256 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=loop-vectorize -S %s | FileCheck %s --check-prefix=CHECK3; RUN: opt -passes=loop-vectorize -debug-only=loop-vectorize -disable-output %s 2>&1 | FileCheck %s --check-prefix=CHECK-COST4; REQUIRES: asserts5 6target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"7target triple = "aarch64--linux-gnu"8 9; CHECK-COST-LABEL: sadd10; CHECK-COST: Found an estimated cost of 6 for VF 1 For instruction: %1 = tail call i16 @llvm.sadd.sat.i16(i16 %0, i16 %offset)11; CHECK-COST: Cost of 4 for VF 2: WIDEN-INTRINSIC ir<%1> = call llvm.sadd.sat(ir<%0>, ir<%offset>)12; CHECK-COST: Cost of 1 for VF 4: WIDEN-INTRINSIC ir<%1> = call llvm.sadd.sat(ir<%0>, ir<%offset>)13; CHECK-COST: Cost of 1 for VF 8: WIDEN-INTRINSIC ir<%1> = call llvm.sadd.sat(ir<%0>, ir<%offset>)14 15define void @saddsat(ptr nocapture readonly %pSrc, i16 signext %offset, ptr nocapture noalias %pDst, i32 %blockSize) #0 {16; CHECK-LABEL: @saddsat(17; CHECK-NEXT: entry:18; CHECK-NEXT: [[CMP_NOT6:%.*]] = icmp eq i32 [[BLOCKSIZE:%.*]], 019; CHECK-NEXT: br i1 [[CMP_NOT6]], label [[WHILE_END:%.*]], label [[ITER_CHECK:%.*]]20; CHECK: iter.check:21; CHECK-NEXT: [[TMP0:%.*]] = zext i32 [[BLOCKSIZE]] to i6422; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 423; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]24; CHECK: vector.main.loop.iter.check:25; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[TMP0]], 1626; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]27; CHECK: vector.ph:28; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 1629; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]30; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <8 x i16> poison, i16 [[OFFSET:%.*]], i64 031; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <8 x i16> [[BROADCAST_SPLATINSERT]], <8 x i16> poison, <8 x i32> zeroinitializer32; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]33; CHECK: vector.body:34; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]35; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 236; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[PSRC:%.*]], i64 [[OFFSET_IDX]]37; CHECK-NEXT: [[OFFSET_IDX2:%.*]] = mul i64 [[INDEX]], 238; CHECK-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[PDST:%.*]], i64 [[OFFSET_IDX2]]39; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i16, ptr [[NEXT_GEP]], i64 840; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[NEXT_GEP]], align 241; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <8 x i16>, ptr [[TMP1]], align 242; CHECK-NEXT: [[TMP2:%.*]] = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> [[WIDE_LOAD]], <8 x i16> [[BROADCAST_SPLAT]])43; CHECK-NEXT: [[TMP3:%.*]] = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> [[WIDE_LOAD4]], <8 x i16> [[BROADCAST_SPLAT]])44; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i16, ptr [[NEXT_GEP3]], i64 845; CHECK-NEXT: store <8 x i16> [[TMP2]], ptr [[NEXT_GEP3]], align 246; CHECK-NEXT: store <8 x i16> [[TMP3]], ptr [[TMP4]], align 247; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 1648; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]49; CHECK-NEXT: br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]50; CHECK: middle.block:51; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]52; CHECK-NEXT: br i1 [[CMP_N]], label [[WHILE_END_LOOPEXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]53; CHECK: vec.epilog.iter.check:54; CHECK-NEXT: [[DOTCAST1:%.