1; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX9 %s
2; RUN: llc -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,CIVI,VI %s
3; RUN: llc -march=amdgcn -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,CIVI,CI %s
4
5; GCN-LABEL: {{^}}v_mad_mixhi_f16_f16lo_f16lo_f16lo_undeflo:
6; GFX9: s_waitcnt
7; GFX9-NEXT: v_mad_mixhi_f16 v0, v0, v1, v2
8; GFX9-NEXT: s_setpc_b64
9define <2 x half> @v_mad_mixhi_f16_f16lo_f16lo_f16lo_undeflo(half %src0, half %src1, half %src2) #0 {
10  %src0.ext = fpext half %src0 to float
11  %src1.ext = fpext half %src1 to float
12  %src2.ext = fpext half %src2 to float
13  %result = tail call float @llvm.fmuladd.f32(float %src0.ext, float %src1.ext, float %src2.ext)
14  %cvt.result = fptrunc float %result to half
15  %vec.result = insertelement <2 x half> undef, half %cvt.result, i32 1
16  ret <2 x half> %vec.result
17}
18
19; GCN-LABEL: {{^}}v_mad_mixhi_f16_f16lo_f16lo_f16lo_constlo:
20; GFX9: s_waitcnt
21; GFX9-NEXT: v_mov_b32_e32 v3, 0x3c00
22; GFX9-NEXT: v_mad_mixhi_f16 v3, v0, v1, v2
23; GFX9-NEXT: v_mov_b32_e32 v0, v3
24; GFX9-NEXT: s_setpc_b64
25define <2 x half> @v_mad_mixhi_f16_f16lo_f16lo_f16lo_constlo(half %src0, half %src1, half %src2) #0 {
26  %src0.ext = fpext half %src0 to float
27  %src1.ext = fpext half %src1 to float
28  %src2.ext = fpext half %src2 to float
29  %result = tail call float @llvm.fmuladd.f32(float %src0.ext, float %src1.ext, float %src2.ext)
30  %cvt.result = fptrunc float %result to half
31  %vec.result = insertelement <2 x half> <half 1.0, half undef>, half %cvt.result, i32 1
32  ret <2 x half> %vec.result
33}
34
35; GCN-LABEL: {{^}}v_mad_mixhi_f16_f16lo_f16lo_f16lo_reglo:
36; GFX9: s_waitcnt
37; GFX9-NEXT: v_mad_mixhi_f16 v3, v0, v1, v2
38; GFX9-NEXT: v_mov_b32_e32 v0, v3
39; GFX9-NEXT: s_setpc_b64
40define <2 x half> @v_mad_mixhi_f16_f16lo_f16lo_f16lo_reglo(half %src0, half %src1, half %src2, half %lo) #0 {
41  %src0.ext = fpext half %src0 to float
42  %src1.ext = fpext half %src1 to float
43  %src2.ext = fpext half %src2 to float
44  %result = tail call float @llvm.fmuladd.f32(float %src0.ext, float %src1.ext, float %src2.ext)
45  %cvt.result = fptrunc float %result to half
46  %vec = insertelement <2 x half> undef, half %lo, i32 0
47  %vec.result = insertelement <2 x half> %vec, half %cvt.result, i32 1
48  ret <2 x half> %vec.result
49}
50
51; GCN-LABEL: {{^}}v_mad_mixhi_f16_f16lo_f16lo_f16lo_intpack:
52; GFX9: s_waitcnt
53; GFX9-NEXT: v_mad_mixlo_f16 v0, v0, v1, v2 op_sel_hi:[1,1,1]
54; GFX9-NEXT: v_lshlrev_b32_e32 v0, 16, v0
55; GFX9-NEXT: s_setpc_b64
56define i32 @v_mad_mixhi_f16_f16lo_f16lo_f16lo_intpack(half %src0, half %src1, half %src2) #0 {
57  %src0.ext = fpext half %src0 to float
58  %src1.ext = fpext half %src1 to float
59  %src2.ext = fpext half %src2 to float
60  %result = tail call float @llvm.fmuladd.f32(float %src0.ext, float %src1.ext, float %src2.ext)
61  %cvt.result = fptrunc float %result to half
62  %bc = bitcast half %cvt.result to i16
63  %ext = zext i16 %bc to i32
64  %shr = shl i32 %ext, 16
65  ret i32 %shr
66}
67
68; GCN-LABEL: {{^}}v_mad_mixhi_f16_f16lo_f16lo_f16lo_intpack_sext:
69; GFX9: s_waitcnt
70; GFX9-NEXT: v_mad_mixlo_f16 v0, v0, v1, v2 op_sel_hi:[1,1,1]
71; GFX9-NEXT: v_lshlrev_b32_e32 v0, 16, v0
72; GFX9-NEXT: s_setpc_b64
73define i32 @v_mad_mixhi_f16_f16lo_f16lo_f16lo_intpack_sext(half %src0, half %src1, half %src2) #0 {
74  %src0.ext = fpext half %src0 to float
75  %src1.ext = fpext half %src1 to float
76  %src2.ext = fpext half %src2 to float
