1; Test 32-bit floating-point comparison. 2; 3; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s 4 5declare float @foo() 6 7; Check comparison with registers. 8define i64 @f1(i64 %a, i64 %b, float %f1, float %f2) { 9; CHECK-LABEL: f1: 10; CHECK: cebr %f0, %f2 11; CHECK-NEXT: je 12; CHECK: lgr %r2, %r3 13; CHECK: br %r14 14 %cond = fcmp oeq float %f1, %f2 15 %res = select i1 %cond, i64 %a, i64 %b 16 ret i64 %res 17} 18 19; Check the low end of the CEB range. 20define i64 @f2(i64 %a, i64 %b, float %f1, float *%ptr) { 21; CHECK-LABEL: f2: 22; CHECK: ceb %f0, 0(%r4) 23; CHECK-NEXT: je 24; CHECK: lgr %r2, %r3 25; CHECK: br %r14 26 %f2 = load float *%ptr 27 %cond = fcmp oeq float %f1, %f2 28 %res = select i1 %cond, i64 %a, i64 %b 29 ret i64 %res 30} 31 32; Check the high end of the aligned CEB range. 33define i64 @f3(i64 %a, i64 %b, float %f1, float *%base) { 34; CHECK-LABEL: f3: 35; CHECK: ceb %f0, 4092(%r4) 36; CHECK-NEXT: je 37; CHECK: lgr %r2, %r3 38; CHECK: br %r14 39 %ptr = getelementptr float *%base, i64 1023 40 %f2 = load float *%ptr 41 %cond = fcmp oeq float %f1, %f2 42 %res = select i1 %cond, i64 %a, i64 %b 43 ret i64 %res 44} 45 46; Check the next word up, which needs separate address logic. 47; Other sequences besides this one would be OK. 48define i64 @f4(i64 %a, i64 %b, float %f1, float *%base) { 49; CHECK-LABEL: f4: 50; CHECK: aghi %r4, 4096 51; CHECK: ceb %f0, 0(%r4) 52; CHECK-NEXT: je 53; CHECK: lgr %r2, %r3 54; CHECK: br %r14 55 %ptr = getelementptr float *%base, i64 1024 56 %f2 = load float *%ptr 57 %cond = fcmp oeq float %f1, %f2 58 %res = select i1 %cond, i64 %a, i64 %b 59 ret i64 %res 60} 61 62; Check negative displacements, which also need separate address logic. 63define i64 @f5(i64 %a, i64 %b, float %f1, float *%base) { 64; CHECK-LABEL: f5: 65; CHECK: aghi %r4, -4 66; CHECK: ceb %f0, 0(%r4) 67; CHECK-NEXT: je 68; CHECK: lgr %r2, %r3 69; CHECK: br %r14 70 %ptr = getelementptr float *%base, i64 -1 71 %f2 = load float *%ptr 72 %cond = fcmp oeq float %f1, %f2 73 %res = select i1 %cond, i64 %a, i64 %b 74 ret i64 %res 75} 76 77; Check that CEB allows indices. 78define i64 @f6(i64 %a, i64 %b, float %f1, float *%base, i64 %index) { 79; CHECK-LABEL: f6: 80; CHECK: sllg %r1, %r5, 2 81; CHECK: ceb %f0, 400(%r1,%r4) 82; CHECK-NEXT: je 83; CHECK: lgr %r2, %r3 84; CHECK: br %r14 85 %ptr1 = getelementptr float *%base, i64 %index 86 %ptr2 = getelementptr float *%ptr1, i64 100 87 %f2 = load float *%ptr2 88 %cond = fcmp oeq float %f1, %f2 89 %res = select i1 %cond, i64 %a, i64 %b 90 ret i64 %res 91} 92 93; Check that comparisons of spilled values can use CEB rather than CEBR. 94define float @f7(float *%ptr0) { 95; CHECK-LABEL: f7: 96; CHECK: brasl %r14, foo@PLT 97; CHECK: ceb {{%f[0-9]+}}, 16{{[04]}}(%r15) 98; CHECK: br %r14 99 %ptr1 = getelementptr float *%ptr0, i64 2 100 %ptr2 = getelementptr float *%ptr0, i64 4 101 %ptr3 = getelementptr float *%ptr0, i64 6 102 %ptr4 = getelementptr float *%ptr0, i64 8 103 %ptr5 = getelementptr float *%ptr0, i64 10 104 %ptr6 = getelementptr