1; Test 64-bit addition in which the second operand is variable. 2; 3; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z10 | FileCheck %s --check-prefixes=CHECK,Z10 4; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z196 | FileCheck %s --check-prefixes=CHECK,Z196 5 6declare i64 @foo() 7 8; Check AGR. 9define i64 @f1(i64 %a, i64 %b) { 10; CHECK-LABEL: f1: 11; CHECK: agr %r2, %r3 12; CHECK: br %r14 13 %add = add i64 %a, %b 14 ret i64 %add 15} 16 17; Check AG with no displacement. 18define i64 @f2(i64 %a, i64 *%src) { 19; CHECK-LABEL: f2: 20; CHECK: ag %r2, 0(%r3) 21; CHECK: br %r14 22 %b = load i64, i64 *%src 23 %add = add i64 %a, %b 24 ret i64 %add 25} 26 27; Check the high end of the aligned AG range. 28define i64 @f3(i64 %a, i64 *%src) { 29; CHECK-LABEL: f3: 30; CHECK: ag %r2, 524280(%r3) 31; CHECK: br %r14 32 %ptr = getelementptr i64, i64 *%src, i64 65535 33 %b = load i64, i64 *%ptr 34 %add = add i64 %a, %b 35 ret i64 %add 36} 37 38; Check the next doubleword up, which needs separate address logic. 39; Other sequences besides this one would be OK. 40define i64 @f4(i64 %a, i64 *%src) { 41; CHECK-LABEL: f4: 42; CHECK: agfi %r3, 524288 43; CHECK: ag %r2, 0(%r3) 44; CHECK: br %r14 45 %ptr = getelementptr i64, i64 *%src, i64 65536 46 %b = load i64, i64 *%ptr 47 %add = add i64 %a, %b 48 ret i64 %add 49} 50 51; Check the high end of the negative aligned AG range. 52define i64 @f5(i64 %a, i64 *%src) { 53; CHECK-LABEL: f5: 54; CHECK: ag %r2, -8(%r3) 55; CHECK: br %r14 56 %ptr = getelementptr i64, i64 *%src, i64 -1 57 %b = load i64, i64 *%ptr 58 %add = add i64 %a, %b 59 ret i64 %add 60} 61 62; Check the low end of the AG range. 63define i64 @f6(i64 %a, i64 *%src) { 64; CHECK-LABEL: f6: 65; CHECK: ag %r2, -524288(%r3) 66; CHECK: br %r14 67 %ptr = getelementptr i64, i64 *%src, i64 -65536 68 %b = load i64, i64 *%ptr 69 %add = add i64 %a, %b 70 ret i64 %add 71} 72 73; Check the next doubleword down, which needs separate address logic. 74; Other sequences besides this one would be OK. 75define i64 @f7(i64 %a, i64 *%src) { 76; CHECK-LABEL: f7: 77; CHECK: agfi %r3, -524296 78; CHECK: ag %r2, 0(%r3) 79; CHECK: br %r14 80 %ptr = getelementptr i64, i64 *%src, i64 -65537 81 %b = load i64, i64 *%ptr 82 %add = add i64 %a, %b 83 ret i64 %add 84} 85 86; Check that AG allows an index. 87define i64 @f8(i64 %a, i64 %src, i64 %index) { 88; CHECK-LABEL: f8: 89; CHECK: ag %r2, 524280({{%r4,%r3|%r3,%r4}}) 90; CHECK: br %r14 91 %add1 = add i64 %src, %index 92 %add2 = add i64 %add1, 524280 93 %ptr = inttoptr i64 %add2 to i64 * 94 %b = load i64, i64 *%ptr 95 %add = add i64 %a, %b 96 ret i64 %add 97} 98 99; Check that additions of spilled values can use AG rather than AGR. 100; Note: Z196 is suboptimal with one unfolded reload. 101define i64 @f9(i64 *%ptr0) { 102; CHECK-LABEL: f9: 103; CHECK: brasl %r14, foo@PLT 104; Z10: ag %r2, 168(%r15) 105; Z196: ag %r0, 168(%r15) 106; CHECK: br %r14 107 %ptr1 = getelementptr i64, i64 *%ptr0, i64 2 108 %ptr2 = getelementptr i64, i64 *%ptr0, i64 4 109 %ptr3 = getelementptr i64, i64 *%ptr0, i64 6 110 %ptr4 = getelementptr i64, i64 *%ptr0, i64 8 111 %ptr5 = getelementptr i64, i64 *%ptr0, i64 10 112 %ptr6 = getelementptr i64, i64 *%ptr0, i64 12 113 %ptr7 = getelementptr i64, i64 *%ptr0, i64 14 114 %ptr8 = getelementptr i64, i64 *%ptr0, i64 16 115 %ptr9 = getelementptr i64, i64 *%ptr0, i64 18 116 117 %val0 = load i64, i64 *%ptr0 118 %val1 = load i64, i64 *%ptr1 119 %val2 = load i64, i64 *%ptr2 120 %val3 = load i64, i64 *%ptr3 121 %val4 = load i64, i64 *%ptr4 122 %val5 = load i64, i64 *%ptr5 123 %val6 = load i64, i64 *%ptr6 124 %val7 = load i64, i64 *%ptr7 125 %val8 = load i64, i64 *%ptr8 126 %val9 = load i64, i64 *%ptr9 127 128 %ret = call i64 @foo() 129 130 %add0 = add i64 %ret, %val0 131 %add1 = add i64 %add0, %val1 132 %add2 = add i64 %add1, %val2 133 %add3 = add i64 %add2, %val3 134 %add4 = add i64 %add3, %val4 135 %add5 = add i64 %add4, %val5 136 %add6 = add i64 %add5, %val6 137 %add7 = add i64 %add6, %val7 138 %add8 = add i64 %add7, %val8 139 %add9 = add i64 %add8, %val9 140 141 ret i64 %add9 142} 143