1; Test 32-bit addition in which the second operand is variable. 2; 3; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s 4 5declare i32 @foo() 6 7; Check ALR. 8define zeroext i1 @f1(i32 %dummy, i32 %a, i32 %b, i32 *%res) { 9; CHECK-LABEL: f1: 10; CHECK: alr %r3, %r4 11; CHECK-DAG: st %r3, 0(%r5) 12; CHECK-DAG: ipm [[REG:%r[0-5]]] 13; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 35 14; CHECK: br %r14 15 %t = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b) 16 %val = extractvalue {i32, i1} %t, 0 17 %obit = extractvalue {i32, i1} %t, 1 18 store i32 %val, i32 *%res 19 ret i1 %obit 20} 21 22; Check using the overflow result for a branch. 23define void @f2(i32 %dummy, i32 %a, i32 %b, i32 *%res) { 24; CHECK-LABEL: f2: 25; CHECK: alr %r3, %r4 26; CHECK: st %r3, 0(%r5) 27; CHECK: jgnle foo@PLT 28; CHECK: br %r14 29 %t = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b) 30 %val = extractvalue {i32, i1} %t, 0 31 %obit = extractvalue {i32, i1} %t, 1 32 store i32 %val, i32 *%res 33 br i1 %obit, label %call, label %exit 34 35call: 36 tail call i32 @foo() 37 br label %exit 38 39exit: 40 ret void 41} 42 43; ... and the same with the inverted direction. 44define void @f3(i32 %dummy, i32 %a, i32 %b, i32 *%res) { 45; CHECK-LABEL: f3: 46; CHECK: alr %r3, %r4 47; CHECK: st %r3, 0(%r5) 48; CHECK: jgle foo@PLT 49; CHECK: br %r14 50 %t = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b) 51 %val = extractvalue {i32, i1} %t, 0 52 %obit = extractvalue {i32, i1} %t, 1 53 store i32 %val, i32 *%res 54 br i1 %obit, label %exit, label %call 55 56call: 57 tail call i32 @foo() 58 br label %exit 59 60exit: 61 ret void 62} 63 64; Check the low end of the AL range. 65define zeroext i1 @f4(i32 %dummy, i32 %a, i32 *%src, i32 *%res) { 66; CHECK-LABEL: f4: 67; CHECK: al %r3, 0(%r4) 68; CHECK-DAG: st %r3, 0(%r5) 69; CHECK-DAG: ipm [[REG:%r[0-5]]] 70; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 35 71; CHECK: br %r14 72 %b = load i32, i32 *%src 73 %t = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b) 74 %val = extractvalue {i32, i1} %t, 0 75 %obit = extractvalue {i32, i1} %t, 1 76 store i32 %val, i32 *%res 77 ret i1 %obit 78} 79 80; Check the high end of the aligned AL range. 81define zeroext i1 @f5(i32 %dummy, i32 %a, i32 *%src, i32 *%res) { 82; CHECK-LABEL: f5: 83; CHECK: al %r3, 4092(%r4) 84; CHECK-DAG: st %r3, 0(%r5) 85; CHECK-DAG: ipm [[REG:%r[0-5]]] 86; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 35 87; CHECK: br %r14 88 %ptr = getelementptr i32, i32 *%src, i64 1023 89 %b = load i32, i32 *%ptr 90 %t = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b) 91 %val = extractvalue {i32, i1} %t, 0 92 %obit = extractvalue {i32, i1} %t, 1 93 store i32 %val, i32 *%res 94 ret i1 %obit 95} 96 97; Check the next word up, which should use ALY instead of AL. 98define zeroext i1 @f6(i32 %dummy, i32 %a, i32 *%src, i32 *%res) { 99; CHECK-LABEL: f6: 100; CHECK: aly %r3, 4096(%r4) 101; CHECK-DAG: st %r3, 0(%r5) 102; CHECK-DAG: ipm [[REG:%r[0-5]]] 103; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 35 104; CHECK: br %r14 105 %ptr = getelementptr i32, i32 *%src, i64 1024 106 %b = load i32, i32 *%ptr 107 %t = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b) 108 %val = extractvalue {i32, i1} %t, 0 109 %obit = extractvalue {i32, i1} %t, 1 110 store i32 %val, i32 *%res 111 ret i1 %obit 112} 113 114; Check the high end of the aligned ALY range. 