1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4 2; RUN: llc -mtriple=arm64-apple-macosx -o - %s | FileCheck %s 3 4%struct.zot = type { ptr, i32, ptr, ptr, i32, i32, i32, i32, i32, i32 } 5 6; FIXME: currently the AND is performed on 64 bit unnecessarily. 7 8; Test cases where an AND is extended from i32 -> i64 which is free. Make 9; sure the extends do not get moved to the arguments, which would perform the 10; AND on 64 bits unnecessarily. 11 12define void @avoid_promotion_1_and(ptr nocapture noundef %arg, ptr %p) { 13; CHECK-LABEL: avoid_promotion_1_and: 14; CHECK: ; %bb.0: ; %bb 15; CHECK-NEXT: ldr w8, [x0, #52] 16; CHECK-NEXT: mov w9, #10 ; =0xa 17; CHECK-NEXT: LBB0_1: ; %bb8 18; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 19; CHECK-NEXT: cmp w8, #3 20; CHECK-NEXT: b.lo LBB0_1 21; CHECK-NEXT: ; %bb.2: ; %bb9 22; CHECK-NEXT: ; in Loop: Header=BB0_1 Depth=1 23; CHECK-NEXT: ldr w10, [x0, #32] 24; CHECK-NEXT: ldr w11, [x1, #76] 25; CHECK-NEXT: ldr w12, [x1] 26; CHECK-NEXT: eor w10, w10, w11 27; CHECK-NEXT: and w10, w10, w12 28; CHECK-NEXT: str w10, [x0, #32] 29; CHECK-NEXT: strh w9, [x1, w10, uxtw #1] 30; CHECK-NEXT: b LBB0_1 31bb: 32 %gep = getelementptr inbounds %struct.zot, ptr %arg, i64 0, i32 9 33 %l = load i32, ptr %gep, align 4 34 %icmp = icmp ugt i32 %l, 2 35 %gep1 = getelementptr inbounds %struct.zot, ptr %arg, i64 0, i32 4 36 %gep2 = getelementptr inbounds %struct.zot, ptr %arg, i64 0, i32 7 37 br label %bb8 38 39bb8: ; preds = %bb27, %bb 40 br i1 %icmp, label %bb9, label %bb27 41 42bb9: ; preds = %bb8 43 %l10 = load i32, ptr %gep1, align 8 44 %gep14 = getelementptr inbounds i32, ptr %p, i64 19 45 %l15 = load i32, ptr %gep14, align 1 46 %xor = xor i32 %l10, %l15 47 %l17 = load i32, ptr %p, align 8 48 %and = and i32 %xor, %l17 49 store i32 %and, ptr %gep1, align 8 50 %zext19 = zext i32 %and to i64 51 %gep20 = getelementptr inbounds i16, ptr %p, i64 %zext19 52 store i16 10, ptr %gep20, align 2 53 br label %bb27 54 55bb27: ; preds = %bb9, %bb8 56 br label %bb8 57} 58 59define void @avoid_promotion_2_and(ptr nocapture noundef %arg) { 60; CHECK-LABEL: avoid_promotion_2_and: 61; CHECK: ; %bb.0: ; %entry 62; CHECK-NEXT: add x8, x0, #32 63; CHECK-NEXT: b LBB1_2 64; CHECK-NEXT: LBB1_1: ; %latch 65; CHECK-NEXT: ; in Loop: Header=BB1_2 Depth=1 66; CHECK-NEXT: cmp w9, #2 67; CHECK-NEXT: add x8, x8, #56 68; CHECK-NEXT: b.ls LBB1_4 69; CHECK-NEXT: LBB1_2: ; %loop 70; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 71; CHECK-NEXT: ldr w9, [x8, #20] 72; CHECK-NEXT: cmp w9, #3 73; CHECK-NEXT: b.lo LBB1_1 74; CHECK-NEXT: ; %bb.3: ; %then 