xref: /llvm-project/llvm/test/CodeGen/X86/avx512fp16-fmaxnum.ll (revision f1de9d6dae174feb5000ad6a1b492b8cb717f5b6) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -verify-machineinstrs --show-mc-encoding -mtriple=x86_64-unknown-unknown -mattr=+avx512fp16    | FileCheck %s --check-prefixes=CHECK
3
4declare half @llvm.maxnum.f16(half, half)
5declare <2 x half> @llvm.maxnum.v2f16(<2 x half>, <2 x half>)
6declare <4 x half> @llvm.maxnum.v4f16(<4 x half>, <4 x half>)
7declare <8 x half> @llvm.maxnum.v8f16(<8 x half>, <8 x half>)
8declare <16 x half> @llvm.maxnum.v16f16(<16 x half>, <16 x half>)
9declare <32 x half> @llvm.maxnum.v32f16(<32 x half>, <32 x half>)
10
11define half @test_intrinsic_fmaxh(half %x, half %y) {
12; CHECK-LABEL: test_intrinsic_fmaxh:
13; CHECK:       # %bb.0:
14; CHECK-NEXT:    vmaxsh %xmm0, %xmm1, %xmm2 # encoding: [0x62,0xf5,0x76,0x08,0x5f,0xd0]
15; CHECK-NEXT:    vcmpunordsh %xmm0, %xmm0, %k1 # encoding: [0x62,0xf3,0x7e,0x08,0xc2,0xc8,0x03]
16; CHECK-NEXT:    vmovsh %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf5,0x7e,0x09,0x10,0xd1]
17; CHECK-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
18; CHECK-NEXT:    retq # encoding: [0xc3]
19  %z = call half @llvm.maxnum.f16(half %x, half %y) readnone
20  ret half %z
21}
22
23define <2 x half> @test_intrinsic_fmax_v2f16(<2 x half> %x, <2 x half> %y) {
24; CHECK-LABEL: test_intrinsic_fmax_v2f16:
25; CHECK:       # %bb.0:
26; CHECK-NEXT:    vmaxph %xmm0, %xmm1, %xmm2 # encoding: [0x62,0xf5,0x74,0x08,0x5f,0xd0]
27; CHECK-NEXT:    vcmpunordph %xmm0, %xmm0, %k1 # encoding: [0x62,0xf3,0x7c,0x08,0xc2,0xc8,0x03]
28; CHECK-NEXT:    vmovdqu16 %xmm1, %xmm2 {%k1} # encoding: [0x62,0xf1,0xff,0x09,0x6f,0xd1]
29; CHECK-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
30; CHECK-NEXT:    retq # encoding: [0xc3]
31  %z = call <2 x half> @llvm.maxnum.v2f16(<2 x half> %x, <2 x half> %y) readnone
32  ret <2 x half> %z
33}
34
35define <4 x half> @test_intrinsic_fmax_v4f16(<4 x half> %x, <4 x half> %y) {
36; CHECK-LABEL: test_intrinsic_fmax_v4f16:
37; CHECK:       # %bb.0:
38; CHECK-NEXT:    vmaxph %xmm0, %xmm1, %xmm2 # encoding: [0x62,0xf5,0x74,0x08,0x5f,0xd0]
39; CHECK-NEXT:    vcmpunordph %xmm0, %xmm0, %k1 # encoding: [0x62,0xf3,0x7c,0x08,0xc2,0xc8,0x03]
40; CHECK-NEXT:    vmovdqu16 %xmm1, %xmm2 {%k1} # encoding: [0x62,0xf1,0xff,0x09,0x6f,0xd1]
41; CHECK-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
42; CHECK-NEXT:    retq # encoding: [0xc3]
43  %z = call <4 x half> @llvm.maxnum.v4f16(<4 x half> %x, <4 x half> %y) readnone
44  ret <4 x half> %z
45}
46
47define <8 x half> @test_intrinsic_fmax_v8f16(<8 x half> %x, <8 x half> %y) {
48; CHECK-LABEL: test_intrinsic_fmax_v8f16:
49; CHECK:       # %bb.0:
50; CHECK-NEXT:    vmaxph %xmm0, %xmm1, %xmm2 # encoding: [0x62,0xf5,0x74,0x08,0x5f,0xd0]
51; CHECK-NEXT:    vcmpunordph %xmm0, %xmm0, %k1 # encoding: [0x62,0xf3,0x7c,0x08,0xc2,0xc8,0x03]
52; CHECK-NEXT:    vmovdqu16 %xmm1, %xmm2 {%k1} # encoding: [0x62,0xf1,0xff,0x09,0x6f,0xd1]
53; CHECK-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
54; CHECK-NEXT:    retq # encoding: [0xc3]
55  %z = call <8 x half> @llvm.maxnum.v8f16(<8 x half> %x, <8 x half> %y) readnone
56  ret <8 x half> %z
57}
58
59define <16 x half> @test_intrinsic_fmax_v16f16(<16 x half> %x, <16 x half> %y) {
