240 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -S -passes=loop-vectorize -force-vector-width=2 < %s | FileCheck %s3 4define void @test_ptr_iv_no_inbounds(ptr %p1.start, ptr %p2.start, ptr %p1.end) {5; CHECK-LABEL: @test_ptr_iv_no_inbounds(6; CHECK-NEXT: entry:7; CHECK-NEXT: [[P1_START7:%.*]] = ptrtoint ptr [[P1_START:%.*]] to i648; CHECK-NEXT: [[P1_END6:%.*]] = ptrtoint ptr [[P1_END:%.*]] to i649; CHECK-NEXT: [[P1_START4:%.*]] = ptrtoint ptr [[P1_START]] to i6410; CHECK-NEXT: [[P1_END3:%.*]] = ptrtoint ptr [[P1_END]] to i6411; CHECK-NEXT: [[P1_START2:%.*]] = ptrtoint ptr [[P1_START]] to i6412; CHECK-NEXT: [[P1_END1:%.*]] = ptrtoint ptr [[P1_END]] to i6413; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[P1_END6]], -414; CHECK-NEXT: [[TMP1:%.*]] = sub i64 [[TMP0]], [[P1_START7]]15; CHECK-NEXT: [[TMP2:%.*]] = lshr i64 [[TMP1]], 216; CHECK-NEXT: [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], 117; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP3]], 218; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]19; CHECK: vector.scevcheck:20; CHECK-NEXT: [[TMP4:%.*]] = trunc i64 [[P1_END1]] to i221; CHECK-NEXT: [[TMP5:%.*]] = trunc i64 [[P1_START2]] to i222; CHECK-NEXT: [[TMP6:%.*]] = sub i2 [[TMP4]], [[TMP5]]23; CHECK-NEXT: [[TMP7:%.*]] = zext i2 [[TMP6]] to i6424; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[TMP7]], 025; CHECK-NEXT: br i1 [[IDENT_CHECK]], label [[SCALAR_PH]], label [[VECTOR_MEMCHECK:%.*]]26; CHECK: vector.memcheck:27; CHECK-NEXT: [[TMP8:%.*]] = add i64 [[P1_END3]], -428; CHECK-NEXT: [[TMP9:%.*]] = sub i64 [[TMP8]], [[P1_START4]]29; CHECK-NEXT: [[TMP10:%.*]] = lshr i64 [[TMP9]], 230; CHECK-NEXT: [[TMP11:%.*]] = shl nuw i64 [[TMP10]], 231; CHECK-NEXT: [[TMP12:%.*]] = add i64 [[TMP11]], 432; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[P1_START]], i64 [[TMP12]]33; CHECK-NEXT: [[SCEVGEP5:%.*]] = getelementptr i8, ptr [[P2_START:%.*]], i64 [[TMP12]]34; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[P1_START]], [[SCEVGEP5]]35; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[P2_START]], [[SCEVGEP]]36; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]37; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]38; CHECK: vector.ph:39; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], 240; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF]]41; CHECK-NEXT: [[TMP13:%.*]] = mul i64 [[N_VEC]], 442; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[P1_START]], i64 [[TMP13]]43; CHECK-NEXT: [[TMP14:%.*]] = mul i64 [[N_VEC]], 444; CHECK-NEXT: [[IND_END8:%.*]] = getelementptr i8, ptr [[P2_START]], i64 [[TMP14]]45; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]46; CHECK: vector.body:47; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]48; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 449; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[P1_START]], i64 [[OFFSET_IDX]]50; CHECK-NEXT: [[OFFSET_IDX10:%.*]] = mul i64 [[INDEX]], 451; CHECK-NEXT: [[NEXT_GEP11:%.*]] = getelementptr i8, ptr [[P2_START]], i64 [[OFFSET_IDX10]]52; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, ptr [[NEXT_GEP]], align 4, !alias.scope [[META0:![0-9]+]], !noalias [[META3:![0-9]+]]53; CHECK-NEXT: [[WIDE_LOAD12:%.