75; CHECK-NEXT: ; in Loop: Header=BB1_2 Depth=1 76; CHECK-NEXT: ldp w13, w12, [x8, #12] 77; CHECK-NEXT: ldr w10, [x8] 78; CHECK-NEXT: ldr x11, [x0] 79; CHECK-NEXT: ldr w14, [x8, #8] 80; CHECK-NEXT: lsl w10, w10, w13 81; CHECK-NEXT: ldrb w11, [x11, x12] 82; CHECK-NEXT: eor w10, w10, w11 83; CHECK-NEXT: ldur w11, [x8, #-24] 84; CHECK-NEXT: and w10, w10, w14 85; CHECK-NEXT: ldp x14, x13, [x8, #-16] 86; CHECK-NEXT: str w10, [x8] 87; CHECK-NEXT: and w11, w11, w12 88; CHECK-NEXT: ldrh w15, [x13, w10, uxtw #1] 89; CHECK-NEXT: strh w15, [x14, w11, uxtw #1] 90; CHECK-NEXT: strh w12, [x13, w10, uxtw #1] 91; CHECK-NEXT: b LBB1_1 92; CHECK-NEXT: LBB1_4: ; %exit 93; CHECK-NEXT: ret 94entry: 95 br label %loop 96 97loop: 98 %p = phi i64 [ 0, %entry ], [ %p.next, %latch ] 99 %gep = getelementptr inbounds %struct.zot, ptr %arg, i64 %p, i32 9 100 %l = load i32, ptr %gep, align 4 101 %icmp = icmp ugt i32 %l, 2 102 %gep1 = getelementptr inbounds %struct.zot, ptr %arg, i64 %p, i32 4 103 %gep2 = getelementptr inbounds %struct.zot, ptr %arg, i64 %p, i32 7 104 %gep3 = getelementptr inbounds %struct.zot, ptr %arg, i64 %p, i32 8 105 %gep4 = getelementptr inbounds %struct.zot, ptr %arg, i64 %p, i32 6 106 %gep5 = getelementptr inbounds %struct.zot, ptr %arg, i64 %p, i32 3 107 %gep6 = getelementptr inbounds %struct.zot, ptr %arg, i64 %p, i32 2 108 %gep7 = getelementptr inbounds %struct.zot, ptr %arg, i64 %p, i32 1 109 br i1 %icmp, label %then, label %latch 110 111then: 112 %l10 = load i32, ptr %gep1, align 8 113 %l11 = load i32, ptr %gep2, align 4 114 %shl = shl i32 %l10, %l11 115 %l12 = load ptr, ptr %arg, align 8 116 %l13 = load i32, ptr %gep3, align 8 117 %zext = zext i32 %l13 to i64 118 %gep14 = getelementptr inbounds i8, ptr %l12, i64 %zext 119 %l15 = load i8, ptr %gep14, align 1 120 %zext16 = zext i8 %l15 to i32 121 %xor = xor i32 %shl, %zext16 122 %l17 = load i32, ptr %gep4, align 8 123 %and = and i32 %xor, %l17 124 store i32 %and, ptr %gep1, align 8 125 %l18 = load ptr, ptr %gep5, align 8 126 %zext19 = zext i32 %and to i64 127 %gep20 = getelementptr inbounds i16, ptr %l18, i64 %zext19 128 %l21 = load i16, ptr %gep20, align 2 129 %l22 = load ptr, ptr %gep6, align 8 130 %l23 = load i32, ptr %gep7, align 8 131 %and24 = and i32 %l23, %l13 132 %zext25 = zext i32 %and24 to i64 133 %gep26 = getelementptr inbounds i16, ptr %l22, i64 %zext25 134 store i16 %l21, ptr %gep26, align 2 135 %trunc = trunc i32 %l13 to i16 136 store i16 %trunc, ptr %gep20, align 2 137 br label %latch 138 139latch: 140 %p.next = add i64 %p, 1 141 br i1 %icmp, label %loop, label %exit 142 143exit: 144 ret void 145} 146