1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -basicaa -slp-vectorizer -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7 | FileCheck %s 3 4define void @tiny_tree_fully_vectorizable(double* noalias nocapture %dst, double* noalias nocapture readonly %src, i64 %count) #0 { 5; CHECK-LABEL: @tiny_tree_fully_vectorizable( 6; CHECK-NEXT: entry: 7; CHECK-NEXT: [[CMP12:%.*]] = icmp eq i64 [[COUNT:%.*]], 0 8; CHECK-NEXT: br i1 [[CMP12]], label [[FOR_END:%.*]], label [[FOR_BODY:%.*]] 9; CHECK: for.body: 10; CHECK-NEXT: [[I_015:%.*]] = phi i64 [ [[INC:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ] 11; CHECK-NEXT: [[DST_ADDR_014:%.*]] = phi double* [ [[ADD_PTR4:%.*]], [[FOR_BODY]] ], [ [[DST:%.*]], [[ENTRY]] ] 12; CHECK-NEXT: [[SRC_ADDR_013:%.*]] = phi double* [ [[ADD_PTR:%.*]], [[FOR_BODY]] ], [ [[SRC:%.*]], [[ENTRY]] ] 13; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds double, double* [[SRC_ADDR_013]], i64 1 14; CHECK-NEXT: [[TMP0:%.*]] = bitcast double* [[SRC_ADDR_013]] to <2 x double>* 15; CHECK-NEXT: [[TMP1:%.*]] = load <2 x double>, <2 x double>* [[TMP0]], align 8 16; CHECK-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds double, double* [[DST_ADDR_014]], i64 1 17; CHECK-NEXT: [[TMP2:%.*]] = bitcast double* [[DST_ADDR_014]] to <2 x double>* 18; CHECK-NEXT: store <2 x double> [[TMP1]], <2 x double>* [[TMP2]], align 8 19; CHECK-NEXT: [[ADD_PTR]] = getelementptr inbounds double, double* [[SRC_ADDR_013]], i64 [[I_015]] 20; CHECK-NEXT: [[ADD_PTR4]] = getelementptr inbounds double, double* [[DST_ADDR_014]], i64 [[I_015]] 21; CHECK-NEXT: [[INC]] = add i64 [[I_015]], 1 22; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INC]], [[COUNT]] 23; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END]], label [[FOR_BODY]] 24; CHECK: for.end: 25; CHECK-NEXT: ret void 26; 27entry: 28 %cmp12 = icmp eq i64 %count, 0 29 br i1 %cmp12, label %for.end, label %for.body 30 31for.body: ; preds = %entry, %for.body 32 %i.015 = phi i64 [ %inc, %for.body ], [ 0, %entry ] 33 %dst.addr.014 = phi double* [ %add.ptr4, %for.body ], [ %dst, %entry ] 34 %src.addr.013 = phi double* [ %add.ptr, %for.body ], [ %src, %entry ] 35 %0 = load double, double* %src.addr.013, align 8 36 store double %0, double* %dst.addr.014, align 8 37 %arrayidx2 = getelementptr inbounds double, double* %src.addr.013, i64 1 38 %1 = load double, double* %arrayidx2, align 8 39 %arrayidx3 = getelementptr inbounds double, double* %dst.addr.014, i64 1 40 store double %1, double* %arrayidx3, align 8 41 %add.ptr = getelementptr inbounds double, double* %src.addr.013, i64 %i.015 42 %add.ptr4 = getelementptr inbounds double, double* %dst.addr.014, i64 %i.015 43 %inc = add i64 %i.015, 1 44 %exitcond = icmp eq i64 %inc, %count 45 br i1 %exitcond, label %for.end, label %for.body 46 47for.end: ; preds = %for.body, %entry 48 ret void 49} 50 51define void @tiny_tree_fully_vectorizable2(float* noalias nocapture %dst, float* noalias nocapture readonly %src, i64 %count) #0 { 52; CHECK-LABEL: @tiny_tree_fully_vectorizable2( 53; CHECK-NEXT: entry: 54; CHECK-NEXT: [[CMP20:%.