*]] = load <2 x float>, ptr [[NEXT_GEP11]], align 4, !alias.scope [[META3]]54; CHECK-NEXT: [[TMP19:%.*]] = fadd <2 x float> [[WIDE_LOAD]], [[WIDE_LOAD12]]55; CHECK-NEXT: store <2 x float> [[TMP19]], ptr [[NEXT_GEP]], align 4, !alias.scope [[META0]], !noalias [[META3]]56; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 257; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]58; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]59; CHECK: middle.block:60; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]61; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]62; CHECK: scalar.ph:63; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[P1_START]], [[ENTRY:%.*]] ], [ [[P1_START]], [[VECTOR_SCEVCHECK]] ], [ [[P1_START]], [[VECTOR_MEMCHECK]] ]64; CHECK-NEXT: [[BC_RESUME_VAL9:%.*]] = phi ptr [ [[IND_END8]], [[MIDDLE_BLOCK]] ], [ [[P2_START]], [[ENTRY]] ], [ [[P2_START]], [[VECTOR_SCEVCHECK]] ], [ [[P2_START]], [[VECTOR_MEMCHECK]] ]65; CHECK-NEXT: br label [[LOOP:%.*]]66; CHECK: loop:67; CHECK-NEXT: [[P1:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[P1_NEXT:%.*]], [[LOOP]] ]68; CHECK-NEXT: [[P2:%.*]] = phi ptr [ [[BC_RESUME_VAL9]], [[SCALAR_PH]] ], [ [[P2_NEXT:%.*]], [[LOOP]] ]69; CHECK-NEXT: [[P1_VAL:%.*]] = load float, ptr [[P1]], align 470; CHECK-NEXT: [[P2_VAL:%.*]] = load float, ptr [[P2]], align 471; CHECK-NEXT: [[SUM:%.*]] = fadd float [[P1_VAL]], [[P2_VAL]]72; CHECK-NEXT: store float [[SUM]], ptr [[P1]], align 473; CHECK-NEXT: [[P1_NEXT]] = getelementptr float, ptr [[P1]], i64 174; CHECK-NEXT: [[P2_NEXT]] = getelementptr float, ptr [[P2]], i64 175; CHECK-NEXT: [[C:%.*]] = icmp ne ptr [[P1_NEXT]], [[P1_END]]76; CHECK-NEXT: br i1 [[C]], label [[LOOP]], label [[EXIT]], !llvm.loop [[LOOP8:![0-9]+]]77; CHECK: exit:78; CHECK-NEXT: ret void79;80entry:81 br label %loop82 83loop:84 %p1 = phi ptr [ %p1.start, %entry ], [ %p1.next, %loop ]85 %p2 = phi ptr [ %p2.start, %entry ], [ %p2.next, %loop ]86 %p1.val = load float, ptr %p187 %p2.val = load float, ptr %p288 %sum = fadd float %p1.val, %p2.val89 store float %sum, ptr %p190 %p1.next = getelementptr float, ptr %p1, i64 191 %p2.next = getelementptr float, ptr %p2, i64 192 %c = icmp ne ptr %p1.next, %p1.end93 br i1 %c, label %loop, label %exit94 95exit:96 ret void97}98 99define void @test_ptr_iv_with_inbounds(ptr %p1.start, ptr %p2.start, ptr %p1.end) {100; CHECK-LABEL: @test_ptr_iv_with_inbounds(101; CHECK-NEXT: entry:102; CHECK-NEXT: [[P1_START5:%.*]] = ptrtoint ptr [[P1_START:%.*]] to i64103; CHECK-NEXT: [[P1_END4:%.*]] = ptrtoint ptr [[P1_END:%.*]] to i64104; CHECK-NEXT: [[P1_START2:%.*]] = ptrtoint ptr [[P1_START]] to i64105; CHECK-NEXT: [[P1_END1:%.*]] = ptrtoint ptr [[P1_END]] to i64106; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[P1_END4]], -4107; CHECK-NEXT: [[TMP1:%.*]] = sub i64 [[TMP0]], [[P1_START5]]108; CHECK-NEXT: [[TMP2:%.*]] = lshr i64 [[TMP1]], 2109; CHECK-NEXT: [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], 1110; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP3]], 2111; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]112; CHECK: vector.memcheck:113; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[P1_END1]], -4114; CHECK-NEXT: [[TMP5:%.*]] = sub i64 [[TMP4]], [[P1_START2]]115; CHECK-NEXT: [[TMP6:%.*]] = lshr i64 [[TMP5]], 2116; CHECK-NEXT: [[TMP7:%.