1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc -verify-machineinstrs -mcpu=pwr8 -mtriple=powerpc64le < %s | FileCheck %s
4 define i64 @test1(ptr %a, ptr %b) {
6 ; CHECK: # %bb.0: # %entry
7 ; CHECK-NEXT: ld 5, 0(3)
8 ; CHECK-NEXT: ld 6, 0(4)
9 ; CHECK-NEXT: mtvsrd 34, 5
10 ; CHECK-NEXT: mtvsrd 35, 6
11 ; CHECK-NEXT: add 4, 5, 6
12 ; CHECK-NEXT: vavgsb 2, 2, 3
13 ; CHECK-NEXT: stxsdx 34, 0, 3
17 %lhs = load i64, ptr %a, align 8
18 %rhs = load i64, ptr %b, align 8
19 %sum = add i64 %lhs, %rhs
20 %lv = insertelement <2 x i64> undef, i64 %lhs, i32 0
21 %rv = insertelement <2 x i64> undef, i64 %rhs, i32 0
22 %lhc = bitcast <2 x i64> %lv to <16 x i8>
23 %rhc = bitcast <2 x i64> %rv to <16 x i8>
24 %add = call <16 x i8> @llvm.ppc.altivec.vavgsb(<16 x i8> %lhc, <16 x i8> %rhc)
25 %cb = bitcast <16 x i8> %add to <2 x i64>
26 %fv = extractelement <2 x i64> %cb, i32 0
27 store i64 %fv, ptr %a, align 8
31 define i64 @test2(ptr %a, ptr %b) {
33 ; CHECK: # %bb.0: # %entry
34 ; CHECK-NEXT: ld 5, 0(3)
35 ; CHECK-NEXT: ld 6, 0(4)
36 ; CHECK-NEXT: mtvsrd 34, 5
37 ; CHECK-NEXT: mtvsrd 35, 6
38 ; CHECK-NEXT: add 4, 5, 6
39 ; CHECK-NEXT: vadduhm 2, 2, 3
40 ; CHECK-NEXT: stxsdx 34, 0, 3
44 %lhs = load i64, ptr %a, align 8
45 %rhs = load i64, ptr %b, align 8
46 %sum = add i64 %lhs, %rhs
47 %lv = insertelement <2 x i64> undef, i64 %lhs, i32 0
48 %rv = insertelement <2 x i64> undef, i64 %rhs, i32 0
49 %lhc = bitcast <2 x i64> %lv to <8 x i16>
50 %rhc = bitcast <2 x i64> %rv to <8 x i16>
51 %add = add <8 x i16> %lhc, %rhc
52 %cb = bitcast <8 x i16> %add to <2 x i64>
53 %fv = extractelement <2 x i64> %cb, i32 0
54 store i64 %fv, ptr %a, align 8
58 ; Ensure that vec-ops with multiple uses aren't simplified.
59 define signext i16 @vecop_uses(ptr %addr) {
60 ; CHECK-LABEL: vecop_uses:
61 ; CHECK: # %bb.0: # %entry
62 ; CHECK-NEXT: li 4, 16
63 ; CHECK-NEXT: lxvd2x 0, 3, 4
64 ; CHECK-NEXT: xxswapd 34, 0
65 ; CHECK-NEXT: lxvd2x 0, 0, 3
66 ; CHECK-NEXT: xxswapd 35, 0
67 ; CHECK-NEXT: vminsh 2, 3, 2
68 ; CHECK-NEXT: xxswapd 35, 34
69 ; CHECK-NEXT: vminsh 2, 2, 3
70 ; CHECK-NEXT: xxspltw 35, 34, 2
71 ; CHECK-NEXT: vminsh 2, 2, 3
72 ; CHECK-NEXT: vsplth 3, 2, 6
73 ; CHECK-NEXT: vminsh 2, 2, 3
74 ; CHECK-NEXT: xxswapd 0, 34
75 ; CHECK-NEXT: mffprd 3, 0
76 ; CHECK-NEXT: clrldi 3, 3, 48
77 ; CHECK-NEXT: extsh 3, 3
80 %0 = load <16 x i16>, ptr %addr, align 2
81 %1 = call i16 @llvm.vector.reduce.smin.v16i16(<16 x i16> %0)
85 define signext i32 @vecop_uses2(ptr %a, ptr %b, ptr %c) {
86 ; CHECK-LABEL: vecop_uses2:
87 ; CHECK: # %bb.0: # %entry
88 ; CHECK-NEXT: lxvd2x 0, 0, 3
89 ; CHECK-NEXT: xxswapd 34, 0
90 ; CHECK-NEXT: lxvd2x 0, 0, 4
91 ; CHECK-NEXT: xxswapd 35, 0
92 ; CHECK-NEXT: xxsldwi 0, 34, 34, 3
93 ; CHECK-NEXT: vmuluwm 2, 3, 2
94 ; CHECK-NEXT: mffprwz 3, 0
95 ; CHECK-NEXT: extsw 3, 3
96 ; CHECK-NEXT: xxswapd 1, 34
97 ; CHECK-NEXT: stxvd2x 1, 0, 5
100 %0 = load <4 x i32>, ptr %a, align 4
101 %1 = load <4 x i32>, ptr %b, align 4
102 %2 = mul <4 x i32> %1, %0
103 store <4 x i32> %2, ptr %c, align 4
104 %3 = extractelement <4 x i32> %0, i32 3
108 declare <16 x i8> @llvm.ppc.altivec.vavgsb(<16 x i8>, <16 x i8>)
109 declare i16 @llvm.vector.reduce.smin.v16i16(<16 x i16>)