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