*]] = trunc i64 [[N_VEC]] to i3255; CHECK-NEXT: [[IND_END8:%.*]] = sub i32 [[BLOCKSIZE]], [[DOTCAST1]]56; CHECK-NEXT: [[TMP6:%.*]] = mul i64 [[N_VEC]], 257; CHECK-NEXT: [[IND_END10:%.*]] = getelementptr i8, ptr [[PSRC]], i64 [[TMP6]]58; CHECK-NEXT: [[TMP7:%.*]] = mul i64 [[N_VEC]], 259; CHECK-NEXT: [[IND_END13:%.*]] = getelementptr i8, ptr [[PDST]], i64 [[TMP7]]60; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 461; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF3:![0-9]+]]62; CHECK: vec.epilog.ph:63; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]64; CHECK-NEXT: [[N_MOD_VF4:%.*]] = urem i64 [[TMP0]], 465; CHECK-NEXT: [[N_VEC6:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF4]]66; CHECK-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC6]] to i3267; CHECK-NEXT: [[IND_END:%.*]] = sub i32 [[BLOCKSIZE]], [[DOTCAST]]68; CHECK-NEXT: [[TMP8:%.*]] = mul i64 [[N_VEC6]], 269; CHECK-NEXT: [[IND_END9:%.*]] = getelementptr i8, ptr [[PSRC]], i64 [[TMP8]]70; CHECK-NEXT: [[TMP9:%.*]] = mul i64 [[N_VEC6]], 271; CHECK-NEXT: [[IND_END12:%.*]] = getelementptr i8, ptr [[PDST]], i64 [[TMP9]]72; CHECK-NEXT: [[BROADCAST_SPLATINSERT21:%.*]] = insertelement <4 x i16> poison, i16 [[OFFSET]], i64 073; CHECK-NEXT: [[BROADCAST_SPLAT22:%.*]] = shufflevector <4 x i16> [[BROADCAST_SPLATINSERT21]], <4 x i16> poison, <4 x i32> zeroinitializer74; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]75; CHECK: vec.epilog.vector.body:76; CHECK-NEXT: [[INDEX15:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT23:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]77; CHECK-NEXT: [[OFFSET_IDX16:%.*]] = mul i64 [[INDEX15]], 278; CHECK-NEXT: [[NEXT_GEP17:%.*]] = getelementptr i8, ptr [[PSRC]], i64 [[OFFSET_IDX16]]79; CHECK-NEXT: [[OFFSET_IDX18:%.*]] = mul i64 [[INDEX15]], 280; CHECK-NEXT: [[NEXT_GEP19:%.*]] = getelementptr i8, ptr [[PDST]], i64 [[OFFSET_IDX18]]81; CHECK-NEXT: [[WIDE_LOAD20:%.*]] = load <4 x i16>, ptr [[NEXT_GEP17]], align 282; CHECK-NEXT: [[TMP10:%.*]] = call <4 x i16> @llvm.sadd.sat.v4i16(<4 x i16> [[WIDE_LOAD20]], <4 x i16> [[BROADCAST_SPLAT22]])83; CHECK-NEXT: store <4 x i16> [[TMP10]], ptr [[NEXT_GEP19]], align 284; CHECK-NEXT: [[INDEX_NEXT23]] = add nuw i64 [[INDEX15]], 485; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT23]], [[N_VEC6]]86; CHECK-NEXT: br i1 [[TMP11]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]87; CHECK: vec.epilog.middle.block:88; CHECK-NEXT: [[CMP_N14:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC6]]89; CHECK-NEXT: br i1 [[CMP_N14]], label [[WHILE_END_LOOPEXIT]], label [[VEC_EPILOG_SCALAR_PH]]90; CHECK: vec.epilog.scalar.ph:91; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[IND_END]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END8]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[BLOCKSIZE]], [[ITER_CHECK]] ]92; CHECK-NEXT: [[BC_RESUME_VAL11:%.*]] = phi ptr [ [[IND_END9]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END10]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[PSRC]], [[ITER_CHECK]] ]93; CHECK-NEXT: [[BC_RESUME_VAL14:%.*]] = phi ptr [ [[IND_END12]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END13]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[PDST]], [[ITER_CHECK]] ]94; CHECK-NEXT: br label [[WHILE_BODY:%.*]]95; CHECK: while.body:96; CHECK-NEXT: [[BLKCNT_09:%.