1; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,SI %s 2; RUN: llc -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,VI %s 3 4declare i64 @llvm.amdgcn.icmp.i32(i32, i32, i32) #0 5declare i64 @llvm.amdgcn.icmp.i64(i64, i64, i32) #0 6declare i64 @llvm.amdgcn.icmp.i16(i16, i16, i32) #0 7declare i64 @llvm.amdgcn.icmp.i1(i1, i1, i32) #0 8 9; GCN-LABEL: {{^}}v_icmp_i32_eq: 10; GCN: v_cmp_eq_u32_e64 11define amdgpu_kernel void @v_icmp_i32_eq(i64 addrspace(1)* %out, i32 %src) { 12 %result = call i64 @llvm.amdgcn.icmp.i32(i32 %src, i32 100, i32 32) 13 store i64 %result, i64 addrspace(1)* %out 14 ret void 15} 16 17; GCN-LABEL: {{^}}v_icmp_i32: 18; GCN-NOT: v_cmp_eq_u32_e64 19define amdgpu_kernel void @v_icmp_i32(i64 addrspace(1)* %out, i32 %src) { 20 %result = call i64 @llvm.amdgcn.icmp.i32(i32 %src, i32 100, i32 30) 21 store i64 %result, i64 addrspace(1)* %out 22 ret void 23} 24 25; GCN-LABEL: {{^}}v_icmp_i32_ne: 26; GCN: v_cmp_ne_u32_e64 27define amdgpu_kernel void @v_icmp_i32_ne(i64 addrspace(1)* %out, i32 %src) { 28 %result = call i64 @llvm.amdgcn.icmp.i32(i32 %src, i32 100, i32 33) 29 store i64 %result, i64 addrspace(1)* %out 30 ret void 31} 32 33; GCN-LABEL: {{^}}v_icmp_i32_ugt: 34; GCN: v_cmp_gt_u32_e64 35define amdgpu_kernel void @v_icmp_i32_ugt(i64 addrspace(1)* %out, i32 %src) { 36 %result = call i64 @llvm.amdgcn.icmp.i32(i32 %src, i32 100, i32 34) 37 store i64 %result, i64 addrspace(1)* %out 38 ret void 39} 40 41; GCN-LABEL: {{^}}v_icmp_i32_uge: 42; GCN: v_cmp_ge_u32_e64 43define amdgpu_kernel void @v_icmp_i32_uge(i64 addrspace(1)* %out, i32 %src) { 44 %result = call i64 @llvm.amdgcn.icmp.i32(i32 %src, i32 100, i32 35) 45 store i64 %result, i64 addrspace(1)* %out 46 ret void 47} 48 49; GCN-LABEL: {{^}}v_icmp_i32_ult: 50; GCN: v_cmp_lt_u32_e64 51define amdgpu_kernel void @v_icmp_i32_ult(i64 addrspace(1)* %out, i32 %src) { 52 %result = call i64 @llvm.amdgcn.icmp.i32(i32 %src, i32 100, i32 36) 53 store i64 %result, i64 addrspace(1)* %out 54 ret void 55} 56 57; GCN-LABEL: {{^}}v_icmp_i32_ule: 58; GCN: v_cmp_le_u32_e64 59define amdgpu_kernel void @v_icmp_i32_ule(i64 addrspace(1)* %out, i32 %src) { 60 %result = call i64 @llvm.amdgcn.icmp.i32(i32 %src, i32 100, i32 37) 61 store i64 %result, i64 addrspace(1)* %out 62 ret void 63} 64 65; GCN-LABEL: {{^}}v_icmp_i32_sgt: 66; GCN: v_cmp_gt_i32_e64 67define amdgpu_kernel void @v_icmp_i32_sgt(i64 addrspace(1)* %out, i32 %src) #1 { 68 %result = call i64 @llvm.amdgcn.icmp.i32(i32 %src, i32 100, i32 38) 69 store i64 %result, i64 addrspace(1)* %out 70 ret void 71} 72 73; GCN-LABEL: {{^}}v_icmp_i32_sge: 74; GCN: v_cmp_ge_i32_e64 75define amdgpu_kernel void @v_icmp_i32_sge(i64 addrspace(1)* %out, i32 %src) { 76 %result = call i64 @llvm.amdgcn.icmp.i32(i32 %src, i32 100, i32 39) 77 store i64 %result, i64 addrspace(1)* %out 78 ret void 79} 80 81; GCN-LABEL: {{^}}v_icmp_i32_slt: 82; GCN: v_cmp_lt_i32_e64 83define amdgpu_kernel void @v_icmp_i32_slt(i64 addrspace(1)* %out, i32 %src) { 84 %result = call i64 @llvm.amdgcn.icmp.i32(i32 %src, i32 100, i32 40) 85 store i64 %result, i64 addrspace(1)* %out 86 ret void 87} 88; GCN-LABEL: {{^}}v_icmp_i32_sle: 89; GCN: v_cmp_le_i32_e64 90define amdgpu_kernel void @v_icmp_i32_sle(i64 addrspace(1)* %out, i32 %src) { 91 %result = call i64 @llvm.amdgcn.icmp.i32(i32 %src, i32 100, i32 41) 92 store i64 %result, i64 addrspace(1)* %out 93 ret void 94} 95 96; GCN-LABEL: {{^}}v_icmp_i64_eq: 97; GCN: v_cmp_eq_u64_e64 98define amdgpu_kernel void @v_icmp_i64_eq(i64 addrspace(1)* %out, i64 %src) { 99 %result = call i64 @llvm.amdgcn.icmp.i64(i64 %src, i64 100, i32 32) 100 store i64 %result, i64 addrspace(1)* %out 101 ret void 102} 103 104; GCN-LABEL: {{^}}v_icmp_i64_ne: 105; GCN: v_cmp_ne_u64_e64 106define amdgpu_kernel void @v_icmp_i64_ne(i64 addrspace(1)* %out, i64 %src) { 107 %result = call i64 @llvm.amdgcn.icmp.i64(i64 %src, i64 100, i32 33) 108 store i64 %result, i64 addrspace(1)* %out 109 ret void 110} 111 112; GCN-LABEL: {{^}}v_icmp_u64_ugt: 113; GCN: v_cmp_gt_u64_e64 114define amdgpu_kernel void @v_icmp_u64_ugt(i64 addrspace(1)* %out, i64 %src) { 115 %result = call i64 @llvm.amdgcn.icmp.i64(i64 %src, i64 100, i32 34) 116 store i64 %result, i64 addrspace(1)* %out 117 ret void 118} 119 120; GCN-LABEL: {{^}}v_icmp_u64_uge: 121; GCN: v_cmp_ge_u64_e64 122define amdgpu_kernel void @v_icmp_u64_uge(i64 addrspace(1)* %out, i64 %src) { 123 %result = call i64 @llvm.amdgcn.icmp.i64(i64 %src, i64 100, i32 35) 124 store i64 %result, i64 addrspace(1)* %out 125 ret void 126} 127 128; GCN-LABEL: {{^}}v_icmp_u64_ult: 129; GCN: v_cmp_lt_u64_e64 130define amdgpu_kernel void @v_icmp_u64_ult(i64 addrspace(1)* %out, i64 %src) { 131 %result = call i64 @llvm.amdgcn.icmp.i64(i64 %src, i64 100, i32 36) 132 store i64 %result, i64 addrspace(1)* %out 133 ret void 134} 135 136; GCN-LABEL: {{^}}v_icmp_u64_ule: 137; GCN: v_cmp_le_u64_e64 138define amdgpu_kernel void @v_icmp_u64_ule(i64 addrspace(1)* %out, i64 %src) { 139 %result = call i64 @llvm.amdgcn.icmp.i64(i64 %src, i64 100, i32 37) 140 store i64 %result, i64 addrspace(1)* %out 141 ret void 142} 143 144; GCN-LABEL: {{^}}v_icmp_i64_sgt: 145; GCN: v_cmp_gt_i64_e64 146define amdgpu_kernel void @v_icmp_i64_sgt(i64 addrspace(1)* %out, i64 %src) { 147 %result = call i64 @llvm.amdgcn.icmp.i64(i64 %src, i64 100, i32 38) 148 store i64 %result, i64 addrspace(1)* %out 149 ret void 150} 151 152; GCN-LABEL: {{^}}v_icmp_i64_sge: 153; GCN: v_cmp_ge_i64_e64 154define amdgpu_kernel void @v_icmp_i64_sge(i64 addrspace(1)* %out, i64 %src) { 155 %result = call i64 @llvm.amdgcn.icmp.i64(i64 %src, i64 100, i32 39) 156 store i64 %result, i64 addrspace(1)* %out 157 ret void 158} 159 160; GCN-LABEL: {{^}}v_icmp_i64_slt: 161; GCN: v_cmp_lt_i64_e64 162define amdgpu_kernel void @v_icmp_i64_slt(i64 addrspace(1)* %out, i64 %src) { 163 %result = call i64 @llvm.amdgcn.icmp.i64(i64 %src, i64 100, i32 40) 164 store i64 %result, i64 addrspace(1)* %out 165 ret void 166} 167; GCN-LABEL: {{^}}v_icmp_i64_sle: 