1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4 2; Test 32-bit atomic minimum and maximum. Here we match the z10 versions, 3; which can't use LOCR. 4; 5; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z10 | FileCheck %s 6 7; Todo: If-converter no longer producing CondReturns (with AtomicExpand pass). 8 9; Check signed minimum. 10define i32 @f1(i32 %dummy, ptr %src, i32 %b) { 11; CHECK-LABEL: f1: 12; CHECK: # %bb.0: 13; CHECK-NEXT: l %r2, 0(%r3) 14; CHECK-NEXT: j .LBB0_2 15; CHECK-NEXT: .LBB0_1: # %atomicrmw.start 16; CHECK-NEXT: # in Loop: Header=BB0_2 Depth=1 17; CHECK-NEXT: cs %r2, %r0, 0(%r3) 18; CHECK-NEXT: je .LBB0_4 19; CHECK-NEXT: .LBB0_2: # %atomicrmw.start 20; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 21; CHECK-NEXT: lr %r0, %r2 22; CHECK-NEXT: crjle %r2, %r4, .LBB0_1 23; CHECK-NEXT: # %bb.3: # %atomicrmw.start 24; CHECK-NEXT: # in Loop: Header=BB0_2 Depth=1 25; CHECK-NEXT: lr %r0, %r4 26; CHECK-NEXT: j .LBB0_1 27; CHECK-NEXT: .LBB0_4: # %atomicrmw.end 28; CHECK-NEXT: br %r14 29 %res = atomicrmw min ptr %src, i32 %b seq_cst 30 ret i32 %res 31} 32 33; Check signed maximum. 34define i32 @f2(i32 %dummy, ptr %src, i32 %b) { 35; CHECK-LABEL: f2: 36; CHECK: # %bb.0: 37; CHECK-NEXT: l %r2, 0(%r3) 38; CHECK-NEXT: j .LBB1_2 39; CHECK-NEXT: .LBB1_1: # %atomicrmw.start 40; CHECK-NEXT: # in Loop: Header=BB1_2 Depth=1 41; CHECK-NEXT: cs %r2, %r0, 0(%r3) 42; CHECK-NEXT: je .LBB1_4 43; CHECK-NEXT: .LBB1_2: # %atomicrmw.start 44; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 45; CHECK-NEXT: lr %r0, %r2 46; CHECK-NEXT: crjh %r2, %r4, .LBB1_1 47; CHECK-NEXT: # %bb.3: # %atomicrmw.start 48; CHECK-NEXT: # in Loop: Header=BB1_2 Depth=1 49; CHECK-NEXT: lr %r0, %r4 50; CHECK-NEXT: j .LBB1_1 51; CHECK-NEXT: .LBB1_4: # %atomicrmw.end 52; CHECK-NEXT: br %r14 53 %res = atomicrmw max ptr %src, i32 %b seq_cst 54 ret i32 %res 55} 56 57; Check unsigned minimum. 58define i32 @f3(i32 %dummy, ptr %src, i32 %b) { 59; CHECK-LABEL: f3: 60; CHECK: # %bb.0: 61; CHECK-NEXT: l %r2, 0(%r3) 62; CHECK-NEXT: j .LBB2_2 63; CHECK-NEXT: .LBB2_1: # %atomicrmw.start 64; CHECK-NEXT: # in Loop: Header=BB2_2 Depth=1 65; CHECK-NEXT: cs %r2, %r0, 0(%r3) 66; CHECK-NEXT: je .LBB2_4 67; CHECK-NEXT: .LBB2_2: # %atomicrmw.start 68; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 69; CHECK-NEXT: lr %r0, %r2 70; CHECK-NEXT: clrjle %r2, %r4, .LBB2_1 71; CHECK-NEXT: # %bb.3: # %atomicrmw.start 72; CHECK-NEXT: # in Loop: Header=BB2_2 Depth=1 73; CHECK-NEXT: lr %r0, %r4 74; CHECK-NEXT: j .LBB2_1 75; CHECK-NEXT: .LBB2_4: # %atomicrmw.end 76; CHECK-NEXT: br %r14 77 %res = atomicrmw umin ptr %src, i32 %b seq_cst 78 ret i32 %res 79} 80 81; Check unsigned maximum. 