*]] = icmp eq i64 [[COUNT:%.*]], 0 55; CHECK-NEXT: br i1 [[CMP20]], label [[FOR_END:%.*]], label [[FOR_BODY:%.*]] 56; CHECK: for.body: 57; CHECK-NEXT: [[I_023:%.*]] = phi i64 [ [[INC:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ] 58; CHECK-NEXT: [[DST_ADDR_022:%.*]] = phi float* [ [[ADD_PTR8:%.*]], [[FOR_BODY]] ], [ [[DST:%.*]], [[ENTRY]] ] 59; CHECK-NEXT: [[SRC_ADDR_021:%.*]] = phi float* [ [[ADD_PTR:%.*]], [[FOR_BODY]] ], [ [[SRC:%.*]], [[ENTRY]] ] 60; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[SRC_ADDR_021]], i64 1 61; CHECK-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds float, float* [[DST_ADDR_022]], i64 1 62; CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds float, float* [[SRC_ADDR_021]], i64 2 63; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds float, float* [[DST_ADDR_022]], i64 2 64; CHECK-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds float, float* [[SRC_ADDR_021]], i64 3 65; CHECK-NEXT: [[TMP0:%.*]] = bitcast float* [[SRC_ADDR_021]] to <4 x float>* 66; CHECK-NEXT: [[TMP1:%.*]] = load <4 x float>, <4 x float>* [[TMP0]], align 4 67; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds float, float* [[DST_ADDR_022]], i64 3 68; CHECK-NEXT: [[TMP2:%.*]] = bitcast float* [[DST_ADDR_022]] to <4 x float>* 69; CHECK-NEXT: store <4 x float> [[TMP1]], <4 x float>* [[TMP2]], align 4 70; CHECK-NEXT: [[ADD_PTR]] = getelementptr inbounds float, float* [[SRC_ADDR_021]], i64 [[I_023]] 71; CHECK-NEXT: [[ADD_PTR8]] = getelementptr inbounds float, float* [[DST_ADDR_022]], i64 [[I_023]] 72; CHECK-NEXT: [[INC]] = add i64 [[I_023]], 1 73; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INC]], [[COUNT]] 74; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END]], label [[FOR_BODY]] 75; CHECK: for.end: 76; CHECK-NEXT: ret void 77; 78entry: 79 %cmp20 = icmp eq i64 %count, 0 80 br i1 %cmp20, label %for.end, label %for.body 81 82for.body: ; preds = %entry, %for.body 83 %i.023 = phi i64 [ %inc, %for.body ], [ 0, %entry ] 84 %dst.addr.022 = phi float* [ %add.ptr8, %for.body ], [ %dst, %entry ] 85 %src.addr.021 = phi float* [ %add.ptr, %for.body ], [ %src, %entry ] 86 %0 = load float, float* %src.addr.021, align 4 87 store float %0, float* %dst.addr.022, align 4 88 %arrayidx2 = getelementptr inbounds float, float* %src.addr.021, i64 1 89 %1 = load float, float* %arrayidx2, align 4 90 %arrayidx3 = getelementptr inbounds float, float* %dst.addr.022, i64 1 91 store float %1, float* %arrayidx3, align 4 92 %arrayidx4 = getelementptr inbounds float, float* %src.addr.021, i64 2 93 %2 = load float, float* %arrayidx4, align 4 94 %arrayidx5 = getelementptr inbounds float, float* %dst.addr.022, i64 2 95 store float %2, float* %arrayidx5, align 4 96 %arrayidx6 = getelementptr inbounds float, float* %src.addr.021, i64 3 97 %3 = load float, float* %arrayidx6, align 4 98 %arrayidx7 = getelementptr inbounds float, float* %dst.addr.022, i64 3 99 store float %3, float* %arrayidx7, align 4 100 %add.ptr = getelementptr inbounds float, float* %src.addr.021, i64 %i.023 101 %add.ptr8 = getelementptr inbounds float, float* %dst.addr.022, i64 %i.023 102 %inc = add i64 %i.023, 1 103 %exitcond = icmp eq i64 %inc, %count 104 br i1 %exitcond, label %for.end, label %for.body 105 106for.end: ; preds = %for.body, %entry 107 ret void 108} 109 110; We do not vectorize the tiny tree which is not fully vectorizable. 