115define zeroext i1 @f7(i32 %dummy, i32 %a, i32 *%src, i32 *%res) { 116; CHECK-LABEL: f7: 117; CHECK: aly %r3, 524284(%r4) 118; CHECK-DAG: st %r3, 0(%r5) 119; CHECK-DAG: ipm [[REG:%r[0-5]]] 120; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 35 121; CHECK: br %r14 122 %ptr = getelementptr i32, i32 *%src, i64 131071 123 %b = load i32, i32 *%ptr 124 %t = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b) 125 %val = extractvalue {i32, i1} %t, 0 126 %obit = extractvalue {i32, i1} %t, 1 127 store i32 %val, i32 *%res 128 ret i1 %obit 129} 130 131; Check the next word up, which needs separate address logic. 132; Other sequences besides this one would be OK. 133define zeroext i1 @f8(i32 %dummy, i32 %a, i32 *%src, i32 *%res) { 134; CHECK-LABEL: f8: 135; CHECK: agfi %r4, 524288 136; CHECK: al %r3, 0(%r4) 137; CHECK-DAG: st %r3, 0(%r5) 138; CHECK-DAG: ipm [[REG:%r[0-5]]] 139; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 35 140; CHECK: br %r14 141 %ptr = getelementptr i32, i32 *%src, i64 131072 142 %b = load i32, i32 *%ptr 143 %t = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b) 144 %val = extractvalue {i32, i1} %t, 0 145 %obit = extractvalue {i32, i1} %t, 1 146 store i32 %val, i32 *%res 147 ret i1 %obit 148} 149 150; Check the high end of the negative aligned ALY range. 151define zeroext i1 @f9(i32 %dummy, i32 %a, i32 *%src, i32 *%res) { 152; CHECK-LABEL: f9: 153; CHECK: aly %r3, -4(%r4) 154; CHECK-DAG: st %r3, 0(%r5) 155; CHECK-DAG: ipm [[REG:%r[0-5]]] 156; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 35 157; CHECK: br %r14 158 %ptr = getelementptr i32, i32 *%src, i64 -1 159 %b = load i32, i32 *%ptr 160 %t = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b) 161 %val = extractvalue {i32, i1} %t, 0 162 %obit = extractvalue {i32, i1} %t, 1 163 store i32 %val, i32 *%res 164 ret i1 %obit 165} 166 167; Check the low end of the ALY range. 168define zeroext i1 @f10(i32 %dummy, i32 %a, i32 *%src, i32 *%res) { 169; CHECK-LABEL: f10: 170; CHECK: aly %r3, -524288(%r4) 171; CHECK-DAG: st %r3, 0(%r5) 172; CHECK-DAG: ipm [[REG:%r[0-5]]] 173; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 35 174; CHECK: br %r14 175 %ptr = getelementptr i32, i32 *%src, i64 -131072 176 %b = load i32, i32 *%ptr 177 %t = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b) 178 %val = extractvalue {i32, i1} %t, 0 179 %obit = extractvalue {i32, i1} %t, 1 180 store i32 %val, i32 *%res 181 ret i1 %obit 182} 183 184; Check the next word down, which needs separate address logic. 185; Other sequences besides this one would be OK. 186define zeroext i1 @f11(i32 %dummy, i32 %a, i32 *%src, i32 *%res) { 187; CHECK-LABEL: f11: 188; CHECK: agfi %r4, -524292 189; CHECK: al %r3, 0(%r4) 190; CHECK-DAG: st %r3, 0(%r5) 191; CHECK-DAG: ipm [[REG:%r[0-5]]] 192; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 35 193; CHECK: br %r14 194 %ptr = getelementptr i32, i32 *%src, i64 -131073 195 %b = load i32, i32 *%ptr 196 %t = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b) 197 %val = extractvalue {i32, i1} %t, 0 198 %obit = extractvalue {i32, i1} %t, 1 199 store i32 %val, i32 *%res 200 ret i1 %obit 201} 202 203; Check that AL allows an index. 204define zeroext i1 @f12(i64 %src, i64 %index, i32 %a, i32 *%res) { 205; CHECK-LABEL: f12: 206; CHECK: al %r4, 4092({{%r3,%r2|%r2,%r3}}) 207; CHECK-DAG: st %r4, 0(%r5) 208; CHECK-DAG: ipm [[REG:%r[0-5]]] 209; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 35 210; CHECK: br %r14 211 %add1 = add i64 %src, %index 212 %add2 = add i64 %add1, 4092 213 %ptr = inttoptr i64 %add2 to i32 * 214 %b = load i32, i32 *%ptr 215 %t = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b) 216 %val = extractvalue {i32, i1} %t, 0 217 %obit = extractvalue {i32, i1} %t, 1 218 store i32 %val, i32 *%res 219 ret i1 %obit 220} 221 222; Check that ALY allows an index. 