[InstCombine] Unfold masked merge with constant mask

Summary:
As discussed in D45733, we want to do this in InstCombine.

https://rise4fun.com/Alive/LGk

Reviewers: spatel, craig.topper

Reviewed By: spatel

Subscribers: chandlerc, xbolva00, llvm-commits

Differential Revision: https://reviews.llvm.org/D45867

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@331205 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Roman Lebedev 2018-04-30 17:59:33 +00:00
parent 7a3b1c9910
commit fa71b55bb9
3 changed files with 80 additions and 70 deletions

View File

@ -2412,14 +2412,19 @@ Value *InstCombiner::foldXorOfICmps(ICmpInst *LHS, ICmpInst *RHS) {
}
/// If we have a masked merge, in the canonical form of:
/// (assuming that A only has one use.)
/// | A | |B|
/// ((x ^ y) & M) ^ y
/// | D |
/// * If M is inverted:
/// | D |
/// ((x ^ y) & ~M) ^ y
/// If A has one use, and, we want to canonicalize it to non-inverted mask:
/// We can canonicalize by swapping the final xor operand
/// to eliminate the 'not' of the mask.
/// ((x ^ y) & M) ^ x
/// * If M is a constant, and D has one use, we transform to 'and' / 'or' ops
/// because that shortens the dependency chain and improves analysis:
/// (x & M) | (y & ~M)
static Instruction *visitMaskedMerge(BinaryOperator &I,
InstCombiner::BuilderTy &Builder) {
Value *B, *X, *D;
@ -2438,6 +2443,15 @@ static Instruction *visitMaskedMerge(BinaryOperator &I,
return BinaryOperator::CreateXor(NewA, X);
}
Constant *C;
if (D->hasOneUse() && match(M, m_Constant(C))) {
// Unfold.
Value *LHS = Builder.CreateAnd(X, C);
Value *NotC = Builder.CreateNot(C);
Value *RHS = Builder.CreateAnd(B, NotC);
return BinaryOperator::CreateOr(LHS, RHS);
}
return nullptr;
}