111 112define void @tiny_tree_not_fully_vectorizable(double* noalias nocapture %dst, double* noalias nocapture readonly %src, i64 %count) #0 { 113; CHECK-LABEL: @tiny_tree_not_fully_vectorizable( 114; CHECK-NEXT: entry: 115; CHECK-NEXT: [[CMP12:%.*]] = icmp eq i64 [[COUNT:%.*]], 0 116; CHECK-NEXT: br i1 [[CMP12]], label [[FOR_END:%.*]], label [[FOR_BODY:%.*]] 117; CHECK: for.body: 118; CHECK-NEXT: [[I_015:%.*]] = phi i64 [ [[INC:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ] 119; CHECK-NEXT: [[DST_ADDR_014:%.*]] = phi double* [ [[ADD_PTR4:%.*]], [[FOR_BODY]] ], [ [[DST:%.*]], [[ENTRY]] ] 120; CHECK-NEXT: [[SRC_ADDR_013:%.*]] = phi double* [ [[ADD_PTR:%.*]], [[FOR_BODY]] ], [ [[SRC:%.*]], [[ENTRY]] ] 121; CHECK-NEXT: [[TMP0:%.*]] = load double, double* [[SRC_ADDR_013]], align 8 122; CHECK-NEXT: store double [[TMP0]], double* [[DST_ADDR_014]], align 8 123; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds double, double* [[SRC_ADDR_013]], i64 2 124; CHECK-NEXT: [[TMP1:%.*]] = load double, double* [[ARRAYIDX2]], align 8 125; CHECK-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds double, double* [[DST_ADDR_014]], i64 1 126; CHECK-NEXT: store double [[TMP1]], double* [[ARRAYIDX3]], align 8 127; CHECK-NEXT: [[ADD_PTR]] = getelementptr inbounds double, double* [[SRC_ADDR_013]], i64 [[I_015]] 128; CHECK-NEXT: [[ADD_PTR4]] = getelementptr inbounds double, double* [[DST_ADDR_014]], i64 [[I_015]] 129; CHECK-NEXT: [[INC]] = add i64 [[I_015]], 1 130; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INC]], [[COUNT]] 131; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END]], label [[FOR_BODY]] 132; CHECK: for.end: 133; CHECK-NEXT: ret void 134; 135entry: 136 %cmp12 = icmp eq i64 %count, 0 137 br i1 %cmp12, label %for.end, label %for.body 138 139for.body: ; preds = %entry, %for.body 140 %i.015 = phi i64 [ %inc, %for.body ], [ 0, %entry ] 141 %dst.addr.014 = phi double* [ %add.ptr4, %for.body ], [ %dst, %entry ] 142 %src.addr.013 = phi double* [ %add.ptr, %for.body ], [ %src, %entry ] 143 %0 = load double, double* %src.addr.013, align 8 144 store double %0, double* %dst.addr.014, align 8 145 %arrayidx2 = getelementptr inbounds double, double* %src.addr.013, i64 2 146 %1 = load double, double* %arrayidx2, align 8 147 %arrayidx3 = getelementptr inbounds double, double* %dst.addr.014, i64 1 148 store double %1, double* %arrayidx3, align 8 149 %add.ptr = getelementptr inbounds double, double* %src.addr.013, i64 %i.015 150 %add.ptr4 = getelementptr inbounds double, double* %dst.addr.014, i64 %i.015 151 %inc = add i64 %i.015, 1 152 %exitcond = icmp eq i64 %inc, %count 153 br i1 %exitcond, label %for.end, label %for.body 154 155for.end: ; preds = %for.body, %entry 156 ret void 157} 158 159define void @tiny_tree_not_fully_vectorizable2(float* noalias nocapture %dst, float* noalias nocapture readonly %src, i64 %count) #0 { 160; CHECK-LABEL: @tiny_tree_not_fully_vectorizable2( 161; CHECK-NEXT: entry: 162; CHECK-NEXT: [[CMP20:%.