82define i32 @f4(i32 %dummy, ptr %src, i32 %b) { 83; CHECK-LABEL: f4: 84; CHECK: # %bb.0: 85; CHECK-NEXT: l %r2, 0(%r3) 86; CHECK-NEXT: j .LBB3_2 87; CHECK-NEXT: .LBB3_1: # %atomicrmw.start 88; CHECK-NEXT: # in Loop: Header=BB3_2 Depth=1 89; CHECK-NEXT: cs %r2, %r0, 0(%r3) 90; CHECK-NEXT: je .LBB3_4 91; CHECK-NEXT: .LBB3_2: # %atomicrmw.start 92; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 93; CHECK-NEXT: lr %r0, %r2 94; CHECK-NEXT: clrjh %r2, %r4, .LBB3_1 95; CHECK-NEXT: # %bb.3: # %atomicrmw.start 96; CHECK-NEXT: # in Loop: Header=BB3_2 Depth=1 97; CHECK-NEXT: lr %r0, %r4 98; CHECK-NEXT: j .LBB3_1 99; CHECK-NEXT: .LBB3_4: # %atomicrmw.end 100; CHECK-NEXT: br %r14 101 %res = atomicrmw umax ptr %src, i32 %b seq_cst 102 ret i32 %res 103} 104 105; Check the high end of the aligned CS range. 106define i32 @f5(i32 %dummy, ptr %src, i32 %b) { 107; CHECK-LABEL: f5: 108; CHECK: # %bb.0: 109; CHECK-NEXT: l %r2, 4092(%r3) 110; CHECK-NEXT: j .LBB4_2 111; CHECK-NEXT: .LBB4_1: # %atomicrmw.start 112; CHECK-NEXT: # in Loop: Header=BB4_2 Depth=1 113; CHECK-NEXT: cs %r2, %r0, 4092(%r3) 114; CHECK-NEXT: je .LBB4_4 115; CHECK-NEXT: .LBB4_2: # %atomicrmw.start 116; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 117; CHECK-NEXT: lr %r0, %r2 118; CHECK-NEXT: crjle %r2, %r4, .LBB4_1 119; CHECK-NEXT: # %bb.3: # %atomicrmw.start 120; CHECK-NEXT: # in Loop: Header=BB4_2 Depth=1 121; CHECK-NEXT: lr %r0, %r4 122; CHECK-NEXT: j .LBB4_1 123; CHECK-NEXT: .LBB4_4: # %atomicrmw.end 124; CHECK-NEXT: br %r14 125 %ptr = getelementptr i32, ptr %src, i64 1023 126 %res = atomicrmw min ptr %ptr, i32 %b seq_cst 127 ret i32 %res 128} 129 130; Check the next word up, which requires CSY. 131define i32 @f6(i32 %dummy, ptr %src, i32 %b) { 132; CHECK-LABEL: f6: 133; CHECK: # %bb.0: 134; CHECK-NEXT: ly %r2, 4096(%r3) 135; CHECK-NEXT: j .LBB5_2 136; CHECK-NEXT: .LBB5_1: # %atomicrmw.start 137; CHECK-NEXT: # in Loop: Header=BB5_2 Depth=1 138; CHECK-NEXT: csy %r2, %r0, 4096(%r3) 139; CHECK-NEXT: je .LBB5_4 140; CHECK-NEXT: .LBB5_2: # %atomicrmw.start 141; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 142; CHECK-NEXT: lr %r0, %r2 143; CHECK-NEXT: crjle %r2, %r4, .LBB5_1 144; CHECK-NEXT: # %bb.3: # %atomicrmw.start 145; CHECK-NEXT: # in Loop: Header=BB5_2 Depth=1 146; CHECK-NEXT: lr %r0, %r4 147; CHECK-NEXT: j .LBB5_1 148; CHECK-NEXT: .LBB5_4: # %atomicrmw.end 149; CHECK-NEXT: br %r14 150 %ptr = getelementptr i32, ptr %src, i64 1024 151 %res = atomicrmw min ptr %ptr, i32 %b seq_cst 152 ret i32 %res 153} 154 155; Check the high end of the aligned CSY range. 156define i32 @f7(i32 %dummy, ptr %src, i32 %b) { 157; CHECK-LABEL: f7: 158; CHECK: # %bb.0: 159; CHECK-NEXT: ly %r2, 524284(%r3) 160; CHECK-NEXT: j .LBB6_2 161; CHECK-NEXT: .LBB6_1: # %atomicrmw.start 162; CHECK-NEXT: # in Loop: Header=BB6_2 Depth=1 163; CHECK-NEXT: csy %r2, %r0, 524284(%r3) 164; CHECK-NEXT: je .LBB6_4 165; CHECK-NEXT: .LBB6_2: # %atomicrmw.start 166; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 167; CHECK-NEXT: lr %r0, %r2 168; CHECK-NEXT: crjle %r2, %r4, .LBB6_1 169; CHECK-NEXT: # %bb.3: # %atomicrmw.start 170; CHECK-NEXT: # in Loop: Header=BB6_2 Depth=1 171; CHECK-NEXT: lr %r0, %r4 172; CHECK-NEXT: j .LBB6_1 173; CHECK-NEXT: .LBB6_4: # %atomicrmw.end 174; CHECK-NEXT: br %r14 175 %ptr = getelementptr i32, ptr %src, i64 131071 176 %res = atomicrmw min ptr %ptr, i32 %b seq_cst 177 ret i32 %res 178} 179 180; Check the next word up, which needs separate address logic. 