138 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=loop-idiom < %s -S | FileCheck %s3 4target datalayout = "e-p:32:32:32-p1:64:64:64-p2:16:16:16-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:32-n8:16:32:64"5target triple = "x86_64-apple-darwin10.0.0"6 7; Two dimensional nested loop should be promoted to one big memset.8define void @test10(ptr addrspace(2) %X) nounwind ssp {9; CHECK-LABEL: @test10(10; CHECK-NEXT: entry:11; CHECK-NEXT: call void @llvm.memset.p2.i16(ptr addrspace(2) align 1 [[X:%.*]], i8 0, i16 10000, i1 false)12; CHECK-NEXT: br label [[BB_NPH:%.*]]13; CHECK: bb.nph:14; CHECK-NEXT: [[I_04:%.*]] = phi i16 [ 0, [[ENTRY:%.*]] ], [ [[INC12:%.*]], [[FOR_INC10:%.*]] ]15; CHECK-NEXT: [[TMP0:%.*]] = mul nuw nsw i16 [[I_04]], 10016; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr nuw i8, ptr addrspace(2) [[X]], i16 [[TMP0]]17; CHECK-NEXT: br label [[FOR_BODY5:%.*]]18; CHECK: for.body5:19; CHECK-NEXT: [[J_02:%.*]] = phi i16 [ 0, [[BB_NPH]] ], [ [[INC:%.*]], [[FOR_BODY5]] ]20; CHECK-NEXT: [[MUL:%.*]] = mul nsw i16 [[I_04]], 10021; CHECK-NEXT: [[ADD:%.*]] = add nsw i16 [[J_02]], [[MUL]]22; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr addrspace(2) [[X]], i16 [[ADD]]23; CHECK-NEXT: [[INC]] = add nsw i16 [[J_02]], 124; CHECK-NEXT: [[CMP4:%.*]] = icmp eq i16 [[INC]], 10025; CHECK-NEXT: br i1 [[CMP4]], label [[FOR_INC10]], label [[FOR_BODY5]]26; CHECK: for.inc10:27; CHECK-NEXT: [[INC12]] = add nsw i16 [[I_04]], 128; CHECK-NEXT: [[CMP:%.*]] = icmp eq i16 [[INC12]], 10029; CHECK-NEXT: br i1 [[CMP]], label [[FOR_END13:%.*]], label [[BB_NPH]]30; CHECK: for.end13:31; CHECK-NEXT: ret void32;33 34entry:35 br label %bb.nph36 37bb.nph: ; preds = %entry, %for.inc1038 %i.04 = phi i16 [ 0, %entry ], [ %inc12, %for.inc10 ]39 br label %for.body540 41for.body5: ; preds = %for.body5, %bb.nph42 %j.02 = phi i16 [ 0, %bb.nph ], [ %inc, %for.body5 ]43 %mul = mul nsw i16 %i.04, 10044 %add = add nsw i16 %j.02, %mul45 %arrayidx = getelementptr inbounds i8, ptr addrspace(2) %X, i16 %add46 store i8 0, ptr addrspace(2) %arrayidx, align 147 %inc = add nsw i16 %j.02, 148 %cmp4 = icmp eq i16 %inc, 10049 br i1 %cmp4, label %for.inc10, label %for.body550 51for.inc10: ; preds = %for.body552 %inc12 = add nsw i16 %i.04, 153 %cmp = icmp eq i16 %inc12, 10054 br i1 %cmp, label %for.end13, label %bb.nph55 56for.end13: ; preds = %for.inc1057 ret void58}59 60define void @test11_pattern(ptr addrspace(2) nocapture %P) nounwind ssp {61; CHECK-LABEL: @test11_pattern(62; CHECK-NEXT: entry:63; CHECK-NEXT: br label [[FOR_BODY:%.*]]64; CHECK: for.body:65; CHECK-NEXT: [[INDVAR:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVAR_NEXT:%.*]], [[FOR_BODY]] ]66; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr i32, ptr addrspace(2) [[P:%.*]], i64 [[INDVAR]]67; CHECK-NEXT: store i32 1, ptr addrspace(2) [[ARRAYIDX]], align 468; CHECK-NEXT: [[INDVAR_NEXT]] = add i64 [[INDVAR]], 169; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVAR_NEXT]], 1000070; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END:%.