|
| 1 | +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 |
| 2 | +; RUN: llc < %s -mtriple=arm64-eabi -mattr=fullfp16,sve | FileCheck %s |
| 3 | + |
| 4 | +define float @divf32_2(float %a) nounwind { |
| 5 | +; CHECK-LABEL: divf32_2: |
| 6 | +; CHECK: // %bb.0: |
| 7 | +; CHECK-NEXT: fmov s1, #2.00000000 |
| 8 | +; CHECK-NEXT: fdiv s0, s0, s1 |
| 9 | +; CHECK-NEXT: ret |
| 10 | + %r = fdiv float %a, 2.0 |
| 11 | + ret float %r |
| 12 | +} |
| 13 | + |
| 14 | +define float @divf32_2_arcp(float %a) nounwind { |
| 15 | +; CHECK-LABEL: divf32_2_arcp: |
| 16 | +; CHECK: // %bb.0: |
| 17 | +; CHECK-NEXT: fmov s1, #0.50000000 |
| 18 | +; CHECK-NEXT: fmul s0, s0, s1 |
| 19 | +; CHECK-NEXT: ret |
| 20 | + %r = fdiv arcp float %a, 2.0 |
| 21 | + ret float %r |
| 22 | +} |
| 23 | + |
| 24 | +define float @divf32_p75(float %a) nounwind { |
| 25 | +; CHECK-LABEL: divf32_p75: |
| 26 | +; CHECK: // %bb.0: |
| 27 | +; CHECK-NEXT: fmov s1, #0.75000000 |
| 28 | +; CHECK-NEXT: fdiv s0, s0, s1 |
| 29 | +; CHECK-NEXT: ret |
| 30 | + %r = fdiv float %a, 0.75 |
| 31 | + ret float %r |
| 32 | +} |
| 33 | + |
| 34 | +define float @divf32_p75_arcp(float %a) nounwind { |
| 35 | +; CHECK-LABEL: divf32_p75_arcp: |
| 36 | +; CHECK: // %bb.0: |
| 37 | +; CHECK-NEXT: mov w8, #43691 // =0xaaab |
| 38 | +; CHECK-NEXT: movk w8, #16298, lsl #16 |
| 39 | +; CHECK-NEXT: fmov s1, w8 |
| 40 | +; CHECK-NEXT: fmul s0, s0, s1 |
| 41 | +; CHECK-NEXT: ret |
| 42 | + %r = fdiv arcp float %a, 0.75 |
| 43 | + ret float %r |
| 44 | +} |
| 45 | + |
| 46 | +define half @divf16_2(half %a) nounwind { |
| 47 | +; CHECK-LABEL: divf16_2: |
| 48 | +; CHECK: // %bb.0: |
| 49 | +; CHECK-NEXT: fmov h1, #2.00000000 |
| 50 | +; CHECK-NEXT: fdiv h0, h0, h1 |
| 51 | +; CHECK-NEXT: ret |
| 52 | + %r = fdiv half %a, 2.0 |
| 53 | + ret half %r |
| 54 | +} |
| 55 | + |
| 56 | +define half @divf16_32768(half %a) nounwind { |
| 57 | +; CHECK-LABEL: divf16_32768: |
| 58 | +; CHECK: // %bb.0: |
| 59 | +; CHECK-NEXT: mov w8, #30720 // =0x7800 |
| 60 | +; CHECK-NEXT: fmov h1, w8 |
| 61 | +; CHECK-NEXT: fdiv h0, h0, h1 |
| 62 | +; CHECK-NEXT: ret |
| 63 | + %r = fdiv half %a, 32768.0 |
| 64 | + ret half %r |
| 65 | +} |
| 66 | + |
| 67 | +define half @divf16_32768_arcp(half %a) nounwind { |
| 68 | +; CHECK-LABEL: divf16_32768_arcp: |
| 69 | +; CHECK: // %bb.0: |
| 70 | +; CHECK-NEXT: mov w8, #512 // =0x200 |
| 71 | +; CHECK-NEXT: fmov h1, w8 |
| 72 | +; CHECK-NEXT: fmul h0, h0, h1 |
| 73 | +; CHECK-NEXT: ret |
| 74 | + %r = fdiv arcp half %a, 32768.0 |
| 75 | + ret half %r |
| 76 | +} |
| 77 | + |
| 78 | +define double @divf64_2(double %a) nounwind { |
| 79 | +; CHECK-LABEL: divf64_2: |
| 80 | +; CHECK: // %bb.0: |
| 81 | +; CHECK-NEXT: fmov d1, #2.00000000 |
| 82 | +; CHECK-NEXT: fdiv d0, d0, d1 |
