1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=x86_64-- -mcpu=skx | FileCheck %s
4 define void @PR48727() {
5 ; CHECK-LABEL: PR48727:
6 ; CHECK: # %bb.0: # %entry
7 ; CHECK-NEXT: vcvttpd2dqy 0, %xmm0
8 ; CHECK-NEXT: vcvttpd2dqy 128, %xmm1
9 ; CHECK-NEXT: movq (%rax), %rax
10 ; CHECK-NEXT: vcvttpd2dqy 160, %xmm2
11 ; CHECK-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1
12 ; CHECK-NEXT: vcvttpd2dqy (%rax), %xmm2
13 ; CHECK-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0
14 ; CHECK-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
15 ; CHECK-NEXT: vpmovdw %zmm0, %ymm0
16 ; CHECK-NEXT: vmovdqu %ymm0, 16(%rax)
17 ; CHECK-NEXT: vzeroupper
20 %0 = load [100 x [100 x i16]]*, [100 x [100 x i16]]** undef, align 8
21 %wide.load.2 = load <4 x double>, <4 x double>* null, align 16
22 %1 = fptosi <4 x double> %wide.load.2 to <4 x i16>
23 %2 = getelementptr inbounds [100 x [100 x i16]], [100 x [100 x i16]]* %0, i64 0, i64 0, i64 8
24 %3 = bitcast i16* %2 to <4 x i16>*
25 store <4 x i16> %1, <4 x i16>* %3, align 8
26 %wide.load.3 = load <4 x double>, <4 x double>* undef, align 16, !invariant.load !0, !noalias !1
27 %4 = fptosi <4 x double> %wide.load.3 to <4 x i16>
28 %5 = getelementptr inbounds [100 x [100 x i16]], [100 x [100 x i16]]* %0, i64 0, i64 0, i64 12
29 %6 = bitcast i16* %5 to <4 x i16>*
30 store <4 x i16> %4, <4 x i16>* %6, align 8
31 %7 = getelementptr inbounds [100 x [100 x double]], [100 x [100 x double]]* null, i64 0, i64 0, i64 16
32 %8 = bitcast double* %7 to <4 x double>*
33 %wide.load.4 = load <4 x double>, <4 x double>* %8, align 16, !invariant.load !0, !noalias !1
34 %9 = fptosi <4 x double> %wide.load.4 to <4 x i16>
35 %10 = getelementptr inbounds [100 x [100 x i16]], [100 x [100 x i16]]* %0, i64 0, i64 0, i64 16
36 %11 = bitcast i16* %10 to <4 x i16>*
37 store <4 x i16> %9, <4 x i16>* %11, align 8
38 %12 = getelementptr inbounds [100 x [100 x double]], [100 x [100 x double]]* null, i64 0, i64 0, i64 20
39 %13 = bitcast double* %12 to <4 x double>*
40 %wide.load.5 = load <4 x double>, <4 x double>* %13, align 16, !invariant.load !0, !noalias !1
41 %14 = fptosi <4 x double> %wide.load.5 to <4 x i16>
42 %15 = getelementptr inbounds [100 x [100 x i16]], [100 x [100 x i16]]* %0, i64 0, i64 0, i64 20
43 %16 = bitcast i16* %15 to <4 x i16>*
44 store <4 x i16> %14, <4 x i16>* %16, align 8
50 !2 = !{!"buffer: {index:1, offset:0, size:20000}", !3}
51 !3 = !{!"XLA global AA domain"}