168; GCN: v_cmp_le_i64_e64 169define amdgpu_kernel void @v_icmp_i64_sle(i64 addrspace(1)* %out, i64 %src) { 170 %result = call i64 @llvm.amdgcn.icmp.i64(i64 %src, i64 100, i32 41) 171 store i64 %result, i64 addrspace(1)* %out 172 ret void 173} 174 175; VI: v_cmp_eq_u16_e64 176 177; SI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x64 178; SI-DAG: s_and_b32 [[CVT:s[0-9]+]], s{{[0-9]+}}, 0xffff{{$}} 179; SI: v_cmp_eq_u32_e64 s{{\[[0-9]+:[0-9]+\]}}, [[CVT]], [[K]] 180define amdgpu_kernel void @v_icmp_i16_eq(i64 addrspace(1)* %out, i16 %src) { 181 %result = call i64 @llvm.amdgcn.icmp.i16(i16 %src, i16 100, i32 32) 182 store i64 %result, i64 addrspace(1)* %out 183 ret void 184} 185 186; GCN-LABEL: {{^}}v_icmp_i16: 187; GCN-NOT: v_cmp_eq_ 188define amdgpu_kernel void @v_icmp_i16(i64 addrspace(1)* %out, i16 %src) { 189 %result = call i64 @llvm.amdgcn.icmp.i16(i16 %src, i16 100, i32 30) 190 store i64 %result, i64 addrspace(1)* %out 191 ret void 192} 193; GCN-LABEL: {{^}}v_icmp_i16_ne: 194; VI: v_cmp_ne_u16_e64 195 196; SI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x64 197; SI-DAG: s_and_b32 [[CVT:s[0-9]+]], s{{[0-9]+}}, 0xffff{{$}} 198; SI: v_cmp_ne_u32_e64 s{{\[[0-9]+:[0-9]+\]}}, [[CVT]], [[K]] 199define amdgpu_kernel void @v_icmp_i16_ne(i64 addrspace(1)* %out, i16 %src) { 200 %result = call i64 @llvm.amdgcn.icmp.i16(i16 %src, i16 100, i32 33) 201 store i64 %result, i64 addrspace(1)* %out 202 ret void 203} 204 205; GCN-LABEL: {{^}}v_icmp_i16_ugt: 206; VI: v_cmp_gt_u16_e64 207 208; SI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x64 209; SI-DAG: s_and_b32 [[CVT:s[0-9]+]], s{{[0-9]+}}, 0xffff{{$}} 210; SI: v_cmp_gt_u32_e64 s{{\[[0-9]+:[0-9]+\]}}, [[CVT]], [[K]] 211define amdgpu_kernel void @v_icmp_i16_ugt(i64 addrspace(1)* %out, i16 %src) { 212 %result = call i64 @llvm.amdgcn.icmp.i16(i16 %src, i16 100, i32 34) 213 store i64 %result, i64 addrspace(1)* %out 214 ret void 215} 216 217; GCN-LABEL: {{^}}v_icmp_i16_uge: 218; VI: v_cmp_ge_u16_e64 219 220; SI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x64 221; SI-DAG: s_and_b32 [[CVT:s[0-9]+]], s{{[0-9]+}}, 0xffff{{$}} 222; SI: v_cmp_ge_u32_e64 s{{\[[0-9]+:[0-9]+\]}}, [[CVT]], [[K]] 223define amdgpu_kernel void @v_icmp_i16_uge(i64 addrspace(1)* %out, i16 %src) { 224 %result = call i64 @llvm.amdgcn.icmp.i16(i16 %src, i16 100, i32 35) 225 store i64 %result, i64 addrspace(1)* %out 226 ret void 227} 228 229; GCN-LABEL: {{^}}v_icmp_i16_ult: 230; VI: v_cmp_lt_u16_e64 231 232; SI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x64 233; SI-DAG: s_and_b32 [[CVT:s[0-9]+]], s{{[0-9]+}}, 0xffff{{$}} 234; SI: v_cmp_lt_u32_e64 s{{\[[0-9]+:[0-9]+\]}}, [[CVT]], [[K]] 235define amdgpu_kernel void @v_icmp_i16_ult(i64 addrspace(1)* %out, i16 %src) { 236 %result = call i64 @llvm.amdgcn.icmp.i16(i16 %src, i16 100, i32 36) 237 store i64 %result, i64 addrspace(1)* %out 238 ret void 239} 240 241; GCN-LABEL: {{^}}v_icmp_i16_ule: 242; VI: v_cmp_le_u16_e64 243 244; SI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x64 