brintos

brintos / llvm-project-archived public Read only

0
0
Text · 1.9 KiB · b1b9959 Raw
55 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=aarch64-none-linux-gnu | FileCheck %s3 4define i16 @test_add_zext_v8i16(<8 x i8> %a, <8 x i8> %b) local_unnamed_addr #0 {5; CHECK-LABEL: test_add_zext_v8i16:6; CHECK:       // %bb.0:7; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q08; CHECK-NEXT:    // kill: def $d1 killed $d1 def $q19; CHECK-NEXT:    mov v0.d[1], v1.d[0]10; CHECK-NEXT:    uaddlv h0, v0.16b11; CHECK-NEXT:    umov w0, v0.h[0]12; CHECK-NEXT:    ret13  %z1 = zext <8 x i8> %a to <8 x i16>14  %z2 = zext <8 x i8> %b to <8 x i16>15  %z = add <8 x i16> %z1, %z216  %r = call i16 @llvm.vector.reduce.add.v8i16(<8 x i16> %z)17  ret i16 %r18}19 20define i32 @test_add_zext_v4i32(<4 x i16> %a, <4 x i16> %b) local_unnamed_addr #0 {21; CHECK-LABEL: test_add_zext_v4i32:22; CHECK:       // %bb.0:23; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q024; CHECK-NEXT:    // kill: def $d1 killed $d1 def $q125; CHECK-NEXT:    mov v0.d[1], v1.d[0]26; CHECK-NEXT:    uaddlv s0, v0.8h27; CHECK-NEXT:    fmov w0, s028; CHECK-NEXT:    ret29  %z1 = zext <4 x i16> %a to <4 x i32>30  %z2 = zext <4 x i16> %b to <4 x i32>31  %z = add <4 x i32> %z1, %z232  %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %z)33  ret i32 %r34}35 36define i64 @test_add_zext_v2i64(<2 x i32> %a, <2 x i32> %b) local_unnamed_addr #0 {37; CHECK-LABEL: test_add_zext_v2i64:38; CHECK:       // %bb.0:39; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q040; CHECK-NEXT:    // kill: def $d1 killed $d1 def $q141; CHECK-NEXT:    mov v0.d[1], v1.d[0]42; CHECK-NEXT:    uaddlv d0, v0.4s43; CHECK-NEXT:    fmov x0, d044; CHECK-NEXT:    ret45  %z1 = zext <2 x i32> %a to <2 x i64>46  %z2 = zext <2 x i32> %b to <2 x i64>47  %z = add <2 x i64> %z1, %z248  %r = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> %z)49  ret i64 %r50}51 52declare i16 @llvm.vector.reduce.add.v8i16(<8 x i16>)53declare i32 @llvm.vector.reduce.add.v4i32(<4 x i32>)54declare i64 @llvm.vector.reduce.add.v2i64(<2 x i64>)55