1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown -mattr=+sse4.1,-slow-unaligned-mem-16 | FileCheck %s
3
4define <8 x i16> @commute_fold_pblendw(<8 x i16> %a, <8 x i16>* %b) {
5; CHECK-LABEL: commute_fold_pblendw:
6; CHECK:       # %bb.0:
7; CHECK-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0],mem[1,2,3],xmm0[4],mem[5,6,7]
8; CHECK-NEXT:    retq
9  %1 = load <8 x i16>, <8 x i16>* %b
10  %2 = call <8 x i16> @llvm.x86.sse41.pblendw(<8 x i16> %1, <8 x i16> %a, i8 17)
11  ret <8 x i16> %2
12}
13declare <8 x i16> @llvm.x86.sse41.pblendw(<8 x i16>, <8 x i16>, i8) nounwind readnone
14
15define <4 x float> @commute_fold_blendps(<4 x float> %a, <4 x float>* %b) {
16; CHECK-LABEL: commute_fold_blendps:
17; CHECK:       # %bb.0:
18; CHECK-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2],mem[3]
19; CHECK-NEXT:    retq
20  %1 = load <4 x float>, <4 x float>* %b
21  %2 = call <4 x float> @llvm.x86.sse41.blendps(<4 x float> %1, <4 x float> %a, i8 5)
22  ret <4 x float> %2
23}
24declare <4 x float> @llvm.x86.sse41.blendps(<4 x float>, <4 x float>, i8) nounwind readnone
25
26define <2 x double> @commute_fold_blendpd(<2 x double> %a, <2 x double>* %b) {
27; CHECK-LABEL: commute_fold_blendpd:
28; CHECK:       # %bb.0:
29; CHECK-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0,1],mem[2,3]
30; CHECK-NEXT:    retq
31  %1 = load <2 x double>, <2 x double>* %b
32  %2 = call <2 x double> @llvm.x86.sse41.blendpd(<2 x double> %1, <2 x double> %a, i8 1)
33  ret <2 x double> %2
34}
35declare <2 x double> @llvm.x86.sse41.blendpd(<2 x double>, <2 x double>, i8) nounwind readnone
36
37define <4 x i32> @commute_fold_blend_v4i32(<4 x i32>* %a, <4 x i32> %b) {
38; CHECK-LABEL: commute_fold_blend_v4i32:
39; CHECK:       # %bb.0:
40; CHECK-NEXT:    paddd %xmm0, %xmm0
41; CHECK-NEXT:    pblendw {{.*#+}} xmm0 = mem[0,1,2,3,4,5],xmm0[6,7]
42; CHECK-NEXT:    retq
43  %1 = load <4 x i32>, <4 x i32>* %a
44  %2 = add <4 x i32> %b, %b ; force integer domain
45  %3 = shufflevector <4 x i32> %1, <4 x i32> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 7>
46  ret <4 x i32> %3
47}
48
49; Test case for a crash that occurred due to blendi being commuted to
50; movsd during two address instruction pass. The change in number of operands
51; caused a bad call to getOperand. This caused the revert in r354713.
52%struct.spam = type { i64, i64 }
53
54define void @baz(<2 x i64>* %arg, %struct.spam* %arg1) optsize {
55; CHECK-LABEL: baz:
56; CHECK:       # %bb.0: # %bb
57; CHECK-NEXT:    movaps (%rdi), %xmm0
58; CHECK-NEXT:    movaps {{.*#+}} xmm1 = [3,3]
59; CHECK-NEXT:    andps %xmm0, %xmm1
60; CHECK-NEXT:    blendps {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3]
61; CHECK-NEXT:    movups %xmm1, (%rsi)
62; CHECK-NEXT:    retq
63bb:
64  %tmp = load <2 x i64>, <2 x i64>* %arg, align 16
65  %tmp2 = and <2 x i64> %tmp, <i64 3, i64 3>
66  %tmp3 = getelementptr inbounds %struct.spam, %struct.spam* %arg1, i64 0, i32 0
67  %tmp4 = extractelement <2 x i64> %tmp, i32 0
68  store i64 %tmp4, i64* %tmp3, align 8
69  %tmp5 = getelementptr inbounds %struct.spam, %struct.spam* %arg1, i64 0, i32 1
70  %tmp6 = extractelement <2 x i64> %tmp2, i32 1
71  store i64 %tmp6, i64* %tmp5, align 8
72  ret void
73}
74