|
| 1 | +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4 |
| 2 | +; RUN: llc -mtriple=aarch64 < %s | FileCheck %s |
| 3 | + |
| 4 | +declare <8 x i16> @llvm.aarch64.neon.uhadd.v8i16(<8 x i16>, <8 x i16>) |
| 5 | +declare <8 x i16> @llvm.aarch64.neon.urhadd.v8i16(<8 x i16>, <8 x i16>) |
| 6 | +declare <8 x i16> @llvm.aarch64.neon.shadd.v8i16(<8 x i16>, <8 x i16>) |
| 7 | +declare <8 x i16> @llvm.aarch64.neon.srhadd.v8i16(<8 x i16>, <8 x i16>) |
| 8 | + |
| 9 | +define <8 x i16> @haddu_zext(<8 x i8> %a0, <8 x i8> %a1) { |
| 10 | +; CHECK-LABEL: haddu_zext: |
| 11 | +; CHECK: // %bb.0: |
| 12 | +; CHECK-NEXT: ushll v0.8h, v0.8b, #0 |
| 13 | +; CHECK-NEXT: ushll v1.8h, v1.8b, #0 |
| 14 | +; CHECK-NEXT: uhadd v0.8h, v0.8h, v1.8h |
| 15 | +; CHECK-NEXT: bic v0.8h, #254, lsl #8 |
| 16 | +; CHECK-NEXT: ret |
| 17 | + %x0 = zext <8 x i8> %a0 to <8 x i16> |
| 18 | + %x1 = zext <8 x i8> %a1 to <8 x i16> |
| 19 | + %hadd = call <8 x i16> @llvm.aarch64.neon.uhadd.v8i16(<8 x i16> %x0, <8 x i16> %x1) |
| 20 | + %res = and <8 x i16> %hadd, <i16 511, i16 511, i16 511, i16 511,i16 511, i16 511, i16 511, i16 511> |
| 21 | + ret <8 x i16> %res |
| 22 | +} |
| 23 | + |
| 24 | +define <8 x i16> @rhaddu_zext(<8 x i8> %a0, <8 x i8> %a1) { |
| 25 | +; CHECK-LABEL: rhaddu_zext: |
| 26 | +; CHECK: // %bb.0: |
| 27 | +; CHECK-NEXT: ushll v0.8h, v0.8b, #0 |
| 28 | +; CHECK-NEXT: ushll v1.8h, v1.8b, #0 |
| 29 | +; CHECK-NEXT: urhadd v0.8h, v0.8h, v1.8h |
| 30 | +; CHECK-NEXT: bic v0.8h, #254, lsl #8 |
| 31 | +; CHECK-NEXT: ret |
| 32 | + %x0 = zext <8 x i8> %a0 to <8 x i16> |
| 33 | + %x1 = zext <8 x i8> %a1 to <8 x i16> |
| 34 | + %hadd = call <8 x i16> @llvm.aarch64.neon.urhadd.v8i16(<8 x i16> %x0, <8 x i16> %x1) |
| 35 | + %res = and <8 x i16> %hadd, <i16 511, i16 511, i16 511, i16 511, i16 511, i16 511, i16 511, i16 511> |
| 36 | + ret <8 x i16> %res |
| 37 | +} |
| 38 | + |
| 39 | +define <8 x i16> @hadds_zext(<8 x i8> %a0, <8 x i8> %a1) { |
| 40 | +; CHECK-LABEL: hadds_zext: |
| 41 | +; CHECK: // %bb.0: |
| 42 | +; CHECK-NEXT: ushll v0.8h, v0.8b, #0 |
| 43 | +; CHECK-NEXT: ushll v1.8h, v1.8b, #0 |
| 44 | +; CHECK-NEXT: shadd v0.8h, v0.8h, v1.8h |
| 45 | +; CHECK-NEXT: bic v0.8h, #254, lsl #8 |
| 46 | +; CHECK-NEXT: ret |
| 47 | + %x0 = zext <8 x i8> %a0 to <8 x i16> |
| 48 | + %x1 = zext <8 x i8> %a1 to <8 x i16> |
| 49 | + %hadd = call <8 x i16> @llvm.aarch64.neon.shadd.v8i16(<8 x i16> %x0, <8 x i16> %x1) |
| 50 | + %res = and <8 x i16> %hadd, <i16 511, i16 511, i16 511, i16 511, i16 511, i16 511, i16 511, i16 511> |
| 51 | + ret <8 x i16> %res |
| 52 | +} |
| 53 | + |
| 54 | +define <8 x i16> @shaddu_zext(<8 x i8> %a0, <8 x i8> %a1) { |
| 55 | +; CHECK-LABEL: shaddu_zext: |
| 56 | +; CHECK: // %bb.0: |
| 57 | +; CHECK-NEXT: ushll v0.8h, v0.8b, #0 |
| 58 | +; CHECK-NEXT: ushll v1.8h, v1.8b, #0 |
| 59 | +; CHECK-NEXT: srhadd v0.8h, v0.8h, v1.8h |
| 60 | +; CHECK-NEXT: bic v0.8h, #254, lsl #8 |
| 61 | +; CHECK-NEXT: ret |
| 62 | + %x0 = zext <8 x i8> %a0 to <8 x i16> |
| 63 | + %x1 = zext <8 x i8> %a1 to <8 x i16> |
| 64 | + %hadd = call <8 x i16> @llvm.aarch64.neon.srhadd.v8i16(<8 x i16> %x0, <8 x i16> %x1) |