*]] = phi i32 [ [[DEC:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ]97; CHECK-NEXT: [[PSRC_ADDR_08:%.*]] = phi ptr [ [[INCDEC_PTR:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL11]], [[VEC_EPILOG_SCALAR_PH]] ]98; CHECK-NEXT: [[PDST_ADDR_07:%.*]] = phi ptr [ [[INCDEC_PTR3:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL14]], [[VEC_EPILOG_SCALAR_PH]] ]99; CHECK-NEXT: [[INCDEC_PTR]] = getelementptr inbounds i16, ptr [[PSRC_ADDR_08]], i32 1100; CHECK-NEXT: [[TMP12:%.*]] = load i16, ptr [[PSRC_ADDR_08]], align 2101; CHECK-NEXT: [[TMP13:%.*]] = tail call i16 @llvm.sadd.sat.i16(i16 [[TMP12]], i16 [[OFFSET]])102; CHECK-NEXT: [[INCDEC_PTR3]] = getelementptr inbounds i16, ptr [[PDST_ADDR_07]], i32 1103; CHECK-NEXT: store i16 [[TMP13]], ptr [[PDST_ADDR_07]], align 2104; CHECK-NEXT: [[DEC]] = add i32 [[BLKCNT_09]], -1105; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq i32 [[DEC]], 0106; CHECK-NEXT: br i1 [[CMP_NOT]], label [[WHILE_END_LOOPEXIT]], label [[WHILE_BODY]], !llvm.loop [[LOOP5:![0-9]+]]107; CHECK: while.end.loopexit:108; CHECK-NEXT: br label [[WHILE_END]]109; CHECK: while.end:110; CHECK-NEXT: ret void111;112entry:113 %cmp.not6 = icmp eq i32 %blockSize, 0114 br i1 %cmp.not6, label %while.end, label %while.body115 116while.body: ; preds = %entry, %while.body117 %blkCnt.09 = phi i32 [ %dec, %while.body ], [ %blockSize, %entry ]118 %pSrc.addr.08 = phi ptr [ %incdec.ptr, %while.body ], [ %pSrc, %entry ]119 %pDst.addr.07 = phi ptr [ %incdec.ptr3, %while.body ], [ %pDst, %entry ]120 %incdec.ptr = getelementptr inbounds i16, ptr %pSrc.addr.08, i32 1121 %0 = load i16, ptr %pSrc.addr.08, align 2122 %1 = tail call i16 @llvm.sadd.sat.i16(i16 %0, i16 %offset)123 %incdec.ptr3 = getelementptr inbounds i16, ptr %pDst.addr.07, i32 1124 store i16 %1, ptr %pDst.addr.07, align 2125 %dec = add i32 %blkCnt.09, -1126 %cmp.not = icmp eq i32 %dec, 0127 br i1 %cmp.not, label %while.end, label %while.body128 129while.end: ; preds = %while.body, %entry130 ret void131}132 133; CHECK-COST-LABEL: umin134; CHECK-COST: Found an estimated cost of 2 for VF 1 For instruction: %1 = tail call i8 @llvm.umin.i8(i8 %0, i8 %offset)135; CHECK-COST: Cost of 1 for VF 2: WIDEN-INTRINSIC ir<%1> = call llvm.umin(ir<%0>, ir<%offset>)136; CHECK-COST: Cost of 1 for VF 4: WIDEN-INTRINSIC ir<%1> = call llvm.umin(ir<%0>, ir<%offset>)137; CHECK-COST: Cost of 1 for VF 8: WIDEN-INTRINSIC ir<%1> = call llvm.umin(ir<%0>, ir<%offset>)138; CHECK-COST: Cost of 1 for VF 16: WIDEN-INTRINSIC ir<%1> = call llvm.umin(ir<%0>, ir<%offset>)139 140 141define void @umin(ptr nocapture readonly %pSrc, i8 signext %offset, ptr nocapture noalias %pDst, i32 %blockSize) #0 {142; CHECK-LABEL: @umin(143; CHECK-NEXT: entry:144; CHECK-NEXT: [[CMP_NOT6:%.*]] = icmp eq i32 [[BLOCKSIZE:%.*]], 0145; CHECK-NEXT: br i1 [[CMP_NOT6]], label [[WHILE_END:%.*]], label [[ITER_CHECK:%.*]]146; CHECK: iter.check:147; CHECK-NEXT: [[TMP0:%.*]] = zext i32 [[BLOCKSIZE]] to i64148; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 8149; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]150; CHECK: vector.main.loop.iter.check:151; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[TMP0]], 32152; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]153; CHECK: vector.ph:154; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 32155; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]156; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i8> poison, i8 [[OFFSET:%.