View File

@ -8,9 +8,9 @@
define i4 @scalar0 (i4 %x, i4 %y) {
; CHECK-LABEL: @scalar0(
; CHECK-NEXT: [[N0:%.*]] = xor i4 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: [[N1:%.*]] = and i4 [[N0]], 1
; CHECK-NEXT: [[R:%.*]] = xor i4 [[N1]], [[Y]]
; CHECK-NEXT: [[TMP1:%.*]] = and i4 [[X:%.*]], 1
; CHECK-NEXT: [[TMP2:%.*]] = and i4 [[Y:%.*]], -2
; CHECK-NEXT: [[R:%.*]] = or i4 [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret i4 [[R]]
;
%n0 = xor i4 %x, %y
@ -21,9 +21,9 @@ define i4 @scalar0 (i4 %x, i4 %y) {
define i4 @scalar1 (i4 %x, i4 %y) {
; CHECK-LABEL: @scalar1(
; CHECK-NEXT: [[N0:%.*]] = xor i4 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: [[N1:%.*]] = and i4 [[N0]], -2
; CHECK-NEXT: [[R:%.*]] = xor i4 [[N1]], [[Y]]
; CHECK-NEXT: [[TMP1:%.*]] = and i4 [[X:%.*]], -2
; CHECK-NEXT: [[TMP2:%.*]] = and i4 [[Y:%.*]], 1
; CHECK-NEXT: [[R:%.*]] = or i4 [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret i4 [[R]]
;
%n0 = xor i4 %x, %y
@ -91,9 +91,9 @@ declare i4 @gen4()
define i4 @c_1_0_0 (i4 %x, i4 %y) {
; CHECK-LABEL: @c_1_0_0(
; CHECK-NEXT: [[N0:%.*]] = xor i4 [[Y:%.*]], [[X:%.*]]
; CHECK-NEXT: [[N1:%.*]] = and i4 [[N0]], -2
; CHECK-NEXT: [[R:%.*]] = xor i4 [[N1]], [[Y]]
; CHECK-NEXT: [[TMP1:%.*]] = and i4 [[X:%.*]], -2
; CHECK-NEXT: [[TMP2:%.*]] = and i4 [[Y:%.*]], 1
; CHECK-NEXT: [[R:%.*]] = or i4 [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret i4 [[R]]
;
%n0 = xor i4 %y, %x ; swapped order
@ -104,9 +104,9 @@ define i4 @c_1_0_0 (i4 %x, i4 %y) {
define i4 @c_0_1_0 (i4 %x, i4 %y) {
; CHECK-LABEL: @c_0_1_0(
; CHECK-NEXT: [[N0:%.*]] = xor i4 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: [[N1:%.*]] = and i4 [[N0]], -2
; CHECK-NEXT: [[R:%.*]] = xor i4 [[N1]], [[X]]
; CHECK-NEXT: [[TMP1:%.*]] = and i4 [[Y:%.*]], -2
; CHECK-NEXT: [[TMP2:%.*]] = and i4 [[X:%.*]], 1
; CHECK-NEXT: [[R:%.*]] = or i4 [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret i4 [[R]]
;
%n0 = xor i4 %x, %y
@ -119,9 +119,9 @@ define i4 @c_0_0_1 () {
; CHECK-LABEL: @c_0_0_1(
; CHECK-NEXT: [[X:%.*]] = call i4 @gen4()
; CHECK-NEXT: [[Y:%.*]] = call i4 @gen4()
; CHECK-NEXT: [[N0:%.*]] = xor i4 [[X]], [[Y]]
; CHECK-NEXT: [[N1:%.*]] = and i4 [[N0]], -2
; CHECK-NEXT: [[R:%.*]] = xor i4 [[Y]], [[N1]]