60; CHECK-LABEL: test_intrinsic_fmax_v16f16:
61; CHECK:       # %bb.0:
62; CHECK-NEXT:    vmaxph %ymm0, %ymm1, %ymm2 # encoding: [0x62,0xf5,0x74,0x28,0x5f,0xd0]
63; CHECK-NEXT:    vcmpunordph %ymm0, %ymm0, %k1 # encoding: [0x62,0xf3,0x7c,0x28,0xc2,0xc8,0x03]
64; CHECK-NEXT:    vmovdqu16 %ymm1, %ymm2 {%k1} # encoding: [0x62,0xf1,0xff,0x29,0x6f,0xd1]
65; CHECK-NEXT:    vmovdqa %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
66; CHECK-NEXT:    retq # encoding: [0xc3]
67  %z = call <16 x half> @llvm.maxnum.v16f16(<16 x half> %x, <16 x half> %y) readnone
68  ret <16 x half> %z
69}
70
71define <32 x half> @test_intrinsic_fmax_v32f16(<32 x half> %x, <32 x half> %y) {
72; CHECK-LABEL: test_intrinsic_fmax_v32f16:
73; CHECK:       # %bb.0:
74; CHECK-NEXT:    vmaxph %zmm0, %zmm1, %zmm2 # encoding: [0x62,0xf5,0x74,0x48,0x5f,0xd0]
75; CHECK-NEXT:    vcmpunordph %zmm0, %zmm0, %k1 # encoding: [0x62,0xf3,0x7c,0x48,0xc2,0xc8,0x03]
76; CHECK-NEXT:    vmovdqu16 %zmm1, %zmm2 {%k1} # encoding: [0x62,0xf1,0xff,0x49,0x6f,0xd1]
77; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
78; CHECK-NEXT:    retq # encoding: [0xc3]
79  %z = call <32 x half> @llvm.maxnum.v32f16(<32 x half> %x, <32 x half> %y) readnone
80  ret <32 x half> %z
81}
82
83define <4 x half> @maxnum_intrinsic_nnan_fmf_f432(<4 x half> %a, <4 x half> %b) {
84; CHECK-LABEL: maxnum_intrinsic_nnan_fmf_f432:
85; CHECK:       # %bb.0:
86; CHECK-NEXT:    vmaxph %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf5,0x7c,0x08,0x5f,0xc1]
87; CHECK-NEXT:    retq # encoding: [0xc3]
88  %r = tail call nnan <4 x half> @llvm.maxnum.v4f16(<4 x half> %a, <4 x half> %b)
89  ret <4 x half> %r
90}
91
92define half @maxnum_intrinsic_nnan_attr_f16(half %a, half %b) #0 {
93; CHECK-LABEL: maxnum_intrinsic_nnan_attr_f16:
94; CHECK:       # %bb.0:
95; CHECK-NEXT:    vmaxsh %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf5,0x7e,0x08,0x5f,0xc1]
96; CHECK-NEXT:    retq # encoding: [0xc3]
97  %r = tail call half @llvm.maxnum.f16(half %a, half %b)
98  ret half %r
99}
100
101define half @test_maxnum_const_op1(half %x) {
102; CHECK-LABEL: test_maxnum_const_op1:
103; CHECK:       # %bb.0:
104; CHECK-NEXT:    vmaxsh {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # encoding: [0x62,0xf5,0x7e,0x08,0x5f,0x05,A,A,A,A]
105; CHECK-NEXT:    # fixup A - offset: 6, value: {{\.?LCPI[0-9]+_[0-9]+}}-4, kind: reloc_riprel_4byte
106; CHECK-NEXT:    retq # encoding: [0xc3]
107  %r = call half @llvm.maxnum.f16(half 1.0, half %x)
108  ret half %r
109}
110
111define half @test_maxnum_const_op2(half %x) {
112; CHECK-LABEL: test_maxnum_const_op2:
113; CHECK:       # %bb.0:
114; CHECK-NEXT:    vmaxsh {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # encoding: [0x62,0xf5,0x7e,0x08,0x5f,0x05,A,A,A,A]
115; CHECK-NEXT:    # fixup A - offset: 6, value: {{\.?LCPI[0-9]+_[0-9]+}}-4, kind: reloc_riprel_4byte
116; CHECK-NEXT:    retq # encoding: [0xc3]
117  %r = call half @llvm.maxnum.f16(half %x, half 1.0)
118  ret half %r
119}
120
121define half @test_maxnum_const_nan(half %x) {
122; CHECK-LABEL: test_maxnum_const_nan:
123; CHECK:       # %bb.0:
124; CHECK-NEXT:    retq # encoding: [0xc3]
125  %r = call half @llvm.maxnum.f16(half %x, half 0x7fff000000000000)
126  ret half %r
127}
128
129attributes #0 = { "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true"}
130