181define i32 @f8(i32 %dummy, ptr %src, i32 %b) { 182; CHECK-LABEL: f8: 183; CHECK: # %bb.0: 184; CHECK-NEXT: agfi %r3, 524288 185; CHECK-NEXT: l %r2, 0(%r3) 186; CHECK-NEXT: j .LBB7_2 187; CHECK-NEXT: .LBB7_1: # %atomicrmw.start 188; CHECK-NEXT: # in Loop: Header=BB7_2 Depth=1 189; CHECK-NEXT: cs %r2, %r0, 0(%r3) 190; CHECK-NEXT: je .LBB7_4 191; CHECK-NEXT: .LBB7_2: # %atomicrmw.start 192; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 193; CHECK-NEXT: lr %r0, %r2 194; CHECK-NEXT: crjle %r2, %r4, .LBB7_1 195; CHECK-NEXT: # %bb.3: # %atomicrmw.start 196; CHECK-NEXT: # in Loop: Header=BB7_2 Depth=1 197; CHECK-NEXT: lr %r0, %r4 198; CHECK-NEXT: j .LBB7_1 199; CHECK-NEXT: .LBB7_4: # %atomicrmw.end 200; CHECK-NEXT: br %r14 201 %ptr = getelementptr i32, ptr %src, i64 131072 202 %res = atomicrmw min ptr %ptr, i32 %b seq_cst 203 ret i32 %res 204} 205 206; Check the high end of the negative aligned CSY range. 207define i32 @f9(i32 %dummy, ptr %src, i32 %b) { 208; CHECK-LABEL: f9: 209; CHECK: # %bb.0: 210; CHECK-NEXT: ly %r2, -4(%r3) 211; CHECK-NEXT: j .LBB8_2 212; CHECK-NEXT: .LBB8_1: # %atomicrmw.start 213; CHECK-NEXT: # in Loop: Header=BB8_2 Depth=1 214; CHECK-NEXT: csy %r2, %r0, -4(%r3) 215; CHECK-NEXT: je .LBB8_4 216; CHECK-NEXT: .LBB8_2: # %atomicrmw.start 217; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 218; CHECK-NEXT: lr %r0, %r2 219; CHECK-NEXT: crjle %r2, %r4, .LBB8_1 220; CHECK-NEXT: # %bb.3: # %atomicrmw.start 221; CHECK-NEXT: # in Loop: Header=BB8_2 Depth=1 222; CHECK-NEXT: lr %r0, %r4 223; CHECK-NEXT: j .LBB8_1 224; CHECK-NEXT: .LBB8_4: # %atomicrmw.end 225; CHECK-NEXT: br %r14 226 %ptr = getelementptr i32, ptr %src, i64 -1 227 %res = atomicrmw min ptr %ptr, i32 %b seq_cst 228 ret i32 %res 229} 230 231; Check the low end of the CSY range. 232define i32 @f10(i32 %dummy, ptr %src, i32 %b) { 233; CHECK-LABEL: f10: 234; CHECK: # %bb.0: 235; CHECK-NEXT: ly %r2, -524288(%r3) 236; CHECK-NEXT: j .LBB9_2 237; CHECK-NEXT: .LBB9_1: # %atomicrmw.start 238; CHECK-NEXT: # in Loop: Header=BB9_2 Depth=1 239; CHECK-NEXT: csy %r2, %r0, -524288(%r3) 240; CHECK-NEXT: je .LBB9_4 241; CHECK-NEXT: .LBB9_2: # %atomicrmw.start 242; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 243; CHECK-NEXT: lr %r0, %r2 244; CHECK-NEXT: crjle %r2, %r4, .LBB9_1 245; CHECK-NEXT: # %bb.3: # %atomicrmw.start 246; CHECK-NEXT: # in Loop: Header=BB9_2 Depth=1 247; CHECK-NEXT: lr %r0, %r4 248; CHECK-NEXT: j .LBB9_1 249; CHECK-NEXT: .LBB9_4: # %atomicrmw.end 250; CHECK-NEXT: br %r14 251 %ptr = getelementptr i32, ptr %src, i64 -131072 252 %res = atomicrmw min ptr %ptr, i32 %b seq_cst 253 ret i32 %res 254} 255 256; Check the next word down, which needs separate address logic. 