80 lines · plain
1; RUN: opt %s -passes=loop-vectorize -force-vector-interleave=3 -force-vector-width=4 -S | FileCheck --check-prefix=UF3 %s2; RUN: opt %s -passes=loop-vectorize -force-vector-interleave=5 -force-vector-width=4 -S | FileCheck --check-prefix=UF5 %s3 4define i32 @reduction_sum(i64 %n, ptr noalias nocapture %A) {5; UF3-LABEL: vector.body:6; UF3-NEXT: [[IV:%.+]] = phi i64 [ 0, %vector.ph ], [ [[IV_NEXT:%.+]], %vector.body ]7; UF3-NEXT: [[SUM0:%.+]] = phi <4 x i32> [ zeroinitializer, %vector.ph ], [ [[SUM0_NEXT:%.+]], %vector.body ]8; UF3-NEXT: [[SUM1:%.+]] = phi <4 x i32> [ zeroinitializer, %vector.ph ], [ [[SUM1_NEXT:%.+]], %vector.body ]9; UF3-NEXT: [[SUM2:%.+]] = phi <4 x i32> [ zeroinitializer, %vector.ph ], [ [[SUM2_NEXT:%.+]], %vector.body ]10; UF3-NEXT: [[GEP0:%.+]] = getelementptr inbounds i32, ptr %A, i64 [[IV]]11; UF3-NEXT: [[L_GEP1:%.+]] = getelementptr inbounds i32, ptr [[GEP0]], i64 412; UF3-NEXT: [[L_GEP2:%.+]] = getelementptr inbounds i32, ptr [[GEP0]], i64 813; UF3-NEXT: [[L0:%.+]] = load <4 x i32>, ptr [[GEP0]], align 414; UF3-NEXT: [[L1:%.+]] = load <4 x i32>, ptr [[L_GEP1]], align 415; UF3-NEXT: [[L2:%.+]] = load <4 x i32>, ptr [[L_GEP2]], align 416; UF3-NEXT: [[SUM0_NEXT]] = add <4 x i32> [[SUM0]], [[L0]]17; UF3-NEXT: [[SUM1_NEXT]] = add <4 x i32> [[SUM1]], [[L1]]18; UF3-NEXT: [[SUM2_NEXT]] = add <4 x i32> [[SUM2]], [[L2]]19; UF3-NEXT: [[IV_NEXT]] = add nuw i64 [[IV]], 1220; UF3-NEXT: [[EC:%.+]] = icmp eq i64 [[IV_NEXT]], %n.vec21; UF3-NEXT: br i1 [[EC]], label %middle.block, label %vector.body22;23; UF3-LABEL: middle.block:24; UF3-NEXT: [[RDX0:%.+]] = add <4 x i32> [[SUM1_NEXT]], [[SUM0_NEXT]]25; UF3-NEXT: [[RDX1:%.+]] = add <4 x i32> [[SUM2_NEXT]], [[RDX0]]26; UF3-NEXT: call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[RDX1]])27;28 29; UF5-LABEL: vector.body:30; UF5-NEXT: [[IV:%.+]] = phi i64 [ 0, %vector.ph ], [ [[IV_NEXT:%.+]], %vector.body ]31; UF5-NEXT: [[SUM0:%.+]] = phi <4 x i32> [ zeroinitializer, %vector.ph ], [ [[SUM0_NEXT:%.+]], %vector.body ]32; UF5-NEXT: [[SUM1:%.+]] = phi <4 x i32> [ zeroinitializer, %vector.ph ], [ [[SUM1_NEXT:%.+]], %vector.body ]33; UF5-NEXT: [[SUM2:%.+]] = phi <4 x i32> [ zeroinitializer, %vector.ph ], [ [[SUM2_NEXT:%.+]], %vector.body ]34; UF5-NEXT: [[SUM3:%.+]] = phi <4 x i32> [ zeroinitializer, %vector.ph ], [ [[SUM3_NEXT:%.+]], %vector.body ]35; UF5-NEXT: [[SUM4:%.+]] = phi <4 x i32> [ zeroinitializer, %vector.ph ], [ [[SUM4_NEXT:%.+]], %vector.body ]36; UF5-NEXT: [[GEP0:%.+]] = getelementptr inbounds i32, ptr %A, i64 [[IV]]37; UF5-NEXT: [[L_GEP1:%.+]] = getelementptr inbounds i32, ptr [[GEP0]], i64 438; UF5-NEXT: [[L_GEP2:%.+]] = getelementptr inbounds i32, ptr [[GEP0]], i64 839; UF5-NEXT: [[L_GEP3:%.+]] = getelementptr inbounds i32, ptr [[GEP0]], i64 1240; UF5-NEXT: [[L_GEP4:%.+]] = getelementptr inbounds i32, ptr [[GEP0]], i64 1641; UF5-NEXT: [[L0:%.+]] = load <4 x i32>, ptr [[GEP0]], align 442; UF5-NEXT: [[L1:%.+]] = load <4 x i32>, ptr [[L_GEP1]], align 443; UF5-NEXT: [[L2:%.+]] = load <4 x i32>, ptr [[L_GEP2]], align 444; UF5-NEXT: [[L3:%.+]] = load <4 x i32>, ptr [[L_GEP3]], align 445; UF5-NEXT: [[L4:%.+]] = load <4 x i32>, ptr [[L_GEP4]], align 446; UF5-NEXT: [[SUM0_NEXT]] = add <4 x i32> [[SUM0]], [[L0]]47; UF5-NEXT: [[SUM1_NEXT]] = add <4 x i32> [[SUM1]], [[L1]]48; UF5-NEXT: [[SUM2_NEXT]] = add <4 x i32> [[SUM2]], [[L2]]49; UF5-NEXT: [[SUM3_NEXT]] = add <4 x i32> [[SUM3]], [[L3]]50; UF5-NEXT: [[SUM4_NEXT]] = add <4 x i32> [[SUM4]], [[L4]]51; UF5-NEXT: [[IV_NEXT]] = add nuw i64 [[IV]], 2052; UF5-NEXT: [[EC:%.+]] = icmp eq i64 [[IV_NEXT]], %n.vec53; UF5-NEXT: br i1 [[EC]], label %middle.block, label %vector.body54;55; UF5-LABEL: middle.block:56; UF5-NEXT: [[RDX0:%.+]] = add <4 x i32> [[SUM1_NEXT]], [[SUM0_NEXT]]57; UF5-NEXT: [[RDX1:%.+]] = add <4 x i32> [[SUM2_NEXT]], [[RDX0]]58; UF5-NEXT: [[RDX2:%.+]] = add <4 x i32> [[SUM3_NEXT]], [[RDX1]]59; UF5-NEXT: [[RDX3:%.+]] = add <4 x i32> [[SUM4_NEXT]], [[RDX2]]60; UF5-NEXT: call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[RDX3]])61;62 63entry:64 br label %loop65 66loop:67 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]68 %sum.02 = phi i32 [ 0, %entry ], [ %sum.next, %loop ]69 %gep.A = getelementptr inbounds i32, ptr %A, i64 %iv70 %lv.A = load i32, ptr %gep.A, align 471 %sum.next = add i32 %sum.02, %lv.A72 %iv.next = add i64 %iv, 173 %exitcond = icmp eq i64 %iv, %n74 br i1 %exitcond, label %exit, label %loop75 76exit:77 %sum.0.lcssa = phi i32 [ %sum.next, %loop ]78 ret i32 %sum.0.lcssa79}80