77  %result = tail call float @llvm.fmuladd.f32(float %src0.ext, float %src1.ext, float %src2.ext)
78  %cvt.result = fptrunc float %result to half
79  %bc = bitcast half %cvt.result to i16
80  %ext = sext i16 %bc to i32
81  %shr = shl i32 %ext, 16
82  ret i32 %shr
83}
84
85; GCN-LABEL: {{^}}v_mad_mixhi_f16_f16lo_f16lo_f16lo_undeflo_clamp_precvt:
86; GCN: s_waitcnt
87; GFX9-NEXT: v_mad_mix_f32 v0, v0, v1, v2 op_sel_hi:[1,1,1] clamp{{$}}
88; GFX9-NEXT: v_cvt_f16_f32_sdwa v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
89; GFX9-NEXT: s_setpc_b64
90define <2 x half> @v_mad_mixhi_f16_f16lo_f16lo_f16lo_undeflo_clamp_precvt(half %src0, half %src1, half %src2) #0 {
91  %src0.ext = fpext half %src0 to float
92  %src1.ext = fpext half %src1 to float
93  %src2.ext = fpext half %src2 to float
94  %result = tail call float @llvm.fmuladd.f32(float %src0.ext, float %src1.ext, float %src2.ext)
95  %max = call float @llvm.maxnum.f32(float %result, float 0.0)
96  %clamp = call float @llvm.minnum.f32(float %max, float 1.0)
97  %cvt.result = fptrunc float %clamp to half
98  %vec.result = insertelement <2 x half> undef, half %cvt.result, i32 1
99  ret <2 x half> %vec.result
100}
101
102; GCN-LABEL: {{^}}v_mad_mixhi_f16_f16lo_f16lo_f16lo_undeflo_clamp_postcvt:
103; GCN: s_waitcnt
104; GFX9-NEXT: v_mad_mixhi_f16 v0, v0, v1, v2 op_sel_hi:[1,1,1] clamp{{$}}
105; GFX9-NEXT: s_setpc_b64
106define <2 x half> @v_mad_mixhi_f16_f16lo_f16lo_f16lo_undeflo_clamp_postcvt(half %src0, half %src1, half %src2) #0 {
107  %src0.ext = fpext half %src0 to float
108  %src1.ext = fpext half %src1 to float
109  %src2.ext = fpext half %src2 to float
110  %result = tail call float @llvm.fmuladd.f32(float %src0.ext, float %src1.ext, float %src2.ext)
111  %cvt.result = fptrunc float %result to half
112  %max = call half @llvm.maxnum.f16(half %cvt.result, half 0.0)
113  %clamp = call half @llvm.minnum.f16(half %max, half 1.0)
114  %vec.result = insertelement <2 x half> undef, half %clamp, i32 1
115  ret <2 x half> %vec.result
116}
117
118
119; GCN-LABEL: {{^}}v_mad_mixhi_f16_f16lo_f16lo_f16lo_undeflo_clamp_postcvt_multi_use:
120; GCN: s_waitcnt
121; GFX9-NEXT: v_mad_mixlo_f16 v3, v0, v1, v2 op_sel_hi:[1,1,1]{{$}}
122; GFX9-NEXT: global_store_short v{{\[[0-9]+:[0-9]+\]}}, v3
123; GFX9-NEXT: v_mad_mixhi_f16 v0, v0, v1, v2 op_sel_hi:[1,1,1] clamp{{$}}
124; GFX9-NEXT: s_waitcnt vmcnt(0)
125; GFX9-NEXT: s_setpc_b64
126define <2 x half> @v_mad_mixhi_f16_f16lo_f16lo_f16lo_undeflo_clamp_postcvt_multi_use(half %src0, half %src1, half %src2) #0 {
127  %src0.ext = fpext half %src0 to float
128  %src1.ext = fpext half %src1 to float
129  %src2.ext = fpext half %src2 to float
130  %result = tail call float @llvm.fmuladd.f32(float %src0.ext, float %src1.ext, float %src2.ext)
131  %cvt.result = fptrunc float %result to half
132  store volatile half %cvt.result, half addrspace(1)* undef
133  %max = call half @llvm.maxnum.f16(half %cvt.result, half 0.0)
134  %clamp = call half @llvm.minnum.f16(half %max, half 1.0)
135  %vec.result = insertelement <2 x half> undef, half %clamp, i32 1
136  ret <2 x half> %vec.result
137}
138
139declare half @llvm.minnum.f16(half, half) #1
140declare half @llvm.maxnum.f16(half, half) #1
141declare float @llvm.minnum.f32(float, float) #1
142declare float @llvm.maxnum.f32(float, float) #1
143declare float @llvm.fmuladd.f32(float, float, float) #1
144declare <2 x float> @llvm.fmuladd.v2f32(<2 x float>, <2 x float>, <2 x float>) #1
145
146attributes #0 = { nounwind "denormal-fp-math-f32"="preserve-sign,preserve-sign" }
147attributes #1 = { nounwind readnone speculatable }
148