xref: /llvm-project/llvm/test/CodeGen/X86/apx/kmov-postrapseudos.ll (revision 55c6bda01ef5a166a69b43956775272d9d67bda5)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx512f,+egpr -show-mc-encoding | FileCheck --check-prefix=AVX512 %s
3; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx512f,+avx512bw,+egpr -show-mc-encoding | FileCheck --check-prefix=AVX512BW %s
4
5define void @kmovkr_1(i1 %cmp23.not) {
6; AVX512-LABEL: kmovkr_1:
7; AVX512:       # %bb.0: # %entry
8; AVX512-NEXT:    kmovw %edi, %k1 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x92,0xcf]
9; AVX512-NEXT:    vmovsd {{.*#+}} xmm0 {%k1} {z} = [1.0E+0,0.0E+0]
10; AVX512-NEXT:    # encoding: [0x62,0xf1,0xff,0x89,0x10,0x05,A,A,A,A]
11; AVX512-NEXT:    # fixup A - offset: 6, value: {{\.?LCPI[0-9]+_[0-9]+}}-4, kind: reloc_riprel_4byte
12; AVX512-NEXT:    vmovsd %xmm0, 0 # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x11,0x04,0x25,0x00,0x00,0x00,0x00]
13; AVX512-NEXT:    retq # encoding: [0xc3]
14;
15; AVX512BW-LABEL: kmovkr_1:
16; AVX512BW:       # %bb.0: # %entry
17; AVX512BW-NEXT:    kmovd %edi, %k1 # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x92,0xcf]
18; AVX512BW-NEXT:    vmovsd {{.*#+}} xmm0 {%k1} {z} = [1.0E+0,0.0E+0]
19; AVX512BW-NEXT:    # encoding: [0x62,0xf1,0xff,0x89,0x10,0x05,A,A,A,A]
20; AVX512BW-NEXT:    # fixup A - offset: 6, value: {{\.?LCPI[0-9]+_[0-9]+}}-4, kind: reloc_riprel_4byte
21; AVX512BW-NEXT:    vmovsd %xmm0, 0 # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x11,0x04,0x25,0x00,0x00,0x00,0x00]
22; AVX512BW-NEXT:    retq # encoding: [0xc3]
23entry:
24  %0 = select i1 %cmp23.not, double 1.000000e+00, double 0.000000e+00
25  store double %0, ptr null, align 8
26  ret void
27}
28
29define void @kmovkr_2() {
30; AVX512-LABEL: kmovkr_2:
31; AVX512:       # %bb.0: # %alloca_21
32; AVX512-NEXT:    vxorps %xmm0, %xmm0, %xmm0 # encoding: [0xc5,0xf8,0x57,0xc0]
33; AVX512-NEXT:    movw $3, %ax # encoding: [0x66,0xb8,0x03,0x00]
34; AVX512-NEXT:    kmovw %eax, %k1 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x92,0xc8]
35; AVX512-NEXT:    vmovups %zmm0, 0 {%k1} # encoding: [0x62,0xf1,0x7c,0x49,0x11,0x04,0x25,0x00,0x00,0x00,0x00]
36; AVX512-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
37; AVX512-NEXT:    retq # encoding: [0xc3]
38;
39; AVX512BW-LABEL: kmovkr_2:
40; AVX512BW:       # %bb.0: # %alloca_21
41; AVX512BW-NEXT:    vxorps %xmm0, %xmm0, %xmm0 # encoding: [0xc5,0xf8,0x57,0xc0]
42; AVX512BW-NEXT:    movw $3, %ax # encoding: [0x66,0xb8,0x03,0x00]
43; AVX512BW-NEXT:    kmovd %eax, %k1 # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x92,0xc8]
44; AVX512BW-NEXT:    vmovups %zmm0, 0 {%k1} # encoding: [0x62,0xf1,0x7c,0x49,0x11,0x04,0x25,0x00,0x00,0x00,0x00]
45; AVX512BW-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
46; AVX512BW-NEXT:    retq # encoding: [0xc3]
47alloca_21:
48  call void @llvm.masked.store.v4f32.p0(<4 x float> zeroinitializer, ptr null, i32 1, <4 x i1> <i1 true, i1 true, i1 false, i1 false>)
49  ret void
50}
51
52define i32 @kmovrk_1(<4 x ptr> %arg) {
53; AVX512-LABEL: kmovrk_1:
54; AVX512:       # %bb.0: # %bb
55; AVX512-NEXT:    vptest %ymm0, %ymm0 # encoding: [0xc4,0xe2,0x7d,0x17,0xc0]
56; AVX512-NEXT:    jne .LBB2_1 # encoding: [0x75,A]
57; AVX512-NEXT:    # fixup A - offset: 1, value: .LBB2_1-1, kind: FK_PCRel_1
58; AVX512-NEXT:  # %bb.2: # %bb3
59; AVX512-NEXT:    xorl %eax, %eax # encoding: [0x31,0xc0]
60; AVX512-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
61; AVX512-NEXT:    retq # encoding: [0xc3]
62; AVX512-NEXT:  .LBB2_1: # %bb2
63;
64; AVX512BW-LABEL: kmovrk_1:
65; AVX512BW:       # %bb.0: # %bb
66; AVX512BW-NEXT:    vptest %ymm0, %ymm0 # encoding: [0xc4,0xe2,0x7d,0x17,0xc0]
67; AVX512BW-NEXT:    jne .LBB2_1 # encoding: [0x75,A]
68; AVX512BW-NEXT:    # fixup A - offset: 1, value: .LBB2_1-1, kind: FK_PCRel_1
69; AVX512BW-NEXT:  # %bb.2: # %bb3
70; AVX512BW-NEXT:    xorl %eax, %eax # encoding: [0x31,0xc0]
71; AVX512BW-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
72; AVX512BW-NEXT:    retq # encoding: [0xc3]
73; AVX512BW-NEXT:  .LBB2_1: # %bb2
74bb:
75  %icmp = icmp ne <4 x ptr> %arg, zeroinitializer
76  %freeze = freeze <4 x i1> %icmp
77  %bitcast = bitcast <4 x i1> %freeze to i4
78  %icmp1 = icmp ne i4 %bitcast, 0
79  br i1 %icmp1, label %bb2, label %bb3
80bb2:
81  unreachable
82bb3:
83  ret i32 0
84}
85
86declare void @llvm.masked.store.v4f32.p0(<4 x float>, ptr nocapture, i32 immarg, <4 x i1>)
87