1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5 2; RUN: opt -p loop-vectorize -force-vector-width=1 -force-vector-interleave=2 -S %s | FileCheck %s 3 4define float @for_load_interleave_only(ptr %src) { 5; CHECK-LABEL: define float @for_load_interleave_only( 6; CHECK-SAME: ptr [[SRC:%.*]]) { 7; CHECK-NEXT: [[ENTRY:.*]]: 8; CHECK-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 9; CHECK: [[VECTOR_PH]]: 10; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[SRC]], i64 16000 11; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] 12; CHECK: [[VECTOR_BODY]]: 13; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 14; CHECK-NEXT: [[VECTOR_RECUR:%.*]] = phi float [ 0.000000e+00, %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ] 15; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 16 16; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[OFFSET_IDX]], 0 17; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[OFFSET_IDX]], 16 18; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[TMP0]] 19; CHECK-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[TMP1]] 20; CHECK-NEXT: [[TMP2:%.*]] = load float, ptr [[NEXT_GEP]], align 4 21; CHECK-NEXT: [[TMP3]] = load float, ptr [[NEXT_GEP2]], align 4 22; CHECK-NEXT: store float 0.000000e+00, ptr [[NEXT_GEP]], align 4 23; CHECK-NEXT: store float 0.000000e+00, ptr [[NEXT_GEP2]], align 4 24; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 25; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1000 26; CHECK-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 27; CHECK: [[MIDDLE_BLOCK]]: 28; CHECK-NEXT: br i1 true, label %[[EXIT:.*]], label %[[SCALAR_PH]] 29; CHECK: [[SCALAR_PH]]: 30; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 1001, %[[MIDDLE_BLOCK]] ], [ 1, %[[ENTRY]] ] 31; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi ptr [ [[IND_END]], %[[MIDDLE_BLOCK]] ], [ [[SRC]], %[[ENTRY]] ] 32; CHECK-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi float [ [[TMP3]], %[[MIDDLE_BLOCK]] ], [ 0.000000e+00, %[[ENTRY]] ] 33; CHECK-NEXT: br label %[[LOOP:.*]] 34; CHECK: [[LOOP]]: 35; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] 36; CHECK-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[BC_RESUME_VAL1]], %[[SCALAR_PH]] ], [ [[PTR_IV_NEXT:%.*]], %[[LOOP]] ] 37; CHECK-NEXT: [[FOR:%.*]] = phi float [ [[SCALAR_RECUR_INIT]], %[[SCALAR_PH]] ], [ [[L:%.*]], %[[LOOP]] ] 38; CHECK-NEXT: [[IV_NEXT]] = add i32 [[IV]], 1 39; CHECK-NEXT: [[PTR_IV_NEXT]] = getelementptr i8, ptr [[PTR_IV]], i64 16 40; CHECK-NEXT: [[L]] = load float, ptr [[PTR_IV]], align 4 41; CHECK-NEXT: store float 0.000000e+00, ptr [[PTR_IV]], align 4 42; CHECK-NEXT: [[EC:%.*]] = icmp eq i32 [[IV]], 1000 43; CHECK-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP3:![0-9]+]] 44; CHECK: [[EXIT]]: 45; CHECK-NEXT: [[FOR_LCSSA:%.*]] = phi float [ [[FOR]], %[[LOOP]] ], [ [[TMP2]], %[[MIDDLE_BLOCK]] ] 46; CHECK-NEXT: ret float [[FOR_LCSSA]] 47; 48entry: 49 br label %loop 50 51loop: 52 %iv = phi i32 [ 1, %entry ], [ %iv.next, %loop ] 53 %ptr.iv = phi ptr [ %src, %entry ], [ %ptr.iv.next, %loop ] 54 %for = phi float [ 0.000000e+00, %entry ], [ %l, %loop ] 55 %iv.next = add i32 %iv, 1 56 %ptr.iv.next = getelementptr i8, ptr %ptr.iv, i64 16 57 %l = load float, ptr %ptr.iv, align 4 58 store float 0.000000e+00, ptr %ptr.iv, align 4 59 %ec = icmp eq i32 %iv, 1000 60 br i1 %ec, label %exit, label %loop 61 62exit: 63 ret float %for 64} 65;. 66; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]} 67; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1} 68; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"} 69; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META1]]} 70;. 71