1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc -ppc-gpr-icmps=all -verify-machineinstrs -mcpu=pwr8 < %s | FileCheck %s 3target datalayout = "e-m:e-i64:64-n32:64" 4target triple = "powerpc64le-unknown-linux-gnu" 5 6@zeroEqualityTest01.buffer1 = private unnamed_addr constant [3 x i32] [i32 1, i32 2, i32 4], align 4 7@zeroEqualityTest01.buffer2 = private unnamed_addr constant [3 x i32] [i32 1, i32 2, i32 3], align 4 8@zeroEqualityTest02.buffer1 = private unnamed_addr constant [4 x i32] [i32 4, i32 0, i32 0, i32 0], align 4 9@zeroEqualityTest02.buffer2 = private unnamed_addr constant [4 x i32] [i32 3, i32 0, i32 0, i32 0], align 4 10@zeroEqualityTest03.buffer1 = private unnamed_addr constant [4 x i32] [i32 0, i32 0, i32 0, i32 3], align 4 11@zeroEqualityTest03.buffer2 = private unnamed_addr constant [4 x i32] [i32 0, i32 0, i32 0, i32 4], align 4 12@zeroEqualityTest04.buffer1 = private unnamed_addr constant [15 x i32] [i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14], align 4 13@zeroEqualityTest04.buffer2 = private unnamed_addr constant [15 x i32] [i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 13], align 4 14 15declare signext i32 @memcmp(i8* nocapture, i8* nocapture, i64) local_unnamed_addr #1 16 17; Check 4 bytes - requires 1 load for each param. 18define signext i32 @zeroEqualityTest02(i8* %x, i8* %y) { 19; CHECK-LABEL: zeroEqualityTest02: 20; CHECK: # %bb.0: 21; CHECK-NEXT: lwz 3, 0(3) 22; CHECK-NEXT: lwz 4, 0(4) 23; CHECK-NEXT: xor 3, 3, 4 24; CHECK-NEXT: cntlzw 3, 3 25; CHECK-NEXT: srwi 3, 3, 5 26; CHECK-NEXT: xori 3, 3, 1 27; CHECK-NEXT: blr 28 %call = tail call signext i32 @memcmp(i8* %x, i8* %y, i64 4) 29 %not.cmp = icmp ne i32 %call, 0 30 %. = zext i1 %not.cmp to i32 31 ret i32 %. 32} 33 34; Check 16 bytes - requires 2 loads for each param (or use vectors?). 35define signext i32 @zeroEqualityTest01(i8* %x, i8* %y) { 36; CHECK-LABEL: zeroEqualityTest01: 37; CHECK: # %bb.0: 38; CHECK-NEXT: ld 5, 0(3) 39; CHECK-NEXT: ld 6, 0(4) 40; CHECK-NEXT: cmpld 5, 6 41; CHECK-NEXT: bne 0, .LBB1_2 42; CHECK-NEXT: # %bb.1: # %loadbb1 43; CHECK-NEXT: ld 3, 8(3) 44; CHECK-NEXT: ld 4, 8(4) 45; CHECK-NEXT: cmpld 3, 4 46; CHECK-NEXT: li 3, 0 47; CHECK-NEXT: beq 0, .LBB1_3 48; CHECK-NEXT: .LBB1_2: # %res_block 49; CHECK-NEXT: li 3, 1 50; CHECK-NEXT: .LBB1_3: # %endblock 51; CHECK-NEXT: clrldi 3, 3, 32 52; CHECK-NEXT: blr 53 %call = tail call signext i32 @memcmp(i8* %x, i8* %y, i64 16) 54 %not.tobool = icmp ne i32 %call, 0 55 %. = zext i1 %not.tobool to i32 56 ret i32 %. 57} 58 59; Check 7 bytes - requires 3 loads for each param. 