154 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -verify-machineinstrs -mtriple=aarch64-none-linux-gnu -global-isel=0 | FileCheck %s --check-prefixes=CHECK,CHECK-SD3; RUN: llc < %s -verify-machineinstrs -mtriple=aarch64-none-linux-gnu -global-isel=1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI4 5define <2 x i32> @shuffle(ptr %P) {6; CHECK-SD-LABEL: shuffle:7; CHECK-SD: // %bb.0:8; CHECK-SD-NEXT: ld1r { v0.2s }, [x0]9; CHECK-SD-NEXT: ret10;11; CHECK-GI-LABEL: shuffle:12; CHECK-GI: // %bb.0:13; CHECK-GI-NEXT: ldr d0, [x0]14; CHECK-GI-NEXT: dup v0.2s, v0.s[0]15; CHECK-GI-NEXT: ret16 %lv2i32 = load <2 x i32>, ptr %P17 %B = shufflevector <2 x i32> %lv2i32, <2 x i32> undef, <2 x i32> zeroinitializer18 ret <2 x i32> %B19}20 21define <4 x i32> @shuffle2(ptr %P) {22; CHECK-LABEL: shuffle2:23; CHECK: // %bb.0:24; CHECK-NEXT: ld1r { v0.4s }, [x0]25; CHECK-NEXT: ret26 %lv2i32 = load <4 x i32>, ptr %P27 %B = shufflevector <4 x i32> %lv2i32, <4 x i32> undef, <4 x i32> zeroinitializer28 ret <4 x i32> %B29}30 31define <4 x i32> @shuffle2_multiuse(ptr %P) {32; CHECK-SD-LABEL: shuffle2_multiuse:33; CHECK-SD: // %bb.0:34; CHECK-SD-NEXT: ldr q0, [x0]35; CHECK-SD-NEXT: dup v1.4s, v0.s[0]36; CHECK-SD-NEXT: dup v0.4s, v0.s[1]37; CHECK-SD-NEXT: add v0.4s, v1.4s, v0.4s38; CHECK-SD-NEXT: ret39;40; CHECK-GI-LABEL: shuffle2_multiuse:41; CHECK-GI: // %bb.0:42; CHECK-GI-NEXT: ldr q0, [x0]43; CHECK-GI-NEXT: ld1r { v1.4s }, [x0]44; CHECK-GI-NEXT: dup v0.4s, v0.s[1]45; CHECK-GI-NEXT: add v0.4s, v1.4s, v0.4s46; CHECK-GI-NEXT: ret47 %lv2i32 = load <4 x i32>, ptr %P48 %B = shufflevector <4 x i32> %lv2i32, <4 x i32> undef, <4 x i32> zeroinitializer49 %C = shufflevector <4 x i32> %lv2i32, <4 x i32> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 1>50 %D = add <4 x i32> %B, %C51 ret <4 x i32> %D52}53 54define <4 x i16> @shuffle3(ptr %P) {55; CHECK-SD-LABEL: shuffle3:56; CHECK-SD: // %bb.0:57; CHECK-SD-NEXT: ld1r { v0.4h }, [x0]58; CHECK-SD-NEXT: ret59;60; CHECK-GI-LABEL: shuffle3:61; CHECK-GI: // %bb.0:62; CHECK-GI-NEXT: ldr d0, [x0]63; CHECK-GI-NEXT: dup v0.4h, v0.h[0]64; CHECK-GI-NEXT: ret65 %lv4i16 = load <4 x i16>, ptr %P66 %sv4i16 = shufflevector <4 x i16> %lv4i16, <4 x i16> undef, <4 x i32> zeroinitializer67 ret <4 x i16> %sv4i1668}69 70define <8 x i16> @shuffle4(ptr %P) {71; CHECK-LABEL: shuffle4:72; CHECK: // %bb.0:73; CHECK-NEXT: ld1r { v0.8h }, [x0]74; CHECK-NEXT: ret75 %lv8i16 = load <8 x i16>, ptr %P76 %sv8i16 = shufflevector <8 x i16> %lv8i16, <8 x i16> undef, <8 x i32> zeroinitializer77 ret <8 x i16> %sv8i1678}79 80define <8 x i8> @shuffle5(ptr %P) {81; CHECK-SD-LABEL: shuffle5:82; CHECK-SD: // %bb.0:83; CHECK-SD-NEXT: ld1r { v0.8b }, [x0]84; CHECK-SD-NEXT: ret85;86; CHECK-GI-LABEL: shuffle5:87; CHECK-GI: // %bb.0:88; CHECK-GI-NEXT: ldr d0, [x0]89; CHECK-GI-NEXT: dup v0.8b, v0.b[0]90; CHECK-GI-NEXT: ret91 %lv8i8 = load <8 x i8>, ptr %P92 %sv8i8 = shufflevector <8 x i8> %lv8i8, <8 x i8> undef, <8 x i32> zeroinitializer93 ret <8 x i8> %sv8i894}95 96define <16 x i8> @shuffle6(ptr %P) {97; CHECK-LABEL: shuffle6:98; CHECK: // %bb.0:99; CHECK-NEXT: ld1r { v0.16b }, [x0]100; CHECK-NEXT: ret101 %lv16i8 = load <16 x i8>, ptr %P102 %sv16i8 = shufflevector <16 x i8> %lv16i8, <16 x i8> undef, <16 x i32> zeroinitializer103 ret <16 x i8> %sv16i8104}105 106define <2 x i64> @shuffle7(ptr %P) {107; CHECK-LABEL: shuffle7:108; CHECK: // %bb.0:109; CHECK-NEXT: ld1r { v0.2d }, [x0]110; CHECK-NEXT: ret111 %lv2i64 = load <2 x i64>, ptr %P112 %sv2i64 = shufflevector <2 x i64> %lv2i64, <2 x i64> undef, <2 x i32> zeroinitializer113 ret <2 x i64> %sv2i64114}115 116define <2 x ptr> @shuffle8(ptr %P) {117; CHECK-SD-LABEL: shuffle8:118; CHECK-SD: // %bb.0:119; CHECK-SD-NEXT: ld1r { v0.2d }, [x0]120; CHECK-SD-NEXT: ret121;122; CHECK-GI-LABEL: shuffle8:123; CHECK-GI: // %bb.0:124; CHECK-GI-NEXT: ldr q0, [x0]125; CHECK-GI-NEXT: dup v0.2d, v0.d[0]126; CHECK-GI-NEXT: ret127 %lv2ptr = load <2 x ptr>, ptr %P128 %sv2ptr = shufflevector <2 x ptr> %lv2ptr, <2 x ptr> undef, <2 x i32> zeroinitializer129 ret <2 x ptr> %sv2ptr130}131 132define <4 x i32> @multiblock_aliasing(ptr %P, i1 %c) {133; CHECK-LABEL: multiblock_aliasing:134; CHECK: // %bb.0: // %entry135; CHECK-NEXT: ldr q0, [x0]136; CHECK-NEXT: tbz w1, #0, .LBB9_2137; CHECK-NEXT: // %bb.1: // %then138; CHECK-NEXT: stp xzr, xzr, [x0]139; CHECK-NEXT: .LBB9_2: // %else140; CHECK-NEXT: dup v0.4s, v0.s[0]141; CHECK-NEXT: ret142entry:143 %lv2ptr = load <4 x i32>, ptr %P144 br i1 %c, label %then, label %else145 146then:147 store <4 x i32> zeroinitializer, ptr %P148 br label %else149 150else:151 %sv2ptr = shufflevector <4 x i32> %lv2ptr, <4 x i32> undef, <4 x i32> zeroinitializer152 ret <4 x i32> %sv2ptr153}154