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
4 define <8 x i16> @commute_fold_pblendw(<8 x i16> %a, ptr %b) {
5 ; CHECK-LABEL: commute_fold_pblendw:
7 ; CHECK-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0],mem[1,2,3],xmm0[4],mem[5,6,7]
9 %1 = load <8 x i16>, ptr %b
10 %2 = call <8 x i16> @llvm.x86.sse41.pblendw(<8 x i16> %1, <8 x i16> %a, i8 17)
13 declare <8 x i16> @llvm.x86.sse41.pblendw(<8 x i16>, <8 x i16>, i8) nounwind readnone
15 define <4 x float> @commute_fold_blendps(<4 x float> %a, ptr %b) {
16 ; CHECK-LABEL: commute_fold_blendps:
18 ; CHECK-NEXT: blendps {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2],mem[3]
20 %1 = load <4 x float>, ptr %b
21 %2 = call <4 x float> @llvm.x86.sse41.blendps(<4 x float> %1, <4 x float> %a, i8 5)
24 declare <4 x float> @llvm.x86.sse41.blendps(<4 x float>, <4 x float>, i8) nounwind readnone
26 define <2 x double> @commute_fold_blendpd(<2 x double> %a, ptr %b) {
27 ; CHECK-LABEL: commute_fold_blendpd:
29 ; CHECK-NEXT: blendps {{.*#+}} xmm0 = xmm0[0,1],mem[2,3]
31 %1 = load <2 x double>, ptr %b
32 %2 = call <2 x double> @llvm.x86.sse41.blendpd(<2 x double> %1, <2 x double> %a, i8 1)
35 declare <2 x double> @llvm.x86.sse41.blendpd(<2 x double>, <2 x double>, i8) nounwind readnone
37 define <4 x i32> @commute_fold_blend_v4i32(ptr %a, <4 x i32> %b) {
38 ; CHECK-LABEL: commute_fold_blend_v4i32:
40 ; CHECK-NEXT: paddd %xmm0, %xmm0
41 ; CHECK-NEXT: pblendw {{.*#+}} xmm0 = mem[0,1,2,3,4,5],xmm0[6,7]
43 %1 = load <4 x i32>, ptr %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>
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 }
54 define void @baz(ptr %arg, ptr %arg1) optsize {
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)
64 %tmp = load <2 x i64>, ptr %arg, align 16
65 %tmp2 = and <2 x i64> %tmp, <i64 3, i64 3>
66 %tmp4 = extractelement <2 x i64> %tmp, i32 0
67 store i64 %tmp4, ptr %arg1, align 8
68 %tmp5 = getelementptr inbounds %struct.spam, ptr %arg1, i64 0, i32 1
69 %tmp6 = extractelement <2 x i64> %tmp2, i32 1
70 store i64 %tmp6, ptr %tmp5, align 8