122 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "^scalar.ph:" --version 52; RUN: opt -passes=loop-vectorize -enable-epilogue-vectorization=false -mattr=+neon -S < %s | FileCheck %s3 4target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"5target triple = "aarch64-none-unknown-elf"6 7define i32 @not_dotp(ptr %a, ptr %b) {8; CHECK-LABEL: define i32 @not_dotp(9; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0:[0-9]+]] {10; CHECK-NEXT: [[ENTRY:.*:]]11; CHECK-NEXT: br label %[[VECTOR_PH:.*]]12; CHECK: [[VECTOR_PH]]:13; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]14; CHECK: [[VECTOR_BODY]]:15; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]16; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <16 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP13:%.*]], %[[VECTOR_BODY]] ]17; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <16 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP14:%.*]], %[[VECTOR_BODY]] ]18; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]19; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[TMP1]], i64 1620; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 121; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 122; CHECK-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>23; CHECK-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>24; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]25; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[TMP6]], i64 1626; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP6]], align 127; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP8]], align 128; CHECK-NEXT: [[TMP9:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>29; CHECK-NEXT: [[TMP10:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>30; CHECK-NEXT: [[TMP11:%.*]] = mul <16 x i32> [[TMP9]], [[TMP4]]31; CHECK-NEXT: [[TMP12:%.*]] = mul <16 x i32> [[TMP10]], [[TMP5]]32; CHECK-NEXT: [[TMP13]] = add <16 x i32> [[TMP11]], [[VEC_PHI]]33; CHECK-NEXT: [[TMP14]] = add <16 x i32> [[TMP12]], [[VEC_PHI1]]34; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 3235; CHECK-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], 99236; CHECK-NEXT: br i1 [[TMP15]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]37; CHECK: [[MIDDLE_BLOCK]]:38; CHECK-NEXT: [[BIN_RDX:%.*]] = add <16 x i32> [[TMP14]], [[TMP13]]39; CHECK-NEXT: [[TMP16:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[BIN_RDX]])40; CHECK-NEXT: br label %[[SCALAR_PH:.*]]41; CHECK: [[SCALAR_PH]]:42;43entry:44 br label %for.body45 46for.body: ; preds = %for.body, %entry47 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]48 %accum = phi i32 [ 0, %entry ], [ %add, %for.body ]49 %gep.a = getelementptr i8, ptr %a, i64 %iv50 %load.a = load i8, ptr %gep.a, align 151 %ext.a = zext i8 %load.a to i3252 %gep.b = getelementptr i8, ptr %b, i64 %iv53 %load.b = load i8, ptr %gep.b, align 154 %ext.b = zext i8 %load.b to i3255 %mul = mul i32 %ext.b, %ext.a56 %add = add i32 %mul, %accum57 %iv.next = add i64 %iv, 158 %exitcond.not = icmp eq i64 %iv.next, 100059 br i1 %exitcond.not, label %for.exit, label %for.body60 61for.exit: ; preds = %for.body62 ret i32 %add63}64 65; Test to ensure that we don't crash when evaluating an extend from a type66; that is not a factor of the target type.67define i40 @partial_reduce_not_known_factor(i32 %a, i32 %b, i16 %N) {68; CHECK-LABEL: define i40 @partial_reduce_not_known_factor(69; CHECK-SAME: i32 [[A:%.*]], i32 [[B:%.*]], i16 [[N:%.*]]) #[[ATTR0]] {70; CHECK-NEXT: [[ENTRY:.*:]]71; CHECK-NEXT: [[SMAX:%.*]] = call i16 @llvm.smax.i16(i16 [[N]], i16 0)72; CHECK-NEXT: [[TMP0:%.*]] = zext nneg i16 [[SMAX]] to i3273; CHECK-NEXT: [[TMP1:%.*]] = add nuw nsw i32 [[TMP0]], 174; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP1]], 475; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]76; CHECK: [[VECTOR_PH]]:77; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP1]], 478; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP1]], [[N_MOD_VF]]79; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[B]], i64 080; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer81; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <2 x i32> poison, i32 [[A]], i64 082; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT1]], <2 x i32> poison, <2 x i32> zeroinitializer83; CHECK-NEXT: [[TMP2:%.*]] = trunc i32 [[N_VEC]] to i1684; CHECK-NEXT: [[TMP3:%.*]] = sext <2 x i32> [[BROADCAST_SPLAT2]] to <2 x i40>85; CHECK-NEXT: [[TMP4:%.*]] = sext <2 x i32> [[BROADCAST_SPLAT]] to <2 x i40>86; CHECK-NEXT: [[TMP5:%.*]] = or <2 x i40> [[TMP4]], [[TMP3]]87; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]88; CHECK: [[VECTOR_BODY]]:89; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]90; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x i40> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ]91; CHECK-NEXT: [[VEC_PHI3:%.*]] = phi <2 x i40> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP8:%.*]], %[[VECTOR_BODY]] ]92; CHECK-NEXT: [[TMP6]] = or <2 x i40> [[VEC_PHI]], [[TMP5]]93; CHECK-NEXT: [[TMP8]] = or <2 x i40> [[VEC_PHI3]], [[TMP5]]94; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 495; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]96; CHECK-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]97; CHECK: [[MIDDLE_BLOCK]]:98; CHECK-NEXT: [[BIN_RDX:%.*]] = or <2 x i40> [[TMP8]], [[TMP6]]99; CHECK-NEXT: [[TMP9:%.*]] = call i40 @llvm.vector.reduce.or.v2i40(<2 x i40> [[BIN_RDX]])100; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP1]], [[N_VEC]]101; CHECK-NEXT: br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]102; CHECK: [[SCALAR_PH]]:103;104entry:105 br label %for.body106 107for.body:108 %red = phi i40 [ 0, %entry ], [ %1, %for.body ]109 %iv = phi i16 [ 0, %entry ], [ %iv.next, %for.body ]110 %resize = sext i32 %a to i40111 %resize4 = sext i32 %b to i40112 %0 = or i40 %resize4, %resize113 %1 = or i40 %red, %0114 %iv.next = add i16 %iv, 1115 %cmp = icmp slt i16 %iv, %N116 br i1 %cmp, label %for.body, label %exit117 118exit:119 %result.lcssa = phi i40 [ %1, %for.body ]120 ret i40 %result.lcssa121}122