; CHECK-NEXT: [[TMP1:%.*]] = and i4 [[X]], -2
; CHECK-NEXT: [[TMP2:%.*]] = and i4 [[Y]], 1
; CHECK-NEXT: [[R:%.*]] = or i4 [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret i4 [[R]]
;
%x = call i4 @gen4()
@ -134,9 +134,9 @@ define i4 @c_0_0_1 () {
define i4 @c_1_1_0 (i4 %x, i4 %y) {
; CHECK-LABEL: @c_1_1_0(
; CHECK-NEXT: [[N0:%.*]] = xor i4 [[Y:%.*]], [[X:%.*]]
; CHECK-NEXT: [[N1:%.*]] = and i4 [[N0]], -2
; CHECK-NEXT: [[R:%.*]] = xor i4 [[N1]], [[X]]
; CHECK-NEXT: [[TMP1:%.*]] = and i4 [[Y:%.*]], -2
; CHECK-NEXT: [[TMP2:%.*]] = and i4 [[X:%.*]], 1
; CHECK-NEXT: [[R:%.*]] = or i4 [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret i4 [[R]]
;
%n0 = xor i4 %y, %x ; swapped order
@ -148,9 +148,9 @@ define i4 @c_1_1_0 (i4 %x, i4 %y) {
define i4 @c_1_0_1 (i4 %x) {
; CHECK-LABEL: @c_1_0_1(
; CHECK-NEXT: [[Y:%.*]] = call i4 @gen4()
; CHECK-NEXT: [[N0:%.*]] = xor i4 [[Y]], [[X:%.*]]
; CHECK-NEXT: [[N1:%.*]] = and i4 [[N0]], -2
; CHECK-NEXT: [[R:%.*]] = xor i4 [[Y]], [[N1]]
; CHECK-NEXT: [[TMP1:%.*]] = and i4 [[X:%.*]], -2
; CHECK-NEXT: [[TMP2:%.*]] = and i4 [[Y]], 1
; CHECK-NEXT: [[R:%.*]] = or i4 [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret i4 [[R]]
;
%y = call i4 @gen4()
@ -163,9 +163,9 @@ define i4 @c_1_0_1 (i4 %x) {
define i4 @c_0_1_1 (i4 %y) {
; CHECK-LABEL: @c_0_1_1(
; CHECK-NEXT: [[X:%.*]] = call i4 @gen4()
; CHECK-NEXT: [[N0:%.*]] = xor i4 [[X]], [[Y:%.*]]
; CHECK-NEXT: [[N1:%.*]] = and i4 [[N0]], -2
; CHECK-NEXT: [[R:%.*]] = xor i4 [[X]], [[N1]]
; CHECK-NEXT: [[TMP1:%.*]] = and i4 [[Y:%.*]], -2
; CHECK-NEXT: [[TMP2:%.*]] = and i4 [[X]], 1
; CHECK-NEXT: [[R:%.*]] = or i4 [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret i4 [[R]]
;
%x = call i4 @gen4()
@ -179,9 +179,9 @@ define i4 @c_1_1_1 () {
; CHECK-LABEL: @c_1_1_1(
; CHECK-NEXT: [[X:%.*]] = call i4 @gen4()
; CHECK-NEXT: [[Y:%.*]] = call i4 @gen4()
; CHECK-NEXT: [[N0:%.*]] = xor i4 [[Y]], [[X]]
; CHECK-NEXT: [[N1:%.*]] = and i4 [[N0]], -2
; CHECK-NEXT: [[R:%.*]] = xor i4 [[X]], [[N1]]
; CHECK-NEXT: [[TMP1:%.*]] = and i4 [[Y]], -2
; CHECK-NEXT: [[TMP2:%.*]] = and i4 [[X]], 1
; CHECK-NEXT: [[R:%.*]] = or i4 [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret i4 [[R]]
;
%x = call i4 @gen4()

