98 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --filter-out-after "vector.ph:" --version 52; RUN: opt %s -passes=loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -S | FileCheck %s3; RUN: opt %s -passes=loop-vectorize -force-vector-interleave=2 -force-vector-width=4 -S | FileCheck %s -check-prefix=UNROLL4 5target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"6 7@b = common global [1000 x i32] zeroinitializer, align 168@c = common global [1000 x i32] zeroinitializer, align 169@a = common global [1000 x i32] zeroinitializer, align 1610 11; Generate min.iters.check to skip the vector loop and jump to scalar.ph directly when loop iteration number is less than VF * UF.12define void @foo(i64 %N) {13; CHECK-LABEL: define void @foo(14; CHECK-SAME: i64 [[N:%.*]]) {15; CHECK-NEXT: [[ENTRY:.*:]]16; CHECK-NEXT: [[C:%.*]] = icmp sgt i64 [[N]], 017; CHECK-NEXT: br i1 [[C]], label %[[LOOP_PREHEADER:.*]], [[EXIT:label %.*]]18; CHECK: [[LOOP_PREHEADER]]:19; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 420; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], [[SCALAR_PH:label %.*]], label %[[VECTOR_PH:.*]]21; CHECK: [[VECTOR_PH]]:22;23; UNROLL-LABEL: define void @foo(24; UNROLL-SAME: i64 [[N:%.*]]) {25; UNROLL-NEXT: [[ENTRY:.*:]]26; UNROLL-NEXT: [[C:%.*]] = icmp sgt i64 [[N]], 027; UNROLL-NEXT: br i1 [[C]], label %[[LOOP_PREHEADER:.*]], [[EXIT:label %.*]]28; UNROLL: [[LOOP_PREHEADER]]:29; UNROLL-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 830; UNROLL-NEXT: br i1 [[MIN_ITERS_CHECK]], [[SCALAR_PH:label %.*]], label %[[VECTOR_PH:.*]]31; UNROLL: [[VECTOR_PH]]:32;33entry:34 %c = icmp sgt i64 %N, 035 br i1 %c, label %loop, label %exit36 37loop:38 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]39 %arrayidx = getelementptr inbounds [1000 x i32], ptr @b, i64 0, i64 %iv40 %tmp = load i32, ptr %arrayidx, align 441 %arrayidx1 = getelementptr inbounds [1000 x i32], ptr @c, i64 0, i64 %iv42 %tmp1 = load i32, ptr %arrayidx1, align 443 %add = add nsw i32 %tmp1, %tmp44 %arrayidx2 = getelementptr inbounds [1000 x i32], ptr @a, i64 0, i64 %iv45 store i32 %add, ptr %arrayidx2, align 446 %iv.next = add nuw nsw i64 %iv, 147 %ec = icmp eq i64 %iv.next, %N48 br i1 %ec, label %exit, label %loop49 50exit:51 ret void52}53 54define void @min_iters_known_via_loop_guards_add(i32 %start, i32 %end, ptr %src) {55; CHECK-LABEL: define void @min_iters_known_via_loop_guards_add(56; CHECK-SAME: i32 [[START:%.*]], i32 [[END:%.*]], ptr [[SRC:%.*]]) {57; CHECK-NEXT: [[ENTRY:.*:]]58; CHECK-NEXT: [[SUB:%.*]] = sub i32 [[END]], [[START]]59; CHECK-NEXT: [[PRE:%.*]] = icmp sgt i32 [[SUB]], 10060; CHECK-NEXT: call void @llvm.assume(i1 [[PRE]])61; CHECK-NEXT: [[ADD_1:%.*]] = add i32 [[SUB]], 162; CHECK-NEXT: [[IV_START:%.*]] = zext i32 [[ADD_1]] to i6463; CHECK-NEXT: [[TMP0:%.*]] = sub i64 101, [[IV_START]]64; CHECK-NEXT: br label %[[VECTOR_PH:.*]]65; CHECK: [[VECTOR_PH]]:66;67; UNROLL-LABEL: define void @min_iters_known_via_loop_guards_add(68; UNROLL-SAME: i32 [[START:%.*]], i32 [[END:%.*]], ptr [[SRC:%.*]]) {69; UNROLL-NEXT: [[ENTRY:.*:]]70; UNROLL-NEXT: [[SUB:%.*]] = sub i32 [[END]], [[START]]71; UNROLL-NEXT: [[PRE:%.*]] = icmp sgt i32 [[SUB]], 10072; UNROLL-NEXT: call void @llvm.assume(i1 [[PRE]])73; UNROLL-NEXT: [[ADD_1:%.*]] = add i32 [[SUB]], 174; UNROLL-NEXT: [[IV_START:%.*]] = zext i32 [[ADD_1]] to i6475; UNROLL-NEXT: [[TMP0:%.*]] = sub i64 101, [[IV_START]]76; UNROLL-NEXT: br label %[[VECTOR_PH:.*]]77; UNROLL: [[VECTOR_PH]]:78;79entry:80 %sub = sub i32 %end, %start81 %pre = icmp sgt i32 %sub, 10082 call void @llvm.assume(i1 %pre)83 %add.1 = add i32 %sub, 184 %iv.start = zext i32 %add.1 to i6485 br label %loop86 87loop:88 %iv = phi i64 [ %iv.start, %entry ], [ %iv.next, %loop ]89 %gep = getelementptr inbounds i64, ptr %src, i64 %iv90 store i64 %iv, ptr %gep91 %iv.next = add i64 %iv, 192 %ec = icmp eq i64 %iv, 10093 br i1 %ec, label %exit, label %loop94 95exit:96 ret void97}98