*]] = icmp eq i64 [[COUNT:%.*]], 0 163; CHECK-NEXT: br i1 [[CMP20]], label [[FOR_END:%.*]], label [[FOR_BODY:%.*]] 164; CHECK: for.body: 165; CHECK-NEXT: [[I_023:%.*]] = phi i64 [ [[INC:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ] 166; CHECK-NEXT: [[DST_ADDR_022:%.*]] = phi float* [ [[ADD_PTR8:%.*]], [[FOR_BODY]] ], [ [[DST:%.*]], [[ENTRY]] ] 167; CHECK-NEXT: [[SRC_ADDR_021:%.*]] = phi float* [ [[ADD_PTR:%.*]], [[FOR_BODY]] ], [ [[SRC:%.*]], [[ENTRY]] ] 168; CHECK-NEXT: [[TMP0:%.*]] = load float, float* [[SRC_ADDR_021]], align 4 169; CHECK-NEXT: store float [[TMP0]], float* [[DST_ADDR_022]], align 4 170; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[SRC_ADDR_021]], i64 4 171; CHECK-NEXT: [[TMP1:%.*]] = load float, float* [[ARRAYIDX2]], align 4 172; CHECK-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds float, float* [[DST_ADDR_022]], i64 1 173; CHECK-NEXT: store float [[TMP1]], float* [[ARRAYIDX3]], align 4 174; CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds float, float* [[SRC_ADDR_021]], i64 2 175; CHECK-NEXT: [[TMP2:%.*]] = load float, float* [[ARRAYIDX4]], align 4 176; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds float, float* [[DST_ADDR_022]], i64 2 177; CHECK-NEXT: store float [[TMP2]], float* [[ARRAYIDX5]], align 4 178; CHECK-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds float, float* [[SRC_ADDR_021]], i64 3 179; CHECK-NEXT: [[TMP3:%.*]] = load float, float* [[ARRAYIDX6]], align 4 180; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds float, float* [[DST_ADDR_022]], i64 3 181; CHECK-NEXT: store float [[TMP3]], float* [[ARRAYIDX7]], align 4 182; CHECK-NEXT: [[ADD_PTR]] = getelementptr inbounds float, float* [[SRC_ADDR_021]], i64 [[I_023]] 183; CHECK-NEXT: [[ADD_PTR8]] = getelementptr inbounds float, float* [[DST_ADDR_022]], i64 [[I_023]] 184; CHECK-NEXT: [[INC]] = add i64 [[I_023]], 1 185; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INC]], [[COUNT]] 186; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END]], label [[FOR_BODY]] 187; CHECK: for.end: 188; CHECK-NEXT: ret void 189; 190entry: 191 %cmp20 = icmp eq i64 %count, 0 192 br i1 %cmp20, label %for.end, label %for.body 193 194for.body: ; preds = %entry, %for.body 195 %i.023 = phi i64 [ %inc, %for.body ], [ 0, %entry ] 196 %dst.addr.022 = phi float* [ %add.ptr8, %for.body ], [ %dst, %entry ] 197 %src.addr.021 = phi float* [ %add.ptr, %for.body ], [ %src, %entry ] 198 %0 = load float, float* %src.addr.021, align 4 199 store float %0, float* %dst.addr.022, align 4 200 %arrayidx2 = getelementptr inbounds float, float* %src.addr.021, i64 4 201 %1 = load float, float* %arrayidx2, align 4 202 %arrayidx3 = getelementptr inbounds float, float* %dst.addr.022, i64 1 203 store float %1, float* %arrayidx3, align 4 204 %arrayidx4 = getelementptr inbounds float, float* %src.addr.021, i64 2 205 %2 = load float, float* %arrayidx4, align 4 206 %arrayidx5 = getelementptr