View File

@ -8,9 +8,9 @@
define <2 x i4> @splat (<2 x i4> %x, <2 x i4> %y) {
; CHECK-LABEL: @splat(
; CHECK-NEXT: [[N0:%.*]] = xor <2 x i4> [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: [[N1:%.*]] = and <2 x i4> [[N0]], <i4 -2, i4 -2>
; CHECK-NEXT: [[R:%.*]] = xor <2 x i4> [[N1]], [[Y]]
; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i4> [[X:%.*]], <i4 -2, i4 -2>
; CHECK-NEXT: [[TMP2:%.*]] = and <2 x i4> [[Y:%.*]], <i4 1, i4 1>
; CHECK-NEXT: [[R:%.*]] = or <2 x i4> [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret <2 x i4> [[R]]
;
%n0 = xor <2 x i4> %x, %y
@ -21,9 +21,9 @@ define <2 x i4> @splat (<2 x i4> %x, <2 x i4> %y) {
define <3 x i4> @splat_undef (<3 x i4> %x, <3 x i4> %y) {
; CHECK-LABEL: @splat_undef(
; CHECK-NEXT: [[N0:%.*]] = xor <3 x i4> [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: [[N1:%.*]] = and <3 x i4> [[N0]], <i4 -2, i4 undef, i4 -2>
; CHECK-NEXT: [[R:%.*]] = xor <3 x i4> [[N1]], [[Y]]
; CHECK-NEXT: [[TMP1:%.*]] = and <3 x i4> [[X:%.*]], <i4 -2, i4 undef, i4 -2>
; CHECK-NEXT: [[TMP2:%.*]] = and <3 x i4> [[Y:%.*]], <i4 1, i4 undef, i4 1>
; CHECK-NEXT: [[R:%.*]] = or <3 x i4> [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret <3 x i4> [[R]]
;
%n0 = xor <3 x i4> %x, %y
@ -34,9 +34,9 @@ define <3 x i4> @splat_undef (<3 x i4> %x, <3 x i4> %y) {
define <2 x i4> @nonsplat (<2 x i4> %x, <2 x i4> %y) {
; CHECK-LABEL: @nonsplat(
; CHECK-NEXT: [[N0:%.*]] = xor <2 x i4> [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: [[N1:%.*]] = and <2 x i4> [[N0]], <i4 -2, i4 1>
; CHECK-NEXT: [[R:%.*]] = xor <2 x i4> [[N1]], [[Y]]
; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i4> [[X:%.*]], <i4 -2, i4 1>
; CHECK-NEXT: [[TMP2:%.*]] = and <2 x i4> [[Y:%.*]], <i4 1, i4 -2>
; CHECK-NEXT: [[R:%.*]] = or <2 x i4> [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret <2 x i4> [[R]]
;
%n0 = xor <2 x i4> %x, %y
@ -73,9 +73,8 @@ define <2 x i4> @in_constant_varx_14(<2 x i4> %x, <2 x i4> %mask) {
define <2 x i4> @in_constant_varx_14_nonsplat(<2 x i4> %x, <2 x i4> %mask) {
; CHECK-LABEL: @in_constant_varx_14_nonsplat(
; CHECK-NEXT: [[N0:%.*]] = xor <2 x i4> [[X:%.*]], <i4 -2, i4 7>
; CHECK-NEXT: [[N1:%.*]] = and <2 x i4> [[N0]], <i4 1, i4 1>
; CHECK-NEXT: [[R:%.*]] = xor <2 x i4> [[N1]], <i4 -2, i4 7>
; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i4> [[X:%.*]], <i4 1, i4 1>
; CHECK-NEXT: [[R:%.*]] = or <2 x i4> [[TMP1]], <i4 -2, i4 6>
; CHECK-NEXT: ret <2 x i4> [[R]]
;
%n0 = xor <2 x i4> %x, <i4 14, i4 7> ; %x
@ -86,9 +85,8 @@ define <2 x i4> @in_constant_varx_14_nonsplat(<2 x i4> %x, <2 x i4> %mask) {
define <3 x i4> @in_constant_varx_14_undef(<3 x i4> %x, <3 x i4> %mask) {
; CHECK-LABEL: @in_constant_varx_14_undef(
; CHECK-NEXT: [[N0:%.*]] = xor <3 x i4> [[X:%.*]], <i4 -2, i4 undef, i4 7>
; CHECK-NEXT: [[N1:%.*]] = and <3 x i4> [[N0]], <i4 1, i4 undef, i4 1>
; CHECK-NEXT: [[R:%.*]] = xor <3 x i4> [[N1]], <i4 -2, i4 undef, i4 7>
