1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc -O3 -mtriple=x86_64-linux-generic -mattr=avx < %s | FileCheck %s
5 ; The LowerMLOAD() method AVX masked load branch should
6 ; use the operand vector type rather than the mask type.
8 ; v4f64,ch = masked_load ..
9 ; The select should be:
14 define <16 x double> @bug45563(ptr %addr, <16 x double> %dst, <16 x i64> %e, <16 x i64> %f) {
15 ; CHECK-LABEL: bug45563:
17 ; CHECK-NEXT: pushq %rbp
18 ; CHECK-NEXT: .cfi_def_cfa_offset 16
19 ; CHECK-NEXT: .cfi_offset %rbp, -16
20 ; CHECK-NEXT: movq %rsp, %rbp
21 ; CHECK-NEXT: .cfi_def_cfa_register %rbp
22 ; CHECK-NEXT: andq $-32, %rsp
23 ; CHECK-NEXT: subq $32, %rsp
24 ; CHECK-NEXT: vextractf128 $1, %ymm7, %xmm8
25 ; CHECK-NEXT: vmovdqa 112(%rbp), %xmm9
26 ; CHECK-NEXT: vmovdqa 128(%rbp), %xmm10
27 ; CHECK-NEXT: vpcmpgtq %xmm8, %xmm10, %xmm8
28 ; CHECK-NEXT: vpcmpgtq %xmm7, %xmm9, %xmm7
29 ; CHECK-NEXT: vinsertf128 $1, %xmm8, %ymm7, %ymm7
30 ; CHECK-NEXT: vextractf128 $1, %ymm6, %xmm8
31 ; CHECK-NEXT: vmovdqa 80(%rbp), %xmm9
32 ; CHECK-NEXT: vmovdqa 96(%rbp), %xmm10
33 ; CHECK-NEXT: vpcmpgtq %xmm8, %xmm10, %xmm8
34 ; CHECK-NEXT: vpcmpgtq %xmm6, %xmm9, %xmm6
35 ; CHECK-NEXT: vinsertf128 $1, %xmm8, %ymm6, %ymm6
36 ; CHECK-NEXT: vextractf128 $1, %ymm5, %xmm8
37 ; CHECK-NEXT: vmovdqa 48(%rbp), %xmm9
38 ; CHECK-NEXT: vmovdqa 64(%rbp), %xmm10
39 ; CHECK-NEXT: vpcmpgtq %xmm8, %xmm10, %xmm8
40 ; CHECK-NEXT: vpcmpgtq %xmm5, %xmm9, %xmm5
41 ; CHECK-NEXT: vinsertf128 $1, %xmm8, %ymm5, %ymm5
42 ; CHECK-NEXT: vextractf128 $1, %ymm4, %xmm8
43 ; CHECK-NEXT: vmovdqa 16(%rbp), %xmm9
44 ; CHECK-NEXT: vmovdqa 32(%rbp), %xmm10
45 ; CHECK-NEXT: vpcmpgtq %xmm8, %xmm10, %xmm8
46 ; CHECK-NEXT: vpcmpgtq %xmm4, %xmm9, %xmm4
47 ; CHECK-NEXT: vinsertf128 $1, %xmm8, %ymm4, %ymm4
48 ; CHECK-NEXT: vmaskmovpd (%rdi), %ymm4, %ymm8
49 ; CHECK-NEXT: vblendvpd %ymm4, %ymm8, %ymm0, %ymm0
50 ; CHECK-NEXT: vmaskmovpd 32(%rdi), %ymm5, %ymm4
51 ; CHECK-NEXT: vblendvpd %ymm5, %ymm4, %ymm1, %ymm1
52 ; CHECK-NEXT: vmaskmovpd 64(%rdi), %ymm6, %ymm4
53 ; CHECK-NEXT: vblendvpd %ymm6, %ymm4, %ymm2, %ymm2
54 ; CHECK-NEXT: vmaskmovpd 96(%rdi), %ymm7, %ymm4
55 ; CHECK-NEXT: vblendvpd %ymm7, %ymm4, %ymm3, %ymm3
56 ; CHECK-NEXT: movq %rbp, %rsp
57 ; CHECK-NEXT: popq %rbp
58 ; CHECK-NEXT: .cfi_def_cfa %rsp, 8
60 %mask = icmp slt <16 x i64> %e, %f
61 %res = call <16 x double> @llvm.masked.load.v16f64.p0(ptr %addr, i32 4, <16 x i1>%mask, <16 x double> %dst)
62 ret <16 x double> %res
65 declare <16 x double> @llvm.masked.load.v16f64.p0(ptr %addr, i32 %align, <16 x i1> %mask, <16 x double> %dst)