float *%ptr0, i64 12 105 %ptr7 = getelementptr float *%ptr0, i64 14 106 %ptr8 = getelementptr float *%ptr0, i64 16 107 %ptr9 = getelementptr float *%ptr0, i64 18 108 %ptr10 = getelementptr float *%ptr0, i64 20 109 110 %val0 = load float *%ptr0 111 %val1 = load float *%ptr1 112 %val2 = load float *%ptr2 113 %val3 = load float *%ptr3 114 %val4 = load float *%ptr4 115 %val5 = load float *%ptr5 116 %val6 = load float *%ptr6 117 %val7 = load float *%ptr7 118 %val8 = load float *%ptr8 119 %val9 = load float *%ptr9 120 %val10 = load float *%ptr10 121 122 %ret = call float @foo() 123 124 %cmp0 = fcmp olt float %ret, %val0 125 %cmp1 = fcmp olt float %ret, %val1 126 %cmp2 = fcmp olt float %ret, %val2 127 %cmp3 = fcmp olt float %ret, %val3 128 %cmp4 = fcmp olt float %ret, %val4 129 %cmp5 = fcmp olt float %ret, %val5 130 %cmp6 = fcmp olt float %ret, %val6 131 %cmp7 = fcmp olt float %ret, %val7 132 %cmp8 = fcmp olt float %ret, %val8 133 %cmp9 = fcmp olt float %ret, %val9 134 %cmp10 = fcmp olt float %ret, %val10 135 136 %sel0 = select i1 %cmp0, float %ret, float 0.0 137 %sel1 = select i1 %cmp1, float %sel0, float 1.0 138 %sel2 = select i1 %cmp2, float %sel1, float 2.0 139 %sel3 = select i1 %cmp3, float %sel2, float 3.0 140 %sel4 = select i1 %cmp4, float %sel3, float 4.0 141 %sel5 = select i1 %cmp5, float %sel4, float 5.0 142 %sel6 = select i1 %cmp6, float %sel5, float 6.0 143 %sel7 = select i1 %cmp7, float %sel6, float 7.0 144 %sel8 = select i1 %cmp8, float %sel7, float 8.0 145 %sel9 = select i1 %cmp9, float %sel8, float 9.0 146 %sel10 = select i1 %cmp10, float %sel9, float 10.0 147 148 ret float %sel10 149} 150 151; Check comparison with zero. 152define i64 @f8(i64 %a, i64 %b, float %f) { 153; CHECK-LABEL: f8: 154; CHECK: ltebr %f0, %f0 155; CHECK-NEXT: je 156; CHECK: lgr %r2, %r3 157; CHECK: br %r14 158 %cond = fcmp oeq float %f, 0.0 159 %res = select i1 %cond, i64 %a, i64 %b 160 ret i64 %res 161} 162 163; Check the comparison can be reversed if that allows CEB to be used, 164; first with oeq. 165define i64 @f9(i64 %a, i64 %b, float %f2, float *%ptr) { 166; CHECK-LABEL: f9: 167; CHECK: ceb %f0, 0(%r4) 168; CHECK-NEXT: je {{\.L.*}} 169; CHECK: lgr %r2, %r3 170; CHECK: br %r14 171 %f1 = load float *%ptr 172 %cond = fcmp oeq float %f1, %f2 173 %res = select i1 %cond, i64 %a, i64 %b 174 ret i64 %res 175} 176 177; ...then one. 178define i64 @f10(i64 %a, i64 %b, float %f2, float *%ptr) { 179; CHECK-LABEL: f10: 180; CHECK: ceb %f0, 0(%r4) 181; CHECK-NEXT: jlh {{\.L.*}} 182; CHECK: lgr %r2, %r3 183; CHECK: br %r14 184 %f1 = load float *%ptr 185 %cond = fcmp one float %f1, %f2 186 %res = select i1 %cond, i64 %a, i64 %b 187 ret i64 %res 188} 189 190; ...then olt. 191define i64 @f11(i64 %a, i64 %b, float %f2, float *%ptr) { 192; CHECK-LABEL: f11: 193; CHECK: ceb %f0, 0(%r4) 194; CHECK-NEXT: jh {{\.L.*}} 195; CHECK: lgr %r2, %r3 196; CHECK: br %r14 197 %f1 = load float *%ptr 198 %cond = fcmp olt float %f1, %f2 199 %res = select i1 %cond, i64 %a, i64 %b 200 ret i64 %res 201} 202 203; ...then ole. 204define i64 @f12(i64 %a, i64 %b, float %f2, float *%ptr) { 205; CHECK-LABEL: f12: 206; CHECK: ceb %f0, 0(%r4) 207; CHECK-NEXT: jhe {{\.L.