1 ; RUN: opt -loop-vectorize -vectorizer-maximize-bandwidth -mcpu=corei7-avx -debug-only=loop-vectorize -S < %s 2>&1 | FileCheck %s --check-prefix=CHECK-AVX1
2 ; RUN: opt -loop-vectorize -vectorizer-maximize-bandwidth -mcpu=core-avx2 -debug-only=loop-vectorize -S < %s 2>&1 | FileCheck %s --check-prefix=CHECK-AVX2
5 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
6 target triple = "x86_64-unknown-linux-gnu"
8 @a = global [1000 x i8] zeroinitializer, align 16
9 @b = global [1000 x i8] zeroinitializer, align 16
10 @c = global [1000 x i8] zeroinitializer, align 16
11 @u = global [1000 x i32] zeroinitializer, align 16
12 @v = global [1000 x i32] zeroinitializer, align 16
13 @w = global [1000 x i32] zeroinitializer, align 16
15 ; Tests that the vectorization factor is determined by the smallest instead of
16 ; widest type in the loop for maximum bandwidth when
17 ; -vectorizer-maximize-bandwidth is indicated.
20 ; CHECK-AVX1: LV: Selecting VF: 16.
21 ; CHECK-AVX2: LV: Selecting VF: 32.
30 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
31 %arrayidx = getelementptr inbounds [1000 x i8], [1000 x i8]* @b, i64 0, i64 %indvars.iv
32 %0 = load i8, i8* %arrayidx, align 1
33 %arrayidx2 = getelementptr inbounds [1000 x i8], [1000 x i8]* @c, i64 0, i64 %indvars.iv
34 %1 = load i8, i8* %arrayidx2, align 1
36 %arrayidx6 = getelementptr inbounds [1000 x i8], [1000 x i8]* @a, i64 0, i64 %indvars.iv
37 store i8 %add, i8* %arrayidx6, align 1
38 %arrayidx8 = getelementptr inbounds [1000 x i32], [1000 x i32]* @v, i64 0, i64 %indvars.iv
39 %2 = load i32, i32* %arrayidx8, align 4
40 %arrayidx10 = getelementptr inbounds [1000 x i32], [1000 x i32]* @w, i64 0, i64 %indvars.iv
41 %3 = load i32, i32* %arrayidx10, align 4
42 %add11 = add nsw i32 %3, %2
43 %arrayidx13 = getelementptr inbounds [1000 x i32], [1000 x i32]* @u, i64 0, i64 %indvars.iv
44 store i32 %add11, i32* %arrayidx13, align 4
45 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
46 %exitcond = icmp eq i64 %indvars.iv.next, 1000
47 br i1 %exitcond, label %for.cond.cleanup, label %for.body
50 ; We should not choose a VF larger than the constant TC.
51 ; VF chosen should be atmost 16 (not the max possible vector width = 32 for AVX2)
52 define void @not_too_small_tc(i8* noalias nocapture %A, i8* noalias nocapture readonly %B) {
53 ; CHECK-LABEL: not_too_small_tc
54 ; CHECK-AVX2: LV: Selecting VF: 16.
59 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
60 %arrayidx = getelementptr inbounds i8, i8* %B, i64 %indvars.iv
61 %l1 = load i8, i8* %arrayidx, align 4, !llvm.access.group !13
62 %arrayidx2 = getelementptr inbounds i8, i8* %A, i64 %indvars.iv
63 %l2 = load i8, i8* %arrayidx2, align 4, !llvm.access.group !13
64 %add = add i8 %l1, %l2
65 store i8 %add, i8* %arrayidx2, align 4, !llvm.access.group !13
66 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
67 %exitcond = icmp eq i64 %indvars.iv.next, 16
68 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !4
73 !3 = !{!3, !{!"llvm.loop.parallel_accesses", !13}}