245; SI-DAG: s_and_b32 [[CVT:s[0-9]+]], s{{[0-9]+}}, 0xffff{{$}} 246; SI: v_cmp_le_u32_e64 s{{\[[0-9]+:[0-9]+\]}}, [[CVT]], [[K]] 247define amdgpu_kernel void @v_icmp_i16_ule(i64 addrspace(1)* %out, i16 %src) { 248 %result = call i64 @llvm.amdgcn.icmp.i16(i16 %src, i16 100, i32 37) 249 store i64 %result, i64 addrspace(1)* %out 250 ret void 251} 252 253; GCN-LABEL: {{^}}v_icmp_i16_sgt: 254; VI: v_cmp_gt_i16_e64 255 256; SI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x64 257; SI-DAG: s_sext_i32_i16 [[CVT:s[0-9]+]], s{{[0-9]+}} 258; SI: v_cmp_gt_i32_e64 s{{\[[0-9]+:[0-9]+\]}}, [[CVT]], [[K]] 259define amdgpu_kernel void @v_icmp_i16_sgt(i64 addrspace(1)* %out, i16 %src) #1 { 260 %result = call i64 @llvm.amdgcn.icmp.i16(i16 %src, i16 100, i32 38) 261 store i64 %result, i64 addrspace(1)* %out 262 ret void 263} 264 265; GCN-LABEL: {{^}}v_icmp_i16_sge: 266; VI: v_cmp_ge_i16_e64 267 268; SI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x64 269; SI-DAG: s_sext_i32_i16 [[CVT:s[0-9]+]], s{{[0-9]+}} 270; SI: v_cmp_ge_i32_e64 s{{\[[0-9]+:[0-9]+\]}}, [[CVT]], [[K]] 271define amdgpu_kernel void @v_icmp_i16_sge(i64 addrspace(1)* %out, i16 %src) { 272 %result = call i64 @llvm.amdgcn.icmp.i16(i16 %src, i16 100, i32 39) 273 store i64 %result, i64 addrspace(1)* %out 274 ret void 275} 276 277; GCN-LABEL: {{^}}v_icmp_i16_slt: 278; VI: v_cmp_lt_i16_e64 279 280; SI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x64 281; SI-DAG: s_sext_i32_i16 [[CVT:s[0-9]+]], s{{[0-9]+}} 282; SI: v_cmp_lt_i32_e64 s{{\[[0-9]+:[0-9]+\]}}, [[CVT]], [[K]] 283define amdgpu_kernel void @v_icmp_i16_slt(i64 addrspace(1)* %out, i16 %src) { 284 %result = call i64 @llvm.amdgcn.icmp.i16(i16 %src, i16 100, i32 40) 285 store i64 %result, i64 addrspace(1)* %out 286 ret void 287} 288; GCN-LABEL: {{^}}v_icmp_i16_sle: 289; VI: v_cmp_le_i16_e64 290 291; SI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x64 292; SI-DAG: s_sext_i32_i16 [[CVT:s[0-9]+]], s{{[0-9]+}} 293; SI: v_cmp_le_i32_e64 s{{\[[0-9]+:[0-9]+\]}}, [[CVT]], [[K]] 294define amdgpu_kernel void @v_icmp_i16_sle(i64 addrspace(1)* %out, i16 %src) { 295 %result = call i64 @llvm.amdgcn.icmp.i16(i16 %src, i16 100, i32 41) 296 store i64 %result, i64 addrspace(1)* %out 297 ret void 298} 299 300; GCN-LABEL: {{^}}v_icmp_i1_ne0: 301; GCN: v_cmp_gt_u32_e64 s[[C0:\[[0-9]+:[0-9]+\]]], 302; GCN: v_cmp_gt_u32_e64 s[[C1:\[[0-9]+:[0-9]+\]]], 303; GCN: s_and_b64 s[[SRC:\[[0-9]+:[0-9]+\]]], s[[C0]], s[[C1]] 304; SI-NEXT: s_mov_b32 s{{[0-9]+}}, -1 305; GCN-NEXT: v_mov_b32_e32 306; GCN-NEXT: v_mov_b32_e32 307; GCN: {{global|flat|buffer}}_store_dwordx2 308define amdgpu_kernel void @v_icmp_i1_ne0(i64 addrspace(1)* %out, i32 %a, i32 %b) { 309 %c0 = icmp ugt i32 %a, 1 310 %c1 = icmp ugt i32 %b, 2 311 %src = and i1 %c0, %c1 312 %result = call i64 @llvm.amdgcn.icmp.i1(i1 %src, i1 false, i32 33) 313 store i64 %result, i64 addrspace(1)* %out 314 ret void 315} 316 317attributes #0 = { nounwind readnone convergent } 318