*]], label [[FOR_BODY]]71; CHECK: for.end:72; CHECK-NEXT: ret void73;74entry:75 br label %for.body76 77for.body: ; preds = %entry, %for.body78 %indvar = phi i64 [ 0, %entry ], [ %indvar.next, %for.body ]79 %arrayidx = getelementptr i32, ptr addrspace(2) %P, i64 %indvar80 store i32 1, ptr addrspace(2) %arrayidx, align 481 %indvar.next = add i64 %indvar, 182 %exitcond = icmp eq i64 %indvar.next, 1000083 br i1 %exitcond, label %for.end, label %for.body84 85for.end: ; preds = %for.body86 ret void87}88 89; PR9815 - This is a partial overlap case that cannot be safely transformed90; into a memcpy.91@g_50 = addrspace(2) global [7 x i32] [i32 0, i32 0, i32 0, i32 0, i32 1, i32 0, i32 0], align 1692 93 94define i32 @test14() nounwind {95; CHECK-LABEL: @test14(96; CHECK-NEXT: entry:97; CHECK-NEXT: br label [[FOR_BODY:%.*]]98; CHECK: for.body:99; CHECK-NEXT: [[TMP5:%.*]] = phi i32 [ [[INC:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]100; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP5]], 4101; CHECK-NEXT: [[IDXPROM:%.*]] = sext i32 [[ADD]] to i64102; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [7 x i32], ptr addrspace(2) @g_50, i32 0, i64 [[IDXPROM]]103; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr addrspace(2) [[ARRAYIDX]], align 4104; CHECK-NEXT: [[ADD4:%.*]] = add nsw i32 [[TMP5]], 5105; CHECK-NEXT: [[IDXPROM5:%.*]] = sext i32 [[ADD4]] to i64106; CHECK-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds [7 x i32], ptr addrspace(2) @g_50, i32 0, i64 [[IDXPROM5]]107; CHECK-NEXT: store i32 [[TMP2]], ptr addrspace(2) [[ARRAYIDX6]], align 4108; CHECK-NEXT: [[INC]] = add nsw i32 [[TMP5]], 1109; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[INC]], 2110; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END:%.*]]111; CHECK: for.end:112; CHECK-NEXT: [[TMP8:%.*]] = load i32, ptr addrspace(2) getelementptr inbounds ([7 x i32], ptr addrspace(2) @g_50, i32 0, i64 6), align 4113; CHECK-NEXT: ret i32 [[TMP8]]114;115 116entry:117 br label %for.body118 119for.body: ; preds = %for.inc, %for.body.lr.ph120 %tmp5 = phi i32 [ %inc, %for.body ], [ 0, %entry ]121 %add = add nsw i32 %tmp5, 4122 %idxprom = sext i32 %add to i64123 %arrayidx = getelementptr inbounds [7 x i32], ptr addrspace(2) @g_50, i32 0, i64 %idxprom124 %tmp2 = load i32, ptr addrspace(2) %arrayidx, align 4125 %add4 = add nsw i32 %tmp5, 5126 %idxprom5 = sext i32 %add4 to i64127 %arrayidx6 = getelementptr inbounds [7 x i32], ptr addrspace(2) @g_50, i32 0, i64 %idxprom5128 store i32 %tmp2, ptr addrspace(2) %arrayidx6, align 4129 %inc = add nsw i32 %tmp5, 1130 %cmp = icmp slt i32 %inc, 2131 br i1 %cmp, label %for.body, label %for.end132 133for.end: ; preds = %for.inc134 %tmp8 = load i32, ptr addrspace(2) getelementptr inbounds ([7 x i32], ptr addrspace(2) @g_50, i32 0, i64 6), align 4135 ret i32 %tmp8136}137 138