*]] = shl nuw i64 [[TMP6]], 2117; CHECK-NEXT: [[TMP8:%.*]] = add i64 [[TMP7]], 4118; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[P1_START]], i64 [[TMP8]]119; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[P2_START:%.*]], i64 [[TMP8]]120; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[P1_START]], [[SCEVGEP3]]121; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[P2_START]], [[SCEVGEP]]122; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]123; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]124; CHECK: vector.ph:125; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], 2126; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF]]127; CHECK-NEXT: [[TMP9:%.*]] = mul i64 [[N_VEC]], 4128; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[P1_START]], i64 [[TMP9]]129; CHECK-NEXT: [[TMP10:%.*]] = mul i64 [[N_VEC]], 4130; CHECK-NEXT: [[IND_END6:%.*]] = getelementptr i8, ptr [[P2_START]], i64 [[TMP10]]131; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]132; CHECK: vector.body:133; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]134; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 4135; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[P1_START]], i64 [[OFFSET_IDX]]136; CHECK-NEXT: [[OFFSET_IDX8:%.*]] = mul i64 [[INDEX]], 4137; CHECK-NEXT: [[NEXT_GEP9:%.*]] = getelementptr i8, ptr [[P2_START]], i64 [[OFFSET_IDX8]]138; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, ptr [[NEXT_GEP]], align 4, !alias.scope [[META9:![0-9]+]], !noalias [[META12:![0-9]+]]139; CHECK-NEXT: [[WIDE_LOAD10:%.*]] = load <2 x float>, ptr [[NEXT_GEP9]], align 4, !alias.scope [[META12]]140; CHECK-NEXT: [[TMP15:%.*]] = fadd <2 x float> [[WIDE_LOAD]], [[WIDE_LOAD10]]141; CHECK-NEXT: store <2 x float> [[TMP15]], ptr [[NEXT_GEP]], align 4, !alias.scope [[META9]], !noalias [[META12]]142; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2143; CHECK-NEXT: [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]144; CHECK-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP14:![0-9]+]]145; CHECK: middle.block:146; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]147; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]148; CHECK: scalar.ph:149; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[P1_START]], [[ENTRY:%.*]] ], [ [[P1_START]], [[VECTOR_MEMCHECK]] ]150; CHECK-NEXT: [[BC_RESUME_VAL7:%.*]] = phi ptr [ [[IND_END6]], [[MIDDLE_BLOCK]] ], [ [[P2_START]], [[ENTRY]] ], [ [[P2_START]], [[VECTOR_MEMCHECK]] ]151; CHECK-NEXT: br label [[LOOP:%.*]]152; CHECK: loop:153; CHECK-NEXT: [[P1:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[P1_NEXT:%.*]], [[LOOP]] ]154; CHECK-NEXT: [[P2:%.*]] = phi ptr [ [[BC_RESUME_VAL7]], [[SCALAR_PH]] ], [ [[P2_NEXT:%.*]], [[LOOP]] ]155; CHECK-NEXT: [[P1_VAL:%.*]] = load float, ptr [[P1]], align 4156; CHECK-NEXT: [[P2_VAL:%.*]] = load float, ptr [[P2]], align 4157; CHECK-NEXT: [[SUM:%.*]] = fadd float [[P1_VAL]], [[P2_VAL]]158; CHECK-NEXT: store float [[SUM]], ptr [[P1]], align 4159; CHECK-NEXT: [[P1_NEXT]] = getelementptr inbounds float, ptr [[P1]], i64 1160; CHECK-NEXT: [[P2_NEXT]] = getelementptr inbounds float, ptr [[P2]], i64 1161; CHECK-NEXT: [[C:%.