; CHECK-NEXT: [[TMP1:%.*]] = and <3 x i4> [[X:%.*]], <i4 1, i4 undef, i4 1>
; CHECK-NEXT: [[R:%.*]] = or <3 x i4> [[TMP1]], <i4 -2, i4 undef, i4 6>
; CHECK-NEXT: ret <3 x i4> [[R]]
;
%n0 = xor <3 x i4> %x, <i4 14, i4 undef, i4 7> ; %x
@ -123,9 +121,8 @@ define <2 x i4> @in_constant_14_vary(<2 x i4> %y, <2 x i4> %mask) {
define <2 x i4> @in_constant_14_vary_nonsplat(<2 x i4> %y, <2 x i4> %mask) {
; CHECK-LABEL: @in_constant_14_vary_nonsplat(
; CHECK-NEXT: [[N0:%.*]] = xor <2 x i4> [[Y:%.*]], <i4 -2, i4 7>
; CHECK-NEXT: [[N1:%.*]] = and <2 x i4> [[N0]], <i4 1, i4 1>
; CHECK-NEXT: [[R:%.*]] = xor <2 x i4> [[N1]], [[Y]]
; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i4> [[Y:%.*]], <i4 -2, i4 -2>
; CHECK-NEXT: [[R:%.*]] = or <2 x i4> [[TMP1]], <i4 0, i4 1>
; CHECK-NEXT: ret <2 x i4> [[R]]
;
%n0 = xor <2 x i4> %y, <i4 14, i4 7> ; %x
@ -136,9 +133,8 @@ define <2 x i4> @in_constant_14_vary_nonsplat(<2 x i4> %y, <2 x i4> %mask) {
define <3 x i4> @in_constant_14_vary_undef(<3 x i4> %y, <3 x i4> %mask) {
; CHECK-LABEL: @in_constant_14_vary_undef(
; CHECK-NEXT: [[N0:%.*]] = xor <3 x i4> [[Y:%.*]], <i4 -2, i4 undef, i4 7>
; CHECK-NEXT: [[N1:%.*]] = and <3 x i4> [[N0]], <i4 1, i4 undef, i4 1>
; CHECK-NEXT: [[R:%.*]] = xor <3 x i4> [[N1]], [[Y]]
; CHECK-NEXT: [[TMP1:%.*]] = and <3 x i4> [[Y:%.*]], <i4 -2, i4 undef, i4 -2>
; CHECK-NEXT: [[R:%.*]] = or <3 x i4> [[TMP1]], <i4 0, i4 undef, i4 1>
; CHECK-NEXT: ret <3 x i4> [[R]]
;
%n0 = xor <3 x i4> %y, <i4 14, i4 undef, i4 7> ; %x
@ -156,9 +152,9 @@ declare <2 x i4> @gen4()
define <2 x i4> @c_1_0_0 (<2 x i4> %x, <2 x i4> %y) {
; CHECK-LABEL: @c_1_0_0(
; CHECK-NEXT: [[N0:%.*]] = xor <2 x i4> [[Y:%.*]], [[X:%.*]]
; CHECK-NEXT: [[N1:%.*]] = and <2 x i4> [[N0]], <i4 -2, i4 -2>
; CHECK-NEXT: [[R:%.*]] = xor <2 x i4> [[N1]], [[Y]]
; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i4> [[X:%.*]], <i4 -2, i4 -2>
; CHECK-NEXT: [[TMP2:%.*]] = and <2 x i4> [[Y:%.*]], <i4 1, i4 1>
; CHECK-NEXT: [[R:%.*]] = or <2 x i4> [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret <2 x i4> [[R]]
;
%n0 = xor <2 x i4> %y, %x ; swapped order
@ -169,9 +165,9 @@ define <2 x i4> @c_1_0_0 (<2 x i4> %x, <2 x i4> %y) {
define <2 x i4> @c_0_1_0 (<2 x i4> %x, <2 x i4> %y) {
; CHECK-LABEL: @c_0_1_0(
; CHECK-NEXT: [[N0:%.*]] = xor <2 x i4> [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: [[N1:%.*]] = and <2 x i4> [[N0]], <i4 -2, i4 -2>
; CHECK-NEXT: [[R:%.*]] = xor <2 x i4> [[N1]], [[X]]
; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i4> [[Y:%.*]], <i4 -2, i4 -2>
; CHECK-NEXT: [[TMP2:%.*]] = and <2 x i4> [[X:%.*]], <i4 1, i4 1>
; CHECK-NEXT: [[R:%.*]] = or <2 x i4> [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret <2 x i4> [[R]]
;
%n0 = xor <2 x i4> %x, %y
@ -184,9 +180,9 @@ define <2 x i4> @c_0_0_1 () {
; CHECK-LABEL: @c_0_0_1(
; CHECK-NEXT: [[X:%.*]] = call <2 x i4> @gen4()
; CHECK-NEXT: [[Y:%.*]] = call <2 x i4> @gen4()
; CHECK-NEXT: [[N0:%.*]] = xor <2 x i4> [[X]], [[Y]]
