xref: /aosp_15_r20/external/llvm/test/CodeGen/X86/avx512-bugfix-26264.ll (revision 9880d6810fe72a1726cb53787c6711e909410d58)
1*9880d681SAndroid Build Coastguard Worker; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2*9880d681SAndroid Build Coastguard Worker; RUN: llc -mtriple=x86_64-apple-darwin -mattr=avx512f,avx512bw < %s | FileCheck %s --check-prefix=AVX512BW
3*9880d681SAndroid Build Coastguard Worker
4*9880d681SAndroid Build Coastguard Workerdefine <32 x double> @test_load_32f64(<32 x double>* %ptrs, <32 x i1> %mask, <32 x double> %src0)  {
5*9880d681SAndroid Build Coastguard Worker; AVX512BW-LABEL: test_load_32f64:
6*9880d681SAndroid Build Coastguard Worker; AVX512BW:       ## BB#0:
7*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vpsllw $7, %ymm0, %ymm0
8*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vpmovb2m %zmm0, %k1
9*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vmovupd (%rdi), %zmm1 {%k1}
10*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    kshiftrd $16, %k1, %k2
11*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vmovupd 128(%rdi), %zmm3 {%k2}
12*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    kshiftrw $8, %k1, %k1
13*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vmovupd 64(%rdi), %zmm2 {%k1}
14*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    kshiftrw $8, %k2, %k1
15*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vmovupd 192(%rdi), %zmm4 {%k1}
16*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vmovaps %zmm1, %zmm0
17*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vmovaps %zmm2, %zmm1
18*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vmovaps %zmm3, %zmm2
19*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vmovaps %zmm4, %zmm3
20*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    retq
21*9880d681SAndroid Build Coastguard Worker  %res = call <32 x double> @llvm.masked.load.v32f64.p0v32f64(<32 x double>* %ptrs, i32 4, <32 x i1> %mask, <32 x double> %src0)
22*9880d681SAndroid Build Coastguard Worker  ret <32 x double> %res
23*9880d681SAndroid Build Coastguard Worker}
24*9880d681SAndroid Build Coastguard Worker
25*9880d681SAndroid Build Coastguard Workerdefine <32 x i64> @test_load_32i64(<32 x i64>* %ptrs, <32 x i1> %mask, <32 x i64> %src0)  {
26*9880d681SAndroid Build Coastguard Worker; AVX512BW-LABEL: test_load_32i64:
27*9880d681SAndroid Build Coastguard Worker; AVX512BW:       ## BB#0:
28*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vpsllw $7, %ymm0, %ymm0
29*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vpmovb2m %zmm0, %k1
30*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vmovdqu64 (%rdi), %zmm1 {%k1}
31*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    kshiftrd $16, %k1, %k2
32*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vmovdqu64 128(%rdi), %zmm3 {%k2}
33*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    kshiftrw $8, %k1, %k1
34*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vmovdqu64 64(%rdi), %zmm2 {%k1}
35*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    kshiftrw $8, %k2, %k1
36*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vmovdqu64 192(%rdi), %zmm4 {%k1}
37*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vmovaps %zmm1, %zmm0
38*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vmovaps %zmm2, %zmm1
39*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vmovaps %zmm3, %zmm2
40*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    vmovaps %zmm4, %zmm3
41*9880d681SAndroid Build Coastguard Worker; AVX512BW-NEXT:    retq
42*9880d681SAndroid Build Coastguard Worker  %res = call <32 x i64> @llvm.masked.load.v32i64.p0v32i64(<32 x i64>* %ptrs, i32 4, <32 x i1> %mask, <32 x i64> %src0)
43*9880d681SAndroid Build Coastguard Worker  ret <32 x i64> %res
44*9880d681SAndroid Build Coastguard Worker}
45*9880d681SAndroid Build Coastguard Worker
46*9880d681SAndroid Build Coastguard Workerdeclare <32 x i64> @llvm.masked.load.v32i64.p0v32i64(<32 x i64>* %ptrs, i32, <32 x i1> %mask, <32 x i64> %src0)
47*9880d681SAndroid Build Coastguard Workerdeclare <32 x double> @llvm.masked.load.v32f64.p0v32f64(<32 x double>* %ptrs, i32, <32 x i1> %mask, <32 x double> %src0)
48