102 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "scalar.ph:" --version 52; RUN: opt -S -passes=loop-vectorize -force-vector-width=4 -force-vector-interleave=1 < %s | FileCheck %s3 4target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"5 6; PR394177; Check that the need for overflow check prevents vectorizing a loop with tiny8; trip count (which implies opt for size).9define void @func_34() {10; CHECK-LABEL: define void @func_34() {11; CHECK-NEXT: [[ENTRY:.*]]:12; CHECK-NEXT: br label %[[LOOP:.*]]13; CHECK: [[LOOP]]:14; CHECK-NEXT: [[IV:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]15; CHECK-NEXT: [[SEXT:%.*]] = shl i32 [[IV]], 1616; CHECK-NEXT: [[STEP:%.*]] = ashr exact i32 [[SEXT]], 1617; CHECK-NEXT: [[IV_NEXT]] = add nsw i32 [[STEP]], 118; CHECK-NEXT: [[IV_NEXT_TRUNC:%.*]] = trunc i32 [[IV_NEXT]] to i1619; CHECK-NEXT: [[EC:%.*]] = icmp slt i16 [[IV_NEXT_TRUNC]], 320; CHECK-NEXT: br i1 [[EC]], label %[[LOOP]], label %[[EXIT:.*]]21; CHECK: [[EXIT]]:22; CHECK-NEXT: ret void23;24entry:25 br label %loop26 27loop:28 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]29 %sext = shl i32 %iv, 1630 %step = ashr exact i32 %sext, 1631 %iv.next = add nsw i32 %step, 132 %iv.next.trunc = trunc i32 %iv.next to i1633 %ec = icmp slt i16 %iv.next.trunc, 334 br i1 %ec, label %loop, label %exit35 36exit:37 ret void38}39 40; Check that a loop under opt-for-size is vectorized, w/o checking for41; stride==1.42; NOTE: Some assertions have been autogenerated by utils/update_test_checks.py43define void @scev4stride1(ptr noalias nocapture %a, ptr noalias nocapture readonly %b, i32 %k) #0 {44; CHECK-LABEL: define void @scev4stride1(45; CHECK-SAME: ptr noalias captures(none) [[A:%.*]], ptr noalias readonly captures(none) [[B:%.*]], i32 [[K:%.*]]) #[[ATTR0:[0-9]+]] {46; CHECK-NEXT: [[ENTRY:.*:]]47; CHECK-NEXT: br label %[[VECTOR_PH:.*]]48; CHECK: [[VECTOR_PH]]:49; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[K]], i64 050; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer51; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]52; CHECK: [[VECTOR_BODY]]:53; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]54; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]55; CHECK-NEXT: [[TMP4:%.*]] = mul nsw <4 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]56; CHECK-NEXT: [[TMP5:%.*]] = extractelement <4 x i32> [[TMP4]], i32 057; CHECK-NEXT: [[TMP7:%.*]] = extractelement <4 x i32> [[TMP4]], i32 158; CHECK-NEXT: [[TMP9:%.*]] = extractelement <4 x i32> [[TMP4]], i32 259; CHECK-NEXT: [[TMP11:%.*]] = extractelement <4 x i32> [[TMP4]], i32 360; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[B]], i32 [[TMP5]]61; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[B]], i32 [[TMP7]]62; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[B]], i32 [[TMP9]]63; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds i32, ptr [[B]], i32 [[TMP11]]64; CHECK-NEXT: [[TMP13:%.*]] = load i32, ptr [[TMP6]], align 465; CHECK-NEXT: [[TMP14:%.*]] = load i32, ptr [[TMP8]], align 466; CHECK-NEXT: [[TMP15:%.*]] = load i32, ptr [[TMP10]], align 467; CHECK-NEXT: [[TMP16:%.*]] = load i32, ptr [[TMP12]], align 468; CHECK-NEXT: [[TMP17:%.*]] = insertelement <4 x i32> poison, i32 [[TMP13]], i32 069; CHECK-NEXT: [[TMP18:%.*]] = insertelement <4 x i32> [[TMP17]], i32 [[TMP14]], i32 170; CHECK-NEXT: [[TMP19:%.*]] = insertelement <4 x i32> [[TMP18]], i32 [[TMP15]], i32 271; CHECK-NEXT: [[TMP20:%.*]] = insertelement <4 x i32> [[TMP19]], i32 [[TMP16]], i32 372; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[INDEX]]73; CHECK-NEXT: store <4 x i32> [[TMP20]], ptr [[TMP21]], align 474; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 475; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw nsw <4 x i32> [[VEC_IND]], splat (i32 4)76; CHECK-NEXT: [[TMP24:%.*]] = icmp eq i32 [[INDEX_NEXT]], 102477; CHECK-NEXT: br i1 [[TMP24]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]78; CHECK: [[MIDDLE_BLOCK]]:79; CHECK-NEXT: br label %[[EXIT:.*]]80; CHECK: [[EXIT]]:81; CHECK-NEXT: ret void82;83entry:84 br label %loop85 86loop:87 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]88 %mul = mul nsw i32 %iv, %k89 %gep.b = getelementptr inbounds i32, ptr %b, i32 %mul90 %0 = load i32, ptr %gep.b, align 491 %gep.a = getelementptr inbounds i32, ptr %a, i32 %iv92 store i32 %0, ptr %gep.a, align 493 %iv.next = add nuw nsw i32 %iv, 194 %ec = icmp eq i32 %iv.next, 102495 br i1 %ec, label %exit, label %loop96 97exit:98 ret void99}100 101attributes #0 = { optsize }102