29 lines · plain
1; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=fiji -passes=loop-vectorize < %s | FileCheck %s2 3 4; For AMDGPU, loop unroll in loop vectorizer is disabled when VF==1.5;6; CHECK-LABEL: @small_loop(7; CHECK: store i328; CHECK-NOT: store i329; CHECK: ret10define amdgpu_kernel void @small_loop(ptr nocapture %inArray, i32 %size) nounwind {11entry:12 %0 = icmp sgt i32 %size, 013 br i1 %0, label %loop, label %exit14 15loop: ; preds = %entry, %loop16 %iv = phi i32 [ %iv1, %loop ], [ 0, %entry ]17 %1 = getelementptr inbounds i32, ptr %inArray, i32 %iv18 %2 = load i32, ptr %1, align 419 %3 = add nsw i32 %2, 620 store i32 %3, ptr %1, align 421 %iv1 = add i32 %iv, 122; %lftr.wideiv = trunc i64 %indvars.iv.next to i3223 %cond = icmp eq i32 %iv1, %size24 br i1 %cond, label %exit, label %loop25 26exit: ; preds = %loop, %entry27 ret void28}29