.splat
+  %div.rev = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %div)
+  ret <vscale x 4 x float> %div.rev
+}
+
+define <vscale x 4 x float> @reverse_binop_reverse_splat_LHS(<vscale x 4 x float> %a, float %b) {
+; CHECK-LABEL: @reverse_binop_reverse_splat_LHS(
+; CHECK-NEXT:    [[B_INSERT:%.*]] = insertelement <vscale x 4 x float> poison, float [[B:%.*]], i64 0
+; CHECK-NEXT:    [[B_SPLAT:%.*]] = shufflevector <vscale x 4 x float> [[B_INSERT]], <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    [[DIV1:%.*]] = fdiv <vscale x 4 x float> [[B_SPLAT]], [[A:%.*]]
+; CHECK-NEXT:    ret <vscale x 4 x float> [[DIV1]]
+;
+  %a.rev = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %a)
+  %b.insert = insertelement <vscale x 4 x float> poison, float %b, i32 0
+  %b.splat = shufflevector <vscale x 4 x float> %b.insert, <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer
+  %div = fdiv <vscale x 4 x float> %b.splat, %a.rev
+  %div.rev = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %div)
+  ret <vscale x 4 x float> %div.rev
+}
+
+define <vscale x 4 x i1> @reverse_fcmp_reverse(<vscale x 4 x float> %a, <vscale x 4 x float> %b) {
+; CHECK-LABEL: @reverse_fcmp_reverse(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> [[B:%.*]])
+; CHECK-NEXT:    [[CMP:%.*]] = fcmp olt <vscale x 4 x float> [[A_REV]], [[B_REV]]
+; CHECK-NEXT:    [[CMP_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[CMP]])
+; CHECK-NEXT:    ret <vscale x 4 x i1> [[CMP_REV]]
+;
+  %a.rev = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %a)
+  %b.rev = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %b)
+  %cmp = fcmp olt <vscale x 4 x float> %a.rev, %b.rev
+  %cmp.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %cmp)
+  ret <vscale x 4 x i1> %cmp.rev
+}
+
+define <vscale x 4 x float> @reverse_select_reverse(<vscale x 4 x i1> %a, <vscale x 4 x float> %b, <vscale x 4 x float> %c) {
+; CHECK-LABEL: @reverse_select_reverse(
+; CHECK-NEXT:    [[A_REV:%.*]] = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> [[A:%.*]])
+; CHECK-NEXT:    [[B_REV:%.*]] = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> [[B:%.*]])
+; CHECK-NEXT:    [[C_REV:%.*]] = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> [[C:%.*]])
+; CHECK-NEXT:    [[SELECT:%.*]] = select <vscale x 4 x i1> [[A_REV]], <vscale x 4 x float> [[B_REV]], <vscale x 4 x float> [[C_REV]]
+; CHECK-NEXT:    [[SELECT_REV:%.*]] = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> [[SELECT]])
+; CHECK-NEXT:    ret <vscale x 4 x float> [[SELECT_REV]]
+;
+  %a.rev = tail call <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1> %a)
+  %b.rev = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %b)
+  %c.rev = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %c)
+  %select = select <vscale x 4 x i1> %a.rev, <vscale x 4 x float> %b.rev, <vscale x 4 x float> %c.rev
+  %select.rev = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %select)
+  ret <vscale x 4 x float> %select.rev
+}
+
+define <vscale x 4 x float> @reverse_unop_reverse(<vscale x 4 x float> %a) {
+; CHECK-LABEL: @reverse_unop_reverse(
+; CHECK-NEXT:    [[NEG1:%.*]] = fneg <vscale x 4 x float> [[A:%.*]]
 ; CHECK-NEXT:    ret <vscale x 4 x float> [[NEG1]]
 ;
-  %reva = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %a)
-  %neg = fneg fast <vscale x 4 x float> %reva
-  %revneg = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %neg)
-  ret <vscale x 4 x float> %revneg
+  %a.rev = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %a)
+  %neg = fneg <vscale x 4 x float> %a.rev
+  %neg.rev = tail call <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float> %neg)
+  ret <vscale x 4 x float> %neg.rev
 }
 
-declare <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float>)
-declare <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32>)
 
+declare void @use_nxv4i1(<vscale x 4 x i1>)
+declare void @use_nxv4i32(<vscale x 4 x i32>)
+declare void @use_nxv4f32(<vscale x 4 x float>)
 
+declare <vscale x 4 x i1> @llvm.experimental.vector.reverse.nxv4i1(<vscale x 4 x i1>)
+declare <vscale x 4 x i32> @llvm.experimental.vector.reverse.nxv4i32(<vscale x 4 x i32>)
+declare <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float>)