*]], i64 0157; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i8> [[BROADCAST_SPLATINSERT]], <16 x i8> poison, <16 x i32> zeroinitializer158; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]159; CHECK: vector.body:160; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]161; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[PSRC:%.*]], i64 [[INDEX]]162; CHECK-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[PDST:%.*]], i64 [[INDEX]]163; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 16164; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[NEXT_GEP]], align 2165; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP1]], align 2166; CHECK-NEXT: [[TMP2:%.*]] = call <16 x i8> @llvm.umin.v16i8(<16 x i8> [[WIDE_LOAD]], <16 x i8> [[BROADCAST_SPLAT]])167; CHECK-NEXT: [[TMP3:%.*]] = call <16 x i8> @llvm.umin.v16i8(<16 x i8> [[WIDE_LOAD3]], <16 x i8> [[BROADCAST_SPLAT]])168; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[NEXT_GEP2]], i64 16169; CHECK-NEXT: store <16 x i8> [[TMP2]], ptr [[NEXT_GEP2]], align 2170; CHECK-NEXT: store <16 x i8> [[TMP3]], ptr [[TMP4]], align 2171; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32172; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]173; CHECK-NEXT: br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]174; CHECK: middle.block:175; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]176; CHECK-NEXT: br i1 [[CMP_N]], label [[WHILE_END_LOOPEXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]177; CHECK: vec.epilog.iter.check:178; CHECK-NEXT: [[DOTCAST6:%.*]] = trunc i64 [[N_VEC]] to i32179; CHECK-NEXT: [[IND_END7:%.*]] = sub i32 [[BLOCKSIZE]], [[DOTCAST6]]180; CHECK-NEXT: [[IND_END9:%.*]] = getelementptr i8, ptr [[PSRC]], i64 [[N_VEC]]181; CHECK-NEXT: [[IND_END12:%.*]] = getelementptr i8, ptr [[PDST]], i64 [[N_VEC]]182; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 8183; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF7:![0-9]+]]184; CHECK: vec.epilog.ph:185; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]186; CHECK-NEXT: [[N_MOD_VF4:%.*]] = urem i64 [[TMP0]], 8187; CHECK-NEXT: [[N_VEC5:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF4]]188; CHECK-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC5]] to i32189; CHECK-NEXT: [[IND_END:%.*]] = sub i32 [[BLOCKSIZE]], [[DOTCAST]]190; CHECK-NEXT: [[IND_END8:%.*]] = getelementptr i8, ptr [[PSRC]], i64 [[N_VEC5]]191; CHECK-NEXT: [[IND_END11:%.*]] = getelementptr i8, ptr [[PDST]], i64 [[N_VEC5]]192; CHECK-NEXT: [[BROADCAST_SPLATINSERT18:%.*]] = insertelement <8 x i8> poison, i8 [[OFFSET]], i64 0193; CHECK-NEXT: [[BROADCAST_SPLAT19:%.*]] = shufflevector <8 x i8> [[BROADCAST_SPLATINSERT18]], <8 x i8> poison, <8 x i32> zeroinitializer194; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]195; CHECK: vec.epilog.vector.body:196; CHECK-NEXT: [[INDEX14:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT20:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]197; CHECK-NEXT: [[NEXT_GEP15:%.*]] = getelementptr i8, ptr [[PSRC]], i64 [[INDEX14]]198; CHECK-NEXT: [[NEXT_GEP16:%.*]] = getelementptr i8, ptr [[PDST]], i64 [[INDEX14]]199; CHECK-NEXT: [[WIDE_LOAD17:%.*]] = load <8 x i8>, ptr [[NEXT_GEP15]], align 2200; CHECK-NEXT: [[TMP6:%.