1 ; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s 2 3 declare float @llvm.fma.f32(float %f1, float %f2, float %f3) 4 5 define float @f1(float %f1, float %f2, float %acc) { 6 ; CHECK-LABEL: f1: 7 ; CHECK: msebr %f4, %f0, %f2 8 ; CHECK: ler %f0, %f4 9 ; CHECK: br %r14 10 %negacc = fsub float -0.0, %acc 11 %res = call float @llvm.fma.f32 (float %f1, float %f2, float %negacc) 12 ret float %res 13 } 14 15 define float @f2(float %f1, float *%ptr, float %acc) { 16 ; CHECK-LABEL: f2: 17 ; CHECK: mseb %f2, %f0, 0(%r2) 18 ; CHECK: ler %f0, %f2 19 ; CHECK: br %r14 20 %f2 = load float *%ptr 21 %negacc = fsub float -0.0, %acc 22 %res = call float @llvm.fma.f32 (float %f1, float %f2, float %negacc) 23 ret float %res 24 } 25 26 define float @f3(float %f1, float *%base, float %acc) { 27 ; CHECK-LABEL: f3: 28 ; CHECK: mseb %f2, %f0, 4092(%r2) 29 ; CHECK: ler %f0, %f2 30 ; CHECK: br %r14 31 %ptr = getelementptr float *%base, i64 1023 32 %f2 = load float *%ptr 33 %negacc = fsub float -0.0, %acc 34 %res = call float @llvm.fma.f32 (float %f1, float %f2, float %negacc) 35 ret float %res 36 } 37 38 define float @f4(float %f1, float *%base, float %acc) { 39 ; The important thing here is that we don't generate an out-of-range 40 ; displacement. Other sequences besides this one would be OK. 41 ; 42 ; CHECK-LABEL: f4: 43 ; CHECK: aghi %r2, 4096 44 ; CHECK: mseb %f2, %f0, 0(%r2) 45 ; CHECK: ler %f0, %f2 46 ; CHECK: br %r14 47 %ptr = getelementptr float *%base, i64 1024 48 %f2 = load float *%ptr 49 %negacc = fsub float -0.0, %acc 50 %res = call float @llvm.fma.f32 (float %f1, float %f2, float %negacc) 51 ret float %res 52 } 53 54 define float @f5(float %f1, float *%base, float %acc) { 55 ; Here too the important thing is that we don't generate an out-of-range 56 ; displacement. Other sequences besides this one would be OK. 57 ; 58 ; CHECK-LABEL: f5: 59 ; CHECK: aghi %r2, -4 60 ; CHECK: mseb %f2, %f0, 0(%r2) 61 ; CHECK: ler %f0, %f2 62 ; CHECK: br %r14 63 %ptr = getelementptr float *%base, i64 -1 64 %f2 = load float *%ptr 65 %negacc = fsub float -0.0, %acc 66 %res = call float @llvm.fma.f32 (float %f1, float %f2, float %negacc) 67 ret float %res 68 } 69 70 define float @f6(float %f1, float *%base, i64 %index, float %acc) { 71 ; CHECK-LABEL: f6: 72 ; CHECK: sllg %r1, %r3, 2 73 ; CHECK: mseb %f2, %f0, 0(%r1,%r2) 74 ; CHECK: ler %f0, %f2 75 ; CHECK: br %r14 76 %ptr = getelementptr float *%base, i64 %index 77 %f2 = load float *%ptr 78 %negacc = fsub float -0.0, %acc 79 %res = call float @llvm.fma.f32 (float %f1, float %f2, float %negacc) 80 ret float %res 81 } 82 83 define float @f7(float %f1, float *%base, i64 %index, float %acc) { 84 ; CHECK-LABEL: f7: 85 ; CHECK: sllg %r1, %r3, 2 86 ; CHECK: mseb %f2, %f0, 4092({{%r1,%r2|%r2,%r1}}) 87 ; CHECK: ler %f0, %f2 88 ; CHECK: br %r14 89 %index2 = add i64 %index, 1023 90 %ptr = getelementptr float *%base, i64 %index2 91 %f2 = load float *%ptr 92 %negacc = fsub float -0.0, %acc 93 %res = call float @llvm.fma.f32 (float %f1, float %f2, float %negacc) 94 ret float %res 95 } 96 97 define float @f8(float %f1, float *%base, i64 %index, float %acc) { 98 ; CHECK-LABEL: f8: 99 ; CHECK: sllg %r1, %r3, 2 100 ; CHECK: lay %r1, 4096({{%r1,%r2|%r2,%r1}}) 101 ; CHECK: mseb %f2, %f0, 0(%r1) 102 ; CHECK: ler %f0, %f2 103 ; CHECK: br %r14 104 %index2 = add i64 %index, 1024 105 %ptr = getelementptr float *%base, i64 %index2 106 %f2 = load float *%ptr 107 %negacc = fsub float -0.0, %acc 108 %res = call float @llvm.fma.f32 (float %f1, float %f2, float %negacc) 109 ret float %res 110 } 111