257define i32 @f11(i32 %dummy, ptr %src, i32 %b) { 258; CHECK-LABEL: f11: 259; CHECK: # %bb.0: 260; CHECK-NEXT: agfi %r3, -524292 261; CHECK-NEXT: l %r2, 0(%r3) 262; CHECK-NEXT: j .LBB10_2 263; CHECK-NEXT: .LBB10_1: # %atomicrmw.start 264; CHECK-NEXT: # in Loop: Header=BB10_2 Depth=1 265; CHECK-NEXT: cs %r2, %r0, 0(%r3) 266; CHECK-NEXT: je .LBB10_4 267; CHECK-NEXT: .LBB10_2: # %atomicrmw.start 268; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 269; CHECK-NEXT: lr %r0, %r2 270; CHECK-NEXT: crjle %r2, %r4, .LBB10_1 271; CHECK-NEXT: # %bb.3: # %atomicrmw.start 272; CHECK-NEXT: # in Loop: Header=BB10_2 Depth=1 273; CHECK-NEXT: lr %r0, %r4 274; CHECK-NEXT: j .LBB10_1 275; CHECK-NEXT: .LBB10_4: # %atomicrmw.end 276; CHECK-NEXT: br %r14 277 %ptr = getelementptr i32, ptr %src, i64 -131073 278 %res = atomicrmw min ptr %ptr, i32 %b seq_cst 279 ret i32 %res 280} 281 282; Check that indexed addresses are not allowed. 283define i32 @f12(i32 %dummy, i64 %base, i64 %index, i32 %b) { 284; CHECK-LABEL: f12: 285; CHECK: # %bb.0: 286; CHECK-NEXT: l %r2, 0(%r4,%r3) 287; CHECK-NEXT: agr %r3, %r4 288; CHECK-NEXT: j .LBB11_2 289; CHECK-NEXT: .LBB11_1: # %atomicrmw.start 290; CHECK-NEXT: # in Loop: Header=BB11_2 Depth=1 291; CHECK-NEXT: cs %r2, %r0, 0(%r3) 292; CHECK-NEXT: je .LBB11_4 293; CHECK-NEXT: .LBB11_2: # %atomicrmw.start 294; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 295; CHECK-NEXT: lr %r0, %r2 296; CHECK-NEXT: crjle %r2, %r5, .LBB11_1 297; CHECK-NEXT: # %bb.3: # %atomicrmw.start 298; CHECK-NEXT: # in Loop: Header=BB11_2 Depth=1 299; CHECK-NEXT: lr %r0, %r5 300; CHECK-NEXT: j .LBB11_1 301; CHECK-NEXT: .LBB11_4: # %atomicrmw.end 302; CHECK-NEXT: br %r14 303 %add = add i64 %base, %index 304 %ptr = inttoptr i64 %add to ptr 305 %res = atomicrmw min ptr %ptr, i32 %b seq_cst 306 ret i32 %res 307} 308 309; Check that constants are handled. 310define i32 @f13(i32 %dummy, ptr %ptr) { 311; CHECK-LABEL: f13: 312; CHECK: # %bb.0: 313; CHECK-NEXT: l %r2, 0(%r3) 314; CHECK-NEXT: j .LBB12_2 315; CHECK-NEXT: .LBB12_1: # %atomicrmw.start 316; CHECK-NEXT: # in Loop: Header=BB12_2 Depth=1 317; CHECK-NEXT: cs %r2, %r0, 0(%r3) 318; CHECK-NEXT: je .LBB12_4 319; CHECK-NEXT: .LBB12_2: # %atomicrmw.start 320; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 321; CHECK-NEXT: lr %r0, %r2 322; CHECK-NEXT: cijl %r2, 43, .LBB12_1 323; CHECK-NEXT: # %bb.3: # %atomicrmw.start 324; CHECK-NEXT: # in Loop: Header=BB12_2 Depth=1 325; CHECK-NEXT: lhi %r0, 42 326; CHECK-NEXT: j .LBB12_1 327; CHECK-NEXT: .LBB12_4: # %atomicrmw.end 328; CHECK-NEXT: br %r14 329 %res = atomicrmw min ptr %ptr, i32 42 seq_cst 330 ret i32 %res 331} 332