223define zeroext i1 @f13(i64 %src, i64 %index, i32 %a, i32 *%res) { 224; CHECK-LABEL: f13: 225; CHECK: aly %r4, 4096({{%r3,%r2|%r2,%r3}}) 226; CHECK-DAG: st %r4, 0(%r5) 227; CHECK-DAG: ipm [[REG:%r[0-5]]] 228; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 35 229; CHECK: br %r14 230 %add1 = add i64 %src, %index 231 %add2 = add i64 %add1, 4096 232 %ptr = inttoptr i64 %add2 to i32 * 233 %b = load i32, i32 *%ptr 234 %t = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b) 235 %val = extractvalue {i32, i1} %t, 0 236 %obit = extractvalue {i32, i1} %t, 1 237 store i32 %val, i32 *%res 238 ret i1 %obit 239} 240 241; Check that additions of spilled values can use AL rather than ALR. 242define zeroext i1 @f14(i32 *%ptr0) { 243; CHECK-LABEL: f14: 244; CHECK: brasl %r14, foo@PLT 245; CHECK: al %r2, 16{{[04]}}(%r15) 246; CHECK: br %r14 247 %ptr1 = getelementptr i32, i32 *%ptr0, i64 2 248 %ptr2 = getelementptr i32, i32 *%ptr0, i64 4 249 %ptr3 = getelementptr i32, i32 *%ptr0, i64 6 250 %ptr4 = getelementptr i32, i32 *%ptr0, i64 8 251 %ptr5 = getelementptr i32, i32 *%ptr0, i64 10 252 %ptr6 = getelementptr i32, i32 *%ptr0, i64 12 253 %ptr7 = getelementptr i32, i32 *%ptr0, i64 14 254 %ptr8 = getelementptr i32, i32 *%ptr0, i64 16 255 %ptr9 = getelementptr i32, i32 *%ptr0, i64 18 256 257 %val0 = load i32, i32 *%ptr0 258 %val1 = load i32, i32 *%ptr1 259 %val2 = load i32, i32 *%ptr2 260 %val3 = load i32, i32 *%ptr3 261 %val4 = load i32, i32 *%ptr4 262 %val5 = load i32, i32 *%ptr5 263 %val6 = load i32, i32 *%ptr6 264 %val7 = load i32, i32 *%ptr7 265 %val8 = load i32, i32 *%ptr8 266 %val9 = load i32, i32 *%ptr9 267 268 %ret = call i32 @foo() 269 270 %t0 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %ret, i32 %val0) 271 %add0 = extractvalue {i32, i1} %t0, 0 272 %obit0 = extractvalue {i32, i1} %t0, 1 273 %t1 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %add0, i32 %val1) 274 %add1 = extractvalue {i32, i1} %t1, 0 275 %obit1 = extractvalue {i32, i1} %t1, 1 276 %res1 = or i1 %obit0, %obit1 277 %t2 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %add1, i32 %val2) 278 %add2 = extractvalue {i32, i1} %t2, 0 279 %obit2 = extractvalue {i32, i1} %t2, 1 280 %res2 = or i1 %res1, %obit2 281 %t3 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %add2, i32 %val3) 282 %add3 = extractvalue {i32, i1} %t3, 0 283 %obit3 = extractvalue {i32, i1} %t3, 1 284 %res3 = or i1 %res2, %obit3 285 %t4 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %add3, i32 %val4) 286 %add4 = extractvalue {i32, i1} %t4, 0 287 %obit4 = extractvalue {i32, i1} %t4, 1 288 %res4 = or i1 %res3, %obit4 289 %t5 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %add4, i32 %val5) 290 %add5 = extractvalue {i32, i1} %t5, 0 291 %obit5 = extractvalue {i32, i1} %t5, 1 292 %res5 = or i1 %res4, %obit5 293 %t6 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %add5, i32 %val6) 294 %add6 = extractvalue {i32, i1} %t6, 0 295 %obit6 = extractvalue {i32, i1} %t6, 1 296 %res6 = or i1 %res5, %obit6 297 %t7 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %add6, i32 %val7) 298 %add7 = extractvalue {i32, i1} %t7, 0 299 %obit7 = extractvalue {i32, i1} %t7, 1 300 %res7 = or i1 %res6, %obit7 301 %t8 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %add7, i32 %val8) 302 %add8 = extractvalue {i32, i1} %t8, 0 303 %obit8 = extractvalue {i32, i1} %t8, 1 304 %res8 = or i1 %res7, %obit8 305 %t9 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %add8, i32 %val9) 306 %add9 = extractvalue {i32, i1} %t9, 0 307 %obit9 = extractvalue {i32, i1} %t9, 1 308 %res9 = or i1 %res8, %obit9 309 310 ret i1 %res9 311} 312 313declare {i32, i1} @llvm.uadd.with.overflow.i32(i32, i32) nounwind readnone 314 315