35 lines · plain
1; RUN: opt -mtriple=amdgcn-- -mcpu=gfx90a -passes=loop-vectorize %s -S -o - | FileCheck %s2 3; CHECK-LABEL: @test4; CHECK-LABEL: vector.body:5; CHECK: br i1 %{{[0-9]+}}, label %middle.block, label %vector.body, !llvm.loop !06; CHECK-LABEL: middle.block:7; CHECK-LABEL: scalar.ph:8; CHECK-LABEL: loop.body:9; CHECK: br i1 %cond, label %exit, label %loop.body, !llvm.loop !210; CHECK: !0 = distinct !{!0, !1}11; CHECK: !1 = !{!"llvm.loop.isvectorized", i32 1}12; CHECK: !2 = distinct !{!2, !3, !1}13; CHECK: !3 = !{!"llvm.loop.unroll.runtime.disable"}14 15 16define amdgpu_kernel void @test(ptr addrspace(1) %out, ptr addrspace(3) %lds, i32 %n) {17 18entry:19 br label %loop.body20 21loop.body:22 %counter = phi i32 [0, %entry], [%inc, %loop.body]23 %ptr_lds = getelementptr i32, ptr addrspace(3) %lds, i32 %counter24 %val = load i32, ptr addrspace(3) %ptr_lds25 %ptr_out = getelementptr i32, ptr addrspace(1) %out, i32 %counter26 store i32 %val, ptr addrspace(1) %ptr_out27 %inc = add i32 %counter, 128 %cond = icmp sge i32 %counter, %n29 br i1 %cond, label %exit, label %loop.body30 31exit:32 ret void33}34 35