*]] = call <8 x i8> @llvm.umin.v8i8(<8 x i8> [[WIDE_LOAD17]], <8 x i8> [[BROADCAST_SPLAT19]])201; CHECK-NEXT: store <8 x i8> [[TMP6]], ptr [[NEXT_GEP16]], align 2202; CHECK-NEXT: [[INDEX_NEXT20]] = add nuw i64 [[INDEX14]], 8203; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT20]], [[N_VEC5]]204; CHECK-NEXT: br i1 [[TMP7]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]205; CHECK: vec.epilog.middle.block:206; CHECK-NEXT: [[CMP_N13:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC5]]207; CHECK-NEXT: br i1 [[CMP_N13]], label [[WHILE_END_LOOPEXIT]], label [[VEC_EPILOG_SCALAR_PH]]208; CHECK: vec.epilog.scalar.ph:209; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[IND_END]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END7]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[BLOCKSIZE]], [[ITER_CHECK]] ]210; CHECK-NEXT: [[BC_RESUME_VAL10:%.*]] = phi ptr [ [[IND_END8]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END9]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[PSRC]], [[ITER_CHECK]] ]211; CHECK-NEXT: [[BC_RESUME_VAL13:%.*]] = phi ptr [ [[IND_END11]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END12]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[PDST]], [[ITER_CHECK]] ]212; CHECK-NEXT: br label [[WHILE_BODY:%.*]]213; CHECK: while.body:214; CHECK-NEXT: [[BLKCNT_09:%.*]] = phi i32 [ [[DEC:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ]215; CHECK-NEXT: [[PSRC_ADDR_08:%.*]] = phi ptr [ [[INCDEC_PTR:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL10]], [[VEC_EPILOG_SCALAR_PH]] ]216; CHECK-NEXT: [[PDST_ADDR_07:%.*]] = phi ptr [ [[INCDEC_PTR3:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL13]], [[VEC_EPILOG_SCALAR_PH]] ]217; CHECK-NEXT: [[INCDEC_PTR]] = getelementptr inbounds i8, ptr [[PSRC_ADDR_08]], i32 1218; CHECK-NEXT: [[TMP8:%.*]] = load i8, ptr [[PSRC_ADDR_08]], align 2219; CHECK-NEXT: [[TMP9:%.*]] = tail call i8 @llvm.umin.i8(i8 [[TMP8]], i8 [[OFFSET]])220; CHECK-NEXT: [[INCDEC_PTR3]] = getelementptr inbounds i8, ptr [[PDST_ADDR_07]], i32 1221; CHECK-NEXT: store i8 [[TMP9]], ptr [[PDST_ADDR_07]], align 2222; CHECK-NEXT: [[DEC]] = add i32 [[BLKCNT_09]], -1223; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq i32 [[DEC]], 0224; CHECK-NEXT: br i1 [[CMP_NOT]], label [[WHILE_END_LOOPEXIT]], label [[WHILE_BODY]], !llvm.loop [[LOOP9:![0-9]+]]225; CHECK: while.end.loopexit:226; CHECK-NEXT: br label [[WHILE_END]]227; CHECK: while.end:228; CHECK-NEXT: ret void229;230entry:231 %cmp.not6 = icmp eq i32 %blockSize, 0232 br i1 %cmp.not6, label %while.end, label %while.body233 234while.body: ; preds = %entry, %while.body235 %blkCnt.09 = phi i32 [ %dec, %while.body ], [ %blockSize, %entry ]236 %pSrc.addr.08 = phi ptr [ %incdec.ptr, %while.body ], [ %pSrc, %entry ]237 %pDst.addr.07 = phi ptr [ %incdec.ptr3, %while.body ], [ %pDst, %entry ]238 %incdec.ptr = getelementptr inbounds i8, ptr %pSrc.addr.08, i32 1239 %0 = load i8, ptr %pSrc.addr.08, align 2240 %1 = tail call i8 @llvm.umin.i8(i8 %0, i8 %offset)241 %incdec.ptr3 = getelementptr inbounds i8, ptr %pDst.addr.07, i32 1242 store i8 %1, ptr %pDst.addr.07, align 2243 %dec = add i32 %blkCnt.09, -1244 %cmp.not = icmp eq i32 %dec, 0245 br i1 %cmp.not, label %while.end, label %while.body246 247while.end: ; preds = %while.body, %entry248 ret void249}250 251declare i16 @llvm.sadd.sat.i16(i16, i16)252declare i8 @llvm.umin.i8(i8, i8)253 254;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:255; CHECK-COST: {{.*}}256