*}} 208; CHECK: lgr %r2, %r3 209; CHECK: br %r14 210 %f1 = load float *%ptr 211 %cond = fcmp ole float %f1, %f2 212 %res = select i1 %cond, i64 %a, i64 %b 213 ret i64 %res 214} 215 216; ...then oge. 217define i64 @f13(i64 %a, i64 %b, float %f2, float *%ptr) { 218; CHECK-LABEL: f13: 219; CHECK: ceb %f0, 0(%r4) 220; CHECK-NEXT: jle {{\.L.*}} 221; CHECK: lgr %r2, %r3 222; CHECK: br %r14 223 %f1 = load float *%ptr 224 %cond = fcmp oge float %f1, %f2 225 %res = select i1 %cond, i64 %a, i64 %b 226 ret i64 %res 227} 228 229; ...then ogt. 230define i64 @f14(i64 %a, i64 %b, float %f2, float *%ptr) { 231; CHECK-LABEL: f14: 232; CHECK: ceb %f0, 0(%r4) 233; CHECK-NEXT: jl {{\.L.*}} 234; CHECK: lgr %r2, %r3 235; CHECK: br %r14 236 %f1 = load float *%ptr 237 %cond = fcmp ogt float %f1, %f2 238 %res = select i1 %cond, i64 %a, i64 %b 239 ret i64 %res 240} 241 242; ...then ueq. 243define i64 @f15(i64 %a, i64 %b, float %f2, float *%ptr) { 244; CHECK-LABEL: f15: 245; CHECK: ceb %f0, 0(%r4) 246; CHECK-NEXT: jnlh {{\.L.*}} 247; CHECK: lgr %r2, %r3 248; CHECK: br %r14 249 %f1 = load float *%ptr 250 %cond = fcmp ueq float %f1, %f2 251 %res = select i1 %cond, i64 %a, i64 %b 252 ret i64 %res 253} 254 255; ...then une. 256define i64 @f16(i64 %a, i64 %b, float %f2, float *%ptr) { 257; CHECK-LABEL: f16: 258; CHECK: ceb %f0, 0(%r4) 259; CHECK-NEXT: jne {{\.L.*}} 260; CHECK: lgr %r2, %r3 261; CHECK: br %r14 262 %f1 = load float *%ptr 263 %cond = fcmp une float %f1, %f2 264 %res = select i1 %cond, i64 %a, i64 %b 265 ret i64 %res 266} 267 268; ...then ult. 269define i64 @f17(i64 %a, i64 %b, float %f2, float *%ptr) { 270; CHECK-LABEL: f17: 271; CHECK: ceb %f0, 0(%r4) 272; CHECK-NEXT: jnle {{\.L.*}} 273; CHECK: lgr %r2, %r3 274; CHECK: br %r14 275 %f1 = load float *%ptr 276 %cond = fcmp ult float %f1, %f2 277 %res = select i1 %cond, i64 %a, i64 %b 278 ret i64 %res 279} 280 281; ...then ule. 282define i64 @f18(i64 %a, i64 %b, float %f2, float *%ptr) { 283; CHECK-LABEL: f18: 284; CHECK: ceb %f0, 0(%r4) 285; CHECK-NEXT: jnl {{\.L.*}} 286; CHECK: lgr %r2, %r3 287; CHECK: br %r14 288 %f1 = load float *%ptr 289 %cond = fcmp ule float %f1, %f2 290 %res = select i1 %cond, i64 %a, i64 %b 291 ret i64 %res 292} 293 294; ...then uge. 295define i64 @f19(i64 %a, i64 %b, float %f2, float *%ptr) { 296; CHECK-LABEL: f19: 297; CHECK: ceb %f0, 0(%r4) 298; CHECK-NEXT: jnh {{\.L.*}} 299; CHECK: lgr %r2, %r3 300; CHECK: br %r14 301 %f1 = load float *%ptr 302 %cond = fcmp uge float %f1, %f2 303 %res = select i1 %cond, i64 %a, i64 %b 304 ret i64 %res 305} 306 307; ...then ugt. 308define i64 @f20(i64 %a, i64 %b, float %f2, float *%ptr) { 309; CHECK-LABEL: f20: 310; CHECK: ceb %f0, 0(%r4) 311; CHECK-NEXT: jnhe {{\.L.*}} 312; CHECK: lgr %r2, %r3 313; CHECK: br %r14 314 %f1 = load float *%ptr 315 %cond = fcmp ugt float %f1, %f2 316 %res = select i1 %cond, i64 %a, i64 %b 317 ret i64 %res 318} 319