1 ; RUN: opt -mcpu=skx -S -passes=loop-vectorize -force-vector-width=8 -force-vector-interleave=1 -enable-interleaved-mem-accesses < %s | FileCheck %s
3 target datalayout = "e-m:e-p:32:32-f64:32:64-f80:32-n8:16:32-S128"
5 ; This test checks the fix for PR39099.
7 ; Check that the predicated load is not vectorized as an
8 ; interleaved-group (which requires proper masking, currently unsupported)
9 ; but rather as a scalarized accesses.
10 ; (For SKX, Gather is not supported by the compiler for chars, therefore
11 ; the only remaining alternative is to scalarize).
13 ; void masked_strided(const unsigned char* restrict p,
14 ; unsigned char* restrict q,
15 ; unsigned char guard) {
16 ; for(ix=0; ix < 1024; ++ix) {
24 ;CHECK-LABEL: @masked_strided(
26 ;CHECK-NEXT: %index = phi i32
27 ;CHECK-NEXT: %[[VECIND:.+]] = phi <8 x i32> [ <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
28 ;CHECK-NEXT: %[[VMASK:.+]] = icmp ugt <8 x i32> %[[VECIND]], %{{broadcast.splat*}}
29 ;CHECK-NEXT: %{{.*}} = shl nuw nsw <8 x i32> %[[VECIND]], splat (i32 1)
30 ;CHECK-NEXT: %[[M:.+]] = extractelement <8 x i1> %[[VMASK]], i32 0
31 ;CHECK-NEXT: br i1 %[[M]], label %pred.store.if, label %pred.store.continue
32 ;CHECK-NOT: %{{.+}} = load <16 x i8>, ptr %{{.*}}, align 1
34 define dso_local void @masked_strided(ptr noalias nocapture readonly %p, ptr noalias nocapture %q, i8 zeroext %guard) local_unnamed_addr {
36 %conv = zext i8 %guard to i32
40 %ix.09 = phi i32 [ 0, %entry ], [ %inc, %for.inc ]
41 %cmp1 = icmp ugt i32 %ix.09, %conv
42 br i1 %cmp1, label %if.then, label %for.inc
45 %mul = shl nuw nsw i32 %ix.09, 1
46 %arrayidx = getelementptr inbounds i8, ptr %p, i32 %mul
47 %0 = load i8, ptr %arrayidx, align 1
48 %arrayidx3 = getelementptr inbounds i8, ptr %q, i32 %ix.09
49 store i8 %0, ptr %arrayidx3, align 1
53 %inc = add nuw nsw i32 %ix.09, 1
54 %exitcond = icmp eq i32 %inc, 1024
55 br i1 %exitcond, label %for.end, label %for.body