1 ; Test 32-bit square root. 2 ; 3 ; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s 4 5 declare float @llvm.sqrt.f32(float %f) 6 7 ; Check register square root. 8 define float @f1(float %val) { 9 ; CHECK-LABEL: f1: 10 ; CHECK: sqebr %f0, %f0 11 ; CHECK: br %r14 12 %res = call float @llvm.sqrt.f32(float %val) 13 ret float %res 14 } 15 16 ; Check the low end of the SQEB range. 17 define float @f2(float *%ptr) { 18 ; CHECK-LABEL: f2: 19 ; CHECK: sqeb %f0, 0(%r2) 20 ; CHECK: br %r14 21 %val = load float *%ptr 22 %res = call float @llvm.sqrt.f32(float %val) 23 ret float %res 24 } 25 26 ; Check the high end of the aligned SQEB range. 27 define float @f3(float *%base) { 28 ; CHECK-LABEL: f3: 29 ; CHECK: sqeb %f0, 4092(%r2) 30 ; CHECK: br %r14 31 %ptr = getelementptr float *%base, i64 1023 32 %val = load float *%ptr 33 %res = call float @llvm.sqrt.f32(float %val) 34 ret float %res 35 } 36 37 ; Check the next word up, which needs separate address logic. 38 ; Other sequences besides this one would be OK. 39 define float @f4(float *%base) { 40 ; CHECK-LABEL: f4: 41 ; CHECK: aghi %r2, 4096 42 ; CHECK: sqeb %f0, 0(%r2) 43 ; CHECK: br %r14 44 %ptr = getelementptr float *%base, i64 1024 45 %val = load float *%ptr 46 %res = call float @llvm.sqrt.f32(float %val) 47 ret float %res 48 } 49 50 ; Check negative displacements, which also need separate address logic. 51 define float @f5(float *%base) { 52 ; CHECK-LABEL: f5: 53 ; CHECK: aghi %r2, -4 54 ; CHECK: sqeb %f0, 0(%r2) 55 ; CHECK: br %r14 56 %ptr = getelementptr float *%base, i64 -1 57 %val = load float *%ptr 58 %res = call float @llvm.sqrt.f32(float %val) 59 ret float %res 60 } 61 62 ; Check that SQEB allows indices. 63 define float @f6(float *%base, i64 %index) { 64 ; CHECK-LABEL: f6: 65 ; CHECK: sllg %r1, %r3, 2 66 ; CHECK: sqeb %f0, 400(%r1,%r2) 67 ; CHECK: br %r14 68 %ptr1 = getelementptr float *%base, i64 %index 69 %ptr2 = getelementptr float *%ptr1, i64 100 70 %val = load float *%ptr2 71 %res = call float @llvm.sqrt.f32(float %val) 72 ret float %res 73 } 74 75 ; Test a case where we spill the source of at least one SQEBR. We want 76 ; to use SQEB if possible. 77 define void @f7(float *%ptr) { 78 ; CHECK-LABEL: f7: 79 ; CHECK: sqeb {{%f[0-9]+}}, 16{{[04]}}(%r15) 80 ; CHECK: br %r14 81 %val0 = load volatile float *%ptr 82 %val1 = load volatile float *%ptr 83 %val2 = load volatile float *%ptr 84 %val3 = load volatile float *%ptr 85 %val4 = load volatile float *%ptr 86 %val5 = load volatile float *%ptr 87 %val6 = load volatile float *%ptr 88 %val7 = load volatile float *%ptr 89 %val8 = load volatile float *%ptr 90 %val9 = load volatile float *%ptr 91 %val10 = load volatile float *%ptr 92 %val11 = load volatile float *%ptr 93 %val12 = load volatile float *%ptr 94 %val13 = load volatile float *%ptr 95 %val14 = load volatile float *%ptr 96 %val15 = load volatile float *%ptr 97 %val16 = load volatile float *%ptr 98 99 %sqrt0 = call float @llvm.sqrt.f32(float %val0) 100 %sqrt1 = call float @llvm.sqrt.f32(float %val1) 101 %sqrt2 = call float @llvm.sqrt.f32(float %val2) 102 %sqrt3 = call float @llvm.sqrt.f32(float %val3) 103 %sqrt4 = call float @llvm.sqrt.f32(float %val4) 104 %sqrt5 = call float @llvm.sqrt.f32(float %val5) 105 %sqrt6 = call float @llvm.sqrt.f32(float %val6) 106 %sqrt7 = call float @llvm.sqrt.f32(float %val7) 107 %sqrt8 = call float @llvm.sqrt.f32(float %val8) 108 %sqrt9 = call float @llvm.sqrt.f32(float %val9) 109 %sqrt10 = call float @llvm.sqrt.f32(float %val10) 110 %sqrt11 = call float @llvm.sqrt.f32(float %val11) 111 %sqrt12 = call float @llvm.sqrt.f32(float %val12) 112 %sqrt13 = call float @llvm.sqrt.f32(float %val13) 113 %sqrt14 = call float @llvm.sqrt.f32(float %val14) 114 %sqrt15 = call float @llvm.sqrt.f32(float %val15) 115 %sqrt16 = call float @llvm.sqrt.f32(float %val16) 116 117 store volatile float %val0, float *%ptr 118 store volatile float %val1, float *%ptr 119 store volatile float %val2, float *%ptr 120 store volatile float %val3, float *%ptr 121 store volatile float %val4, float *%ptr 122 store volatile float %val5, float *%ptr 123 store volatile float %val6, float *%ptr 124 store volatile float %val7, float *%ptr 125 store volatile float %val8, float *%ptr 126 store volatile float %val9, float *%ptr 127 store volatile float %val10, float *%ptr 128 store volatile float %val11, float *%ptr 129 store volatile float %val12, float *%ptr 130 store volatile float %val13, float *%ptr 131 store volatile float %val14, float *%ptr 132 store volatile float %val15, float *%ptr 133 store volatile float %val16, float *%ptr 134 135 store volatile float %sqrt0, float *%ptr 136 store volatile float %sqrt1, float *%ptr 137 store volatile float %sqrt2, float *%ptr 138 store volatile float %sqrt3, float *%ptr 139 store volatile float %sqrt4, float *%ptr 140 store volatile float %sqrt5, float *%ptr 141 store volatile float %sqrt6, float *%ptr 142 store volatile float %sqrt7, float *%ptr 143 store volatile float %sqrt8, float *%ptr 144 store volatile float %sqrt9, float *%ptr 145 store volatile float %sqrt10, float *%ptr 146 store volatile float %sqrt11, float *%ptr 147 store volatile float %sqrt12, float *%ptr 148 store volatile float %sqrt13, float *%ptr 149 store volatile float %sqrt14, float *%ptr 150 store volatile float %sqrt15, float *%ptr 151 store volatile float %sqrt16, float *%ptr 152 153 ret void 154 } 155