inbounds float, float* %dst.addr.022, i64 2 207 store float %2, float* %arrayidx5, align 4 208 %arrayidx6 = getelementptr inbounds float, float* %src.addr.021, i64 3 209 %3 = load float, float* %arrayidx6, align 4 210 %arrayidx7 = getelementptr inbounds float, float* %dst.addr.022, i64 3 211 store float %3, float* %arrayidx7, align 4 212 %add.ptr = getelementptr inbounds float, float* %src.addr.021, i64 %i.023 213 %add.ptr8 = getelementptr inbounds float, float* %dst.addr.022, i64 %i.023 214 %inc = add i64 %i.023, 1 215 %exitcond = icmp eq i64 %inc, %count 216 br i1 %exitcond, label %for.end, label %for.body 217 218for.end: ; preds = %for.body, %entry 219 ret void 220} 221 222define void @store_splat(float*, float) { 223; CHECK-LABEL: @store_splat( 224; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds float, float* [[TMP0:%.*]], i64 0 225; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds float, float* [[TMP0]], i64 1 226; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds float, float* [[TMP0]], i64 2 227; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, float* [[TMP0]], i64 3 228; CHECK-NEXT: [[TMP7:%.*]] = insertelement <4 x float> undef, float [[TMP1:%.*]], i32 0 229; CHECK-NEXT: [[TMP8:%.*]] = insertelement <4 x float> [[TMP7]], float [[TMP1]], i32 1 230; CHECK-NEXT: [[TMP9:%.*]] = insertelement <4 x float> [[TMP8]], float [[TMP1]], i32 2 231; CHECK-NEXT: [[TMP10:%.*]] = insertelement <4 x float> [[TMP9]], float [[TMP1]], i32 3 232; CHECK-NEXT: [[TMP11:%.*]] = bitcast float* [[TMP3]] to <4 x float>* 233; CHECK-NEXT: store <4 x float> [[TMP10]], <4 x float>* [[TMP11]], align 4 234; CHECK-NEXT: ret void 235; 236 %3 = getelementptr inbounds float, float* %0, i64 0 237 store float %1, float* %3, align 4 238 %4 = getelementptr inbounds float, float* %0, i64 1 239 store float %1, float* %4, align 4 240 %5 = getelementptr inbounds float, float* %0, i64 2 241 store float %1, float* %5, align 4 242 %6 = getelementptr inbounds float, float* %0, i64 3 243 store float %1, float* %6, align 4 244 ret void 245} 246 247define void @store_const(i32* %a) { 248; CHECK-LABEL: @store_const( 249; CHECK-NEXT: entry: 250; CHECK-NEXT: [[PTR0:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 0 251; CHECK-NEXT: [[PTR1:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 1 252; CHECK-NEXT: [[PTR2:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 2 253; CHECK-NEXT: [[PTR3:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 3 254; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[PTR0]] to <4 x i32>* 255; CHECK-NEXT: store <4 x i32> <i32 10, i32 30, i32 20, i32 40>, <4 x i32>* [[TMP0]], align 4 256; CHECK-NEXT: ret void 257; 258entry: 259 %ptr0 = getelementptr inbounds i32, i32* %a, i64 0 260 store i32 10, i32* %ptr0, align 4 261 %ptr1 = getelementptr inbounds i32, i32* %a, i64 1 262 store i32 30, i32* %ptr1, align 4 263 %ptr2 = getelementptr inbounds i32, i32* %a, i64 2 264 store i32 20, i32* %ptr2, align 4 265 %ptr3 = getelementptr inbounds i32, i32* %a, i64 3 266 store i32 40, i32* %ptr3, align 4 267 ret void 268} 269