60define signext i32 @zeroEqualityTest03(i8* %x, i8* %y) { 61; CHECK-LABEL: zeroEqualityTest03: 62; CHECK: # %bb.0: 63; CHECK-NEXT: lwz 5, 0(3) 64; CHECK-NEXT: lwz 6, 0(4) 65; CHECK-NEXT: cmplw 5, 6 66; CHECK-NEXT: bne 0, .LBB2_3 67; CHECK-NEXT: # %bb.1: # %loadbb1 68; CHECK-NEXT: lhz 5, 4(3) 69; CHECK-NEXT: lhz 6, 4(4) 70; CHECK-NEXT: cmplw 5, 6 71; CHECK-NEXT: bne 0, .LBB2_3 72; CHECK-NEXT: # %bb.2: # %loadbb2 73; CHECK-NEXT: lbz 3, 6(3) 74; CHECK-NEXT: lbz 4, 6(4) 75; CHECK-NEXT: cmplw 3, 4 76; CHECK-NEXT: li 3, 0 77; CHECK-NEXT: beq 0, .LBB2_4 78; CHECK-NEXT: .LBB2_3: # %res_block 79; CHECK-NEXT: li 3, 1 80; CHECK-NEXT: .LBB2_4: # %endblock 81; CHECK-NEXT: clrldi 3, 3, 32 82; CHECK-NEXT: blr 83 %call = tail call signext i32 @memcmp(i8* %x, i8* %y, i64 7) 84 %not.lnot = icmp ne i32 %call, 0 85 %cond = zext i1 %not.lnot to i32 86 ret i32 %cond 87} 88 89; Validate with > 0 90define signext i32 @zeroEqualityTest04() { 91; CHECK-LABEL: zeroEqualityTest04: 92; CHECK: # %bb.0: 93; CHECK-NEXT: addis 3, 2, .LzeroEqualityTest02.buffer1@toc@ha 94; CHECK-NEXT: addis 4, 2, .LzeroEqualityTest02.buffer2@toc@ha 95; CHECK-NEXT: addi 6, 3, .LzeroEqualityTest02.buffer1@toc@l 96; CHECK-NEXT: addi 5, 4, .LzeroEqualityTest02.buffer2@toc@l 97; CHECK-NEXT: ldbrx 3, 0, 6 98; CHECK-NEXT: ldbrx 4, 0, 5 99; CHECK-NEXT: cmpld 3, 4 100; CHECK-NEXT: bne 0, .LBB3_2 101; CHECK-NEXT: # %bb.1: # %loadbb1 102; CHECK-NEXT: li 4, 8 103; CHECK-NEXT: ldbrx 3, 6, 4 104; CHECK-NEXT: ldbrx 4, 5, 4 105; CHECK-NEXT: li 5, 0 106; CHECK-NEXT: cmpld 3, 4 107; CHECK-NEXT: beq 0, .LBB3_3 108; CHECK-NEXT: .LBB3_2: # %res_block 109; CHECK-NEXT: cmpld 3, 4 110; CHECK-NEXT: li 3, 1 111; CHECK-NEXT: li 4, -1 112; CHECK-NEXT: isel 5, 4, 3, 0 113; CHECK-NEXT: .LBB3_3: # %endblock 114; CHECK-NEXT: extsw 3, 5 115; CHECK-NEXT: neg 3, 3 116; CHECK-NEXT: rldicl 3, 3, 1, 63 117; CHECK-NEXT: xori 3, 3, 1 118; CHECK-NEXT: blr 119 %call = tail call signext i32 @memcmp(i8* bitcast ([4 x i32]* @zeroEqualityTest02.buffer1 to i8*), i8* bitcast ([4 x i32]* @zeroEqualityTest02.buffer2 to i8*), i64 16) 120 %not.cmp = icmp slt i32 %call, 1 121 %. = zext i1 %not.cmp to i32 122 ret i32 %. 123} 124 125; Validate with < 0 126define signext i32 @zeroEqualityTest05() { 127; CHECK-LABEL: zeroEqualityTest05: 128; CHECK: # %bb.0: 129; CHECK-NEXT: addis 3, 2, .LzeroEqualityTest03.buffer1@toc@ha 130; CHECK-NEXT: addis 4, 2, .LzeroEqualityTest03.buffer2@toc@ha 131; CHECK-NEXT: addi 6, 3, .LzeroEqualityTest03.buffer1@toc@l 132; CHECK-NEXT: addi 5, 4, .LzeroEqualityTest03.buffer2@toc@l 133; CHECK-NEXT: ldbrx 3, 0, 6 134; CHECK-NEXT: ldbrx 4, 0, 5 135; CHECK-NEXT: cmpld 3, 4 136; CHECK-NEXT: bne 0, .LBB4_2 137; CHECK-NEXT: # %bb.1: # %loadbb1 138; CHECK-NEXT: li 4, 8 139; CHECK-NEXT: ldbrx 3, 6, 4 140; CHECK-NEXT: ldbrx 4, 5, 4 141; CHECK-NEXT: li 5, 0 