| 83 | +; CHECK-NEXT: ret |
| 84 | + %r = fdiv double %a, 2.0 |
| 85 | + ret double %r |
| 86 | +} |
| 87 | + |
| 88 | +define <4 x float> @divv4f32_2(<4 x float> %a) nounwind { |
| 89 | +; CHECK-LABEL: divv4f32_2: |
| 90 | +; CHECK: // %bb.0: |
| 91 | +; CHECK-NEXT: movi v1.4s, #64, lsl #24 |
| 92 | +; CHECK-NEXT: fdiv v0.4s, v0.4s, v1.4s |
| 93 | +; CHECK-NEXT: ret |
| 94 | + %r = fdiv <4 x float> %a, <float 2.0, float 2.0, float 2.0, float 2.0> |
| 95 | + ret <4 x float> %r |
| 96 | +} |
| 97 | + |
| 98 | +define <4 x float> @divv4f32_2_arcp(<4 x float> %a) nounwind { |
| 99 | +; CHECK-LABEL: divv4f32_2_arcp: |
| 100 | +; CHECK: // %bb.0: |
| 101 | +; CHECK-NEXT: movi v1.4s, #63, lsl #24 |
| 102 | +; CHECK-NEXT: fmul v0.4s, v0.4s, v1.4s |
| 103 | +; CHECK-NEXT: ret |
| 104 | + %r = fdiv arcp <4 x float> %a, <float 2.0, float 2.0, float 2.0, float 2.0> |
| 105 | + ret <4 x float> %r |
| 106 | +} |
| 107 | + |
| 108 | +define <4 x float> @divv4f32_3(<4 x float> %a) nounwind { |
| 109 | +; CHECK-LABEL: divv4f32_3: |
| 110 | +; CHECK: // %bb.0: |
| 111 | +; CHECK-NEXT: fmov v1.4s, #3.00000000 |
| 112 | +; CHECK-NEXT: fdiv v0.4s, v0.4s, v1.4s |
| 113 | +; CHECK-NEXT: ret |
| 114 | + %r = fdiv <4 x float> %a, <float 3.0, float 3.0, float 3.0, float 3.0> |
| 115 | + ret <4 x float> %r |
| 116 | +} |
| 117 | + |
| 118 | +define <4 x float> @divv4f32_3_arcp(<4 x float> %a) nounwind { |
| 119 | +; CHECK-LABEL: divv4f32_3_arcp: |
| 120 | +; CHECK: // %bb.0: |
| 121 | +; CHECK-NEXT: mov w8, #43691 // =0xaaab |
| 122 | +; CHECK-NEXT: movk w8, #16042, lsl #16 |
| 123 | +; CHECK-NEXT: dup v1.4s, w8 |
| 124 | +; CHECK-NEXT: fmul v0.4s, v0.4s, v1.4s |
| 125 | +; CHECK-NEXT: ret |
| 126 | + %r = fdiv arcp <4 x float> %a, <float 3.0, float 3.0, float 3.0, float 3.0> |
| 127 | + ret <4 x float> %r |
| 128 | +} |
| 129 | + |
| 130 | +define <4 x float> @divv4f32_24816(<4 x float> %a) nounwind { |
| 131 | +; CHECK-LABEL: divv4f32_24816: |
| 132 | +; CHECK: // %bb.0: |
| 133 | +; CHECK-NEXT: adrp x8, .LCPI12_0 |
| 134 | +; CHECK-NEXT: ldr q1, [x8, :lo12:.LCPI12_0] |
| 135 | +; CHECK-NEXT: fdiv v0.4s, v0.4s, v1.4s |
| 136 | +; CHECK-NEXT: ret |
| 137 | + %r = fdiv <4 x float> %a, <float 2.0, float 4.0, float 8.0, float 16.0> |
| 138 | + ret <4 x float> %r |
| 139 | +} |
| 140 | + |
| 141 | +define <vscale x 4 x float> @divnxv4f32_2(<vscale x 4 x float> %a) nounwind { |
| 142 | +; CHECK-LABEL: divnxv4f32_2: |
| 143 | +; CHECK: // %bb.0: |
| 144 | +; CHECK-NEXT: fmov z1.s, #2.00000000 |
| 145 | +; CHECK-NEXT: ptrue p0.s |
| 146 | +; CHECK-NEXT: fdiv z0.s, p0/m, z0.s, z1.s |
| 147 | +; CHECK-NEXT: ret |
| 148 | + %r = fdiv <vscale x 4 x float> %a, splat (float 2.0) |
| 149 | + ret <vscale x 4 x float> %r |
| 150 | +} |
0 commit comments