| 65 | + %res = and <8 x i16> %hadd, <i16 511, i16 511, i16 511, i16 511, i16 511, i16 511, i16 511, i16 511> |
| 66 | + ret <8 x i16> %res |
| 67 | +} |
| 68 | + |
| 69 | +; ; negative tests |
| 70 | + |
| 71 | +define <8 x i16> @haddu_sext(<8 x i8> %a0, <8 x i8> %a1) { |
| 72 | +; CHECK-LABEL: haddu_sext: |
| 73 | +; CHECK: // %bb.0: |
| 74 | +; CHECK-NEXT: sshll v0.8h, v0.8b, #0 |
| 75 | +; CHECK-NEXT: sshll v1.8h, v1.8b, #0 |
| 76 | +; CHECK-NEXT: uhadd v0.8h, v0.8h, v1.8h |
| 77 | +; CHECK-NEXT: bic v0.8h, #254, lsl #8 |
| 78 | +; CHECK-NEXT: ret |
| 79 | + %x0 = sext <8 x i8> %a0 to <8 x i16> |
| 80 | + %x1 = sext <8 x i8> %a1 to <8 x i16> |
| 81 | + %hadd = call <8 x i16> @llvm.aarch64.neon.uhadd.v8i16(<8 x i16> %x0, <8 x i16> %x1) |
| 82 | + %res = and <8 x i16> %hadd, <i16 511, i16 511, i16 511, i16 511,i16 511, i16 511, i16 511, i16 511> |
| 83 | + ret <8 x i16> %res |
| 84 | +} |
| 85 | + |
| 86 | +define <8 x i16> @urhadd_sext(<8 x i8> %a0, <8 x i8> %a1) { |
| 87 | +; CHECK-LABEL: urhadd_sext: |
| 88 | +; CHECK: // %bb.0: |
| 89 | +; CHECK-NEXT: sshll v0.8h, v0.8b, #0 |
| 90 | +; CHECK-NEXT: sshll v1.8h, v1.8b, #0 |
| 91 | +; CHECK-NEXT: urhadd v0.8h, v0.8h, v1.8h |
| 92 | +; CHECK-NEXT: bic v0.8h, #254, lsl #8 |
| 93 | +; CHECK-NEXT: ret |
| 94 | + %x0 = sext <8 x i8> %a0 to <8 x i16> |
| 95 | + %x1 = sext <8 x i8> %a1 to <8 x i16> |
| 96 | + %hadd = call <8 x i16> @llvm.aarch64.neon.urhadd.v8i16(<8 x i16> %x0, <8 x i16> %x1) |
| 97 | + %res = and <8 x i16> %hadd, <i16 511, i16 511, i16 511, i16 511,i16 511, i16 511, i16 511, i16 511> |
| 98 | + ret <8 x i16> %res |
| 99 | +} |
| 100 | + |
| 101 | +define <8 x i16> @hadds_sext(<8 x i8> %a0, <8 x i8> %a1) { |
| 102 | +; CHECK-LABEL: hadds_sext: |
| 103 | +; CHECK: // %bb.0: |
| 104 | +; CHECK-NEXT: sshll v0.8h, v0.8b, #0 |
| 105 | +; CHECK-NEXT: sshll v1.8h, v1.8b, #0 |
| 106 | +; CHECK-NEXT: shadd v0.8h, v0.8h, v1.8h |
| 107 | +; CHECK-NEXT: bic v0.8h, #254, lsl #8 |
| 108 | +; CHECK-NEXT: ret |
| 109 | + %x0 = sext <8 x i8> %a0 to <8 x i16> |
| 110 | + %x1 = sext <8 x i8> %a1 to <8 x i16> |
| 111 | + %hadd = call <8 x i16> @llvm.aarch64.neon.shadd.v8i16(<8 x i16> %x0, <8 x i16> %x1) |
| 112 | + %res = and <8 x i16> %hadd, <i16 511, i16 511, i16 511, i16 511, i16 511, i16 511, i16 511, i16 511> |
| 113 | + ret <8 x i16> %res |
| 114 | +} |
| 115 | + |
| 116 | +define <8 x i16> @shaddu_sext(<8 x i8> %a0, <8 x i8> %a1) { |
| 117 | +; CHECK-LABEL: shaddu_sext: |
| 118 | +; CHECK: // %bb.0: |
| 119 | +; CHECK-NEXT: sshll v0.8h, v0.8b, #0 |
| 120 | +; CHECK-NEXT: sshll v1.8h, v1.8b, #0 |
| 121 | +; CHECK-NEXT: srhadd v0.8h, v0.8h, v1.8h |
| 122 | +; CHECK-NEXT: bic v0.8h, #254, lsl #8 |
| 123 | +; CHECK-NEXT: ret |
| 124 | + %x0 = sext <8 x i8> %a0 to <8 x i16> |
| 125 | + %x1 = sext <8 x i8> %a1 to <8 x i16> |
| 126 | + %hadd = call <8 x i16> @llvm.aarch64.neon.srhadd.v8i16(<8 x i16> %x0, <8 x i16> %x1) |
| 127 | + %res = and <8 x i16> %hadd, <i16 511, i16 511, i16 511, i16 511, i16 511, i16 511, i16 511, i16 511> |
| 128 | + ret <8 x i16> %res |
| 129 | +} |
0 commit comments