; CHECK-NEXT: [[N1:%.*]] = and <2 x i4> [[N0]], <i4 -2, i4 -2>
; CHECK-NEXT: [[R:%.*]] = xor <2 x i4> [[Y]], [[N1]]
; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i4> [[X]], <i4 -2, i4 -2>
; CHECK-NEXT: [[TMP2:%.*]] = and <2 x i4> [[Y]], <i4 1, i4 1>
; CHECK-NEXT: [[R:%.*]] = or <2 x i4> [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret <2 x i4> [[R]]
;
%x = call <2 x i4> @gen4()
@ -199,9 +195,9 @@ define <2 x i4> @c_0_0_1 () {
define <2 x i4> @c_1_1_0 (<2 x i4> %x, <2 x i4> %y) {
; CHECK-LABEL: @c_1_1_0(
; CHECK-NEXT: [[N0:%.*]] = xor <2 x i4> [[Y:%.*]], [[X:%.*]]
; CHECK-NEXT: [[N1:%.*]] = and <2 x i4> [[N0]], <i4 -2, i4 -2>
; CHECK-NEXT: [[R:%.*]] = xor <2 x i4> [[N1]], [[X]]
; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i4> [[Y:%.*]], <i4 -2, i4 -2>
; CHECK-NEXT: [[TMP2:%.*]] = and <2 x i4> [[X:%.*]], <i4 1, i4 1>
; CHECK-NEXT: [[R:%.*]] = or <2 x i4> [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret <2 x i4> [[R]]
;
%n0 = xor <2 x i4> %y, %x ; swapped order
@ -213,9 +209,9 @@ define <2 x i4> @c_1_1_0 (<2 x i4> %x, <2 x i4> %y) {
define <2 x i4> @c_1_0_1 (<2 x i4> %x) {
; CHECK-LABEL: @c_1_0_1(
; CHECK-NEXT: [[Y:%.*]] = call <2 x i4> @gen4()
; CHECK-NEXT: [[N0:%.*]] = xor <2 x i4> [[Y]], [[X:%.*]]
; CHECK-NEXT: [[N1:%.*]] = and <2 x i4> [[N0]], <i4 -2, i4 -2>
; CHECK-NEXT: [[R:%.*]] = xor <2 x i4> [[Y]], [[N1]]
; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i4> [[X:%.*]], <i4 -2, i4 -2>
; CHECK-NEXT: [[TMP2:%.*]] = and <2 x i4> [[Y]], <i4 1, i4 1>
; CHECK-NEXT: [[R:%.*]] = or <2 x i4> [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret <2 x i4> [[R]]
;
%y = call <2 x i4> @gen4()
@ -228,9 +224,9 @@ define <2 x i4> @c_1_0_1 (<2 x i4> %x) {
define <2 x i4> @c_0_1_1 (<2 x i4> %y) {
; CHECK-LABEL: @c_0_1_1(
; CHECK-NEXT: [[X:%.*]] = call <2 x i4> @gen4()
; CHECK-NEXT: [[N0:%.*]] = xor <2 x i4> [[X]], [[Y:%.*]]
; CHECK-NEXT: [[N1:%.*]] = and <2 x i4> [[N0]], <i4 -2, i4 -2>
; CHECK-NEXT: [[R:%.*]] = xor <2 x i4> [[X]], [[N1]]
; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i4> [[Y:%.*]], <i4 -2, i4 -2>
; CHECK-NEXT: [[TMP2:%.*]] = and <2 x i4> [[X]], <i4 1, i4 1>
; CHECK-NEXT: [[R:%.*]] = or <2 x i4> [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret <2 x i4> [[R]]
;
%x = call <2 x i4> @gen4()
@ -244,9 +240,9 @@ define <2 x i4> @c_1_1_1 () {
; CHECK-LABEL: @c_1_1_1(
; CHECK-NEXT: [[X:%.*]] = call <2 x i4> @gen4()
; CHECK-NEXT: [[Y:%.*]] = call <2 x i4> @gen4()
; CHECK-NEXT: [[N0:%.*]] = xor <2 x i4> [[Y]], [[X]]
; CHECK-NEXT: [[N1:%.*]] = and <2 x i4> [[N0]], <i4 -2, i4 -2>
; CHECK-NEXT: [[R:%.*]] = xor <2 x i4> [[X]], [[N1]]
; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i4> [[Y]], <i4 -2, i4 -2>
; CHECK-NEXT: [[TMP2:%.*]] = and <2 x i4> [[X]], <i4 1, i4 1>
; CHECK-NEXT: [[R:%.*]] = or <2 x i4> [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret <2 x i4> [[R]]
;
%x = call <2 x i4> @gen4()