1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -loop-vectorize -force-target-max-vector-interleave=1 -mtriple=riscv64 -mattr=+v -riscv-v-vector-bits-min=128 -scalable-vectorization=off -S | FileCheck %s --check-prefix=LMUL1 3; RUN: opt < %s -loop-vectorize -force-target-max-vector-interleave=1 -mtriple=riscv32 -mattr=+v -riscv-v-vector-bits-min=128 -scalable-vectorization=off -S | FileCheck %s --check-prefix=LMUL1 4; RUN: opt < %s -loop-vectorize -force-target-max-vector-interleave=1 -mtriple=riscv64 -mattr=+v -riscv-v-vector-bits-min=128 -scalable-vectorization=off -riscv-v-register-bit-width-lmul=2 -S | FileCheck %s --check-prefix=LMUL2 5; RUN: opt < %s -loop-vectorize -force-target-max-vector-interleave=1 -mtriple=riscv32 -mattr=+v -riscv-v-vector-bits-min=128 -scalable-vectorization=off -riscv-v-register-bit-width-lmul=2 -S | FileCheck %s --check-prefix=LMUL2 6 7; Function Attrs: nounwind 8define i32* @array_add(i32* noalias nocapture readonly %a, i32* noalias nocapture readonly %b, i32* %c, i32 %size) { 9; LMUL1-LABEL: @array_add( 10; LMUL1-NEXT: entry: 11; LMUL1-NEXT: [[CMP10:%.*]] = icmp sgt i32 [[SIZE:%.*]], 0 12; LMUL1-NEXT: br i1 [[CMP10]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]] 13; LMUL1: for.body.preheader: 14; LMUL1-NEXT: [[TMP0:%.*]] = add i32 [[SIZE]], -1 15; LMUL1-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 16; LMUL1-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 17; LMUL1-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 4 18; LMUL1-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 19; LMUL1: vector.ph: 20; LMUL1-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 4 21; LMUL1-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]] 22; LMUL1-NEXT: br label [[VECTOR_BODY:%.*]] 23; LMUL1: vector.body: 24; LMUL1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 25; LMUL1-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 0 26; LMUL1-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP3]] 27; LMUL1-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[TMP4]], i32 0 28; LMUL1-NEXT: [[TMP6:%.*]] = bitcast i32* [[TMP5]] to <4 x i32>* 29; LMUL1-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, <4 x i32>* [[TMP6]], align 4 30; LMUL1-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i64 [[TMP3]] 31; LMUL1-NEXT: [[TMP8:%.*]] = getelementptr inbounds i32, i32* [[TMP7]], i32 0 32; LMUL1-NEXT: [[TMP9:%.*]] = bitcast i32* [[TMP8]] to <4 x i32>* 33; LMUL1-NEXT: [[WIDE_LOAD1:%.*]] = load <4 x i32>, <4 x i32>* [[TMP9]], align 4 34; LMUL1-NEXT: [[TMP10:%.*]] = add nsw <4 x i32> [[WIDE_LOAD1]], [[WIDE_LOAD]] 35; LMUL1-NEXT: [[TMP11:%.*]] = getelementptr inbounds i32, i32* [[C:%.*]], i64 [[TMP3]] 36; LMUL1-NEXT: [[TMP12:%.*]] = getelementptr inbounds i32, i32* [[TMP11]], i32 0 37; LMUL1-NEXT: [[TMP13:%.*]] = bitcast i32* [[TMP12]] to <4 x i32>* 38; LMUL1-NEXT: store <4 x i32> [[TMP10]], <4 x i32>* [[TMP13]], align 4 39; LMUL1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 40; LMUL1-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 41; LMUL1-NEXT: br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 42; LMUL1: middle.block: 43; LMUL1-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] 44; LMUL1-NEXT: br i1 [[CMP_N]], label [[FOR_END_LOOPEXIT:%.*]], label [[SCALAR_PH]] 45; LMUL1: scalar.ph: 46; LMUL1-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER]] ] 47; LMUL1-NEXT: br label [[FOR_BODY:%.*]] 48; LMUL1: for.body: 49; LMUL1-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] 50; LMUL1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV]] 51; LMUL1-NEXT: [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4 52; LMUL1-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDVARS_IV]] 53; LMUL1-NEXT: [[TMP16:%.*]] = load i32, i32* [[ARRAYIDX2]], align 4 54; LMUL1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP16]], [[TMP15]] 55; LMUL1-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i32, i32* [[C]], i64 [[INDVARS_IV]] 56; LMUL1-NEXT: store i32 [[ADD]], i32* [[ARRAYIDX4]], align 4 57; LMUL1-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1 58; LMUL1-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32 59; LMUL1-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[SIZE]] 60; LMUL1-NEXT: br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]] 61; LMUL1: for.end.loopexit: 62; LMUL1-NEXT: br label [[FOR_END]] 63; LMUL1: for.end: 64; LMUL1-NEXT: ret i32* [[C]] 65; 66; LMUL2-LABEL: @array_add( 67; LMUL2-NEXT: entry: 68; LMUL2-NEXT: [[CMP10:%.