142; CHECK-NEXT: cmpld 3, 4 143; CHECK-NEXT: beq 0, .LBB4_3 144; CHECK-NEXT: .LBB4_2: # %res_block 145; CHECK-NEXT: cmpld 3, 4 146; CHECK-NEXT: li 3, 1 147; CHECK-NEXT: li 4, -1 148; CHECK-NEXT: isel 5, 4, 3, 0 149; CHECK-NEXT: .LBB4_3: # %endblock 150; CHECK-NEXT: srwi 3, 5, 31 151; CHECK-NEXT: xori 3, 3, 1 152; CHECK-NEXT: clrldi 3, 3, 32 153; CHECK-NEXT: blr 154 %call = tail call signext i32 @memcmp(i8* bitcast ([4 x i32]* @zeroEqualityTest03.buffer1 to i8*), i8* bitcast ([4 x i32]* @zeroEqualityTest03.buffer2 to i8*), i64 16) 155 %call.lobit = lshr i32 %call, 31 156 %call.lobit.not = xor i32 %call.lobit, 1 157 ret i32 %call.lobit.not 158} 159 160; Validate with memcmp()?: 161define signext i32 @equalityFoldTwoConstants() { 162; CHECK-LABEL: equalityFoldTwoConstants: 163; CHECK: # %bb.0: # %loadbb 164; CHECK-NEXT: li 3, 1 165; CHECK-NEXT: blr 166 %call = tail call signext i32 @memcmp(i8* bitcast ([15 x i32]* @zeroEqualityTest04.buffer1 to i8*), i8* bitcast ([15 x i32]* @zeroEqualityTest04.buffer2 to i8*), i64 16) 167 %not.tobool = icmp eq i32 %call, 0 168 %cond = zext i1 %not.tobool to i32 169 ret i32 %cond 170} 171 172define signext i32 @equalityFoldOneConstant(i8* %X) { 173; CHECK-LABEL: equalityFoldOneConstant: 174; CHECK: # %bb.0: 175; CHECK-NEXT: ld 4, 0(3) 176; CHECK-NEXT: li 5, 1 177; CHECK-NEXT: sldi 5, 5, 32 178; CHECK-NEXT: cmpld 4, 5 179; CHECK-NEXT: bne 0, .LBB6_2 180; CHECK-NEXT: # %bb.1: # %loadbb1 181; CHECK-NEXT: li 4, 3 182; CHECK-NEXT: ld 3, 8(3) 183; CHECK-NEXT: sldi 4, 4, 32 184; CHECK-NEXT: ori 4, 4, 2 185; CHECK-NEXT: cmpld 3, 4 186; CHECK-NEXT: li 3, 0 187; CHECK-NEXT: beq 0, .LBB6_3 188; CHECK-NEXT: .LBB6_2: # %res_block 189; CHECK-NEXT: li 3, 1 190; CHECK-NEXT: .LBB6_3: # %endblock 191; CHECK-NEXT: cntlzw 3, 3 192; CHECK-NEXT: srwi 3, 3, 5 193; CHECK-NEXT: blr 194 %call = tail call signext i32 @memcmp(i8* bitcast ([15 x i32]* @zeroEqualityTest04.buffer1 to i8*), i8* %X, i64 16) 195 %not.tobool = icmp eq i32 %call, 0 196 %cond = zext i1 %not.tobool to i32 197 ret i32 %cond 198} 199 200define i1 @length2_eq_nobuiltin_attr(i8* %X, i8* %Y) { 201; CHECK-LABEL: length2_eq_nobuiltin_attr: 202; CHECK: # %bb.0: 203; CHECK-NEXT: mflr 0 204; CHECK-NEXT: std 0, 16(1) 205; CHECK-NEXT: stdu 1, -32(1) 206; CHECK-NEXT: .cfi_def_cfa_offset 32 207; CHECK-NEXT: .cfi_offset lr, 16 208; CHECK-NEXT: li 5, 2 209; CHECK-NEXT: bl memcmp 210; CHECK-NEXT: nop 211; CHECK-NEXT: cntlzw 3, 3 212; CHECK-NEXT: rlwinm 3, 3, 27, 31, 31 213; CHECK-NEXT: addi 1, 1, 32 214; CHECK-NEXT: ld 0, 16(1) 215; CHECK-NEXT: mtlr 0 216; CHECK-NEXT: blr 217 %m = tail call signext i32 @memcmp(i8* %X, i8* %Y, i64 2) nobuiltin 218 %c = icmp eq i32 %m, 0 219 ret i1 %c 220} 221 222