1; RUN: opt -mcpu=skx -S -loop-vectorize -force-vector-width=8 -force-vector-interleave=1 -enable-interleaved-mem-accesses < %s | FileCheck %s 2 3target datalayout = "e-m:e-p:32:32-f64:32:64-f80:32-n8:16:32-S128" 4 5; This test checks the fix for PR39099. 6; 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). 12; 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) { 17; if (ix > guard) { 18; char t = p[2*ix]; 19; q[ix] = t; 20; } 21; } 22; } 23 24;CHECK-LABEL: @masked_strided( 25;CHECK: vector.body: 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]], <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1> 30;CHECK-NEXT: %[[M:.+]] = extractelement <8 x i1> %[[VMASK]], i32 0 31;CHECK-NEXT: br i1 %[[M]], label %pred.load.if, label %pred.load.continue 32;CHECK-NOT: %[[WIDEVEC:.+]] = load <16 x i8>, <16 x i8>* %{{.*}}, align 1 33;CHECK-NOT: %{{.*}} = shufflevector <16 x i8> %[[WIDEVEC]], <16 x i8> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14> 34 35define dso_local void @masked_strided(i8* noalias nocapture readonly %p, i8* noalias nocapture %q, i8 zeroext %guard) local_unnamed_addr { 36entry: 37 %conv = zext i8 %guard to i32 38 br label %for.body 39 40for.body: 41 %ix.09 = phi i32 [ 0, %entry ], [ %inc, %for.inc ] 42 %cmp1 = icmp ugt i32 %ix.09, %conv 43 br i1 %cmp1, label %if.then, label %for.inc 44 45if.then: 46 %mul = shl nuw nsw i32 %ix.09, 1 47 %arrayidx = getelementptr inbounds i8, i8* %p, i32 %mul 48 %0 = load i8, i8* %arrayidx, align 1 49 %arrayidx3 = getelementptr inbounds i8, i8* %q, i32 %ix.09 50 store i8 %0, i8* %arrayidx3, align 1 51 br label %for.inc 52 53for.inc: 54 %inc = add nuw nsw i32 %ix.09, 1 55 %exitcond = icmp eq i32 %inc, 1024 56 br i1 %exitcond, label %for.end, label %for.body 57 58for.end: 59 ret void 60} 61