*]] = icmp ne ptr [[P1_NEXT]], [[P1_END]]162; CHECK-NEXT: br i1 [[C]], label [[LOOP]], label [[EXIT]], !llvm.loop [[LOOP15:![0-9]+]]163; CHECK: exit:164; CHECK-NEXT: ret void165;166entry:167 br label %loop168 169loop:170 %p1 = phi ptr [ %p1.start, %entry ], [ %p1.next, %loop ]171 %p2 = phi ptr [ %p2.start, %entry ], [ %p2.next, %loop ]172 %p1.val = load float, ptr %p1173 %p2.val = load float, ptr %p2174 %sum = fadd float %p1.val, %p2.val175 store float %sum, ptr %p1176 %p1.next = getelementptr inbounds float, ptr %p1, i64 1177 %p2.next = getelementptr inbounds float, ptr %p2, i64 1178 %c = icmp ne ptr %p1.next, %p1.end179 br i1 %c, label %loop, label %exit180 181exit:182 ret void183}184 185define void @store_pointer_induction(ptr %start, ptr %end) {186; CHECK-LABEL: @store_pointer_induction(187; CHECK-NEXT: entry:188; CHECK-NEXT: [[START2:%.*]] = ptrtoint ptr [[START:%.*]] to i64189; CHECK-NEXT: [[END1:%.*]] = ptrtoint ptr [[END:%.*]] to i64190; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[END1]], -8191; CHECK-NEXT: [[TMP1:%.*]] = sub i64 [[TMP0]], [[START2]]192; CHECK-NEXT: [[TMP2:%.*]] = lshr i64 [[TMP1]], 3193; CHECK-NEXT: [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], 1194; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP3]], 2195; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]196; CHECK: vector.ph:197; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], 2198; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF]]199; CHECK-NEXT: [[TMP4:%.*]] = mul i64 [[N_VEC]], 8200; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP4]]201; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]202; CHECK: vector.body:203; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]204; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[START]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]205; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <2 x i64> <i64 0, i64 8>206; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x ptr> [[TMP5]], i32 0207; CHECK-NEXT: store <2 x ptr> [[TMP5]], ptr [[TMP6]], align 4208; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2209; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 16210; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]211; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]212; CHECK: middle.block:213; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]214; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]215; CHECK: scalar.ph:216; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[START]], [[ENTRY:%.*]] ]217; CHECK-NEXT: br label [[LOOP:%.*]]218; CHECK: loop:219; CHECK-NEXT: [[IV:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]220; CHECK-NEXT: store ptr [[IV]], ptr [[IV]], align 4221; CHECK-NEXT: [[IV_NEXT]] = getelementptr inbounds ptr, ptr [[IV]], i32 1222; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq ptr [[IV_NEXT]], [[END]]223; CHECK-NEXT: br i1 [[EXITCOND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP17:![0-9]+]]224; CHECK: exit:225; CHECK-NEXT: ret void226;227entry:228 br label %loop229 230loop:231 %iv = phi ptr [ %start, %entry ], [ %iv.next, %loop ]232 store ptr %iv, ptr %iv, align 4233 %iv.next = getelementptr inbounds ptr, ptr %iv, i32 1234 %exitcond = icmp eq ptr %iv.next, %end235 br i1 %exitcond, label %exit, label %loop236 237exit:238 ret void239}240