*]] = icmp sgt i32 [[SIZE:%.*]], 0 69; LMUL2-NEXT: br i1 [[CMP10]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]] 70; LMUL2: for.body.preheader: 71; LMUL2-NEXT: [[TMP0:%.*]] = add i32 [[SIZE]], -1 72; LMUL2-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 73; LMUL2-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 74; LMUL2-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 8 75; LMUL2-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 76; LMUL2: vector.ph: 77; LMUL2-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 8 78; LMUL2-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]] 79; LMUL2-NEXT: br label [[VECTOR_BODY:%.*]] 80; LMUL2: vector.body: 81; LMUL2-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 82; LMUL2-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 0 83; LMUL2-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP3]] 84; LMUL2-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[TMP4]], i32 0 85; LMUL2-NEXT: [[TMP6:%.*]] = bitcast i32* [[TMP5]] to <8 x i32>* 86; LMUL2-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i32>, <8 x i32>* [[TMP6]], align 4 87; LMUL2-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i64 [[TMP3]] 88; LMUL2-NEXT: [[TMP8:%.*]] = getelementptr inbounds i32, i32* [[TMP7]], i32 0 89; LMUL2-NEXT: [[TMP9:%.*]] = bitcast i32* [[TMP8]] to <8 x i32>* 90; LMUL2-NEXT: [[WIDE_LOAD1:%.*]] = load <8 x i32>, <8 x i32>* [[TMP9]], align 4 91; LMUL2-NEXT: [[TMP10:%.*]] = add nsw <8 x i32> [[WIDE_LOAD1]], [[WIDE_LOAD]] 92; LMUL2-NEXT: [[TMP11:%.*]] = getelementptr inbounds i32, i32* [[C:%.*]], i64 [[TMP3]] 93; LMUL2-NEXT: [[TMP12:%.*]] = getelementptr inbounds i32, i32* [[TMP11]], i32 0 94; LMUL2-NEXT: [[TMP13:%.*]] = bitcast i32* [[TMP12]] to <8 x i32>* 95; LMUL2-NEXT: store <8 x i32> [[TMP10]], <8 x i32>* [[TMP13]], align 4 96; LMUL2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 97; LMUL2-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 98; LMUL2-NEXT: br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 99; LMUL2: middle.block: 100; LMUL2-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] 101; LMUL2-NEXT: br i1 [[CMP_N]], label [[FOR_END_LOOPEXIT:%.*]], label [[SCALAR_PH]] 102; LMUL2: scalar.ph: 103; LMUL2-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER]] ] 104; LMUL2-NEXT: br label [[FOR_BODY:%.*]] 105; LMUL2: for.body: 106; LMUL2-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] 107; LMUL2-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV]] 108; LMUL2-NEXT: [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4 109; LMUL2-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDVARS_IV]] 110; LMUL2-NEXT: [[TMP16:%.*]] = load i32, i32* [[ARRAYIDX2]], align 4 111; LMUL2-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP16]], [[TMP15]] 112; LMUL2-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i32, i32* [[C]], i64 [[INDVARS_IV]] 113; LMUL2-NEXT: store i32 [[ADD]], i32* [[ARRAYIDX4]], align 4 114; LMUL2-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1 115; LMUL2-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32 116; LMUL2-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[SIZE]] 117; LMUL2-NEXT: br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]] 118; LMUL2: for.end.loopexit: 119; LMUL2-NEXT: br label [[FOR_END]] 120; LMUL2: for.end: 121; LMUL2-NEXT: ret i32* [[C]] 122; 123entry: 124 %cmp10 = icmp sgt i32 %size, 0 125 br i1 %cmp10, label %for.body.preheader, label %for.end 126 127for.body.preheader: ; preds = %entry 128 br label %for.body 129 130for.body: ; preds = %for.body.preheader, %for.body 131 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ] 132 %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv 133 %0 = load i32, i32* %arrayidx, align 4 134 %arrayidx2 = getelementptr inbounds i32, i32* %b, i64 %indvars.iv 135 %1 = load i32, i32* %arrayidx2, align 4 136 %add = add nsw i32 %1, %0 137 %arrayidx4 = getelementptr inbounds i32, i32* %c, i64 %indvars.iv 138 store i32 %add, i32* %arrayidx4, align 4 139 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 140 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 141 %exitcond = icmp eq i32 %lftr.wideiv, %size 142 br i1 %exitcond, label %for.end.loopexit, label %for.body 143 144for.end.loopexit: ; preds = %for.body 145 br label %for.end 146 147for.end: ; preds = %for.end.loopexit, %entry 148 ret i32* %c 149} 150